Neural Mechanisms Underlying Human Auditory Evoked Responses Revealed By Human Neocortical Neurosolver — TEI-tagged view

KohlpersName, ParviainenpersName, JonespersName (2022date). “Neural Mechanisms Underlying Human Auditory Evoked Responses Revealed By Human Neocortical Neurosolver.” Brain Topographytitle·j.

Object of study — the cell and circuit mechanisms that generate the macroscale auditory evoked field, and whether a modelterm·mth built for somatosensory cortex can account for them

Research question — whether the laminar organisation and relay of sensoryterm information is preserved well enough across sensoryterm areas that the same cortical inputterm sequence reproduces auditory and somatosensory evoked responses

Contribution — the same feedforward-feedback-feedforward inputterm sequence reproduces the auditory evoked field, and the right-hemisphere and contralateralterm dominancesterm can each be reproduced by changing a small number of inputterm parametersterm

Full text — tagged

The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.

Neural Mechanisms Underlying Humanterm Auditory Evoked Responses Revealed By Humanterm Neocortical Neurosolver Carmen KohlpersName Carmen KohlpersName 1 Department of Neuroscience, Carney Institute for Brain Sciences, Brown University, Providence, USA 1, ✉ , Tiina ParviainenpersName Tiina ParviainenpersName 2 Centre for Interdisciplinary Brain Research, Department of Psychology, University of Jyväskylä, P.O. Box 35, 40014 Jyväskylä, Finland 3 Megterm Core Aalto Neuroimaging, Aalto University, AALTO, P.O. Box 15100, 00076 Espoo, Finland 2, 3 , Stephanie R JonespersName Stephanie R JonespersName 1 Department of Neuroscience, Carney Institute for Brain Sciences, Brown University, Providence, USA 4 Center for Neurorestoration and Neurotechnology, Providence VAMC, Providence, USA 1, 4 1 Department of Neuroscience, Carney Institute for Brain Sciences, Brown University, Providence, USA 2 Centre for Interdisciplinary Brain Research, Department of Psychology, University of Jyväskylä, P.O. Box 35, 40014 Jyväskylä, Finland 3 Megterm Core Aalto Neuroimaging, Aalto University, AALTO, P.O. Box 15100, 00076 Espoo, Finland 4 Center for Neurorestoration and Neurotechnology, Providence VAMC, Providence, USA Handling Editor: Jeremie Lefebvre.

✉ Corresponding author.

Received 2020 Dec 29; Accepted 2021 Mar 30; Issue date 2022.

© The Author(s) 2021 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/ .

PMCID: PMC8813713 PMID: 33876329 Abstract Auditory evoked fields (AEFsterm) are commonly studied, yet their underlying neural mechanisms remain poorly understood. Here, we used the biophysical modelling software Humanterm Neocortical Neurosolver (HNNterm) whose foundation is a canonicalterm·dom neocortical circuit modelterm·mth to interpret the cell and network mechanisms contributing to macroscale AEFsterm elicited by a simple tone, measured with magnetoencephalography. We found that AEFsterm can be reproduced by activating the neocortical circuit through a layer specific sequence of feedforwardterm and feedbackterm excitatory synaptic drives, similar to prior simulationterm·mth of somatosensory evoked responses, supporting the notion that basic structures and activation patterns are preserved across sensoryterm regions. We also applied the modeling framework to develop and test predictionsterm on neural mechanisms underlying AEFterm differences in the left and right hemispheresterm, as well as in hemispheresterm contralateralterm and ipsilateralterm to the presentation of the auditory stimulus. We found that increasing the strength of the excitatory synaptic cortical feedback inputsterm·def to supragranular layers simulates the commonly observed right hemisphereterm dominanceterm, while decreasing the inputterm latencies and simultaneously increasing the number of cells contributing to the signal accounted for the contralateralterm dominanceterm. These results provide a direct link between humanterm data and prior animalterm studies and lay the foundation for future translational research examining the mechanisms underlying alteration in this fundamental biomarkerterm of auditory processing in healthy cognition and neuropathology.

Supplementary Information The online version contains supplementary material available at 10.1007/s10548-021-00838-0.

Keywords: Auditory processing, AEFterm, MEGterm, Biophysical modelterm·mth, HNNterm Introduction Brain activity evoked by auditory stimulation has been studied for many decades and remains not only commonly used in cognitive neuroscience (Wagner et al. 2017 ; ParviainenpersName et al. 2019 ) but also clinically relevant (Paulraj et al. 2015 ; Samatra et al. 2020 ). In humansterm, the sequence of neural activation evoked by auditory stimulation can be measured using electroencephalography (EEGterm) or magnetoencephalography (MEGterm), and the surface-recorded responses are typically divided into three categories based on their latencyterm. Broadly speaking, early responses (within ~ 10 ms) primarily reflect brain stem activity, middle-latency auditory responses (10–50 ms) are thought to reflect processing in thalamocortical structures, and late-latency responses (50–250 ms) are associated with cortical activity (Picton et al. 1974 ). Late latencyterm responses are elicited in the primary auditory cortexterm and surrounding areas and typically consist of components labelled P50m-N100m-P200m (or P1-N1-P2), which peak at around 50, 100 and 180 ms respectively (Eggermont and Ponton 2002 ). Here, we refer to this cortical P50m-N100m-P200m sequence as auditory evoked fields (AEF)term·def. Although the P50m-N100m-P200m waveform is commonly found in response to any auditory event, the shape and amplitudeterm of its components depends on a number of characteristics of the evoking stimuli, such as their acoustic complexity, intensity, duration, and frequency (Picton 2011 ). Here, we use simple pure tones at 1 kHz, which have been shown to produce robust N1 responses and are neutral in terms of linguistic associations, and adopted a paradigm that has been successfully used to evidence the fundamental response properties of the humanterm auditory processing pathway (Mäkelä et al. 1994 ; Salmelin et al. 1999 ).

While AEFsterm are commonly used as correlates for a number of cognitive processes (Shahin et al. 2003 ; Fan et al. 2017 ), developmental stages (ParviainenpersName et al. 2019 ; van Bijnen et al. 2019 ), and clinical observations (Goodin et al. 1978 ; Stephen et al. 2017 ), several fundamental questions about their neural origin remain unanswered. A deeper understanding of the cellular and circuit mechanisms generating AEFsterm is essential to understanding the role of AEFterm components in cognition and to developing treatments based on regularizing neuropathological AEFterm features.

Insights into the biophysical mechanisms underlying auditory processing have been gained using invasive recordings in animalterm modelsterm·mth. An early description of a fundamental anatomical circuit of sensoryterm processing was based on cat visual cortex (Gilbert 1983 ; Gilbert and Wiesel 1989 ) and states that excitatory cells in layer IVterm of the neocortex receive thalamicterm inputsterm and project to superficial layers, which in turn project to deeper layers. This fundamental layer specific relay of sensoryterm information has since been confirmed in other sensoryterm systems, including somatosensory (Di et al. 1990 ; Schroeder et al. 1995 ) and auditory cortexterm (Ojima et al. 1991 , 1992 ), with confirmation from a number of studies and speciesterm (e.g. Mitani and Shimokouchi 1985 ; Mitani et al. 1985 ; Pandya and Rosene 1993 ; Huang and Winer 2000 ; Sakata and Harris 2009 ; Atencio and Schreiner 2010 ; Ji et al. 2016 ). Importantly, the commonality of this laminar sequence of sensoryterm information flow in different sensoryterm areas suggests that the basic laminar structure and organization is similar across sensoryterm regions and deterministic of the early sensoryterm response, particularly in primary sensoryterm cortices (Douglas and Martin 2004 ; Barbour and Callaway 2008 ; Atencio and Schreiner 2010 ).

Although animalterm modelsterm·mth provide insight into cell and circuit-level dynamics underlying sensoryterm evoked responses, determining if and how this information translates to macroscale humanterm signals, as measured with EEGterm and MEGterm, is a non-trivial problem. In recent years, computational neuroscience has begun to bridge the gap between macroscale humanterm brain signals and network-dynamics (Kiebel et al. 2009 ; Sanz Leon et al. 2013 ; Hagen et al. 2018 ). One approach to achieve this is proposed by the modeling software Humanterm Neocortical Neurosolver (HNNterm; Neymotin et al. 2020 ), namely to create a modelterm·mth of neuronal circuitry informed by invasive animalterm recordings and use this modelterm·mth to simulate humanterm macroscale brain signals based on their biophysical origin. The modelterm·mth underlying HNNterm consists of a canonicalterm·dom neocortical circuit, with individual excitatory and inhibitoryterm neurons across the cortical layers, and layer specific thalamocortical and cortico-cortical inputterm pathways. The primary electrical currents generating EEGterm and MEGterm signals are simulated from the intracellular current flow in the long and spatially aligned cortical pyramidalterm neuron apical dendrites (for a detailed description, see Methods , supplement S1, or visit hnnterm.brown.edu). HNNterm has been applied to study the cellular and circuit level neural mechanism underlying a number of commonly measured EEGterm and MEGterm signals, such as alpha and beta frequency oscillations (JonespersName et al. 2009 ; Ziegler et al. 2010 ; Sherman et al. 2016 ), gamma oscillations (Lee and JonespersName 2013 ), and somatosensory evoked fields (SEFterm; JonespersName et al. 2007 , 2009 ; Ziegler et al. 2010 ; Sliva et al. 2018 ). Here, we build from the prior literature and known commonalities in the cortical structure and information relay in sensoryterm areas to apply HNNterm to study the circuit mechanisms underlying MEGterm measured AEFsterm. We focus on interpreting the neural mechanisms generating (1) the AEFterm waveform in response to a simple auditory tone (ParviainenpersName et al. 2019 ), including the P50m-N100m-P200m sequence, (2) observed differences in AEFsterm in the right and left hemisphereterm, and (3) between contralateralterm and ipsilateralterm tone presentationsterm·var. We hypothesized that the laminar organization and relay of sensoryterm information is preserved across sensoryterm areas, and that, as a result, AEFsterm could be simulated using similar cortical inputterm sequences as those shown to reproduce SEFsterm in our prior studies. Note that, while the laminar organization and exogenous drives simulated in HNNterm are based on animalterm work (see Discussion ), previous work has shown that the resulting simulationsterm·mth can be applied to humanterm data and support canonicalterm·dom inputterm sequences (JonespersName et al. 2007 , 2009 ; Ziegler et al. 2010 ; Sliva et al. 2018 ; Neymotin et al. 2020 ). Building from this hypothesisterm, we then investigated the observed phenomenon that the AEFterm recorded over the right hemisphereterm, and specifically the N100mterm component, is often larger in amplitudeterm compared to the left AEFterm (Peronnet et al. 1974 ; Mononen and Seitz 1977 ; Wolpaw and Penry 1977 ; Hine and Debener 2007 ; Howard and Poeppel 2009 ; Kimura 2011 ; Shaw et al. 2013 ). The biophysical origin of this difference is unknown, and we apply HNNterm to develop targeted predictionsterm on neural mechanisms that may generate these differences.

Lastly, we examine the mechanisms underlying AEFterm differences that depend on the site of auditory stimulation, i.e. between contralateralterm and ipsilateralterm responses. In early AEFterm research, it was found that when simple sounds were presented monaurally, responses in the hemisphereterm contralateralterm to the sound presentation were more pronounced than responses in the ipsilateralterm hemisphereterm (Tunturi 1946 ; Rosenzweig 1951 ). This increased magnitude in contralateralterm auditory responses has since been confirmed in a number of experimental paradigms and recording modalities (Peronnet et al. 1974 ; Andreassi et al. 1975 ; Taub et al. 1976 ; Wolpaw and Penry 1977 ; Reite et al. 1981 ; Pantev et al. 1986 , 1998 ; Yoshiura et al. 1994 ; Loveless et al. 1994 ; Mäkelä et al. 1994 ; Jäncke et al. 2002 ; Devlin et al. 2003 ; Petkov et al. 2004 ; ParviainenpersName et al. 2019 ). In addition, the contralateralterm response has been described as not only larger in magnitude, but also as faster than its ipsilateralterm equivalent (Mononen and Seitz 1977 ; Wolpaw and Penry 1977 ; Mäkelä et al. 1994 ; Pantev et al. 1998 ). However, this temporal effect appears to be less clear than the effect on magnitude, as several studies failed to confirm any temporal differences (Tunturi 1946 ; Andreassi et al. 1975 ; Yoshiura et al. 1994 ).

While there is strong evidence suggesting that the contralateralterm dominanceterm effect is robust (although there is some debate, e.g. Hine and Debener 2007 ), few studies have focused on the biophysical mechanisms underlying this phenomenon. These studies guide our neural modelterm·mth investigation of observed contralateralterm AEFterm dominanceterm in our data. One explanation for contralateralterm dominanceterm is that, while auditory processing is not as lateralized as, for example, somatosensory processing, those pathways between ear and cortex which cross over nevertheless appear dominant, with greater numbers of fibers and faster transmission speeds (Rosenzweig 1951 ; Kimura 1961 ). This feature is most likely to account for temporal differences, as stronger pathways may lead to faster transmission of information, resulting in earlier contralateralterm AEFterm peaks. However, it is not immediately apparent that the integrity of fiber tracts alone necessarily leads to responses of greater magnitude. A second explanation states that larger areas of cortex are activated by contralateralterm stimulation (Rosenzweig 1951 ; Gross et al. 1967 ; Pantev et al. 1986 ), which would lead to greater amplitudeterm responses. Gross et al. ( 1967 ) found that the cortical area in which responses could be evoked was larger when tones were presented contralaterally as opposed to ipsilaterally. Importantly, these explanations are based on evidence almost exclusively from animalterm modelsterm·mth, and it is unknown if they can account for the contralateralterm dominanceterm observed in humanterm macroscale AEFsterm. We therefore applied our HNNterm modeling framework to test the hypothesisterm that changes in modelterm·mth parametersterm that represent the speed of activation of the cortex, and number of cells contributing to the signal, could account for the observed differences in contralateralterm vs. ipsilateralterm MEGterm measured AEFsterm.

In summary, the goals of the current study were threefold. (1) We aimed to simulate an AEFterm waveform in HNNterm and hypothesized that the cortical inputsterm underlying the waveform share important features with those previously shown to be involved in other sensoryterm evoked responses, due to the canonicalterm·dom structure of sensoryterm neocortex. (2) We used HNNterm to explore which biophysical mechanisms underlie the commonly observed difference between right hemisphereterm and left hemisphereterm AEFsterm. (3) We tested the hypothesisterm that differences in the scale of the underlying network as well as the latencies of the inputsterm into the network could account for the previously reported dominanceterm of contralateralterm AEFsterm.

Methods Participants Participant recruitment, MEGterm recording and source-localization were performed for a previous study. For a more detailed description, see ParviainenpersName et al. ( 2005 , 2019 ).

Ten neurotypical participants (age 23–39; five females) were recruited for MEGterm recording.

Auditory Stimulation Participants were presented with simple 1 kHz sine wave tones, 50 ms in duration with 10 ms fade-in and fade-out time, created in Sound Edit (MacroMedia, San Francisco, CA, USA). The tones were presented alternatingly to the left and right ear (ParviainenpersName et al. 2019 ). Tones were separated by inter-stimulus-intervals varying between 0.8 and 1.2 s and were presented at 60 dB above the subjective hearing level.

MEGterm Acquisition and Analysis MEGterm was recorded using a helmet-shaped 306-channel whole-head system (Vectorview, Neuromag Ltd, Helsinki, Finland), band-pass filtered at 0.03–200 Hz and sampled at 600 Hz. Segments of − 200 ms to 800 ms relative to auditory stimulus onset were averaged off-line for each participant, separately for left and right ear sounds. Electro-oculogram (EOG) was recorded and epochs in which the EOG channel exceeded a threshold of 150 µV were considered contaminated by blinks or saccades, and were excluded from the averages. After artifact rejection, an average of 102 (± 4)/105 (± 4) (mean ± SD) epochs remained per subject for left/right ear sounds (see ParviainenpersName et al. 2005 , 2019 ).

The resulting averages were source-localized using equivalent current dipole (ECD)term·def modeling, where ECDs represent the average distribution of electric current within a cortical patch, giving an estimate of the location, strength, and direction of local current flow (Hämäläinen et al. 1993 ). For determining the ECD, the temporally varying field pattern was visually inspected between 80 and 120 ms relative to the auditory stimulus to identify symmetric dipolar fields, indicative of separable, active neuron population. The N100mterm peak visible at the sensor-level was always associated with salient dipolar field pattern (cf. Fig. 2 d, e). ECDs were determined for each participant (one dipoleterm per hemisphereterm) from a standard subset of 46 planar gradiometers that covered the 100 ms auditory field pattern (Fig. 2 ). The determined ECDs in each hemisphereterm were then used to account for the MEGterm signals by keeping ECD location and orientations fixed and varying only their amplitudeterm. For further analysis in this study, source-localized AEFsterm were segmented into 0 ms to 250 ms epochs, relative to the auditory stimulus, and averaged across participants. The same pair of ECDs, at an individual level, accounted for the activation patterns evoked by both ipsilateralterm and contralateralterm stimulation. Moreover, the ECD that was fitted to the 100 ms response (N100mterm) was also able to account for the preceding response at 50 ms (P50mterm). The current direction of the N100mterm peak was into the cortex, and the P50mterm and P200mterm have amplitudesterm in the opposite direction representing current out of the cortex.

Fig. 2.

Source-localized grand average AEFsterm: a , b AEFterm waveforms in the left ( a ) and right ( b ), as well as the contralateralterm (blue) and ipsilateralterm (red) hemisphereterm. Inserts show average source locations. c For ease of comparison, ( c ) shows left (dotted line, corresponding to blue line in ( a ) and right (solid line, corresponding to blue line in ( b ) contralateralterm AEFsterm overlaid. Shaded areas indicate SE. *Significantterm differences in N100m amplitudeterm·var at p < 0.05. d , e Magnetic field patterns (at 100 ms) of one example subject in the left ( d ) and right ( e ), as well as contralateralterm (blue) and ipsilateralterm (red) hemispheresterm Source localization is an important step since we intended to modelterm·mth the AEFterm waveforms using the Humanterm Neocortical Neurosolver software (HNNterm, see below), which is designed to study the origin of source localized signals. Inverse solution methods applied to sensor-level MEGterm (or EEGterm) estimate the primary current generator of the sensor data. These primary currents are estimated as current dipolesterm in units of current × distance (Ampere × meters) and known to be generated by intracellular post-synaptic current flow in the long and spatially aligned cortical pyramidalterm neuron dendrites (Ikeda and Shibasaki 1992 ; Hämäläinen et al. 1993 ; Okada et al. 1997 ; Murakami et al. 2003 ; Murakami and Okada 2006 ; Sacchet et al. 2015 ; Neymotin et al. 2020 ). As detailed below, HNNterm simulates these primary currents from the modelterm·mth pyramidalterm neurons and produces dipoleterm outputs in units of ampere–meters (Am), allowing for direct comparisons between simulated and empiricalterm AEFsterm. Currents that flow into the cortex (e.g. N100mterm) correspond to current flow down the pyramidalterm neuron dendrites in HNNterm, as detailed further below.

Statistical Analysis In order to quantify the differences in the AEFterm waveform between left and right, as well as between contralateralterm and ipsilateralterm hemispheresterm, we focused on the analysis of the N100mterm response. Although we do not expect AEFterm differences to be expressed exclusively in the N100mterm, this component is typically the most prominent of the AEFterm components, and therefore lends itself to quantification. To this end, we conducted a total of four 2 × 2 repeated-measures ANOVAsterm·stat with factors Tone Presentationterm·var (contralateralterm/ipsilateralterm) and Hemisphereterm (right/left) to determine differences in N100m amplitudeterm·var, N100m latencyterm·var, P50m-N100m slopeterm·var, and N100m-P200m slopeterm·var. The N100mterm trough was defined per participant average as the absolute signal maximum during the time interval from 80 to 120 ms relative to the auditory stimulus onset. The P50m-N100m, and N100m-P200m slopesterm·var were determined by fitting straight lines to each participant’s average waveformterm·stat between 70–90 and 110–130 ms respectively.

We applied the same ANOVAsterm·mth to the modelterm·mth simulationterm·mth (see below), by running 10 simulationsterm·mth per modelled condition as a source of variabilityterm. While the reduced variabilityterm of simulationsterm·mth compared to humanterm MEGterm data means that these statistics are not directly comparable to those of the empiricalterm data, they nevertheless give insight into whether specific qualitative features, such as differences in slopesterm or latencies, are reproduced by the modelterm·mth.

Modelterm·mth The Humanterm Neocortical Neurosolver (HNNterm) We used the computational neural modeling software Humanterm Neocortical Neurosolver (HNNterm; Neymotin et al. 2020 ) to study the neural mechanisms generating the observed source localized, grand-average AEFterm waveforms. HNNterm is an open-source software (for tutorials and detailed descriptions of the modelterm·mth, please visit http://hnnterm.brown.edu), which simulates the primary electrical currents underlying macro-scale EEGterm and MEGterm data, based on their biophysical origin by modeling neocortical cellular and circuit-level activity. The modelterm·mth contains multi-compartment pyramidalterm neurons as well as single-compartment inhibitoryterm interneurons (basketterm cells) in supragranular layers (layer II/IIIterm) and infragranular layers (layer Vterm, see Fig. 1 ). The morphology of the pyramidalterm neurons was modelled based on cat visual cortex pyramidalterm neurons (Bush and Sejnowski 1993 ), and adjusted in accordance with anatomical findings in the humanterm brain (Geyer et al. 1997 ; Fischl and Dale 2000 ; Elston et al. 2001 ). The intracellular electrical current flow in the long and spatially aligned apical dendrites of the pyramidalterm neurons across the infragranular and supragranular layers are the main contributors to macroscale primary current dipolesterm that generate macroscale signals that can be observed on the scalp (Murakami and Okada 2006 ). The cells are connected with glutamatergic and GABAergic synapses, and each cell’s activity is simulated using Hodgkin-Huxley dynamics (Fig. 1 a). The primary current dipoleterm is calculated by summing the intracellular current flow across the network of pyramidalterm neurons (see red and green arrows in Fig. 1 c and d). Importantly, the current dipoleterm generated in HNNterm is expressed in units of ampere-meters (Am), the same units as those estimated by source localization methods of EEGterm and MEGterm data, enabling one to one comparison between modelterm·mth and empiricalterm data. For computational tractability, a reduced network size is simulated and a scalingterm factor is multiplied by the aggregate current dipoleterm to estimate the size of the network contributing to the recorded data, as detailed below. To assess the correspondence between averaged evoked responses and modelled dipolesterm, simulationsterm·mth are smoothed (30 ms Hamming window, see Table S2). There are many more sources of variabilityterm in the larger network contributing to the recorded humanterm data, such as spike variabilityterm across the network and individual subject differences. We assume high-frequency components created by this variabilityterm are more likely to average out in the larger network and multi-subject averages contributing to the recorded AEFsterm than in modelterm·mth simulationsterm·mth. We therefore apply smoothingterm·mth to modelterm·mth simulationsterm·mth to allow for a direct comparison between simulationterm·mth and data (see also Fig. S2). We also simulate N = 10 trials per example, each of which has some intrinsic variabilityterm, and average across trials.

Fig. 1.

HNNterm Modelterm·mth schematics. a Pyramidalterm neurons in layer II/IIIterm and V, and inhibitoryterm fast-spiking basketterm cells (empty circles). Excitatory and inhibitoryterm synaptic coupling is indicated by black lines with filled circles and bars respectively. Within-layer excitatory–excitatory and inhibitoryterm–inhibitoryterm connections are not shown, but exist for each cell type (see Table S2). b Visualization of the spatial alignment of a network of layer II/IIIterm and V pyramidalterm neurons. c , d The network is activated by proximalterm/feedforwardterm ( c ) and distalterm/feedbackterm ( d ) inputsterm which deliver trains of action potentials via canonicalterm·dom pathways The baseline network consists of 100 pyramidalterm neurons per layer (Fig. 1 b). A scaling parameterterm·var is then applied by multiplying the dipoleterm simulated by the baseline network by a constant. This constant represents the number of neurons contributing to a given recorded signal, assuming the signal represents the summed activity of a larger population of synchronous pyramidalterm neurons. Since a typical MEGterm/EEGterm response has a magnitude of 10-100nAm, and a single pyramidalterm neuron contributes approximately 0.2 pAm, between 50,000 and 500,000 cells are contributing to a typical macroscale signal, corresponding to a scalingterm factor of between 250 and 25,000 (Murakami and Okada 2006 ; Neymotin et al. 2020 ).

In order to simulate macroscale MEGterm/EEGterm signals, the network modelled in HNNterm is activated through exogenous excitatory synaptic inputsterm generated by predefined trains of action potentials that contact layer specific post-synaptic targets via two canonicalterm·dom inputterm pathways, which have previously been shown in animalterm modelsterm·mth (Rockland and Pandya 1979 ; Friedman and JonespersName 1980 ; Kulics and Cauller 1986 ; Cauller and Kulics 1991 ; Douglas and Martin 2004 ). We refer to the two types of inputsterm used here as proximalterm or feedforwardterm and distalterm or feedback inputsterm·def. Feedforwardterm drives represent signals that reach the cortex from the lemniscal thalamus to granular layers, which then propagate directly to supragranular and infragranular layers where they effectively target the proximalterm dendrites of the pyramidalterm neurons (Fig. 1 c). Feedbackterm drives represent inputsterm from non-lemniscal thalamus or cortico-cortical connections that target the distalterm dendrites in supragranular layers (Fig. 1 d). The level of detail included in HNNterm’s modelterm·mth enables simulationterm·mth of the macroscale current dipolesterm along with microscale circuit information, including layer specific spiking activity in individual cells. Detailed descriptions of the modelterm·mth underlying HNNterm can be found in the Supplementary Materials (S1: Supplemental Modelterm·mth Description).

Although the cortical columnterm modelled in HNNterm was originally generated to account for the anatomy and behavior of primary somatosensory cortex (JonespersName et al. 2007 , 2009 ), it is based on canonicalterm·dom features shared between different neocortical circuits and hence can be applied to interpret signals in other cortical areas (Neymotin et al. 2020 ). This is particularly the case for sensoryterm areas, as the relay of sensoryterm information has been shown to follow remarkably stereotypical patterns across modalities (Douglas and Martin 2004 ; Barbour and Callaway 2008 ).

Since HNNterm is a complex modelterm·mth with a large number of parametersterm, many of which interact, we cannot rule out that some alternative parameterterm sets could not reproduce similar results (see “ Discussion ” section). HNNterm was designed to be a hypothesisterm development and testing tool. The hypothesisterm testing here was motivated from our prior studies of sensoryterm evoked responses, which provided the baseline assumptions for the pattern of exogenous proximalterm and distal drivesterm·var underlying an evoked response, and prior literature on neural mechanisms underlying hemispheric dominanceterm (see “ Introduction ” section). While we cannot show that any given modelterm·mth provides a uniqueterm solution, it is important to note that the dipoleterm shape produced by HNNterm depends crucially on the underlying assumptions and we can examine the dipoleterm response under alternative hypotheses. To illustrate this, we show modelsterm·mth using alternative assumptions for two of our examples (see Fig. 5 and supplement S2, Fig. S1).

Fig. 5.

Alternative simulationsterm·mth for left hemisphereterm AEFsterm: left hemisphereterm AEFsterm (contralateralterm) are displayed in light blue, while modelterm·mth simulationsterm·mth are displayed in dark blue dotted lines. Individual simulationsterm·mth are displayed in gray. Inserts show the dipolesterm associated with layer II/IIIterm and layer Vterm separately. The fit provided by the alternative simulationsterm·mth was noticeably worse than that of the initial modelterm·mth, which only adjusted inputterm parametersterm (cf. Fig. 4b ). a Synaptic gains were decreased in all connections targeting inhibitoryterm interneurons. b Layer Vterm pyramidalterm calciumterm channel densities were decreased. All other parametersterm were equal to the modelterm·mth of the right hemisphereterm AEFterm (see Fig. 4d –f; Table 1 ) (Color figure online) Results Contralateralterm and Right Hemisphereterm AEFsterm Display More Prominent N100mterm Peaks Source-localized grand average AEFsterm and their descriptive statistics are displayed in Figs. 2 and 3 respectively. The differences between contralateralterm and ipsilateralterm, as well as left and right hemispheresterm were quantified for the N100mterm component using Tone Presentationterm·var × Hemisphereterm ANOVAsterm·mth. As expected based on prior studies (e.g. Peronnet et al. 1974 ; Wolpaw and Penry 1977 ; Pantev et al. 1998 ), we found that N100mterm magnitudes were larger in contralateralterm compared to ipsilateralterm ( F (1,9) = 7.72, p = 0.021, η 2 = 0.46) and right compared to left hemispheresterm ( F (1,9) = 5.71, p = 0.041, η 2 = 0.39; interaction p > 0.67). N100m latencyterm·var differences did not reach significance, but there was a trendterm·stat towards shorter latencies in contralateralterm compared to ipsilateralterm N100msterm ( F (1,9) = 5.08, p = 0.051, η 2 = 0.17; all other effects p > 0.21). The ascending and descending slopesterm of the N100mterm trough from the P50mterm to N100mterm, and N100mterm to P200mterm, were also greater in contralateralterm compared to ipsilateralterm AEFsterm ( F P50mterm−N100mterm (1,9) = 10.55, p P50mterm−N100mterm = 0.01, η P50mterm−N100mterm = 0.54; F N100mterm−P200mterm 2 (1,9) = 5.42, p N100mterm−P200mterm = 0.045, η N100mterm−P200mterm 2 = 0.38; all other effects p > 0.092; Fig. 3 ).

Fig. 3.

Means of empiricalterm and simulated AEFsterm for each quantified N100mterm characteristic and each condition. p values are displayed for each ANOVAterm·mth. Effects demonstrated in both empiricalterm and simulated AEFsterm are printed in bold. Statistics associated with simulated AEFsterm are printed in gray as the limited variabilityterm in simulationsterm·mth does not allow for direct comparison with tests performed on empiricalterm data. Error bars indicate standard error.term·stat * p < 0.05; *** p < 0.001. a N100m amplitudeterm·var, b N100m latencyterm·var, c P50-N100m slopeterm, d N100m-N200m slopeterm (Color figure online) A Sequence of Feedforward-Feedback-Feedforward Activation of the Cortical Circuit Reproduces AEFterm in HNNterm The tone evoked AEFsterm in Fig. 2 exhibit peak timings and polarities that are similar to source localized SEFsterm evoked by brief perceptual threshold level tactile stimuli observed in our prior studies (e.g. JonespersName et al. 2007 , 2009 ), where there is an initial small amplitudeterm positive peak (confirmed to correspond to current flow out of cortex), followed by a prominent negative peak (current flow into cortex), followed by a subsequent positive peak. Motivated by this consistency along with the homology of canonicalterm·dom cortical circuitry in sensoryterm areas built into HNNterm, we applied the inputterm sequence of the existing SEFterm modelterm·mth distributed with HNNterm as a starting point for our AEFterm simulationterm·mth (namely the ‘default.param’ parameterterm file). This sequence consisted of a proximalterm inputterm, followed by a distalterm inputterm, followed by a second subsequent proximalterm inputterm (JonespersName et al. 2007 ; Neymotin et al. 2020 ), and represents a canonicalterm·dom feedforwardterm, feedbackterm, feedforwardterm pattern of activation. This pattern of activation was able to account for rough features of the AEFterm waveform. To produce a closer fit to the AEFterm data we began by adjusting only the parametersterm defining the strength and timing of each inputterm (all inputterm parametersterm are displayed in Table 1 ), as well as the scale of the waveform. We first hand-tuned these parametersterm until there was visually close agreement between modelterm·mth and data, and subsequently used the automated parameterterm optimization feature in HNNterm to improve the modelterm·mth fit further. The optimization tool automatically adjusted parametersterm controlling inputterm timing and strengths to minimize the error (root mean squared error; RMSEterm·stat) between the simulated and empiricalterm dipoleterm waveforms (Neymotin et al. 2020 ). This was initially performed for the AEFterm recorded over the contralateralterm, right hemisphereterm. The resulting modelterm·mth simulationterm·mth (RMSEterm: 1.0) and its inputterm parametersterm are displayed in Fig. 4 d–f and Table 1 , respectively (see Supplementary Materials S2 for alternatives of this modelterm·mth).

Table 1.

Inputterm parametersterm used to simulate AEFsterm in the left and right hemisphereterm after contralateralterm tone presentationterm·var Parameterterm Left hemisphereterm modelterm·mth Right hemisphereterm Proximalterm Distalterm Proximalterm Proximalterm Distalterm Proximalterm Inputterm time 54.898 82.99 161.307 47.355 81.09 150.826 SD 5.401 13.208 19.844 2.963 12.26 11.061 Layer II/IIIterm Pyramidalterm AMPAterm 0.991 0.607 0.854 0.659 0.607 0.346 NMDA 1.714 0.242 0.067 0.535 0.243 0.027 Layer II/IIIterm Basketterm AMPAterm 0.997 0.624 0.758 0.997 0.523 0.995 NMDA 0.984 0.953 0.851 0.987 0.959 0.994 Layer Vterm Pyramidalterm AMPAterm 0.004 0.258 0.012 0.004 0.964 0.005 NMDA 0.01 0.157 0.004 0.009 0.158 0.006 Layer Vterm Basketterm AMPAterm 0.615 0.98 0.427 0.984 NMDA 0.062 0.902 0.037 0.998 The corresponding simulationsterm·mth are displayed in Fig. 4 . Parametersterm which were identified as most relevant to account for the difference between left and right hemispheresterm are printed in bold. AMPAterm and NMDA synaptic weights (i.e. maximal conductances) are displayed in units of µS, while mean and standard deviation of the inputterm times are displayed in ms. Note that distalterm inputsterm do not project onto layer Vterm basketterm cells (see Fig. 1 ). Parameterterm values rounded to three decimal places. All other parametersterm are displayed in Table S2 Fig. 4.

HNNterm simulationterm·mth of the AEFterm recorded in response to contralateralterm tone presentationterm·var over the right hemisphereterm (right panels) and over the left hemisphereterm (left panels): a , d Inputterm sequence: Inputterm spikes are sampled from a Gaussian distribution (mean and sd are defined by inputterm time and sd, see Table 1 ) on each trial. The resulting temporal profile of the spiking activity arriving into the network is displayed in red (proximalterm) and green (distalterm) histograms. A proximalterm inputterm drives the network, before a distalterm inputterm and a second proximalterm inputterm arrive (see Fig. 1 c, d for proximalterm/distalterm inputsterm). Corresponding inputterm parameterterm values are displayed in Table 1 . b , e Dipoleterm Simulationterm·mth: mean AEFterm modelterm·mth (dark blue) as well as 10 individual trial simulationsterm·mth (gray). The empiricalterm AEFterm (here: contralateralterm AEFterm) is displayed in light blue (cf. Fig. 2 ). Insert at the bottom right shows the dipolesterm of layer II/IIIterm and layer Vterm separately. Left inset in b shows modelterm·mth fit of manually fitted modelterm·mth (with no automatic optimization applied). All dipolesterm were smoothed using the default settings in HNNterm (30 ms Hamming window; see Table S2). An unsmoothed equivalent to panel e is displayed in Fig. S2. c , f Simulated spiking activity: spiking associated with the dipoleterm displayed in ( b ) (one example trial selected) (Color figure online) Briefly, the inputterm sequence affects the network in the following way: first, a feedforwardterm inputterm reaches the proximalterm dendrites, as well as inhibitoryterm interneurons of layer II/IIIterm and layer Vterm. This inputterm is thought of as feedforwardterm sensoryterm information reaching the auditory cortexterm via the lemniscal thalamus. It reaches the network around 47 ms after the onset of the auditory stimulus, strongly driving layer II/IIIterm cells (Table 1 ). This timing is broadly in line with previous findings, suggesting that AEFterm components occur after auditory signals reach the cortex at around 50 ms (Picton et al. 1974 ). This inputterm causes both pyramidalterm and basketterm cells in layer II/IIIterm to fire, with pyramidalterm firing quickly inhibited by the inhibitoryterm basketterm cells (Fig. 4 f). The spiking in the pyramidalterm neurons creates backpropagation of current flow up the pyramidalterm apical dendrites towards the surface of the cortex, leading to the P50mterm peak seen in the AEFterm (Fig. 4 e).

While the effects of the proximalterm inputterm are ongoing, a feedback inputterm·def arrives at the distalterm dendrites at around 81 ms (although with a wider temporal distribution, Fig. 4 d), strongly activating layer II/IIIterm basketterm and pyramidalterm cells and layer Vterm pyramidalterm cells and inducing the N100mterm peak. This drive is presumed to represent inputsterm from either cortico-cortical, or non-lemniscal thalamocortical connections. The distal driveterm·var induces several changes in the network dynamics, the net effect of which is a downward deflecting current. Excitatory synaptic inputsterm on distalterm apical dendrites of layer II/IIIterm and layer Vterm pyramidalterm neurons push the current flow down (Fig. 1 d). At the same time, somatic inhibition mediated by activation of basketterm cells in both layers (note, the layer Vterm basketsterm cells are driven by the layer II/IIIterm excitatory connections to layer Vterm, Fig. 1 a) pulls current further down the pyramidalterm neuron dendrites. In addition, activation of calciumterm dynamics, triggered by layer Vterm pyramidalterm excitation pulls current down towards infragranular layers. The distal driveterm·var also generates spiking activity in the layer Vterm pyramidalterm neurons, which leads to backpropagation, i.e. current flow in the opposite direction up the dendrites. However, the size of this effect is smaller and not visible in averaged and smoothed data (see Fig. S2 and “ Methods ” section) and the overall effect is current flow down the dendrites, leading to a large N100mterm trough (Fig. 4 e).

Lastly, as the spiking activity in the network would begin to relax back to baseline after the distal driveterm·var, a second proximalterm inputterm reaches the network at around 151 ms, driving primarily layer II/IIIterm and layer Vterm basketterm cells, and some pyramidalterm cells, causing continuous basketterm firing, with burst-like behavior occurring at around 200 ms (Fig. 4 f). This activation of postsynaptic excitatory synapses on the basal dendrites of the pyramidalterm neurons again pushes current flow towards the surface of the cortex. Additionally, the layer II/IIIterm basketterm cells inhibit the distalterm dendrites of the layer Vterm pyramidalterm neurons (Fig. 1 a), helping to pull current flow up the dendrites toward the cortical surface. Together these effects create the P200mterm peak in the AEFterm.

Dividing the net current dipoleterm into its layer specific components shows that the shape of the AEFterm simulationterm·mth is largely driven by responses in layer Vterm, while layer II/IIIterm responses contribute only to the earliest component (see inserts in Fig. 4 e, and Fig. S2b).

We also calculated the simulated firing ratesterm of each neuronal population by averaging the number of action potentials over trials and cells. We found the highest firing ratesterm in basketterm cells (layer Vterm = 13.97 spikes/s; layer II/IIIterm = 13.46 spikes/s), followed by layer Vterm pyramidalterm neurons (10.60 spikes/s), and low firing ratesterm in layer II/IIIterm pyramidalterm neurons (1.85 spikes/s, cf. Fig. 4 f). Although firing ratesterm were not considered as part of the modelterm·mth fitting procedures, these firing ratesterm appear to be within physiological ranges (Wallace and Palmer 2008 ; Atencio and Schreiner 2010 ), further supporting the model-derived predictionsterm.

HNNterm Predicts Smaller Left Hemisphereterm AEFterm N100mterm Peak can be Generated by Decreased Feedbackterm Drive With the AEFterm recorded from the right hemisphereterm (for contralateralterm tone presentationterm·var) simulated, we set out to modelterm·mth the AEFterm recorded over the left hemisphereterm. Since we did not have any clear hypotheses predicting the differences between left and right hemispheresterm based on prior literature, we used modelterm·mth derived knowledge of how the inputsterm to the network influence the timing and magnitude of the peaks to examine how differences in the right and left hemispheresterm could emerge. Using right AEFterm as a starting point, we began by tuning the inputterm parametersterm to fit the left AEFterm, using both hand-tuning and automatic optimization. The targeted differences we aimed to create were the significantly smaller N100m amplitudeterm·var (Figs. 2 c and 3 ). By hand tuning each inputterm parameterterm, we found that modulation of only one inputterm parameterterm was able to account fairly well for the qualitative differences in the right and left waveforms. Specifically, we found that by decreasing the AMPAterm mediated excitation of layer Vterm pyramidalterm cells by the distal driveterm·var (see Table 1 and Table S2), and keeping all other parametersterm fixed between hemispheresterm, we were able to reproduce the AEFterm waveform associated with the left hemisphereterm (Table 1 ). Since distalterm layer Vterm pyramidalterm excitation pushes currents down the dendrites, creating the N100mterm trough, decreasing the excitatory inputsterm to these cells leads to a smaller N100mterm magnitude as seen in left hemisphereterm AEFsterm. We refer to the modelterm·mth in which only this parameterterm was changed to account for left AEFsterm as the ‘Manually Fitted Modelterm·mth’, and used automated parameterterm optimization to further improve the modelterm·mth fit. While further small parametersterm changes were needed to optimize the modelterm·mth fit (Fig. 4 b), fitting the modelterm·mth with only this one parameterterm was sufficient to reproduce the targeted differences (see “Manually Fitted Modelterm·mth” in Fig. 4 b). The optimized modelterm·mth (Fig. 4 b, large panel) was used for all further analyses. The resulting dipoleterm simulationsterm·mth displayed the same N100mterm peak amplitudeterm difference between left and right hemispheresterm as the empiricalterm AEFterm ( p < 0.001).

While our initial testing of the hypothesisterm that strength of the exogenous distalterm inputsterm would be essential to defining the amplitudeterm of the AEFterm peaks confirmed that alteration in this parameterterm can reproduce the observed differences, we cannot rule out that alternative mechanisms could reproduce similar results. To address this issue, we also tested two alternative modelsterm·mth that could logically account for N100m amplitudeterm·var differences through alteration of local network features, rather than inputsterm into the network. Since simulationsterm·mth show that both somatic inhibition and layer Vterm calciumterm activity can lead to current flow down the dendrite (i.e. larger N100m amplitudesterm·var), we decreased the parameterterm values representing each one at a time to test if these changes could also lead to less current flow down the dendrite and therefore smaller N100m amplitudesterm·var as seen in left hemisphereterm AEFsterm. We found that by decreasing the parametersterm controlling the synaptic weights of all local network connections targeting inhibitoryterm cells, as shown in Fig. 1 a, by a factor of 10 from the right hemisphereterm AEFterm parametersterm, the modelterm·mth was indeed able to approximate the decreased N100m amplitudesterm·var of left hemisphereterm AEFsterm (Fig. 5 a). Specifically, we decreased five parametersterm, representing the conductance of the synaptic connections targeting layer Vterm basketterm cells from layer II/IIIterm and layer Vterm pyramidalterm cells, and layer Vterm basketterm cells, as well as connections targeting layer II/IIIterm basketterm cells from layer II/IIIterm pyramidalterm and basketterm cells. However, this manipulation also impacted the subsequent P200mterm peak, which was significantly larger when the local connections to the inhibitoryterm neurons were decreased due to increased pyramidalterm neuron firing (data not shown), and hence a larger RMSEterm between the modelterm·mth and recorded data were observed; RMSEterm = 7.84 compared to RMSEterm = 1.96 in Fig. 4 . Similarly, we found that the N100m amplitudeterm·var decrease could also be reproduced by a simulationterm·mth in which the calciumterm channel density in the soma and along the dendrites of the layer Vterm pyramidalterm neurons (Table S2), was decreased by approximately 94% from the right hemisphereterm AEFterm parametersterm (Fig. 5 b). In this case, the subsequent P200mterm amplitudeterm was smaller than recorded data due to decreased pyramidalterm neuron firing (data not shown), and hence the RMSEterm between the modelterm·mth and recorded data was again larger than before; RMSEterm = 6.57 compared to RMSEterm = 1.96. These results suggest that initial manipulations of the strength of the distalterm inputterm as in Fig. 4 ; Table 1 , was more effective at reproducing multiple features of the empiricalterm data and we therefore used the initial left AEFterm modelterm·mth for all following analyses.

Differences in the Feedforwardterm and Feedback Inputterm·def Latencies, and Network Size, Can Reproduce Contralateralterm/Ipsilateralterm AEFterm Differences in HNNterm After establishing mechanisms that can reproduce the contralateralterm AEFsterm in both the left and right hemisphereterm, we next tested the specific hypotheses based on prior literature that differences between contralateralterm and ipsilateralterm AEFsterm emerge from differences in the size of the active network (Rosenzweig 1951 ; Gross et al. 1967 ), and alterations in the timing of the exogenous inputterm to the network (Tunturi 1946 ; Kimura 1961 ). The differences that we targeted to reproduce in the modelterm·mth were the decreased N100m amplitudeterm·var, and smaller P50m-N100m and N100m-P200m slopesterm·var as observed in the empiricalterm data (Figs. 2 and 3 ). We specifically hypothesized fewer cells contributed to the ipsilateralterm compared to the contralateralterm AEFsterm, and thus we first hand-tuned the scaling parameterterm·var in the modelterm·mth that provides an estimate of the number of neurons contributing to the observed data (see Methods), decreasing from a value of 1500 in the contralateralterm AEFterm modelterm·mth to 1200 for the ipsilateralterm AEFterm simulationterm·mth. This change was able to account well for the decreased N100mterm magnitude in the ipsilateralterm versus contralateralterm AEFterm, and also improved the modelterm·mth fit to the ipsilateralterm AEFterm around the P50mterm and P200mterm components, compared to the contralateralterm simulationterm·mth. This change in scale also accounted for the decreased P50m-N100m and N100m-P200m slopesterm·var associated with ipsilateralterm AEFsterm (Fig. 3 ). As such, the modelterm·mth results predict that a smaller network with on the order of 240,000 pyramidalterm neurons (1200 × 200 PN per layer) generated the ipsilateralterm response compared to 300,000 (1500 × 200 PN per layer) for the contralateralterm response (Fig. 6 ). Since previous findings further suggest a difference in latencyterm between contralateralterm and ipsilateralterm AEFsterm (Pantev et al. 1986 , 1998 ), in a second step, we hand tuned the mean latencyterm of the inputsterm into the network and found that the modelterm·mth fit to the data was improved when all three inputsterm were delayed by 5 ms. Since the latencies of all inputsterm were adjusted equally, this manipulation did not change the shape of the dipoleterm, but shifted it along the time axis. Although latencyterm differences between contralateralterm and ipsilateralterm AEFsterm were not statistically significantterm in our data, N100mterm latencies were slightly but consistently larger in ipsilateralterm AEFsterm (Fig. 3 ). This trendterm was captured well by simulationsterm·mth incorporating this 5 ms shift. The resulting AEFterm simulationsterm·mth are displayed in Fig. 6 .

Fig. 6.

HNNterm simulationterm·mth of the contralateralterm dominanceterm effect for the right hemisphereterm (right panel) and the left hemisphereterm (left panel) AEFterm. Average simulationsterm·mth (based on 10 individual trial simulationsterm·mth) of contralateralterm AEFsterm (dark blue, cf. Fig. 4 ) and ipsilateralterm (dark red) AEFsterm. Ipsilateralterm responses were generated by decreasing the modelterm·mth scalingterm factor, representing the number of cells contributing to the signal, and increasing the inputterm latencies, as compared to the contralateralterm simulationsterm·mth. Empiricalterm AEFsterm are shown in light colors (cf. Fig. 2 ) to indicate modelterm·mth fit (Color figure online) Both scalingterm and time parametersterm were hand-tuned for the difference between contralateralterm and ipsilateralterm AEFsterm to account first for only the right hemisphereterm. The same modulations, i.e. decreasing the scale by 300 and increasing the inputterm times by 5 ms, were then applied to the modelterm·mth of the left hemisphereterm AEFterm without further fitting steps.

By following these two steps, the modelterm·mth provided a good fit to the ipsilateralterm AEFsterm of both hemispheresterm (RMSEterm right: 2.15, RMSEterm left: 2.19), displaying the same N100m amplitudeterm·var and slopeterm differences as the empiricalterm data ( p < 0.006, Fig. 3 ), without further fitting or optimization steps, confirming the hypothesisterm that differences in network size and timing of the exogenous inputsterm can account for waveform shape differences between contralateralterm and ipsilateralterm AEFsterm.

Discussion Auditory evoked fields are a commonly used measurement in humanterm studies of healthy cognitive processing and as biomarkersterm of neuropathology, yet the biophysical mechanisms underlying these scalp-recorded responses are not yet fully understood. While humanterm AEFterm studies and laminar recordings of the auditory cortexterm in animalterm modelsterm·mth often proceed in relative isolation, here, we connected the information of the circuitry architecture, provided by animalterm modelsterm·mth, with humanterm MEGterm recordings using biophysically principled computational neural modeling. We used the Humanterm Neocortical Neurosolver (HNNterm), a software whose foundation is a modelterm·mth of a canonicalterm·dom neocortical circuit, which allows users to define inputsterm to activate the network and simulates the primary electrical currents underlying scalp-recorded fields or potentials (JonespersName et al. 2007 ; Neymotin et al. 2020 ), to develop and test predictionsterm on the neural mechanisms underlying two well-established characteristics in the AEFterm response, namely the lateralizationterm·dom effect in which responses recorded over the right hemisphereterm display stronger magnitudes in response to simple auditory stimuli than those recorded over the left hemisphereterm (Hine and Debener 2007 ; Howard and Poeppel 2009 ; Kimura 2011 ), as well as the phenomenon that AEFsterm in response to contralaterally presented tones show larger, and often faster, responses than ipsilaterally presented tones (Tunturi 1946 ; Rosenzweig 1951 ; Kimura 1961 ; Pantev et al. 1998 ).

We were able to modelterm·mth the AEFterm waveform elicited by a brief monaural contralateralterm tone presentationterm·var using the same inputterm sequence that has previously been shown to simulate somatosensory evoked responses (JonespersName et al. 2009 ), with parametersterm tuned to match the AEFterm data. According to this sequence, auditory responses can be accounted for by an initial excitatory thalamocortical feedforwardterm drive to layer II/IIIterm and V, via layer IVterm inducing the P50mterm, followed by a cortico-cortical or non-lemniscal thalamicterm feedback inputterm·def to supragranular layers inducing the N100mterm, and a subsequent second feedforwardterm inputterm inducing the P200mterm. This finding is in line with evidence suggesting that all sensoryterm areas share the same basic structure as well as many defining features, such as the basic laminar organization and activation patterns (Douglas and Martin 2004 ). Specifically, the initial activation of cortical circuits by the thalamus, which follows a layer IVterm, layer II/IIIterm, layer Vterm information processing sequence, a simplified version of which is recreated here and generates the P50mterm using HNNterm, has been shown to be similar across sensoryterm modalities (Douglas and Martin 2004 ; Barbour and Callaway 2008 ; Atencio and Schreiner 2010 ).

It is important to note that the connectivity within the auditory cortexterm is not as well understood as in other sensoryterm regions, and that focusing solely on similarities between cortical regions is likely an oversimplified description of auditory cortical activation. There are known differences between auditory cortexterm and other sensoryterm regions, as for example, laminar differences are less prominent than in visual or somatosensory regions, and there is some evidence to suggest that the functional role of auditory layer IVterm may differ from those in other sensoryterm regions (Linden and Schreiner 2003 ; Barbour and Callaway 2008 ; Ji et al. 2016 ; Oviedo 2017 ). However, we show that a manually fitted neural modelterm·mth of a cortical columnterm with canonicalterm·dom layer specific inputterm patterns is able to account for scalp-recorded auditory responses, supporting the notion that the basic structure and activation patterns are similar across sensoryterm regions, and providing means to approach the biophysical basis of well-known characteristics and effects reported in humanterm AEFterm research.

Further, we found that simulationsterm·mth associated with AEFsterm recorded over left and right hemispheresterm were remarkably similar, not only in the inputterm sequence, but also in firing ratesterm and dipoleterm waveforms (compare Fig. 4 a-c and d-f). Interestingly, the only parameterterm necessary to account for qualitative differences in the waveform shape was the strength of the distalterm inputterm into layer Vterm pyramidalterm neurons. In other words, a 100 ms distalterm feedback inputterm·def drives layer Vterm pyramidalterm neurons more strongly in right than in left hemisphereterm AEFsterm (note that since parameterterm optimization was performed all inputterm parametersterm differed somewhat between left and right hemisphereterm modelsterm·mth). These findings are in general agreement with research demonstrating functional asymmetries between left and right hemispheresterm. For example, right hemisphereterm dominanceterm has been shown especially in processing of pure tones, although the hemispheric balance seems to depend on the nature of the stimulus, as well as context (Peronnet et al. 1974 ; Wolpaw and Penry 1977 ; Schönwiesner et al. 2007 ; Howard and Poeppel 2009 ). It could be speculated that the relative specialization of the right hemisphereterm to process pure tones is associated with optimized connectivity, leading to stronger inputsterm. It would be interesting, in the future, to explore whether the diminished rightward asymmetry, or leftward asymmetry to speech sounds could be explained by similar underlying dynamics. However, the functional lateralizationterm·dom of auditory processing continues to be debated (e.g. Boemio et al. 2005 ), and its relation to biophysical mechanisms is unknown. Additionally, our interpretations are model-derived predictionsterm, since we did not have clear pre-defined hypotheses regarding the mechanisms underlying the differences in left and right AEFsterm.

HNNterm is designed to be a hypothesisterm development and testing tool, however in its current form not all possible hypotheses can be explored. For example, macroanatomical differences not captured in HNNterm between left and right hemispheresterm could contribute to, if not fully account for, waveform differences observed in AEFsterm (Galaburda et al. 1978 ; Rademacher et al. 1993 ; Penhune et al. 1996 ; Howard and Poeppel 2009 ). In fact, Shaw et al. ( 2013 ) argued that what may appear as a functional asymmetry in auditory processing is, at least partially, caused by structuralterm asymmetries, as increased cortical foldingterm in the left auditory regions results in increased MEGterm/EEGterm signal cancelation. Here, we do not account for these anatomical differences since HNNterm has been developed to account for the microanatomical principles underlying electrophysiological recordings, and does not include cortical foldingterm in its parametersterm. As such, we cannot rule out the possibility that the parameterterm differences between left and right hemisphereterm AEFsterm we found in our modelterm·mth are simply compensating for the effect of structuralterm differences, and do not accurately reflect underlying functional mechanisms. However, we found that alternative modelterm·mth adjustments which manipulated local network features, instead of external inputterm characteristics to account for hemispheric AEFterm differences, led to a poor modelterm·mth fit, providing some support for our modelterm·mth. Ultimately, the predictionsterm made with HNNterm can guide targeted experiments to validate or negate modelterm·mth results, and new data informs HNNterm expansion and improvement.

We also simulated the contralateralterm dominanceterm effect, i.e. the larger N100m amplitudeterm·var in the AEFterm waveform associated with contralateralterm compared to ipsilateralterm tone presentationterm·var. In the AEFterm literature, the mechanistic differences proposed to underlie the finding that increased AEFterm amplitudesterm are recorded in hemispheresterm contralateralterm to the auditory stimulus, are that auditory pathways that cross over are stronger and/or more numerous than ipsilateralterm ones (Tunturi 1946 ; Rosenzweig 1951 ; Kimura 1961 ), and that the cortical region activated by the auditory stimulus is larger in the contralateralterm hemispheresterm (Rosenzweig 1951 ; Gross et al. 1967 ). We tested these predictionsterm using HNNterm. We expected the strength of a pathway to correlate primarily with the speed of the relay of information, which corresponds to the inputterm time parametersterm in HNNterm. The notion of differences in the size of the cortical area can be accounted for by HNNterm’s scaling parameterterm·var, which gives an estimate of the number of cells contributing to a given dipoleterm. We were able to recreate the differences between contralateralterm and ipsilateralterm AEFsterm, solely by changing the time and scaling parametersterm·var, supporting the hypotheses that these manipulations based on animalterm research can indeed account for the differences in the humanterm signal.

These simple manipulations led to greatly improved modelterm·mth fits to ipsilateralterm data, in both the left and the right hemisphereterm. The ipsilateralterm modelterm·mth fits the ipsilateralterm AEFterm less well than the contralateralterm modelterm·mth fits its corresponding data which is noticeable particularly in the P200mterm peak fit in the right hemisphereterm. However, note that the contralateralterm modelsterm·mth were established by hand-tuning and then optimizing all 28 inputterm parametersterm (Table 1 ), while the ipsilateralterm modelterm·mth was generated by adjusting only four parametersterm, three of which by a fixed value. Importantly, the four parametersterm, as well as the directions in which to adjust them (i.e. increasing the time of each of three inputsterm by a fixed amount and decreasing the network scale), were chosen based on predictionsterm from the literature, not on the patterns identified in our data, and no further steps were taken to fit the modelterm·mth simulationsterm·mth to the empiricalterm AEFsterm. In view of this procedure, we consider the ipsilateralterm modelsterm·mth to reproduce the corresponding AEFsterm remarkably well.

We found that, compared to ipsilateralterm AEFsterm, contralateralterm AEFsterm can be simulated by increasing the scalingterm factor by 20% and decreasing the time of each inputterm by 5 ms. The difference of 20% found here is also approximately in line with predictionsterm made by Rosenzweig ( 1951 ) who suggested that the size of ipsilateralterm presentations is around 25% smaller than contralateralterm representations. Although in our data, we were not able to demonstrate clear latencyterm differences between contralateralterm and ipsilateralterm AEFsterm, we found that a difference of 5 ms provided improved modelterm·mth fits. This small difference is in line with previous studies which have reported contralateralterm responses to occur between 4 and 14 ms faster than ipsilateralterm ones (Mononen and Seitz 1977 ; Mäkelä et al. 1994 ; Pantev et al. 1998 ). Since this is a fairly small difference in latencyterm, it may overlap with individually differing factors that influence AEFterm measures at these latencies, and may not be easily demonstrated. We speculate that this may explain why other studies, like our own, have not been able to confirm latencyterm differences between ipsilateralterm and contralateralterm AEFsterm (Yoshiura et al. 1994 ). Overall, we found that ipsilateralterm AEFsterm, compared to contralateralterm ones, can be simulated by delaying the activation of the cortical network and decreasing the size of the network, supporting previous hypotheses predicting that contralateralterm pathways are faster and activate larger regions of the cortex.

It is important to note that, since HNNterm is a large-scale modelterm·mth with thousands of differential equations and parametersterm, we cannot claim that any given waveform has a uniqueterm way in which to modelterm·mth it. With large numbers of parametersterm, many of which interact and trade off, HNNterm is primarily a hypothesisterm testing tool as it is not feasible to test the entire parameter spaceterm. However, a large proportion of the modelterm·mth, including cell morphologies and physiologies, and local connectivity patterns were fixed based on a large body of animalterm modelsterm·mth, introducing biologically realistic constraints (JonespersName et al. 2007 ; Neymotin et al. 2020 ). Further, we drastically limited the number of parametersterm free to vary, by making the assumption that many basic features of the network are preserved across sensoryterm areas, which allowed us to fix the vast majority of parametersterm to values previously established for somatosensory response modelsterm·mth (JonespersName et al. 2007 , 2009 ). Based on the same assumption, we also fixed the inputterm sequence which activates the network, but not the timing and strength of the inputsterm. This resulted in a modelterm·mth with 28 inputterm parametersterm and one scaling parameterterm·var, while all other parametersterm were fixed (seven additional parametersterm, representing calciumterm activity and inhibitoryterm connection strength, were adjusted to test alternative modelsterm·mth, but were not used in the final modelsterm·mth). In the case of interpreting the difference between left and right hemispheresterm, we attempted to narrow down the parameter spaceterm of interest by identifying which changes made the largest contribution to the difference in waveform shape. However, we cannot rule out the possibility that other parameterterm values could account for the finding in a different way. In the case of contralateralterm compared to ipsilateralterm AEFsterm on the other hand, we had clear hypotheses predicting not only the parametersterm of interest, but also the direction of their adjustment, enabling much clearer conclusions.

Additionally, it is important to note that HNNterm is designed to bridge the gap between humanterm macro-scale data and circuit-level findings from animalterm modelsterm·mth. Many of the circuit features built into HNNterm, including those regulating inputterm drives into sensoryterm cortices, were based on animalterm studies in rodents and non-human primates and have not been directly demonstrated in humansterm (see Neymotin et al. 2020 ; JonespersName et al. 2007 , 2009 ). Although many basic structures and mechanisms seem to be shared across cortical regions and speciesterm (Douglas and Martin 2004 ), and HNNterm has been able to account for a variety of humanterm signals (JonespersName et al. 2009 ; Sliva et al. 2018 ; Neymotin et al. 2020 ), including those with remarkable homology between humansterm, rodents and non-human primates (Sherman et al. 2016 ; Shin et al. 2017 ) we cannot rule out the possibility that some cell- or circuit-level dynamics which are specific to humansterm are misrepresented in HNNterm.

Overall, we argue that biophysical modelling of humanterm macro-scale brain signals, using modelsterm·mth such as HNNterm, is an important step to bridge the gap between humanterm cognitive neuroscience, and cell- and circuit-level insights provided by animalterm modelsterm·mth. While we aimed to characterize commonly observed phenomena on a biophysical basis, future research may apply similar approaches to study similar scalp potentials, providing insights into mechanistic differences between developmental stages or clinical populations. Future research may also build on the current capabilities of HNNterm and incorporate sensitivity analyses to allow for an improved interpretation of parameterterm values.

In summary, we found that HNNterm, a modelterm·mth of a canonicalterm·dom neocortical circuit, can be used to simulate humanterm AEFsterm by activating the network using an inputterm sequence similar to one used to modelterm·mth other sensoryterm responses, supporting the notion that the basic structure and activation patterns are preserved across sensoryterm regions. We show that AEFterm differences between left and right, as well as contralateralterm and ipsilateralterm hemispheresterm, can be simulated by adjusting a small number of parametersterm representing network scale and inputterm characteristics. Our simulationsterm·mth establish a connection between scalp-recorded correlates of auditory processing and network-level insights from animalterm modelsterm·mth, providing a first step to understand the mechanisms underlying this cognitively and clinically relevant biomarkerterm.

Supplementary Information Below is the link to the electronic supplementary material.

Supplementary Material 1 (DOCX 195 KB) (194.2KB, docx) Author Contributions SRJ and TP devised the project. TP provided the data. CK carried out the simulationsterm·mth. SRJ supervised the modelling work. SRJ, TP and CK contributed to the interpretation of the results. CK drafted the manuscript, which all authors critically revised and approved.

Funding This study was supported by National Institutes of Health (Grant Nos. NIBIB RO1 EB022889 , NIMH RO1 MH106174 ).

Data Availability Averaged data, simulationsterm·mth, and parameterterm files are available on http://github.com/kohl-carmen/HNN-AEF .

Code Availability HNNterm is available at http://hnnterm.brown.edu and http://github.com/jonescompneurolab/hnnterm .

Declarations Conflict of interest The authors declare that no competing interests exist.

Footnotes This is one of several papers published together in Brain Topography on the “Special Issue: Computational Modeling and M/EEGterm”.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.