www.sciencedirect.com

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9

Available online at

ScienceDirect

Journal homepage: www.elsevier.com/locate/cortex

Research report

Dynamics of brain activity underlying workingmemory for music in a naturalistic condition

Iballa Burunat a,*, Vinoo Alluri a, Petri Toiviainen a, Jussi Numminen b andElvira Brattico c,d,e

aFinnish Centre for Interdisciplinary Music Research, Department of Music, University of Jyvaskyla, FinlandbHelsinki Medical Imaging Center at Toolo Hospital, University of Helsinki, FinlandcBrain & Mind Lab, Department of Biomedical Engineering and Computational Science (BECS),

Aalto University School of Science, FinlanddCognitive Brain Research Unit (CBRU), Institute of Behavioral Sciences, University of Helsinki, FinlandeAdvanced Magnetic Imaging (AMI) Centre, Aalto University School of Science, Finland

a r t i c l e i n f o

Article history:

Received 21 August 2013

Reviewed 22 October 2013

Revised 17 January 2014

Accepted 24 April 2014

Action editor Asaf Gilboa

Published online 9 May 2014

Keywords:

Working memory (WM)

Music

Naturalistic

Functional magnetic resonance

imaging (fMRI)

Hippocampus

* Corresponding author. Department of MusiE-mail address: iballa.burunat@jyu.fi (I. B

http://dx.doi.org/10.1016/j.cortex.2014.04.0120010-9452/ª 2014 Elsevier Ltd. All rights rese

a b s t r a c t

We aimed at determining the functional neuroanatomy of working memory (WM) recog-

nition of musical motifs that occurs while listening to music by adopting a non-standard

procedure. Western tonal music provides naturally occurring repetition and variation of

motifs. These serve as WM triggers, thus allowing us to study the phenomenon of motif

tracking within real music. Adopting amodern tango as stimulus, a behavioural test helped

to identify the stimulus motifs and build a time-course regressor of WM neural responses.

This regressor was then correlated with the participants’ (musicians’) functional magnetic

resonance imaging (fMRI) signal obtained during a continuous listening condition. In order

to fine-tune the identification of WM processes in the brain, the variance accounted for by

the sensory processing of a set of the stimulus’ acoustic features was pruned from par-

ticipants’ neurovascular responses to music. Motivic repetitions activated prefrontal and

motor cortical areas, basal ganglia, medial temporal lobe (MTL) structures, and cerebellum.

The findings suggest that WM processing of motifs while listening to music emerges from

the integration of neural activity distributed over cognitive, motor and limbic subsystems.

The recruitment of the hippocampus stands as a novel finding in auditory WM. Effective

connectivity and agglomerative hierarchical clustering analyses indicate that the hippo-

campal connectivity is modulated by motif repetitions, showing strong connections with

WM-relevant areas (dorsolateral prefrontal cortex e dlPFC, supplementary motor area e

SMA, and cerebellum), which supports the role of the hippocampus in the encoding of the

musical motifs in WM, and may evidence long-term memory (LTM) formation, enabled by

the use of a realistic listening condition.

ª 2014 Elsevier Ltd. All rights reserved.

c, University of Jyvaskylaurunat).

rved.

, PL 35(M), 40014 Jyvaskyla, Finland.

1 Method that consists in ‘occupying’ or ‘using’ the phonolog-ical loop by, i.e., repeating a word. By engaging the phonologicalloop in this way, it can be determined whether other tasksrequiring the same loop are inhibited.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9 255

1. Introduction

Music is ubiquitous and seems to be associated with a distinct

brain architecture. In recent years there has been a significant

increase in studies on low- and high-level music processing

in the brain, including phenomena such as perception of

psycho-acoustic features, performance, and music-driven

emotion and memory, all aimed at describing and under-

standing musicebrain interaction: how music engages the

brain and how it affects cognition in different ways. Working

memory (WM), necessary for the integration of sound events

over time, is crucial for making sense of the continuity of our

musical experience (i.e., by keeping auditory items accessible

to the memory system), allowing for expectations and pre-

dictions about upcoming events, which in turn facilitates

appreciation and emotional reactions in music. Ultimately,

the study of music-related memory circuits in the brain could

elucidate the distinct ways in which our selective brains listen

tomusic. Finally it is important to emphasize that the study of

how memory encodes music will also tell us about the nature

of human memory in general.

According to the unifying definition proposed by Miyake

and Shah (1999), WM is those mechanisms dependent on

various subsystems, involved in controlling, regulating and

maintaining active task-relevant information needed by

complex cognitive functions. Moreover, WM is neither found

in a fixed location in the cognitive architecture nor are its

limits fixed. More accurately, WM is dependent upon several

factors, and it may even be an emergent property of the

various mechanisms at work. Limitations deriving from our

WM capacity influence how humans perceive temporal pat-

terns of events and the boundaries between them, guiding our

decisions about how they relate to one another in order to

comprehend them as a whole and predict future events

(Snyder, 2000). Thus, WM mechanisms are required to form a

coherent representation of the auditory flow by enabling the

retention of information over time.

Although themapping of the functional architecture ofWM

in the brain is a very intricate research subject, there is

consensus in the research community about the critical

involvement of areas in the prefrontal cortex (PFC) in WM

functions, as indicated by Fuster’s (1987) main results from

single unit recordings in animals. Further research confirmed

that the PFC along with the basal ganglia are the most consis-

tently recruitedbrainstructures inWM(Fuster, 1999;Goldman-

Rakic, 1995; Gruber, Dayan, Gutkin, & Solla, 2006). Within the

PFC, the dorsolateral prefrontal cortex (dlPFC) has been

particularly and consistently found active in tasks requiring

executive functions (Kane & Engle, 2002), i.e., regulation of

encoding, strategy selection, andmanipulation and retrieval of

information, as opposed to active maintenance functions (i.e.,

keeping information available online), for which consensus

regarding its loci in the functional brain anatomy has thus far

not been reached. In connection with the dlPFC, the hippo-

campus, amedial temporal lobe (MTL) structure, seems tohave

a role in encoding and retrieval (but not in maintenance)

functions while learning new material, being it stored within

WM or for longer periods of time in long-term memory (LTM;

e.g., Karlsgodt, Shirinyan, van Erp, Cohen, & Cannon, 2005).

The neural networks engaged in auditory WM for music

have only recently begun to be investigated. Preliminary

studies on music-driven WM revealed activity in the supra-

temporal lobe, inferior frontal cortex and motor areas (Gaab,

Gaser, Zaehle, Jancke, & Schlaug, 2003; Koelsch et al., 2009;

Schulze, Zysset, Mueller, Friederici, & Koelsch, 2011), only

partly confirming what has been found on WM via other

sensory modalities in animals or humans. In addition to the

scarcity ofmusic-relatedWMstudies, in neuroscienceWMhas

thus far not been studied in naturalistic listening situations,

but rather using artificial target detection tasks (i.e., n-back

and Sternberg) with simpler, manipulated materials, all of

which might create mental states not characteristic of brain’s

behaviour in more natural, attentive situations. However, as

stated by Janata, Tillmann, and Bharucha (2002), humans have

evolved in a natural complex auditory scene environment,

capable of segregating auditory objects for interaction and

survival. Hence, in studying music-driven cognitive processes

in the brain,more naturalistic approaches are crucial if we aim

to (a)map those functional brain areas engaged in acoustically

complex environment-conditioned processing, and (b)

compare the experimental findings resulting from the use of

artificially created stimuli with more natural and complex

approaches that more reliably replicate the acoustic environ-

ments our brains have adapted to.

Using functional magnetic resonance imaging (fMRI) in a

verbal and tonal WM task involving nonsense sentences and

novel piano melodies, Hickok, Buchsbaum, and Humphries

(2003) observed recruitment of part of the left posterior pla-

num temporale, Broca’s area, and left premotor region in non-

musicians. Similar areas were found by Koelsch et al. (2009)

also in non-musicians employing sung syllables, for which

either pitch or text had to be remembered during rehearsal

and articulatory suppression1 (Broca’s area, ventrolateral

premotor cortex e vlPMC, dorsal premotor cortex, the planum

temporale, inferior parietal lobe, the anterior insula, basal

ganglia and thalamus, and the cerebellum, with subcentral

gyrus and putamen only in the verbal, and globus pallidus in

the tonal condition). Schulze et al. (2011) included musicians

and non-musicians and showed overlapping topography be-

tween the verbal and non-verbal conditions, validating pre-

vious findings. However, a number of core WM structures

(Broca’s area, left premotor cortex, pre-SMA/SMA, left insular

cortex, left inferior parietal lobe) differed significantly be-

tween groups in terms of their weightings for both tonal and

verbal WM. Additionally, in musicians specific structures

were recruited only during verbal (right insular cortex) or only

during tonal WM (right globus pallidus, right caudate nucleus,

left cuneus and left cerebellum; plus the right premotor cor-

tex, the left putamen, and the right cerebellum were more

strongly engaged than in the verbal condition). These results

point to the existence of both a tonal and a phonological loop

in musicians (Schulze & Koelsch, 2012). In general, it is

reasonable to expect an overlap of areas for language and

music since music processing shares some circuitry with

2 MAX/MSP 5.1 (Cycling ’74, San Francisco, CA 94103, USA).3 These excerpts correspond to higher level section boundaries

of the stimulus starting at (1) 0:00, (2) 1:08, (3) 1:28, (4) 2:23, (5) 2:56,(6) 3:39, (7) 4:36, (8) 5:08, (9) 6:17, (10) 7:07 with an average length of50.6 sec each.

4 Audio CD (May 23, 2006). Label: Milan Records; ASIN:B000F6ZID8. Performers: Astor Piazzolla (bandoneon); HoracioMalvicino (guitar); Carlos Nozzi (violoncello); Angel Ridolfi (doublebass, upright bass); Daniel Binelli (bandoneon); Gerardo Gandini(piano). The track is available in Spotify: http://open.spotify.com/track/6X5SzbloyesrQQb3Ht4Ojx.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9256

spoken language processing (Koelsch et al., 2002; Patel, 2003;

Patel, Gibson, Ratner, Besson, & Holcomb, 1998; Steinbeis &

Koelsch, 2008).

The above-described studies on musical WM are not

conclusive in defining the mechanisms implicated in WM

processes involving musical information, nor do they employ

a naturalistic paradigm. Studies on music-related WM only

partially provided support for the importance of dlPFC and

hippocampus for WM processes and instead highlighted the

role of auditory, motor and parietal areas. Natural environ-

ments represent an added value, which usually requires “the

parallel processing of information related to different objects

or events that have to be kept apart to allow sensory seg-

mentation and goal-directed behaviour” (Engel & Singer,

2001). For instance, even though the relation between

reading capacities in children and verbal WM has been widely

studied, the online formation of short lasting memories for

repetitive words has not been correlated thus far with the

fMRI signal change (Cain, Oakhill, & Bryant, 2004; Casey et al.,

1995; Georgiewa et al., 1999). Additionally, as evidenced by

studies using a naturalistic approach with visual stimuli, the

presentation of complex stimulation highly resembling real

life activates more powerfully the brain, allowing the

recruitment of neural structures otherwise not reaching sig-

nificance level (Alluri et al., 2012; Hasson, Nir, Levy,

Fuhrmann, & Malach, 2004). Until now, more naturalistic ap-

proaches to studying auditory WM have not been feasible.

This has been due to methodological limitations in the

extraction of time course predictors from naturalistic stimuli

and the lack of appropriate models for explaining the fMRI

variance, capable of dealing with the larger variability

intrinsic in such naturalistic stimuli.

Here we aimed to overcome previous methodological lim-

itations, while allowing for a systematic control of confound-

ing factors, by combining knowledge from three sources:

computational analysis of themusical stimulus, a behavioural

test, and neuroimaging, with the motivation to determine for

the first time the topography of music-elicited WM formed

online while listening to real music. To accomplish this, we

used a novel holistic approach. First, we adopted a naturalistic

setting, denoting both (a) a complex, non-manipulated piece of

music; and (b) a continuous, free listening setting, whereby

participants attentively listened to the piece frombeginning to

end without performing any tasks. This allowed participants

to avoid possible mental states arising from such target

detection tasks that may not reliably reflect brain’s func-

tioning in a more natural setting. Second, we removed from

the participants’ brain responses the variance explained by a

set of perceptually validated acoustical features computa-

tionally extracted from the music. Additionally, effective

connectivity analyses were conducted between identified

WM-relevant areas to deepen the understanding about their

interactions.

Hence, the present paradigm constitutes a novel approach

to the standard practice in research studies focussing on

music processing in the brain (e.g., Brattico et al., 2011;

Koelsch et al., 2009; Levitin & Menon, 2003; Pallesen et al.,

2010; Pereira et al., 2011; ). Third, the activation of music-

related neural networks involved in WM was investigated by

tracking the repetition of the salient motifs in the piece. More

specifically, a regressor modelling predicted responses to

motif repetitions was built to help retrieve the brain areas that

highly correlated with it. We expected to observe brain ac-

tivity in structures previously associated toWMas revealed by

conventional tasks, but also additional structures involved in

the phenomenon under investigation that could be evidenced

by this naturalistic paradigm (Alluri et al., 2012; Hasson et al.,

2004).

2. Methods

2.1. Overview

A perceptual test was conducted to fragment the music into

small segments, a step necessary for subsequently extracting

the relevant motifs therein and building the WM regressor of

responses to motif repetitions. Next, the regressor was

correlated against the fMRI brain signal at each voxel in the

brain to retrieve the relevant brain areas involved in WM for

music. To ensure that the observed activations were due to

memory formation and retrieval rather than sensory pro-

cessing, time series of perceptually validated acoustic features

of the music were also added to the analysis as nuisance re-

gressors to remove the variance accounted for by them from

the brain responses and isolate the effect of WM processes. In

the next sections we first proceed to describe the perceptual

experiment aimed at retrieving the motifs in the piece, fol-

lowed by an in-depth explanation of the method adopted in

the fMRI analysis.

2.2. Stimulus material

The musical piece used in the experiment was Adios Nonino, a

tango piece by the Argentinian composer Astor Piazzolla

(1921e1992) lasting about 8 min. The recording is of an

emblematic performance during Astor Piazzolla’s Sexteto

European tour (1989e1990) recorded live at MAD (Moulin a

Danses), Lausanne, Switzerland on November 4th, 19894. The

piece, written in October 1959while in NewYork inmemory of

his father, Vicente “Nonino” Piazzolla, a few days after his

father’s death, has proven to be one of Piazzolla’s most well-

known compositions. However, only four out of 26 partici-

pants reported to be familiar with the piece, hence approxi-

mately replicating the frequency observed in the fMRI

listening experiment (two out of eleven). In music, WM gets

activated even if participants already have a LTM of the piece

overall (Jonides et al., 2008), thus this did not represent a

drawback to the study.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9 257

Adios Nonino is characterized by two well-defined, salient

motifs. Musical recursion is a specific feature ofWestern tonal

classical tradition, in which naturally occurring repetition and

variation patterns in the music constitute a distinct attribute,

hence allowing the study of motif tracking in the context of

real music. Moreover, Adios Nonino is a very challengingmusic

stimulus rich in timbre and modulations, and particularly in

this very vibrant and dynamic performance, exhibiting strong

shifts in tempo, dynamics, and rhythm, which is in effect

representative of a complex auditory scene.

2.3. Perceptual experiment

As an unfolding sequence of temporal events, all musical

properties are at the mercy of the human WM thresholds,

limiting our capacity for retention. Segmentation of the un-

differentiated auditory stream in smaller units is a funda-

mental process in music perception, necessary for the

coherent representation of themusical flow due to our limited

memory capacity. Thus, by means of a perceptual test, the

music stimulus was fragmented into small segments, from

which motivic units (expected to trigger WM) were identified

and used to build the WM regressor. Participants’ task was to

segment the piece in real-time as they listened to it.

2.3.1. ParticipantsA total of 26 participants (female ¼ 11; age range ¼ 18e56;

mean age ¼ 30 � 8 SD; musical training: starting age ¼ 11 � 5

SD years; mean total training ¼ 11 � 8 SD years; mean

listening time ¼ 7 � 3 SD hours/week; mean practicing time:

30 � 25 SD min/day) took part in the experiment. We decided

to recruit both professional and amateurmusicians capable of

performing this music-specific task. The motivation to use

musicians was based on the evidence for a distinction be-

tween musicians and non-musicians regarding musical WM

(Schulze et al., 2011), to avoid musicianship becoming an

additional factor. Indeed, musicians have also been shown to

have superior task performance and larger blood-oxygen-

level dependent (BOLD) responses in attention and cognitive

control-related networks during a WM task for musical

sounds than non-musicians (Pallesen et al., 2010). Overall,

brain responses to music have been observed to be enhanced

in musicians (Brattico, Bogert, & Jacobsen, 2013; Schneider

et al., 2002; Tervaniemi, 2009).

2.3.2. ProcedureThe segmentation protocol was conducted on a computer

running a Max/MSP2 environment-based platform. Partici-

pants were instructed to (1) fill in a personal background in-

formation form; (2) listen to the entire piece of music from

beginning to end once without performing the segmentation

task; (3) perform the segmentation task in real-time (to avoid

fatigue effects, listeners did not segment the music from the

beginning to the end; instead ten3 long sections of the music

were presented to the participants twice in randomized

order); (4) answer general questions about their performance.

Participants were instructed to click with a computer mouse

on a grey button (labelled “CLICK TO PLAY THE MUSIC EX-

CERPTS”) to trigger the playback of one random excerpt at a

time and immediately proceed to segment it by clicking on the

red button beside it (labelled “CLICK TO SEGMENT THEMUSIC

EXCERPTS”) every time they heard a boundary in the music,

i.e., the end or beginning of a segment. Clicks were logged as

time coordinates (in milliseconds) at which they occurred for

further analysis. The segmenting task was defined to the

participants in terms of cutting the excerpts into small units

where appropriate boundaries were found while the music

was playing. The criteria to define these units were up to the

participants.

2.3.3. ResultsThere was great variation among participants’ number of

clicks (mean¼ 60� 38 SD; range¼ 15e137), but the location of

the clicks (segmentation points) was consistent: participants

that clicked a greater number of times only revealed a more

fine-grained segmentation within hierarchical boundaries

set by participants that performed fewer clicks. Subse-

quently, Kernel Density Estimation (KDE; Silverman, 1986)

was used to estimate the unknown underlying segmentation

probability distribution from the data, whereby participants’

segmentation points were merged into a time series (see

Appendix A). The estimate was computed using a Gaussian

kernel function with a standard deviation of one second,

which was found to yield the optimal compromise between

variance and bias (see Fig. 1). Next, maxima in the estimated

segmentation probability function were extracted by

applying different density thresholds. Sixty-four peaks (seg-

mentation points) were found to be an appropriate number

for the piece lasting approximately eight minutes and thus

yielding segments with an average length of 7 sec, which is

in accordance with short-term memory (STM) capacity re-

strictions. The resulting segmentation points from the anal-

ysis are visualized in Fig. 2.

2.3.4. MOTIF identificationFrom the sixty-four segments we proceeded to identify the

main motifs plus their repetitions and variations, which are

the building blocks of the stimulus, thereby discarding the

remaining musical content, of a more introductory, transi-

tional or developmental nature. Two main salient motifs and

their occurrences were identified: motif A and motif B (see

Fig. 3 for a schema of the temporal sequence of motifs in the

piece). A short sequence of repetitivemusical segments (#45 to

#48) deviating rhythmically and melodically from the main

two motifs was observed. This was included in the WM hy-

pothesis as a third motif (C) assuming it would trigger WM.

Similarly as proceeded formotifs A and B, the firstmotif of this

sequence (segment #45) was discarded and the rest was

included in the regressor as part of the ‘on’ condition.

Motif A is characterized by slow tempo, rubato, expres-

siveness, and cadential quality, whereas motifs B and C are

fast with a clear beat. With regards to timbre, different oc-

currences of motif A’s melody are featured by the piano, cello

and bandoneon, whereas in case of motif B and C, melody is

always featured by the bandoneon. Motif A has hefty tempo

changes, so it is expanding and retracting constantly, whereas

the tempo for motif B and C remains stable throughout the

piece. Motif repetitions vary in melodic contour, display or-

naments (especially motif A), and are continuously trans-

posed (except for motif C due to its brevity).

0 60 120 180 240 300 360 420 480time(sec)

Kernel density estimation

Fig. 1 e Probability distribution of the participants’ segmentation points (clicks), estimated with KDE.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9258

2.4. FMRI experiment

2.4.1. FMRI data acquisition2.4.1.1. PARTICIPANTS. Eleven healthy participants with no his-

tory of neurological or psychological disorders, with formal

musical training (styles: classical ¼ 5, folk ¼ 2, pop/rock ¼ 4;

instruments: string ¼ 4, percussion ¼ 3, wind ¼ 2,

keyboard ¼ 2; mean starting age ¼ 9.1 � 3.4 SD years; mean

total training ¼ 16.1 � 6 SD years; mean practice

time ¼ 2.5 � 1.2 SD hours/day) took part in the fMRI experi-

ment (females ¼ 5; age range ¼ 19e31; mean ¼ 23.2 � 3.7 SD;

handedness: 9 right-handed, 2 n/a). The present fMRI dataset

has been used by Alluri et al. (2012) in a previous study.

2.4.1.2. TASK SPECIFICATION. Participants were instructed to

listen to the stimulus (average sound level ¼ 80 dB) with

gradient noise attenuating (w30 dB) headphones (plus extra

attenuation via cotton inserted in the headset), and remain

still and relaxed as they underwent fMRI scanning while

0 60 120 180 240−.8

−.6

−.4

−.2

0

.2

.4

.6

.8Perceptual segme

time

ampl

itude

Fig. 2 e Segmentation of the music according to the participant

cadence and the applause from the live performance, was disca

keeping their eyes open. The study protocol proceeded upon

acceptance by the ethics committee of the Coordinating Board

of the Helsinki and Uusimaa Hospital District.

2.4.1.3. FMRI IMAGES. Functional images were acquired at the

3T scanner (3.0 T Signa VH/I General Electric) at the Advanced

Magnetic Imaging (AMI) Centre (Aalto University, Espoo,

Finland) with a temporal resolution of 2 sec. Thirty-three

oblique slices (thickness ¼ 4 mm, interslice skip ¼ 0 mm)

providing whole-brain coverage were imaged (echo

time ¼ 32 msec; flip angle¼ 75�) per participant, with a field of

view of 200 � 200 mm and a matrix size of 64 � 64, using a

single-shot gradient echo planar imaging (EPI) pulse sequence.

To measure BOLD changes, each EPI acquisition was run for

256 volumes (the first 4 volumes were later discarded to allow

for the magnetization to stabilize) per participant. T1-

weighted structural images (field of view ¼ 260 � 260 mm;

matrix ¼ 256 � 256; thickness ¼ 1 mm; interslice skip ¼ 0 mm)

were also collected for individual coregistration.

300 360 420 480 540

ntation of Adios Nonino

(sec)

s’ KDE. The last segment, corresponding to the final long

rded.

Fig. 3 e Temporal evolution of only the motivic segments, used to build the binary WM regressor (represented below the

motifs diagram). The numbers above identify the number of the segment in the piece. The first OFF condition for motif B

extends to the first two motifs due to their undesirable short length for the haemodynamic delay to differentiate between

the two conditions (WM off and WM on).

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9 259

2.4.2. FMRI data analysis2.4.2.1. FMRI DATA PREPROCESSING. Functional MRI scans were

first partly preprocessed in SPM5 (Statistical Parametric

Mapping 5).5 For each participant low-resolution images were

realigned (translation and rotation corrections did not exceed

2 mm and 2� respectively), segmented into grey matter, white

matter, and cerebrospinal fluid (voxel-based morphometry e

VBM5 for SPM]; Ashburner & Friston, 2000), and registered to

the corresponding segmented high-resolution T1-weighted

structural images. These were in turn normalized to the MNI

(Montreal Neurological Institute; Evans, Kamber, Collins, &

MacDonald, 1994) segmented standard brain template using

a 12-parameter affine transformation. Functional imageswere

then blurred to best accommodate anatomical and functional

variations across participants aswell as to enhance the signal-

to-noise by means of spatial smoothing using a 6 mm full-

width-at-half-maximum Gaussian filter (see Alluri et al.,

2012 for more details on preprocessing steps).

The rest of the fMRI data processing was carried out in

Matlab using customized scripts developed by the present

authors. Following the spatial smoothing operation, the fMRI

signal was detrended by means of spline interpolation (using

six anchor points). In addition to subtracting the low fre-

quency fluctuations, removal of high-frequency fluctuations

in the data is a typical procedure aimed at maximizing the

signal-to-noise ratio (Jezzard, Matthews, & Smith, 2001). To

this end and following the detrending operation, temporal

filtering (Gaussian smoothing with kernel width ¼ 5 sec) was

applied to the fMRI signal.

Movement-related variance components in fMRI time se-

ries resulting from residual motion artefacts from the regis-

tration operation were defined in six variable parameters

(vector of x, y, and z translations and rotations about themajor

axes at each time point describing each voxel displacement

relative to the template image) and treated as nuisance terms

by regressing them out from each voxel time series.

2.4.2.2. STATISTICAL MODELLING. In the following we will explain

processing of the regressors of interest used for further

correlational analysis. Questions about orthogonality between

the regressor of interest and nuisance regressors are also dealt

5 Wellcome Trust Centre for Neuroimaging, London, UK (http://www.fil.ion.ucl.ac.uk/spm).

with, after which we proceed with the intrasubject and sub-

sequent group-level statistical analyses.

2.4.2.2.1. BUILDING THE WM REGRESSOR. To build the hypothesis

of WM, we selected those motifs thought to trigger WM,

namely all occurrences of motif A, B and C except the first

instance thereof (see Fig. 3). These repetitions and variations

constitute the ‘on’ condition, while the rest of the music

(including the first appearance of each motif) is assumed not

to elicit WM. The regressor is defined as a time series vector

(sampling rate¼ 1 Hz) of binary values having ones for the ‘on’

condition and zeros elsewhere. This regressor time series was

convolved with a canonical double-gamma haemodynamic

response function (HRF) in order to match the haemodynamic

response delay typical of BOLD signals, and downsampled to

.5 Hz to match the sampling rate of the fMRI scanner. This

regressor will be called ‘WM’ in the following.

Next, the first 13 samples (26 sec) of theWM regressor were

removed. Those corresponded to the first 4 volumes excluded

after the fMRI acquisition due to T1 stabilizing effects, and to

the subsequent 9 samples (matching the length of the HRF) to

avoid artefacts resulting from the convolution operation

(Alluri et al., 2012). Finally the last 12 samples (24 sec) of brain

volumes and regressors corresponding to the applause from

the live performance were also excluded. The final time series

had a length of 231 samples (see Fig. 4).

2.4.2.2.2. AC NUISANCE REGRESSORS. Time courses of perceptually

validated acoustic features from the Piazzolla stimulus (in the

following referred to as ‘AC’) provided by Alluri et al. (2012)

were added to the analysis as nuisance regressors to remove

the variation evoked by the AC in the brain signal and best

isolate theWMeffect in the brain. The AC (n¼ 6) correspond to

themain timbral, tonal, and rhythmic features in the stimulus

(Fullness, Brightness, Timbral complexity, Key clarity, Pulse

clarity, and Activity; see Alluri et al., 2012 for an in-depth

acoustic description). Similarly as with the WM regressor,

the AC underwent truncation of the first 13 and last 12 sam-

ples, convolution with a double-gamma HRF, detrending, and

downsampling, to match the fMRI data.

2.4.2.2.3. ORTHOGONALITY OF REGRESSORS. Orthogonality between

the AC and the regressors that could reduce the validity of the

analysis was examined. To this end, Pearson’s correlation

Fig. 4 e Binary and convolved WM regressor. The motif being featured in the ‘on’ condition can be seen on top. The y axis

shows the number of samples.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9260

coefficients (r) were calculated between theWM regressor and

each of the AC. These were mostly weak (r range ¼ .09e.21);

Corrected p-values [due to serial correlation present in the

time series data, estimated following Pyper and Peterman’s

(1998) method; see Appendix B for details] were large

(range ¼ .10e.46), indicating with confidence that the effect of

interest and the individual AC were independent.

We also tested whether a linear combination of the AC,

rather than individual components, could account for ourWM

regressor. Thus we performed multiple regression of the WM

regressor against the whole set of AC. For the WM regressor a

moderate r2 of .13 (corrected p ¼ .02) was obtained, suggesting

there is some variance in the WM regressor that can be

explained by a linear combination of the AC. Thus some of the

signal of interest may be removed, which cannot be avoided

using a naturalistic setting, where independence between the

motivic structure and the temporal evolution of the AC cannot

be controlled for.

2.4.2.3. STATISTICAL INFERENCE

2.4.2.3.1. FIRST-LEVEL ANALYSIS. A non-parametric approach was

followed in order to analyse both the fMRI data serieswith and

without the variance accounted for by the AC. Pearson’s cor-

relation coefficients were computed at participant level for

both regressors against each voxel time course. These r co-

efficients were Z-Fisher transformed to make the sampling

distribution approximately normal, and their significance was

corrected for serial correlation using Pyper and Peterman’s

(1998) method (see Appendix B).

2.4.2.3.2. GROUP-LEVEL ANALYSIS. Group-level analysis was car-

ried out on these results to produce a groupmap of significant

voxels. To this end, and following the approach described in

Lazar (2008), Fisher’s method (1950) was used to obtain the

pooled group-level map from the individual p-value images

(see Appendix C). The resulting groupmapwas thresholded at

a significance level of p � .001 (Z ¼ 3.29). The multiple com-

parisons problem was tackled by means of a cluster-wise

significance approach described in Ledberg, Akerman, and

Roland (1998) based on a Monte Carlo procedure to obtain an

approximation of the cluster size (CS) distribution at a

particular significance level, from which the critical CS

threshold can be selected (see Appendix D). Consequently, the

obtained corrected critical CS was of 17 voxels for a cluster-

defining threshold set at p � .001.

2.4.2.3.3. REGION ON INTEREST (ROI) EXTRACTION. MarsBaR v0.43

(http://marsbar.sourceforge.net) was used to extract the

regions falling under each resulting cluster. For each cluster,

the x y z coordinates (in MNI space) of the maximum voxel

Z-value was determined. Voxels identified as white matter

or voxels encroaching very small regions within the cluster

(k � 4 voxels) were discarded from the resulting ROI analysis.

Anatomical areas were determined using Automated

Anatomical Labelling (AAL; Tzourio-Mazoyer et al., 2002).

2.4.2.3.4. EFFECTIVE CONNECTIVITY ANALYSIS. To help clarify the

functional interactions between the identified brain regions

relevant for WM, effective connectivity analyses were per-

formed. To this end, and following a similar approach to that

used by Menon and Levitin (2005), PPI (Psychophysiological

Interactions) was employed. PPI is a straightforward and

intuitive approach aimed at investigating task-specific

changes in the connectivity pattern of selected seeds. The

selected seed voxels to undergo PPI analyses were chosen

from the resulting WM activation map based on their cluster

membership and Z-value: the voxel with the highest Z-value

from each cluster was chosen, and clusters were selected

based on brain area relevance to WM according to previous

research.

First, an interaction regressor (PPI regressor), as the

explanatory variable, was built. The PPI regressor is the

product of the WM regressor and the BOLD time series of

the seed being tested for connectivity, and it is positively

correlated with the seed time course at times when WM has

high values (is ‘on’) and negatively correlated with the seed

time course at times when WM is off. PPI analysis detects

precisely the regions that are more connected with the seed

voxel when WM is active than at other times (and vice

versa: regions that are more strongly connected with the

seed voxel when WM is off). Following this, multiple

regression was performed on the participants’ whole brain

responses using the PPI regressor as explanatory variable.

To avoid including either task-driven correlating voxels or

voxels with a similar activation pattern as the seed voxel in

the analyses, both the WM regressor and seed voxel time

series were added in the regression equation as nuisance

terms. The resulting beta coefficients were then averaged

across participants, and their significance was assessed by

means of permutation tests (number of trials: 150,000). The

resulting map of beta coefficients was subsequently

thresholded at a significance level of p � .05, and then

intersected with the map of WM significant areas. The

purpose of the intersection was to examine seed connec-

tivity only with those brain areas that are relevant for WM

in our present study.

Fig. 5 e Selected transversal views of the distributed activated/deactivated brain areas by the WM regressor, with a

significance level of p £ .001 (corrected; Z [ 3.29) and a CS ‡ 17 voxels. Voxels are colour-coded red and blue for positive and

negative correlations, respectively.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9 261

3. Results

3.1. Correlation analysis

The results revealed the brain areas strongly responding to

repetition of motifs in the music, hypothesized to subserve

WM executive processes driven by music. Brain responses to

the WM regressor were predominantly right-lateralized. We

observed 6 and 12 positive correlating clusters in cerebellar

and cerebral regions respectively, and 5 clusters that corre-

lated negatively in response to motif repetitions. Transversal,

orthographic and lateral/mid-sagittal views of the resulting

map are shown in Fig. 5, Fig. 6, and Fig. 7, respectively. See

Table 1 for the listed clusters and brain regions obtained.

In the cerebellum, two clusters were observed in the left

posterior lobe (left crus IeII; vermis VIII, IX and lobule VIII, IX);

another cluster was identified in the right posterior lobe

(lobule VIII); and three clusters were located in the right

anterior lobe (vermis IIIeIV, V; vermis VI, VIII; and lobule III).

In the frontal cerebral lobe, five clusters were present. These

were located in the left middle frontal gyrus (MFG, BA9); right

MFG and superior frontal gyrus (SFG; BA10 and BA9), thus

revealing the recruitment of the right dlPFC; triangular and

orbital part of the right inferior frontal gyrus (IFG, BA47), thus

revealing the recruitment of the right ventrolateral prefrontal

cortex (vlPFC); opercular and triangular part of the right IFG

(BA44); and right supplementarymotor area (SMA, BA6). In the

temporal pole of the left middle temporal gyrus (MTG) one

cluster was found (BA38). Additionally, some subcortical

structures in the forebrain were revealed by correlational

analysis. The basal ganglia responded to the WM regressor

with a rightward bias: one cluster comprised the left putamen

and left caudate nucleus; two clusters were observed in the

right caudate nucleus, and an additional cluster was identified

in the right putamen. Significant activity was also found in

other two clusters: the left hippocampus, and right para-

hippocampal gyrus and amygdala.

Four additional clusters correlated negatively with theWM

regressor: two thereof were observed in the left hemisphere

(middle occipital gyrus e MOG and MTG [BA19]; and the

medial part of the SFG [BA10]). Another cluster was identified

in the right hemisphere (MTG), and an additional cluster

encroached on both hemispheres (gyrus rectus).

In order to observe the extent to which musical feature

processing networks might overlap with WM processes rele-

vant to music in the brain, correlational analysis was re-run

on the brain data without using the AC as nuisance re-

gressors. The inclusion of the AC variance did not have a

significant impact on the WM-driven activation, suggesting

brain processing of acoustic features does not overlap with

WM networks. The activation in the temporal pole of the left

MTG was absent in the responses including the acoustic

variance, as were the deactivations in the bilateral gyrus

rectus and right MFG-IFG. Five additional clusters were,

however, identified that were absent in the responses

excluding the acoustic variance: one cluster in the left IFG,

three clusters in the posterior lobe of the cerebellum, one in

the right superior temporal gyrus (STG), and a deactivation in

the MTG.

3.2. Effective connectivity analyses

PPI analyses were conducted on nine selected seeds with the

highest Z-value from different clusters in the resulting WM

map (see Table 2 for a list of the selected seeds). The clusters

were selected based on the relevance of the brain area to WM

as per previous research on WM (areas within the PFC and

basal ganglia). The hippocampus was also included due to its

novel recruitment in auditory WM in the present study, and

crucial function in LTM.

PPI analysis revealed (a) the modulatory effect of WM

(p � .05) in the effective connectivity pattern of the selected

seed voxels with other WM-relevant areas (see Fig. 8). Overall

we found connectivity with a larger number of areas during

Fig. 6 e Orthographic projection ofWM-driven activations at different maxima in the basal ganglia, hippocampus, dlPFC and

vlPFC: (a) left putamen, (b) right caudate nucleus, (c) left hippocampus, (d) right IFG, triangular part (BA47), and (e) right MFG

(BA10). Each coordinate label (in MNI space) indicates the value at the cross-section, shown in green lines.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9262

Fig. 7 e Left and right lateral ([a] [d]) andmid-sagittal ([b] [c]) views of the thresholded statistical map displaying positive (red)

and negative (blue) correlations with the WM regressor. The hippocampal activation is indicated in the left hemisphere.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9 263

the ‘off’ condition (i.e., periods when there are no motif rep-

etitions), suggesting thatmost of theWM-relevant areas seem

to function independently only during the WM process. We

contrasted this analysis with additional functional connec-

tivity analyses based on voxelwise correlation with the seed

voxels, which revealed a very different connectivity pattern,

based on anatomical proximity.

To further investigate the relationship between the effec-

tive connectivity patterns of these selected areas, agglomer-

ative hierarchical clustering using Ward’s minimum variance

method was performed using as input the PPI vectorized

spatial maps (averaged across participants and intersected

with the WM significant map) for the nine seeds. The metric

used was cosine distance (see Fig. 9 for the resulting dendro-

gram). Cluster analysis identified two main clusters with

cosine distance > 1, thus their voxelwise connectivity modu-

lation patterns correlated negatively with each other: (a) one

cluster comprises the right IFG and the right caudate nucleus

(seeds #3 and #6), and (b) another one consists of two

orthogonal subclusters (with cosine distance ¼ 1): (1) the right

IFG, the right putamen, and the left hippocampus (seeds #7,

Table 1 e Results of the region on interest (ROI) analysis on the cin SPM. Positive and negative correlations are reported. The signreports hemispheric location, within-cluster region size (k; i.e.,cluster, and its respective MNI coordinates. Anatomical areas wAnatomical Labelling (AAL; Tzourio-Mazoyer et al., 2002). Abbrevgyrus), IFG (inferior frontal gyrus); SMA (supplementary motor a

Left k Max Z x y z

Positive correlations

Crus IeII of cerebellum 56 4.02 �22 �84 �26 MFG (BA10), S

Putamen, caudate nucleus 35 4.19 �20 18 6 IFG, opercula

Vermis VIII, IX; lobule VIII,

IX of cerebellum

35 3.77 �2 �62 �34 SMA (BA6)

MFG (BA9) 34 3.82 �26 44 18 Caudate nucl

Hippocampus 19 3.85 �32 �34 �2 Vermis III, IV

Temporal pole, MTG (BA38) 18 4.80 �40 10 �30 Lobule VIII of

IFG, triangula

Parahippocam

Vermis VI, VI

Caudate nucl

Putamen

Lobule III of c

Negative correlations

MOG (BA19), MTG (BA19) 263 �4.73 �34 �82 36 MTG

SFG, medial part (BA10) 23 �4.06 �10 60 2 Gyrus rectus

Gyrus rectus (BA11) 11 �3.5 2 34 �18

#9, and #1), and (2) the right MFG, the left putamen, the right

caudate nucleus, and the left MFG (seeds #2, #4, #8, and #5).

Clustering using the Euclidean distance as metric gave iden-

tical results, denoting that the effective connectivity of the

nine seeds is similar not only in what brain areas are effec-

tively connected with them, but also in their absolute sizes.

In short, results from the PPI analysis and hierarchical

clustering converge: seeds that cluster together display

similar connectivity modulation patterns (most prominently

observed for seeds #7, #9, and #1). For the other clusters it

might not be as patent from the PPI spatial maps since these

show only the areas significant at the p � .05 level.

4. Discussion

We studied music-related WM in musicians using a natural-

istic non-standard procedure: (a) participants’ task was to

listen attentively to a piece of music while their brain re-

sponses were recorded; (b) a complex stimulus was used

containing strong shifts in tempo, timbre, dynamics, tonality

lusters obtained via the 26-connectivity scheme employedificance threshold was set to p [ .001 (Z [ 3.29). The tablenumber of voxels), peak Z-value per region within theithin the clusters were determined using Automatediations: MFG (middle frontal gyrus), MTG (middle temporalrea), MOG (middle occipital gyrus).

Right k Max Z x y z

FG (BA9) 184 4.61 32 50 20

r part (BA44); IFG, triangular part (BA44) 133 4.11 56 18 8

111 4.02 2 8 54

eus 81 4.22 12 10 18

eV of cerebellum 60 4.35 2 �44 �16

cerebellum 42 4.75 26 �44 �48

r part (BA47); IFG, orbital part (BA47) 29 4.07 48 34 �2

pal gyrus (BA34), amygdala 29 4.25 14 �6 �20

II of cerebellum 15 3.7 4 �60 �26

eus 14 3.99 16 22 8

13 3.95 24 20 �6

erebellum 12 4.17 16 �38 �26

14 �3.71 40 �64 18

(BA11) 11 �4.01 4 34 �20

Table 2 e Selected seeds for PPI analyses.

Seed N Brain area Abbreviations Z-value x y z

#1 Hippocampus L Hippo 3.85 �32 �34 �2

#2 MFG (BA10) R MFG 4.61 32 50 20

#3 IFG, opercular part (BA44) R IFG 4.11 56 18 8

#4 Putamen L Put 4.19 �20 18 6

#5 MFG (BA9) L MFG 3.82 �26 44 18

#6 Caudate nucleus R Cau1 4.22 12 10 18

#7 IFG, triangular part (BA47) R IFG 4.07 48 34 �2

#8 Caudate nucleus R Cau2 3.99 16 22 8

#9 Putamen R Put 3.95 24 20 �6

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9264

and rhythm, more representative of the complex auditory

scene environment our brains have evolved to respond to; (c)

activation of WM-related neural networks was studied by

tracking the temporal evolution of motivic repetition that

naturally occurs in Western tonal music, assumed to trigger

recognition-related WM; and (d) in order to fine-tune the

identification of WM-related function in the brain, the vari-

ance accounted for by the stimulus’ acoustic features was

pruned from participants’ brain responses.

The recruited regions are consistent with previous work,

and expand the areas implicated in WM for music. Areas

within the dlPFC (BA9, BA10) and vlPFC (BA47), typically

associated with executive control and active maintenance in

Fig. 8 e PPI analyses results (p £ .05) showing the modulatory ef

selected seeds: (1) left hippocampus, (2) right MFG (BA10), (3) rig

caudate nucleus, (7) right IFG (BA47), (8) right caudate nucleus,

with the seed voxel during motif repetitions (‘on’ condition) tha

connectivity with the seed voxel elsewhere (‘off’ condition) tha

frontal gyrus), STG (superior frontal gyrus), IFG (inferior frontal

(parahippocampal gyrus).

WM (D’Esposito, Postle, & Rypma, 2000) correlated with the

WM regressor. Some regions reported in Koelsch et al. (2009)

and Schulze et al. (2011) involved in verbal and tonal WM

were found (cerebellum, basal ganglia structures, and SMA).

However, other areas did not show up in our study (premotor

cortex, parietal lobe, inferior frontal sulcus e IFS). We did not

find the supramarginal gyrus (SMG), although it has been

present in previous studies using non-naturalistic pitch WM

tasks (Gaab et al., 2003; Zatorre, Evans, & Meyer, 1994). Simi-

larly, the STG seems to be important in short-term auditory

retention (Gaab et al., 2003; Zatorre et al., 1994; Zatorre &

Samson, 1991), but our analysis did not reveal this region,

most likely because of our novel methodological approach of

fect of motif repetitions in the effective connectivity of nine

ht IFG (BA44), (4) left putamen, (5) left MFG (BA9), (6) right

(9) right putamen. Red areas indicate stronger connectivity

n elsewhere, whereas green areas indicate stronger

n during motif repetitions. Abbreviations: MFG (middle

gyrus); SMA (supplementary motor area), PHG

Fig. 9 e Results of the agglomerative hierarchical clustering using as input the PPI spatial maps (averaged across

participants and intersected with the WM significant map) for the nine seeds. The distance metric used was cosine. See

Table 2 for abbreviations.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9 265

regressing out the variance explained by the stimulus’

acoustic features. Indeed, when this variance was included in

our analysis, one region of the right STG was activated.

Additionally, MTL structures, including the left hippocam-

pus, were identified in our study that had not been observed in

previous research on auditory WM. The recruitment of the

hippocampus in a WM study during continuous listening

supports previous findings on spatial and visualWM functions

(Gazzaley, Rissman, & D’Esposito, 2004; Nichols, Kao,

Verfaellie, & Gabrieli, 2006). Considering the crucial role of

hippocampus in LTM (c.f., e.g., Eichenbaum, 2012), it also raises

questions regarding hippocampal function. As indicated by

the effective connectivity and hierarchical clustering analyses

performed: (a) the hippocampus functions independently of

the other WM-relevant areas during the motif recognition

process; and (b) the effective connectivity of the hippocampus

with certain WM-relevant areas (dlPFC, cerebellum and SMA)

is modulated by motif repetitions. A network comprising the

dlPFC and the hippocampus was found by Ranganath, Cohen,

and Brozinsky (2005) to mediate successful LTM formation

during a WM task involving delayed recognition of line draw-

ings of complex objects, supporting the hypothesis that MTL

structures contribute to bothWMand LTM. Similarly, previous

research has found that activity in both MTL structures

and PFC successfully predicts subsequent LTM formation

(Davachi, Mitchell, & Wagner, 2003; Otten & Rugg, 2001).

Evidence for this overlap between WM and LTM neuroan-

atomical structures draws mainly on animal studies

(Friedman & Goldman-Rakic, 1988; Zola-Morgan & Squire,

1986), but also on fMRI studies with humans (Cabeza, Dolcos,

Graham, & Nyberg, 2002; Davachi & Wagner, 2002). Karlsgodt

et al. (2005) stressed this overlap between WM and LTM

when they identified the hippocampus as being selectively

active during WM, and indicated that its engagement in WM

may derive from its noted role in recognition/recollection. The

authors consequently postulated that hippocampal activation

in WM would reflect episodic aspects of information being

maintained. Thus, similar to its function in LTM, the hippo-

campus may be involved in recollecting the specific motifs

that are maintained during the listening of the music, since

this is the main aspect defined by the designed regressor.

Disentangling whether the hippocampal activity in our

study is exclusively related to WM, or both WM and LTM is an

intricate question that would need further investigation using

a targeted design to be answered. However, it is reasonable to

assume that the continuous listening of a piece of music

characterized by highly distinct musical motifs may have

imprinted in the participants an overall memory of the piece.

Thus, we hypothesize the hippocampal activation to also be

related to LTM formation, enabled by the use of a realistic

listening condition.

With regard to lateralization, studies on verbal WM have

found support for a left hemispheric pattern (Wager & Smith,

2003). However, we observed in our study a rightward bias in

response to the music-related WM effect. Likewise, right-

hemispheric predominance has been implicated in a study

of stream segmentation in natural settings which engaged a

ventral network (including the recruitment of vlPFC [BA47, 44/

45] for detecting salient events, and dorsal network (including

dlPFC [BA9]) for maintaining attention and updating WM

(Sridharan, Levitin, Chafe, Berger, &Menon, 2007). These same

regions were observed in the present study to be active in

music-driven WM, and in fact it seems sensible to consider

these segmentation processes as sharing brain circuits with

WM, since stream segmentation into perceptually meaningful

chunks is a necessary requirement for WM encoding.

Furthermore, tempo tracking structures inmusic are bilateral,

whereas similar structures in speech are predominantly left

lateralized (Levitin & Tirovolas, 2009). This rightward trend

has been dominant in the literature on memory-related ex-

periments with music (e.g., Zatorre et al., 1994; Griffiths,

Johnsrude, Dean, & Green, 1999).

In what follows, each of the main anatomical areas found

in the present study is discussed in more details.

4.1. Neuroanatomy of WM for music

4.1.1. CerebellumCerebellar activation to motif repetitions might be related to

the temporal processing of the stimuli used (Mathiak,

Hertrich, Grodd, & Ackermann, 2004). In a study by Leung

and Alain (2011) right cerebellar activity in the tonsil (lobule

XI), culmen (vermis IVeV) and pyramis (vermis VIII) was

revealed in a WM task relevant to location and category. FMRI

and tractography analysis of diffusion weighted MRI images

by Salmi et al., (2010) suggest that cerebellar areas subserve

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9266

not only motor but cognitive processes. More specifically, the

posterior lobe of cerebellum (crus IeII) seems to be connected

to the lateral prefrontal areas activated by cognitive load in-

crease, suggesting this cerebellar area to subserve cognitive

functions during demanding tasks implicated in the optimi-

zation of the response speed. In Alluri et al. (2012), increased

activation in cerebellar areas including declive (vermis VI) and

pyramis (vermis VIII) was found to correlate with high values

of the acoustic features fullness and activity in the Piazzolla

stimulus. Thus neural substrate underlying these acoustic

features might aid in the encoding of the music in WM. In

Schulze et al. (2011), musicians’ right cerebellum responded to

verbal and tonalWM,while the left cerebellumwas selectively

recruited in response to tonal WM.

4.1.2. Prefrontal cortexThe PFC, in cooperation with other brain areas, seems to

subserve the common underlying process to its many pro-

posed functions (i.e., attention, memory, planning; Fuster,

2001): temporal integration, a uniform requirement of all

WM tasks, as well as of many others (Duncan & Owen, 2000).

Regions in the right vlPFC (BA47) and predominantly right

dlPFC (BA9 and BA10) were observed active during motif rep-

etitions in the music. These regions correspond to areas that

within the PFC have been consistently found active in tasks

requiring executive functions (Kane & Engle, 2002), with the

IFG reflecting WM aspects in the integration of information

over time (Nan, Knosche, Zysset, & Friederici, 2008). Activity in

BA47 has in the context of music been linked with the pro-

cessing of musical temporal structure, specifically with

extracting the correct temporal information in a sequence

(Chen, Penhune, & Zatorre, 2008).

Another prefrontal cortical area found to be active during

motifs’ repetitions pertains to an area in the opercular and

triangular part of the right IFG (BA44), considered to be the

right homologue of Broca’s area. In the left hemisphere, the

role of Broca’s area as a syntactic processor or as a WM

resource in sentence comprehension has been debated over

the last 30 years and remains controversial (Rogalsky,Matchin,

& Hickok, 2008). In the context of music, fMRI studies (Koelsch

& Siebel, 2005) using chord sequence and melody paradigms

have linked music syntactic processing with a predominantly

right activation in the opercular part. Activation of this region

has been likewise observed in WM for pitch (Koelsch & Siebel,

2005; Zatorre et al. 1994) as well as in music-driven rhythmic

tasks or while playing music (Peretz & Zatorre, 2003).

4.1.3. SMAA predominantly right-lateralized region in the SFG (SMA,

BA6) was found active in response to the WM condition. This

area has been observed active while continuously updating

information andmaintaining temporal order in WM (Wager &

Smith, 2003). The role of the SMA also seems to be crucial in

the processing of temporal structure. Distributional differ-

ences of haemodynamic activations emerge along a ros-

trocaudal axis within the SMA depending on different types of

temporal processing (Schwartze, Rothermich, & Kotz, 2012).

Additionally, this region was reported in Schulze et al. (2011)

as part of the common core areas underpinning verbal and

tonal WM in musicians and non-musicians.

4.1.4. Basal gangliaThe bilateral putamen and caudate nucleus, mainly right-

lateralized, correlated with the WM regressor. The role of the

basal ganglia has been shown to be primarily related to motor

skills (DeLong et al., 1984), but it has been shown to be also

involved in implicit and motivational learning and memory

(Packard & Knowlton, 2002). During tonal rehearsal in non-

musicians, Koelsch et al. (2009) observed activation in basal

ganglia structures (right caudate nucleus and left globus pal-

lidus). Similarly, in Schulze’s (2011) study on verbal and tonal

WMpartof thebasal ganglia (leftputamen) respondedtoverbal

and tonal WM (though more strongly in the latter) exclusively

in musicians. Likewise, Pallesen et al. (2010) observed activa-

tion in the bilateral putamen during a musical chord n-back

task in both non-musicians andmusicians, showing the latter

group larger BOLD responses in the right putamen.

Upon the recruitment of both basal ganglia and prefrontal

cortical regions in our study, it is important to stress the long-

known dense interconnections between basal ganglia and PFC

(Alexander, DeLong, & Strick, 1986; Miller & Cohen, 2001). In

relation toWM, it has been proposed that activity in prefrontal

cortical areas along with basal ganglia structures is respon-

sible for the filtering of relevant information into WM (Mcnab

& Klingberg, 2007). Indeed, PFC and basal ganglia (along with

the midbrain dopamine nuclei) appear to be the anatomical

structures most deeply implicated in WM (Fuster, 1999;

Goldman-Rakic, 1995; Gruber et al., 2006).

4.1.5. Temporal pole of the MTGA region in the temporal pole of the MTG (BA38) was found to

be active in response to the WM regressor. This area has been

thought to contribute to the overall functioning of a semantic

WM network (Martin & Chao, 2001). However, its functioning

is not well understood. A review study by Olson, Plotzker, and

Ezzyat (2007) indicates that this area concerns both social and

emotional processes, face recognition and theory of mind.

They suggest that the temporal pole binds complex, percep-

tual inputs to visceral emotional responses.

4.1.6. MTL structures: hippocampus, parahippocampal gyrusand amygdalaThe MTL comprises a critical system of anatomically con-

nected regions, including the hippocampus, amygdala, and

adjacent parahippocampal regions that have a critical role in

long-term declarative memory (Squire et al. 2004). The hip-

pocampus has been known to be crucial in the formation of

new long-term memories based on episodic or autobio-

graphical events (Cohen & Eichenbaum, 1993; Scoville &

Milner, 1957; Squire & Schacter, 2002), as well as a novelty

detector (VanElzakker, Fevurly, Breindel, & Spencer, 2008). If

we consider that a) the hippocampus is widely known to be

critical in the consolidation of long-term memories; b)

emotionally arousing experiences modulate LTM consolida-

tion (McGaugh, 2013); and c) the naturalistic continuous

listening of a real piece of music has the potential to trigger an

emotionally arousing response from the listeners; we could

propose this hippocampal activation to be enabled by the use

of such realistic listening condition. In other words, one could

speculate that the current naturalistic paradigm has evi-

denced the formation of long-term memories for the music.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9 267

Although in the animal literature the hippocampus is

strongly linked toWM and also to LTM (Friedman & Goldman-

Rakic, 1988; Zola-Morgan & Squire, 1986), this region has so far

and to our knowledge not been reported in auditory WM

studies, and thus its recruitment in the presentwork stands as

a novel finding in auditory WM. Hippocampal activation dur-

ing WM has however been found in human studies in other

modalities. For instance, Nichols et al. (2006) did find the

hippocampus to have an important role during maintenance

in face recognition. Likewise, Gazzaley et al. (2004) found

activation in the hippocampus, parahippocampus and

amygdala also during a face recognition WM task. Other

studies linking hippocampus and WM have been mentioned

earlier in the discussion.

4.1.7. Negative correlating areasA sizable area that correlated negatively in response to motif

repetitions was the left MOG (BA19) encroaching into the left

MTG, an area implicated in visual processing and known to be

deactivated during both auditory and visual processing

(Laurienti et al., 2002). Additional areas that showed deacti-

vation during the appearance of the motifs in the music were

themedial part of the left SFG (BA10), the bilateral gyrus rectus

(BA 11) and a smaller region in the right MTG.

5. Conclusions and further research

As investigated naturalistically in our present study, WM

recognition for music predicted by motif repetitions seems to

result from the ad-hoc activation of motoric, cognitive, and

limbic areas. Hence the contribution of different areas seems

to be relevant to mnemonic processing. However, inferences

about recruitment of specific areas only tell about their

engagement, and not requirement, in the process under

investigation.

To further investigate the novel hippocampal activation in

connection with the hypothesis of LTM formation of the

stimulus’ motifs, participants could perform a short listening

test after a week from the fMRI session, thus testing their LTM

of the piece.

To deepen investigation into questions of brain speciali-

zation,musicians’ responses could be contrasted against non-

musicians’. Evidence suggests that musicians possess an

enhanced ability to retrieve, monitor, and chunk information

over non-musicians (Chen et al., 2008), crucial processes for

WM efficiency which could be detected in the brain dynamics.

For instance, we might see not only differences in asymmetry

across groups for WM, but also within groups across different

conditions, i.e., due to different levels of complexity in the

music. For instance, although musicians tend to rely on the

left hemisphere for music processing to a greater extent than

non-musicians (Fujioka, Trainor, Ross, Kakigi, & Pantev, 2005),

possibly explained by a more consciously learned or analytic

approach to the musical input, complex music has been re-

ported to drive even trained musicians into strongly using

their ‘right brain’ (McGilchrist, 2010; Vollmer-Haase, Finke,

Hartje, & Bulla-Hellwig, 1998). Vollmer-Haase’s explanation

lies in the increased WM requirement for analysing the

complex musical material, while McGilchrist proposes it may

lie in the perceptually “new” experience of the music on

different hearings, due to the impossibility of attending to all

parts as a whole.

In the present study, the use of a naturalistic setting to

study musical WM recognition using the motivic repetitions

as our WM trigger proved viable, which should motivate

similar approaches to studying, for instance, verbal WM.

Similarly to how musical motifs are the iterated, recognizable

musical structures upon which a piece is built that render it

coherent, a linguistic discourse is characterized by a theme or

subject matter embodied by recurring ideas across the text.

We are just now beginning to unravel the systems problem

of WM for music, and so mechanisms of WM in the brain

remain an open issue. This study points in the direction of

exploring the functional brain topology of music-elicited WM

during naturalistic listening in musicians. In view of the

recruitment of hippocampal regions within the present study,

we cannot ignore another component to memory: emotion,

which is thought to play an influential role in the mnemonic

power of music as to how and what we remember. Evidence

on the robust integration of musical memory and emotion,

i.e., in autobiographical memories, has been reported (Ashley

& Luce, 2004; Eschrich, Munte, & Altenmuller, 2008; Janata,

2009; Jancke, 2008). However, research on how auditory WM

interacts with cognitive and emotional processes is still at an

early stage (cf. Pallesen, Brattico, Bailey, Korvenoja, & Gjedde,

2009).

Acknowledgement

This research was supported by the Academy of Finland

(Centre of Excellence program, project number 141106). I

would like to thank David Ellison and Emily Carlson for

proofreading this manuscript.

Supplementary data

Supplementary data related to this article can be found at

http://dx.doi.org/10.1016/j.cortex.2014.04.012.

r e f e r e n c e s

Alexander, G. E., DeLong, M. R., & Strick, P. L. (1986). Parallelorganization of functionally segregated circuits linking basalganglia and cortex. Annual Review of Neuroscience, 9, 357e381.

Alluri, V., Toiviainen, P., Jaaskelainen, I. P., Glerean, E., Sams, M.,& Brattico, E. (2012). Large-scale brain networks emerge fromdynamic processing of musical timbre, key and rhythm.NeuroImage, 59(4), 3677e3689.

Ashburner, J., & Friston, K. J. (2000). Voxel-based morphometry d

the methods. NeuroImage, 11(6), 805e821.Ashley, R., & Luce, K. (2004). Music, autobiographical memory,

and emotion. Emotion, 451e452.Brattico, E., Alluri, V., Bogert, B., Jacobsen, T., Vartiainen, N.,

Nieminen, S., et al. (2011). A functional MRI study of happyand sad emotions in music with and without lyrics. Frontiers inPsychology, 2, 308.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9268

Brattico, E., Bogert, B., & Jacobsen, T. (2013). Toward a neuralchronometry for the aesthetic experience of music. Frontiers inPsychology, 4.

Cabeza, R., Dolcos, F., Graham, R., & Nyberg, L. (2002). Similaritiesand differences in the neural correlates of episodic memoryretrieval and working memory. NeuroImage, 16, 317e330.

Cain, K., Oakhill, J., & Bryant, P. (2004). Children’s readingcomprehension ability: concurrent prediction by workingmemory, verbal ability, and component skills. Journal ofEducational Psychology, 96(1), 31e42.

Casey, B. J., Cohen, J. D., Jezzard, P., Turner, R., Noll, D. C.,Trainor, R. J., et al. (1995). Activation of prefrontal cortex inchildren during a nonspatial working memory task withfunctional MRI. NeuroImage, 2(3), 221e229.

Chen, J. L., Penhune, V. B., & Zatorre, R. J. (2008). Moving on time:brain network for auditory e motor synchronization ismodulated by rhythm complexity and musical training.Journal of Cognitive Neuroscience, 226e239.

Cohen, N. J., & Eichenbaum, H. (1993). Memory, amnesia and thehippocampal system. Cambridge, MA: MIT Press.

Davachi, L., Mitchell, J. P., & Wagner, A. D. (2003). Multiple routesto memory: distinct medial temporal lobe processes builditem and source memories. Proceedings of the National Academyof Sciences USA, 100, 2157e2162.

Davachi, L., & Wagner, A. D. (2002). Hippocampal contributions toepisodic encoding: insights from relational and item-basedlearning. Journal of Neurophysiology, 88, 982e990.

DeLong, M. R., Alexander, G. E., Georgopoulos, A. P.,Crutcher, M. D., Mitchell, S. J., & Richardsoon, R. T. (1984). Roleof basal ganglia in limb movements. Abstract HumanNeurobiology, 2(4), 235e324.

Duncan, J., & Owen, A. M. (2000). Common regions of the humanfrontal lobe recruited by diverse cognitive demands. Trends inNeurosciences, 10, 475e483.

D’Esposito, M., Postle, B. R., & Rypma, B. (2000). Prefrontalcortical contributions to working memory: evidence fromevent-related fMRI studies. Experimental Brain Research, 133,3e11.

Eichenbaum, H. (2012). The cognitive neuroscience of memory: Anintroduction. New York: Oxford University Press.

Engel, A. K., & Singer, W. (2001). Temporal binding and the neuralcorrelates of sensory awareness. Trends in Cognitive Sciences,5(1), 16e25.

Eschrich, S., Munte, T. F., & Altenmuller, E. O. (2008).Unforgettable film music: the role of emotion in episodic long-term memory for music. BMC Neuroscience, 9, 48.

Evans, A., Kamber, M., Collins, D., & MacDonald, D. (1994). AnMRI-based probabilistic atlas of neuroanatomy. InS. D. Shorvon (Ed.), Magnetic resonance scanning and epilepsy (pp.263e274). New York: Plenum Press.

Fisher, R. A. (1950). Statistical methods for research workers (11thed.). London: Oliver and Boy.

Friedman, H. R., & Goldman-Rakic, P. S. (1988). Activation of thehippocampus and dentate gyrus by working-memory: a2-deoxyglucose study of behaving rhesus monkeys. Journal ofNeuroscience, 8, 4693e4706.

Fujioka, T., Trainor, L. J., Ross, B., Kakigi, R., & Pantev, C. (2005).Automatic encoding of polyphonic melodies in musiciansand nonmusicians. Journal of Cognitive Neuroscience, 17,1578e1592.

Fuster, J. M. (1987). Single-unit studies of the prefrontal cortex. InE. Perecman (Ed.), The Frontal lobes revisited (pp. 109e120). NewYork: IRBN Press.

Fuster, J. M. (1999). Memory in the cerebral cortex: An empiricalapproach to neural networks in the human and nonhuman primate(p. 358). MIT Press.

Fuster, J. M. (2001). The prefrontal cortex d an update: time is ofthe essence. Neuron, 30, 319e333.

Gaab, N., Gaser, C., Zaehle, T., Jancke, L., & Schlaug, G. (2003).Functional anatomy of pitch memorydan fMRI study withsparse temporal sampling. NeuroImage, 19(4), 1417e1426.

Gazzaley, A., Rissman, J., & D’Esposito, M. (2004). Functionalconnectivity during working memory maintenance. Cognitive,Affective & Behavioral Neuroscience, 4(4), 580e599.

Georgiewa, P., Rzanny, R., Hopf, J. M., Knab, R., Glauche, V.,Kaiser, W. A., et al. (1999). fMRI during word processing indyslexic and normal reading children. NeuroReport, 10(16),3459e3465.

Goldman-Rakic, P. (1995). Cellular basis of working memory.Neuron, 14, 477e485.

Griffiths, T. D., Johnsrude, I., Dean, J. L., & Green, G. G. (1999). Acommonneural substrate for the analysis of pitch anddurationpattern in segmented sound? NeuroReport, 10, 3825e3830.

Gruber, A. J., Dayan, P., Gutkin, B. S., & Solla, S. A. (2006).Dopamine modulation in the basal ganglia locks the gate toworking memory. Journal of Computational Neuroscience, 20(2),153e166.

Hasson, U., Nir, Y., Levy, I., Fuhrmann, G., & Malach, R. (2004).Intersubject synchronization of cortical activity during naturalvision. Science (New York, N.Y.), 303(5664), 1634e1640.

Hickok, G., Buchsbaum, B., & Humphries, C. (2003). Auditory e

motor interaction revealedby fMRI: speech,music, andworkingmemory in area Spt. Journal of Cognitive Neuroscience, 673e682.

Janata, P. (2009). The neural architecture of music-evokedautobiographical memories. Cerebral Cortex, 19(11), 2579.

Janata, P., Tillmann, B., & Bharucha, J. J. (2002). Listening topolyphonic music recruits domain-general attention andworking memory circuits. Cognitive, Affective & BehavioralNeuroscience, 2(2), 121e140.

Jancke, L. (2008). Music, memory and emotion. Journal of Biology,7(6), 21.

Jezzard, P., Matthews, P. M., & Smith, S. M. (2001). Functional MRI:An introduction to methods. Oxford University Press.

Jonides, J., Lewis, R. L., Nee, D. E., Lustig, C. A., Berman, M. G., &Moore, K. S. (2008). The mind and brain of short-termmemory.Annual Review of Psychology, 59, 193e224.

Kane, M. J., & Engle, R. W. (2002). The role of prefrontal cortex inworking-memory capacity, executive attention, and generalfluid intelligence: an individual-differences perspective.Psychonomic Bulletin & Review, 9(4), 637e671.

Karlsgodt, K. H., Shirinyan, D., van Erp, T. G. M., Cohen, M. S., &Cannon, T. D. (2005). Hippocampal activations duringencoding and retrieval in a verbal working memory paradigm.NeuroImage, 25(4), 1224e1231.

Koelsch, S., Gunter, T. C., von Cramon, D. Y., Zysset, S., Lohmann,& Friederici, A. D. (2002). Bach speaks: a cortical ‘language-network’ serves the processing of music. NeuroImage, 17,956e966.

Koelsch, S., Schulze, K., Sammler, D., Fritz, T., Muller, K., &Gruber, O. (2009). Functional architecture of verbal and tonalworking memory: an FMRI study. Human Brain Mapping, 30(3),859e873.

Koelsch, S., & Siebel, W. A. (2005). Towards a neural basis ofmusic perception. Trends in Cognitive Sciences, 9(12), 578e584.

Laurienti, P. J., Burdette, J. H., Wallace, M. T., Yen, Y.-F.,Field, A. S., & Stein, B. E. (2002). Deactivation of sensory-specific cortex by cross-modal stimuli. Journal of CognitiveNeuroscience, 14, 420e429.

Lazar, N. A. (2008). The statistical analysis of functional MRI data.New York: Springer.

Ledberg, A., Akerman, S., & Roland, P. E. (1998). Estimation of theprobabilities of 3D clusters in functional brain images.NeuroImage, 8(2), 113e128.

Leung, A. W., & Alain, C. (2011). Working memory load modulatesthe auditory “What” and “Where” neural networks.NeuroImage, 55(3), 1260e1269.

c o r t e x 5 7 ( 2 0 1 4 ) 2 5 4e2 6 9 269

Levitin, D., & Menon, V. (2003). Musical structure is processed in“language” areas of the brain: a possible role for Brodmannarea 47 in temporal coherence. NeuroImage, 20(4), 2142e2152.

Levitin, D. J., & Tirovolas, A. K. (2009). Current advances in thecognitive neuroscience of music. Annals of the New YorkAcademy of Sciences, 1156(1), 211e231.

Martin, A., & Chao, L. L. (2001). Semantic memory and the brain:structure and processes. Current Opinion in Neurobiology, 11(2),194e201.

Mathiak, K., Hertrich, I., Grodd, W., & Ackermann, H. (2004).Discrimination of temporal information at the cerebellum:functional magnetic resonance imaging of nonverbal auditorymemory. NeuroImage, 21, 154e162.

McGaugh, J. L. (2013). Making lasting memories: remembering thesignificant. Proceedings of the National Academy of Sciences, 1e6.

McGilchrist, I. (2010). The master and his emissary: The divided brainand the making of the Western world. Yale University Press.

Mcnab, F., & Klingberg, T. (2007). Prefrontal cortex and basalganglia control access to working memory. NatureNeuroscience, 11(1), 103e107.

Menon, V., & Levitin, D. J. (2005). The rewards of music listening:response and physiological connectivity of the mesolimbicsystem. NeuroImage, 28(1), 175e184.

Miller, E. K., & Cohen, J. D. (2001). An integrative theory ofprefrontal cortex function. Annual Review of Neuroscience, 24,167e202.

Miyake, A., & Shah, P. (1999). Models of working memory:Mechanisms of active maintenance and executive control.Cambridge University Press.

Nan, Y., Knosche, T. R., Zysset, S., & Friederici, A. D. (2008). Cross-cultural music phrase processing: an fMRI study. Human BrainMapping, 29(3), 312e328.

Nichols, E. A., Kao, Y., Verfaellie, M., & Gabrieli, J. D. E. (2006).Working memory and long-term memory for faces: evidencefrom fMRI and global amnesia for involvement of the medialtemporal lobes. Hippocampus, 616, 604e616.

Olson, I. R., Plotzker, A., & Ezzyat, Y. (2007). The enigmatictemporal pole: a review of findings on social and emotionalprocessing. Brain: A Journal of Neurology, 130(Pt 7), 1718e1731.

Otten, L. J., & Rugg, M. D. (2001). When more means less: neuralactivity related to unsuccessful memory encoding. CurrentBiology, 11, 1528e1530.

Packard, M. G., & Knowlton, B. J. (2002). Learning and memoryfunctions of the Basal Ganglia. Annual Review of Neuroscience,25, 563e593.

Pallesen, K. J., Brattico, E., Bailey, C. J., Korvenoja, A., & Gjedde, A.(2009). Cognitive and emotional modulation of brain defaultoperation. Journal of Cognitive Neuroscience, 1065e1080.

Pallesen, K. J., Brattico, E., Bailey, C. J., Korvenoja, A., Koivisto, J.,Gjedde, A., et al. (2010). Cognitive control in auditory workingmemory is enhanced in musicians. PloS One, 5(6), 1e12.

Patel, A. D. (2003). Language, music, syntax and the brain. NatureNeuroscience, 6(7), 674e681.

Patel, A. D., Gibson, E., Ratner, J., Besson, M., & Holcomb, P. J.(1998). Processing syntactic relations in language and music:an event-related potential study. Journal of CognitiveNeuroscience, 10, 717e733.

Pereira, C. S., Teixeira, J., Figueiredo, P., Xavier, J., Castro, S. L., &Brattico, E. (2011). Music and emotions in the brain: familiaritymatters. PloS One, 6, 11.

Peretz, I., & Zatorre, R. J. (2003). The cognitive neuroscience of music.Oxford: Oxford University Press.

Pyper, B. J., & Peterman, R.M. (1998). Comparison ofmethods to accountfor autocorrelation in correlation analyses of fish data (pp. 2127e2140).

Ranganath, C., Cohen, M. X., & Brozinsky, C. J. (2005). Workingmemory maintenance contributes to long-term memoryformation: neural and behavioral evidence. Journal of CognitiveNeuroscience, 17(7), 994e1010.

Rogalsky, C., Matchin, W., & Hickok, G. (2008). Broca’s area,sentence comprehension, and working memory: an fMRIStudy. Frontiers in Human Neuroscience, 2.

Salmi, J., Pallesen, K. J., Neuvonen, T., Brattico, E., Korvenoja, A.,Salonen, O., et al. (2010). Cognitive and motor loops of thehuman cerebro-cerebellar system. Journal of CognitiveNeuroscience, 22(11), 2663e2676.

Schneider, P., Scherg, M., Dosch, H. G., Specht, H. J., Gutschalk, A.,& Rupp, A. (2002). Morphology of Heschl’s gyrus reflectsenhanced activation in the auditory cortex of musicians.Nature Neuroscience, 5(7), 688e694.

Schulze, K., & Koelsch, S. (2012). Working memory for speech andmusic. Annals of the New York Academy of Sciences, 1252(1),229e236.

Schulze, K., Zysset, S., Mueller, K., Friederici, A. D., & Koelsch, S.(2011). Neuroarchitecture of verbal and tonal working memoryin nonmusicians and musicians. Human Brain Mapping, 32(5),771e783.

Schwartze, M., Rothermich, K., & Kotz, S. A. (2012). Functionaldissociation of pre-SMA and SMA-proper in temporalprocessing. NeuroImage, 60(1), 290e298.

Scoville, W. B., & Milner, B. (1957). Loss of recent memory afterbilateral hippocampal lesions. Journal of Neurology,Neurosurgery & Psychiatry, 20, 11e21.

Silverman, B. W. (1986). Density estimation for statistics and dataanalysis. London: Chapman and Hall.

Snyder, B. (2000). Music and memory: An introduction. CambridgeMass: MIT Press.

Squire, L. R., & Schacter, D. (Eds.). (2002). Neuropsychology ofmemory. New York: Guildford Press.

Squire, L. R., Stark, C. E., & Clark, R. E. (2004). The medial temporallobe. Annual Review of Neuroscience, 27, 279e306.

Sridharan, D., Levitin, D. J., Chafe, C. H., Berger, J., & Menon, V.(2007). Neural dynamics of event segmentation in music:converging evidence for dissociable ventral and dorsalnetworks. Neuron, 521e532.

Steinbeis,N.,&Koelsch, S. (2008). Sharedneural resourcesbetweenmusic and language indicate semantic processing of musicaltension-resolution patterns. Cerebral Cortex, 18, 1169e1178.

Tervaniemi, M. (2009). Musiciansdsame or different? Annals of theNew York Academy of Sciences, 1169(1), 151e156.

Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F.,Etard, O., Delcroix, N., et al. (2002). Automated anatomicallabeling of activations in SPM using a macroscopic anatomicalparcellation of the MNI MRI single-subject brain. NeuroImage,15(1), 273e289.

VanElzakker, M., Fevurly, R. D., Breindel, T., & Spencer, R. L.(2008). Environmental novelty is associated with aselective increase in Fos expression in the output elementsof the hippocampal formation and the perirhinal cortex.Learning & Memory (Cold Spring Harbor, N.Y.), 15(12),899e908.

Vollmer-Haase, J., Finke, K., Hartje, W., & Bulla-Hellwig, M. (1998).Hemispheric dominance in the processing of J. S. Bach fugues:a transcranial Doppler sonography (TCD) study withmusicians. Neuropsychologia, 36(9), 857e867.

Wager, T. D., & Smith, E. E. (2003). Neuroimaging studies ofworking memory: a meta-analysis. Cognitive, Affective &Behavioral Neuroscience, 3(4), 255e274.

Zatorre, R. J., Evans, A., & Meyer, E. (1994). Neural mechanismsunderlying melodic perception and memory for pitch. TheJournal of Neuroscience, 14, 1908e1919.

Zatorre, R. J., & Samson, S. (1991). Role of the right temporalneocortex in retention of pitch in auditory short-termmemory. Brain, 114(6), 2403e2417.

Zola-Morgan, S., & Squire, L. R. (1986). Memory impairment inmonkeys following lesions limited to the hippocampus.Behavioral Neuroscience, 100, 155e160.