ORIGINAL PAPER
The Neural Bases of Word Encoding and Retrieval:
A fMRI-Guided Transcranial Magnetic Stimulation Study
R. Manenti
•
M. Tettamanti
•
M. Cotelli
•
C. Miniussi
•
S. F. Cappa
Received: 27 October 2009 / Accepted: 27 November 2009 / Published online: 12 December 2009
Ó Springer Science+Business Media, LLC 2009
Abstract There is evidence that the human prefrontal
cortex is asymmetrically involved in long-term episodic
memory processing. Moreover, abstract and concrete words
processing has been reported to differentially involve pre-
frontal and parietal areas. We implemented a two-stages
functional magnetic resonance imaging (fMRI)–repetitive
transcranial magnetic stimulation (rTMS) paradigm to
investigate the role of the dorsolateral prefrontal cortices
(DLPFCs) and parietal cortices (PARCs) in encoding and
retrieval of abstract and concrete words. Using this para-
digm we could select areas to be stimulated on the basis of
single-subject (SS) anatomical and functional data, inves-
tigating the usefulness of this integration approach. With
respect to fMRI, abstract and concrete words differed only
for a greater left fusiform gyrus activation for concrete
words. In turn, significant rTMS effects were found, but
only for the retrieval of abstract words. Consistent with
previous findings, repetitive stimulation of the right DLPFC
had a specific impact on episodic retrieval. Memory
retrieval performance was also disrupted when rTMS was
applied to the left PARC. Finally, we found a significant
positive correlation between the effect sizes of SS right
PARC activations for abstract word retrieval and the con-
sequent rTMS interference effects. Taken together these
data provide for the first time evidence that also the PARC
has a necessary role in episodic retrieval of abstract words.
Importantly, from a methodological perspective, our data
demonstrate that fMRI-guided rTMS with a SS approach
provides a powerful tool to investigate the neural under-
pinnings of cognitive functions.
Keywords rTMS Single-subject Memory
Combining Youngs
Introduction
While functional magnetic resonance imaging (fMRI) data
reveal the correlations existing between brain functions and
behavior, transcranial magnetic stimulation (TMS) is a
complementary technique that can address hypotheses
about the necessity of one or more brain areas to a par-
ticular aspect of cognitive performance (Walsh and Pasc-
ual-Leone 2003), and it has been extensively used to map
the flow of information across different brain regions
during the execution of a cognitive task (Walsh and
Rushworth 1999). The use of combined TMS–fMRI has
been therefore receiving a growing interest over the last
decade, with the development of both an online (TMS
applied at the same time of fMRI acquisition) and an off-
line (TMS and fMRI separated in time) approach. fMRI
R. Manenti S. F. Cappa
Vita-Salute San Raffaele University, Milan, Italy
R. Manenti (&) M. Cotelli C. Miniussi
IRCCS San Giovanni di Dio Fatebenefratelli, Via Pilastroni 4,
25125 Brescia, Italy
e-mail: rosa.manenti@cognitiveneuroscience.it
M. Tettamanti
Nuclear Medicine Unit, San Raffaele Scientific Institute,
Milan, Italy
M. Tettamanti S. F. Cappa
Division of Neuroscience, San Raffaele Scientific Institute,
Milan, Italy
M. Tettamanti C. Miniussi S. F. Cappa
National Institute of Neuroscience, Torino, Italy
C. Miniussi
Department of Biomedical Sciences and Biotechnologies,
University of Brescia, Brescia, Italy
123
Brain Topogr (2010) 22:318–332
DOI 10.1007/s10548-009-0126-1
can be followed by rTMS and this type of combination
might be particularly promising for the study of memory
and other cognitive functions, in which there is high
interindividual anatomo-functional variability. By submit-
ting each experimental subject to a fMRI investigation
before rTMS, it becomes possible to define the anatomical
site of rTMS on the subject-specific (SS) activations rather
than on the mean group activations. The importance of this
aspect has been confirmed by a recent methodological
study that highlighted an elevated discrepancy between the
effects induced by rTMS in cognitive tasks according to
target area selection (Sack et al. 2009). Sparing et al.
addressed this question by evaluating the accuracy and
efficiency of different localization strategies for the pri-
mary motor cortex and found the highest accuracy with SS
fMRI-guided stimulation (Sparing et al. 2008). Coil posi-
tioning becomes even more important when applying TMS
to ‘‘silent’’ areas (i.e., areas in which the stimulation does
not induce a visible behavioral effect, such as muscle
contraction) that are typically involved in higher-order
cognitive functions, including memory.
In the last few years a number of studies have been
conducted to study memory using fMRI-guided stimula-
tion. Herwig et al. (2003a) used for the first time a com-
bined fMRI and rTMS approach to study the functional
role of premotor and parietal areas during phonological
rehearsal. The results of this study showed that the pre-
motor cortex was the only necessary brain region for
rehearsal, probably because of its implication in top–down
control processes (Herwig et al. 2003a). More recently,
Osaka et al. (2007) underlined the critical importance of
the left Brodmann area 9 in a reading span test. The
combined experimental approach has also been systemati-
cally applied by another research group to the study of
working memory (Feredoes and Postle 2007; Feredoes
et al. 2006, 2007; Postle et al. 2006). These series of studies
showed that only the stimulation of SS loci yielded an
accuracy effect (Feredoes and Postle 2007; Feredoes et al.
2007).
The fMRI literature on long term memory emphasized
the functional asymmetry of the frontal lobes. The HERA
(Hemispherical Encoding Retrieval Asymmetry) model
was based on the observation of the predominance of left
prefrontal cortex activation during encoding and of the
prevalence of right prefrontal cortex activation during
retrieval (Nyberg et al. 2000; Tulving et al. 1994). In the
last few years, it has become evident that the HERA model
may be an oversimplification, as both the nature of the
encoded material and the type of task also need to be taken
into account (Kelley et al. 1998; Wagner et al. 1998a;
Wagner et al. 1998b). Furthermore, differences between
encoding and retrieval are not restricted to the prefrontal
cortex. Neuroimaging studies have demonstrated the
involvement of a distributed neural network constituted by
the dorsolateral prefrontal cortices (DLPFCs), the medial
temporal lobes, the parietal cortices (PARCs), and the
precuneus (Buckner and Wheeler 2001; Cabeza et al. 2003;
Cabeza and Nyberg 2003; Fletcher and Henson 2001; Rugg
and Wilding 2000; Simons and Spiers 2003; Wagner et al.
1998a).
With respect to the nature of the encoded material, a
number of fMRI studies have shown different activations
for abstract and concrete words during episodic memory
tasks. A recent fMRI study analyzed the concreteness
effect during both encoding and recognition, and high-
lighted a greater effect of stimulus type compared to the
effect of process (Fliessbach et al.
2006). The intentional
encoding of abstract words compared to concrete words
was associated to a stronger activation of the left IFG,
while the recognition of concrete words compared to
abstract words was associated to a stronger activation of
the bilateral inferior parietal cortex and of the angular
gyrus (Fliessbach et al. 2006). The authors concluded that
the parietal areas showing a concreteness effect were more
engaged during the recognition task than during the
encoding task because of their role in item identification
(Weis et al. 2004).
The first study based on the ‘‘interference’’ approach
with repetitive TMS (rTMS) to assess the prefrontal cortex
functional asymmetries during encoding and retrieval was
conducted by Rossi et al. (2001). During encoding, subjects
were asked to discriminate between complex coloured
pictures (indoor–outdoor), whereas during retrieval sub-
jects had to recognize the previously seen pictures from
new ones. The right DLPFC was found to be crucial for the
retrieval of the encoded pictorial information, whereas the
left DLPFC was specifically involved during encoding.
Moreover, Sandrini et al. (2003) tested the influence of the
material by studying the encoding and the retrieval of
words, and found that the encoding of verbal material was
disrupted by both right and left prefrontal cortex stimula-
tion, whereas the retrieval was disrupted by right prefrontal
cortex stimulation. In conclusion, rTMS data concur with
neuroimaging data, in indicating that both the nature of the
material and the type of memory process may affect the
lateralization of frontal activation during memory tasks
(Fletcher and Henson 2001). With respect to posterior
brain areas, in a more recent work, Rossi et al. (2006)
investigated the functional role of the parietal cortex in
encoding and retrieval, and found that the activity of the
intraparietal sulci, unlike that of the DLPFC, is not caus-
ally engaged in the encoding and retrieval of visual
scenes. The authors suggested that parietal activations
accompanying the memorization processes reflect the
engagement of a widespread brain attentional network
(Rossi et al. 2006).
Brain Topogr (2010) 22:318–332 319
123
The main goal of our study was to adopt the SS com-
bined experimental approach, consisting in an fMRI-based
target area selection on an individual basis followed by
rTMS, to verify the usefulness of this approach in the study
of cognitive functions and, in particular, in the assessment
of the causal role of prefrontal and parietal areas in
memory encoding and retrieval of abstract and concrete
words.
Materials and Methods
Subjects
A group of 11 subjects [mean age = 30 years (range:
25–40); mean education = 18.5 years (range: 18–22)]
participated in the experiment. All subjects were native
Italian speakers and had normal or corrected-to-normal
visual acuity. Participants reported being free of neuro-
logical disorders or history of seizures. All were right
handed, with a mean score on the Edinburgh Handedness
Inventory (Oldfield 1971)of?88% (range =?44 to
?100%). Participants were informed about the procedures
and the possible risks of rTMS and informed consent was
obtained after a safety screening. The experimental meth-
ods had ethical approval from the locals Human Ethics
Committees (Ethics Committee of the San Raffaele Sci-
entific Institute, Milano, Italy and Ethics Committee of the
IRCCS San Giovanni di Dio Fatebenefratelli, Brescia,
Italy). Each subject first underwent an fMRI investigation,
followed by rTMS.
Stimuli
For the encoding condition, a total of 90 abstract words and
90 concrete words were selected from the ‘‘Corpus e Lessico
di Frequenza dell’Italiano Scritto (CoLFIS)’’ (Laudanna
et al. 1995). For the retrieval condition, we further selected
45 abstract and 45 concrete ‘‘new’’ words. The retrieval
word list thus consisted of half of the old words (45 concrete,
45 abstract) and 90 new words (45 concrete, 45 abstract).
The mean word length was 6.4 (±1.5) letters, and 2.7 (±0.6)
syllables. Abstract and concrete words were balanced for
word length and for variables known to influence memory
performance, i.e., word frequency (mean ± standard devi-
ation = 59 ± 20) and familiarity (mean ± standard devi-
ation = 6.0 ± 0.7). There were significant differences with
respect to concreteness (concrete = 6.0; abstract = 3.6;
t(1, 99) = 29.47, P \ 0.05) and imageability (concrete =
5.7; abstract = 3.3; t(1, 99) = 30.89, P \ 0.05) based on
CoLFIS (Laudanna et al. 1995).
In total, both the encoding and the retrieval word lists
included 180 words. Each word list (encoding and retrie-
val) was then divided into 9 blocks of concrete words and 9
blocks of abstract words, each block consisting of ten
words. Each encoding block was associated with one
retrieval block containing an equal number of ‘‘new’’ and
‘‘old’’ words in randomized order. Each encoding block
began with the instruction ‘‘Encode’’, while each retrieval
block began with the instruction ‘‘Old or new?’’. The
experimental paradigm also included a baseline condition,
with one baseline block associated to each encoding or
retrieval block. In the baseline blocks, the instruction
‘‘Count’’ was presented, followed by a cross remaining on
the screen for the entire block duration: the subject were
instructed to fixate the cross and to covertly count from one
to ten in a non-stop mode. The baseline condition served
both as a subtraction baseline for fMRI, and as a delay
period between the stimulations of different cortical areas
for rTMS. All blocks (encoding, retrieval or baseline)
lasted for 30 s, preceded by 4 s of instructions.
The blocks were grouped into three encoding and three
associated retrieval sessions for each word category
(abstract or concrete). Each encoding session included 3
encoding and 3 baseline blocks, while each associated
retrieval session comprised the 3 corresponding retrieval
blocks and 3 baseline blocks. This behavioural paradigm,
reflecting a 2 by 2 factorial combination of task (encoding
or retrieval) and word category (abstract or concrete),
resulted in four experimental conditions (encoding of
abstract words: EncA; encoding of concrete words: EncC;
retrieval of abstract words: RetA; retrieval of concrete
words: RetC). Since each encoding or retrieval block was
associated to one baseline block, four baseline conditions
were also constructed, reflecting one level of a further two-
levels experimental factor (experimental condition: mem-
ory or baseline). The four baseline conditions were formed
by dividing the baseline blocks according to the session
(encoding or retrieval of abstract or concrete words) in
which they were included (baseline for encoding of
abstract words: BasEncA; baseline for encoding of con-
crete words: BasEncC; baseline for retrieval of abstract
words: BasRetA; baseline for retrieval of concrete words:
BasRetC). This allowed us to treat the 4 baseline condi-
tions as orthogonal data sources and to compute interaction
and conjunction effects (see below). The three associated
encoding and retrieval abstract sessions were presented
consecutively followed (half of the subjects) or preceded
(the other half of the subjects) by the three associated
encoding and retrieval concrete sessions. Between each
encoding and retrieval sessions, we included a 5 min delay
to allow for working memory wash out and trace consoli-
dation. During this delay, the subjects simply rested in the
scanner listening to music (Fig. 1b).
320 Brain Topogr (2010) 22:318–332
123
Behavioural Task
The stimuli were presented using Presentation software
(Version 10.3, www.neurobs.com). All stimuli were pre-
sented in black lower-case letters on a white background.
During the fMRI phase, stimuli were projected from out-
side the magnet room onto a translucent screen placed at
the foot of the magnet bore. A mirror attached to the top of
the head coil allowed the participants to view the translu-
cent screen from inside the magnet. During both the fMRI
and rTMS studies, the subjects were told to read and
encode the presented words. After a delay, they were asked
to decide whether the presented word was from the pre-
viously encoded word list or not, by making a speeded
decision via a two-choice button press, using the right and
left forefingers (hands counterbalanced across subjects).
During both encoding and retrieval, each word was pre-
sented for 1500 ms in the centre of the screen and was
followed by a 1500 ms delay (Fig. 1a).
At the end of the experimental phase (both during the
fMRI and the rTMS phase) subjects were asked to fill in an
‘‘Encoding strategies questionnaire’’. This questionnaire
comprised twelve possible strategies that could be used
during the task and subjects had to assign a score from 1 to
ten (1 = never, 10 = always) to each strategy according to
how often they had used each strategy during the task. The
12 listed strategies were: (i) to use words’ initials, (ii) to
create sentences including some of the presented words,
(iii) to imagine the pictures corresponding to the presented
words, (iv) to repeat the words, (v) to create songs
including some of the presented words, (vi) to create
rhymes between the displayed words, (vii) to translate the
Fig. 1 Experimental design.
a Experimental task: during
encoding subjects had to read
the presented words and were
instructed to memorize them;
during retrieval they had to
decide if a word was old (O) or
new (N). b fMRI and rTMS
sessions: the displayed scheme
was repeated twice (one time for
each word category, i.e.,
abstract and concrete, during
both fMRI and rTMS phase).
Each encoding session was
composed of 3 encoding blocks
(E) alternating with 3 baseline
blocks (B); the same pattern was
also used for the retrieval
sessions, with alternating
retrieval blocks (R) and baseline
blocks. Coil icons represent the
sequence of rTMS stimulation
conditions, reported for clarity
in c, in the example of one of
the experimental subjects. The
combination between the kind
of stimulation applied during
the encoding and the one
applied during the retrieval (see
b) results in the experimental
condition (see c). The words
blocks were presented exactly in
the same order in fMRI and
rTMS experiments
Brain Topogr (2010) 22:318–332 321
123
words in a foreign language, (viii) to create associations of
words, (ix) to create a brief story including the presented
words, (x) to associate each word to a personal event, (xi)
to classify each word as easy/difficult, abstract/concrete,
positive/negative, etc., (xii) to imagine the words’ sound,
color, shape, etc.
fMRI Study
MRI Data Acquisition
MRI scans were acquired on a 3T Achieva Philips body
scanner (Philips Medical Systems, Best, NL) using an 8
channels-sense head coil (sense reduction factor = 2).
Whole-brain functional images were obtained with a T2*-
weighted gradient echo, echo-planar sequence, using
blood-oxygenation-level-dependent contrast. Each func-
tional image comprised 30 contiguous axial slices (4 mm
thick), acquired in interleaved mode, and with a repetition
time of 2000 ms (echo time: 30 ms; field of view:
240 mm 9 240 mm; matrix size: 128 9 128). The para-
digm was a block-design each participant underwent 12
functional scanning sessions (half encoding sessions and
half for concrete words). Baseline blocks were included in
all sessions. The duration of each session was 110 scans,
preceded by 10 dummy scans that were discarded prior to
data analysis.
For anatomical localization of brain activations and pre-
cise coil positioning during rTMS phase, we acquired one
high-resolution whole-brain structural T1 weighted scan
(resolution 1 mm 9 1mm9 1 mm) of each participant. In
order to maximize the coil localization accuracy, we
acquired 200 slices, covering on average the entire brain and
skull down to the midbrain. The normalized structural
images of all participants were then averaged in one single
image for visualization of group brain activations.
Data Analysis
Behavioural Data Behavioural data were analyzed eval-
uating both accuracy and reaction times (RTs) during the
retrieval sessions. It is important to consider that the
behavioural performance measured in this way may reflect
the functioning during both the encoding and the associated
retrieval sessions. For both accuracy and RTs data, we
conducted a paired t-Student test between abstract and
concrete words, verifying if, as predicted by the concrete-
ness effect, concrete words were recognized better and
faster than abstract words. We also calculated two-one-way
ANOVAs (one for each word category) with the block as a
factor, in order to exclude performance differences
between blocks.
Functional MRI Data Statistical parametric mapping
(SPM5, Wellcome Department of Imaging Neuroscience,
London, UK) was used for image realignment (Andersson
et al. 2001), normalization to the Montreal Neurological
Institute (MNI) standard space (yielding normalization
parameter files used for the inverse definition of individual
rTMS loci in the subject native brain space, see below),
smoothing by a 6 mm FWHM Gaussian kernel, and Gen-
eral Linear Model statistical analysis (Friston et al. 2002).
We adopted both: (i) a fixed-effects single subject analysis,
in order to identify individual stimulation loci; (ii) a two-
stage random-effects group analysis approach, in order to
identify group brain activations.
Fixed-effects Single Subject Analysis
At the first stage, the time series of each participant were
high-pass filtered at 67 s and pre-whitened by means of an
autoregressive model AR(1) (Friston et al. 2002). Global
normalization was performed to account for the global
between sessions effects confounding the comparisons
between experimental conditions. For each participant, we
modelled a 2 (experimental condition: memory or base-
line) 9 (tasks: encoding or retrieval) 9 2 (word category:
abstract or concrete) factorial design with 12 separate
sessions, reflecting conditions EncA, EncC, BasEncA,
BasEncC, RetA, RetC, BasRetA, BasRetC. We then calcu-
lated effect-specific interaction contrasts and, for the purpose
of the random effects group analysis only, a set of condition-
specific contrasts, each contrast including a weight of one
for a particular condition of interest and a weight of zero
for all the other conditions. The t-Student effect-specific
interaction contrasts included: (i) activations for encoding
versus retrieval of abstract words: (EncA - BasEncA)
-(RetA - BasRetA) inclusively masked by (EncA
- BasEncA), (ii) activations for encoding versus retrieval of
concrete words: (EncC - BasEncC) - (RetC - BasRetC)
inclusively masked by (EncC - BasEncC), (iii) activations
for retrieval versus encoding of abstract words: (RetA
-BasRetA) - (EncA - BasEncA) inclusively masked by
(RetA - BasRetA), and (iv) activations for retrieval versus
encoding of concrete words: (RetC - BasRetC) -(EncC
- BasEncC) inclusively masked by (RetC -BasRetC).
Second-level Random Effects Group Analysis
Approach
At the second stage of analysis, the contrast images
obtained at the SS level were used to compute a within-
subjects one way ANOVA assessing their significance at the
group-level (n = 11 participants). The ANOVA included
the set of all first-level contrast images, one image per
322 Brain Topogr (2010) 22:318–332
123
