Categorical and coordinate spatial relations in working memory: An fMRI study

Abstract

Spatial relations within and between objects can be represented either coordinately or categorically. Coordinate representations concern metric and precise relations, and are strongly associated with right parietal cortex activity, while categorical representations relate to more qualitative, abstract relations, and have shown to have a, somewhat weaker, relationship with left parietal cortex activation [Trojano et al., 2002. Coordinate and categorical judgements in spatial imagery. An fMRI study. Neuropsychologia, 40, 1666-1674]. In the current study, a functional magnetic resonance imaging (fMRI) experiment enabled a closer examination of this proposed hemispheric lateralization within a working memory paradigm. A visual half field task in a match-to-sample format was conducted to examine these lateralization effects with a short (500 ms) and a long (2000 ms) interval between two stimuli, with either a categorical or a coordinate instruction. In the behavioural data, the hypothesized hemispheric specialization was found for the brief interval. The imaging data support the hemispheric lateralization as well. The proposed lateralization effect is present during spatial relation processing, but only within the superior parietal cortex and with certain temporal constraints. Additionally, categorical trials show a clear involvement of the left and right premotor and posterior parietal areas during the brief interval, while coordinate trials are related to higher activity in the left and right insula, during the long interval. We propose a refined view on lateralization of spatial relation processing, keeping in mind the temporal restrictions shown by this study.

Introduction

In order to interact with our spatial environment it is critical to be aware of the locations of objects, not only in an absolute sense but also relative to other objects and ourselves. This enables us to navigate accurately through space and to communicate about our environment. Spatial relations between objects can be classified into two types; categorical and coordinate spatial relations, as first proposed by Kosslyn (1987). Categorical relations can be expressed by prepositional, abstract labels such as above/below and right of/left of. Coordinate relations are more precise and metric and can be expressed in absolute measures of distance. Along with this classification, Kosslyn (1987) and Kosslyn et al. (1989) proposed a distinct difference in hemispheric processing. Categorical relations are thought to be processed predominantly by the left hemisphere, whereas the right hemisphere is assumed to be mostly concerned with coordinate relation processing. These hemispheric advantages are suggested to be due to the linguistic characteristics and involvement in navigation of the left and right hemisphere respectively, which are functionally linked to the preferred types of processing. Later simulation studies on this lateralization suggest a dominant role of receptive field sizes. Relatively large and overlapping field sizes are related to coordinate processing, while relatively small receptive field sizes are more suitable for categorical spatial processing (Chabris and Kosslyn, 1998, Jacobs and Kosslyn, 1994, Kosslyn et al., 1992).

Evidence for the foregoing functional lateralization has been found in many different experiments (see Jager and Postma, 2003), including a large number of visual half field experiments on healthy subjects (Christman, 2002, Hellige and Michimata, 1989, Laeng and Peters, 1995, Rybash and Hoyer, 1992) and patient studies (Laeng, 1994, Palermo et al., 2008). In the majority of these studies the proposed hemispheric dissociation was demonstrated. In particular, a right hemispheric advantage for coordinate tasks has been reported most often (for a review see Jager and Postma, 2003). These results have been found for studies focusing on perception (e.g. Hellige and Michimata, 1989, Rybash and Hoyer, 1992) and working memory (Laeng and Peters, 1995, Van der Ham et al., 2007) as well as on mental imagery (e.g. Palermo et al., 2008). Additionally, computer model simulations have indicated that the computation of coordinate and categorical relations is performed better by separate networks (Baker et al., 1999, Kosslyn et al., 1992). Apart from these confirming results, the left hemispheric advantage for categorical tasks is found less often and the idea of a complete double dissociation has been rejected by some (Niebauer, 2001, Sergent, 1991).

Remarkably, the number of direct neurophysiological and imaging studies addressing the distinction is limited. The only EEG study focusing on spatial relation processing offered a more critical view on this double dissociation (Van der Lubbe et al., 2006), proposing that differences between categorical and coordinate processing are mainly of a quantitative nature, caused by differential difficulty, than a qualitative one. Nevertheless the hypothesized lateralization effect has been found to some extent in a PET experiment for rather extensive networks of activation, including parietal and frontal regions (Kosslyn et al., 1998). An rTMS study has also provided proof for this lateralization pattern. Trojano et al. (2006) reported the hypothesized dissociation in the left and right parietal cortex involvement during categorical and coordinate processing. Stimulation of the left parietal cortex affected categorical processing and reduced the categorical learning effect during coordinate processing, whereas right parietal stimulation affected only the coordinate task.

So far, two fMRI studies (Baciu et al., 1999, Trojano et al., 2002) have shown that an increase in activity in the left and right parietal cortex was related to categorical and coordinate relation processing, respectively. While Baciu et al. (1999) report lateralization for the angular gyrus, Trojano et al. (2002) propose that the superior parietal lobule holds a central role in lateralized spatial relation processing. Additionally, Baciu et al. (1999) noted that the initial right hemispheric advantage found for coordinate processing decreased and eventually changed into a left hemispheric advantage in the angular gyrus, probably due to a practice effect, which caused categorization in the coordinate trials. Not only the parietal cortex seems to be of importance, there are some reports focusing on frontal areas, showing similar lateralization patterns (e.g. Slotnick and Moo, 2006) and others with a focus on larger parieto-frontal networks (Kosslyn et al., 1998).

In a more recent fMRI study Martin et al. (2008) made use of a working memory paradigm, in contrast to the older studies reported here, which have focused on perception in designs entailing single stimulus presentation (Baciu et al., 1999, Kosslyn et al., 1998) or on mental imagery (Trojano et al., 2002, Trojano et al., 2006). This working memory paradigm enabled Martin et al. (2008) to examine memory load by changing the number of spatial relations to be remembered, which served as a measure of task difficulty. Accordingly, their imaging study lent support to a different view on spatial relation processing: the continuous spatial coding hypothesis. This proposal entails that both hemispheres are involved in categorical and coordinate relation processing, while factors as attentional and executive processes affect the balance between left and right hemisphere involvement. In particular, higher task complexity would lead to the recruitment of more right hemispheric resources. Coordinate spatial relation processing tasks, being typically more difficult, thus would show a right hemispheric lateralization. However some objections can be made to the idea of continuous spatial coding. The issue of task difficulty has been addressed previously and variation in task difficulty did not affect the direction of lateralization in these studies, only the extent of the difference (Kosslyn et al., 1992, Slotnick et al., 2001). The lateralization thus might only be apparent when the task at hand is sufficiently demanding. Furthermore, the computational analyses mentioned earlier further support the separate coding hypothesis by showing that the use of two separate processing mechanisms is more efficient than the use of a single one (Baker et al., 1999, Kosslyn et al., 1992).

Evidently, the diverse outcome of the limited number of imaging studies indicates the need for more clarity on the issue. Some criticize the lateralization hypothesis, and even if the expected pattern has been found, the location where it would be most apparent in the brain is also a matter of debate. From the literature discussed above, two views emerge; either the two types of spatial processing can be dissociated and show distinguishable patterns of activation, or the two types of processing are actually different expressions of the same underlying mechanisms, influenced by task properties such as difficulty and involvement of spatial attention. In the first case lateralization differences are caused directly by the different characteristics of categorical and coordinate processing, while in the latter case, lateralization differences are assumed to arise from quantitative differences during task execution.

Within the current study we focused on a working memory paradigm, reported previously (Van der Ham et al., 2007), because that type of match-to-sample design combined with neuroimaging allows us to further separate the two views and may provide new insights in the processing of spatial relations. We have selected the cross-and-dot stimuli for this task, as we did in our previous studies, where the relation of a dot to a cross has to be assessed, memorized, and compared to a second stimulus. The match-to-sample design overcomes some of the shortcomings of a simple, perceptual paradigm (see Van der Lubbe et al., 2006). More importantly, with this design we could manipulate the working memory demands by varying the temporal characteristics of the task, which permits the comparison of brain activation patterns between categorical and coordinate processing at different working memory conditions. Given the differential results at a behavioural level (Postma et al., 2006, Van der Ham et al., 2007) this variable was expected to have a substantial influence on brain activity during task execution and to provide for well comparable trial characteristics.

The comparisons of spatial relation type as well as temporal characteristics can be placed in the context of the two views on spatial relation processing. If the original hypothesis on lateralization (Kosslyn, 1987) is fully correct we would expect a clear categorical-left parietal cortex and a coordinate-right parietal cortex advantage, regardless of retention interval. However, if the continuous coding hypothesis is more accurate as proposed by Martin et al. (2008), we would expect to discover lateralization patterns affected mainly by working memory requirements and task difficulty. Arguably, increasing the retention interval length may lead to increased working memory load (Haxby et al., 1995). Additionally, an increasing involvement of working memory could instigate a stronger right hemisphere advantage (e.g. Jonides et al., 1993), regardless of instruction type (categorical or coordinate).