NeuroImage 207 (2020) 116432

Contents lists available at ScienceDirect

NeuroImage
journal homepage: www.elsevier.com/locate/neuroimage

Mapping adaptation, deviance detection, and prediction error in

auditory processing
Christina Hofmann-Shen a, b, *, Bob O. Vogel a, Maximillian Kaffes b, Armin Rudolph a,
Elliot C. Brown c, Cumhur Tas d, Martin Brüne e, Andres H. Neuhaus a, f
a
Department of Psychiatry, Charit
e University Medicine, Berlin, Germany
b
Department of Neurology, Charit
e University Medicine, Berlin, Germany
c
Department of Decision Neuroscience and Nutrition, German Institute of Human Nutrition, Potsdam, Rehbruecke, Germany
d
Department of Psychology, Faculty of Humanities and Social Sciences, Uskudar University, Istanbul, Turkey
e
Department of Psychiatry, Ruhr University, Bochum, Germany
f
Department of Psychiatry, Medical School, Brandenburg, Neuruppin, Germany

A R T I C L E I N F O A B S T R A C T

Keywords: Various studies have suggested that auditory deviance detection is organized in a hierarchical manner with
Mismatch negativity ascending levels of complexity. Event-related potentials (ERP) are considered to reflect different cortical pro-
N1 cessing stages. In the current electroencephalographic study, we employed an auditory sequence oddball para-
Repetition suppression
digm to investigate different levels of cortical auditory processing and the contribution of neuronal habituation
Prediction error
Deviance detection
and prediction error mechanism to N1 and Mismatch Negativity (MMN). Our findings suggest that N1 reflects a
lower cortical process primarily involved in the encoding of simple physical features and is thus mainly modu-
lated by neuronal attenuation and not complex top-down mechanisms. By analyzing within-sequence signal
differences, we divided the MMN into distinct subcomponents reflecting different hierachical levels of auditory
processing. We determined a “first-order” MMN that reflects the processing of simple deviant features (such as
frequency) and “higher-order” MMNs that occur at regularity violation of complex patterns or unexpected inputs
that do not allow further predictions. In our source localization analysis, both the primary auditory cortex and left
IFG were primarily involved in the detection of simple, physically deviant features, while the right IFG was
associated with the processing of novel, unexpected auditory inputs and the ACC with regularity violation of
known patterns. Summarizing, our results might contribute to a better understanding of the different complexities
of neuronal habituation and prediction error mechanisms at different levels of cortical auditory processing.

1. Introduction 2016). Previous studies have shown early deviant-related event related
potentials (ERP) elicited by simple regularity violations that are located
The predictive coding theory is considered as a unifying theory of in lower subcortical regions and occur long before a cortical ERP is
cortical processing (Friston, 2005). The predictive coding framework generated (Grimm et al., 2011; Recasens et al., 2014). However, deviants
assumes a hierachical organization of the brain in which predictions of complex regularities, e.g. in terms of tone alterations, lead to gener-
generated by higher cortical areas are constantly compared to bottom-up ation of cortical ERPs but failed to elicit early subcortical ERP responses
sensory inputs. Thus, sensory processing is characterized by deviance (Althen et al., 2013; Cornella et al., 2012).
detection and generation of prediction error throughout all hierarchical It has been suggested that ERPs, interpreted in the light of the pre-
levels. This view is supported by previous studies investigating the sen- dictive coding theory, reflect cortical processing stages of prediction
sory processing along the auditory pathway (Carbajal and Malmierca, error detection and error minimization (Rentzsch et al., 2015). The most
2018; Parras et al., 2017). It has been suggested that auditory deviance well-studied ERPs considered as correlates of cortical prediction are
detection is organized in a hierarchical manner with ascending levels of mismatch negativity (MMN) and stimulus-specific adaptation, or repe-
complexity along the hierarchy (Escera et al., 2014; Aghamolaei et al., tition suppression (RS). RS describes the reduction of neuronal activity

* Corresponding author. Department of Neurology, Campus Benjamin Franklin Charit
e University Medicine, Hindenburgdamm 30, Berlin, Germany.
E-mail address: [email protected] (C. Hofmann-Shen).

https://doi.org/10.1016/j.neuroimage.2019.116432
Received 4 July 2019; Received in revised form 13 November 2019; Accepted 1 December 2019
Available online 3 December 2019
1053-8119/© 2019 Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
C. Hofmann-Shen et al. NeuroImage 207 (2020) 116432

when a stimulus is repeatedly presented. It is commonly accepted that RS study. Four participants had to be excluded from the study because of
represents an lower cortical effect caused by repeated sensory stimula- technical artifacts. The remaining 16 participants (12 males, 4 females)
tion leading to changes in the responsivity and adaptation of the involved had a mean age of 31.3
6.6 years. Histories of any psychiatric or
neuronal population (Grill-Spector et al., 2006). Other studies, however, neurological disorders (including family history) led to exclusion from
suggest that RS is generated due to of top-down neuronal mechanisms the study. All participants had normal or corrected-to-normal vision,
(Summerfield et al., 2008; Todorovic et al., 2011), since RS has been were right-handed (laterality index 82.9
17.5; Edinburgh Handedness
described to be significantly larger for expected than unexpected repe- Inventory; Oldfield, 1971), and were of normal intelligence, as estimated
titions (Todorovic et al., 2011). MMN is an ERP component that occurs by a vocabulary (IQ ¼ 110.6
12.48; Lehrl, 1999) and a non-verbal
when a stimulus qualitatively deviates from a frequently presented intelligence test (IQ ¼ 114.31
14.08; Horn, 1983). The study proto-
stimulus and has been described in the auditory (N€a€at€anen et al., 2007), col was approved by the ethics committee of the Charit
e University
visual (Neuhaus et al., 2013; Stefanics et al., 2012; Vogel et al., 2015), or Medicine Berlin and was conducted in accordance with the Declaration
somatosensory modality (Kekoni et al., 1997). MMN is thought to reflect of Helsinki and its amendments. All subjects gave written informed
a comparison mechanism based on sensory memory by some authors consent before participating and were reimbursed.
(e.g. N€a€
at€
anen et al., 2005) or is understood as the discrepancy between
higher cortical predictions and sensory information resulting in predic- 2.2. Procedure and paradigm
tion error by others (e.g. Friston, 2005; Lecaignard et al., 2015;
Wacongne et al., 2012). The auditory MMN is computed as the difference The experiment was carried out in a windowless, dimly lit, electrically
wave between the ERP response to a deviant stimulus and that to a shielded, and sound-attenuated room. Participants were asked to take a
standard stimulus that occurs about 150–250 ms after stimulus onset and seat in a comfortable chair in front of a 22” widescreen monitor and to
is mainly generated in both supratemporal cortices and predominantly visually fixate the monitor. Standardized instructions for the experi-
right frontal cortex (N€ a€
at€anen et al., 2007). However, a recent study by mental task were given by the experimenter verbally and on the screen.
MacLean et al. (2015) showed that cortical generators of auditory MMN Participants were then instructed to watch a silent movie during the
vary with paradigm. While temporal cortex activity seems to be more experiment. During the whole experimental session, subjects were visu-
predominant in monotonic oddball paradigms, frontal cortex activity ally monitored by the experimenter through a window from a neigh-
seems to be more prevalent in more complex auditory paradigms. boring room.
Altogether, the cortical auditory deviance detection mechanism re- Stimulus pairs were presented binaurally via headphones at a 60 dB
mains to be further studied with respect to the underlying mechanisms of sensation level. All auditory stimuli consisted of sinuoidal tones with 50
neuronal adaptation and prediction error. Further, the generation of ms duration (incl. 5 ms rise and 5 ms fall) and frequencies of either 800
MMN from varying cortical sources in dependance of deviance Hz (tone A) or 1,600 Hz (tone B). The standard sequence consisted of the
complexity needs to be addressed. stimulus pair A-B, presented with a probablity of p ¼ .7; three deviant
To account for these aspects, in the current electroencephalographic sequences, A-A, B–B, and B-A were presented with a probability of p ¼ .1
(EEG) study, we employed an auditory sequence oddball paradigm, each. The inter-stimulus interval was 450 ms; the inter-trial interval was
similar to the sound pair paradigm used by Wacongne et al. (2012). This jittered between 4,950 ms and 5,450 ms. After a learning phase with
paradigm, consisting of a standard sequence A-B and deviant sequences presentation of 40 standard sequences, 400 experimental stimulus pairs,
A-A, B–B, B-A, allowed us to investigate cortical processing differences i.e. 280 standard and 3*40 deviant auditory sequences, were presented
following repetitive versus non-repetitive single tones as well as cortical (see also Fig. 1a and b). Four blocks with different pseudo-randomized
mechanisms elicited by expected versus unexpected auditory patterns, trial orders were presented with a short break after each block.
i.e. tone pairs. We hypothesized that detection of single stimulus repe-
tition and detection of pattern violation are represented on different 2.3. EEG acquisition and analysis
levels of the cortical auditory processing hierarchy, while contributing to
the same ‘omnibus MMN’ component. Consequently, we hypothesized EEG was recorded with a 64-channel DC amplifier (Advanced Neuro
that the auditory MMN can be dissected into distinct sub-processes by Technology, Enschede, The Netherlands) with a sampling rate of 512 Hz
analyzing within-sequence signal differences. Sole differences between and 64 sintered Ag–AgCl electrodes mounted on an elastic cap according
second and preceding first stimuli are primarily geared towards cortical to the extended International 10/20 System. The ground electrode was
comparisons of physical stimulus properties between tone 1 and tone 2 (B placed on the forehead at position AFz. Electrode Cz served as internal
minus A for the standard; A minus A, B minus B, B minus A, and A minus reference for the online recording. Electrode impedances were kept
B for the deviant sequences), i.e. first-order MMN. Subsequently sub- below 5 kΩ.
tracting the first-order standard sequence MMN from first-order deviant Brain Vision Analyzer 2.03 (Brain Products, Munich, Germany) was
sequence MMN waveforms (calculated as ([A minus A] minus [B minus used for offline EEG analysis. EEG was digitally filtered at 0.5 Hz high-
A]); ([B minus B] minus [B minus A]); and ([A minus B] minus [B minus pass and 20 Hz low-pass with 24 dB/octave each, re-referenced to
A])) should largely remove low-level processing of physical stimulus common average, and down-sampled to 500 Hz. Ocular artifacts were
properties and thereby expose underlying processing of more complex corrected by using an independent component analysis approach (Jung
regularity violations, i.e higher-order MMN. Given an auditory sensory et al., 2000), which involved eliminating an average of 7.35
1.58
memory duration of approximately 10 s (N€a€at€anen and Escera, 2000), we artifact-contaminated components across participants. Remaining arti-
chose an intermediate inter-trial interval of about 5 s that allows for facts (80 μV at any electrode) were marked for later removal of
eliciting an MMN component while concurrently also minimizing po- contaminated epochs. The EEG was then segmented according to the
tential adaptation effects of preceding trials on directly following first experimental conditions. After baseline correction from 100 ms to 0 ms,
stimuli (e.g. A-A followed by A-B). Source localization was applied to ERPs were separately averaged for stimuli, conditions, and individuals;
estimate the neuronal generators of first-order and higher-order MMN this step also involved removal of segments previously tagged as
components at different stages of the cortical prediction hierarchy. artifact-contaminated; on average, 0.66
1.77% of all segments were
removed in this step. Averaged ERP segments consisted of periods from
2. Methods 100 ms before to 400 ms after stimulus onset. The N1 component was
determined as the first maximum negative peak of a butterfly plot
2.1. Participants averaged across all first stimuli with a latency range from 50 to 150 ms
after stimulus onset. A corresponding topographical map led to pooling
Twenty healthy participants (15 males, 5 females) participated in this the N1 signal across electrodes FC1, FCz, FC2, F1, Fz, and F2 (see also

2
C. Hofmann-Shen et al. NeuroImage 207 (2020) 116432

Fig. 1. Paradigm structure, paramatrization, and principal hypotheses. (a) The experiment consisted of repeated presentations of stimulus pairs. All auditory stimuli
were sinuoidal tones with 50 ms duration and frequencies of either 800 Hz (tone A, gray) or 1,600 Hz (tone B, black). Inter-stimulus intervals were 450 ms, inter-trial
intervals varied from 4,950 ms to 5,450 ms. (b) After a learning phase with 40 presentations of the standard A-B sequence, the experimental phase consisted of the
standard sequence A-B that was presented with a probablity of p ¼ .7 and three deviant sequences A-A, B–B, and B-A that were presented with a probability of p ¼ .1
each. (c) To compute within-sequence contrasts (i.e. first-order MMN), ERP waveforms to first stimuli were subtracted from ERPs to the subsequent second stimuli. (d)
In order to eliminate cortical signals elicited by physical stimulus differences within sequences, between-sequence contrasts (i.e. higher-order MMN) were computed
by subtracting the first-order MMN elicited by the standard sequence (A–B) from the first-order MMN components elicited by the different deviant stimulus sequences.
(e) Detection of physical stimulus deviance was hypothesized to occur in sequences consisting of different tones, i.e. A-B and B-A. (f) Violations of established
predictions were hypothesized to occur in sequence A-A (minus A-B), but not in sequences B-A (minus A-B) or B–B (minus A-B) as sequences B-A and B–B were
equiprobable.

3
C. Hofmann-Shen et al. NeuroImage 207 (2020) 116432

Fig. 2. (a) Butterfly plot of all first stimuli with an N1 component around 100 ms after stimulus onset with maximal negativities over frontal and fronto-central
electrodes F1, Fz, F2, FC1, FCz, and FC2. (b) Butterfly plot of all second minus corresponding first stimuli with an MMN component around 200 ms after stimulus
onset with maximal negativities over fronto-central and central electrodes FC1, FCz, FC2, C1, Cz, and C2. (c) Grand averaged N1 responses stratified by sequences and
stimuli. Between 150 and 250 ms, significant amplitude negativations to second stimuli (thin lines) as compared to their preceding first stimuli (thick lines) suggest a
clear MMN component. Between 50 and 150 ms, within-sequence amplitude reductions in response to second stimuli are consistent with adaptive processes that are
larger in sequences with identical stimuli (A-A and B–B) than in sequences with different stimuli (A-B and B-A). (d) Adaptive pattern of the N1 response stratified for
auditory sequences.

Fig. 2a,b,c). We then assessed negative differences between first and the MMN waveform elicited by the standard stimulus sequence (A-B) was
second stimuli as mean amplitudes to statistically confirm the presence of subtracted from waveforms elicited by deviant stimulus sequences (A-A,
significant negativities at a typical MMN latency of 150–250 ms for every B-A, B–B) to isolate the hypothesized higher-order MMN (see also Fig. 1c
condition. Conversely, this latency range was again confirmed by visually and d). First-order and higher-order MMN were then submitted to source
inspecting a butterfly plot of within-sequence difference waves averaged analysis to estimate differential cortical contributions to MMN signals at
across conditions. Based on the corresponding topographical map, MMN different orders.
signals were pooled across C1, Cz, C2, FC1, FCz, and FC2.
First-order MMN waveforms were computed by subtracting ERPs to 2.4. Source localization analysis
first stimuli from ERPs to second stimuli and by assessing the local
negative maximum between 150 and 250 ms. Next, in order to eliminate The cortical distribution of electrical activity recorded from scalp
cortical signals elicited by physical within-sequence stimulus differences, electrodes in response to averaged adapters was computed with

4
C. Hofmann-Shen et al. NeuroImage 207 (2020) 116432

Standardized Low Resolution Brain Electromagnetic Tomography tone in the standard A-B sequence (3.02
0.90 μV) compared with the
(sLORETA; Pascual-Marqui, 2002). sLORETA employs a three-shell head A-A (2.32
1.39 μV; t(15) ¼ 2.82; p ¼ .013) and B–B repetition
model registered to the Talairach atlas of the human brain (Talairach and sequences (2.20
1.41 μV; t(15) ¼ 3.03; p < .01). Similarly, N1
Tournoux, 1998) with a three-dimensional solution space restricted to suppression to the second tone was smaller in the B-A sequence (2.84

the cortical gray matter and the hippocampus, thus compriseing a total of 1.30 μV) than in the A-A (t(15) ¼ 3.03; p < .01) and the BB sequences
6,239 voxels at 5 mm spatial resolution. Without a priori assumptions on (t(15) ¼ 2.07; p ¼ .06), although the latter only showed a trend towards
number and location of active sources, this solution to the inverse significance. N1 results thus largely conform to an adaptive pattern as a
problem computes the standardized current density at each voxel as the function of physical attributes’ similarity between successive auditory
weighted sum of the scalp electric potentials. The time frames of interest stimuli (see Fig. 2d).
for statistical imaging were selected on a data-driven base corresponding
to the ERP analysis, i.e. using identical time frames for MMN components 3.2. MMN responses
and their respective baselines across conditions.
All source imaging results were based on difference ERPs to allow for We confirmed the presence of a significant negativity at 150–250 ms
directly testing MMN against baseline instead of statistical double dif- after stimulus onset by subjecting corresponding mean amplitudes to a
ferences. To avoid double dipping (Kriegeskorte et al., 2009), we used 2*4 ANOVA that showed a significant main effect of stimulus order (F ¼
the within-sequence MMN in response to the auditory standard sequence 36.719; p < .001; partial η2 ¼ 0.710), but not condition. This main effect
A-B to define regions of interest (ROI) for confined statistical source was driven by a significantly more negative mean ERP amplitudes in
estimation of first-order predictions instead of calculating full cortex response to second compared to first stimuli (S1 ¼ 1.75
1.35 μV; S2 ¼
solutions for all experimental sequences. Cortical source estimation of 0.67
1.14 μV; t(15) ¼ 6.06; p < .001).
higher-order predictions were based on between-sequence MMN, where Planned MMN comparisons showed a significantly smaller first-order
the A-B standard auditory sequence was subtracted from all deviant MMN amplitude elicited by the standard A-B condition (1.32
1.56
conditions. To avoid circularity, an average of all between-sequence μV) compared to the deviant conditions A-A (2.76
2.31 μV; t(15) ¼
MMN components was used to define ROIs that were then selectively 2.74; p ¼ .015) and B–B (2.86
1.59 μV; t(15) ¼ 2.98; p < .01); a
applied to single conditions. ROIs included the center voxels of activated similar result was found in comparison with the B-A condition, although
Brodmann Areas and surrounding voxels within a 10 mm radius. To statistical significance was missed (2.34
3.60 μV; t(15) ¼ 1.539; p ¼
avoid alpha error inflation, a maximum of three ROIs was chosen to .145). No significant differences were found between higher-order MMN
analyze cortical activation patterns associated with first-order and amplitudes. These results principally confirm the idea that first-order and
higher-order predictions. higher-order prediction processes are superimposed within the auditory
MMN, which, in turn, can again be dissected into distinct prediction
2.5. Statistical analysis processes (see also Figs. 3 and 4). First-order and higher-order MMN
components were then submitted to source localization of respective
SPSS 20.0 (IBM, Armonk, US) was used for statistical computations. cortical prediction processes.
ERP components were analyzed using repeated measures analyses of
variance (rmANOVA) and post hoc linear contrasts and paired t-tests, as 3.3. First-order cortical activation
appropriate. Omnibus rmANOVAs were applied to N1 amplitude (stim-
ulus order[2]*condition[4]) and mean amplitude at 150–250 ms (stim- Cortical sources of the first-order MMN following the standard A-B
ulus order[2]*condition[4]). Finally, planned comparisons of MMN sequence were estimated using a two-sided paired t-tests with a conser-
amplitudes between conditions were done using paired t-tests. Alpha was vative p < .01. The resulting treshold of t ¼ 4.202 was exceeded by ac-
set at p < .05 for all tests. Greenhouse-Geisser correction was done in case tivations of primary auditory cortex in Brodmann Area 41 bilaterally
of violation of the sphericity assumption. Partial η2 served as an estimator (right: MNI X, Y, Z ¼ 41, 30, 7; t ¼ 6.73; left: MNI X, Y, Z ¼ 48, 28,
of the variance accounted for by the model. Statistical imaging of current 15; t ¼ 4.61) as well as left inferior frontal gyrus (IFG) in Brodmann Area
density differences was done based on non-parametric voxel-by-voxel t- 47 (MNI X, Y, Z ¼ 37, 21, 5; t ¼ 5.14).
tests (Holmes et al., 1996). This maximum t-statistic offers a procedure of Corresponding ROIs were defined to selectively calculate statistical
5,000 bootstrap resampling operations across conditions, which pro- activation differences within each sequence. Here, primary auditory
duces threshold values applicable to single voxel p’s. cortex (averaged across hemispheres) was significantly activated in se-
quences A-B (threshold t ¼ 3.155; ROI t ¼ 4.18) and B-A (threshold t ¼
2.6. Data availability statement 3.314; ROI t ¼ 3.37), but not in sequences A-A (threshold t ¼ 3.346; ROI
t ¼ 2.57) or B–B (threshold t ¼ 3.339; ROI t ¼ 2.58). Similarly, left IFG
Raw data were generated at the Department of Psychiartry, Charit
e was significantly activated during sequences A-B (threshold t ¼ 3.155;
University Medicine. Derived data supporting the findings of this study ROI t ¼ 4.03) and B-A (threshold t ¼ 3.314; ROI t ¼ 3.47), but not during
are available from the corresponding author on request. sequences A-A (threshold t ¼ 3.346; ROI t ¼ 2.78) or B–B (threshold t ¼
3.339; ROI t ¼ 2.33). This response pattern functionally conforms to
3. Results detecting auditory stimulus deviance (see Fig. 3).

3.1. N1 effects 3.4. Higher-order cortical activation

Omnibus ANOVA of the N1 component showed a significant main The MMN signal to the standard sequence A-B was then subtracted
effect of the within-subject factor stimulus order (F(1,15) ¼ 33.946; p < from each deviant sequence’s MMN (A-A, B-A, B–B) to isolate higher-
.001; partial η2¼ .694) due to larger, i.e. more negative, N1 amplitudes order MMN signals, which were averaged and submitted to an explor-
evoked by first stimuli (4.44
1.26 μV) compared with second stimuli atory source analysis, again using a two-sided test with a conservative p
(2.60
1.10 μV; t(15) ¼ 5.83; p < .001). Within-subject linear < .01. We found widespread significant activations at a treshold of t ¼
contrast analysis revealed a significant interaction of stimulus order*- 3.898; in order to reduce the number of statistical contrast and in keeping
condition (F(1,15) ¼ 8.498; p ¼ .011; partial η2¼ .362), driven by with ROI definition of first-order MMN, subsequent analyses were
significantly smaller adaptation effects of N1 amplitudes to the second restricted to the two clusters with highest t values, i.e. anterior cingulate

5
C. Hofmann-Shen et al. NeuroImage 207 (2020) 116432

Fig. 3. (a) First-order MMN components, where 0 ms corresponds to onset of second stimuli of the standard A-B (black) and the deviant A-A (red), B-A (blue), and B–B
(green) sequences. The standard sequence A-B served as subtrahend for subsequent isolations of higher-order MMN components and was identified at 150–250 ms
after stimulus onset. (b) Cortical source analysis of the standard sequence A-B revealed significant activations of primary auditory cortex bilaterally (Brodmann Area
41) and left inferior frontal gyrus (BA 47). (c) Left IFG and bilateral auditory cortex were defined as regions of interest for analysis of cortical sources of first-order
MMN. (d) ROIs in auditory cortex (top) and in left IFG (bottom) were significantly activated in sequences A-B and B-A, but not A-A or B–B, which conforms to a pattern
consistent with detection of physical stimulus deviance. Red lines indicate single ROI t-thresholds for statistical activation at p < .05 across conditions. (e) Linear co-
activation of auditory cortex and left IFG consistent with deviance detection.

cortex (ACC) in Brodmann Area 24 (MNI X, Y, Z: 0, 25, 20; t ¼ 6.42) and uncertainty in right IFG (see Fig. 4).
right IFG in Brodmann Area 47 (MNI X, Y, Z: 40, 15, 10; t ¼ 5.94).
A significant ACC activation was found in sequence A-A (threshold t 4. Discussion
¼ 3.775; ROI t ¼ 4.13), but not B-A (threshold t ¼ 3.494; ROI t ¼ 3.11) or
B–B (threshold t ¼ 3.575; ROI t ¼ 3.25). By contrast, IFG was signifi- In our EEG study, we employed an auditory sound pair paradigm to
cantly activated in sequences B-A (threshold t ¼ 3.494; ROI t ¼ 4.55) and assess the contribution of neuronal adaptation and prediction error
B–B (threshold t ¼ 3.575; ROI t ¼ 4.11), but not A-A (threshold t ¼ 3.775; mechanisms to auditory deviance detection and detect MMN generators
ROI t ¼ 2.20). This pattern suggests coding of prediction error in ACC and in dependance of deviance complexity.

6
C. Hofmann-Shen et al. NeuroImage 207 (2020) 116432

Fig. 4. (a) Higher-order MMN components elicited in response to second stimuli in deviant sequences A-A (red), B-A (blue), and B–B (green); the standard sequence A-
B is subtracted from all signals. (b) Cortical source analysis averaged across all deviant sequences showed activations of anterior cingulate cortex (BA 24) and right IFG
(BA 47). (c) Corresponding cortical ROIs in right IFG and in ACC for source estimation of higher-order MMN. (d) ACC (top) was significantly activated in sequence A-
A, but not B-A or B–B, thus conforming to a prediction error signal. Right IFG (bottom) was significantly activated in sequences B-A and B–B, but not A-A, which is
consistent with signalling absence of predictions, i.e. uncertainty. Red lines indicate single ROI t-thresholds for statistical activation at p < .05 across conditions. (e)
Anti-correlated activation of ACC and right IFG consistent with cortical prediction error signalling.

Adaptation of N1 were found both in the standard and the deviant general, irrespective of frequency, leads to neuronal attenuation. The
conditions. Additionally, in the deviant conditions, presentation of finding of maximal N1-adaptation in the deviant conditions with
identical stimuli (A-A and B–B) lead to increased stimulus-specific repeated identical stimuli, but not the standard condition with repeated
adaptation. It is known that stimulus repetition leads to decreased in- patterns supports the assumption that N1 reflects a lower cortical process
tensity of activity of frequency-specific neurons, whereas adaptation to a primarily involved in the encoding of physical features such as frequency,
lesser degree also takes place in other frequency-specific neuronal pop- but not of, even simple, auditory patterns. Our findings argue against
ulations responding best to other frequencies (Herrmann et al., 2013a; top-down neuronal mechanisms in terms of detection of regularity and its
a€
J€ askel€ainen et al., 2011). Our findings confirm that, apart from violation in the generation of N1 as previously purposed (Todorovic
frequency-specific N1-adaptation, repeated auditory stimulation in et al., 2011), but suggest that cortical encoding of prediction only plays a

7
C. Hofmann-Shen et al. NeuroImage 207 (2020) 116432

minor role in the modulation of N1. Summarizing, our study aimed to assess different levels of cortical
Further, we found significant first-order MMN response not only in auditory deviance detection. Our findings suggest that N1 reflects a lower
the deviant, but also in the standard condition; however, the first-order cortical process primarily involved in the encoding of physical features
MMN elicited by deviant conditions is larger compared to that elicited by underlying neuronal habituation mechanisms, while MMN consists of
by the standard condition. A previous work by Sussmann and Winkler subcomponents reflecting different hierachical levels of auditory pro-
(2001) studying deviant tone pairs has shown that no MMN is elicited cessing from the detection of simple physical features to the prediction of
when pairs consisting of two different tones are repeated with 100% uncertainty and regularity violation. While detection of physical devi-
probability, suggesting that those deviant tone pairs with high repetition ance is processed in the temporal cortices, detection of prediction error is
probability are unified into a single event during the cortical processing. in the frontal and prefrontal areas.
Contradictory results were presented by Tavano et al. (2014) who found
an attenuated MMN elicited by deviant tones with high repetition Data availability statement
probability. In our experiment, the standard tone pair, presented with
high repetition probability, yielded a clear first-order MMN. Our findings Raw data were generated at the Department of Psychiartry, Charit
e
argue against a unified cortical processing of tone pairs and in favor of University Medicine. Derived data supporting the findings of this study
separate processing of each tone. It has been previously proposed that are available from the corresponding author on request.
auditory MMN consists of both memory-based and memory-independent,
sensorial effects (Jacobsen and Schr€ oger, 2001; Maess et al., 2007). Our
Author contributions section
results are in line with those previous findings and suggest that physical
sound deviance per se, although embedded in an repeatedly presented
Christina Hofmann-Shen: conceptualization, methodology, inves-
auditory pattern, leads to the generation of first-order MMN, while
tigation, formal analysis, data curation, writing – original draft, writing –
complex regularity violations are reflected in the higher-order, more
review & editing.
evaluative MMN.
Bob O. Vogel: methodology, investigation, data curation, writing –
Various studies have located the neural source of MMN to superior
original draft.
temporal cortex and predominantly right inferior frontal gyrus (MacLean
Maximillian Kaffes: formal analysis, data curation, writing – original
et al., 2015; N€a€
at€
anen et al., 2007). It has been suggested that change in
draft.
certain physical features, such as frequency or duration, are detected in
Armin Rudolph: formal analysis, data curation, writing – original
auditory cortex, while the frontal cortex plays an more important role in
draft.
attention switching, whenever salient environmental changes are
Elliot C. Brown: Resources, supervision, writing – original draft.
detected (Tse et Penney, 2008; Tse et al., 2013). Our source localization
Cumhur Tas: writing – original draft.
analysis of the first-order MMN revealed a significant activation of pri-
Martin Brüne: Resources, supervision, writing – original draft.
mary auditory cortex only when tone pairs with different frequencies,
Andres H. Neuhaus: conceptualization, methodology, formal anal-
irresponsive of probability, were presented, confirming that only fre-
ysis, data curation, writing – original draft, writing – review & editing,
quency deviance, but not regularity violation was processed in the pri-
resources, supervision.
mary auditory cortex. Interestingly, the same activation pattern was
found for the left IFG. Involvement of left IFG in auditory deviance
detection has previously been described, however, its precise role re- Declaration of competing interest
mains unclear (Doeller et al., 2003; Molholm et al., 2005; Phillips et al.,
2016). Our results suggest that left IFG, like primary auditory cortex, is There is no conflict of interest, financial or otherwise, related to this
primarily involved in the detection of simple, physically deviant features, work for any of the authors.
while right IFG is involved in the detection of pattern changes. Although
significant activity in left IFG was not found for higher-order MMN, an
References
involvement of left IFG in the detection of complex pattern changes
cannot be excluded. Further investigations on the role of left and right Aghamolaei, M., Zarnowiec, K., Grimm, S., Escera, C., 2016. Functional dissociation
IFG in generating the MMN component could be valuable. between regularity encoding and deviance detection along the auditory hierarchy.
After substracting the MMN signal to the standard sequence from Eur. J. Neurosci. 43, 529–535. https://doi.org/10.1111/ejn.13138.
Alexander, W.H., Brown, J.W., 2017. The role of the anterior cingulate cortex in
deviant sequences, our source analysis revealed significant ACC activa- prediction error and signaling surprise. Top. Cogn. Sci. 11, 119–135. https://doi.org/
tion for the sequence A-A and right IFG activation for the sequences B-A 10.1111/tops.12307.
and B–B. While an expected first stimulus is unexpectedly repeated in the Althen, H., Grimm, S., Escera, C., 2013. Simple and complex acoustic regularities are
encoded at different levels of the auditory hierarchy. Eur. J. Neurosci. 38,
A-A condition, an unexpected first stimulus is presented in the B-A and 3448–3455. https://doi.org/10.1111/ejn.12346.
B–B condition, with uncertainty what follows next. Thus, our source Carbajal, G.V., Malmierca, M.S., 2018. The neuronal basis of predictive coding along the
analysis pattern suggests that right IFG does not encode complex auditory auditory pathway: from the subcortical roots to cortical deviance detection. Trends
Hear 22. https://doi.org/10.1177/2331216518784822, 2331216518784822.
changes in general, but in terms of novel, unexpected auditory input that Cornella, M., Leung, S., Grimm, S., Escera, C., 2012. Detection of simple and pattern
does not allow for specific predictions. Our result is thus consistent with regularity violations occurs at different levels of the auditory hierarchy. PLoS One 7,
previous findings concerning coding of uncertainty in right IFG (Kluger e43604. https://doi.org/10.1371/journal.pone.0043604.
Doeller, C.F., Opitz, B., Mecklinger, A., Krick, C., Reith, W., Schr€
oger, E., 2003. Prefrontal
and Schubotz, 2017; Meyniel and Dehaene, 2017; Nastase et al., 2014).
cortex involvement in preattentive auditory deviance detection: neuroimaging and
Regularity violation of known patterns was associated with ACC activa- electrophysiological evidence. Neuroimage 20, 1270–1282. https://doi.org/
tion in our study, consistent with involvement of ACC in auditory devi- 10.1016/S1053-8119(03)00389-6.
Escera, C., Leung, S., Grimm, S., 2014. Deviance detection based on regularity encoding
ance processing in general (Molholm et al., 2005), coding of error
along the auditory hierarchy: electrophysiological evidence in humans. Brain Topogr.
likelihood (Modirrousta and Fellows, 2008), as well as signalling pre- 27, 527–538. https://doi.org/10.1007/s10548-013-0328-4.
diction error in particular (Alexander and Brown, 2017). However, ACC Friston, K., 2005. A theory of cortical responses. Philos. Trans. R. Soc. Lond. B Biol. Sci.
is more known to be involved in the processing of behavorial prediction 360, 815–836. https://doi.org/10.1098/rstb.2005.1622.
Grill-Spector, K., Henson, R., Martin, A., 2006. Repetition and the brain: neural models of
error in terms of distinguishing between positive and negative valence of stimulus-specific effects. Trends Cogn. Sci. 10, 14–23. https://doi.org/10.1016/
an unexpected event (Alexander and Brown, 2017; Hyman et al., 2017). j.tics.2005.11.006.
Our results are consistent with previous reports on ACC playing a crucial Grimm, S., Escera, C., Slabu, L., Costa-Faidella, J., 2011. Electrophysiological evidence for
the hierarchical organization of auditory change detection in the human brain.
role in prediction of likely events and detection of prediction error in Psychophysiology 48, 377–384. https://doi.org/10.1111/j.1469-
general. 8986.2010.01073.x.

8
C. Hofmann-Shen et al. NeuroImage 207 (2020) 116432

Herrmann, B., Henry, M.J., Obleser, J., 2013. Frequency-specific adaptation in human Nastase, S., Iacovella, V., Hasson, U., 2014. Uncertainty in visual and auditory series is
auditory cortex depends on the spectral variance in the acoustic stimulation. coded by modality-general and modality-specific neural systems. Hum. Brain Mapp.
J. Neurophysiol. 109, 2086–2096. https://doi.org/10.1152/jn.00907.2012. 35, 1111–1128. https://doi.org/10.1002/hbm.22238.
Holmes, A.P., Blair, R.C., Watson, J.D., Ford, I., 1996. Nonparametric analysis of statistic Neuhaus, A.H., Brandt, E.S., Goldberg, T.E., Bates, J.A., Malhotra, A.K., 2013. Evidence
images from functional mapping experiments. J. Cereb. Blood Flow Metab. 16, 7–22. for impaired visual prediction error in schizophrenia. Schizophr. Res. 147, 326–330.
https://doi.org/10.1097/00004647-199601000-00002. https://doi.org/10.1016/j.schres.2013.04.004.
Horn, W., 1983. Leistungsprüfsystem: LPS. Hogrefe, G€ ottingen. Oldfield, R.C., 1971. The assessment and analysis of handedness: the Edinburgh
Hyman, J.M., Holroyd, C.B., Seamans, J.K., 2017. A novel neural prediction error found in inventory. Neuropsychologia 9, 97–113.
anterior cingulate cortex ensembles. Neuron 95, 447–456. https://doi.org/10.1016/ Parras, G.G., Nieto-Diego, J., Carbajal, G.V., Vald
es-Baizabal, C., Escera, C.,
j.neuron.2017.06.021. Malmierca, M.S., 2017. Neurons along the auditory pathway exhibit a hierarchical
a€
J€ askel€ainen, I.P., Ahveninen, J., Andermann, M.L., Belliveau, J.W., Raij, T., Sams, M., organization of prediction error. Nat. Commun. 8, 2148. https://doi.org/10.1038/
2011. Short-term plasticity as a neural mechanism supporting memory and s41467-017-02038-6.
attentional functions. Brain Res. 1422, 66–81. https://doi.org/10.1016/ Pascual-Marqui, R.D., 2002. Standardized low-resolution brain electromagnetic
j.brainres.2011.09.031. tomography (sLORETA): technical details. Methods Find. Exp. Clin. Pharmacol. 24,
Jung, T.P., Makeig, S., Humphries, C., Lee, T.W., McKeown, M.J., Iragui, V., 5–12.
Sejnowski, T.J., 2000. Removing electroencephalographic artifacts by blind source Phillips, H.N., Blenkmann, A., Hughes, L.E., Kochen, S., Bekinschtein, T.A., Cam-Can,
separation. Psychophysiology 37, 163–178. Rowe, J.B., 2016. Convergent evidence for hierarchical prediction networks from
Jacobsen, T., Schr€ oger, E., 2001. Is there pre-attentive memory-based comparison of human electrocorticography and magnetoencephalography. Cortex 82, 192–205.
pitch? Psychophysiology 38, 723–727. https://doi.org/10.1016/j.cortex.2016.05.001.
Kekoni, J., H€ am€ al€
ainen, H., Saarinen, M., Gr€ohn, J., Reinikainen, K., Lehtokoski, A., Recasens, M., Grimm, S., Wollbrink, A., Pantev, C., Escera, C., 2014. Encoding of nested
N€at€
anen, R., 1997. Rate effect and mismatch responses in the somatosensory system: levels of acoustic regularity in hierarchically organized areas of the human auditory
ERP-recordings in humans. Biol. Psychol. 46, 125–142. cortex. Hum. Brain Mapp. 35, 5701–5716. https://doi.org/10.1002/hbm.22582.
Kluger, D.S., Schubotz, R.I., 2017. Strategic adaptation to non-reward prediction error Rentzsch, J., Shen, C., Jockers-Scherübl, M.C., Gallinat, J., Neuhaus, A.H., 2015. Auditory
qualities and irreducible uncertainty in fMRI. Cortex 97, 32–48. https://doi.org/ mismatch negativity and repetition suppression deficits in schizophrenia explained
10.1016/j.cortex.2017.09.017. by irregular computation of prediction error. PLoS One 10, e0126775. https://
Kriegeskorte, N., Simmons, W.K., Bellgowan, P.S., Baker, C.I., 2009. Circular analysis in doi.org/10.1371/journal.pone.0126775.
systems neuroscience: the dangers of double dipping. Nat. Neurosci. 12 (5), 535–540. Summerfield, C., Trittschuh, E.H., Monti, J.M., Mesulam, M.M., Egner, T., 2008. Neural
https://doi.org/10.1038/nn.2303. repetition suppression reflects fulfilled perceptual expectations. Nat. Neurosci. 11,
Lecaignard, F., Bertrand, O., Gimenez, G., Mattout, J., Caclin, A., 2015. Implicit learning 1004–1006. https://doi.org/10.1038/nn.2163.
of predictable sound sequences modulates human brain responses at different levels Stefanics, G., Csukly, G., Koml
osi, S., Czobor, P., Czigler, I., 2012. Processing of
of the auditory hierarchy. Front. Hum. Neurosci. 9, 505. https://doi.org/10.3389/ unattended facial emotions: a visual mismatch negativity study. Neuroimage 59,
fnhum.2015.00505. 3042–3049. https://doi.org/10.1016/j.neuroimage.2011.10.041.
Lehrl, S., 1999. Mehrfachwahl-Wortschatz-Intelligenztest: MWT-B. Balingen: Spitta. Sussman, E., Winkler, I., 2001. Dynamic sensory updating in the auditory system. Brain
MacLean, S.E., Blundon, E.G., Ward, L.M., 2015. Brain regional networks active during Res. Cogn. Brain Res. 12, 431–439.
the mismatch negativity vary with paradigm. Neuropsychologia 75, 242–251. Tavano, A., Widmann, A., Bendixen, A., Trujillo-Barreto, N., Schr€ oger, E., 2014. Temporal
https://doi.org/10.1016/j.neuropsychologia.2015.06.019. regularity facilitates higher-order sensory predictions in fast auditory sequences. Eur.
Maess, B., Jacobsen, T., Schr€ oger, E., Friederici, A.D., 2007. Localizing pre-attentive J. Neurosci. 39, 308–318. https://doi.org/10.1111/ejn.12404.
auditory memory-based comparison: magnetic mismatch negativity to pitch change. Talairach, J., Tournoux, P., 1988. Co-Planar Stereotaxic Atlas of the Human Brain: 3-
Neuroimage 37, 561–571. https://doi.org/10.1016/j.neuroimage.2007.05.040. Dimensional Proportional System: an Approach to Cerebral Imaging. Thieme Medical
Meyniel, F., Dehaene, S., 2017. Brain networks for confidence weighting and hierarchical Publishers, New York.
inference during probabilistic learning. Proc. Natl. Acad. Sci. U. S. A. 114, Todorovic, A., van Ede, F., Maris, E., de Lange, F.P., 2011. Prior expectation mediates
E3859–E3868. https://doi.org/10.1073/pnas.1615773114. neural adaptation to repeated sounds in the auditory cortex: an MEG Study.
Modirrousta, M., Fellows, L.K., 2008. Dorsal medial prefrontal cortex plays a necessary J. Neurosci. 31, 9118–9123. https://doi.org/10.1523/JNEUROSCI.1425-11.2011.
role in rapid error prediction in humans. J. Neurosci. 28, 14000–14005. https:// Tse, C.Y., Penney, T.B., 2008. On the functional role of temporal and frontal cortex
doi.org/10.1523/JNEUROSCI.4450-08.2008. activation in passive detection of auditory deviance. Neuroimage 41, 1462–1470.
Molholm, S., Martinez, S., Ritter, S., Javitt, D.C., Foxe, J.J., 2005. The neural circuitry of https://doi.org/10.1016/j.neuroimage.2008.03.043.
pre-attentive auditory change-detection: a fMRI study of pitch and duration mismatch Tse, C.Y., Rinne, T., Ng, K.K., Penney, T.B., 2013. The functional role of the frontal cortex
negativity generators. Cerebr. Cortex 15, 545–551. https://doi.org/10.1093/cercor/ in pre-attentive auditory change detection. Neuroimage 83, 870–879. https://
bhh155. doi.org/10.1016/j.neuroimage.2013.07.037.
a€
N€ at€anen, R., Escera, C., 2000. Mismatch negativity: clinical and other applications. Vogel, B.O., Shen, C., Neuhaus, A.H., 2015. Emotional context facilitates cortical
Audiol. Neuro. Otol. 5, 105–110. https://doi.org/10.1159/000013874. prediction error responses. Hum. Brain Mapp. 36, 3641–3652. https://doi.org/
a€
N€ at€anen, R., Jacobsen, T., Winkler, I., 2005. Memory-based or afferent processes in 10.1002/hbm.22868.
mismatch negativity (MMN): a review of the evidence. Psychophysiology 42, 25–32. Wacongne, C., Changeux, J.P., Dehaene, S., 2012. A neuronal model of predictive coding
https://doi.org/10.1111/j.1469-8986.2005.00256.x. accounting for the mismatch negativity. J. Neurosci. 32 (11), 3665–3678. https://
a€
N€ at€anen, R., Paavilainen, P., Rinne, T., Alho, K., 2007. The mismatch negativity (MMN) doi.org/10.1523/JNEUROSCI.5003-11.2012.
in basic research of central auditory processing: a review. Clin. Neurophysiol. 118,
2544–2590. https://doi.org/10.1016/j.clinph.2007.04.026.