Examining encoding imprecision in spatial working memory in schizophrenia

1. Introduction 

Working Memory deficits are considered a cardinal cognitive feature of schizophrenia (Park and Holzman, 1992, Lee and Park, 2005). Spatial working memory (SWM) deficits in particular have been proposed as potential cognitive endophenotypic markers of schizophrenia, consequently SWM tasks are increasingly likely to be used as tools to detect individuals at risk for schizophrenia and in research aimed at developing preventative interventions (Brewer et al., 2006, Cannon et al., 2000, Saperstein et al., 2006, Smith et al., 2006). Equally important, the neurophysiological basis of some SWM tasks, such as delayed-response (DR) tasks, is reasonably well delineated (Park and Lee, 2002).

DR tasks used in schizophrenia research typically involve the brief presentation and removal of a simple visual target followed by a series of probe target locations after a variable delay. Subjects engaged in these tasks must, therefore, make use of an internal representation of the original stimulus to guide their responses. Correct responding on DR tasks may, then, be determined by two separable cognitive processes, namely encoding (primarily linked to occipitoparietal cortex) and maintenance (primarily mediated by prefrontal cortex) (Tek et al., 2002). Whilst maintenance entails sustaining activated representations in WM, encoding comprises the ability to attend to and select a stimulus for entry into WM and activate its internal representation (Hartman et al., 2002). Recent studies have endeavoured to determine more precisely those processes which are, or are not, affected as part of the SWM deficit in schizophrenia; specifically, whilst attentional control of WM encoding appears to be intact (Gold et al., 2006) other studies tend to converge on the notion that there is probably disruption/loss of spatial representations during encoding of spatial information (Glahn et al., 2003, Lee and Park, 2005, Park, 1999, Tek et al., 2002).

SWM is not a unitary “sketch pad” (Baddeley, 1986); for example, there is increasing evidence for separable subsystems for processing visual identity and spatial location in working memory (Logie, 1995, Darling et al., 2006). Whilst it is still a matter of debate, there also appears to be a differential response of prefrontal cortical regions supporting the maintenance of information in WM; specifically, visual and spatial representations preferentially involve ventral and dorsal regions respectively (Sala and Courtney, 2007). Furthermore, there is now strong evidence for multiple representations of spatial location in SWM, including independent encoding of target distance and direction (radius and angle in polar space) (Chieffi and Allport, 1997). Huttenlocher et al. (1991) proposed that encoding of distance and direction information in memory entails at least two levels of detail; a fine-grained value and a category level. The former comprises precise representation of stimulus location based on distance and direction dimensions, whilst the latter involves estimation processes incorporating the spontaneous use of spatial reference frames, such as horizontal and vertical axes, and weighting by a prototypical value (Huttenlocher et al., 1991). Huttenlocher and colleagues proposed that memory for spatial location depends on the combined effects of both levels of representation: when memory load is low (e.g. when responses are immediate) the fine-grained level of representation has more influence whilst when memory load is increased (e.g. location information must be maintained over a delay) the category level has more weight. In the latter condition the category level of representation produces systematic biases, in both the radial and angular dimensions, when subjects are asked to briefly memorize the position of a target. For example, when remembering the location of a point in a larger display (e.g. a circle) subjects typically impose horizontal and vertical axes, dividing the circle mentally into quadrants. Delayed responses are typically misplaced towards a central (prototypic) location in each quadrant, i.e. biased away from the horizontal and vertical axes (Nelson and Chaiklin, 1980).

Previous DR studies of patients with schizophrenia have not considered the different dimensions (distance and direction) or levels (fine-grained or category) of spatial representation. More recently a small number of studies have reported deficits in terms of distance (only) from target, however, these studies have usually cued participants on delay with a limited set of probe response locations, which restricts the range of distance and direction responses produced, and inevitably introduces processes of selective attention and inhibition of distracter location which may differentially impair performance (e.g. Myles-Worsley and Park, 2002, Park, 1999, Park and O'Driscoll, 1996). Elsewhere, Javitt and colleagues have argued that impaired precision may underlie working memory dysfunction in schizophrenia (Javitt et al., 1999), however, these studies did not specifically examine spatial working memory.

The aim of the current study was to examine more specific sub-processes that might be underlying poor encoding of visuospatial targets in schizophrenia. The VisuoSpatial Working Memory (VSWM) test (Badcock et al., 2004a) is a modified DR task, which was developed to assess the accuracy of spatial representations by examining both distance and direction coding in working memory and the maintenance of spatial location codes over a short delay, in the absence of distracters. The VSWM test differs, therefore, from other popular measures such as the SWM task in the Cambridge Neuropsychological Test Automated Battery (Sahakian and Owen, 1992), which assesses attentional, memory capacity and strategic ability, and Delayed Match to Sample (DMTS) tasks which examine visual identity-based DR performance over variable delays.

We considered two separate (though not mutually exclusive) hypotheses regarding the processes that may contribute to poor SWM in schizophrenia: 1) reduced precision in only one dimension (e.g. distance) when coding spatial location, 2) reduced precision at only the fine-grained level of encoding spatial location. Furthermore since previous studies have suggested that poor encoding in working memory results from slowed information processing in schizophrenia (Hartman et al., 2002) the VSWM test controls for individual differences in the speed of encoding.

2. Method 

Ethics approval for this project was granted by the Human Research Ethics Committee, University of Western Australia and the Institutional Ethics Committee, Graylands Hospital. Written, informed consent was obtained from all participants prior to their participation in the study.

2.2. Stimuli and procedures 

The VSWM test was designed to assess the encoding and maintenance of visuospatial information over short delays, after controlling for individual differences in the speed of encoding. Participants were required to respond by pointing a stylus, as accurately as possible, to the visual location of target stimuli (white circles) presented on a touch sensitive screen. All movements began from a fixed starting hand position, in the midline, at the base of the screen (see Fig. 1).

View Large Image Download to PowerPoint

Fig. 1. Distribution of target locations in the VSWM test.

The speed of encoding of visual target locations was assessed first by establishing the target duration required to achieve 79% correct detection of spatial locations during a ‘pre-test’ phase and does not require a speeded response. The target stimulus (circle, 4.7 mm radius) was presented at variable (i.e. unpredictable) locations on the monitor: on each trial the target stimulus was presented at one of twenty possible locations. An auditory stimulus at target offset signalled to the participant to point to (touch) the target stimulus as accurately as possible. A rapid, adaptive staircase procedure was used to control stimulus duration (Edwards et al., 1998). This staircase was designed to begin with a target duration that is easily detectable (1166 ms or 70 frames), then rapidly drop to near threshold, which is then estimated with maximum precision (final step size=1 frame or 16.67 ms).

During the test phase two blocks of stimuli were presented at each participant’s individualized target duration for recall following either no delay (immediate) or 4-second delay (counterbalanced), reflecting the accuracy of encoding and maintenance respectively. The distribution of target positions for the VSWM test was determined by a hidden grid comprised of four concentric arcs and five radial arms (see Fig. 1). This configuration yielded twenty target positions. Stimuli were presented twice at each position (in random order); hence the total number of targets per block was forty. The advantage of this stimulus configuration was twofold. First, working memory for spatial location could be assessed for targets at different distances along the same angle of direction, conversely for targets at the same distance from the starting position but appearing at different directions. Second, participants remained unaware of the structured grid controlling target positions. Accuracy was assessed in terms of the discrepancy between target position and response coordinates relative to the starting hand position and reported separately for target distance (mm) and direction (°). Absolute (i.e. unsigned) direction and distance accuracy was used to assess fine-grained spatial representations, whilst signed accuracy values (+/−) reflect response biases produced by category level representations (+distance values represent undershoots, +direction values represent overshoots).

3. Results 

Table 1 includes basic demographic and general cognitive performance variables for the patient and control groups, together with group comparisons. The patient group contained a greater male to female ratio (20:1) than the control group (2.69:1). Patients were also significantly younger and had completed fewer years of education than controls. In addition, healthy controls demonstrated a higher estimated IQ than patients, based on either the NART-R or the Shipley estimate of intelligence, however, there were no significant group differences in the laterality quotient from the Edinburgh Handedness Inventory, reflecting a similar degree of handedness in these groups.

Table 1. Mean (SD) and group comparisons of demographic and general cognitive variables for schizophrenia patients and healthy controls

		Schizophrenia patients N=42	Healthy controls N=48	Group comparisons
	Gender	40 male	35 male	Chi square=8.03 sig.=.005
	Age	32.2 (8.68)	41.48 (13.29)	Mann Whitney U=595.00 sig.=.001
	Range	20–53	19–61
	Years of education	10.62 (1.49)	13.17 (2.50)	Mann Whitney U=379.00 sig.=.000
	Range	8–15	10–21
	Handedness			Mann Whitney U=540.50 sig.=.062
	Laterality quotient	56.24 (46.8)	72.43 (39.1)
	Shipley IQ	87.74 (12.21)	110.24 (6.65)	Mann Whitney U=124.00 sig.=.000
	NART-R IQ	94.02 (10.71)	109.84 (6.56)	Mann Whitney U=222.0 sig.=.000

Note: Two healthy controls had missing Shipley scores. One healthy control had missing NART-R score.

3.1. Speed of encoding 

Initial inspection of the detection threshold data indicated significant skewness and formal tests of normality were significant, consequently nonparametric analysis was employed throughout. On average, patients with schizophrenia required targets to be of longer duration in order to be correctly detected compared to healthy controls (Mann Whitney U=615.00, sig.=0.001; mean (SD)=811.96 (608.15) ms — schizophrenia patients; mean (SD)=426.95 (394.50) ms — healthy controls) consistent with a slower speed of encoding of visuospatial information. Longer stimulus durations required by patients were not simply due to age since, on average, controls were older than patients. There were also no significant within group correlations between threshold values for stimulus duration and age, measures of intelligence or education (all Spearman r values <.186).

In order to examine the potential impact of stimulus duration on target detectability the proportion of individuals in each group achieving criterion performance as stimulus duration increased was calculated, as illustrated in Fig. 2. For example, at stimulus duration of 500 ms it can be seen that 71% of healthy controls had achieved the detection threshold compared with only 40% of patients.

View Large Image Download to PowerPoint

Fig. 2. Impact of stimulus duration on target detection.

3.2. Spatial working memory 

Responses more than 40° or 48 mm from each target location (i.e. more than two dot locations away from the actual target location — see Fig. 1) were considered to be outliers. This resulted in fewer than 3% (patients) and 2% (controls) of all responses being removed from the analysis. In order to test the first hypothesis it was necessary to demonstrate that target distance and location were indeed coded independently and may, therefore, be subject to selective impairment. Within group correlation coefficients (Pearson's r) between individual signed direction and distance accuracy for all target location were calculated. At each delay, the correlation coefficients approximated zero (0-s: schizophrenia r=−.043, controls r=−.033; 4-s: schizophrenia r=−.070, controls r=.012), consistent with the notion of independent encoding of these dimensions in spatial working memory.

Summary variables were then computed. An average of the two responses gained at each location was calculated (only a single value was used if the other value was an outlier), followed by the mean (SD) absolute accuracy of direction and distance responses for each group, collapsed across all target locations (Table 2). Inspection of signed data (Fig. 3) suggested a differential pattern of undershoots and overshoots across the stimuli in the array, consequently signed data were examined separately at each angle and radius presented. The distribution of dependent variables was skewed and formal tests of normality were significant, consequently nonparametric analyses were employed.

Table 2. Mean (SD), group comparisons and effect sizes for absolute distance and direction accuracy (averaged across all target locations) between schizophrenia patients and healthy controls

		Schizophrenia patients	Healthy controls	Group comparisons	Effect size
	Absolute accuracy			Mann Whitney U	Cohens's d
	0 s Direction	1.1366 (0.975)	0.9220 (0.490)	842.00, sig.=.179	.278
	0 s Distance	3.9697 (1.332)	3.5284 (1.194)	833.00, sig.=.157	.349
	4 s Direction	2.0969 (0.875)	1.4967 (0.777)	557.00, sig.=.000	.725
	4 s Distance	6.3845 (2.060)	4.8234 (1.648)	542.00, sig.=.000	.837

View Large Image Download to PowerPoint

Fig. 3. Signed accuracy data at 0-second (upper panel) and 4-second (lower panel) delay across all target locations for schizophrenia patients and healthy controls.

3.2.1. 0-second delay (encoding) 

At the 0-second delay, there were no significant group differences in mean absolute accuracy (Table 2) of either direction or distance responses collapsed across all target locations suggesting that, when differences in speed of processing are controlled for, the precision of fine-grained representations encoded in working memory is intact in patients with schizophrenia. Accuracy immediately after stimulus offset, as shown in Fig. 3, was sufficient to ensure that responses from both groups fell closely within the area of the target stimulus.

Examination of signed accuracy averaged across each of the five angles of the stimulus array showed no significant group differences (50° Mann Whitney U=978.00, sig.=.808; 70° Mann Whitney U=983.00, sig.=.840; 90° Mann Whitney U=966.00, sig.=.734; 110° Mann Whitney U=884.00, sig.=.316; 130 Mann Whitney U=825.00, sig.=.139). Similarly, responses from schizophrenia patients and controls were not significantly different when averaged across each of the four distances in the stimulus array (133 mm Mann Whitney U=916.00, sig.=.457; 157 mm Mann Whitney U=872.00, sig.=.271; 181 mm Mann Whitney U=833.00, sig.=.157; 206 mm Mann Whitney U=949.00, sig.=.633).

3.2.2. 4-second delay (maintenance) 

A series of nonparametric paired-samples tests was used to compare mean absolute direction accuracy, collapsed across all locations, at the 0- and 4-second delay and absolute distance accuracy at the 0- and 4-second delay. In both patient and control groups performance was less accurate at the 4-second delay compared with the 0-second delay (Wilcoxon signed rank tests: all p values <.001). However, at the 4-second delay schizophrenia patients were less accurate (see Table 2) in both mean absolute direction and mean absolute distance responses, collapsed across all target locations, compared to healthy controls indicating that patients with schizophrenia have a relatively greater difficulty maintaining precise direction and distance information in working memory. In order to examine whether group differences were limited to specific distances or directions within the stimulus array, analyses were repeated based on average absolute performance accuracy at each of the five angles and each of the four distances used. All group comparisons were significant (all p-values <.05).

Since fine-grained representations were less exact for all participants following a brief delay category level representations were likely to have been used to estimate target location. This was examined by inspection of response biases, as revealed in signed data. Averaged accuracy at angles <90° in both groups were positively signed, indicating a bias away from the vertical axis towards the oblique at 45° whilst angles ≥90° were negatively signed, indicating bias away from the vertical axis to the oblique at 135°. Significant group differences emerged in signed direction responses at stimulus angles ≥90° only i.e. in left hemi-space (50° Mann Whitney U=947.00, sig.=.622; 70° Mann Whitney U=894.00, sig.=.357; 90° Mann Whitney U=765.00, sig.=.049; 110° Mann Whitney 727.00, sig.=.023; 130° Mann Whitney U=694.00, sig.=.011), indicating relatively greater inaccuracy in schizophrenia patients for spatial representations in this region. Examination of signed distance accuracy revealed no significant group differences.

4. Discussion 

Recent evidence has focussed attention on impaired encoding as a major contributor to working memory deficits in schizophrenia (Lee and Park, 2005). The VSWM task was employed in the current study to examine potential factors underlying encoding imprecision; including multiple representations of spatial location and individual differences in speed of processing. The latter is important in the context of substantial evidence from previous research which shows that patients with schizophrenia exhibit slower information processing compared to healthy controls (Badcock et al., 2004a, Badcock et al., 2004b, Hartman et al., 2002). As expected, the current results confirmed the view that speed of encoding was significantly slower in patients with schizophrenia: the mean stimulus duration required to correctly detect simple visual targets was almost twice as long for patients compared to controls. This result does not appear to be due simply to age-related slowing, since the healthy control group were significantly older than the patients; though ageing processes in schizophrenia may be different from the normal population. Badcock et al., 2004a, Badcock et al., 2004b have also shown that medication effects could not account for slower processing in schizophrenia patients, whilst Gold et al. (2006) recently demonstrated that patients with schizophrenia are able to use attention to encode relevant information in WM, hence alternative explanations may be needed to explain the current data. Indeed, previous DR studies have often selected fixed, short (e.g. <500 ms) target durations (e.g. Fucetola et al., 1999, Park et al., 1999, Pukrop et al., 2003, Snitz et al., 1999) which are likely to have compromised the SWM performance of patients. In the current experimental conditions (Fig. 2) less than half of the patient group had achieved the 79% detection threshold (i.e. could reliably detect the target) at this duration compared to almost three-quarters of healthy controls. Consequently previous reports of impairments in encoding in SWM (Lee and Park, 2005) in schizophrenia may largely reflect these differences in speed of processing.

Importantly, with differences in processing speed equated, the accuracy of spatial representations encoded in working memory was not significantly different between patients with schizophrenia and healthy controls. Specifically, there were no significant group differences in either absolute or signed direction or distance accuracy when responses were made immediately after stimulus offset. Therefore, poor SWM in schizophrenia can not be accounted for by either a selective loss of precision (for example, in the ability to encode distance information only) or a general loss of precision in the fine-grained encoding of spatial location. The accuracy of responses in the 0-second delay condition from both groups across all target locations in the current study is clearly shown in Fig. 3 (upper panel). It can be seen that targets were encoded in sufficient detail to ensure that the majority of immediate responses fell within the area of the target. This conclusion contrasts with previous suggestions of pan-modal processing imprecision in schizophrenia (Javitt et al., 1999), however, the visual task used by this group (A-X type continuous performance task) is more likely to have assessed transient memory for visual identity rather than visuo-spatial working memory.

The current data also replicate previous studies reporting impaired retention of spatial information, even after controlling for differences in speed of processing (O'Donnell et al., 1996, Tek et al., 2002). Both groups showed a loss of absolute accuracy over the 4-second delay, indicating that detailed direction and distance representations tend to fade with time. However, compared to healthy controls, patients with schizophrenia were significantly less accurate on both direction and distance dimensions, consistent with the notion that schizophrenia is associated with a greater difficulty maintaining fine-grained spatial information in working memory (Tek et al., 2002).

It is unclear why there is a greater loss of fine-grained spatial information in schizophrenia. Visual memory decay in healthy observers does not occur randomly; rather there may be a systematic removal of fine details from the representation determined by a time-dependent blurring function (Gold et al., 2005, Harvey, 1986). It is conceivable, therefore, that patients differ in this type of decay function. It is also possible that reduced processing speed in schizophrenia contributes to the problem, since patients may be unable to maintain information as quickly as necessary (e.g. through rehearsal) and as a consequence information is more rapidly lost. Alternatively, maintenance of the target location must take into account eye-movements made during the delay period, since this information (i.e. corollary discharge) is normally used in computing target position. Abnormalities of eye-movement and corollary discharge previously described in patients with schizophrenia (Ford and Mathalon, 2004) may, therefore, also underlie the maintenance problem. Clearly, such possibilities should be addressed in future studies. Furthermore, it appears that the WM systems responsible for visual object identity and spatial location are affected differentially in schizophrenia. For example, in contrast to the current findings, Gold et al. (2003) reported that patients with schizophrenia are able to process two features of visual objects (orientation and colour) in working memory and maintain these representations over a short delay (see also Tek et al., 2002).

As spatial representations become inexact with increasing memory delay the influence of category level spatial representations is assumed to increase, as revealed in the distribution of response biases (Huttenlocher et al., 1991). Noticeably, both groups exhibited a typical pattern of response biases away from the vertical axis, crowding the obliques, suggesting patients with schizophrenia use similar spatial boundaries or reference frames to healthy controls. Nonetheless, significant group differences emerged in signed accuracy at angles ≥90°, i.e. left hemi-space. It is important to note that no central fixation point was employed in the current task; however, this finding suggests that memory traces representing this region of space may decay more rapidly than elsewhere in patients with schizophrenia. There was also no evidence of greater encoding inaccuracy (0-second delay condition) in left hemi-space in schizophrenia, arguing against a general left-sided neglect and no group differences in laterality quotient (i.e. degree of handedness). The current findings suggest that spatial representations are maintained over a short delay – though less accurately – and perhaps especially in left hemi-space; though clearly this result warrants replication. Interestingly, Park (1999) showed that targets presented to the right visual field are more commonly associated with the total loss of spatial representation in working memory in schizophrenia, as indexed by never-corrected errors. Whilst Park's study utilized very brief target durations (200 ms) which may have contributed to the working memory deficit observed in patients, this would not be expected to produce an asymmetric pattern of errors. Clearly further investigation of hemispheric asymmetry of spatial working memory in schizophrenia is required.

In summary, considerable progress has been made in clarifying those component processes of SWM which are not impaired in schizophrenia (Gold et al., 2006). The current results add to this picture, showing that encoding of fine-grained distance and direction information is also intact. The precision of immediate responses obtained in the VSWM test argues against the notion that these patients performed poorly as a result of a generalized deficit in performance, poor motivation or as a result of negative effects of medication. These findings are consistent with the conclusions of Lee and Park's (2005) meta-analysis, which stated that “encoding/or early part of maintenance may be problematic” (p. 599) and clarify that processing speed rather than fine-grained precision contribute to the encoding problem in schizophrenia. It will be important to control for individual differences in speed of processing in future DR studies aimed at identifying individuals at risk of developing the illness.

Role of funding source 

This research was supported by the National Health and Medical Research Council, Australia and infrastructure funding from the Department of Health of Western Australia (CPP Grant 9937102 NH&MRC, HP Grant 960579 NH&MRC). The NHMRC and Department of Health had no further role in study design; in the collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication.

Contributors 

J.C. Badcock designed the study, analysed the data and wrote the first draft of the manuscript, D. R. Badcock wrote the software and developed the protocol for the VSWM test and contributed to data analysis. C. Read managed the literature search and collected the data. A. Jablensky conducted diagnostic reviews for all participants. All authors contributed to and approved the final manuscript.

Conflict of interest 

All authors declare that they have no conflicts of interest.