Evidence for a Two-Stage Model ofSpatial Working Memory Processingwithin the Lateral Frontal Cortex: APositron Emission Tomography Study

Adrian M. Owen, Alan C. Evans, and Michael Petrides

Montreal Neurological Institute, McGill University, Montreal,Canada

Previous work in nonhuman primates and in patients with frontal lobedamage has suggested that the frontal cortex plays a critical role inthe performance of both spatial and nonspatial working memory tasks.The present study used positron emission tomography with magneticresonance imaging to demonstrate the existence, within the humanbrain, of two functionally distinct subdivisions of the lateral frontalcortex, which may subserve different aspects of spatial working mem-ory. Five spatial memory tasks were used, which varied in terms ofthe extent to which they required different executive processes. Whenthe task required the organization and execution of a sequence ofspatial moves retained in working memory, significant changes inblood flow were observed in ventrolateral frontal cortex (area 47) bi-laterally. By contrast when the task required active monitoring andmanipulation of spatial information within working memory, additionalactivation foci were observed in mid-dorsolateral frontal cortex (areas46 and 9). These findings support a two-stage model of spatial workingmemory processing within the lateral frontal cortex.

There is considerable evidence that the frontal cortex playsa critical role in certain aspects of working memory. This ev-idence comes both from the study of patients with excisionsof frontal cortex (Petrides and Milner, 1982; Owen et al., 1990,1995) and from lesion and electrophysiological recordingwork on nonhuman primates (see Goldman-Rakic, 1987, forreview). In the monkey, it has been shown that lesions con-fined to one part of the dorsolateral frontal cortex, namelythe cortex lining the sulcus principalis (i.e., area 46) result insevere impairments on tests of spatial working memory, suchas the spatial delayed alternation and delayed response tasks(see Goldman-Rakic, 1987; Fuster, 1989).

On the basis of an analysis of the nature of the impairmenton nonspatial self-ordered working memory tasks shown bymonkeys with lesions of the mid-dorsal lateral frontal cortex(Petrides, 1991a,b, 1995), a general theoretical framework re-garding the role of the frontal cortex in mnemonic processingand its relationship to planning and other executive processeshas recently been proposed (Petrides, 1994). According to thisview, there are two executive processing systems within thelateral frontal cortex. The middle portion of the ventrolateralfrontal cortex (i.e., areas 45 and 47) underlies active compar-isons made about stimuli held in short-term memory as wellas the active organization of sequences of responses basedon conscious, explicit retrieval of information from posteriorcortical association systems. In this sense, this region servesas one level of interaction between short-term and long-termmemory systems and executive processing. By contrast, themid-dorsolateral frontal cortex (dorsal area 46 and area 9) isassumed to constitute another level of interaction of execu-tive processes with memory and is recruited only when ac-tive manipulation and monitoring of information within work-ing memory is required. By monitoring, we refer to the activechecking of information as occurs, for example, when a sub-ject is required to decide which items in a given array ofstimuli have been selected previously, and which ones havenot. This two-stage model of lateral frontal cortical function

describes how both spatial and nonspatial stimuli are retainedand manipulated within working memory. The model makesa number of specific predictions, some of •which have re-cently been tested with nonspatial stimuli in functional acti-vation studies with positron emission tomography (PET).Thus, Petrides et al. (1993) have reported significant increasesin regional cerebral blood flow (rCBF) within the mid-dor-solateral frontal cortex (areas 46 and 9) on a visual nonspatialself-ordered working memory task in which the subjects wererequired to monitor which ones of a set of stimuli had beenselected and which ones had not. In a related study, whensubjects were required to make judgements about, but not tomanipulate, similar visual stimuli within working memory, asignificant increase in blood flow was observed only in ven-trolateral frontal cortex (Petrides et al., unpublished observa-tions).

The present PET study was designed to investigate wheth-er this two-stage model of lateral frontal cortical function mayalso apply to the contribution of the frontal cortex to spatialworking memory. Normal subjects were scanned while per-forming five different spatial working memory tasks that dif-fered in terms of the extent to which they required activemonitoring and manipulation of the stored information. It waspredicted that the tasks requiring active judgements aboutthe contents of working memory and the organization of ap-propriate responses but that had minimal monitoring requi-rements would only activate ventrolateral areas within thefrontal lobe. By contrast, performance on tasks that requiredmonitoring and manipulation of spatial information withinworking memory would activate areas within the mid-dorso-lateral frontal cortex (i.e., areas 46 and 9).

Materials and Methods

Scanning Methods and Data AnalysisPET scans were obtained with the Scanditronix PC-2048 system,which produces 15 image slices at an intrinsic resolution 5.0 X 5.0X 6.0 mm (Evans et al., 1991a). In this study, the resultant "Beld ofview," within which PET data from all 16 subjects was obtained, ex-tended from 24 mm below the anterior-posterior commissure line(AC-PC line) to 61 mm above it. The relative distribution of regionalcerebral blood flow (rCBF) was measured with the bolus Hj"O meth-odology (Raichle et al., 1983), without arterial sampling (Fox andRaichle, 1984). For each subject, a high-resolution magnetic reso-nance imaging (MRI) study (whole brain, 1 mm3 voxels, 3-D sagittalacquisition) was also obtained from a Philips Gyroscan 1.5T and resl-iced so as to be coregistered with the PET data (Evans et al., 1991b).An orthogonal coordinate frame was then established based on theAC-PC line as defined in the MRI volume (Evans et al., 1992). Thesecoordinates were used to apply a trilinear resampling of each pair ofMRI and PET data sets into a standardized stereotaxic coordinatesystem (Talairach and Toumoux, 1988). To overcome residual ana-tomical variability persisting after stereotaxic standardization, the PETimages were reconstructed with a 20 mm filter and then normalizedfor global rCBF and averaged across subjects within each scanningcondition. The mean state-dependent change rCBF image volume wasobtained (Fox et al., 1985) and converted to a t statistic volume by

Cerebral Cortex Jan/Feb 1996;6:31-38; 1047-3211/96/$4.00

Monitoring I Monitoring III Fixed Spatial SequenceMonitoring II Spatial Span

Task

Figure 1. Performance data from the five spatial memory tasks. In each case, the per-centage of total trials completed without an error is shown. Errors bars are SEM.

dividing each voxel by the mean standard deviation in normalizedCBF for all intracerebral voxels (Worsley et al., 1992).

•Individual MRI images were subjected to the same averaging pro-cedure, such that composite stereotaxic image volumes sampled atapproximately 1.5 mm in each dimension were obtained for both tstatistic and MRI volumes. Anatomical and functional images weremerged to allow direct localization on the MRI images of t statisticpeaks identified by an automatic peak-detection algorithm.

The significance of a given change in rCBF was assessed by ap-plication of an intensity threshold to the t statistic images (Worsleyet al., 1992). This threshold, based on 3-D Gaussian random fieldtheory, predicts the likelihood of obtaining a false positive in an ex-tended 3-D field. For an exploratory search involving all peaks withinthe gray matter volume of 600 cm3 or 200 resolution elements (re-sels), the threshold for reporting a peak as significant was set at t =3.5, corresponding to an uncorrected probability of p < 0.0002 (onetailed). Correcting for multiple comparisons, a t value of 3-5 yields afalse positive rate of only 0.58 in 200 resels (each of which hasdimensions 20 X 20 X 7.6 mm), which approximates the volume ofcortex scanned. For the directed search within the mid-dorsolateraland mid-ventrolateral frontal regions for predicted activation foci inspecific cytoarchitectonic areas, we selected a search volume of 150cm3 or 50 resels. On this basis, the threshold for significance withinthese regions was set at a conservative value of t = 3.00, correspond-ing to an uncorrected probability of p < 0.0013.

SubjectsSixteen normal right-handed volunteer subjects, eight male and eightfemale, participated in the study. Each subject underwent seven, 60sec PET scans within a single session and an MRI scan on a differentday. Six of the seven scanning conditions administered pertain to thecurrent study. The ages of the subjects ranged from 21 to 25 years(mean age, 21.56 yrs). All subjects gave informed, written consentfor participation in the study after its nature and possible conse-quences were explained to diem. The study was approved by theEthics Committee of the Montreal Neurological Institute.

Table 1Stereotaxic coordinates of activation obtained when spatial monitoring I was compared with the Con-trol condition

Region

Spatial monitoring I minus control condition

Right hemisphereMid-dorsolateral frontal cortex (area 9/46)Premotor cortex (area 6|Sensorimotor cortex (area 1/2)Intraparietal sulcal cortex (area 7/40)Posterior parietal cortex (area 7)Primary visual cortex (area 17)Parieto-occipital sulcus (area 7/19)Primary visual cortex (area 17)

Left hemispherePremotor cortex (area 6)Posterior parietal cortex (area 7)Medial posterior parietal cortex (area 7)Primary visual cortex (area 17)

Control condition minus spatial monitoring I

Right hemisphereCingulate cortex (area 32)Ventrolateral frontal cortex (area 45)Inferior frontal cortex (area 44)Insula

Left hemisphereFrontopolar cortex (area 10)Orbitofrontal cortex (area 11)Ventrolateral frontal cortex (area 45)Supplementary motor cortex (area 6)Middle temporal cortex (area 21)Posterior parietal cortex (area 39)

Activation foci in this and the other tables represent peaks of statistically significant (see text) changesin normalized rCBF. The stereotaxic coordinates are expressed in mm. x, medial-to-lateral distancerelative to the midline (positive = right hemisphere); y, anterior-to-posterior distance relative to theanterior commissure (positive = anterior); z, superior-to-inferior distance relative to the anterior com-missure-posterior commissure line (positive = superior). Significance level is given in (test units (seeMaterials and Methods for details).

Stimuli and Testing ConditionsThe stimuli used in all six conditions of this study were coloredcircles (1 cm in radius) presented, on a black background, on a high-resolution, touch-sensitive screen. The screen was suspended approx-imately 50 cm above the subject and was therefore within comfort-able reach. The order in which the six conditions were administeredwas randomly arranged across subjects with the restriction that notwo subjects performed the tasks in the same order. Each PET scanlasted 60 sec and testing on the task was initiated 10 sec beforescanning began. All subjects completed the same fixed number oftrials in each condition, the performance lasting for approximately90 sec in total. Performance data were collected during this 90 secperiod. The scans were separated by approximately 10 min, duringwhich time the requirements of the task to be administered in thenext scanning condition were explained to the subject and practiceproblems were administered to ensure that the task had been fullyunderstood.

There were five experimental conditions and one control condi-

Stereotaxic coordinates

X

3530544427163113

-28-21-12-17

8474736

-9-21-51-4

-55-56

Y

303

- 2 6 .-40-62-64-76-83

1-49-69-74

30226

-2

634218

-9-54-57

Z

295041515412305

54455411

-116

110

12-9

2625

29

fstau'stic

3.095.693.977.157.786.725.195.45

5.014.458.094.87

5.304.985.463.58

5.715.385.224.694.275.32

Figure 2. Spatial monitoring II minus control condition: merged PET-MRI sections illustrating rCBF increases averaged for all 16 subjects. The schematic outline of the brain indicates,in red, the level (y-coordinate) of the coronal sections rostral to the anterior commissure. The green dots indicate the sites of activation within the mid-dorsolateral and ventrolateralfrontal cortex presented in these sections. The subject's left is on the left side of the images. The top right section (y = +37) shows activation within the right mid-dorsolateralfrontal cortex (area 46). The bottom left (y = +20) and bottom right (y = +24) images show bilateral activation within the ventrolateral frontal cortex (area 47). The activation fociwithin the ventrolateral frontal cortex are located directly below the horizontal ramus, which is marked on the images with a white arrow. Significant activation foci within thepremotor cortex (area 6), the mid-dorsolateral frontal cortex (area 9), and the anterior cingulate cortex are also clearly visible on these lower images.Figure 1 Spatial span minus control condition: merged PET-MRI sections illustrating rCBF increases averaged for all 16 subjects. The schematic outline of the brain indicates thelevel (red lines) of the coronal sections shown and the green dot indicates the site of activation within the ventrolateral frontal cortex (area 47) presented in those sections (y =+20 and +24). The subject's left is on the left side in these images. Note that the activation foci are located directly below the horizontal ramus, which is marked on the imageswith a white arrow. A significant activation focus within the premotor cortex (area 6) is also clearly visible on these images.

32 Spatial Working Memory and the Frontal Lobe • Owen et al.

Cerebral Cortex Jan/Feb 1996. V 6 N 1 33

Table 2Stereotaxic coordinates of activation obtained when spatial monitoring II was compared with thecontrol condition

Table 3Stereotaxic coordinates of activation obtained when spatial monitoring III was compared with thecontrol condition

Region

Spatial monitoring II minus control conditionRight hemisphere

Frontopolar cortex (area 10)Mid-dorsolateral frontal (area 9)Mid-dorsolateral frontal (area 46)Anterior circulate cortex (area 32)Ventrolateral frontal cortex (area 47)Premotor cortex (area 46)Posterior parietal cortex (area 7/40)Posterior parietal cortex (area 7)Posterior parietal cortex (area 7/19)Lateral prestriate cortex (area 18)Primary visual cortex (area 17)Medial prestriate cortex (area 18)

Left hemisphereVentrolateral frontal cortex (area 47)Premotor cortex (area 6)Medial posterior parietal cortex (area 7)Posterior parietal cortex (area 7)Lateral prestriate cortex larea 19)Primary visual cortex (area 17)Lateral prestriate cortex (area 19)

Control condition minus spatial monitoring IIRight hemisphere

Primary motor cortex (area 4)Superior longitudinal fasc.

Left hemisphereFrontopolar cortex (area 10)Medial orb'rtofrontal cortex (area 11)Anterior cingulate cortex (area 32)Ventrolateral frontal cortex (area 45)

InsulaPostcentral cortex (area 43)Middle temporal cortex (area 21)

Stereotaxic coordinates

X

3439395

35254020253489

-29- 2 7

- 5-20-38-11-24

4834

-13-17-4

-51

-44-59

Y

512537251810

-44-64-73-83-88-90

201

- 6 9-69-71-81-85

1-30

65393632

«1

- 1 4-52

Z

6362036-356515338235

20

- 15453440

1227

1127

18-9-5

34 1

1 1206

tistic

1533.524.861551756.476.538.016.266.367.165.43

1545.858.155.804.566.114.95

6.45174

6.026.307.125.25C AQ3.40

4.775,90

Region

Spatial monitoring III minus control condition

Right hemispherePremotor cortex (area 6)Anterior paracingulate cortex (area 32)Mid-dorsolateral frontal (area 9/46)Mid-dorsolateral/frontopolar cortex (area 9/10)Parieto-occipital sulcus larea 7/40)Precuneus (area 31)Lateral prestriate cortex (area 19)Primary visual cortex (area 17)

Left hemisphereVentrolateral frontal cortex (area 47)Premotor cortex (area 6)Posterior parietal cortex (area 7)Primary visual cortex larea 17)Lateral prestriate cortex (area 19)

Control condition minus spatial monitoring III

Right hemisphereVentrolateral frontal cortex (area 47)Ventrolateral frontal cortex (area 45)Primary motor cortex (area 4)Superior temporal sulcus (area 21/22)Inferior parietal cortex (area 40)

Left hemisphereMedial orbitofrontal cortex (area 11)Ventrolateral frontal cortex larea 45)Cingulate cortex (area 32)Ventrolateral frontal cortex (area 45)InsulaSuperior temporal cortex (area 22)Middle temporal cortex larea 21)

See Table 1 note.

Stereotaxic coordinates

X

293

312840212713

-28-24-7

-12-29

394747604

-17-47-3

-46-31-64-58

Y

6243749

-42-59-85-85

205

-69-76-85

3932-1

-31-31

393732203

-21-52

Z

5139239

5023223

-354531024

-86

123

27

-123

-5121105

tistic

6.761605.143.406.285.456.609.01

3.275.428.507.365.94

3.783.836.665.884.10

7.536.417.085.076.824.395.96

See Table 1 note.

tion in this study. Three of the five experimental tasks required var-ious degrees of monitoring and manipulating of spatial informationin working memory. We refer to these three conditions as spatialmonitoring I-HI, respectively. All three spatial monitoring conditionsrequired that the subject monitor the contents of working memoryin order to make various judgements, such as whether particular stim-uli had or had not been presented (spatial monitoring I) or whichones of a set of denned locations had already been selected (spatialmonitoring n and HI). These tasks have mnemonic requirements thatare similar to those that have been shown to be critical in accountingfor the impairment after mid-dorsolateral frontal lesions in the mon-key (Petrides, 1991a,b, 1995). It was predicted that performance ofthese tasks would result in a greater blood flow response within themid-dorsolateral frontal cortex in comparison with the control task.The other two experimental conditions also involved spatial •workingmemory, but these tasks were not expected to activate mid-dorsolat-eral frontal cortex, but, rather, the ventrolateral frontal region. One ofthese tasks, the spatial span task, required that the subject remembera given sequence of spatial locations and then reproduce them im-mediately afterward. The other task, the fixed spatial sequence, re-quired that the subject reproduce a fixed sequence of locations thathad been learned prior to scanning. Note that these two tasks do notrequire any manipulation of spatial information within workingmemory but only the retention of a perceived or learned spatial se-quence and the organization of its execution. The fixed spatial se-quence task was similar to a fixed sequence tasks that monkeys withmid-dorsolateral frontal lesions perform normally (Petrides, 1995).

Finally, there was a control condition diat provided a baselineagainst which to examine the extent of activation within the frontalcortex in the other five experimental conditions. The control condi-tion had similar visual, spatial, and motor requirements as the five

experimental tasks, except for their specific mnemonic requirements.Eight identical red circles were presented, scattered randomly, exceptfor one of these circles that occupied the central location. Once ev-ery second the central circle changed its color from red to blue and,at this point, the subjects were required to touch it. Once touched,the central circle turned red again.

Spatial Monitoring IWithin each trial, three blue circles were presented on the computerscreen, one at a time and in random locations. Each of these circlesremained on the screen for 0.25 sec and then disappeared as thenext circle appeared. After the third circle had been presented, adelay of 3 sec ensued, during which time the screen remained blank.At the end of the delay period, eight red circles appeared simulta-neously on the screen. The location occupied by three of these cir-cles was identical to the location of the three blue circles shownbefore the 3 sec delay, the remaining five being randomly positioned.The subjects were required to touch, in any order they wished, eachone of the three locations that were presented before the delay. Im-mediately after the third response, the screen cleared for 1 sec andthe next trial began. Within the testing period, each subject com-pleted 12 trials. Note that the requirements of this task are similar tothose of the nonspatial working memory tasks that monkeys withmid-dorsal lateral frontal lesions fail (Petrides, 1991a, 1995) and, inthis sense, it is procedurally quite different from the spatial span taskand fixed spatial sequence task described below. Thus, during stim-ulus presentation, each of the three locations to be remembered isselected randomly from a large number of possible positions on thescreen. At this stage, the subject has no information about the subsetof eight locations from which these three target stimuli are to beselected. Consequently, during recall, subjects must consider each ofthese eight locations in turn, and decide, with reference to the con-tents of working memory, whether each location has been presentedpreviously or not. The fact that the three target stimuli cannot be

34 Spatial Working Memory and the Frontal lobe • Owen et al.

Table 4Stereotaxic coordinates of activation obtained when spatial span was compared with the controlcondition

Region

Spatial span minus control conditionRight hemisphere

Ventrolateral frontal cortex (area 47)Premotor cortex (area 6)Posterior parietal cortex (area 7)Posterior parietal cortex (area 7)Primary visual cortex (area 17)Medial prestriate cortex (area 18)

Left hemispherePremotor cortex (area 6)Medial posterior parietal cortex (area 7)Precuneus (area 7)Primary visual cortex (area 17)Lateral prestriate cortex (area 18)

Control condition minus spatial spanRight hemisphere

Ventrolateral frontal cortex (area 45)Primary motor cortex (area 4|Superior temporal cortex (area 22)

Left hemisphereFrontopolar cortex (area 10)Ventrolateral frontal cortex (area 45)Superior temporal cortex larea 22)InsulaSupplementary motor area (area 6)Superior temporal cortex larea 22)Middle temporal cortex (area 21)Posterior parietal cortex (area 39)

Stereotaxic coordinates

X

362524271311

-21-12-19-12-38

464860

-11-46-46-34-4

-47-55-55

Y

208

-64-73-83-90

5-69-69-73-85

393

-30

65418

-1-11-28-50-57

Z

-55056353

20

5954249

17

3115

146

-171160125

29

tjstic

1474.447.786.725.886.23

4318.683.636.523.70

4.655.643.80

4.984.753.774.534.484.583.896.14

See Table 1 note.

encoded as part of a known array effectively removes the tendencyfor subjects to simply remember the stimuli as a spatial sequence, astrategy that clearly is used in the spatial span task included in thisstudy (see below). Analysis of response patterns during the spatialmonitoring I task supports this suggestion and demonstrates that sub-jects tend to respond according to the position of target stimuli onthe screen (i.e., left to right), rather than according to the temporalorder in which they were presented.

Spatial Monitoring IIThis task is based directly on one used previously to assess spatialworking memory in neurosurgical patients with frontal or temporallobe damage (Owen et al., 1990,1995). On each trial, eight red circleswere presented in random locations on the screen. The subjects wererequired to "search through" these red circles by touching each oneof them until one of the touched circles turned blue. The circle, then,returned to its original red color and the subject was required toinitiate a new "search" through the circles until another one turnedblue. The subjects knew that once a particular location had turnedblue, it would never turn blue again and therefore the point of thetask was to avoid touching locations that had turned blue on earlier"searches." The subjects could search the boxes in any order theywished, but were explicitly instructed to search in random fashionand not to use any systematic spatial strategies. When all eight loca-tions had turned blue, a new random arrangement of the eight redcircles was presented, and the subject had to "search though" them,as before, until each one had turn blue. In order to keep the numberof searches consistent across individuals, the number of locationsvisited before each location turned blue was determined by the com-puter. Within the testing period, each subject completed four trials,each of which required eight searches. Thus, in this spatial monitor-ing condition (and also in spatial monitoring HI; see below), the sub-ject was required to refer to a continually updated on-line record ofwhich of a defined set of previously selected locations had beenmarked with a blue circle.

Spatial Monitoring IIIThis condition was exactly the same as the spatial monitoring II con-dition described above, except for the fact that 12, rather than 8,spatial locations were used. Consequently, within each trial, 12searches were required and a total of 12 circles would have to turn

. blue to complete the trial. Within the testing period, each subjectcompleted three trials, each of which required 12 searches.

Spatial SpanThis task was based directly on the Corsi block tapping test de-scribed by Milner (1971). On each trial, eight red circles were pre-sented in random locations on the screen. One of these red circleswould then turn blue, for 0.5 sec, before returning to its red color.Another circle would then turn blue in the same manner until fiveof the eight stimuli had changed color in this way. Immediately fol-lowing the presentation of the fifth stimulus, the subjects were re-quired to touch each of these five "target" locations in any order theywished. After five responses by the subject; the screen was clearedand the next trial began with the eight red circles occupying newlocations, five of which would sequentially turn blue. Within the test-ing period, each subject completed seven trials.

Note that in this spatial span task, the subject has merely to watcha spatial sequence, hold it in short-term memory and program itsreproduction. In this sense, it is quite different from the spatial mon-itoring I task described above in that, during presentation, the fivetarget stimuli are encoded as a subset of a known array. This provi-sion effectively encourages subjects to encode the target stimuli asa sequence, and then reproduce that sequence in exactly the sametemporal order. Thus, no manipulation of the stored sequence is re-quired. Analysis of response patterns during the spatial span tasksupports this suggestion and demonstrates that subjects invariablyreproduce the target stimuli in the same temporal order in whichthey were presented.

Fixed Spatial SequenceIn this condition, a single array of eight randomly positioned redcircles were presented on the computer screen. Prior to scanning,the subjects learned to touch each one of these red circles in a fixedrandom sequence. Scanning did not begin until each subject couldreproduce this sequence perfectly from memory, five times in a row.During scanning, the subjects were required simply to reproduce thesequence by touching each one of the circles in the correct order,returning to the beginning once the end of the sequence had beenreached. When a circle was touched, it changed color from red toblue for 0.5 sec and then returned to red to indicate that the nextcircle in the sequence should be touched. In this way, the rate ofresponse was controlled and kept approximately the same as in theother conditions. Within the testing period, each subject completedseven trials. Again, this task requires very little monitoring or manip-ulation of information within working memory but simply the reten-tion and reproduction of a learned sequence of responses.

Results

PerformanceThe proportion of trials completed without an error on eachone of the five behavioral tasks is presented in Figure 1. Ofthe three monitoring tasks, the first one was clearly the easi-est, with subjects making very few errors, whereas the spatialmonitoring TJ (eight boxes) and spatial monitoring in (12boxes) tasks were more difficult, the subjects completing 73%and 65%, respectively, of all trials •without an error. During thespatial span task, 96% of all trials were completed without anerror, which is exactly the same as the mean score for thespatial monitoring I condition. Finally, in the fixed spatial se-quence task, 98% of all performed sequences were withouterror.

Blood FlowThis study was designed to permit specific previously de-signed comparisons, accomplished via subtractions, betweeneach one of the five experimental conditions and the controlcondition. The results of these subtractions, in terms of statis-

Cerebral Cortex Jan/Feb 1996, V 6 N 1 35

Stereotaxic coordinates

X

502512161611

-26-9

-20-1

-15

526262

-48-3

-59-56

Y

24-62-66-74-76-85

18-69-71-74-78

6-30-52

39-14-45-52

Z

-9549

425

20

-554422411

125

11

-314275

r statistic

3.384.446355.S66.865.28

3.166.551695.675.84

3.964.283.91

4.673.853.724.45

Table 5Stereotaxic coordinates of activation obtained when the fixed spatial sequence condition was com-pared with the control condition

Region

Fixed spatial sequence minus control conditionRight hemisphere

Ventrolateral frontal cortex (area 47)Posterior parietal cortex (area 7)Primary visual cortex (area 17)Posterior parietal cortex (area 7)Lateral prestriate cortex (area 18)Medial prestriate cortex (area 18)

Left hemisphereVentrolateral frontal cortex (area 11/47)Medial posterior parietal cortex (area 7)Posterior parietal cortex (area 7)Medial prestriate cortex (area 18)Primary visual cortex (area 17)

Control condition minus fixed spatial sequenceRight hemisphere

Precentral cortex (area 6)Superior temporal cortex (area 22)Superior temporal sulcus (area 21/22)

Left hemisphereVentrolateral frontal cortex (area 10/47)ThalamusPosterior parietal cortex (area 40)Middle temporal cortex (area 21)

See Table 1 note.

tically significant changes in rCBF are given in Tables 1-5,together with the corresponding stereotaxic coordinates.These coordinates are based on the system used in the brainatlas of Talairach and Tournoux (1988).

When activity in the spatial monitoring I condition wascompared with that in the control condition, there was sig-nificantly greater rCBF, as predicted, in the mid-dorsolateralfrontal cortex (i.e., areas 46 and 9) in the right hemisphere(Table 1). Other significant rCBF changes were located mainlyin primary visual cortex, posterior parietal region, and pre-motor cortex bilaterally. No significant changes in blood flowwere observed in the mid-ventrolateral region of the frontalcortex following this subtraction, the maximum t value ob-served in this region of cortex being t = 0.45 (see summaryin Table 6).

The results of the comparisons between spatial monitoringII and the control conditions (Table 2, Fig. 2) and betweenspatial monitoring III and the control conditions (Table 3) arevery similar and will be discussed together. Again, our predic-tions were clearly confirmed as both the spatial monitoringII (eight-location) and the spatial monitoring HI (12-location)tasks resulted in significantly greater rCBF in the right mid-dorsolateral frontal cortex (areas 46 and 9). In both condi-tions, significant changes in blood flow were also observed inright frontopolar cortex (area 10) and in ventrolateral frontalarea 47. In the eight-location version of the task this changewas clearly bilateral (see Fig. 2), while in the 12-location ver-sion it failed to reach statistical significance in the right hemi-sphere (f = 2.66). Again, nonfrontal peaks of activation werelargely confined to visual cortical areas, the posterior parietalregion, and lateral premotor cortex. Finally, in both conditionsa significant change in blood flow was observed within area32 of the cingulate region.

The results of the comparison between the spatial spancondition and the control condition (Table 4, Fig. 3) also clear-ly confirmed our prediction that a significant change in bloodflow -would be observed in the mid-ventrolateral frontal cor-

Table 6Summary of results

Maximum f value

Mid-dorsolateralfrontal cortex

Mid-ventrolateralfrontal cortex

Spatial monitoring ISpatial monitoring IISpatial monitoring IIISpatial spanFixed spatial sequence

( = 3.09r = 4JS( = 5.14t = 1.58 NSr = 1.53 NS

r = 0.45NSr=i75r=lZ7*1 = 147[=138

NS, nonsignificant All peaks reported are in the right hemisphere except the one marked with anasterisk (*), which was in the left hemisphere. The i values shown represent the maximum valuesobserved in the mid-ventrolateral and mid-dorsolateral frontal regions when the control condition wascompared with each one of the five experimental tasks. See also Table 1 note.

tex (area 47) in the right hemisphere but not in the mid-dorsolateral frontal region. The maximum t value observed inthe mid-dorsolateral frontal cortex was t = 1.58 for this com-parison (see summary in Table 6). Elsewhere in the cortex,the peaks of activation observed were very similar to thoseseen in the other subtractions, namely occipital visual areas,the posterior parietal cortex, and in the lateral premotor cor-tex.

The comparison between the fixed spatial sequence con-dition and the control condition revealed a significant changein blood flow in mid-ventrolateral frontal cortex, as predicted(Fig. 4, Table 5). Again, t values in the mid-dorsolateral frontalcortex were low, the maximum value being t = 1.53 (seesummary in Table 6). Other significant regions of activationwere observed again, in occipital visual areas and in the pos-terior parietal cortex.

DiscussionThe present study used positron emission tomography withmagnetic resonance imaging to demonstrate the existence,within the human brain, of two functionally distinct subdivi-sions of the lateral frontal cortex that subserve different as-pects of spatial working memory. Five spatial memory taskswere used that were similar in the type and mode of stimuluspresentation and response required, but differed in terms oftheir executive processing requirements. The tasks were de-signed to test the hypothesis that the middle sections of thedorsolateral and ventrolateral frontal cortex are differentiallyinvolved in executive processes (Petrides 1991a,b, 1994). Thefirst major issue addressed in the present investigation waswhether frontal activation would be confined to the mid-ven-trolateral region of die frontal cortex when the experimentaltask (in comparison with the control task) required the or-ganization and execution of a remembered series of spatialmoves. In the spatial span task, the subject was required toorganize a sequence of spatial moves in order to reproducea sequence of spatial stimuli previously presented and cur-rently held in working memory. In the fixed spatial sequencetask, the subject had to organize and guide the execution ofa fixed learned sequence of moves. As predicted, in compar-ison with the control task that did not require the organiza-tion of a sequence of responses, but controlled for the motorcomponent of reaching and touching die stimuli displayed,both the spatial span and the fixed spatial sequence tasksactivated ventrolateral frontal cortical area 47 (Tables 4, 5;Figs. 3,4): It is important to note here that, in the spatial spantask, the subject is simply required to hold the presented se-quence of moves in working memory until its reproductionis organized and executed. As can be seen from the summaryin Table 6, this requirement was clearly not sufficient to yieldany significant activation in mid-dorsolateral frontal areas 46

36 Spatial Working Memory and the Frontal Lobe • Owen et al.

Figure 4. Hxed spatial sequence minus control condition: merged PET-MRI sections illustrating rCBF increases averaged for all 16 subjects. The schematic outline of the brainindicates the level {red lines) of the coronal sections shown and the green dot indicates the site of activation within the ventrolateral frontal cortex (area 47) presented in thesesections (y = +20 and +24). The subject's left is on the left side of images. The activation foci are located directly below the horizontal ramus, which is marked on the imageswith a white arrow.

and 9 when either the spatial span task or the fixed spatialsequence task were compared with the control condition.

The second major question addressed in the present in-vestigation was whether there would be activation within themid-dorsolateral frontal cortex (i.e., areas 46 and 9) when theexecutive requirements of the spatial working memory taskswere changed to increase the monitoring and manipulationof information required within working memory. By monitor-ing, we refer to the active checking of information as occurs,for example, when the subject is required to decide which ofa given array of locations have been selected and which oneshave not (e.g., spatial monitoring I). The three monitoringtasks (spatial monitoring I-ffl) differed in terms of the num-ber of such moves that had to be monitored, but they allinvolved this process to a considerable extent. In all threetasks, activation foci were observed in mid-dorsolateral frontalcortex (i.e., areas 46 and 9) when blood flow in these con-ditions was compared with the control condition (see Table6). Furthermore, there were clear differences in the extent towhich mid-ventrolateral frontal cortex was also activated. Thespatial monitoring I task required that the subjects decidewhich ones of a set of eight stimuli had been presented ear-lier, but, unlike the spatial span task (see Materials and Meth-

ods), had minimal requirements (compared with the controlcondition) in terms of the organization of a sequence of re-sponses. As expected, this task activated only mid-dorsolateralfrontal cortex (see Table 6). By contrast, the spatial monitor-ing II and III tasks required that the subject monitor whichones of a set of spatial stimuli had been marked with a bluecircle and also that a sequence of selections through the setbe organized and executed. These tasks activated both mid-dorsolateral and mid-ventrolateral frontal cortex. It is impor-tant to note that task difficulty was not related to the ob-served activation of mid-dorsolateral or ventrolateral frontalcortex. For instance, the proportion of trials completed with-out an error was the same in the spatial monitoring I and thespatial span conditions, yet one of these conditions resultedin activation in mid-dorsolateral frontal cortex and the otherin the mid-ventrolateral frontal cortex.

In light of these findings, some of the apparently conflict-ing results from previous functional neuroimaging studies ofspatial working memory can be understood. For example, Mc-Carthy et al. (1994) used functional MRI to measure changesin rCBF while subjects judged whether each of a series of 14or 15 stimuli was located in a position that had already beenoccupied earlier in the sequence. The single slice chosen for

Cerebral Cortex Jan/Fcb 1996, V 6 N 1 37

the functional MRI study allowed examination of frontal areas9, 46, 23, and often 47, and revealed a significantly increasedMR signal in area 46 across all eight subjects tested. This par-adigm is conceptually similar to the spatial monitoring II andIII conditions used in the present study, and therefore the twostudies concur with respect to the observed activation fociin area 46. Contrasting results, however, have been reportedby Jonides et al. (1993), who used PET and a behavioral taskthat was intended to simulate a paradigm that has been usedto test spatial working memory in nonhuman primates (Fu-nahashi et al., 1989, 1990). The subjects were required to re-member the location of three dots presented simultaneouslyon a computer screen for a delay period of 3 sec and then todecide whether or not a probe circle was presented in oneof those same three locations. Within the frontal cortex, sig-nificant changes in blood flow were observed, ventrolaterally,in area 41 in the right hemisphere. Interestingly however, nosignificant activity was reported in area 46 in the mid-dorso-lateral frontal cortex. Unlike the task described by McCarthyet al. (1994) and the three spatial monitoring conditions in-cluded in the present experiment, the task used by Jonideset al. (1993) had minimal monitoring requirements. It simplyrequired that the subject carry out an active comparison be-tween the probe presented and the information stored inworking memory in order to decide whether or not the twomatched. Our results concur fully therefore with both formerstudies and clearly demonstrate that both ventrolateral anddorsolateral frontal areas can be activated in spatial workingmemory tasks, depending on the precise executive processesthat are called upon by the task being performed. Thus,whereas the ventrolateral frontal cortex appears to be in-volved when working memory tasks require comparisonswith, or reproduction of, stored information, the mid-dorso-lateral frontal cortex becomes involved when active decisionsabout the occurrence or nonoccurrence of stimuli from a giv-en set are required.

NotesWe thank the staff of the McConneU Brain Imaging Center and theMedical Cyclotron Unit for assistance with this study and P Neelin,S. Milot, and E. Meyer for technical expertise and advice. This workwas supported by the McDonnell-Pew Program in Cognitive Neuro-science and by the Medical Research Council (Canada) Special Pro-ject Grant SP-30.

Address correspondence to Adrian M. Owen, Neuropsychology/Cognitive Neuroscience Unit, Montreal Neurological Institute, 3801University Street, Montreal, Quebec, H3A 2B4, Canada.

ReferencesEvans AC, Thompson CJ, Marrett S, et al. (1991a) Performance char-

acteristics of the PC-2048: a new 15 slice encoded crystal PETscanner for neurological studies. IEEE Transact Med Imaging 10:90-98.

Evans AC, Marrett S.TorrescorzoJ.et al. (1991b) MRI-PET correlativeanalysis using a volume of interest (VOI) atlas. J Cereb Blood FlowMetab 11:A69-A78.

Evans AC, Marrett S, Neelin P, et al. (1992) Anatomical mapping offunctional activation in stereotactic coordinate space. Neuroimage1:43-63-

Fox FT, Raichle ME (1984) Stimulus rate dependence of regionalcerebral blood flow in human striate cortex, demonstrated withpositron emission tomography. J Neurophysiol 51:1109-1121.

Fox FT, Perimurter JS, Raichle ME (1985) A stereotactic method ofanatomical localization for positron emission tomography. J Corn-put Assist Tomogr 9:141-153.

Funahashi S, Bruce CJ, Goldman-Rakic PS (1989) Mnemonic codingof visual space in the monkey's dorsolateral prefrontal cortex. JNeurophysiol 61:1-19.

Funahashi S, Bruce CJ, Goldman-Rakic PS (1990) Visuospatial codingof primate prefrontal neurons revealed by oculomotor paradigms.J Neurophysiol 63:814-831.

Fuster JM (1989) The prefrontal cortex. Anatomy, physiology andneuropsychology of the frontal lobe. New York: Raven.

Goldman-Rakic PS (1987) Circuitry of primate prefrontal cortex andthe regulation of behavior by representational memory. In: Hand-book of physiology, Sec 1, Vol 5, The nervous system (Plum, F,Mountcastle V, eds), pp 373-417. Bethesda, MD: American Physi-ological Society.

Jonides J, Smith EE, Koeppe RA, Awh E, Minoshima S, Minrun MA(1993) Spatial working memory in humans as revealed by PET.Nature 363:623-625.

McCarthy G, Blamire AM, Puce A, Nobre AC, Bloch G, Hyder F, Gold-man-Rakic P, Shulman RG (1994) Functional magnetic resonanceimaging of human prefrontal cortex activation during a spatialworking memory task. Proc Natl Acad Sci USA 91:8690-8694.

Milner, B (1971) Interhemispheric differences in the localization ofpsychological processes in man. Br Med Bull 27:272-277.

Owen AM, Downes JD, Sahakian BJ, Polkey CE, Robbins TW (1990)Planning and spatial working memory following frontal lobe le-sions in man. Neuropsychologia 28:1021-1034.

Owen AM, Sahakian BJ, Semple J, Polkey CE, Robbins TW (1995)Visuo-spatial short term recognition memory and learning aftertemporal lobe excisions, frontal lobe excisions or amygdalo-hip-pocampectomy in man. Neuropsychologia 33:1-24.

PetridesM (1991a) Monitoring of selections of visual stimuli and theprimate frontal cortex. Proc R Soc Lond [Biol] 246:293-298.

Petrides M (1991b) Functional specialization within the dorsolateralfrontal cortex for serial order memory. Proc R Soc Lond [Biol]246:299-306.

Petrides M (1994) Frontal lobes and working memory: evidencefrom investigations of the effects of cortical excisions in nonhu-man primates. In: Handbook of neuropsychology, Vol 9 (Boiler F,Grafman J, eds), pp 59-82. Amsterdam: FJsevier.

Petrides M (1995) Impairments on nonspatial self-ordered and ex-ternally ordered working memory tasks after lesions of the mid-dorsal lateral frontal cortex in the monkey. J Neurosci, in press.

Petrides M, Milner B (1982) Deficits on subject-ordered tasks afterfrontal- and temporal-lobe lesions in man. Neuropsychologia 20:249-262.

Petrides M, Alivisatos B, Evans AC, Meyer E (1993) Dissociation ofhuman mid-dorsolateral from posterior dorsolateral frontal cortexin memory processing. Proc Natl Acad Sci USA 90:873-877.

Raichle JE, Martin WRW, Herscovitch P, Mintum MA, Markham J(1983) Brain blood flow measured with intravenous H2"O. II. Im-plementation and validation. J Nucl Med 24:790-798.

Talairach J, Tournoux P (1988) Co-planar stereotactic atlas of thehuman brain: 3-dimensionaI proportional system: an approach tocerebral imaging. Stuttgart: TTiieme.

Worsley KJ, Evans AC, Marrett S, Neelin P (1992) Determining thenumber of statistically significant areas of activation in subtractedactivation studies from PET. J Cereb Blood Flow Metab 12:900-918.

38 Spatial Working Memory and the Frontal Lobe • Owen et al.