Four healthy adult Macaca mulatta (M1, M2, M3, and M4) were used for these experiments. The data from these experiments have also been used in a previous study that examined reorganization in area V1 (Smirnakis et al., 2005). The experimental and surgical procedures were performed with care, in full compliance with the German Law for the Protection of Animals, the European Community guidelines for the care and use of laboratory animals (EUVS 86/609/EEC), and the recommendations of the Weatherall report for the use of non-human primates in research. The regional authorities (Regierungspräsidium Tübingen) approved our experimental protocol and the institutional representatives for animal protection supervised all procedures.

The monkeys were anesthetized during the fMRI experiments. Details on the anesthesia protocol have been given previously (Logothetis et al., 1999). Briefly, the animals were intubated after induction with fentanyl (31 μg/kg), thiopental (5 mg/kg) and succinylcholine chloride (3 mg/kg); anesthesia was maintained with isoflurane (M2 and M4) or remifentanyl (0.5–2 μg/kg/min, M1 and M3). Mivacurium chloride (5–7 mg/kg/h) was used after induction to ensure the suppression of eye movements. Heart rate and blood oxygen saturation were monitored continuously with a pulse-oxymeter. Body temperature was kept at 38–39°.

Homonymous retinal lesions were induced by using a photocoagulation laser (NIDEKGYC-2000; 532 nm) under general anesthesia (Figure 1A) as described in Smirnakis et al. (2005). Histological results confirmed that the photoreceptor and bipolar cell layers as well as most of the ganglion cell layer of the extrafoveal retina were destroyed (Figure 1B). Histological confirmation was obtained in two of the four animals. Animal fundi were photographed in each experimental session, and we confirmed that all lesions remained stable throughout the course of the experiments (see Smirnakis et al., 2005 for the details).

Stimuli were presented monocularly, at resolution of 640 × 480 pixels with a 60 Hz frame rate using an SVGA fiber-optic system (AVOTEC). Stimuli were centered on the fovea by using a modified fundus camera. Animals were fitted appropriate lenses to ensure the stimulus remained in focus. Standard expanding ring stimuli (outer radius expanded from either 0.3° or 0.9°–6.9° in steps of 0.6°, i.e., 12 or 11 annuli, frame interval = 6 s) and rotating wedge (90° wedges rotated in steps of 30°, i.e., 12 annuli, frame interval = 6 s) were presented to the subjects monocularly, always in the same eye for each animal. The full retinotopic maps were obtained pre-lesion but after lesioning only the eccentricity maps were followed to maximize repetitions and the signal to noise ratio of our measurements. The first measurement was obtained on the day of the lesion. The earliest subsequent measurement was obtained 14 days later to comply with the accepted animal protocol and allow for retinal recovery (Smirnakis et al., 2005).

One important control condition involved presenting the same expanding ring stimuli to the subjects and at the same time occluding a region in the normal half of the visual field. This region was designed to mirror the approximate location of the retinal lesion. We refer to this as the “artificial scotoma” (AS) condition. For all four subjects, the ASs were centered at (3.7°, 0) or (–3.7°, 0), on the opposite side of the scotoma resulting from the retinal lesion, and had a diameter of 3.7°.

FMRI experiments were performed on a 4.7T vertical scanner (Bruker Biospec, Bruker Biospin GmbH, Ettlingen, Germany). Multi-slice fMRI was performed by the use of 8 segmented gradient-recalled echo-planar imaging (EPI). The acquisition parameters were TE = 20 ms, TR = 750 or 805 ms, flip angle = 40°. Either 15 or 17 axial slices were collected at 1 × 1 mm² in-plane resolution and 2-mm thickness. A full-brain anatomical scan was acquired before lesioning at 0.5 × 0.5 × 0.5 mm³ resolution for co-registration with the EPI images by using an MDEFT sequence (Logothetis et al., 1999; Keliris et al., 2007).

FMRI data were reconstructed and imported into a MATLAB based toolbox (mrVista)¹ (Amano et al., 2009; Levin et al., 2010). The gray-white matter boundary was manually segmented using itkGray from the high resolution 3D-MDEFT anatomical images, and 3D cortical surface and flat mesh models were created and realigned with the functional data by using mrMesh/mrVista (Wandell et al., 2000).

In a typical experiment 5–10 repeats of the expanding ring and rotating wedge stimulation paradigm were performed and the average BOLD signal time course was generated. To obtain the retinotopic maps we have used the traveling wave method (Engel et al., 1997). The strength of the visual modulation was assessed using the measure of coherence, computed with the expanding ring stimuli. Coherence is defined as the BOLD signal spectral amplitude at the stimulus presentation frequency (12 or 11 cycles per scan in our experiments) divided by the average square root of the power over a range of nearby frequencies (Smirnakis et al., 2005). Normalized coherence was defined by dividing the coherence with the average coherence in a corresponding ROI in the contralateral (intact) hemisphere on the same day; this can account for variability across scanning days. The retinotopic maps were then fitted into a template of expected eccentricity and angle maps (atlas fitting) (Dougherty et al., 2003). During this procedure, the four sides of the visual field map, from fovea to periphery and from upper to lower vertical meridian were defined manually by simultaneously looking at angle and eccentricity gradients and were found to obey expected anatomical landmarks. For example, the dorsal V1/V2 border was confirmed to lay ∼2 mm from the lip of the lunate, the ventral V1/V2 border along the inferior occipital sulcus (Brewer et al., 2002), and the dorsal V2/V3 boundary at the bottom of the lunate sulcus. The calcarine sulcus was at ∼6.5° eccentricity as reported in Gattass et al. (2005). The fitting algorithm deforms these templates to match the eccentricity and angle data, allowing local deformations but no tears or folds in the atlas. The atlas with the least error compared with the data was generated, following (Dougherty et al., 2003). Visual inspection confirmed no major errors.

To find the V2/V3 LPZ we performed the following steps: (1) we selected V1 voxels with coherence below threshold. The value of 0.28 was chosen as a threshold based on the expected value of coherence in the absence of visual modulation that depends on the bandwidth (Δf = 12 cycles; thr = 1/(Δf + 1)^0.5 = 0.28). This is a very conservative threshold for selecting the V1 LPZ as it equals the expected noise level ignoring its variability and thus selects the core of the V1 LPZ. Then, the voxels below threshold were merged to define the V1 LPZ (Figures 2C,D). Note that the area V1 LPZ has been shown to remain unchanged over time in this data set (Smirnakis et al., 2005), (2) we fitted the pre-lesion eccentricity and polar angle maps into an atlas per monkey using the process described in Dougherty et al. (2003), and (3) based on the pre-lesion retinotopic atlas-fit, the voxels in V2/V3 that corresponded retinotopically to the voxels belonging to the V1 LPZ, to within 0.05° of eccentricity and polar angle (on the atlas) were selected. These selected voxels were highly clustered and continuous inside the dorsal and ventral V2/V3. They were then merged to define the V2/V3 LPZ (Figures 2C,D). Once the V2/V3 LPZ was identified on the anatomical template, all quantitative analysis was done on the original data derived from voxels located inside the anatomical region corresponding to the LPZs. For each monkey, we followed the same procedure in order to define the artificial scotoma projection zones (ASPZ) in areas V1 and V2/V3 on the non-deafferented hemisphere (ipsilateral to the scotoma). Note that the monkeys were always scanned in the same position with the help of implanted fMRI-compatible headposts, and that the LPZ was defined on the high-resolution anatomical scan that serves as a template for aligning the functional data. These procedures assured we could follow the activity of the same anatomical region over time.

Before lesioning the retina, retinotopic maps were measured using standard expanding rings and rotating wedge stimuli (see section “Materials and Methods”; Engel et al., 1997; Baseler et al., 1999). As shown in Figures 2A,B, all subjects showed normal retinotopic organization. The V1/V2 and V2/V3 borders were identified by the location of the vertical and horizontal meridians, respectively (Figure 2A). The border location was confirmed anatomically, as the macaque visual area boundaries have stereotypical anatomical locations (see section “Materials and Methods”).

Following a homonymous retinal lesion, parts of the visual cortex become deafferented (Heinen and Skavenski, 1991; Yinon et al., 1993; Smirnakis et al., 2005; Giannikopoulos, 2006). This results in a series of LPZs, one per retinotopic visual area. Identifying the LPZ of V1 is relatively straightforward given that, according to its role as primary visual cortex, V1 is receiving input from the lateral geniculate nucleus, it has relatively small receptive field sizes. In addition, a major part of central V1 which encompasses the LPZ in our study is exposed on the relatively flat operculum. On the contrary, identifying the LPZs in higher visual areas such as V2/V3 is more challenging given their anatomical location, receptive field sizes, and fewer numbers of voxels. Our goal was to define the LPZ in areas V2/V3 and follow how the strength of its visual modulation changed over time. Note that given that these areas share the horizontal meridian their LPZs are expected to be joined and dissected by the horizontal meridians and moreover to be positioned along the V2-V3 border both dorsally and ventrally. In order to determine a stable and appropriate LPZ for our V2/V3 analysis, we used the LPZ retinotopic coordinates carefully identified in V1 to guide our identification of the V2/V3 LPZ, instead of attempting to define the V2/V3 LPZ based on activity levels in the region which are expected to be variable over time. Importantly, because the areas V2/V3 have larger receptive fields in comparison to V1, this V2/V3 LPZ region is expected to be larger than the absolute V2/V3 LPZ. To this end, to be able to estimate the extent of this activity we have also used the stimuli that included an artificial scotoma in the healthy hemisphere and evaluated the timecourse of activity within the V2/V3 artificial scotoma projection zone (ASPZ) defined in an identical way as the V2/V3 LPZ.

The retinal lesion crosses the horizontal meridian and so the V2/V3 LPZ is split into two parts, one across the ventral and the other across the dorsal V2/V3 border (Figure 2C). From the coherence maps (Figure 2C) it is evident, as we expected, that even on the day of the lesion (D1), the V2/V3 LPZs are, on average, significantly visually modulated. Therefore the V2/V3 LPZ appears to receive visually modulated input outside the retinotopically corresponding V1 LPZ, which shows no significant visual modulation over its entire extent (Smirnakis et al., 2005). The average fraction of voxels inside the V2/V3 LPZ that were visually modulated (> 2 standard deviations above noise level; note that this is a conservative definition of modulated voxels inside an area that was defined based on V1 LPZ selected with a threshold at noise level) on the day of the lesion (D1) was 50.4 ± 12% (standard deviation across subjects). This was consistent with observations made in the artificial scotoma (AS) control condition: the average fraction of voxels that were visually modulated inside the ASPZ was similar at 53.5 ± 10% (standard deviation, Figure 3).

Importantly, in addition to the voxels that were visually modulated immediately after the lesion, a fraction of voxels that were not significantly modulated on the day of the lesion (D1) recovered the ability to be visually modulated by day 14 post lesion (D14+, Figure 2D and Supplementary Figure 1). Figure 4A, top row, plots the percent modulation of the BOLD signal as a function of the stimulus cycle of voxels taken from monkey M1’s V2v/V3v LPZ with low coherence (< 2 standard deviations above noise level) on the day of lesion. Note how the strength of the visual modulation initially drops following the lesion (D1) but then recovers over time from day 1 (D1) to day 14 (D14). The same result is also demonstrated in Figure 4A, bottom row, which shows the signal amplitude as a function of temporal frequency for the same voxels and the modulations of spectral peak at the stimulus frequency (12 cycles/scan).

To better understand the reorganization of responses across days, we compared the distribution of coherence values measured across all the LPZ voxels pre-lesion (PRE) with the distribution on the day of the lesion (D1) and 14 days post-lesion (D14+) (Figure 4B). Because the overall level of the BOLD signal can fluctuate on different measuring sessions, the coherence of each voxel inside the LPZ was expressed as a fraction of the average coherence computed across all voxels within an iso-angular ROI (see Figures 2C,D) from a non-deafferented region of V2 in the same hemisphere (normalized coherence). Note that the distribution of normalized coherences derived from all voxels inside the V2/V3 LPZ (Figure 4B) shifts initially toward lower coherence values on the day of the lesion (D1) but returns close to pre-lesion levels 14 days later (D14+). One way ANOVA tests were performed comparing average normalized coherence over time (PRE, D1 and D14+) across all the V2v/V3v and V2d/V3d LPZ voxels in all 4 monkeys. There were significant differences of coherence over time in both the dorsal and ventral LPZ (F = 76.48, p < 10^–8 for V2v/V3v LPZ and F = 11.45, p < 10^–5 for V2d/V3d LPZ). Statistics were performed using the functional voxels (a total of 58, 45 voxels for the V2v/V3v and V2d/V3d LPZ, respectively, across all monkeys). Tukey post hoc pairwise comparisons showed significant differences between the mean level of coherence pre-lesion (PRE) and day 1 (D1), D1 and day 14 (D14+) for both LPZs, as well as the differences between PRE and D14+ for V2d/V3d LPZ (p < 0.05; Figure 4C). In fact, the number of voxels with low coherence inside the V2/V3 LPZ also changes following the lesion. The average fraction of voxels inside the V2/V3 LPZ that had low coherence (< 2 standard deviations above noise level) across subjects was 14.1 ± 8% (standard deviation) pre-lesion, increased to 49.6 ± 12% on D1, and dropped back to 14.8 ± 9% on D14+, close to pre-lesion levels.

To study how the strength of visual modulation inside the LPZ of area V2/V3 evolves over time, we focused on voxels in the center of the V2/V3 LPZ whose absolute coherence was low on day 1 (D1) following the lesion (< 2 standard deviations above noise level). For the sake of comparison, the coherence of each voxel in the LPZ was normalized by the average coherence across voxels of the non-deafferented control V2 ROI as described before (see Figure 2). As shown in Figure 4D, the mean coherence of these voxels was high pre-lesion, dropped on the lesion day, increased markedly as early as 2 weeks after the lesion and then remained approximately at the same level over time. In contrast, the same analysis in control voxels outside the V2/V3 LPZ did not show any significant changes in coherence levels (Figure 4D). Overall, for all four subjects, we found that the strength of the visual modulation inside the V2/V3 LPZ increased over time following the lesion, arguing for potential reorganization in area V2/V3. This contrasts with the case of the V1 LPZ where the signal dropped to noise levels following the lesion and did not change significantly over time (Smirnakis et al., 2005).

The next step was to look for the source of the input leading to the increase in visual modulation inside the V2/V3 LPZ over time. Although part of the sensory input of the voxels inside the LPZ has been cut off because of the retinal lesion, neurons inside the LPZ may still receive input from visual field locations that are outside the border of the scotoma (Barton and Brewer, 2015). To test this, we compared the range of eccentricities of the voxels inside the V2/V3 LPZ before and 14 days after the lesion. We focused on the voxels whose strength of visual modulation recovered as a function of time following the lesion. Such voxels had low coherence (< 2 standard deviations above noise) on D1, but were modulated above noise level both before and 14 days following the lesion. For each subject, we performed the eccentricity analysis in the dominant (largest) V2/V3 LPZ, dorsal or ventral. This was compared to controls (Figure 5A) from an approximately iso-angular non-deafferented area V2 ROI in each subject (see Figure 2). Eccentricity measurements in control ROIs (Figure 5A) demonstrate the reliability of this analysis. As expected, all control ROIs showed strong correlation between eccentricities measured pre-lesion and on day 14, with slopes near one (b = 1.54, 0.98, 0.71, 0.94; R² = 0.91, 0.93, 0.79, 0.86; p < 10^–7 for M1, M2, M3 and M4, respectively, illustrated in dashed lines). Figure 5B presents the eccentricity scatter plots of voxels inside the dominant V2/V3 LPZ before and after the lesion. Voxel eccentricities pre-lesion were significantly correlated with eccentricities measured on day 14 in 3 out of 4 subjects, but slopes were significantly smaller than one (b = 0.14, 0.28, 0.52; R² = 0.17, 0.22, 0.67; p < 0.01 for M1, M3, and M4, respectively, and b = –0.05, R² = 0.0084, p = 0.56 for M2; illustrated in dashed lines). Overall, the eccentricities inside the LPZ span a wider range following the lesion compared to pre-lesion for each subject. Voxels with small eccentricity pre-lesion tend to have smaller eccentricity after the lesion, while voxels with large eccentricity pre-lesion tend to have larger eccentricity after the lesion. This suggests that neurons inside the LPZ may regain their activity by receiving input either from the foveal or the peripheral border of the LPZ, depending on which one is closer.