Perisylvian Language Networks of the
Human Brain
Marco Catani, MD,1
Derek K. Jones, PhD,1,2
and Dominic H. ffytche, MD1
Early anatomically based models of language consisted of an arcuate tract connecting Broca’s speech and Wernicke’s
comprehension centers; a lesion of the tract resulted in conduction aphasia. However, the heterogeneous clinical presentations
of conduction aphasia suggest a greater complexity of perisylvian anatomical connections than allowed for in
the classical anatomical model. This article re-explores perisylvian language connectivity using in vivo diffusion tensor
magnetic resonance imaging tractography. Diffusion tensor magnetic resonance imaging data from 11 right-handed
healthy male subjects were averaged, and the arcuate fasciculus of the left hemisphere reconstructed from this data using
an interactive dissection technique. Beyond the classical arcuate pathway connecting Broca’s and Wernicke’s areas directly,
we show a previously undescribed, indirect pathway passing through inferior parietal cortex. The indirect pathway
runs parallel and lateral to the classical arcuate fasciculus and is composed of an anterior segment connecting Broca’s
territory with the inferior parietal lobe and a posterior segment connecting the inferior parietal lobe to Wernicke’s
territory. This model of two parallel pathways helps explain the diverse clinical presentations of conduction aphasia. The
anatomical findings are also relevant to the evolution of language, provide a framework for Lichtheim’s symptom-based
neurological model of aphasia, and constrain, anatomically, contemporary connectionist accounts of language.
Ann Neurol 2005;57:8–16
After Paul Broca’s report1
of a patient with disordered
speech production and a lesion of the left posterior inferior
frontal gyrus, Carl Wernicke2
described a posterior
temporal area devoted to auditory word processing.
Wernicke postulated the existence of a direct
connection between the two areas, and that a lesion of
this theoretical pathway would cause an aphasia characterized
by normal language comprehension and fluent
conversational speech but the inability to repeat
what had just been heard. In fact, a prominent pathway
connecting posterior frontal and superior temporal
lobes had already been described by Burdach and was
later confirmed by Dejerine who referred to the pathway
as Burdach’s arcuate fasciculus.3
Clinical evidence
for Wernicke’s theoretical “conduction” aphasia has accumulated
over many years.4
However, such descriptions
fall short of supporting Wernicke’s model. The
trouble is that many of the cases have lesions involving
both the arcuate fasciculus and its overlying cortex. In
some patients, the lesion extends directly into Wernicke’s
or Broca’s areas, whereas in others it extends to
the inferior parietal lobe, a region now recognized as
having an important linguistic function in its own
right.5
Wernicke’s conduction aphasia thus could be
equally well explained by cortical deficits as by a true
disconnection of Wernicke’s and Broca’s areas.
The advent of structural imaging allowed this issue
to be addressed in aphasic patients without recourse to
postmortem dissection. Such studies confirmed that
Wernicke’s conduction aphasia could occur with lesions
restricted to white matter tracts6,7
; however, it
also became clear that Wernicke’s original anatomical
formulation was incomplete. Wernicke’s model predicted
that lesions at any point along the course of the
arcuate fasciculus should result in an identical aphasia.
Yet, clinically, this is known not to be the case with
conduction aphasias forming a heterogeneous group
ranging from “Broca-like” to “Wernicke-like” defi-
cits.6,8
Clearly, Wernicke’s model requires revision.
What has been missing is a detailed description of the
arcuate fasciculus that takes into account developments
in linguistic anatomy since Burdach’s and Dejerine’s
original arcuate dissections.
In 1985, the modification of a conventional magnetic
resonance imaging (MRI) sequence permitted
quantification of the diffusion characteristics of water
molecules in vivo.9
Within cerebral white matter, water
molecules diffuse more freely along myelinated tracts
From the 1
Centre for Neuroimaging Sciences, Institute of Psychiatry,
King’s College London, London, United Kingdom; and 2
Section on
Tissue Biophysics and Biomimetics, Laboratory of Integrative Medicine
and Biophysics, National Institute of Child Health and Human
Development, National Institutes of Health, Bethesda, MD.
Received Jul 2, 2004, and in revised form Sep 13. Accepted for
publication Sep 16, 2004.
Published online Dec 13, 2004 in Wiley InterScience
(www.interscience.wiley.com). DOI: 10.1002/ana.20319
Address correspondence to Dr Catani, Institute of Psychiatry, PO
89, King’s College London, DeCrespigny Park, London SE5 8AZ,
United Kingdom. E-mail: m.catani@iop.kcl.ac.uk
ORIGINAL ARTICLES
8 © 2004 American Neurological Association
Published by Wiley-Liss, Inc., through Wiley Subscription Services
than across them10
; such directional dependence of diffusivity
is termed anisotropy. By combining anisotropy
data with the directional dependence of the hindrance
of water diffusion, it is possible to obtain estimates of
fiber orientation.11
This has led to diffusion tensor
“tractography”12–16
in which the three-dimensional
pathways of white matter tracts are reconstructed by
sequentially piecing together discrete and shortly
spaced estimates of fiber orientation to form continuous
trajectories. Although the fiber tracts dissected are
virtual and require further anatomical validation, the
technique has been successfully used to study human
brain thalamic connections,17
occipitotemporal con-
nections,18
and other white matter tracts,12,19,20
and
therefore appears to be a powerful technique with
which to re-examine the anatomy of the arcuate fasciculus
and the connectivity of perisylvian language areas.
Subjects and Methods
Subjects and Data Acquisition
Eleven subjects (mean age, 33.3 Ϯ 4.7 years) were recruited.
All gave written informed consent, and the study was approved
by the Institute of Psychiatry research ethics committee.
Because sex21
and handedness influence language laterality,
the study contained only right-handed male subjects.
Data were acquired on a GE Signa 1.5 Tesla LX MRI system
(General Electric, Milwaukee, WI) with 40mT/m gradients,
using an acquisition sequence fully optimized for diffusion
tensor magnetic resonance imaging (DT-MRI) of white matter,
providing isotropic resolution (2.5 ϫ 2.5 ϫ 2.5 mm)
and coverage of the whole head. The acquisition was gated
to the cardiac cycle using a peripheral gating device placed
on the subjects’ forefingers. Full details of the acquisition
sequence are provided in Jones and colleagues.22
After correction
for the image distortions introduced by the application
of the diffusion-encoding gradients, the diffusion tensor
was determined in each voxel following the method of Basser
and colleagues.11
After diagonalization of the diffusion tensor,
the fractional anisotropy,23
a scalar measure that reflects
the degree to which the diffusivity depends on the orientation
in which it is measured, was computed in each voxel.
Spatial Normalization and Averaging of Diffusion
Tensor Magnetic Resonance Imaging Data Sets
The DT-MRI data sets were spatially normalized to the anatomical
reference space defined by the MNI EPI template
supplied as part of the of the functional imaging analysis
software package SPM99 (Statistical Parametric Mapping;
Wellcome Department of Cognitive Neurology, London,
UK; www.fil.ion.ucl.ac.uk/spm/spm99.html). Spatial normalization
of tensor volumes is more complicated than that
for scalar volumes (eg, T1-weighted intensity maps), because
the orientations of the tensor must be correctly preserved.24
Once normalized, a mean of 10 diffusion tensors in each
voxel was computed to produce a mean diffusion tensor volume
data set. Full details of the spatial normalization procedure
and generation of an average DT-MRI volume have
been provided previously.25
Tractography: Definition of the Regions of Interest
and Three-Dimensional Visualization
A virtual dissection of the arcuate fasciculus of the left hemisphere
was performed using approaches with one and two
regions of interest (ROIs), as described in Catani and col-
leagues.19
Guided by color fiber orientation maps,26
an ROI
was defined on the fractional anisotropy map computed
from the average DT-MRI volume to encompass the horizontal
fibers lateral to the corona radiata and medial to the
cortex extending from Talairach z ϭ 22 to z ϭ 28 (Fig 1).
All fibers passing through this ROI (a one-ROI approach)
were reconstructed in three dimensions and visualized using
illuminated streamtubes.19
The fractional anisotropy volumes
were resliced in axial, coronal, and sagittal planes and displayed
in conjunction with tractography results to allow approximate
neuroanatomical location of the tract reconstructions.
A two-ROI approach was used to perform further
detailed dissection of the arcuate fasciculus, allowing us to
separate different sets of fibers within the arcuate bundle.
Here, two spatially separated regions are defined in the fractional
anisotropy volume, and all fibers passing through both
are visualized, as described earlier. The approach does not
constrain tracts to start and end within the defined regions,
only to pass through them. The delineation of regions used
in two-ROI dissections was guided by the results of the oneROI
dissection (see later).
Results
Figure 2 shows the reconstructed arcuate tract defined
using a one-ROI approach. The tract originated in the
cortex of the posterior part of the ventrolateral frontal
lobe, passing backward to the inferior parietal lobe
where it arched around the lateral fissure to terminate
in the posterior part of the superior and middle temporal
gyrus (the terms origin and termination are arbitrary
because tractography is blind to fiber direction).
The ventral portion of the frontoparietal tract segment
ran dorsal to the external capsule and to the insular
cortex (see Fig 2). These reconstructions correspond
exactly to the descriptions of the arcuate fasciculus
made by Burdach and Dejerine. An unexpected feature
was that, although in its medial portion the arcuate
fasciculus appeared homogenous along its entire length,
at its most lateral aspect, its fibers projected to three
distinct cortical regions (see Fig 2). The presence of a
parietal projection is inconsistent with Wernicke’s
model, and this result prompted us to perform further
virtual dissections to explore the anatomy of the arcuate
fasciculus in more detail. Using a two-ROI approach
based on the three distinct regions identified in
Figure 2, we identified three types of connections: (1) a
long segment corresponding to the classical arcuate fasciculus
connecting frontal and temporal lobes located
medially (red fibers in Fig 3); (2) a posterior lateral
segment connecting temporal and parietal lobes (yellow
fibers in Fig 3); and (3) an anterior lateral segment connecting
frontal and parietal lobes (green fibers in Fig 3).
To confirm that this pattern of tractography-defined
Catani et al: Language Networks 9
connectivity was not an artifact of averaging (ie, the
results in Figs 2 and 3 were obtained from the average
brain DT-MRI data set), we examined six individual
brains chosen at random from the group for evidence
of the same perisylvian network. Figure 4 shows the
tractography results for the six subjects. The three distinct
segments of the arcuate fasciculus are present in
all, although individual variations are also evident. For
Fig 2. Tractography reconstruction of the arcuate fasciculus using the one-region of interest approach. Sagittal sections (Talairach x
ϭ Ϫ38 to Ϫ52) of the arcuate fasciculus (red) in the average brain superimposed on the fractional anisotropy image. Note the
color coding used in tractography images differs from that used for the color fiber orientation maps. The reconstruction has a close
correspondence to classical postmortem anatomical descriptions of the arcuate bundle that originates in the posterior part of the ventrolateral
frontal lobe, arches around the external capsule medially (left panel) and insula cortex laterally (middle panel), and ends
in the temporal lobe. The lateralmost section (Talairach x ϭ Ϫ52) shows three cortical termination territories (numbered 1–3).
Fig 1. Delineation of the region of interest around the arcuate fasciculus. The top row shows axial color fiber orientation maps from a
single subject (Talairach z ϭ 22–28). Fibers that are predominantly oriented lateral-to-lateral are shown in red, anterior-posteriorly or
vice versa in green, and superior-inferiorly or vice versa in blue. Green fibers lateral to the vertical corona radiata belong to the arcuate
fasciculus (yellow arrows indicate the anterior and posterior extent of the arcuate fibers). Guided by the color fiber orientation map, a
region of interest (encircled in white) was defined around the fibers in the average fractional anisotropy image.
10 Annals of Neurology Vol 57 No 1 January 2005
example, in Subject 1, the anterior segment (see Fig 4,
green) is more prominent than the posterior segment
(see Fig 4, yellow), whereas the opposite is true in Subject
6. The temporal lobe terminations also vary across
subjects. For example, the long posterior segment terminations
extend further along the temporal lobe in
Subjects 3, 5, and 6 than in Subjects 1, 2, and 4.
Discussion
DT-MRI tractography helped visualize putative pathways
in vivo between perisylvian language areas of the
left hemisphere. The results point to two parallel pathways
connecting temporal and frontal regions: a direct
pathway (long segment) corresponding to the pathway
anticipated by Wernicke, and a novel, indirect pathway
connecting temporal with parietal (posterior segment)
and parietal with frontal regions (anterior segment). Although
the indirect pathway requires further anatomical
validation, several independent lines of evidence
support its existence. First, classical anatomists had
noted the presence of cortico-cortical fibers lateral to
the arcuate fasciculus3
; however, unaware of the
linguistic functions of the parietal lobe, they had no
reason to consider the fibers as part of a perisylvian
language network. Second, intraoperative electrocorticography
studies investigating perisylvian language networks
in humans have shown bidirectional connectivity
from Broca’s area to both superior temporal and inferior
parietal regions.27
Of further relevance to the indirect
pathway, an unexplained delay was found in the
response evoked in superior temporal cortex by stimulation
of Broca’s area and vice versa.27
The delay is
consistent with the presence of an extra synapse in the
pathway, and although the authors suggest a possible
Broca–subcortical–Wernicke pathway to account for
the results, they are equally explicabley by our indirect
Broca–parietal–Wernicke pathway. Third, path analysis
of functional MRI activity during linguistic processing
points to a functional connection between Broca’s area
and parietal cortex unaccounted for by the classical arcuate
model.28
In the next section, we propose functions
for the two parallel pathways based on previous
studies of aphasic patients, place the anatomical model
in an evolutionary context, and discuss its implication
for neurological models of language.
The Function of Direct and Indirect
Perisylvian Pathways
Our study suggests a greater complexity of connectivity
between frontal and temporal language regions than
previously supposed, allowing for a variety of distinct
disconnection syndromes not predicted by the classical
arcuate model. McCarthy and Warrington29
reported
Fig 3. Tractography reconstruction of the arcuate fasciculus using the two-region of interest approach. Broca’s and Wernicke’ territories
are connected through direct and indirect pathways in the average brain. The direct pathway (long segment shown in red) runs
medially and corresponds to classical descriptions of the arcuate fasciculus. The indirect pathway runs laterally and is composed of
an anterior segment (green) connecting the inferior parietal cortex (Geschwind’s territory) and Broca’s territory and a posterior segment
(yellow) connecting Geschwind’s and Wernicke’s territories. Note the color coding in this figure differs from that used in Figures
1 and 2.
Catani et al: Language Networks 11
three aphasic patients, two of which had a complementary
deficit to the third. The two patients had a temporoparietal
lesion that resulted in an impaired performance
in repetition tasks but relatively preserved
spontaneous speech and comprehension (classical conduction
aphasia). In contrast, the third patient had a
superficial parietal lesion resulting in impaired spontaneous
speech but relatively intact repetition (transcortical
motor aphasia). The patients with a deep lesion
were facilitated in repetition tasks that required active
semantic processing, whereas the patient with a superficial
lesion was impaired. The opposite dissociation
was observed in tasks that required passive repetition.
The patients with a deep lesion were impaired, whereas
the patient with a superficial lesion was facilitated. The
authors proposed a two-route model of speech production:
a direct pathway between Wernicke’s and Broca’s
areas acting in fast, automatic word repetition, and an
indirect pathway where a stage of verbal comprehension
and semantic/phonological transcoding intervened
between verbal input and articulatory output. Our
tractography findings appear to provide an anatomical
substrate for the findings of McCarthy and War-
rington.29
The deep intracerebral lesions in the first
two patients would have affected predominantly the direct
long segment pathway (the classical arcuate fasciculus)
that lies medial to the indirect pathway. In contrast,
the superficial cortical lesion of the third patient
Fig 4. Tractography reconstructions of the direct and indirect pathways in six individual subjects superimposed on sagittal fractional
anisotropy data. The three segments are color coded using the conventions of Figure 3. Both direct and indirect pathways are present
in each subject, but an intersubject variance is evident.
12 Annals of Neurology Vol 57 No 1 January 2005
would have affected predominantly the indirect pathway
or its parietal cortical relay station.
Our revised arcuate anatomy also suggests the possibility
of two further aphasic disconnection syndromes:
one based on an isolated lesion of the anterior segment,
and the other based on an isolated lesion of the posterior
segment. We would speculate that, for the McCarthy
and Warrington model,29
semantic/phonological
transcoding would be affected in both syndromes
but in different ways. A lesion of the anterior segment
would be expected to alter connectivity between Broca’s
area and the parietal region, resulting in a failure
to vocalize semantic content. In contrast, a lesion of
the posterior segment would be expected to alter connectivity
between Wernicke’s area and the parietal region,
resulting in a failure of auditory semantic comprehension.
These two theoretical syndromes bear
some resemblance to the subtypes of aphasia first described
by Lichtheim,30
which have later been referred
to as transcortical motor and sensory aphasia.31
In the
former, spontaneous speech is reduced, but comprehension
and repetition are intact. In the latter, comprehension
is reduced, but fluency and repetition are
spared. The location of the lesion in transcortical sensory
aphasia is entirely consistent with damage to the
posterior segment between the posterior temporal and
parietal lobes.31,32
Transcortical motor aphasia is associated
with widespread frontal lesions, some of which
are likely to have affected the anterior segment.33
In summary, the combined evidence from patients
with aphasic disconnection syndromes, and our interpretation
of such evidence in light of the parallel pathway
model, suggests the following functions for indirect
and direct pathways: the indirect pathway appears
to relate to semantically based language functions (such
as auditory comprehension and vocalization of semantic
content), whereas the direct pathway relates to phonologically
based language functions (such as automatic
repetition). This is not to say that these functions are
restricted to perisylvian areas, but merely that within
the parallel pathways we describe, the two functions are
anatomically dissociable. Exactly how other functions
such as verbal working memory, syntax, reading, and
writing relate to the parallel pathways, or how the
pathways themselves connect to other extrasylvian language
areas, remains unclear.
Broca-like and Wernicke-like Conduction Aphasia
The direct and indirect pathways model provides a
novel explanation for the clinical observation that conduction
aphasias fall into two distinct groups: the
Broca-like syndrome in which the deficit in repetition
is accompanied by a relative impairment in fluency and
the Wernicke-like syndrome in which the deficit in
repetition is accompanied by a relative impairment in
comprehension.6,8
One explanation for this dichotomy
is that more anterior lesions encroach on Broca’s area,
whereas more posterior lesions encroach on Wernicke’s
area; the two resulting aphasias being a mixture of cortical
deficits (Broca’s or Wernicke’s) and subcortical
conduction deficits. However, cases of Broca-like or
Wernicke-like conduction aphasias have been described
in patients with pure subcortical lesions,6
an observation
that suggests this explanation is incorrect. We suggest
that Broca-like and Wernicke-like conduction
aphasias result from lesions involving direct and indirect
pathways at different points along their course. A
lesion located anteriorly and involving the long and anterior
segments would lead to Broca-like conduction
aphasia, whereas a lesion located posteriorly involving
the long and posterior segments would lead to a
Wernicke-like conduction aphasia. Our explanation is
consistent with clinical evidence that Broca-like conduction
aphasias occur with subcortical frontal lesions,
whereas Wernicke-like conduction aphasias occur with
subcortical temporoparietal lesions.6
From Language Centers to Language Territories
The distribution of arcuate fiber terminations found by
tractography extends beyond the classical limits of Broca’s
areas (Brodmann Area, BA44 and 45)34
to include
part of the middle frontal gyrus and inferior precentral
gyrus. This wider distribution is consistent with several
lines of evidence that Broca’s area (in the restricted
sense used earlier) is surrounded by cortical regions
with specializations for higher level aspects of lan-
guage.27,34,35
We refer to this extended region as Broca’s
territory. Interestingly, the tractography evidence
appears to suggest that long segment direct fibers
project to anterior Broca’s territory, whereas the anterior
segment of the indirect pathway projects more
posteriorly (see Fig 3). This anatomical segregation of
arcuate fibers suggests a rostrocaudal segregation of
function within Broca’s territory, a segregation that
echoes the findings of functional imaging studies.36,37
Although Wernicke’s original description was of a
temporal lobe language area, the term Wernicke’s area
subsequently has been used to include inferior parietal
areas, as well as posterior temporal areas,38
a region that
encompasses BA22, 37, 39, and 40.34
Our tractography
results suggest that the posterior temporal and inferior
parietal regions of Wernicke’s area are extensively connected
but distinct, an observation that agrees with functional
anatomical studies of Wernicke’s area that have
argued that the term should be reserved for the posterior
temporal region.39
As is the case for Broca’s territory,
the distribution of tractography-defined fibers in the
posterior temporal region is wider than the classical
model, suggesting a segregation of function and an extended
Wernicke’s territory (including the posterior part
of both the superior and middle temporal gyrus) rather
than a localized center. This view is supported by funcCatani
et al: Language Networks 13
tional imaging and neuroanatomical studies suggesting a
segregation of language functions in the posterior temporal
lobe.37,40
Our tractography results highlight the importance of
the inferior parietal cortex as a separate primary language
area with dense connections to classical language
areas through the previously undescribed indirect pathway.
The region corresponds to BA39 and 40, and although
its importance as a linguistic area has been recognized
for some time, recent neuroimaging studies
have highlighted its importance for semantic process-
ing.41
Norman Geschwind5
emphasized the importance
of the region in ideational speech, suggesting
that, through cortico-cortical interactions, the convergence
of multimodality sensory inputs in the inferior
parietal lobe allowed the development of semantic content.
The development of this area and its perisylvian
connections is thought to have coincided with the
emergence of language in humans5,34
(see later). In recognition
of Geschwind’s contribution, we refer to this
region as Geschwind’s territory.
Comparative Anatomy and Phylogeny of Perisylvian
Language Networks
Perisylvian connections in the monkey brain have been
studied extensively using fiber tracing techniques; however,
their significance with respect to language remains
controversial because the homologies between cortical
areas in monkeys and humans are unclear.42,43
Initial
studies concluded that there was no equivalent of the
arcuate fasciculus in the monkey, with fibers originating
in the posterior superior temporal gyrus (the putative
homologue of Wernicke’s area) projecting to a region of
frontal cortex, dorsal and anterior to the putative homologue
of Broca’s area (in proximity to the inferior extent
of the arcuate sulcus).44
However, later work using more
sensitive techniques showed a specific pattern of connectivity
between both the inferior parietal cortex and the
putative Wernicke homologue with the Broca homologue
and areas surrounding it.45,46
A recent review42
concludes that in the macaque, a rudiment of the arcuate
fasciculus may exist; however, both its origin and termination
sites appear to be more diverse than in the human
brain, and prominent connections are found
between the inferior partial lobule and Broca’s homologue,
for which “there is no information of such corresponding
connections in the human brain.” Our tractography
results perfectly bridge these apparent
discrepancies between monkey and human anatomy.
First, our fiber tracts extend beyond the classical language
areas to form the territories described earlier, corresponding
to the diversity of origin and termination
sites found in the macaque. Second, our finding of an
anterior segment connecting Geschwind’s and Broca’s
territories mirrors the prominent inferior parietal to
frontal fiber pathway described in the monkey. Finally,
the tractography-defined cortical distribution of fibers
within Broca’s territory bears a striking resemblance to
that described in the monkey. In the monkey, a rostrocaudal
separation of projections to inferior frontal cortex
is found; that is, the more rostral regions receive projections
from the Wernicke homologue and the more caudal
regions receive projections from the inferior parietal
lobe.45
This mirrors our finding that long segment fibers
(from Wernicke’s) connect to anterior portions of Broca’s
territory, whereas anterior segment fibers (from Geschwind’s)
connect to the posterior portions of Broca’s
territory.
One theory of the evolution of language from monkey
to human is that it involves a change in the
strengths of perisylvian connections. Aboitiz and Gar-
cia34
argue that two evolutionary tendencies are involved.
First, posterior superior temporal and inferior
parietal regions became increasingly connected, linking
the auditory system and a pre-existing parietal–premotor
loop involved in the generation of complex vocalizations.
Second, the development of connections between
posterior superior temporal and inferior frontal
regions links auditory information to orofacial premotor
regions. We would argue that these two tendencies
correspond to the evolution of posterior and long segments,
respectively, the anterior segment being, phylogenetically,
the oldest component of the perisylvian
network. An alternative theory argues that human language
evolved from the mirror neuron system in the
monkey through a sequence of stages involving grasp
imitation, protosigns, and protospeech.42
Although the
theory does not invoke the strengthening of connections,
the mirror neuron system includes inferior frontal,
inferior parietal, and superior temporal regions
overlapping those in our model.
Methodological Issues
We should stress the difference between measuring anatomical
connections directly and inferring white matter
trajectories with DT-MRI tractography. The former
would require the ability to identify the cell bodies that
give rise to the axons of interest, then to identify the
neurons with which the axonal terminals synapse, in
addition to the trajectory of the axons from one site to
the other. Obtaining this information in the human
brain in vivo noninvasively is, of course, currently impossible.
The intrinsic resolution of our DT-MRI data
was 2.5 ϫ 2.5 ϫ 2.5 mm. With such resolution, one
cannot hope to resolve individual axons, neurons, and
synapses. Therefore, we have to rely on proxy techniques
to infer information about white matter pathways.
DT-MRI tractography takes the bulk-averaged
tissue properties in each voxel and, by fitting a model
to the data, infers the dominant fiber orientation
within each voxel. These discrete estimates of fiber orientation
are pieced together to provide estimates of the
14 Annals of Neurology Vol 57 No 1 January 2005
trajectory of the fasciculus of interest. Each estimate of
fiber orientation is subject to noise contamination,47
which can lead to accumulated errors and low repeatability
(precision) in tract reconstructions (although
our use of the two-ROI approach serves to minimize
the effect of this confound). Furthermore, inadequacies
in the model (eg, incomplete modeling of the diffusion
process at intersections/crossing of fibers48
) can also
contribute to errors in tract reconstruction. Therefore,
one should perhaps always refer to the trajectories elucidated
by DT-MRI tractography as “tractographydefined”
pathways, and, clearly, any results obtained
through tractography should be investigated using
“gold standard” techniques such as postmortem dissection.
However, even within these confounds, the
tractography-defined direct pathways presented in this
article correspond exactly to previously published postmortem
descriptions, and the indirect pathway is
largely consistent across the six subjects examined individually.
Furthermore, electrocorticographic and functional
MRI path analysis studies provide independent
evidence for a functional connection between Geschwind’s
and Broca’s territories, as would be predicted
by the tractography-defined parallel pathway mod-
el.27,28
Tractography is also unable to determine the
exact cortical origin and termination of fibers, therefore
we can only infer the territories of projection.18,25
It
should be noted that in this article we only studied the
connections of perisylvian cortex and that other brain
regions are likely to connect to the network (eg, the
insula). Also, our network is limited to long associative
fiber connections and does not include superficial
U-shaped fibers or thalamic connections, which may
also play a role in language.38
Finally, our study is restricted
to the left hemisphere of right-handed male
subjects. Whether the pattern of connectivity identified
in this article varies with hemisphere, sex, or handedness
are questions for future study.
From Associationist to Connectionist Models
of Language
In 1885, Lichtheim30
proposed an extension to the existing
Broca–arcuate–Wernicke model, based on the
careful observation of aphasic symptoms he encountered
clinically. He hypothesized an additional pathway between
Wernicke’s and Broca’s areas through a theoretical
third center thought to be related to semantic processing,
which he referred to as the concept center. Although
the existence of a concept center and its connections
remained theoretical through lack of anatomical
evidence, Lichtheim’s model proved a useful clinical
classificatory tool, widely adopted by clinicians for the
bedside assessment of aphasia. Our tractography findings
provide anatomical support for Lichtheim’s original
scheme. In fact, the direct and indirect pathway model
suggested by our tractography dissections closely
matches the theoretical wiring diagram outlined in Lichtheim’s
1885 article. This is not to say that Lichtheim’s
model is entirely consistent with contemporary views.
Lichtheim conceived language in associationist terms of
centers and pathways, the general assumption being that
visual and auditory linguistic information was processed
in localized cortical regions with the serial passage of information
between such regions through white matter
tracts. However, such associationist models have been criticized
on several grounds.49
An alternative “connectionist”
account has been proposed in which linguistic functions
are conceived as resulting from parallel distributed processing
performed by distributed groups of connected
neurons rather than individual centers.50,51
Although tractography
evidence alone cannot distinguish between these
alternative models, some aspects of our results are more
supportive of the connectionist account. First, our finding
of extended language territories argues against the associationist
concept of localized centers. Second, the fiber bundles
identified appear unnecessarily large for the simple
serial transfer of signals from one region to another. Instead,
their size suggests that subsets of fibers connect different
linguistic subregions in parallel. The picture that
emerges is of multiple parallel connections among subspecialized
cortical regions of Broca’s, Wernicke’s, and Geschwind’s
territories consistent with Mesulam’s large-scale
neural network model of language.52
Conclusions
We have shown that the perisylvian language areas of
the left hemisphere are connected by two parallel pathways.
The findings may provide an explanation for the
observation that lesions disconnecting Broca’s and Wernicke’s
areas cause different clinical aphasia syndromes
depending on their location. Although the pathways
shown by our tractography method are closely related to
those hypothesized by Lichtheim, they go beyond his
model in their support of a connectionist account of linguistic
function with processing distributed between
brain territories rather than localized within specific centers.
Tractography appears to be a valuable technique
both in helping to understand normal and aphasic language
function and in providing anatomical constraints
for connectionist accounts of language.
This study was supported by the Wellcome Trust (054030/2/98,
067437/7/02/A, D.K.J., D.H.ff.).
We thank Drs R. Howard and A. Baddeley for their suggestions.
References
1. Broca P. Perte de la parole, ramollissement chronique et destruction
partielle du lobe ante´rieur gauche. Bull Soc Anthropol
1861;2:235–238.
2. Wernicke C. Der aphasische Symptomencomplex. Breslau: Cohen
and Weigert, 1874.
Catani et al: Language Networks 15
3. Dejerine J. Anatomie des centres nerveux. Vol 1. Paris: Rueff et
Cie, 1895.
4. Benson DF, Sheremata WA, Bouchard R, et al. Conduction
aphasia: a clinicopathological study. Arch Neurol 1973;28:
339–346.
5. Geschwind N. Disconnexion syndromes in animals and man.
Brain 1965;88:237–294.
6. Naeser MA, Alexander MP, Helm-Estabrooks N, et al. Aphasia
with predominantly subcortical lesion sites: description of three
capsular/putaminal aphasia syndromes. Arch Neurol 1982;39:
2–14.
7. Alexander MP, Naeser MA, Palumbo CL. Correlations of subcortical
CT lesion sites and aphasia profiles. Brain 1987;110:
961–991.
8. Kempler D, Metter EJ, Jackson CA, et al. Disconnection and
cerebral metabolism: the case of conduction aphasia. Arch Neurol
1988;45:275–279.
9. Le Bihan D, Breton E. Imagerie de diffusion in vivo par resonance
magnetique nucleaire. Cr Acad Sci (Paris) 1985;301:
1109–1112.
10. Moseley ME, Cohen Y, Kucharczyk J, et al. Diffusion-weighted
MR imaging of anisotropic water diffusion in cat central nervous
system. Radiology 1990;176:439–445.
11. Basser PJ, Mattiello J, Le Bihan D. MR diffusion tensor spectroscopy
and imaging. Biophys J 1994;66:259–267.
12. Mori S, Crain BJ, Chacko VP, van Zijl PC. Three-dimensional
tracking of axonal projections in the brain by magnetic resonance
imaging. Ann Neurol 1999;45:265–269.
13. Conturo TE, Lori NF, Cull TS, et al. Tracking neuronal fiber
pathways in the living human brain. Proc Natl Acad Sci U S A
1999;96:10422–10427.
14. Jones DK, Simmons A, Williams SCR, Horsfield MA. Noninvasive
assessment of axonal fiber connectivity in the human
brain via diffusion tensor MRI. Magn Reson Med 1999;42:
37–41.
15. Poupon C, Clark CA, Frouin V, et al. Regularization of
diffusion-based direction maps for the tracking of brain white
matter fasciculi. NeuroImage 2000;12:184–195.
16. Basser PJ, Pajevic S, Pierpaoli C, et al. In vivo tractography
using DT-MRI data. Magn Reson Med 2000;44:625–632.
17. Behrens TEJ, Johansen-Berg H, Woolrich MW, et al. Noninvasive
mapping of connections between human thalamus and
cortex using diffusion imaging. Nat Neurosci 2003;6:750–757.
18. Catani M, Jones DK, Donato R, ffytche DH. Occipitotemporal
connections in the human brain. Brain 2003;126:
2093–2107.
19. Catani M, Howard RJ, Pajevic S, Jones DK. Virtual in vivo
interactive dissection of white matter fasciculi in the human
brain. Neuroimage 2002;17:77–94.
20. Lehe´ricy S, Ducros M, Van de Morteele PF, et al. Diffusion
tensor fibre tracking shows distinct corticostriatal circuits in humans.
Ann Neurol 2004;55:522–529.
21. Shaywitz BA, Shaywitz SE, Pugh KR, et al. Sex differences in
the functional organization of the brain for language. Nature
1995;373:607–609.
22. Jones DK, Williams SCR, Gasston D, et al. Isotropic resolution
diffusion tensor imaging with whole brain acquisition in a clinically
acceptable time. Hum Brain Mapp 2002;15:216–230.
23. Basser PJ, Pierpaoli C. Microstructural and physiological features
of tissue elucidated by quantitative-diffusion-tensor MRI.
J Mag Reson B 1996;111:209–219.
24. Alexander DC, Pierpaoli C, Basser PJ, Gee JC. Spatial transformations
of diffusion tensor magnetic resonance images. IEEE
Trans Med Imaging 2001;20:1131–1139.
25. Jones DK, Griffin LD, Alexader DC, et al. Spatial normalization
and averaging of diffusion tensor MRI data sets. Neuroimage
2002;17:592–617.
26. Pajevic S, Pierpaoli C. Color schemes to represent the orientation
of anisotropic tissues from diffusion tensor data: application
to white matter fiber tract mapping in the human brain.
Magn Reson Med 1999;42:526–540.
27. Matsumoto R, Nair DR, LaPresto E, et al. Functional connectivity
in the human language system: a cortico-cortical evoked
potential study. Brain 2004;127:2316–2330.
28. Bullmore E, Horwitz B, Honey G, et al. How good is good
enough in path analysis of fMRI data? Neuroimage 2000;11:
289–301.
29. McCarthy R, Warrington E. A two-route model of speech production.
Brain 1984;107:463–485.
30. Lichtheim L. On aphasia. Brain 1885;7:433–484.
31. Damasio AR, Geschwind N. The neural basis of language. Ann
Rev Neurosci 1984;7:127–147.
32. Boatman D, Gordon B, Hart J, et al. Transcortical sensory
aphasia: revisited and revised. Brain 2000;123:1634–1642.
33. Schiff HB, Alexander MP, Naeser MA, et al. Aphemia: clinicalanatomical
correlations. Arch Neurol 1983;40:720–727.
34. Aboitiz F, Garcia RV. The evolutionary origin of the language
areas in the human brain. A neuroanatomical perspective. Brain
Res Rev 1987;25:381–396.
35. Damasio AR. Aphasia. N Engl J Med 1992;326:531–539.
36. Paulesu E, Goldacre B, Scifo P, et al. Functional heterogeneity
of left inferior frontal cortex as revealed by fMRI. Neuroreport
1997;8:2011–2017.
37. Cannestra AF, Bookheimer SY, Pouratian N, et al. Temporal and
topographical characterization of language cortices using intraoperative
optical intrinsic signals. Neuroimage 2000;12:41–54.
38. Penfield W, Roberts L. Speech and brain-mechanisms.
Princeton: Princeton University Press, 1959.
39. Wise RJS, Scott SK, Blank SC, et al. Separate neural subsystems
within “Wernicke’s area.” Brain 2001;124:83–95.
40. Castillo EM, Simos PG, Davis RN, et al. Levels of word processing
and incidental memory: dissociable mechanisms in the
temporal lobe. Neuroreport 2001;12:3561–3566.
41. Price CJ. The anatomy of language: contributions from functional
neuroimaging. J Anat 2000;197:335–359.
42. Arbib M, Bota M. Language evolution: neural homologies and
neuroinformatics. Neural Netw 2003;16:1237–1260.
43. Deacon T. Monkey homologues of language areas: computing
the ambiguities. Trends Cogn Sci 2004;8:288–290.
44. Petrides M, Pandya DN. Association fiber pathways to the
frontal cortex from the superior temporal region in the rhesus
monkey. J Comp Neurol 1988;273:52–66.
45. Deacon TW. Cortical connections of the inferior arcuate sulcus
cortex in the macaque brain. Brain Res 1992;573:8–26.
46. Petrides M, Pandya DN. Comparative cytoarchitectonic analysis
of the human and the macaque ventrolateral prefrontal cortex
and corticocortical connection patterns in the monkey. Eur
J Neurosci 2002;16:291–310.
47. Jones DK. Determining and visualizing uncertainty in estimates
of fiber orientation from diffusion tensor MRI. Magn Reson
Med 2003;49:7–12.
48. Pierpaoli C, Barnett A, Pajevic S, et al. Water diffusion changes
in wallerian degeneration and their dependence on white matter
architecture. Neuroimage 2001;13:1174–1185.
49. Gainotti G. Development of the concept of aphasia. In: Denes
G, Pizzamiglio L, eds. Handbook of clinical and experimental
neuropsychology. London, UK: Psychology Press, 1999.
50. Nadeau SE. Phonology: a review and proposals from a connectionist
perspective. Brain Lang 2001;79:511–579.
51. McClelland JL, Rogers TT. The parallel distributed processing
approach to semantic cognition. Nat Neurosci 2003;4:310–322.
52. Mesulam M-M. Large-scale neurocognitive networks and distributed
processing for attention, language, and memory. Ann
Neurol 1990;28:597–613.
16 Annals of Neurology Vol 57 No 1 January 2005