THE FUNCTIONAL NEUROANATOMY OF WORKING MEMORY: CONTRIBUTIONS OF HUMAN BRAIN LESION STUDIESWhy patient studies remain indispensableThe functional neuroanatomy of the visuospatial sketchpadThe functional neuroanatomy of the phonological loopWM and the frontal lobesCONCLUSIONSHow to study patients todayREFERENCESTHE FUNCTIONAL NEUROANATOMY OF WORKING MEMORY:CONTRIBUTIONS OF HUMAN BRAIN LESION STUDIESN. G. MÜLLERa* AND R. T. KNIGHTb,caCognitive Neurology Unit and Brain Imaging Center, Department ofNeurology, Johann Wolfgang Goethe-University, Theodor-Stern-Kai 7,60590 Frankfurt am Main, GermanybDepartment of Psychology, University of California, Berkeley, CA,USAcThe Helen Wills Neuroscience Institute, University of California,Berkeley, CA, USAAbstract—Studies of patients with focal brain lesions remaincritical components of research programs attempting to un-derstand human brain function. Whereas functional imagingtypically reveals activity in distributed brain regions thatare involved in a task, lesion studies can define which ofthese brain regions are necessary for a cognitive process.Further, lesion studies are less critical regarding the selec-tion of baseline conditions needed in functional brain imag-ing research.Lesion studies suggest a functional subdivision of thevisuospatial sketchpad of working memory with a ventralstream reaching from occipital to temporal cortex supportingobject recognition and a dorsal stream connecting the occip-ital with parietal cortex enabling spatial operations. The pho-nological loop can be divided into a phonological short-termstore in inferior parietal cortex and an articulatory subvocalrehearsal process relying on brain areas necessary forspeech production, i.e. Broca’s area, the supplementary mo-tor association area and possibly the cerebellum.More uncertainty exists regarding the role of the prefron-tal cortex in working memory. Whereas single cell studies innon-human primates and functional imaging studies in hu-mans have suggested an extension of the ventral and dorsalpath into different subregions of the prefrontal cortex, lesionstudies together with recent single-cell and imaging studiespoint to a non-mnemonic role of the prefrontal cortex, includ-ing attentional control of sensory processing, integration ofinformation from different domains, stimulus selection andmonitoring of information held in memory. Our own dataargue against a modulatory view of the prefrontal cortex andsuggest that processes supporting working memory are dis-tributed along ventral and dorsal lateral prefrontal cortex.© 2005 IBRO. Published by Elsevier Ltd. All rights reserved.Key words: working memory, visuospatial sketchpad, pho-nological loop, brain lesion, prefrontal cortex.“Patients with brain lesions provide a unique window intobrain function, and this approach will fill an important nichein future research.” (Rorden and Karnath, 2004).In the days of Broca and Wernicke lesion studies were theonly method to assign cognitive function to brain anatomy.Currently, functional imaging (functional magnetic reso-nance imaging [fMRI], positron emission tomography[PET], magnetoencephalography [MEG], electroencepha-lography [EEG]) and related techniques such as transcra-nial magnetic stimulation (TMS) allow the non-invasivestudy of cognitive processes with good spatial and tempo-ral resolution in the healthy brain. Thus, the ongoing needfor patient studies may be questioned given the numerouslimitations inherent in this approach. For instance, lesionsin humans are usually not limited by the boundaries ofthe underlying function but are, for example, determined bya common blood supply or a regional susceptibility totrauma. Thus, lesions in the brain are not randomly distrib-uted which is further complicated by the fact that only asmall fraction of a disease-affected brain region may ac-tually sustain the function under investigation. Additionalcomplications of patient research are that the borders oflesions are not always clear and brain tissue that seemsintact on CT or MR scans may nevertheless be functionallyimpaired, for example through disconnection from an areawhich is part of a key network for task performance. Le-sions of different etiology are also difficult to compare. Forexample, tumors often cause less symptoms than a com-parably sized stroke perhaps due to re-organization offunction in slow growing tumors (Anderson et al., 1990).On the other hand, ischemia may have altered function inbrain tissue that macroscopically seems intact.Another major problem one encounters is plasticity,because damage to the brain is often followed by somere-organization complicating inferences about normal brainfunction. Homogenous groups of patients are also difficultto establish and maintain whereas neuroimaging studiescan rely on easily defined young and older control sub-jects. Last but not least, lesion studies do not seem toaccount for the now popular view that complex cognitivefunctions are supported by neural networks rather thandiscrete and specialized modules. However, this notionmay also be incorrect since lesion studies coupled withother neuroimaging techniques such as EEG and fMRI canprovide key evidence on neural network contributions tocognition (Barcelo et al., 2000; Yago et al., 2004).Why patient studies remain indispensableThe reason why patient studies in the era of functionalimaging are still necessary is based on a principal draw-back inherent in functional imaging (see (Rorden and Kar-nath, 2004) for a review): Imaging reveals correlation notcausality. In other words, as only lesion studies address*Corresponding author. Tel: ⫹49-69-6301-6657; fax: ⫹49-69-6301-6842.E-mail address: [email protected] (N. G. Müller).Abbreviations: DL, dorsolateral subregion of the prefrontal cortex;DVL, dorso- plus ventrolateral subregion of the prefrontal cortex; EEG,electroencephalography; fMRI, functional magnetic resonance imag-ing; PFC, prefrontal cortex; VM, ventromedial subregion of the pre-frontal cortex; WM, working memory.Neuroscience 139 (2006) 51–580306-4522/06$30.00⫹0.00 © 2005 IBRO. Published by Elsevier Ltd. All rights reserved.doi:10.1016/j.neuroscience.2005.09.01851disruption instead of activation only these studies can re-veal structures that are indispensably required for a certainfunction. For example, activation of a brain region during adelay in a working memory (WM) task may not imply thatthis area’s basic function is to