Blindsight in action: what can the different sub-types of blindsight tell us about the control of visually guided actions?

Transcription

1 Neuroscience and Biobehavioral Reviews 29 (2005) Review Blindsight in action: what can the different sub-types of blindsight tell us about the control of visually guided actions? James Danckert a, *, Yves Rossetti b,c,d a Department of Psychology, University of Waterloo, 200 University Avenue West, Waterloo, Ont., Canada N2L 3G1 b Espace et Action, INSERM Unité 534 and Université Claude Bernard, 16 Avenue doyen Lépine, Bron Cedex, France c Service de Rééducation Neurologique, Hôpital Henry Gabrielle, Université Claude Bernard and Hospices Civils de Lyon, Route de Vourles, St Genis Laval Cedex, France d Institut Fédératif des neurosciences de Lyon, 59 Boulevard Pinel, Lyon, France Received 27 July 2004; revised 13 December 2004; accepted 7 February 2005 Abstract Blindsight broadly refers to the paradoxical neurological condition where patients with a visual field defect due to a cortical lesion nevertheless demonstrate implicit residual visual sensitivity within their field cut. The aim of this paper is twofold. First, through a selective review of the blindsight literature we propose a new taxonomy for the subtypes of residual abilities described in blindsight. Those patients able to accurately act upon blind field stimuli (e.g. by pointing or saccading towards them) are classified as having action-blindsight, those whose residual functions can be said to rely to some extent upon attentive processing of blind field stimuli are classified as demonstrating attention-blindsight, while finally, patients who have somewhat accurate perceptual judgements for blind field stimuli despite a complete lack of any conscious percept, are classified as having agnosopsia literally meaning not knowing what one sees. We also address the possible neurological substrates of these residual sensory processes. Our second aim was to investigate the most striking subtype of blindsight, action-blindsight. We review the data relevant to this subtype and the hypotheses proposed to account for it, before speculating on how action-blindsight may inform our normal models of visuomotor control. q 2005 Elsevier Ltd. All rights reserved. Keywords: Blindsight; Visuomotor control; Parietal cortex Contents 1. Introduction A new taxonomy for residual behaviours in blindsight Parietal cortex and action-blindsight Action-blindsight operates in the here and now Action-blindsight and the automatic pilot Conclusion: action-blindsight the automatic pilot in slow motion? Acknowledgements References Introduction * Corresponding author. Tel.: C x7014; fax: C addresses: jdancker@watarts.uwaterloo.ca (J. Danckert), jdancker@uwaterloo.ca (J. Danckert), rossetti@lyon.inserm.fr (Y. Rossetti) /$ - see front matter q 2005 Elsevier Ltd. All rights reserved. doi: /j.neubiorev Blindsight refers to the residual visual abilities that some patients with visual field defects demonstrate for stimuli placed in their blind fields (Pöppel et al., 1973; Weiskrantz et al., 1974; Perenin and Jeannerod, 1975). That is, although patients with primary occipital (area V1) lesions are essentially blind in one visual hemifield, they can

2 1036 J. Danckert, Y. Rossetti / Neuroscience and Biobehavioral Reviews 29 (2005) nevertheless demonstrate above chance performance when responding to stimuli placed in their blind field. For example, when asked to guess, under the appropriate conditions, the location of a target that was briefly illuminated in the blind hemifield, some patients guess the location accurately on greater than 50% of trials (Weiskrantz et al., 1974; Zihl and Werth, 1984a,b). Initially, the most common residual ability demonstrated by blindsight patients was the ability to localize, either by pointing or eye movements, targets presented to the blind field. This ability to localize blind field targets has also been demonstrated in hemidecorticated patients (Perenin and Jeannerod, 1978; Ptito et al., 1991). Taken together, these results suggest subcortical involvement in the residual functions of these patients (see Jeannerod and Rossetti (1993) and Rossetti and Pisella (2002) for review). However, since the earliest work on blindsight, a wide range of residual functions have been described, ranging from motion, form and wavelength discrimination, to remarkable demonstrations of semantic priming from words presented to the blind field (Danckert et al., 1998; Magnussen and Mathiesen, 1989; Marcel, 1998; Morland et al., 1999; Stoerig and Cowey, 1989). Although still somewhat controversial, the performance of blindsight patients suggests that visual information is able to reach extrastriate visual cortex via pathways that do not depend on processing in area V1 (see Stoerig and Cowey (1997) for review). That is, it has been suggested that the residual pathway which runs from the eye directly to the superior colliculus and from there to the pulvinar nucleus of the thalamus is responsible for the ability to localize blind field targets (Weiskrantz et al., 1974; Zihl and Werth, 1984a,b). The many and varied residual abilities demonstrated by some blindsight patients may suggest, however, that blindsight relies on not one, but many residual pathways (Danckert and Goodale, 2000). 1 Accordingly, visual projections from subcortical structures, and in particular from the pulvinar, may project not only onto parietal but also onto temporal cortex. In addition, there is some recent anatomical evidence from the macaque monkey that demonstrated direct koniocellular inputs from the interlaminar layers of the LGN to the middle temporal (MT) motion-sensitive region of visual cortex (Sincich et al., 2004). This finding provides evidence for an alternate residual pathway that may subserve the Riddoch phenomenon (see below for a more detailed description of Riddoch phenomenon; Zeki and Ffytche (1998); see also Benevento and Yoshida (1981) for discussion of other LGN inputs to prestriate cortex). 1 There has been substantial debate in the literature concerning the possibility that the residual abilities observed in blindsight patients are in fact subserved by spared islands of cortex within V1. Recent neuroimaging evidence demonstrating extrastriate activation in GY in the absence of any such spared islands would seem to suggest this explanation does not suffice for all patients (see Danckert and Goodale (2000) for discussion of this issue). In this selective review we will suggest a new taxonomy for describing the various residual capacities demonstrated by blindsight patients. It is important to note that this taxonomy is intended to describe distinct types of residual behaviours demonstrated by blindsight patients. While some discussion of the neural networks underlying these distinct behaviours is obviously warranted, at this stage such a discussion is necessarily speculative. We will then explore in more detail one of the proposed definitions of a blindsight capability namely action-blindsight in which patients with V1 lesions are able to localize blind field targets by virtue of motor actions (e.g. pointing, grasping or saccades). Finally, we will examine how action-blindsight can inform models of visually guided action. 2. A new taxonomy for residual behaviours in blindsight The earliest demonstrations of blindsight involved asking the patient to motorically guess the location of a target that had been briefly flashed in the blind field (Pöppel et al., 1973; Weiskrantz et al., 1974; Perenin and Jeannerod, 1975). Weiskrantz first coined the term blindsight to account for the paradoxical observation of accurate eye and arm movements directed toward a visual target that was not consciously perceived. Since these early demonstrations, localization of targets presented to the blind field by various means has been by far the most common residual ability demonstrated (e.g. Weiskrantz et al., 1974; Perenin and Jeannerod, 1975; Zihl and Werth, 1984a,b; Danckert et al., 2003; Kentridge et al., 1999a,b). The ability to point or saccade towards a blind field target strongly supported the notion that the extrageniculate pathway directly from the eye to the superior colliculus was responsible for this residual ability (sometimes referred to as the retino-tectal pathway; Fig. 1). This was especially true for saccades made to blind field targets given the wealth of literature demonstrating collicular involvement in the control of eye movements (see Gaymard and Pierrot-Deseillingy (1999) for review). This hypothesis gained even further support from findings in hemidecorticated patients (Perenin and Jeannerod, 1978; Ptito et al., 1991). That is, despite having little or no remaining cortex in the damaged hemisphere, these patients were nevertheless able to show above chance localization of blind field targets, heavily implicating the retinofugal pathway from the eye to the superior colliculus (Perenin and Jeannerod, 1978; Ptito et al., 1991). The requirement that an action pointing or saccading be used to demonstrate above chance localization of blind field targets leads us to call this kind of residual function action-blindsight. As distinct from action-blindsight, some patients demonstrate residual abilities that do not completely lack a conscious percept (Magnussen and Mathiesen, 1989; Morland et al., 1999). For example, some patients may be able to discriminate the direction of motion of a stimulus presented in their blind field and furthermore, will report

3 J. Danckert, Y. Rossetti / Neuroscience and Biobehavioral Reviews 29 (2005) optic nerve optic chiasm pulvinar nucleus lateral geniculate nucleus optic tract lateral geniculate nucleus superior colliculi optic radiations extrastriate visual cortex Primary geniculo-striate pathway primary visual cortex (V1) retina optic tract LGNd V1 Retino-tectal pathway retina optic tract SC pulvinar extrastriate cortex Geniculo-extrastriate pathway retina optic tract interlaminar layers (koniocellular) of LGN extrastriate cortex Fig. 1. Schematic representation of the various pathways for visual information from the retina to striate (V1) and extrastriate cortex. The primary geniculostriate pathway is indicated by the by the dashed line from the temporal hemretina of the left eye and the widely space dotted line from the nasal portion of the right eye. The two secondary pathways indicated are shown originating from the optic tract for clarity, with the retino-tectal pathway indicated by the dashed/dotted line and the geniculostriate pathway indicated by the closely spaced dotted line. The pathways are also represented in simple box and arrow form below the schematic. Note, that recent anatomical work in the monkey has shown direct koniocellular projections to area MT (Sincich et al., 2004). The possibility exists for other such pathways from the interlaminar layers of the LGN to regions of extrastriate cortex other than area MT. experiencing a sensation of that stimulus albeit a sensation that is qualitatively distinct from actual vision of the stimulus. That is, the patient reports that they do not see, but rather have a sense or feeling that something had moved within their blind field, indicating some level of awareness of that stimulus the so-called Riddoch phenomenon (Morland et al., 1999; Zeki and Ffytche, 1998). This kind of residual ability in blindsight (awareness without seeing) may be related to alerting functions and may also depend on the integrity of the human homologue of monkey are MT (sometime referred to as area V5), which is known to subserve the processing of visual motion (e.g. see Dukelow et al. (2001)). Given that recent fmri research in humans has demonstrated increased activation in some areas of the human MT complex for both contralateral and ipsilateral motion stimuli (Dukelow et al., 2001) it may even be

4 1038 J. Danckert, Y. Rossetti / Neuroscience and Biobehavioral Reviews 29 (2005) Table 1 A new taxonomy for residual behaviours demonstrated in blindsight Action-blindsight Attention-blindsight Agnosopsia Residual behaviours Grasping, pointing, saccades Covert spatial orienting, inhibition of return, motion detection and discrimination Wavelength and form discrimination, semantic priming Paradigms Direct behaviour towards blind field stimuli Forced-choice guessing, implicit processing Forced-choice guessing paradigms Residual neural pathways SC pulvinar posterior parietal cortex (dorsal stream) SC pulvinar extrastriate visual cortex (MT and dorsal stream) Interlaminar layers of the dlgn extrastriate visual cortex (ventral stream) Examples from the literature Danckert et al. (2003), Jackson (1999), Perenin and Rossetti (1996), Weiskrantz et al. (1974), Zihl and Werth (1984a,b) and Perenin and Jeannerod (1975, 1978) Danziger et al. (1997), Kentridge et al. (1999a,b), Magnussen and Mathiesen (1989), Morland et al. (1999) and Walker et al. (2000) Marcel (1998), Stoerig and Cowey (1989) and Zeki and Ffytche (1998) the case that this residual ability is supported, at least in part, by the undamaged hemisphere. Further research would be needed to explore this possibility, as well as examining in more detail the contribution of subcortical structures. In addition to motion discrimination, other residual abilities have been demonstrated in blindsight that differ from those described for action-blindsight. These include some aspects of covert spatial orienting including inhibition of return and implicit task interference effects (e.g. the flanker task with flankers presented to the blind field) (Danckert et al., 1998; Kentridge et al., 1999a,b; Danziger et al., 1997; Walker et al., 2000). These abilities also appear to rely on attentional processes and are not necessarily associated with a specific action or effector. We refer to these residual abilities as attention-blindsight as a means of distinguishing them from action-blindsight. It is important to note that attention and action-blindsight are closely related and may depend upon aspects of the same residual neural pathways. For example, intact covert orienting abilities in blindsight patients may rely on the pathway from the eye to the superior colliculus described above (Danziger et al., 1997) as being responsible for action-blindsight. As we will suggest later, the differences in action and attentionblindsight may lie in the regions of extrastriate visual cortex involved. That is, it may turn out that the behavioural distinction proposed here also reflects subtle differences in the terminal region of extrastriate cortex that the residual neural pathways responsible for those functions project to. Furthermore, it may be the case that to demonstrate actionblindsight the patient may also need attentional processes of the type described in attention-blindsight to be functioning. An important determinant of the presence of attention- or action-blindsight may be the nature of the task used to examine residual functioning. Obviously, actionblindsight requires the patient to perform some kind of action towards blind field stimuli whereas attention blindsight can be demonstrated using implicit processing paradigms such as the flanker task in which the effect of blind field stimuli on sighted field stimuli is the critical measure (e.g. Danckert et al. (1998); Table 1). Research employing both kinds of tasks within the same patients will be needed to determine the extent to which attention- and action-blindsight co-exist or co-vary. Finally, residual abilities such as form or wavelength discrimination are more perceptual in nature and may rely on very different residual pathways, perhaps from the interlaminar layers of the LGN and from there to regions of extrastriate cortex that subserve the various functions demonstrated (Jeannerod and Rossetti, 1993; Rossetti and Pisella, 2002; Stoerig and Cowey, 1989, 1997; Danckert and Goodale, 2000; Girard et al., 1992; Perenin and Rossetti, 1996; Rossetti, 1998; Stoerig, 1996). We refer to this kind of blindsight as agnosopsia, a term first used by Zeki and Ffytche (1998) which literally means to not know what one sees (Zeki and Ffytche, 1998) that is, although the patient is completely unaware of blind field stimuli, they can nevertheless guess the correct perceptual characteristic at above chance levels (Table 1). We would suggest that the residual capacities falling under this classification depend on very distinct residual neural pathways from those underlying either attention- or action-blindsight. This may be evident at the level of thalamic nuclei such that the pulvinar nucleus is implicated in attention- and action-blindsight while the interlaminar layers of the LGN may be implicated in agnosopsia (Fig. 1 and Table 1). The latter pathway has recently been shown to exist in the macaque and to maintain direct connections with MT (Sincich et al., 2004). It remains to be seen whether or not there are similar projections from the interlaminar layers of the LGN to other regions of extrastriate cortex. In addition, the projections to extrastriate cortex from these residual pathways are also likely to differ such that the projections for agnosopsia may terminate in ventral extrastriate cortex known to be responsible for form and colour perception, while the projections for action-blindsight are more likely to terminate in dorsal extrastriate and posterior parietal cortices known to be important for the control of visually guided actions 2 2 The final pathway for attention-blindsight is less clear. While residual capacities such as motion discrimination may terminate in area MT (importantly, this region is not generally considered to be either dorsal or ventral in the human) other implicit processing capacities may terminate in distinct regions of extrastriate cortex.

5 J. Danckert, Y. Rossetti / Neuroscience and Biobehavioral Reviews 29 (2005) (Danckert and Goodale, 2000; Milner and Goodale, 1995). Finally, the distinction between attention-blindsight and agnosopsia is reminiscent of Weiskrantz s Type I and Type II blindsight distinction (Weiskrantz, 1998). That is, having a sense of something in the blind field despite not seeing it per se Weiskrantz s Type II blindsight is what characterises some of the phenomenon we are including under attention-blindsight (e.g. Riddoch phenomenon; see Zeki and Ffytche (1998)). In contrast, Type I blindsight is more akin to agnosopsia in which the patient s above chance performance is never accompanied by a conscious percept (albeit a degraded one; see Table 1). Our distinction focuses on the behaviours demonstrated and to some extent the methodologies used to elucidate those behaviours, rather than the direct association between the behaviour and various levels of awareness. What is important to emphasise here is that different residual abilities in blindsight are likely to rely on different residual neural pathways Parietal cortex and action-blindsight Although the description of action-blindsight discussed above suggests involvement of the superior colliculus in the ability to localize blind field targets, this explanation may not be sufficient to explain the same ability when the required response was not an eye movement but a pointing movement. Indeed, Weiskrantz and colleagues (1974) found that in one patient the ability to localize blind field targets was dramatically better when the patient was required to point to those targets rather than making an eye movement (Fig. 2). Such a performance suggests cortical involvement in the residual ability to localize blind field targets a point made by Weiskrantz and colleagues in their original work. This is not to suggest that the superior colliculus is not involved in the control of reaching movements (note, we are using the term reaching here rather than pointing). Recent work in the cat has demonstrated that stimulation of the superior colliculus just after the onset of a reaching movement leads to a perturbation in movement trajectory (Courjon et al., 2004). Importantly, the perturbed reaching movements are then corrected on-line such that the target of the movement is still accurately acquired (Courjon et al., 2004). In addition, the latency of saccades in the rhesus monkey are reduced when they are accompanied by an arm movement in the same direction as the saccade (Snyder et al., 2002), suggesting a role for the superior colliculus in hand eye co-ordination (see Lunenburger et al. (2001) for review). While this work demonstrates a role for the colliculus in the transport component of reaching movements it does not necessarily implicate the midbrain in the end stage 3 Stoerig (1996) discusses the different residual abilities related to lesions at different levels in the visual system (e.g. from the optic nerve to the LGN and through to V1). The distinction we are making here is within the final residual pathway Stoerig discusses the extra-geniculo-striate pathways and extrastriate cortical areas (p. 402). Eye position (degrees) Eye movements Target position (degrees) Hand movements components of pointing or grasping movements. Recent work in the macaque monkey suggests that complex coding of eye in head position may depend on structures downstream from the colliculus (Klier et al., 2003). Results of this kind do seem to suggest that complex components of visually guided movements over and above specification of direction and amplitude may depend on processing in neural regions beyond the colliculus. What is important to emphasise here, is that the residual pathway from the eye to the colliculus may not fully account for residual abilities such as pointing to blind field targets. Instead, additional processing in the cortical regions within which this pathway terminates may be required for accurate action-blindsight to occur. 4 Interestingly, pointing is not the only action that has been shown to be above chance in patients with blindsight. More recently the ability of two blindsight patients to process both size and orientation of a target presented in the blind field was investigated using three types of response: perceptual matching, goal directed action (i.e. grasping to examine action-discrimination of size differences and posting a card to assess action-discrimination of orientation differences; see Perenin and Vighetto (1988) for original description of these tasks) and verbal report (Perenin and Rossetti, 1996; Rossetti, 1998). They found that the patient s performance was above chance only for the goal-directed actions (i.e. grasping or posting). Finger position (degrees) Target position (degrees) Fig. 2. Eye position (left panel) and finger pointing position (right panel) for one blindsight patient (DB) tested by Weiskrantz and colleagues (1974). The patient was required to localize blind field targets in both cases. Bars represent the range of responses recorded at each target eccentricity. (Note: the left panel is adapted from Fig. 2 of Weiskrantz et al. (1974) in which the patient made saccades following the presentation of targets subtending 28 of visual angle, while the right panel is adapted from Fig. 3b of Weiskrantz et al. (1974) in which the patient pointed to targets subtending of visual angle). 4 Presumably, hemidecorticate patients who are capable of localising blind field targets at above chance levels are relying on only collicular and subcortical processing to perform this task. There is some controversy concerning the ability of such patients to localize blind field targets when the target is of lower luminance than the background (King et al., 1996; Scharli et al., 1999). The variable performance of hemidecorticate patients under such conditions suggests that light scatter from high luminance targets may actually be informing their performance. In addition, to our knowledge there has been no study examining more complex residual abilities such as grasping or wavelength discrimination in these patients.

6 1040 J. Danckert, Y. Rossetti / Neuroscience and Biobehavioral Reviews 29 (2005) That is, their patient was able to scale his grip aperture (i.e. the distance between forefinger and thumb) appropriately to the target s size but performed below chance when perceptually matching the target s size or verbally reporting it. A similar performance was observed for orientation judgements in the posting task. That is, the patient performed above chance levels when required to post an object through a slot that varied in orientation from trial to trial, but performed at or below chance level when matching or verbally reporting the slot s orientation (Perenin and Rossetti, 1996; Rossetti and Pisella (2002)). Similarly, Jackson (1999) examined the ability of patient GY, a well studied blindsight patient, to grasp objects that extended into his blind field (the objects in Perenin and Rossetti s (1996) study were completely within the blind field). That is, GY was asked to grasp objects whose horizontal aspect covered the same extent within his sighted field from trial to trial but extended into his blind field to differing degrees. As was the case for the patient in the Perenin and Rossetti (1996) study, GY was able to demonstrate accurate grip scaling under these conditions, such that the distance between his forefinger and thumb was accurately scaled to the object size despite the fact that he was unable to see the degree to which objects extended into his blind field (Jackson, 1999). It has also been demonstrated that the ability to accurately scale grip aperture to blind field objects is lost when there is a short delay between target presentation and the onset of the movement into the blind field (note: this distinction was based on the patient s own movement initiation time, such that movements with slower RTs showed a poorer relationship to object size than did movements with faster RTs; Rossetti, 1998). This result highlighted the importance of on-line or automatic motor control in this patient s performance (Rossetti and Pisella, 2002). The on-line control of such grasping movements has been shown to depend on the dorsal action pathway, which runs from V1 through to posterior parietal cortex (Jeannerod and Rossetti, 1993; Milner and Goodale, 1995; Goodale and Milner, 1992). While subcortical structures such as the superior colliculus are also likely to play a role in such movements, it is important to emphasise that grip scaling and orientation judgements require some degree of cortical involvement (Culham et al., 2003). This is especially so for orientation judgements as there have been no demonstrations to the best of our knowledge of orientation selectivity in the monkey superior colliculus. Therefore, for a blindsight patient to demonstrate accurate grip scaling and orientation judgements towards blind field stimuli it is necessary for visual information to reach the dorsal extrastriate and posterior parietal cortex even in the face of no input from V1. However, the input signals to this area may arise from several different regions (Rossetti and Pisella, 2002). As a matter of fact, patients with occipital lesions exhibit poorer performance in the card posting task described above (Perenin and Rossetti, 1996) than do patients with a lesion of the ventral stream (Goodale and Milner, 1992). A comparison of the two types of performance is shown in Rossetti and Pisella (2002), p. 70, Fig. 4.2). Therefore, the optimal processing performed in the posterior parietal cortex is likely to require inputs from both V1 and subcortical structures. We recently demonstrated that an intact posterior parietal cortex (PPC) is indeed necessary for demonstrating action-blindsight (Danckert et al., 2003). We explored the ability of two patients with hemianopia to localize blind field targets by pointing. In one patient (JR), the PPC was generally spared, while in the other (YP) there was more extensive damage of extrastriate cortex extending well into the PPC. On a touch screen version of the pointing task only patient JR demonstrated above chance localization. We then explored the kinematics of pointing movements made to blind field targets (this unavoidably led to significant changes in stimulus setup and target types; Danckert et al., 2003). Although patient YP now demonstrated above chance localization of blind field targets, his performance showed several qualitative differences when compared with patient JR. That is, the strength of the relationship between actual target locations and the patients responses was strongest for patient JR in which there was greater sparing of extrastriate and posterior parietal cortex. In addition, a strong relationship was found between peak velocity, time to reach peak velocity and target location only for patient JR (i.e. both peak velocity and time to reach peak increased with increasing target distance from fixation a common effect observed in healthy individuals; Danckert et al., 2003). What this work clearly demonstrates is that an intact PPC (and perhaps a greater degree of sparing of dorsal extrastriate cortex) is needed for action-blindsight to be apparent in patients with hemianopia an assertion initially made by Weiskrantz and colleagues (1974). Put another way, for action-blindsight to be evident in a hemianopic patient, the residual neural pathway implicated is likely to terminate in the dorsal action stream given the nature of the residual functions observed. This pathway is ideally placed to subserve the residual abilities we are calling action-blindsight as it terminates in regions of dorsal extrastriate cortex which in turn project to the PPC areas known to be crucial for the control of eye and hand movements (Jeannerod and Rossetti, 1993; Milner and Goodale, 1995; Culham et al., 2003; Connolly et al., 2000). Recent functional magnetic resonance imaging (fmri) in patient GY revealed activation in dorsal extrastriate cortex in the damaged hemisphere in response to visual stimuli placed in the blind field (Baseler et al., 1999). Similar residual activity in dorsal stream structures has been observed in monkeys following cooling of area V1 (Girard et al., 1992, 1991). Interestingly no activity was found in the ventral stream in spite of the anatomical connectivity

7 J. Danckert, Y. Rossetti / Neuroscience and Biobehavioral Reviews 29 (2005) mentioned above. These results provide strong support for the suggestion that action-blindsight depends on the integrity of residual pathways that terminate in the dorsal action stream Action-blindsight operates in the here and now The nature of the tasks used to explore residual visual functions in blindsight may determine to a large extent the performance of the patient (Danckert and Goodale, 2000; Stoerig, 1996). For example, the discrimination of wavelength and form (i.e. what we and others have termed agnosopsia; Zeki and Ffytche, 1998) is typically only demonstrated in forced choice guessing paradigms and over a very large number of trials (Table 1; Stoerig and Cowey, 1989). The residual abilities characteristic of attention-blindsight (e.g. implicit interference effects, motion discrimination) can be demonstrated using both forced choice and implicit processing paradigms. 5 In contrast, action-blindsight for both pointing (or saccades) and grasping movements, is typically demonstrated using tasks in which the patients behaviour is directly measured as they perform the action in their blind field (although this is not always true, as localization of blind field targets has also been explored using forced choice procedures; e.g. Danziger et al. (1997)). Furthermore, significant demonstrations of action-blindsight have been observed after a relatively small number of trials (Danckert et al., 2003; Danziger et al., 1997; Perenin and Rossetti, 1996). Finally, when a brief delay is introduced between the presentation of the target in the blind field and the initiation of the action directed towards that target, performance returns to chance level (Rossetti, 1998). 6 Taken together, this evidence suggests that for action-blindsight to be evident the behaviour must be performed in direct response to or immediately after the presentation of a target. The suggestion that action-blindsight operates only for immediate actions fits nicely with work in healthy individuals demonstrating both qualitative and quantitative differences in the control of skilled actions performed immediately in response to a target or following a delay between target presentation and action onset (for review see Rossetti and Pisella (2002) and Rossetti (1998)). One elegant example of the effects of a delay on the control of actions can be seen when healthy individuals are required to grasp objects embedded within pictorial illusions (e.g. size-contrast illusions including the Ebbinghaus and Müller-Lyer illusions) (Aglioti et al., 1995; Gentilucci et al., 1996). Typically, when subjects reach to grasp target objects imbedded within a size-contrast illusion, no effect is seen 5 Of course for motion discrimination based on actual, albeit degraded percepts, overt discriminations can be made (Morland et al., 1999). 6 Interestingly, the same effect of delay has been reported for the somatosensory equivalent of blindsight that is, numbsense for both tactile and proprioceptive stimuli (Rossetti et al., 1995). on grip scaling. In other words, the action is unaffected by the illusion. 7 In contrast, when asked to delay their action for a period of 2 s or more, grip scaling now shows a dramatic influence of the illusory context (Hu et al., 1999; Hu and Goodale, 2000; see also Gentilucci et al. (1996)). Interestingly, while the visual form agnosic patient DF is able to accurately grasp objects (despite impaired perceptual discrimination abilities with respect to those same objects), when she performs the same actions following a delay imposed between target presentation and movement onset, her performance is now greatly impaired (Goodale et al., 1991, 1994). Milner and Goodale (1995) suggested that the differences in the immediate control of actions compared with the control of the same actions following a delay, is indicative of the division of labour between the dorsal action and ventral perception pathways. That is, while the dorsal pathway, which runs from V1 to posterior parietal cortex depends on moment-to-moment, or on-line calculations of spatial relationships to control skilled actions, the ventral perception pathway, which runs from V1 to inferotemporal cortex, operates on stable visual representations of objects and their spatial relationship to one another stored in long term memory (see Rossetti and Pisella (2002), Rossetti (1998), Pisella and Rossetti (2000) and Milner and Dijkerman (in press) for review). If actionblindsight also depends on the dorsal pathway, then it too should operate on a moment-to-moment basis. In other words, action-blindsight should only be evident when the patient initiates their action in immediate response to the presentation of a target (obviously while maintaining central fixation; see Danckert et al. (2003) for an example) or immediately after it has been extinguished (as is the case when patients are allowed to look towards the blind field location they believe a target was presented in Weiskrantz et al. (1974)). The fact that introducing a delay between blind field target presentation and action initiation eliminated any statistical relationship between grip scaling and target size in one hemianopic patient, suggests that actionblindsight does rely on the immediate initiation of actions towards blind field targets to be successful (Rossetti, 1998) Action-blindsight and the automatic pilot So far we have suggested that action-blindsight depends on the integrity of visual pathways that terminate in 7 It is important to note that some influence of the illusion on grasping can be seen in these studies although the magnitude of the effect is far smaller than the influence observed on perception (Westwood et al., 2000). In addition, recent work has demonstrated that the immunity of grasping actions to illusory contexts is evident throughout the course of a movement, suggesting that the programming of the movement, and not on-line control, is responsible for the fact that actions are impervious to illusions (Danckert et al., 2002). This contention is far from uncontroversial (Glover and Dixon, 2001). What we are intending to suggest here is not a distinction between planning and on-line control but a distinction between the immediate and delayed control of actions.

8 1042 J. Danckert, Y. Rossetti / Neuroscience and Biobehavioral Reviews 29 (2005) the PPC (i.e. the dorsal visual stream) and that rely upon immediate initiation of actions in response to blind field stimuli (see Rossetti (1998) for review). Given that these residual abilities are carried out by definition in the absence of visual awareness of the targets, one further suggestion would imply that action-blindsight operates in an automatic manner. Indeed, recent evidence from healthy individuals and other neurological disorders does imply that the PPC often functions automatically, rapidly modifying visually guided hand movements often in contradiction to conscious commands (Pisella et al., 2000). Some of the evidence for automatic processing in the PPC will be reviewed here (see also Rossetti and Pisella (2002), Pisella and Rossetti (2000) and Rossetti et al. (in press)) before we discuss the implications of this work for action-blindsight. As mentioned above, recent research in patients with a neurological disorder known as optic ataxia suggests that the dorsal action pathway often operates in a largely automatic manner (Milner and Dijkerman, in press). Optic ataxia is an impairment observed following lesions of superior parietal cortex, often bilaterally (Jeannerod and Rossetti, 1993; Perenin and Vighetto, 1988; De Renzi, 1974, 1982; Vighetto and Perenin, 1981). Although no motor or visual sensory deficits are typically observed on clinical testing, closer examination shows that the patient has great difficulty directing their arm towards a visual target presented in their peripheral visual field (Perenin and Vighetto, 1988; Vighetto and Perenin, 1981). Typically, an interaction is found between the hand used to carry out the action and the visual field in which the action is directed, such that the patient s performance is worse with the hand contralateral to the lesion when (s)he reaches into the contralateral peripheral visual field, whereas the deficit is minimal when the ipsilateral hand reaches into the ipsilateral field. In addition, these impairments are most evident for movements made in immediate response to the appearance of a visual target. Recent work has recast the impairments observed in optic ataxia as a disruption to what has been dubbed a parietal automatic pilot (Pisella et al., 2000). Primary evidence for this account of optic ataxia comes from the double-step pointing task (Prablanc et al., 1986). In this paradigm subjects are required to point to targets that on a small percentage of trials can be perturbed after movement onset. In healthy individuals, pointing movements can be corrected on-line to such perturbations in location or even size of the target object (Rossetti and Pisella, 2002; Pisella and Rossetti, 2000; Prablanc et al., 1986; Desmurget et al., 1995; Goodale et al., 1986; Gréa et al., 2000; Pélisson et al., 1986; Prablanc and Martin, 1992; Rossetti et al., 2000). That is, subjects are able to make rapid adjustments to goal-directed actions when some aspect of the target (e.g. location or size) is perturbed (see Rossetti and Pisella (2002) and Rossetti et al. (2000)) for review). In addition, after a fast pointing movement has been programmed and initiated toward a visual target, it can be corrected without a significant increase in movement time (Goodale et al., 1986; Pélisson et al., 1986; Prablanc and Martin, 1992). That is, the rapid adjustment of the movement to the perturbed target location occurs without the need to reprogram a new motor output. Finally, such on-line corrections can be observed whether or not the target displacement is consciously perceived (Goodale et al., 1986; Pélisson et al., 1986; Prablanc and Martin, 1992). These results suggest that such corrections are executed automatically. Neuroimaging evidence has revealed increased activation in a network of structures including the PPC when such automatic corrections to pointing movements are made (Desmurget et al., 2001). In addition, when functioning in the PPC is disrupted due to transcranial magnetic stimulation (TMS), the ability to correct movements to target perturbations is also disrupted (Desmurget et al., 1999). This evidence strongly implicates the PPC in the on-line, automatic control of visually guided actions. In light of this research, the impairments characteristic of optic ataxia can be reconsidered as an impairment of real-time control of visually guided actions rather than of motor programming (see Rossetti and Pisella (2002) for review). It may be the case that the same areas implicated in the on-line control of movements discussed above are also responsible for actionblindsight. In at least one neuroimaging study of a blindsight patient activity in dorsal extrastriate regions that project to the PPC was observed (Baseler et al., 1999). Further neuroimaging work that explicitly examines behaviours such as pointing or saccading to blind field targets and not simply passive neural responses to blind field stimuli, will be needed to determine the veracity of the claim that actionblindsight depends on processing in the PPC. If the ability to correct rapid goal-directed actions can bypass conscious awareness, to what extent can such an automatic system process visual information? A modification of the double-step pointing paradigm was used to explore this question (Rossetti and Pisella, 2002; Pisella and Rossetti, 2000; Pisella et al., 1998, 1999). In this version of the paradigm target perturbations could be used as a go-signal to perform an in-flight correction to acquire the new target location (the location-go task) or alternatively, as a stop-signal (the location-stop task) to interrupt the ongoing movement. Therefore, the two types of response, go or stop, required alteration to an already planned (and presumably ongoing) pointing action. For both versions of the task subjects were instructed to initiate a rapid pointing movement towards the target. Movement duration was tightly controlled using auditory feedback so that a sample of movement times was taken (i.e. if the subject moved too slowly a tone sounded to inform them that a quicker movement was required). Successful performance in the location-stop task would presumably lead to a cessation of the rapid pointing movement. Initially, it was expected that any failure to comply with task instructions would lead to a movement that was completed to the original target location in other words an inability to alter the first

9 J. Danckert, Y. Rossetti / Neuroscience and Biobehavioral Reviews 29 (2005) motor program made towards the initial target location (Pisella et al., 2000). In striking contrast to this prediction, a significant percentage of corrective movements were performed in the direction of the target jump in spite of the instruction to halt their movements. These corrections were considered to be automatic because they were produced spontaneously by naive subjects against their own intention to stop their movement in accordance with instructions. After touching the displaced target, subjects were fully aware of their mistakes and impulsively expressed frustration at their error. This automatic pilot (see also Place (2000)) systematically activated during movement execution led subjects to produce disallowed corrective movements over a narrow range of movement times (between 150 and 300 ms). Importantly, the same rate of movement corrections observed in the location-stop condition were also found in the control location-go condition where subjects were allowed to adjust their pointing movement in response to target perturbations. That is, whether they were asked to adjust or stop their pointing movements in response to target perturbations, subjects made approximately the same number of corrections. Only movements slower than 300 ms could be fully controlled by voluntary processes (Pisella et al., 2000). In contrast to the performance of healthy individuals on the location-stop task, a patient with a bilateral lesion of the PPC (patient IG) showed a complete lack of on-line automatic corrective processes, whereas the slower intentional motor processes were preserved (Pisella et al., 2000). That is, when asked to interrupt her rapid pointing movements to target perturbations IG was fully able to do so, unlike healthy controls, suggestive of a disruption to her automatic pilot (the same pattern of behaviour has been observed in another bilateral optic ataxia patient (Pisella, personal communication). This result suggests that rapid pointing movements are controlled by a posterior parietal automatic pilot located in the dorsal stream. In contrast, slow movements are controlled by intentional motor processes that do not necessarily rely as heavily on the posterior parietal cortex. Thus the notion of an automatic pilot extended that of hand-sight (Rossetti et al., 2000) in the sense that it refers not only to unconscious visual processing by the action system, but also to an autonomous use of visual information which bypasses and even counteracts intentional control. For the purposes of the current paper we are suggesting that the parietal automatic pilot may also be the final point in the neural pathway that is responsible for action-blindsight. 3. Conclusion: action-blindsight the automatic pilot in slow motion? We have previously demonstrated that a greater degree of sparing of the PPC is associated with more robust action-blindsight (Danckert et al., 2003). In those same patients we attempted to explore the possibility that automatic corrections of pointing movements would also be possible even though the patients were never aware of the presence of the targets or perturbations in target locations (Danckert and Rossetti, unpublished data). That is, if actionblindsight does indeed depend on the integrity of a residual pathway that terminates in the PPC the same pathway we suggest is responsible for automatic corrections of pointing movements to target perturbations then similar corrections should be evident in action-blindsight patients for target perturbations occurring in the blind field. We found no such evidence for an ability to correct movements to blind field target perturbations (Danckert and Rossetti, unpublished data; Fig. 3). The absence of significant adjustments to target perturbations in the blind field could be due to several factors that we raise here for potential future studies. Perhaps the most striking factor involved is movement duration. The movement durations of both patients for perturbed trials were generally longer than around 400 ms outside the range of automatic corrections (or involuntary errors) seen in healthy individuals (Pisella et al., 2000). This is an important point. When asked to adjust their rapid pointing movements to perturbations in target location, healthy individuals were only able to do so on approximately 15% of trials, all at movement durations of between 150 and 300 ms (Pisella et al., 2000). The evidence to suggest that these were automatic corrections comes from the observation that a similar percentage of movement corrections were observed even when subjects were told to stop their movement in response to a location perturbation once again, all occurring at movement durations of around 300 ms or less (Pisella et al., 2000). This time frame is quite a bit shorter than the average movement duration seen in either of our hemianopic patients (for JR mean MDZ540.2 ms; for YP mean MDZ ms). What implications does this have for the suggestion that the secondary visual pathway from the colliculus to the PPC is responsible both for actionblindsight and the automatic pilot? The first, most obvious explanation would suggest that both behaviours (actionblindsight and the parietal automatic pilot) are not supported by the same pathway. Second, the automatic pilot (V1 intact) requires a heavier visual input than is required for action-blindsight. In other words, in order to show corrections to rapid pointing movements the PPC may require input from V1. Alternatively, the same neural pathway may be involved but for patients with blindsight this pathway may operate in a different manner. The suggestion here is that to localize blind field targets or to adjust a pointing movement to a target perturbation, the same pathway from the colliculus to the PPC via the pulvinar is involved. But for action-blindsight patients the uncertainty induced by the conscious awareness of their field defect makes this pathway function more slowly than would otherwise be the case the automatic pilot operating

10 1044 J. Danckert, Y. Rossetti / Neuroscience and Biobehavioral Reviews 29 (2005) A fixation LED targets B 300 end x-position Patient JR target fixation auditory tone target 1 movement target 2 starting position perturbed target trial 2 secs touch screen monitor backward perturbation forward perturbation end x-position C end x-position end x-position Patient YP target backward (3 to 2) backward (3 to 2) target 3 target 3 Patient JR forward (3 to 4) Patient YP forward (3 to 4) Fig. 3. Panel A. Schematic representation of the experimental setup examining pointing behaviour in two blindsight patients. The patient started by resting their finger on a starting position at the bottom of a touch screen monitor. The patient fixated a constantly illuminated LED taped to the far side of the monitor. For unperturbed trials an auditory tone was presented coincident with the onset of the target which remained on the screen for a period of 2 s. Perturbed trials always began with a target at the central location (location 3 in the schematic). For perturbed trials (50% of all trials) the location of the initial target moved either to the right (termed a forward perturbation) or the left (termed a backward perturbation) initiated by movement onset (i.e. when the patient lifted their finger from the touch screen). Panel B. Mean end point along the x-axis for patient JR (above) and YP (below) for unperturbed targets. Patient JR s movements were significantly correlated with target location while YP s were not (but see Danckert et al. (2003) for a different version of this task in which YP does show some degree of action blindsight). Panel C. Mean end point along the x-axis for pointing movements to perturbed targets for JR (above) and YP (below). Neither patients pointing movements showed any significant relationship to the perturbed target locations. In both panels B and C the dotted lines represent linear regression lines fitted to the patients data. in slow motion. This hypothesis can be easily tested simply by forcing patients with hemianopic field defects to point to blind field targets within tightly controlled time frames (less than 300 ms). While one may not expect blindsight patients to perform at the same level as controls, the hypothesis proposed here would suggest that performance for both static and perturbed blind field targets would improve relative to performance at longer time frames. Although both of our blindsight patients tended to have long movement durations, JR s movements were on average faster than YP s. Given that the ability to localize unperturbed blind field targets was more robust for JR than it was for YP (Danckert et al., 2003), this provides some support for the notion that fast, automatic control of movements is required for hemianopic patients to demonstrate action-blindsight. The present comparison between action-blindsight and optic ataxia suggests that at least two types of afferences to the PPC may participate in the control of action. The dorsal action pathway from V1 through to the PPC and the more limited subcortico-cortical pathway from the colliculus to the PPC via the pulvinar. While it may be the case that this more minor subcortico-cortical pathway cannot enable normal visuo-motor behaviour (see the comparison between visual agnosia and blindsight in Rossetti and Pisella (2002)), it is nevertheless true that it supports some control of skilled actions. Importantly, this control is likely to be automatic in nature (Rossetti and Pisella, 2003). The main challenge arising from the present result is to delineate the contribution of these two pathways, as well as other possible pathways, to the integrated control of visually guided actions both in healthy individuals and in patient populations.