Lesion analysis of the brain areas involved in language comprehension

Abstract

The cortical regions of the brain traditionally associated with the comprehension of language are Wernicke's area and Broca's area. However, recent evidence suggests that other brain regions might also be involved in this complex process. This paper describes the opportunity to evaluate a large number of brain-injured patients to determine which lesioned brain areas might affect language comprehension. Sixty-four chronic left hemisphere stroke patients were evaluated on 11 subtests of the Curtiss–Yamada Comprehensive Language Evaluation – Receptive (CYCLE-R; Curtiss, S., & Yamada, J. (1988). Curtiss–Yamada Comprehensive Language Evaluation. Unpublished test, UCLA). Eight right hemisphere stroke patients and 15 neurologically normal older controls also participated. Patients were required to select a single line drawing from an array of three or four choices that best depicted the content of an auditorily-presented sentence. Patients' lesions obtained from structural neuroimaging were reconstructed onto templates and entered into a voxel-based lesion-symptom mapping (VLSM; Bates, E., Wilson, S., Saygin, A. P., Dick, F., Sereno, M., Knight, R. T., & Dronkers, N. F. (2003). Voxel-based lesion-symptom mapping. Nature Neuroscience, 6(5), 448–450.) analysis along with the behavioral data. VLSM is a brain–behavior mapping technique that evaluates the relationships between areas of injury and behavioral performance in all patients on a voxel-by-voxel basis, similar to the analysis of functional neuroimaging data. Results indicated that lesions to five left hemisphere brain regions affected performance on the CYCLE-R, including the posterior middle temporal gyrus and underlying white matter, the anterior superior temporal gyrus, the superior temporal sulcus and angular gyrus, mid-frontal cortex in Brodmann's area 46, and Brodmann's area 47 of the inferior frontal gyrus. Lesions to Broca's and Wernicke's areas were not found to significantly alter language comprehension on this particular measure. Further analysis suggested that the middle temporal gyrus may be more important for comprehension at the word level, while the other regions may play a greater role at the level of the sentence. These results are consistent with those seen in recent functional neuroimaging studies and offer complementary data in the effort to understand the brain areas underlying language comprehension.

Introduction

Early work concerning the brain areas involved in language comprehension focused on the deficits observed in brain-injured aphasic patients and the correlation of these deficits with regions of injury in the brain. Recent advances in both structural and functional neuroimaging capabilities have afforded a new means by which earlier neural models of language comprehension can be evaluated. Yet, there has been a paucity of lesion studies that have fully exploited this new technology in defining a more precise role for individual brain areas within the left peri-Sylvian region. This is partly due to the difficulty in recruiting adequate numbers of patients to effectively examine such brain–behavior relationships. More importantly, lesion studies have not had an effective statistically-based tool with which to analyze these data in a way that allows them to explore discrete regions.

This paper will attempt to bridge that gap by utilizing a new method of lesion analysis to explore the brain areas involved in language comprehension in a large number of patients. This technique, designed to be more commensurate with the functional imaging work, was applied to data from a rigidly-screened group of 64 left hemisphere stroke patients, uniform in their handedness, native language, and neurological and psychiatric histories, who underwent the same extensive language and structural neuroimaging evaluations. This large cohort adds a perspective that functional neuroimaging studies, which look at regional activations in normal brains, cannot offer, i.e. the opportunity to examine the effects the absence of these brain areas has on the language comprehension system. Ultimately, findings from both lesion and functional imaging studies would need to converge before an optimal theory of language comprehension in the brain can be realized.

When one thinks of language comprehension, the brain area that first comes to mind is Wernicke's area in the posterior temporal lobe. This is because patients with Wernicke's aphasia have profound language comprehension deficits and often find it difficult to understand even the simplest phrases (Benson and Ardila, 1996, Goodglass, 1993). Those with Broca's aphasia were not originally thought to have a comprehension disorder (Broca, 1861), but were later shown to have difficulty at the sentence level, particularly with sentences requiring the processing of complex morphosyntactic structures (Berndt and Caramazza, 1980, Caramazza and Zurif, 1976, Schwartz et al., 1980, Zurif et al., 1972).

The idea that either a posterior or an anterior lesion could affect aspects of comprehension suggests that it is a more widely distributed process. Functional neuroimaging studies have come to capitalize on this idea by studying normal subjects performing language comprehension tasks under different conditions. In addition to Broca's and Wernicke's areas (Brodmann's areas (BA) 44/45 and posterior BA22, respectively), a number of other left hemisphere cortical regions have emerged in auditory word and sentence recognition studies. These include frontal BA 9 and 47 (e.g. Binder et al., 1996, Muller et al., 1997, Schlosser et al., 1998), parietal area 39 (e.g. Binder et al., 1996, Perani et al., 1996), and temporal areas 20, 21, and 42 (e.g. Binder et al., 1996, Grady et al., 1997) as well as the anterior portion of area 22 (e.g. Friederici et al., 2000, Mazoyer et al., 1993). The question now is whether lesion analysis can concur with the same brain–language relationships if an appropriate and comparable method could be found.

More than a decade ago, we began the systematic collection of language and neuropsychological data on a rigidly-selected group of chronic stroke patients referred to our Center for Aphasia and Related Disorders. At the time we began accumulating data, the most viable measure of language comprehension for aphasic patients that covered the broadest range of word, phrase, and sentence types was the Curtiss–Yamada Comprehensive Language Evaluation – Receptive measures (CYCLE-R) (Curtiss & Yamada, 1988). This test reflected developmental trends in language acquisition from age 2 to 9, and did not require a verbal response that could hinder performance in aphasic or speech apraxic patients. Over time, we were able to collect CYCLE-R data on a large number of left hemisphere stroke patients who had also undergone structural neuroimaging and whose lesions were computer-reconstructed for further analysis.

In initial presentations of this endeavor (Dronkers et al., 1994a, Dronkers et al., 1994b, Dronkers et al., 1996), we used a lesion overlay method to explore areas that might be involved in sentence comprehension. One novel finding concerned the involvement of the anterior portion of Brodmann's area 22 in the anterior superior temporal gyrus (STG). We originally thought this area was related to the comprehension of elaborated morphosyntactic structures, as patients with deficits on the more difficult sentence types of the CYCLE-R were all seen to have lesions involving this region. The present study examines this area more closely and amends this role to some degree.

Since that time, the lesion overlay method for the analysis of this type of data has been superceded by a technique newly available to lesion analysis known as voxel-based lesion-symptom mapping (VLSM; Bates et al., 2003). Past lesion studies have typically used one of two methodologies. The first is to group patients on the basis of lesions to a particular brain region and then test for differences between these patients and a control group (e.g. Chao and Knight, 1998, Friedrich et al., 1998). The second is to begin by defining a behavior and then to overlap the lesions of those patients to find a common area of injury (e.g. Dronkers, 1996, Kertesz, 1979). While these methods have been useful in offering information about the relationships between specific brain areas and certain behavioral functions, valuable information is sometimes lost if patients or behaviors do not meet certain criteria and must be discarded. VLSM avoids such losses. All continuous behavioral data are statistically analyzed on a voxel-by-voxel basis much as in functional imaging research. This also makes the results between these two literatures more compatible.

In the current study, we evaluated CYCLE-R performance of 64 single left hemisphere stroke patients using the more powerful, statistically-based VLSM method. Computerized lesion reconstructions were entered into the VLSM analysis along with patients' CYCLE-R data to evaluate if lesions in specific voxels might influence CYCLE-R performance. The goal was to determine if specific brain regions could be identified that, when lesioned, affected performance on a language comprehension task. We sought to explore several questions, including: What can a statistically-based lesion analysis reveal about the brain areas involved in language comprehension? How do these areas compare to those found in the functional imaging literature? How might each area participate in comprehension? How do these areas relate to each other? Are these areas specific to language or are they shared with other cognitive functions? This study only begins to answer some of these questions. The complexity of language comprehension necessarily dictates that numerous processes and brain areas must be interacting in elaborate ways to achieve this remarkable phenomenon. No single lesion or functional imaging study can provide the definitive answer, though each can contribute to the discussion.

For our contribution, we envision a model that assumes that the brain networks important for lower level functions such as word comprehension feed into networks that support higher level functions such as sentence comprehension. Such an incremental recruitment of brain areas follows from the logic that natural language sentence comprehension depends upon word comprehension, and any brain areas found to support sentences must also depend upon brain areas found to support words if a complete interpretation is to be made. The developing model also assumes that while certain brain areas might be specific to language, higher-level language functions must also interact with other cognitive skills such as executive functioning and short-term verbal memory. Thus, different levels of structural complexity from words to complex sentences are expected to engage distinct regional brain networks while at the same time interacting with other regional networks to effect language comprehension.

It is for these reasons that we chose a sentence comprehension task to explore the brain areas that might contribute to language comprehension from the word level to the level of complex syntax. The use of this measure is described herein. Since excellent reviews on the brain areas contributing to language comprehension can be found elsewhere in the literature (e.g. Cabeza and Nyberg, 2000, Grodzinsky, 2000, Kaan and Swaab, 2002), we will wait until the discussion section of our paper (Section 4) to introduce other research as it pertains to the specific findings of the present study. At that time, we will also discuss how our findings merge with those from functional imaging.