OUP user menu

Spatial and temporal deficits are regionally dissociable in patients with pulvinar lesions

Isabel Arend, Robert Rafal, Robert Ward
DOI: http://dx.doi.org/10.1093/brain/awn135 2140-2152 First published online: 25 July 2008

Summary

The pulvinar is an important structure for visual attention function. Spatial and temporal attention was examined in three patients with varying pulvinar lesions. Spatial and temporal deficits were dissociable. The patient with anterior damage showed strong spatial but not temporal attention deficits, while the patient with posterior damage showed clear temporal attention deficits, but much reduced spatial problems. A third patient with intermediate damage showed intermediate behaviours. These findings are discussed within the scope of models of visual attention in which the pulvinar facilitates communication between different brain areas: depending upon the specifics of pulvinar damage, communication with different cortical areas may be degraded, thereby producing distinct patterns of deficit.

  • human pulvinar
  • visual attention
  • dwell-time
  • feature binding
  • neuropsychology

The pulvinar nucleus of the thalamus is an important structure for visual attention function. The pulvinar has reciprocal connections throughout the brain, including frontal, parietal, temporal, occipital and cingulate cortex, and other subcortical areas including the superior colliculus and amygdala (Romanski et al., 1997). Many specific functions for this hub of connections have been proposed, including attentional engagement (Rafal and Posner, 1987), visual filtering (LaBerge and Buchsbaum, 1990), feature binding (Ward et al., 2002), the analysis of target surround (Michael and Desmedt, 2004) and orienting to feature changes (Michael and Buron, 2005). More general theories of pulvinar function suggest that the widespread reciprocal connections of the pulvinar may facilitate cortico-cortical communication, providing a nexus where the activity of one cortical area can modulate another (Guillery and Sherman, 2002; Shipp, 2004).

Recent reviews of the pulvinar's functional neuroanatomy highlight the well-structured connections between the cortex and the pulvinar (Greive et al., 2000; Shipp, 2003). Shipp (2003) in particular has suggested replacing traditional descriptions of pulvinar subdivisions with a more functional scheme that reflects the pulvinar's cortical connections (Fig. 1). As outlined by Shipp (2003), the pulvinar contains multiple gradients of connections which reflect cortical topography. Figure 1 shows four vectors within the cortex, one running from occipital to infero-temporal cortex, another from parietal to superio-temporal cortex, a third running through cingulate cortex and a fourth from lateral prefrontal (Brodman 8 and 46) to more anterior prefrontal areas including areas 9, 10, 11 and 12. These four vectors define part of a cortical topography which is mirrored in four corresponding gradients of connections within the pulvinar (Shipp, 2003). The gradients of connections illustrated in Fig. 1 share a general characteristic: anterior (and lateral) parts of the pulvinar are more likely to be connected with posterior cortex; posterior (and medial) pulvinar are more likely to be connected with anterior cortex.

Fig. 1

The topographic correspondence between the cortical sheet and the pulvinar presented by Shipp (2003; adapted with permission). (A) The arrows on the cortical flat map show four gradients of topography, as described in the text. These are reflected in gradients of connectivity within the pulvinar. (B) An adaptation of the occipital-IT gradient in the monkey pulvinar (Shipp, 2003) on the human pulvinar. The axial section from the ventral pulvinar is taken from thalamic atlas of Morel et al. (1997; anterior is towards the top of the figure, medial to the left). (C) An axial slice showing a more dorsal section of the pulvinar. The parietal-ST gradient is illustrated, and the cingulate and frontal gradients shown in run in parallel to this one (Shipp, 2003). The labels here in (B) and (C), and also in Figs 2–4, refer to the traditional anatomically defined subdivisions of the pulvinar: PuM = medial pulvinar; PuL = lateral pulvinar; PuA = anterior pulvinar.

This picture reflects the neuroanatomy of the non-human primate. Although studies looking of the functional anatomy of the human pulvinar are still rare, the evidence suggests that the human pulvinar may be organized along similar lines. The effects of damage to different parts of the pulvinar seem to produce deficits related to the processing of connected regions. For example, damage to visuospatial maps in the anterior of the pulvinar produces deficits in spatial localization and feature binding (Ward et al., 2002; Ward and Arend, 2007). However, damage to the medial pulvinar, which carries connections to areas including amygdala and orbitofrontal cortex (Romanski et al., 1997), can produce an inability to recognize briefly presented fearful expressions (Ward et al., 2007) and delays in processing visual threat (Ward et al., 2005). These findings suggest that different parts of the pulvinar assist different visual functions.

Here we investigate whether different parts of the pulvinar may be differentially involved in spatial and temporal aspects of attention. Posterior cortical areas including striate, extrastriate and parietal cortex are vital in the control and use of spatial attention (Corbetta et al., 1998; Kastner et al., 1998). If the human pulvinar is organized as in non-human primates, so that the more anterior parts tend to be connected to more posterior cortical areas, then damage to the anterior would be expected to degrade communications with posterior cortex, and thereby degrade spatial attention processes.

Likewise, the posterior pulvinar is connected with many areas further down the cortical visual streams. Damage to posterior pulvinar would be expected to disrupt communication from areas like TEO and IT, which play a role both in object recognition and in the top-down control of visual attention towards task-relevant objects (Chelazzi et al., 1998; DeWeerd et al., 1999). Lesion in monkey IT can also interfere with identification of objects based on visual conjunctions (Baker et al., 2002). Damage to posterior pulvinar therefore might also be expected to disrupt visual attention and binding. However, disrupted communcation with areas like TEO would not be expected to produce the same spatial specificity as disruption to extrastriate, for example (Kastner et al., 2001). Therefore, we might expect that posterior, compared to anterior pulvinar damage, would likewise produce deficits with less impact on spatial attention processes.

We tested three patients with unilateral lesions affecting the pulvinar. Patient TN suffered a lesion to the anterior and lateral pulvinar, and has previously been seen to show specific spatial deficits (Ward et al., 2002), but has not previously been tested on other types of attention tasks; patient DG's lesion includes some of the the anterior damage found with TN, but his lesion also extends futher posterior; patient CR's is centred on the posterior pulvinar, and appears to have spared much of the anterior.

Spatial and temporal aspects of attention

We assessed spatial and temporal processes using two broad classes of paradigm: dwell time and binding. First we consider the dwell time tasks. Visual identification of an initial target (T1) interferes with identification of a subsequent one (T2) for ∼200–500 ms (Duncan et al., 1994). This so-called ‘dwell time’ reflects the time to identify T1 and then reallocate limited visual-processing resources from T1 to T2. Dwell time can reflect both spatial and non-spatial factors. If T1 and T2 appear in the same location, there is still a marked dwell time, or ‘attentional blink’, following attention to T1. If T1 and T2 appear in different locations, there is not only a cost to reallocating attention between objects, but also a cost and time for shifting spatial attention (Ward et al., 1997). We therefore used two types of dwell time experiment, one that requires spatial shifts between T1 and T2, and one that does not.

Our second set of tasks investigate visual feature binding. Visual processing is distributed in parallel across many specialized brain areas. Coordinating the activity of multiple visual processors to represent the features of a single object is described as the ‘binding problem’. Breakdowns in the process of feature binding can be observed in the form of illusory conjunctions, errors in which visual features are correctly perceived but incorrectly combined. For example, observers might report the presence of a blue X when shown a display containing a blue O and a pink X (Treisman and Schmdit, 1982).

Here we focus on two types of binding tasks, as outlined by Robertson (2003). In spatial binding tasks (Treisman and Schmidt, 1982), object features at one location must be kept separate from features appearing concurrently at other locations. In temporal binding tasks (Botella et al., 2001), features that appear at the same time belong together, and must be kept separate from other features, appearing in the same location, but at different times. That is, in spatial binding, the visual system can use space but not time to determine which features go together; in temporal binding the system can use time but not space.

Methods

Patients

We tested three patients with lesions involving different parts of the pulvinar. Visual fields were intact in all patients and none had signs of visual extinction or hemispatial neglect on confrontation. All are active, ambulatory and independent, with no mental symptoms.

TN is a 60-year-old, right handed, hypertensive woman who suffered a haematoma centred in the right thalamus 8 years before the present testing. She has weakness in the left arm and leg, but can walk with a cane. There remains pseudoathetosis of the left hand and, with her eyes closed, the arm will drift and she will not know its location. She suffers from periodic ‘pins and needles’ pain in the face and left hand. We have previously found that her selective visual attention is good in that she can effectively attend to targets and filter distractors in either field, although there is evidence of reduced response channel activation (Danziger et al., 2004). There is also clear evidence of spatial coding and feature binding deficits in her lower-left quadrant (Ward et al., 2002). TN's spatial coding deficit is also illustrated by her inaccuracy in locating objects presented in her lower-left quadrant (Ward and Arend, 2007). Visual identification of complex emotional stimuli is completely intact in both fields (Ward et al., 2007).

DG is a 70-year-old, right-handed man who suffered an hypertensive haemorrhage in the left thalamus 3 years before the present testing. He has weakness in his right arm and leg, but he can walk with a cane. We found that DG has deficits in correctly locating targets presented across the contralesional field (Ward and Arend, 2007). Like TN, visual recognition of complex emotional face stimuli is normal in both fields, again even with brief exposures (Ward et al., 2007).

CR is a 19-year-old man who suffered a closed head injury in a fall 2 years before testing resulting in a haemorrhage contusion and avulsion of the posterior pole of the left pulvinar and no other contusions to the brain. High-resolution multisprectral MRI revealed no other brain pathology. He has no mental symptoms or motor impairment.

Lesion reconstruction

For each patient we mapped the lesion reconstruction of MRI data to the Atlas of the Human Thalamus (Morel et al., 1997). We used saggital atlas sections, because this view happened to contain the clearest information for the ventral and lateral parts of the pulvinar damaged in TN, and was as useful as the axial and coronal views for DG and CR.

For each patient, we scaled the damaged thalamus to the dimensions used in the saggital atlas sections. Scaling in the anterior–posterior dimension was based on the distance between anterior and posterior commissures within the AC–PC plane. Scaling along the medial-lateral dimension was based on the distance from the wall of the ventricle to the edge of the white matter for the vector defined by the intersection of the AC–PC plane and the perpendicular coronal plane running through the PC. For the ventral-dorsal dimension, we used the distance of the PC to the floor of the ventricle, for the vector defined by the intersection of the AC–PC plane and the perpendicular plane running along the saggital midline. Results are illustrated in Figs 2–4.

Fig. 2

TN reconstructed MRI images and lesion mapping onto atlas (Morel et al., 1997). Upper panel shows axial MRI slices running from 2 mm below the AC–PC plane to 4 mm dorsal of AC–PC. In the lower panel we present four saggital slices from the atlas corresponding to 13, 14, 16 and 19 mm lateral from the midline (posterior to the left). Neurological inspection of the MRI scan also revealed probable lesion (indicated as light gray area on the atlas) in the 14 mm slice that reaches the ventral pulvinar. The two most lateral slices shown (16 and 19 mm) illustrate how the lesion mainly affects the anterior pulvinar. As we move laterally to 19 mm, the lesion extends ventrally below the AC–PC line, coming very close to the lateral retinotopic area (PuL). PuM, PuA and PuL are the anatomically defined medial, anterior and lateral pulvinar regions. Other abbreviations mark different parts of the thalamus and nearby areas, see Morel et al. (1997) for details. The cross marks represent the locations of the anterior and posterior commissures at the midline. The rulers indicate 1 mm spacing.

Fig. 3

Lesion data for DG. The four saggital slices corresponds to 9, 11, 13 and 18 mm. DG's lesion affects almost the entire medial pulvinar nucleus above the AC–PC line (dark gray area on the atlas). As for TN, the light gray area shows probable lesion. The 18 mm saggital slice shows that his lesion still affects the medial pulvinar above the AC–PC line. See Fig. 2 for abbreviations used.

Fig. 4

Axial slices show that CR's lesion produced loss of the posterior pulvinar. The figure indicates a plane that runs almost perpendicular to the AC–PC plane, starting about 5 mm posterior of the PC, and that runs along the breadth of the pulvinar. Only a single slice is shown since this plane would look the same in all slices. Unlike TN and DG, CR's lesion mainly affects the posterior part of the medial pulvinar above and below the AC–PC line. See Fig. 2 for abbreviations used.

Apparatus and stimuli

All studies, expect the Dwell time with spatial shift, were presented by Tobii Clear View software. The spatial dwell time task was run using PsyScope (Cohen et al., 1993). The experiments were ran using a Dell portable PC connected to a Tobii eye-tracker.

Procedures

Dwell time task with spatial shift

The experimental presentation was similar to the previous attentional dwell time studies (Duncan et al., 1994). On each trial, a letter (L or T) and a digit (2 or 5) character were presented in an unpredictable order with a variable temporal delay between them (0–1000 ms). The first character (T1) could appear in any quadrant; the second (T2) was presented in an orthogonally opposed quadrant to T1. Characters were briefly displayed (58–80 ms) and immediately followed by a 250 ms pattern mask (identical to Duncan et al., 1994) to limit visual persistence. The letters were presented 0.5° from fixation. Stimulus exposure was set on each session to avoid floor or ceiling effects on performance. The short exposure combined with the unpredictable location ensured there was no value in attempting to fixate targets. Patients were instructed to keep fixation and to verbally report both targets in their order of appearance. For the simultaneous (0 SOA) condition, the first item reported was considered the T1 item, and the second T2. Responses were typed in a keypad by the experimenter. No feedback was provided after the initial practice.

Dwell time interpretation depends on attention to T1 affecting report of T2. As in numerous other dwell time studies (Raymond et al., 1992; Chun and Potter, 1995), we therefore restricted our analysis of T2 performance to trials in which T1 was correctly reported.

Dwell time task without spatial shift

This task was similar to spatial shift version above, except that both targets appeared successively in the same location, in one of the four visual quadrants. Each target was presented for 60–80 ms, with a 100 ms pattern mask immediately afterwards. Again participants were required to report both targets. To allow full time for T1 presentation, we did not use the 0 and 100 ms SOAs used in the spatial shift version. As in the spatial shift task, patients were instructed to keep fixation, and this was monitored by remote eye-tracker.

Spatial binding task

Patients were asked to report the identity of the white letter presented among three black distractor letters arranged on a 2 × 2 search array. The letters appeared against a gray background. The four letters in each display were randomly selected without replacement from a pool of five letters (QWERT). The letters within the search array measured 0.4 × 0.4 of visual angle and the search array measured 1°×1°. The search array was presented 1.5° from fixation in one of the four visual quadrants. When the patient was ready, the experimenter pressed the space bar to start the trial, after a variable delay between 200 and 450 ms, the search array appeared. The array was present for 80–100 ms, and followed by a pattern mask composed of checkerboard of black lines. The mask appeared at the array location and stayed on the screen for 150 ms. The patients responded verbally and the responses were typed by the experimenter.

In this task two kinds of errors can arise: illusory conjunctions (ICs) and feature errors (FE). FE refers to the situation in which a feature that was not presented in the display is reported as being the target feature. This is possible because there are five possible stimulus identities, but only four are presented on each trial. FEs therefore reflect a failure to correctly encode visual features (Treisman and Schmidt, 1982). In contrast, ICs refer to the situations in which a feature belonging to a distractor stimulus is reported as belonging to the target stimulus. Here we will be looking at both types of error as a function of search array location.

Temporal binding task

This procedure was based on the previous temporal binding studies (see Lawrence, 1971; Botella et al., 2001), in which the target is preceded and followed by distractors. All items appear sequentially in rapid succession at a single location. Like the spatial binding task, patients were asked to report the identity of the only white letter presented among black distractor letters. The target letter was always preceded and followed by two distractor letters. The letters were selected from the set ‘QWERT’, and were presented in random order. The exposure duration of single items varied in different sessions between 80 and 100 ms. The stimulus display contained four boxes, one in each quadrant, 1.5° from fixation, and subtending 1° × 1°. The boxes remained on the screen throughout the trial. During the trial, the stream of letters appeared inside one of the four boxes, the letters subtending 0.5°× 0.5.

In this task, intrusion errors occur when a distractor feature appearing at one point in time, is reported as belonging to the target, appearing at a different time. Intrusions most commonly come from positions temporally near the target, just before (–1) or just after (+1), but intrusions from the +2 and –2 positions are relatively rare (Botella and Eriksen, 1992). That is, there is a distribution of intrusions such that features temporally near the target are more likely to be integrated with the target. In contrast, blind guessing of target identity would produce a distribution of errors that is flat across temporal distance. We therefore focused on the difference between intrusions from distractors temporally near (±1) and far (±2) from the target. An increase in this difference, particularly when the rate of ±2 errors stays constant, would indicate a deficit in the use of temporal information to bind object features.

Eye-tracker recordings

Eye movements were monitored by using a 50 Hz Tobii 1750 remote eye-tracker system connected to a portable Dell PC. Trials in which fixation deviated from screen centre by 1.0° or more were excluded in the binding tasks, and by 0.5°or more in the dwell time task (as stimulus eccentricity was reduced in this task). All patients sucessfuly kept fixation in over 90% of trials, except for TN in the dwell time task without spatial shift (75% of trials). In this exceptional case, we analysed the excluded eye-movement trials but found no differences in the pattern of results compared to the no-eye-movement trials.

Results

We present results on a patient-by-patient basis.

Patient TN

TN has the most anterior lesion of the patients (Fig. 2). Her lesion is somewhat wedge-shaped, such that the ventral extent of her lesion increases moving laterally (Fig. 2). We have previously seen specific deficits for stimuli appearing in her lower-left (LL) quadrant (Ward et al., 2002; Ward and Arend, 2007). This was also true in our current experiments, and our analyses therefore focused on the comparison between her affected quadrant (LL) compared to her intact quadrants (mean performance in the other three quadrants).

Dwell time task with spatial shift

There were 1112 trials for analysis. TN's accuracy to report the first target (T1) was not different when it was in LL (89%) compared to when it was in any other quadrant (90%), χ2(1) < 1, N.S. That is, there was no evidence of a deficit in visual identification, prior to attention reallocation.

Figure 5, first row, shows T2 performance as a function of SOA and target locations. The figure demonstrates a very typical pattern, in which T2 performance reaches a minimum around 200 ms after T1 presentation, and then recovers with increasing SOA (Raymond et al., 1992). We analysed T2 accuracy on the basis of both T1 and T2 location. Figure 5 (left panel) shows that T2 accuracy depended on the location of T1. Specifically, when T1 appeared in the impaired LL quadrant, there was more interference on subsequent T2 report. Using trial as the random factor (excluding trials in which T1 was not reported correctly), we found a main effect of SOA, F(6, 988) = 11.454, P < 0.0005, and a significant main effect of T1 Position F(1, 988) = 8.186, P = 0.004. The interaction between the two factors did not reach significance, F(6, 988) = 1.612, P = 0.14.

Fig. 5

Dwell time task with spatial shift. Shown is T2 accuracy on trials in which T1 is correctly reported. Each row provides the data for a single patient (identified by initials). The left panel shows T2 performance as a function of T1 position and SOA, the right panel shows T2 performance as a function of T2 position and SOA. Because TN's deficit is specific to one quadrant, her data compare the lower-left (LL) quadrant to all others. For the other two patients we compare contralesional to ipsilesional positions. Error bars represent the SEM.

In contrast, T2 accuracy did not depend on the location of T2 (Fig. 5, right panel). Although we did see a main effect of SOA, F(6, 988) = 11.562, P < 0.0005, we observed neither the main effect of T2 position, nor the interaction between Position × SOA (F < 1 in both cases).

In summary, TN shows costs for reallocating attention to a new target when she must shift her attention away from her impaired LL quadrant. Here it is useful to note that this cost, and the majority of other costs we see in this paradigm, is not in the form of a Position × SOA interaction, but a main effect of Position. That is, reallocation from the impaired field resulted in a sustained loss of subsequent capacity, rather than an extended dwell time.

Dwell time task without spatial shift

There were 1632 trials available for data analysis. T1 performance was 84% in her LL quadrant and 85% in the other quadrants. This difference was not significant, χ2(1) <1. Figure 6 shows TN's dwell time data plotted as a function of Position (LL versus all others) and SOA. In this experiment, T1 and T2 are always in the same location, so there is no separate analysis of T1 and T2 position. As shown in Fig. 6, performance in the LL quadrant is certainly not impaired, and in fact, shows a slight advantage over the mean of all other quadrants, F(1, 1338) = 10.50, P = 0.01. There was a significant main effect of SOA, F(4, 1338) = 10.512, P < 0.0005, but no Position × SOA interaction, F < 1.

Fig. 6

Dwell time task without spatial shift. Shown is T2 accuracy on trials in which T1 was correctly reported, and there was no eye-movement. In this experiment, both targets appeared in the same position.

This result supports the idea that TN's deficit in this task is in shifting away from the T1 location. When the shift away was not required, because targets were in the same position, she showed no deficit.

Spatial binding

The frequency of binding errors (illusory conjunctions, or ICs) was significantly greater in the LL quadrant than in the others (42 versus 5%), χ2(1) = 107.47, P < 0.0005 (Fig. 7). This was true even though feature perception as indicated by FE was constant across quadrants, 4% for LL, 6% for all others, χ2(1) = 1.36, P = 0.242. These results generalize previous findings reported by Ward et al. (2002), but using a different search task. TN makes many more binding errors in her affected LL quadrant, even though feature perception seems intact. Table 1 gives detailed information on the amount of ICs and FE across field for this task.

Fig. 7

Comparing patterns of deficit in the spatial and temporal binding tasks. The left panel shows for each patient the mean proportion of illusory conjunction errors (ICs) in the spatial binding task. The right panel shows the mean proportion of –1/+1 intrusions for the temporal binding task. For TN, ‘Impaired’ refers to her lower-left quadrant, and ‘Intact’ to the mean performance in all other quadrants. For DG and CR, ‘Impaired’ refers to the contralesional and ‘Intact’ to the ipsilesional field.

View this table:
Table 1

Frequencies and trial numbers for the binding tasks

Spatial binding task
TNDGCR
ICFEICFEICFE
Intact0.050.040.110.020.030.03
Impaired0.420.060.230.060.150.09
No. intact393355360
No. impaired136340354
Temporal binding task
TNDGCR
−2−1+1+2−2−1+1+2−2−1+1+2
Intact0.010.020.090.020.010.020.080.020.010.030.190.03
Impaired0.020.020.070.020.000.030.240.030.010.060.260.06
No. intact597323429
No. impaired186311432
  • For the spatial binding task, the first two rows give the proportion of illusory conjunction (IC) and feature errors (FE) are given for Impaired and Intact fields. The next two rows give the number of trials for these fields. For TN, ‘Impaired’ refers to her lower-left quadrant, and ‘Intact’ to the mean performance in all other quadrants. For DG and CR, ‘Impaired’ refers to the contralesional and ‘Intact’ to the ipsilesional field. For the temporal binding task, the frequency of intrusions from each distractor position in the stream is given, e.g. −1 refers to the distractor appearing just before the target, +2 to the distractor two places after the target.

Temporal binding

Figure 7 shows the mean proportion of intrusions from distractors temporally near the target, for TN's impaired LL quadrant compared to all others in this task. There was no difference, χ2(1) < 1. Details on TN's response distribution on this task is given in Table 1. Instrusions from the temporally far distractors in ±2 positions were low and did not vary between fields, χ2(1) < 1 (Table 1). TN shows no difficulty in using temporal information to assist feature binding, regardless of where the features appear.

Patient DG

DG's lesion, like TN's, affects the anterior of the pulvinar. There are two notable differences from TN. First, DG's lesion extends further into the posterior pulvinar. Second, the ventral extent of DG's lesion remains consistent moving in the medial to lateral direction (Fig. 3; unlike TN's ‘wedge shape’, Fig. 2). The effects of DG's lesion are observed throughout his contralesional visual field, both in previous work (Ward and Arend, 2007) and in the present results. We therefore focus on the traditional contra- versus ipsi-lesional comparisons.

Dwell task with spatial shifts

DG completed 1008 trials. He showed a small but significant advantage for reporting ipsilesional T1s (95%) compared to contralesional (89%), χ2(1) = 10.08, P = 0.001, suggesting some reduced contralesional capacity for visual identification, prior to attention reallocation.

Following the analysis carried out for TN, we looked at DG's reallocation costs as a function of T1 and T2 position (Ipsi or Contra). Using trial as the random variable, and excluding trials in which T1 was incorrectly reported, we conducted a three-way ANOVA examining T2 accuracy: T1 Field (Ipsi and Contra) × T2 Field (Ipsi and Contra) × SOA. Results are shown in Fig. 5, second row. We found a significant main effect of SOA, F(6, 920) = 4.726, P < 0.0005, and a significant main effect of T1 Field, F(1, 920) = 9.307, P = 0.002. No other effects were significant, in particular, there was no main effect of T2 Field F(1, 920) < 1, nor any interaction of T2 Field and SOA, F(6, 920) < 1. DG then showed a similar pattern of results as TN: an increased cost of reallocation when attention must be shifted away from his impaired field. Like TN, DG also shows a sustained cost on T2, which does not recover to the levels seen after shifts away from the intact field.

Dwell time task without spatial shift

There were 1373 trials submitted for analysis. As in the previous experiment, DG's overall T1 accuracy was slightly but significantly lower for Contra (83%) than Ipsi fields (89%), χ2(1) = 12.30, P < 0.0005. DG's dwell time results are shown in the middle panel of Fig. 6. A two-way ANOVA on T2 accuracy for Field (Ipsi versus Contra) × SOA revealed a main effect of Field F(1, 1135) = 16.048, P < 0.0005, and a significant main effect of SOA, F(6, 1135) = 19.157, P < 0.0005, but no interaction, F < 1. Therefore, for DG, unlike TN, we do see a small but consistent cost for contralesional T2s at all SOAs, even in the absence of a spatial shift, suggesting a reduction in capacity to process contralesional items that is independent of time.

Spatial binding

Figure 7 shows DG's performance in his ipsi and contra field. The number of ICs were significantly higher in his Contra (23%) than in his Ipsi field (10%), χ2(1) = 19.478, P < 0.0005. Table 1 shows that FEs were also significantly higher in his Contra (6%) than in his Ipis (2%) Field, χ2(1) = 14.293, P < 0.0005, although the magnitude of this difference is only 4%.

How should we think about this small difference in FE? It is not an easy question. By any model, the rate of true binding errors must be a positive function of IC rate and a negative function of FE rate: a person making many IC errors and few FE is perceiving features correctly, but miscombining them; someone else making many IC errors and many FE is not perceiving features well, and so more of the IC errors are likely to be the result of errors in perception. The field difference for IC (13%) is greater than for FE (4%), and the absolute levels of FE are low. This in itself might suggest that damage is having a proportionately greater impact on IC than FE. But there are numerous ways in which to take into account the effect of feature processing on FE and IC rates (Triesman and Schmidt, 1982). The most comprehensive model we are aware of is the multinomial modelling approach used by Prinzmetal and colleagues (Prinzmetal et al., 2002), which uses maximum likelihood estimate techniques to solve for parameters relating to perceptual processing, guessing and binding efficiency. Work in progress uses this multinomial model to assess binding deficits in the spatial binding task, and suggests that the models for all three patients can fit the data well using a parameter for contralesional binding deficit.

However, to anticipate our argument in the General Discussion, we are ultimately most interested in the dissociation of attention processes following different kinds of pulvinar damage. From this perspective, it is not so critical to assess the absolute level of binding errors for an individual patient, but to compare error rates between patients. Again, given that the true rate of true binding errors will be a positive function of IC rate and a negative function of FE rate, it is clear from Table 1 and Fig. 7 that any spatial binding deficit for DG is smaller than for TN. We ran ANOVAs on FE rate and IC rate, using trial as the random variable, Patient as a between-subjects variable, and Field as a within-subjects variable. TN compared to DG has a significantly larger field difference in IC (37 versus 12%, respectively), F(1, 1220) = 33.12, P < 0.0005. If TN also had a significantly larger field difference in FE, it would be difficult to know for certain whether her greater IC difference reflected binding or perceptual problems. However, in fact, TN compared to DG has a non-significantly lower field difference in FE (4 versus 2%, respectively), F(1, 1220) = 1.74, P > 0.05. By any model, TN will therefore have a greater binding deficit than DG.

Temporal binding

Figure 7 shows mean proportion of intrusion errors from temporally near (±1) distractors for Ipsi (10%) and Contra (27%) fields. DG's instrusion rate was significantly greater in his contralesional field χ2(1), 31.18, P < 0.0005, showing that his lesion impaired his ability to use time to correctly bind features. As seen in Table 1, the average intrusion rate for the temporally far distractors was low (1.5%) and identical for both fields. Unlike TN, DG shows clear evidence of a temporal binding deficit.

Patient CR

CR's pulvinar lesion is notably different from TN and DG, as he is spared the damage to the anterior of the pulvinar that is found in these patients. His lesion is to the most posterior pulvinar, which is spared in TN, but which appears to overlap slightly with DG.

Dwell time task with spatial shift

CR completed 1008 trials. CR was significantly better in reporting Ipsi (93%) than Contra (86%) T1s, χ2(1) = 7.87, P = 0.005. This result indicates a reduction in capacity to identify contralesional objects independent of attention reallocation.

We performed a three-way ANOVA for T1 position (Ipsi and Contra) × T2 position (Ipsi and Contra) × SOA, examining T2 accuracy (see Fig. 5, lower row). There was a significant main effect of SOA, F(6, 884) = 6.5430, P < 0.0005; a significant main effect of T2 position, F(1, 884) = 16.183, P < 0.0005; and a significant interaction between SOA × T2 position, F(6, 884) = 2.482, P = 0.022. No other effect was significant, in particular, unlike TN and DG, there was no effect of T1 position, F(1, 884) = 2.440, P = 0.119. Unlike these patients, who showed costs to move away from T1 in the impaired field, CR showed a cost for reallocating into his impaired field. Furthermore, in CR, unlike the others, we see an effect of increased dwell time, evidenced by the interaction of T2 position × SOA, sugggesting a slowing of attentional reallocation within his contralesional field.

Dwell time task without spatial shift

There were 726 trials available for analysis. Again, CR's performance was significantly worse to report Contra (90%) compared to Ipsi T1s (96%), χ2(1) = 9.68, P = 0.0018. An ANOVA on T2 accuracy as a function of Field and SOA showed a significant main effect of Field, F(1, 662) = 29.153, P < 0.0005 (Fig. 6). However, neither the main effect of SOA reached significance, F(1, 662) = 1.741, P = 0.139, nor the interaction between Field × SOA, F < 1. For CR, eliminating the spatial shift did not reduce contralesional costs. That is, the contralesional deficit on T2 appears to be maintained regardless of whether a spatial shift is required.

Spatial binding

Figure 7 shows CR's mean proportion of ICs across field. Errors were more frequent in his contralesional field, both for ICs (3 versus 15%), χ2(1) = 29.21, P < 0.0005; and FE (3 versus 9%), χ2(1) = 11.29, P < 0.0005. FE for this task is given in Table 1. All the considerations raised with DG regarding difference in FE apply to CR as well. That is, we suggest he probably does have a spatial binding deficit: ICs show a larger field difference than FEs, and multinomial modelling based on Prinzmetal et al. (2002), indicates a potential deficit in binding. However, the more important point, from the perspective of dissociation of attention functions, is that any spatial binding deficit in CR is clearly smaller than in TN: the field difference in IC rate was larger for TN than CR (37 versus 12%, respectively), F(1, 1239) = 71.9, P < 0.0005, but the field difference in FE rate was non-significantly lower for TN than CR (2 versus 6%, respectively), F(1, 1239) < 1. That is, CR compared to TN shows a smaller difference in IC errors, but an equivalent or larger difference in FE errors, making a strong case for a smaller binding deficit. This same form of analysis did not find a significant difference between DG and CR in FE or IC field differences, F(1, 1405) < 1.

Temporal binding

Figure 7 shows the mean proportion of intrusions from temporally near distractors, in both Ipsi (22%) and Contra (32%) fields, χ2(1) = 11.0, P = 0.0009. There was no significant difference between fields for the temporally far distractors, χ2(1) = 3.7, P > 0.05 (Table 1). CR therefore shows evidence of a temporal binding deficit.

General discussion

We examined the involvement of the human pulvinar in spatial and temporal attention function by testing three patients with unilateral pulvinar lesions on a range of attention tasks. We first summarize the results and deficits for each patient. TN and CR show little overlap in their pulvinar lesions, while DG's lesion is somewhat intermediate in position. So here it is useful to first contrast TN and CR, and then consider DG.

TN

TN's lesion to the anterior of the pulvinar produced marked spatial deficits. First, she showed costs for reallocating attention away from her impaired lower-left quadrant, but not into this quadrant. Consistent with this finding, when the spatial shift was eliminated, she showed no difference between quadrants. That is, whether or not she shows a T2 cost depends upon whether a spatial shift is involved. This shift appears to be something like a ‘disengage’ cost, in that the cost is in moving away from her impaired quadrant. To our knowledge, the disengage difficulty seen here has not previously been reported. It is unlike the ‘disengage’ deficit described in parietal patients (Posner et al., 1984), where the deficit is in disengaging from the intact field. However, in the typical Posner cueing paradigm, like that used to test parietal lesion patients, the task is a simple target detection, and participants are instructed to ignore the cue. In contrast, in the dwell time paradigm employed here, participants are required to engage attention and encode both T1 and T2. The ‘disengage defict’ decribed here, therefore, presumably reflects greater time and effort to encode the contralesional T1 and the resultant prolognation of attentional dwell at the location of the contralesional target. Second, when multiple objects appeared simultaneously in her impaired quadrant, she had marked difficulty in maintaining correct feature bindings. However, when multiple objects appeared within her impaired quadrant, but were separated in time, feature binding was intact.

CR

CR's lesion has little or no obvious overlap with TN's, and indeed they show clearly different patterns of deficit. Unlike TN, CR showed some general costs to identification in his impaired field, evidenced by slightly worse T1 identification in the dwell time tasks, and increased FE in the spatial binding task. In the dwell time task, this general deficit in identification seemed to be his primary impairment, as suggested by two results. First, with the spatial shift, CR showed the opposite pattern to TN: increased costs to T2 identification based on T2, not T1 location. That is, instead of a cost to shift away from his impaired field, a cost to process objects within it. Second, eliminating the spatial shift did not eliminate or even reduce costs on contralesional T2s. In other words, reallocating attention from one place to another does not seem to be nearly as important in determining CR's deficit as for TN.

For spatial binding, the difference between IC and FE rates was greater in his contra- than ipsilesional field, but any spatial binding deficit was much less pronounced than for TN. Temporal feature binding, by contrast, was markedly impaired in CR's contralesional field, particularly compared to TN. We therefore see a double dissociation between patients in the degree of spatial and temporal binding deficits.

So the sum of evidence for CR suggests: (i) a general deficit in identifying contralesional items; (ii) beyond this deficit, no obvious additional cost of reallocation from spatial shifts of attention; (iii) any spatial binding deficit is small compared to TN and (iv) a clear contralesional deficit in using temporal information for feature binding.

DG

DG's lesion had further posterior extent than TN, but not as much as CR. It is interesting that his pattern of deficit is also in many ways intermediate to TN and CR.

Like CR, DG showed some general difficulty with visual processing in his contralesional field, evidenced by reduced T1 accuracy in the dwell time tasks, and increased feature errors in the spatial binding task. While the bulk of data argues for a contralesional idenficiation deficit, we note one inconsistency with this conclusion: namely, DG did not show an effect of T2 position. The basis of this discrepancy is unclear at present, although it is unlikely to be an issue of experimental power, given the number of trials measured.

In the binding tasks, DG showed a similar pattern to CR. DG showed a clear temporal binding deficit. We suggest his spatial binding is likely to be impaired as well, but any such deficit is clearly less than for TN.

However, like TN, DG also showed costs for reallocating away from, but not into, his contralesional field. Removing the spatial shift did not completely equalize performance, as there was still a cost in contralesional T2 processing, perhaps reflecting his general contralesional deficit in identification. However, this cost is smaller than for CR (Fig. 5 middle and right panels). So like TN but unlike CR, there does appear to be a spatial component to DG's deficit, such that he shows a similar cost for ‘disengaging’ from his impaired field. DG therefore shows evidence of both spatial and temporal deficits.

Dissociations and relation to neuroanatomy

Taken together, our results suggest that temporal and spatial attention processes can be regionally dissociable within the human pulvinar. Clearly, damage to the anterior of the pulvinar can produce substantial deficits in spatial feature binding, and in reallocation after spatial shifts of attention, without producing problems in temporal binding. Likewise, a lesion limited to the posterior pulvinar can produce substantial deficits in temporal binding, with only little or no effect on spatial feature binding and the costs of reallocation after a spatial shift from the impaired field. Figures 7 and 8 show that for binding tasks especially, spatial and temporal binding deficits can be freely dissociated between patients. TN shows the greatest spatial binding, but the least temporal binding deficit. CR shows the reverse pattern, and DG demonstrates intermediate levels of error on both tasks, although resembling CR more than TN in these binding tasks.

This particular dissociation, suggesting spatial processing deficits following damage to the anterior of the pulvinar, and temporal processing deficits following damage to the posterior, is especially interesting given what we know about connectivity in the non-human primate (NHP) pulvinar. The gradients of cortical connectivity described by Shipp (2003), and diagrammed in Fig. 1, suggest that connections to striate, extrastriate and parietal cortex are made more in the anterior of the pulvinar. Given the importance of these cortical areas in spatial attention, it is highly relevant that damage to the anterior of the pulvinar would disrupt spatial attention processes. In this context it is also interesting to note that spatial binding errors, shown especially strongly by TN, can also be found in patients with parietal lesions (Cohen and Rafal, 1987).

Of course, it would be a gross over-simplification to suggest that all spatial processing is done in posterior cortex, or that posterior cortex is limited to spatial tasks. Areas like the frontal eye fields are clearly important for the allocation of spatial attention, and an enormous range of tasks show evidence of coordination between frontal and parietal regions (Corbetta et al., 1998). It would also be a mistake to attend only to the anterior–posterior organization of the pulvinar and neglect its medial–lateral organization. The gradients outlined in Fig. 1 run along both dimensions. Furthermore, there appear to be interesting differences in connectivity between dorsal and ventral pulvinar (Shipp, 2003). Our emphasis, on the broad structure of pulvinar-cortical connectivity, is meant (i) to provide a first sketch of how the human pulvinar may be organized with respect to different kinds of attention tasks and (ii) to try and relate current proposals about pulvinar functional anatomy to behaviour. We suggest this is a useful start given how little is currently known about pulvinar function, especially in humans.

Finally, we need to consider the fact that TN's lesion specifically affects her inferior contralesional quadrant, as seen here and in other studies (Ward et al., 2002; Ward and Arend, 2007). Traditional accounts of pulvinar organization subdivide the pulvinar into inferior (PuI), lateral (PuL), dorsomedial (PuM) and anterior (PuA) regions. PuI and PuL contain spatial maps, organized so that the inferior field is represented dorsally and the superior field ventrally. PuI and PuL are located in the ventral and anterior part of the pulvinar. Referring to Fig. 2, the most lateral saggital slice of TN's lesion comes very close to the location PuL (just visible in the 16 mm lateral slice). The specificity of her spatial deficits, within a single visual quadrant, might be expected if her lesion actually contacts spatial maps in PuL, but only the dorsal part, representing the inferior part of the contralateral visual field. In fact, TN's lesion to the lateral, ventral and anterior ‘corner’ of the pulvinar is broadly similar to the locus of activation for contralateral pulvinar maps suggested in recent imaging work (Cotton and Smith, 2007).

This discussion is meant to highlight interesting correspondences between our neuropsychological findings and detailed neuroanatomy of the pulvinar. This correspondence suggests new tests of pulvinar patients, based on the specifics of their lesion location. However, there are limits to this perspective. In particular, lesions to the pulvinar do not seem to have anywhere near the effect of lesions to the connected cortical areas. Indeed, the three patients discussed here, and other patients we have observed, even with complete unilateral loss of the pulvinar (Ward et al., 2005, 2007) have not tended to complain about their vision. This is important when considering proposals that the driving input for cortico-cortical communication may be via the thalamus (Guillery and Sherman, 2002). It seems more likely that the thalamus facilitates cortico-cortical communication rather than serving as the primary communication route.

Acknowledgements

We thank TN, DG and CR, who have made this research possible through their participation, time and effort. The funding statement has been provided by Biotechnology and Biological Sciences Research Council (C501417 to R.W. and R.R.).

Footnotes

  • Abbreviations:
    Abbreviations:
    FEs
    feature errors
    ICs
    illusory conjunctions

References

View Abstract