OUP user menu

Changes in connectivity after visual cortical brain damage underlie altered visual function

Holly Bridge , Owen Thomas , Saâd Jbabdi , Alan Cowey
DOI: http://dx.doi.org/10.1093/brain/awn063 1433-1444 First published online: 9 May 2008

Summary

The full extent of the brain's ability to compensate for damage or changed experience is yet to be established. One question particularly important for evaluating and understanding rehabilitation following brain damage is whether recovery involves new and aberrant neural connections or whether any change in function is due to the functional recruitment of existing pathways, or both. Blindsight, a condition in which subjects with complete destruction of part of striate cortex (V1) retain extensive visual capacities within the clinically blind field, is an excellent example of altered visual function. Since the main pathway to the visual cortex is destroyed, the spared or recovered visual ability must arise from either an existing alternative pathway, or the formation of a new pathway. Using diffusion-weighted MRI, we show that both controls and blindsight subject GY, whose left V1 is destroyed, show an ipsilateral pathway between LGN (lateral geniculate nucleus) and human motion area MT+/V5 (bypassing V1). However, in addition, GY shows two major features absent in controls: (i) a contralateral pathway from right LGN to left MT+/V5, (ii) a substantial cortico-cortical connection between MT+/V5 bilaterally. Both observations are consistent with previous functional MRI data from GY showing enhanced ipsilateral activation in MT+/V5. There is also evidence for a pathway in GY from left LGN to right MT+/V5, although the lesion makes its quantification difficult. This suggests that employing alternative brain regions for processing of information following cortical damage in childhood may strengthen or establish specific connections.

  • visual cortex
  • blindsight
  • diffusion tensor imaging
  • functional MRI
  • lateral geniculate nucleus

Introduction

A crucial question following any damage to the brain is how much neural plasticity can be recruited to minimize the long-term effects of damage and to provide a basis for functional rehabilitation. The traditional view of the brain and spinal cord is that the majority of connections are laid down early in life, and, although some will later be pruned, few will be created beyond early childhood. It has recently been suggested, with traditional fibre tracing (Dancause et al., 2005) and imaging techniques (Bengtsson et al., 2005) that changes in local connectivity can be induced by practice or injury. However, it is less clear whether experience can provoke more fundamental changes in anatomical connectivity patterns, and whether any such changes involve the recruitment and strengthening of existing pathways, or include genuinely new pathways not otherwise present.

A well-studied example of altered behavioural function following cortical damage is the phenomenon of ‘blindsight’. Partial loss of the primary visual cortex (V1), the first cortical visual area, leads to a scotoma corresponding retinotopically to the region that is destroyed. However, many patients with such damage are nevertheless able to make remarkably accurate psychophysical judgements about stimuli presented within the blind field, despite not being able to visually perceive them, hence the term blindsight (Weiskrantz et al., 1974). There is a clear bias in blindsight towards properties processed by the dorsal cortical visual stream (luminance contrast, flicker, motion) that has led to a proposal that there is a direct projection to the middle temporal area (MT/V5) from subcortical areas that bypasses V1 in the undamaged brain. Support for this idea comes from physiological studies in which V1 of the macaque monkey was inactivated, either permanently (Rodman et al., 1989), or reversibly (Girard et al., 1992). Despite lack of V1 activity, neuronal responses to visual stimuli could still be recorded in MT. More recent findings in New World monkeys however, are less clear, with one study reporting some activity in MT several weeks after a permanent lesion to V1 (Rosa et al., 2000), while Collins et al. (2003) found no residual activity after similar lesions.

Initial anatomical investigations into a direct connection between LGN and MT were contradictory (Yoshida and Benevento, 1981; Benevento and Standage, 1982; Horton, 1984; Sorenson and Rodman, 1999). However, more recently a direct input to MT predominantly from the koniocellular interlaminar layers of the LGN has been described in macaque monkeys by Sincich et al. (2004). This LGN pathway, demonstrated by using retrograde tracing, was estimated to be around 10% the size of the V1 to MT pathway.

Blindsight subject GY, whose residual or recovered visual properties have been carefully documented and often attributed to activity of his area MT+/V5, is an ideal subject to investigate changes in connectivity due to altered experience. It allows the comparison of his brain to those of normal subjects in whom there may be a direct connection between LGN and MT+/V5 to determine any changes in the visual pathways. Diffusion-weighted MRI (DW-MRI) (Le Bihan, 2003) offers a unique method to explore non-invasively the organization of white matter in the living human brain. Diffusion imaging characterizes the apparent diffusion properties of water molecules (Basser et al., 1994a, b), which can be used to determine the orientation of large white matter fibres in the brain. The local orientation of the tissues can then be used for tractography, to reveal hitherto untraceable distant connections between brain areas (Mori et al., 1999).

We have used probabilistic diffusion-based tractography (Behrens et al., 2003) to investigate a direct connection between anatomically defined LGN and functionally defined human MT+/V5 in six controls and blindsight subject GY. In 14/14 hemispheres, a pathway between these two areas was present, even though GY's left LGN is shrunken as a result of the destruction of his left V1. In addition, we investigated the strength of the connection between MT+/V5 via the splenium in the two hemispheres. Subject GY differed conspicuously from controls in two major aspects: (i) he showed a transcallosal projection from the LGN to the contralateral MT+/V5 on both sides and (ii) he showed a much more prominent transcallosal connection between MT+/V5 in the two hemispheres. These observations suggest that there is plasticity in anatomical and functional long-range connections even when damage occurs relatively late in development (in his case at 8 years) and that these changes may contribute to the phenomenon of blindsight. More generally, it is possible that the anatomical changes in connectivity are driven by a compensatory reaction to brain damage, an intriguing possibility that awaits further investigation.

Methods

Subjects

This study was conducted under ethical approval from the Oxfordshire NHS Research Ethics Committee (05/Q1605/143). Subjects gave informed consent in accordance with the Declaration of Helsinki and were compensated for their time. Five healthy controls, aged between 25 and 33 (one female) and one male aged 53, with normal or corrected to normal, vision participated in the study. Blindsight subject GY was aged 53 at the time of this study, and had been extensively studied previously (Barbur and Ruddock, 1980; Barbur et al., 1980; Blythe et al., 1986, 1987; Weiskrantz et al., 1991; Baseler et al., 1999; Azzopardi and Cowey, 2001; Morland et al., 2004). He has a large unilateral lesion in the left medial occipital lobe, caused by a traffic accident at the age of 8. Striate cortex is absent in the left hemisphere, except at the occipital pole corresponding to about 3–4° of macular sparing. In addition to the occipital lesion, there is a smaller lesion in the right parietal lobe that has not been investigated behaviourally. GY's visual system has been tested both at a behavioural level, and using fMRI (Sahraie et al., 1997; Zeki and Ffytche, 1998; Baseler et al., 1999; Morland et al., 2004).

Data acquisition

Data were acquired using scanners at the Oxford Centre for Clinical Magnetic Resonance Research (OCMR). All subjects participated in session 1 in which the DW-MRI data and the functional scan to localize MT+/V5 were performed. The data were acquired using an 8-channel head coil with a 1.5 T Siemens Sonata scanner, having a maximum gradient strength of 40 mT/m. Visual stimuli were generated on a VSG 2/5 graphics card (Cambridge Research Systems), and projected using a Sanyo projector. Subjects used mirrors to view the stimuli projected onto a screen near their waist. This arrangement afforded a visual field of ∼20 × 20°.

Diffusion and functional data were acquired axially using echo-planar imaging, with isotropic voxels of 2.5 × 2.5 × 2.5 mm3. Fifty slices were acquired for the fMRI scans, and 60 for the DW-MRI. Standard parameters were used in the fMRI scan.

In the diffusion scans, the diffusion weighting was isotropically distributed through space (Jones et al., 1999) along 60 directions using a b-value of 1000 s/mm2. Three sets of diffusion-weighted data were collected, and for each set five volumes with no diffusion weighting were acquired during the sequence. The total acquisition time for the diffusion imaging was ∼40 min.

A standard method of localizing MT+/V5 was used in which moving dots were contrasted with stationary dots in a block design. Each complete cycle of moving/stationary blocks took 32 s and was presented eight times, giving a total scan length of 256 s.

A proton density scan was acquired on a Siemens Trio 3 T scanner using a 12-channel head coil. Forty 2 mm slices were acquired coronally at an in-plane resolution of 750 × 750 μm2 using a fast spin echo protocol. The combination of a long repetition time and short echo time minimizes the T1 and T2 weighting, such that proton density is the primary tissue contrast (Jackson et al., 1997). Each scan took just less than 6 min, and three repeats of the same scan were acquired for averaging.

For registration and anatomical localization purposes, a T1-weighted anatomical image of 1 × 1 × 1 mm3 was acquired using the 3 T scanner (3D turbo FLASH, TR = 15 ms, TE = 6.9 ms).

Functional data analysis

Functional data to locate MT+/V5 were analysed using FEAT, part of the FSL toolbox (www.fmrib.ox.ac.uk/fsl). Images were pre-processed using a number of steps: motion correction using MCFLIRT, spatial smoothing with a Gaussian kernel FWHM = 5 mm, mean-based intensity normalization, non-linear highpass temporal frequency filtering (Gaussian-weighted straight line fitting, with sigma = 32 s). To functionally define MT+/V5, a Z-statistic (Gaussianized T) image of the contrast moving dots > stationary dots was created. The Z-statistic threshold was chosen individually for each subject (median Z = 3.9) such that the volume of MT+/V5 was 2500 mm3 in each hemisphere. While it is widely accepted that the volume of MT differs considerably, between subjects and even between hemispheres, it is imperative that all masks are the same size when making comparisons using diffusion tractography. By taking the region of most significant activation for each subject, we aimed to maximize the likelihood of restricting the region to MT+/V5. MT+/V5 masks were transformed into diffusion space using FLIRT (Jenkinson et al., 2002) in order to perform tractography.

Analysis of diffusion data

Diffusion data were also analysed using tools in FSL. The following steps were taken using FDT (FMRIB's Diffusion Tool): (i) correction for eddy current and head motion, (ii) data averaged across the three repeats to increase signal to noise and (iii) diffusion tensors (Basser et al., 1994a) were fitted independently to each voxel in the averaged image and served to calculate the FA maps.

Masks for LGN and MT+/V5 were transformed into diffusion space using FLIRT to serve as seed and target, respectively, for the fibre tracking. Probabilistic tractography produces an estimate of the most likely location of a pathway from a seed point using Bayesian techniques, the details of which have been described elsewhere (Behrens et al., 2007). Briefly, a local model for fibre orientations within each voxel is inferred from the data. Crucially, this model allows for the representation of two fibre orientations per voxel, when more than one orientation is supported by the data. This allows modelling of crossing fibres in some voxels. The model also quantifies the fraction of the signal contributed by each fibre orientation. The probabilistic tractography then consists of drawing pathways by following sample orientations in each voxel along the trajectory. Five thousand sample tracts were generated from each seed voxel within the LGN masks, and only tracts entering the MT+/V5 masks were retained. Once tracts entered this target mask, the tract was terminated to prevent it from projecting on to other areas. In the case where contralateral pathways were followed, an additional waypoint mask was added in the corpus callosum. This ensured that any contralateral pathway crossed in that specific structure, rather than at a different level. All displayed tracts were thresholded at a value of 1, such that voxels where either one or no sample passed through were not included in the displayed tract. While this value is arbitrary, it is only for the purpose of displaying the data and has no effect on the quantitative analysis.

To produce a quantitative estimate of the size of the projection between LGN and MT+/V5, the total number of samples from the LGN voxels reaching the MT+/V5 mask was calculated. This enabled a comparison to be made between the strength of connection of the ipsilateral tract and the contralateral tract. Similar methodology was used to carry out probabilistic tractography between the LGN and an early visual mask, corresponding to a posterior region of V1 and probably V2.

To investigate cortico-cortical connections between MT+/V5, the probabilistic tractography was run twice, once with the left MT+/V5 as the seed and right MT+/V5 as the target, and once with the masks reversed. The mean of these two values was used as a measure of connectivity. To ensure that the tract passed through the splenium, a mask placed on the midline was used as a waypoint mask. The same technique was used to measure cortico-cortical connectivity between left and right V1. The connection strength referred to in this type of probabilistic tractography analysis is not straightforward to interpret, but must have some relationship to the probability of connection. A full description of the issues related to connection strength is included in the Discussion.

Results

Identification of target areas

Human area MT+/V5 was reliably activated in all controls and blindsight subject GY by the functional scan. It lies within the ascending limb of the middle temporal sulcus, as expected from the work of Dumoulin et al. (2000). Figure 1A and B shows the activation for subject 2 and GY respectively, while the overlap of the MT+/V5 masks for the six controls is shown in Fig. 1C.

Fig. 1

Definition of visual area masks in controls and blindsight subject GY. The functional activation to moving versus stationary dots designed to activate area MT+/V5 is shown for subject 2 in an axial plane (A) and for GY (B) (arrow indicates lesion in V1). The red scale bar indicates the z-statistic representing the significance of the functional activation in (A) and (B), thresholded at 3.5. The overlap of the restricted MT+/V5 masks and the LGN masks defined from the proton density images for the control subjects is shown in (C). Both regions are thresholded such that masks are from at least two subjects. The blue scale bar shows the number of subjects having overlapping masks. An example coronal slice from a control subject proton density image is shown in (D), and the left and right LGNs are indicated on a magnified section of this image by the white arrows (E). (F) shows a magnified image from the proton density scan of GY in which the LGN can also be seen bilaterally, indicated by the white arrows.

Although the thalamic nuclei are often difficult to distinguish anatomically in a T1-weighted structural image, high-resolution proton density images in this region can be used to reliably visualize the LGN (Devlin et al., 2006). Two of the authors independently identified the location of the LGN on the proton density images, and the final masks were defined as the intersection of these two masks. The mean volumes for the two LGNs in controls using this anatomical definition were 164 ± 16 mm3 and 160 ± 18 mm3 for the left and right sides, respectively. An example of a proton density image is shown in Fig. 1D, with a close-up of the region around the LGN in Fig. 1E. The white arrows indicate the positions of the LGN on the two sides. The overlap of the proton density masks from all subjects is shown in Fig. 1C, illustrating the excellent correspondence.

As the LGN masks were used as seed locations for the tractography, any quantitative comparison between tracts must not be biased by the size of the individual LGN mask. Hence, we restricted the LGN mask to a total volume of 140 mm3 (nine voxels). This is slightly larger than the means of 121 mm3 and 115 mm3 for left and right LGNs measured post mortem (Andrews et al., 1997), probably due in part to some shrinkage in the fixing process, but also to some error in the definition used here. It is, however, considerably lower than the mean activated volumes of 244 mm3 and 234 mm3 found by Kastner et al. (2004) using fMRI. The size of the mask used here reflected the smallest volume of LGN that was clearly identified on the proton density scans. In a few cases voxels at the edge of the mask were removed in order to meet the size constraint.

In subject GY it was straightforward to determine the location of the LGN in the right hemisphere, as it appeared very similar to that of the normal subjects. However, in the left hemisphere the definition was more challenging, presumably due to retrograde and anterograde atrophy from the V1 lesion (Cowey, 1974). It was only possible to unambiguously determine the LGN in two slices, compared with an average of four slices for the normal subjects. One of these slices is shown magnified in Fig. 1F, where the LGN can be seen bilaterally as a brighter region, indicated by the white arrows. As a result of the atrophy, the left LGN mask in GY contained only six voxels, compared to nine voxels for all other LGN masks.

Ipsilateral pathways between LGN and MT+/V5

To investigate the existence of a direct pathway between the LGN and MT+/V5, probabilistic tractography was used on diffusion-weighted images (Behrens et al., 2007). This technique computes a quantitative measure of the probability of connection between individual image voxels. Fibre-tracking was performed for each of the nine voxels in the LGN mask in diffusion space (six in the case of GY's left LGN).

In all subjects, a tract was present bilaterally. Figure 2 shows example data from subject 1 and the age-matched control (AMC). The MT+/V5 masks are shown in blue. The slices show the region of the tract with the highest probability of connection in yellow. Similar tracts were found in all subjects, and the sum of tracts from all control subjects is shown in a 3D representation in Fig. 2. To produce these data, tracts and masks were transformed into the Montreal Neurological Institute brain template (MNI152), then onto the brain of subject 2. The LGN and MT+/V5 masks were summed and thresholded at two subjects.

Fig. 2

Tracts between LGN and ipsilateral MT+/V5 in control subjects. The tracts are shown both in 3D space and on slices from the anatomical scans. 3D representations are shown for subject 1 and the AMC, and for the data summed across controls. The MT+/V5 masks are shown in blue, and LGN in green (LGN only visible in 3D data). In the group data both visual areas and tracts are thresholded to keep those regions that are present in at least 2/6 subjects. In the slices, the colour represents the total number of samples reaching each voxel, as shown by the scale bar. AMC = age-matched control.

Pathways in subject GY

The strong ipsilateral tract between LGN and MT+/V5 seen in all control hemispheres was also present bilaterally in GY, and was of a comparable strength. The path taken by the tract is also similar to that of controls, as seen in the middle row of Fig. 3. In addition, subject GY showed prominent bilateral tracts that crossed in the corpus callosum. Slices illustrating these tracts are shown in the bottom row of Fig. 3, and the 3D rendering of the pathways is shown on the right side of the top row. The entirety of the tracts can be seen in Supplementary Fig. 1.

Fig. 3

Blindsight subject GY has ipsilateral tracts from LGN to MT+/V5, very similar to those of the controls (middle row, and left 3D brain). In addition, and most unusually, he has contralateral tracts both between the left LGN (lesioned side) and the right MT+/V5, and between right LGN and left MT+/V5. Slices showing these contralateral tracts can be seen on the left and right panels of the bottom row, respectively. These crossing tracts can be visualized in the 3D brains in the top row.

In order to quantify the strength of these pathways, the total number of samples from each voxel in the LGN mask reaching either the ipsilateral or the contralateral MT+/V5 mask was computed. The values for all subjects are shown in Table 1. Across the 12 hemispheres of the controls, the average number of samples reaching the ipsilateral mask was 168.1 ± 258.3, while for the contralateral mask was 1.1 ± 1.6. In subject GY the number of samples reaching the ipsilateral MT+/V5 mask from the left (lesioned) side was 160.3 and 200.7 from the right. The contralateral connection from the left LGN to right MT+/V5 was 48.6 samples, just over 30% of the ipsilateral connection size. The contralateral connection from the right was 31.4 samples (16% of the ipsilateral connection size). Although smaller than the left side, this is well outside of the range for the control subjects.

View this table:
Table 1

Number of samples between LGN and MT+/V5 for each subject

SubjectIpsi-Contra-% Contra/Ipsi
Left LGN
136.21.33.6
298.40.10.1
3337.45.71.7
473.60.10.1
5943.81.00.1
AMC113.90.00.0
Mean267.21.40.9
GY160.348.630.3
Right LGN
1112.60.30.3
2132.90.80.6
366.01.82.7
413.60.21.5
549.61.02.0
AMC39.71.02.5
Mean69.10.91.6
GY200.731.415.6

One possible confound of the data obtained in these contralateral tracts, particularly the one from the left LGN to right MT+/V5, is that the presence of a lesion close to the corpus callosum in the left hemisphere could bias the tractography algorithm to enter the splenium. To investigate this further, we performed three additional analyses: (i) probabilistic tractography from the LGN to both ipsilateral and contralateral V1, (ii) reversing the seed and target masks to perform probabilistic tractography from MT+/V5 to the contralateral LGN and (iii) probabilistic tractography from the splenium to left MT+/V5 and splenium to right MT+/V5.

Measurement of tracts from LGN to early visual cortex

A target area was defined in the posterior tip of the occipital lobe on the lesioned side of GY's brain. This region consisted predominantly of spared V1, corresponding to his macula, but also probably included some of the surrounding V2. An equivalent mask was also made in the right hemisphere. These masks were transformed into the space of the controls, and were restricted to 7000 mm3 per hemisphere for all subjects. Probabilistic tractography was performed from LGN to both ipsilateral and contralateral early visual area masks in controls and GY. The mean number of samples reaching the ipsilateral mask in control subjects is 361.2 ± 543.1, and 3.1 ± 2.2 for the contralateral pathway, corresponding to 4.1 ± 9.0% of the ipsilateral tract strength. There is no significant difference between the pathway strength between LGN and MT+/V5 or LGN and V1 in the ipsilateral case due to the large intersubject variability. In GY, the ipsilateral pathways between LGN to V1 are smaller bilaterally than LGN to MT+/V5 (79.6 and 117.6 for left and right LGN, respectively). However, the contralateral pathways between LGN and V1 are considerably smaller (8.1 and 0.9 from left and right LGN, respectively, corresponding to 10% and 0.7% of the ipsilateral tract strength), clearly suggesting that the crossing tract is not necessarily a general property of the damaged cortical visual system. However, it is also clear that although the size of the contralateral tract from left LGN to right V1 is within the range of the controls, it is more than double the mean value. This suggests that the lesion may have some effect on the contralateral pathway from the left LGN. To investigate this possibility explicitly, we performed an additional control to measure the numbers of samples reaching the LGN when MT+/V5 was used as the seed mask, that is to say that we reversed the direction in which the tractography was performed. This analysis is described in the next section. Ipsilateral and contralateral tracts from LGN to MT+/V5 are shown in Fig. 4A for all subjects and GY in standard space. The corresponding tracts from LGN to V1 are shown in Fig. 4B. All GY's tracts are shown in detail superimposed on his anatomical scan in Supplementary Fig. 1 to illustrate their position relative to his lesion.

Fig. 4

Tracts between LGN and MT+/V5 (A) and LGN and V1 (B) for all control subjects (left side) and blindsight subject GY (right brain) shown in standard space. In both panels the ipsilateral tracts are present in all subjects, whereas only GY shows significant contralateral pathways between LGN and MT+/V5 (A). Contralateral pathways between LGN and V1 are either weak or absent in all subjects (B). The colour scale bars indicate the thresholding at one sample reaching the target.

Probabilistic tractography in the opposite direction

Since the tractography algorithm does not provide any information about directionality of pathways, it should be possible to perform probabilistic tractography with the seed and target mask swapped and obtain similar-sized tracts (although due to differences in geometry, not necessarily identical). In controls, there is no significant difference between the number of samples reaching the target when the probabilistic tractography is performed in opposite directions, i.e. from MT+/V5 to contralateral LGN (1.1 ± 1.6 versus 1.0 ± 0.7). If GY's lesion is affecting how the tract enters the corpus callosum, reversing the direction of the tractography should affect the relative size of the contralateral pathways from the two hemispheres. When the probabilistic tractography is performed in the opposite directions, the number of samples reaching the left LGN from the right MT+/V5 is 202 and the number reaching the right LGN from left MT+/V5 is 1282. Since each MT+/V5 consists of 140 voxels, rather than the nine in each LGN, it is necessary to scale the number of samples by the relative size of the masks to make a reasonable comparison with the LGN to contralateral MT+/V5 tract. When this scaling is applied, the number of samples reaching LGN from MT+/V5 is 13 (from right MT+/V5 to left LGN) and 82.5 (from left MT+/V5 to right LGN), while the same tracts from LGN to MT+/V5 have values of 48.6 and 31.4. Since the strength of the pathways is reversed, this is strong evidence that the lesion in the left hemisphere is increasing the likelihood of tracts entering the corpus callosum, and the contralateral pathway from left LGN to right MT+/V5 appears to be artificially strengthened. However, since there remains evidence of a tract even when it is run from right MT+/V5 to left LGN, this suggests such a tract may still be stronger than in controls.

Effect of the lesion on tracts exiting the splenium

The previous section indicates that tracts entering the corpus callosum from the left hemisphere appear artificially stronger due to the effect of GY's left hemisphere lesion in the nearby white matter. It is important, therefore, to ensure that the tracts exiting the corpus callosum in the left hemisphere are not also affected by this change in white matter, as this would affect the contralateral pathway running from the right LGN to left MT+/V5. To explicitly test this, probabilistic tractography was run from a mask in the mid-line of the splenium to either the left or right MT+/V5 mask separately. Since the tractography from the centre of the splenium to the right MT+/V5 will be unaffected by the lesion, a demonstration that the tract to the left MT+/V5 is of similar magnitude would be strong evidence that tracts exiting in the left hemisphere are not affected by the lesion. In fact the number of tracts reaching the right and left masks were 1794 and 1575, suggesting little or no effect of the lesion on tracts crossing the corpus callosum from right to left.

Cortico-cortical connections between left and right MT+/V5

The strength of the pathways from the centre of the splenium to MT+/V5 bilaterally in GY is suggestive of an increase in cortico-cortical connectivity between this area in the two hemispheres. To test whether such an increase is present in GY compared to control subjects, probabilistic tractography was performed from MT+/V5 in one hemisphere to MT+/V5 in the other, with a waypoint mask in the splenium to ensure that any tracts passed through that structure. This procedure was then repeated, starting in the other hemisphere, and the average number of samples was taken as a measure of connectivity. The results of this analysis are shown for all subjects in the upper row of Fig. 5, and the quantitative values shown in Table 2. This analysis in GY supports the earlier assertion that the lesion in GY increases the likelihood of pathways entering the corpus callosum, as there is a 16-fold increase in the strength of the connection when the tract is seeded from the left and runs to the right. Nonetheless, the size of the connection from the unlesioned right hemisphere to the left MT+/V5 is 10 times greater than the mean for the controls.

Fig. 5

Cortico-cortical pathways in control subjects and GY. The top row shows the connection through the splenium between MT+/V5 bilaterally, and the lower row shows the connection between V1 bilaterally. Note the difference in scale, indicating the relative strength of the V1 pathway.

View this table:
Table 2

Number of samples crossing the splenium between MT+/V5 in the two hemispheres and V1 in the two hemispheres

SubjectMTV1
148.5167.7
24.0633.6
314.52553.4
462.5998.3
51.0300.7
AMC0.0665.0
Mean21.8886.4
GY287.0153.4

Again, to ensure that this is not an anomaly of GY's visual system in general, a similar analysis was performed using the early visual masks as the targets. Since these pathways through the splenium project around the forceps major, a major white matter tract, we would predict that this cortico-cortical connection would be considerably stronger than the MT+/V5 connection. The tracts for V1 are shown in the lower row of Fig. 5, with the values in Table 2. The number of samples reaching the contralateral V1 has been normalized to make it comparable to MT, as the V1 mask contains more voxels, and therefore the total number of samples used for the probabilistic tractography was greater. The V1–V1 pathway in GY is around 55% of the MT+/V5-MT+/V5 pathway and shows equivalent size tracts when run from the two hemispheres. This suggests that the increased ability to cross the splenium is not a general feature of the system, and depends on where the tracts originate. All pathways for GY are shown in detail in Supplementary Fig. 3.

Discussion

Our investigation provides evidence of a change in, or strengthening of, anatomical connectivity, using diffusion tractography. Subject GY, from whom many of the ideas about the nature of blindsight are derived, is known to have visual responses in area MT+/V5, as well as in more medial occipito-parietal cortex (Barbur et al., 1993; Morland et al., 2004), when stimuli are presented within his scotoma. In the present investigation we demonstrated ipsilateral pathways between MT+/V5 and LGN in both his lesioned and intact hemispheres, of a similar strength to those in control subjects. In addition, the cortico-cortical pathway between MT+/V5 in the two hemispheres was considerably more prominent than the equivalent pathway in control subjects. Finally, there is evidence for contralateral tracts between LGN in one hemisphere and MT+/V5 in the other. The pathway from LGN in the right hemisphere to MT+/V5 in the lesioned hemisphere has a strength of 16% of the ipsilateral pathway. The strength of the pathway from LGN in the left (lesioned) hemisphere is more difficult to interpret as it is affected by the presence of the lesion near to where the tract enters the corpus callosum. However, the finding that this pathway is still evident when the probabilistic tractography is performed in the opposite direction indicates that this pathway is also present. It is important to mention that no controls showed transcallosal pathways of any significant strength. The equivalent contralateral connection was not present between LGN and V1, and there was no evidence of increased cortico-cortical connectivity between V1 in the two hemispheres in GY. Together, these observations suggest an increase in both the thalamo-cortical and cortico-cortical connectivity of MT+/V5 in GY.

Effect of the lesion on pathways

While the initial tractography following pathways from left LGN to right MT+/V5 uncovered a strong pathway crossing in the splenium, additional control experiments indicated that this is likely to be a result of decreased white matter in the region of the corpus callosum in the left hemisphere. This degeneration could bias the tractography algorithm to enter the corpus callosum particularly if there is retrograde degeneration in the fibres that would have projected to V1. A reduction in tracts running in one direction can selectively expose unaffected fibre bundles running in a different direction, making them easier for the algorithm to follow. However, following of the tract in the opposite direction, which should not be affected by the lesion, still produced a tract considerably stronger than those found in control subjects.

It has been reported previously that, in addition to the V1 damage, there is atrophy of the splenial fibres in GY (Ffytche et al., 2000), a finding that could predict a reduction in the strength of pathways crossing in the splenium. While we emphasize the apparent strengthening of the MT–MT connection, it can be seen from Table 2 and Fig. 5 that the V1–V1 connection is considerably weaker than that seen in control subjects. If the atrophy in the splenium is due to retrograde degeneration, it would not be surprising if the transcallosal V1 connection was weakened. Dougherty et al. (2005) show, using diffusion tractography, an organization of the fibres in the splenium according to the visual area from which they originate. Thus, it is entirely possible that the atrophy in the splenium could affect the connection between V1 in the two hemispheres, while the more rostral MT connection is unaffected.

Functional significance of tracts

The existence of a tract does not necessarily mean that it is of any functional significance, or that any function it mediates is necessarily relevant to understanding blindsight. However, there is considerable evidence for enhanced ipsilateral activation in MT+/V5 in subject GY, particularly in the left hemisphere. Morland et al. (2004) found that a moving stimulus in the left—normal—visual field activated a comparable number of voxels in the lesioned side (left) compared to the non-lesioned, contralateral side. It is accepted that area MST in primates has large receptive fields, including an ipsilateral representation (Tanaka and Saito, 1989), and in fact this has been used a means of separating out MT and MST from the human motion complex (Huk et al., 2002; Smith et al., 2006). So, while controls may be expected to show some ipsilateral activation in MT+/V5, corresponding to MST, the area of activation should be considerably smaller than the contralateral activation. This is indeed what Morland et al. found in controls. Increased ipsilateral representation in GY is consistent with both the contralateral pathway from right LGN to left MT+/V5 described here, and increased interhemispheric transfer of information. Additionally, Goebel et al. (2001) found evidence of ipsilateral activation in MT+/V5 in both hemispheres for subject GY, but not in the other hemianopic blindsight subject studied (Goebel et al. 2001). This latter result is consistent with tracts from both the left and right LGN to contralateral MT+/V5, as well as increased interhemispheric transfer of information.

The finding that there is increased connectivity between MT+/V5 in the two hemispheres is supported further by recent TMS data. Silvanto et al. (2007) found that it was not possible to elicit phosphenes in the blind field of GY by stimulating left MT+/V5 alone with transcranial magnetic stimulation (TMS). However, when TMS was applied to MT+/V5 bilaterally, phosphenes were elicited in his blind field. This suggests that there is interhemispheric transmission of information between the two areas, as shown in the present experiment.

It is reasonably straightforward to relate the changes in connectivity found here to the functional activation and TMS results described above. In particular the pathway from right LGN to left MT+/V5 is supported by bilateral activation to a stimulus in the left visual field (normal side), and increased interhemispheric connectivity is also consistent with this. However, relating the former pathway to GY's blindsight abilities is more challenging. It may be that this pathway allows the use of left MT+/V5 in addition to right MT+/V5 when processing stimuli in the normal hemifield, to maximize use of MT+/V5 in the lesioned hemisphere. In contrast, it is reasonable to hypothesize that information from the blind hemifield is available (blindsight) due to processing by MT+/V5 in either hemisphere. Activation in the right hemisphere may be weaker than that in the left, which could explain why only one previous study has found such activation using fMRI (Goebel et al., 2001). It is also possible that the right MT+/V5 could affect the ability of the left MT+/V5 to process information in the blind field possibly by influencing its physiological condition. The study by Silvanto et al. (2007) mentioned above supports this possibility, and the results of this study are only effectively explained by callosal connections between the two MT+/V5 complexes, a strengthening of which has been shown here. The result of Silvanto et al. suggests that there is a directional specificity as right MT+/V5 must be stimulated prior to the left side in order to perceive phosphenes. However, it is not necessarily a general phenomenon of blindsight.

Types of reorganization underlying functional and anatomical changes

The tractography data suggest two major differences between connections in blindsight subject GY's brain and those of controls. The first is a strengthening of cortico-cortical connections between MT+/V5 bilaterally, while the second is a contralateral pathway between right LGN and left MT+/V5. Since the lesion happened at the age of 8, most brain development has already occurred. However, consistent with the changes that we describe, particularly the cortico-cortical strengthening, the cross-sectional area of the corpus callosum continues to grow at least until early adulthood (Keshavan et al., 2002). Furthermore, recent studies have shown that in healthy adult subjects, repeated performance of a task can alter the fractional anisotropy in tracts (Bengtsson et al., 2005), suggesting that changes measured with diffusion can correlate with increased functionality.

The existence of a contralateral tract from the right LGN to left MT+/V5 cannot be explained as easily in terms of strengthening of an existing pathway, as such a pathway does not appear to exist in controls. However, a recent study by Leh et al. (2006) found that hemispherectomized patients exhibiting blindsight showed both ipsilateral and contralateral projections to multiple areas in the surviving hemisphere from the superior colliculus. In hemispherectomized patients that did not exhibit blindsight, and control subjects the contralateral projections were absent. This suggests that additional pathways can be uncovered even when damage has occurred quite late in development. However, it is not possible to determine with current methods whether the emergence of pathway is due to a strengthening of existing pathways or development of new pathways. While there is currently no independent evidence for the latter, it must be remembered that any form of neural regeneration and/or birth of new neurons in the adult human brain was rejected until relatively recently (Gould, 2007). Furthermore, since GY was aged 8 at the time of his cortical damage, the possibility for regeneration is likely to be greater than in the adult brain.

An alternative explanation that we cannot discount is that the contralateral pathway from right LGN to left MT+/V5 is driven predominantly by the strengthening of the cortico-cortical pathway between MT+/V5 bilaterally. It is possible, using this type of probabilistic tractography, that tracts passing by a region with a very high pathway density can get deviated by the high probability pathway. Using the current methods, it is not possible to determine whether this is the case, but we believe it is highly unlikely for two reasons. First, a similar pattern (LGN to contralateral V1) is not seen in control subjects who show a very strong V1–V1 connection, and secondly there is no evidence of a relationship between the weak contralateral LGN to MT+/V5 tracts in control subjects and the strength of the bilateral connection between MT in the two hemispheres.

Limitations of tractography and inter-subject variability

There are several limitations to the conclusions that can be drawn from the results of diffusion-based tractography. The first, and critical, ambiguity relates to the directionality of the pathway. In this case, our hypothesis is based on a feedforward connection from LGN to MT+/V5. However, there is no information in the data to distinguish this from the established and prominent corticofugal pathway providing feedback to the LGN (Sherman and Guillery, 2001). While there is a recent anatomical study that describes a feedforward connection (Sincich et al., 2004), since this pathway originates predominantly from the koniocellular layers of the LGN, it is relatively weak and constitutes only about 1% of LGN relay neurons. The pathway that we find is, on average, around half the strength of that to early visual cortex. It is likely that the tracts include both feedforward and feedback connections. The second major limitation reflects the inability to determine whether a given tract is functional. When comparing data from tractography to animal tracer studies, the large gulf in resolution must be considered. It is clear that only large fibre bundles will be evident using tractography as the imaging voxel may contain thousands of axons.

A further consideration is how to interpret the values of connectivity that are computed using tractography. The measure used in this study is the total number of samples generated in the seed mask (of the 5000 from each voxel) that reach the target mask. These values have means around 360 for LGN to V1, and 160 for LGN to MT+/V5, corresponding to 0.8% and 0.35% of the total number of samples. Earlier in the manuscript it was suggested that the values calculated from the probabilistic tractography are related to the probability of an anatomical connection. Clearly in normal subjects the probability of an anatomical connection between LGN and V1 is very high, whereas the percentage values given above are very low. The apparent discrepancy arises because the numbers reported here are actually related to the posterior probability of the dominant connection, that is, taking into account the diffusion data and the local diffusion model. This should not be interpreted as the probability that the tract exists, but rather the probability that the dominant tract passes through a particular point. Although the existence of an anatomical pathway and its strength will influence this probability, there is no straightforward relation between the two.

In addition, there is considerable variability between the connectivity in different subjects. Some of the variability may be due to real differences in the underlying anatomy, as there are large differences in cell counts throughout the visual pathway in normal subjects, even in the eye itself (Perry and Cowey, 1985). However, there is also variability in the tractography technique such as data quality and brain geometry. The inherent confounds of tractography mean that it is not possible to infer the size of actual anatomical connections from the tractography derived measure of connection strength, which is a problematic limitation.

One aspect of this article that highlights difficulties in the interpretation of diffusion data is the comparison between the number of samples reaching the early visual mask from ipsilateral LGN compared to those reaching MT+/V5 from the ipsilateral LGN. Although the mean number of samples reaching V1 is around double those reaching MT+/V5, this clearly does not reflect the anatomical connectivity of the two regions with the LGN measured post mortem. There are several possible reasons for this discrepancy. First, as mentioned earlier, the lack of directionality of the tractography technique means that the measure reflects cortical-fugal connections as well as feedforward connections. Second, the two target masks were defined with different methods. The MT+/V5 mask was defined functionally for each subject and restricted to the region of highest activation. In contrast, the early visual mask, determined anatomically from the spared cortex in the left hemisphere of GY, is likely to contain voxels from both V1 and V2, and is not necessarily the most active region of V1. Finally, the distance to V1 from LGN is greater than MT+/V5. As the probabilistic tractography accounts for the local uncertainty on the fibre orientations in each voxel, the strength of a connection will decrease with distance along the tract. This is one of the limitations of probabilistic tractography that makes tract thresholding and interpretation of connectivity strength difficult. While it is possible to correct for distance, this essentially involves assigning greater weight to lower probabilities, which could introduce additional complications.

The issues described above highlight the difficulties in choosing the best metric to illustrate connectivity measured with probabilistic tractography. The total number of samples reaching the target mask that is used here has a large variability both between different tracts within the same subject and between subjects in the same tract. This means that it is difficult to find an acceptable method of normalisation. Furthermore, assessing the evidence for the existence of a connection, given diffusion data alone is hard. Recent advances in probabilistic tractography have begun to address the issue of relating results to anatomy (Jbabdi et al., 2007), but this methodology is too preliminary to incorporate here.

Despite these limitations in tractography, the methodology is a valuable addition to the range of techniques that can be used for studying the intact and damaged brain. Here, we have presented evidence that blindsight subject GY has undergone substantial reorganization of his occipital lobe that may underlie his ability to discriminate moving, and other, stimuli in his blind field.

Supplementary material

Supplementary material is available at Brain online.

Acknowledgements

This work was supported by the Royal Society through a Dorothy Hodgkin Fellowship to HB and an MRC grant to AC. SJ is supported by the Dr Hadwen Trust for Humane Research. We would like to thank R. Cameron-Smail for pilot data analyses, T. E. Behrens for advice on analysis, P. Stoerig for helpful comments on a draft manuscript and J.T. Devlin for detailed appraisal of the methods and the findings.

Footnotes

  • Abbreviations:
    Abbreviations:
    AMC
    age-matched control
    DW-MRI
    diffusion-weighted MRI

References

View Abstract