fMRI and gene therapy in adults with CNGB3 mutation

Achromatopsia is an inherited retinal disease that affects 1 in 30,000 – 50,000 individuals and is characterised by an absence of functioning cone photoreceptors from birth. This results in severely reduced visual acuity, no colour vision, marked sensitivity to light and involuntary oscillations of the eyes (nystagmus). In most cases, a single gene mutation prevents normal development of cone photoreceptors, with mutations in the CNGB3 or CNGA3 gene being responsible for ~80 % of all patients with achromatopsia. There are a growing number of studies investigating recovery of cone function after targeted gene therapy. These studies have provided some promise for patients with the CNGA3 mutation, but thus far have found limited or no recovery for patients with the CNGB3 mutation. Here, we developed colour-calibrated visual stimuli designed to isolate cone photoreceptor responses. We combined these with adapted fMRI techniques and pRF mapping to identify if cortical responses to cone-driven signals could be detected in 9 adult patients with the CNGB3 mutation after receiving gene therapy. We did not detect any change in brain activity after gene therapy when the 9 patients were analysed as a group. However, on an individual basis, one patient self-reported a change in colour perception, corroborated by improved performance on a psychophysical task designed to selectively identify cone function. This suggests a level of cone sensitivity that was lacking pre-treatment, further supported by a subtle but reliable change in cortical activity within their primary visual cortex.


Introduction
Achromatopsia is a rare inherited eye disease that affects 1 in 30,000-50,000 individuals, characterised by an absence of functioning cone photoreceptors (Michaelides, Hunt, and Moore, 2004;Hirji et al., 2018).This results in severely reduced visual acuity, no colour vision, marked sensitivity to light and involuntary oscillations of the eyes (nystagmus) (Hirji et al., 2020;Remmer et al., 2015;Kohl et al., 1993).In most cases, a single gene mutation prevents normal development of cone photoreceptors in early infancy, with mutations in the CNGB3 and CNGA3 gene being responsible for ~50 % and ~30 % of all people with achromatopsia (Michalakis et al., 2022;Kohl et al., 2005).Rod photoreceptors are present and purportedly functioning, hence what appears as a vibrant multi-coloured scene to a normal-sighted individual appears in shades of grey and is poorly defined to an individual with achromatopsia.
At present, there are no curative treatment options.However, great progress is being made in molecular genetic testing and the development of targeted gene-therapies, with promising results in animal models (Michalakis et al., 2022;Sundaram et al., 2014;Hassall, Barnard, and MacLaren, 2017;Carvalho et al., 2011;Komaromy et al., 2010) giving hope for similar interventions in humans.Despite a marked reduction in cone cell density, most patients with achromatopsia have evidence for residual cone cells that could be targeted for rescue (Langlo et al., 2016).Recent gene therapy trials in humans have largely focussed on the CNGB3 and CNGA3 variants (Michaelides et al., 2023;McKyton et al., 2021McKyton et al., , 2023;;Farahbakhsh et al., 2022), aiming to transfer a healthy copy of the disease-causing gene into affected cone photoreceptors.
Cone photoreceptor recovery is generally measured using retinal imaging techniques (e.g.adaptive optics and optical coherence tomography) and visual colour psychophysics (Georgiou et al., 2019;Aboshiha et al., 2014;Patterson et al., 2021;Michaelides et al., 2023).However, for there to be improvement in functional vision, any restoration of cone photoreceptor structure in the retina must be accompanied by measurable responses within the visual cortex.Currently, little is understood about the post-retinal neural pathway in achromatopsia due to the technical challenges involved in measuring visual responses beyond the retina in vivo in this patient group.
Functional magnetic resonance imaging (fMRI) is generally the method of choice, but standard protocols need to be adapted to accommodate individuals with poor vision, nystagmus, and photophobia.Using, carefully calibrated visual stimuli together with highly specialised population receptive field mapping (pRF) techniques (Dumoulin and Wandell, 2008), it is possible to map cortical visual responses in patients with achromatopsia (Farahbakhsh et al., 2022).
Using silent substitution techniques (Estevez and Spekreijse, 1982), we developed visual stimuli that could independently drive rod or cone photoreceptors.These were incorporated into a pRF mapping protocol to measure retinotopic responses in primary visual cortex to these selective stimuli.The cone-selective stimulus was also presented within a simple psychophysical task to evaluate cone sensitivity before and after gene therapy.The expectation being that any change in colour perception would be accompanied by a measurable change in cortical activity to the cone-selective stimulus.
Using these stimuli in a pRF mapping protocol we were able to compare the spatial characteristics of rod-and cone-driven responses in individuals with achromatopsia and compare these to normal-sighted individuals, before and after gene therapy.We looked for differences in pRF parameters (e.g.size) as well as the spatial representation of the visual field on the occipital surface.Particular interest was placed on any change in cone-driven activity after treatment which might indicate successful recovery of cone function.Analysis was performed at both the group and individual level.
Using similar techniques, we have previously reported a change in cone-driven cortical responses after gene therapy in two children with the CNGA3 mutation (Farahbakhsh et al., 2022).However, we found no effect of treatment on children with the CNGB3 mutation.Here, we examine the effects of gene therapy in a group of adult patients with the CNGB3 mutation selected for a gene therapy trial ('Gene therapy for achromatopsia trial' NCT03001310).

Participants
Nine patients diagnosed with achromatopsia due to the CNGB3 mutation were recruited from Moorfields Eye Hospital NHS Foundation Trust, already enrolled in the phase 1/2 clinical trial 'Gene therapy for achromatopsia' (NCT03001310).In addition, we recruited 10 healthy age-matched controls (patients average age 24.2 yrs, controls average age 24.1 yrs, P=0.96) with normal vision and no previous neurological condition.They all gave written informed consent in accordance with local protocols to ensure adherence to MR safety.All procedures were approved by University College London Research Ethics Committee (11/ LO/1229).See Table 1 for participant details.

Stimuli adapted for individuals with achromatopsia
Visual stimuli used for retinotopic mapping are typically broad spectrum and presented at an unspecified luminance within the high photopic range (~100 cd/m 2 ), hence activating all types of cone photoreceptor but evoking minimal (if any) rod response.This approach is thought to achieve maximal retinal response and evoke the most reliable retinotopic maps.However, such stimuli would be uncomfortably bright to a rod achromat and well outside their peak sensitivity range.Therefore, to achieve a reliable retinotopic map in patients with achromatopsia we designed stimuli that independently activated the rod and cone photoreceptors and presented them within a tolerable luminance range.
We generated a set of colour pairs that when flicker-reversed selectively activate rod photoreceptors, whilst leaving cone photoreceptors silent (cone-equated) and a second set of colour pairs that activate cone photoreceptors, whilst leaving rod responses silent (rod-equated) (Fig. 1).These colour pairs were defined in terms of their red (R), green (G) and blue (B) phosphor contributions.Using the spectral output of the presentation screen (measured using a PR670 spectroradiometer, www.photoresearch.com) and standard observer sensitivity functions for rods and cones (http://www.cvrl.org),we created a transformation matrix to convert RGB colour channel values into L-cone (L), M-cone (M) and rod (R) photoreceptor activation levels.We used this matrix to calculate the RGB triplets required to independently increase and decrease rod and cone activity by a set proportion from a consistent baseline (mid-grey).
To maximise rod photoreceptor function, we used our transformation matrix to generate colour pairs that when flicker reversed kept the L and M cone photoreceptors silent whilst evoking robust rod activity.We refer to this as a 'Cone-Equated' stimulus and present it at scotopic light levels (<0.2 cd/m 2 ).Similarly, we generated a different colour pair that evoked a robust response in L and M cone photoreceptors whilst leaving the rods silent and presented this at low photopic light levels (0.8-1.2 cd/m 2 ).We refer to this as 'Rod-Equated'.We also created a 3rd colour pair which did not control for photoreceptor type -'Non-Equated' -and presented this in the mesopic range (0.3-0.7 cd/ m 2 ).
These selective stimuli were then presented within a standard ringwedge retinotopic mapping stimulus that rotated clockwise or anticlockwise and expanded or contracted (Fig. 1).For each stimulus type, one component of the colour pair was used for the background and the other for the moving checkerboard.The full-aperture checkerboard stimulus covered a circular region that subtended 6.5 degrees of visual angle around fixation.All stimuli were generated in MATLAB R2015b (Mathworks) and displayed using Psychtoolbox-3 (Brainard, 1997) on to a BOLDscreen24 LCD (www.crsltd.com)running at 1920×1200 resolution (warmed for a minimum of 1 hour prior to scanning).

Retinotopic Mapping Procedure & pre-scanning psychophysics
The BOLDscreen24 was positioned at the rear of the scanner bore and viewed via a mirror system mounted on the head coil.Different combinations of removable neutral density filter (NDF) sheets were secured to the surface of the BOLDscreen24 to achieve scotopic (luminance range 0.07 -0.2 cd/m 2 ), mesopic (0.3 -0.7 cd/m 2 ) and photopic viewing conditions (luminance range 0.8 -1.2 cd/m 2 ).
To ensure the rod photoreceptors were 'quiet', and hence maximally sensitive, prior to starting the retinotopic mapping procedure, each participant underwent 30 minutes of dark adaptation using a black-out eye mask (http://www.mindfold.com)and all light sources within the scanner room were covered.Participants lay within the scanner bore during the dark-adaptation period and a high-resolution structural scan was acquired (see details below).After 30 minutes the eye mask was removed, whilst keeping all room lights off and the patient's eyes closed.
Before starting the first retinotopic mapping scan, we asked all participants to perform a simple psychophysical task.We presented the cone-activating (rod-equated) stimulus, under the same conditions used in the main fMRI experiment, and asked participants to report whether the wedge rotated clockwise or anti-clockwise and whether the ring was expanding or contracting.The contrast of the colour pair (background/ foreground) was increased or decreased using 21 contrast levels in a simple 1-up/1-down staircase procedure to identify the 'switch point' from visible to invisible (Fig. 2).Both an incorrect response and a report that no checkerboard was visible triggered a step up in cone contrast, while a correct response triggered a step down.Each staircase continued until at least eight reversals had occurred.All colour pairs were visible to a normal sighted individual, but only the very high contrast pairs (if any) were visible to the patients with achromatopsia.Contrast level 15 was chosen a-priori for the rod-equated stimulus used in the fMRI scans.
After the psychophysics task, participants were asked to close their eyes and all lights were turned off for a further 10 minutes to ensure dark adaptation was maintained prior to the first scan.For all participants, the rod-activating stimulus was presented first (2 scan runs), then the cone-activating stimulus (2 scan runs) and then the mixed (non-selective) stimulus for a further 2 scan runs.Between each scan run the participants were asked to close their eyes and rest for 5 minutes.During this period, we changed the neutral density filter (NDF) on the screen, to ensure appropriate luminance levels for each stimulus type.
Throughout all scans participants were asked to maintain fixation on a central black fixation point (0.2 • ) and to press a button each time the fixation target briefly flashed white.Meanwhile, they passively viewed the flickering ring/wedge stimulus with both eyes.To further facilitate central fixation a low-contrast radial pattern was superimposed over the entire stimulus, comprising of radial lines extending from just outside the fixation dot (Fig. 1).Eye-movements were monitored throughout using an Eyelink 1000 MRI compatible eye tracker (http://www.sr-research.com).To limit the time between removing the eye mask and starting the first functional scan, we performed the eye-tracking calibration prior to starting the initial dark-adaptation.All tasks were done with both eyes open, as the amplitude of nystagmus often increases with Neutral density filters were used to ensure each stimulus was presented within the correct luminance range (Scotopic <0.2 cd/m 2 , photopic 0.8-1.2cd/m 2 , mesopic 0.3-0.7 cd/m 2 respectively).The rotating ring/wedge checkboard covered a circular region subtending 6.5 degrees of visual angle around fixation.

Fig. 2. Cone-driven psychophysical responses. A.
For each patient, our rod-equated retinotopic mapping stimulus was presented with increasing/decreasing conecontrast (N=1 highest contrast, N=21 lowest contrast).Patients were asked to report the direction of movement of the flickering checkerboard wedge/ring stimulus (expanding/contracting or clockwise/anti-clockwise). B. Prior to treatment, most of our patients did not see any of the flickering checkerboards, or they only saw the highest cone-contrast levels.This did not change after treatment except for 1 individual (P07) who significantly improved on this task and reached N=11 at the 1st post treatment check and N=13 at the 2nd post treatment check, suggesting a significant improvement in their cone sensitivity post treatment.Cone-contrast N=15 was chosen for the fMRI experiment, prior to any knowledge of the improvement in cone-sensitivity expected.monocular viewing.

Data acquisition and scan timing
All MRI data were acquired using a Siemens 3 T TIM-Trio scanner using a 32-channel head coil.For the functional scans a high-resolution EPI sequence (2.3 mm isotropic, interleaved slice order, 96×96 matrix, slice acquisition time 85 ms, TE 37 ms) was used to acquire 30 near axial slices, positioned parallel to the calcarine sulcus and to optimise coverage of the occipital lobe.The front of the head coil was removed for these scans to maximise the field of view, leaving 20 receiving channels.
To assess the homogeneity of the magnetic field with the front of the head coil removed we acquired B0 fieldmaps after the functional scan runs (double-echo FLASH sequence, short TE 10 ms, long TE 12.46 ms, 3x3x2mm resolution, 1 mm gap).We also acquired two T1-weighted structural images (during the initial dark-adaptation period); one with the front of the coil removed (MPRAGE, 1 mm isotropic voxels, 176 sagittal slices, 256×215 matrix, TE 2.97 ms, TR 1900 ms), which was used as an intermediate step in coregistering functional data to a second high-resolution anatomical image acquired with the front of the coil in place (3D MDEFT, 1 mm isotropic voxels, 176 sagittal slices, 256×240 matrix, TE 2.48 ms, TR 7.92 ms, TI 910 ms).The latter was used for segmentation and cortical reconstruction.
For each pRF mapping scan run we acquired 118 volumes using a TR of 2.55 s.This included 4 'dummy' scans acquired at the beginning of each run to allow the brain to reach steady state magnetisation.The central black fixation dot was presented during the dummy scans and central fixation maintained prior to the first mapping stimulus appearing.Each scan run contained 4 full wedge rotations, 6 contractions/ expansions and a blank period at the end.A full wedge rotation took 61.2 seconds (24 volumes), whilst a ring expansion/contraction took 40.8 seconds (16 volumes) and the blank period lasted 45.9 seconds (18 volumes).For the first run the wedge rotated clockwise and the ring expanded and for the second run the wedge rotated anti-clockwise, and the ring contracted.

MRI data pre-processing
All functional data were pre-processed using SPM12 (http://www.fil.ion.ucl.ac.uk/spm).To compensate for the effects of removing the front of the head coil, all functional images were intensity bias-corrected using in-house software.The dummy volumes were then discarded and the remaining images from the mapping scans were realigned, unwarped (using the B0 field maps to correct any image distortion) and co-registered to the individual's high resolution T1 structural image (using the additional MPRAGE as an intermediate step).
Freesurfer software (http://surfer.nmr.mgh.harvard.edu,version 5.0.0) was used to create separate 3D surface meshes, for the right and left cortical hemispheres; one for the boundary between gray and white matter, and one for the outer (pial) boundary of the white matter.For each vertex in the surface mesh, we determined the voxel in the functional image that lay at the midpoint between the vertex on the greywhite matter boundary, and the same vertex on the pial surface boundary.The cortical surfaces were then inflated, and functional data projected onto the smoothed gray/white matter surface.
All further analyses were performed using an in-house MATLAB toolbox 'SamSrf v7' (https://osf.io/2rgsm)for pRF analysis and for projecting data onto the cortical surface.For speed, analysis was restricted to just the occipital cortex, defined separately for each individual.

pRF modelling
For the pRF analysis we used a forward modelling approach (Dumoulin and Wandell, 2008;Morgan and Schwarzkopf, 2019;Urale et al., 2022), in which the pRF was modelled as a two-dimensional Gaussian in visual space.A linear overlap between the pRF model and a binary mask of the stimulus (aperture file) across time was used to predict the response of the neuronal population at each vertex.This predicted response time series is further convolved with the canonical hemodynamic response.The best fitting pRF model is determined by varying its visual field position (x, y) and size (σ) parameters, and calculating the Pearson correlation between the predicted and observed time series.We ran a first pass coarse fit to find the parameters with the highest correlation between observed and predicted time series.These parameters were then used to seed a subsequent fine fit to the unsmoothed data at each vertex.Because correlation is scale-invariant, we also calculated a linear regression between this fit and the observed time series, to determine the amplitude (β 1 ) and baseline level (β 0 ) of the response.Further details In addition, we ran a second model fit using a stimulus aperture file that incorporated a simulation of the central rod-free zone known to exist in normal-sighted individuals (Curcio et al., 1990).Previous research has shown that pRF estimates derived via forward-modelling are susceptible to potential biases when a scotoma is not explicitly modelled in the stimulus apertures (Binda et al., 2013;Urale et al., 2022).This could produce spurious estimates of pRFs inside the rod-free zone.Therefore, we applied a central 1.25-degree circular scotoma to the aperture file used in the model fit for the 'Cone-Equated' (rod activating) stimulus for all individuals (patients and controls).

Regions of Interest
Visual regions were delineated manually by displaying pseudocolour coded maps of polar angle and eccentricity calculated from the pRF analysis.An average map was created for each individual including data from all 3 stimulus types to ensure all active voxels were included.This included data from all 3 scan sessions for the patients (i.e.pre and post treatment).We applied a surface based smoothing kernel of 3 mm FWHM to the average map to deal with any gaps arising at vertices with poor model fits.Visual areas V1-V4 were then defined using standard criteria (DeYoe et al., 1996;Engel, Glover, and Wandell, 1997;Sereno et al., 1995;Wandell, Dumoulin, and Brewer, 2007).See Fig. 3 for example rod-driven maps prior to treatment.

Data analysis
Modelled data for each participant were denoised using a min/max β cut-off of 0.005 and 5 respectively and had to exceed a 'goodness of fit' threshold of R 2 >0.03 to be included in the group analysis.This is more lenient than the usually accepted R 2 threshold of 0.05 and β minimum threshold of 0.01 to ensure weak, but critical, responses were not missed.See Figs.4A and 5A for group average data for R 2 and β.
For each participant, we calculated the number of vertices that survived the above inclusion criteria and averaged these for six different one-degree eccentricity bins between 0.5 and 6.5 degrees.These are presented as a group average for each stimulus type in Figs.4B and 5B.We then calculated the average pRF size (σ) for the same 6 eccentricity bins (Fig. 4C & 5 C).
In addition, we plotted the visual field eccentricity represented at increasing distance across the cortical surface (foveal representation = 0 mm).For this, we picked a point (vertex) on each individual's retinotopic map that approximated the occipital pole/foveal representation.This was done by interactively inspecting the polar and eccentricity maps on the inflated surface and choosing a vertex within the foveal confluence where iso-eccentricity bands could be detected.In-house software (https://osf.io/2rgsm)was then used to convert the functional maps into anatomical coordinates using the inflated spherical surface mesh (see methods 2.5).Retinotopic eccentricity estimates from the functional maps were then plotted against distance on the cortical surface.The same origin was used for the pre-and post-therapy analyses.To ensure artefactual estimates near the fovea were eliminated we start plotting at a cortical distance of 5 mm, with eccentricity estimates binned for each 1 mm on the cortical surface (Fig. 4D & 5D).
We used a linear regression for plotting pRF size against eccentricity and used a second-degree polynomial for comparing number of active vertices and eccentricity against anatomy.To compare between groups (controls Vs patients and pre-Vs post-treatment) we calculated the difference in (squared) area under the fitted data.To confirm that observed differences were robust we bootstrapped the fit by resampling each group 1000 times (with replacement), refitted the curves and recalculated the difference in squared area under the curve for each iteration.The proportion of bootstrapped differences that were opposite to the observed difference was calculated.

Gene therapy
Treatment involved a single sub-retinal injection of AAV8-hCARp.hCNGB3 as a suspension of viral vector particles to the worse seeing eye (as determined by the participant and clinician using ocular dominance or VA), with the aim of including the central macula in the bleb created.Despite the invasive delivery procedure, the treatment was well-tolerated (Michaelides et al., 2023).The time between gene therapy injection and the 1st and 2nd post-treatment scans are detailed in Table 1 in weeks.

Data quality
We recorded eye and head movement throughout all scan runs.All participants maintained good head stability, both before and after treatment.There was no difference in head movement parameters extracted during the MRI realignment stage before and after gene therapy for any participant, hence the data was averaged across all 3 scan runs.T-tests confirmed there was no significant difference in the mean SD between the control and patient groups for head translation (xhead: p=0.48; yhead: p=0.94; zhead: p=0.21) or rotation (pitch: p=0.23; roll: p=0.84, yaw p=0.68).
Horizontal eye movements are the dominant direction of nystagmus for patients with achromatopsia, therefore we calculated the median deviation in this dimension for all scan runs before and after treatment.To account for drift during the scan, eye position data was segmented into 10 equal sections, all recentred to the fixation point before the median deviation was calculated.This deviation was then averaged across the 10 segments for each scan run.In house software was used to detect and remove outliers associated with blinks and artefacts.Robust Z values were estimated for each datapoint and anything higher than Z=3 was removed.Z values were estimated as distance from median in units of 1.4826 median absolute deviation (Ben-Gal, 2005).
As expected, in all cases, the median deviation of eye movements for the patients were significantly greater than controls (p=0.02),confirmed by a significant main effect of group in a 2-way ANOVA with group (patients/controls) and stimulus (Rod-Equated/Cone-Equated) as factors.However, there was no main effect of stimulus condition (p=0.181) and no interaction between group and stimulus type (P=0.127).Furthermore, there was no significant increase or decrease in the deviation of eye movements after gene therapy (ANOVA main effect group (patients PRE/patients POST) P=0.54, main effect stimulus condition (Rod-Equated/Cone-Equated) p=0.078, and interaction P=0.84).We also looked at the number of outliers (data removed as blinks/artefacts) but this did not differ between groups (P=0.42) or stimulus type (p=0.22) and no interaction between group and stimulus (p=0.71).
Central fixation was further assessed by accuracy on the fixation task which ran throughout all scan runs.All control and patient participants were highly accurate at this task, either under or over pressing by no Individual maps for the 9 patients with achromatopsia prior to treatment.The maps have been restricted to areas V1-V4 (defined using the average map for all 3 stimulus typessee methods), surface smoothed using a kernel of 3 mm and thresholded at R 2 =0.03.The colour wheels indicate the color coding used for plotting polar angle and eccentricity.more than 3 %.There was no difference in accuracy between groups before treatment for either stimulus (CE P=0.66, RE P=0.28), and no difference before and after treatment for the patient group (CE P=1, RE P=0.65).

Psychophysics
Prior to each fMRI session, before and after gene therapy, all patients performed a short psychophysical test targeted at identify visual responses from cone-photoreceptors. Whilst lying in the scanner, after 30+ minutes of dark adaptation we presented each patient with a series of rod-equated stimuli at varying levels of cone-contrast (see methods for details).
Prior to treatment, most patients were unable to see any of the rodequated pairs, even those at the very highest levels of cone-contrast (all easily visible to normal sighted individuals), and none of them could see the cone-contrast chosen for the fMRI retinotopic mapping stimulus (N=15).See Fig. 2.
After treatment, one patient demonstrated a reliable change in their cone-driven colour perception (P07).At their 6-month visit, this patient self-reported being aware of more vibrant lines along the side of roads and that London Underground logos displayed in stations appeared more vivid.Consistent with this, their performance on the psychophysics task was much improved, such that they were able to accurately report the direction of movement of rod-equated (cone-activating) stimuli at contrast steps 11 and 13 of our 21-step range.These were at a cone-contrast level significantly lower than that detected by this patient prior to treatment (Fig. 2B).
For this individual, we analysed their fMRI data in isolation seeking to determine if this change in visual perception was accompanied by an increase in brain activity to rod-equated stimuli (see below).Note that the cone-contrast chosen for use in the fMRI experiment (N=15) was below the threshold of detection for this patient even after treatment, so any activity elicited by this stimulus was expected to be weak.

Comparison of fMRI results for patients and controls
For data to be included in the group analysis the individual model fit had to exceed a 'goodness of fit' threshold greater than R 2 =0.03 and a 'response magnitude' threshold greater than β=0.005.With this

Fig. 4. Group parameter estimates for V1 from pRF model prior to treatment. A)
To determine the reliability of responses for each group (patient/control) and stimulus type (Rod-/Cone-Equated) we plotted the Goodness-of-Fit (R 2 ) and Response Magnitude (β) against eccentricity.The generally accepted threshold for R 2 is 0.05 and β is 0.01.However, for all of the analysis included here we lowered these thresholds to R 2 =0.03 and β=0.005 (indicated by the dashed grey lines) to ensure subtle effects were not missed.Patient responses for the Rod-Equated (cone stimulating) stimulus did not reach either of these criteria and hence have not been included in the data presented in C-D.B) Number of active vertices by eccentricity.The number of active vertices for both patients and controls peaked around 3 deg eccentricity, but there was no difference between groups for the Cone-Equated (rod activating) stimulus (bootstrap test, p=0.444), but there was a significant difference for the Rod-Equated (cone activating) stimulus (bootstrap test, P<0.001).C) pRF size (Sigma) Vs Eccentricity.pRF size estimates for the Cone-Equated (rod activating) stimulus increased with eccentricity for both groups, however the size estimates were significantly larger for patients compared to controls (bootstrap test, P=0.002).Only the controls demonstrated reliable responses to the Rod-Equated (cone-activating) stimulus, so no comparison could be made.D) Eccentricity plotted against cortical location.The central fovea represents 0, with increasing cortical distance moving anteriorly along the cortical surface.Due to ambiguity at the extreme occipital pole we removed data from the central 5 mm of cortex.For all cortical locations (5-30 mm), the patients tended to represent more eccentric parts of the visual field than controls (bootstrap test, P=0.013).NB the maximum eccentricity of our retinotopic mapping stimuli was 6.5 deg.criterion, both patients and controls demonstrated clear and reliable responses to the Cone-Equated stimulus, which aimed to drive rod photoreceptors, labelled in the figures as 'Rod Responses'.See Fig. 3 for retinotopic maps driven by rod-photoreceptors for each patient prior to treatment, and a single example control for comparison.For all patients a polar angle and eccentricity map is evident.In contrast, only the control group showed a measurable response to the Rod-Equated stimulus, aimed at driving cone-photoreceptors, and labelled 'Cone Responses' in the figures.No patient demonstrated a response to this stimulus prior to treatment and there is no evidence of a retinotopic map (Fig. 4A).
To ensure we were not missing an unstructured increase in activity, we also identified how many vertices were active for each individual and each stimulus.For the Cone-Equated (rod-activating) stimulus we observed a peak in active vertices around 3 degrees eccentricity, but no significant differences between the patient and control groups (bootstrap test, p=0.256).For the Rod-Equated (cone activating) stimulus there were few active vertices for the patient group (as expected prior to treatment) and hence a significant difference between the groups (bootstrap test, P<0.001) (Fig. 4B).
Interestingly, when we plotted the modelled parameter for pRF size against eccentricity it was evident that the rod-driven responses differed Fig. 5. Group parameter estimates for V1 from pRF model POST treatment.For patients 1-7 the POST treatment data included here is for their 12month visit.For patients 8 & 9 (who did not attend a 12month scan) we have included their 6 month data.We carried out the same analysis as prior to treatment but instead of comparient patients with controls, here we are comparing patients POST-treatment results with their PRE-treatment results.A) To determine the reliability of responses at each time point (PRE-/POST-treatment) and for each stimulus type (Rod-/Cone-Equated) we plotted the Goodness-of-Fit (R 2 ) and Response Magnitude (β) against eccentricity.As before, we lowered the acceptable threshold for inclusion to R 2 =0.03 and β=0.005 (indicated by the dashed grey lines) to ensure subtle POST-treatment effects were not missed.Patient responses for the Rod-Equated (cone stimulating) stimulus PRE-and POST-treatment did not reach either of these criteria and hence have not been included in the data presented in C-D.B) Number of active vertices by eccentricity.As before, the number of active vertices PREand POST-treatment peaked around 3 deg eccentricity, but there was no increase in active vertices POST-treatment for the Cone-Equated (rod activating) stimulus (bootstrap test, p=0.415), or for the Rod-Equated (cone activating) stimulus (bootstrap test, P=0.169).C) pRF size (Sigma) Vs Eccentricity.There was no change in pRF size estimates at any eccentricity for the Cone-Equated (rod activating) stimulus after treatment (bootsrap test, P=0.464).D) Eccentricity plotted against cortical location.There was no significant change in the eccentricity represented at any cortical location after treatment (bootstrap test, P=0.104).
between the patients and controls.Although the slope of the fit was similar (slope = 0.11 & 0.13 respectively), pRF sizes were significantly greater for patients than controls, as evidenced by significantly greater area under the curve fitted to the pRF size (σ) by eccentricity data (bootstrap test, P=0.001) (Fig. 4C).Only the controls demonstrated reliable responses to the Rod-Equated (cone-activating) stimulus, so no comparison could be made.
We then assessed if a different profile in eccentricity representation existed for rod-driven responses across the cortical surface between patients and controls.We found that at any given location on the cortical surface, the patient group represented more eccentric areas of the visual field than controls (Fig. 4D).

Effect of targeted gene therapy on patient fMRI responses
As before, for data to be included in the group analysis the individual model fit had to exceed a 'goodness of fit' threshold greater than R 2 =0.03 and a 'Response magnitude' threshold greater than β=0.005.There also needed to be sufficient active vertices to ensure robust statistics.
This time we compared the patient responses pre-and post-gene therapy.For patients 1-7 we have used data acquired at their second post-treatment scan (12-18 months post treatment) and for patients 8 and 9 we used data acquired at their first post-treatment scan (~6 months post treatment) as these two patients could not attend a second post-treatment scan.
Disappointingly, for the patient group the R 2 and β parameters remained like those measured prior to treatment (Fig. 5A).There was, however, a suggestion that the number of active vertices increased after gene therapy for near foveal eccentricities (<3 deg).This effect was mainly driven by a single patient (P07) for whom there was some evidence of cone recovery and an increase in active vertices within this central region (see below & Fig. 6B).For the remaining 8 patients, for some there was a small increase in the number of active vertices after treatment, but for others this number decreased (range − 176 to +201, summed across all eccentricity bins 0.5 -6.5).None of these patients showed evidence of a change in perception in everyday life or to our cone-selective stimuli.
We also looked for a change in the rod responses after gene therapy, but these were unaffected by the cone-targeted treatment -pRF size, number of active vertices and anatomy profile all remained unchanged (Fig. 5B, C & D).

Individual patient with signs of cone recovery after gene therapy
For the single patient (P07) that demonstrated subjective changes in colour perception in their everyday life and an improvement in colour perception on our psychophysics task we analysed their individual data in more detail.
Firstly, we measured how many vertices met our criterion for inclusion in primary visual cortex (V1) before and after gene therapy.Prior to treatment only 344 vertices survived an R 2 of 0.03 and β greater than 0.003, but this significantly increased to 1073 by their 12-month post-treatment visit.A similar increase in active vertices was also evident at the stricter threshold of R 2 = 0.005, used for the group analysis.
Despite the promising findings above, the eccentricity and polar angle maps for this individual did not reveal any clear structure in either domain.And although a weak relationship between sigma and eccentricity was initially evident at their 1st post-treatment scan, this was not repeated at their 2nd post-treatment visit (Fig. 6B).
We analysed all the other individual's data in a similar way, but no other patient showed increased activity after treatment that survived even the most lenient criteria for inclusion.

Discussion
For our cohort of 9 adult patients with achromatopsia caused by the CNGB3 mutation, we conclude that there was no measurable recovery of cone-driven cortical responses after gene therapy when analysed as a group.However, on individual analysis there was one patient with an identifiable improvement in colour perception and a subtle accompanying change in cortical activity to cone selective stimuli.To our knowledge, this is the first glimpse of recovery of cortical responses following gene therapy in patients with the CNGB3 mutation.However, the response was weak and due to geographical and technical limitations we have not been able to follow this patient long term, to ensure reliability and repeatability of these results.Our findings add to those of McKyton et al., who found a minor improvement in adult patients with the CNGA3 mutation, providing some promise for older patients with achromatopsia (McKyton et al., 2021(McKyton et al., , 2023)).
One reason for the modest results we and others have observed following gene therapy in adult patients may be due to limited neural plasticity within the mature retino-cortical pathway.This supports earlier findings that suggest the congenital lack of cone input to visual cortex can lead to structural and functional changes in multiple visual areas (Molz et al., 2022;Lowndes et al., 2021;Baseler et al., 2002), which become more entrenched with time.Hence intervention earlier in life, when cones are better preserved and cortical structure is not that different from normal, would likely offer better visual outcomes (Molz et al., 2022).Certainly, preclinical data suggests that therapeutic intervention at a younger age leads to better results (Gray, 2016).Consistent with this, we have previously demonstrated improvements in cone function following gene therapy administered to children aged 10-15 years (Farahbakhsh et al., 2022).The most profound effects being found in children with the CNGA3 mutation.CNGA3 variants more often arise from missense mutations, rather than loss-of-function mutations as is the case in CNGB3, so may be more likely to have residual cone channel activity (Hirji et al., 2018).
A second reason for the modest results may be methodological.The contrast used for the cone-activating ('Rod-Equated') stimulus in our fMRI experiment was selected arbitrarily prior to treatment and without knowledge of the magnitude of improvement to expect in cone-related activity.It is possible that the selected contrast was too weak and in future studies we would recommend using a higher cone-contrast that might elicit a greater response.
In addition to assessing the effects of gene therapy on cone photoreceptor function within this patient group, we also compared the characteristics of pRF mapping to stimuli that isolated rod photoreceptor function with a group of normal-sighted individuals.ERG findings have previously suggested that rod function may not be as preserved as initially thought (Maguire et al., 2018).Similar to two previous fMRI accounts (Farahbakhsh et al., 2022;McKyton et al., 2021), we found pRF size estimates to be significantly greater in the patient group compared to controls, suggesting a coarser representation of the visual field by rod photoreceptors in this group.This could be artefactual, associated with noise in the data due to involuntary eye movements (nystagmus), which have been shown to increase pRF size estimates (Zuiderbaan, Harvey, and Dumoulin, 2012;Dumoulin and Wandell, 2008).However, if this were the case, we would also expect to see poorer model fits (i.e., lower R 2 values) in the patient group, which we do not (Fig. 4A & 4 A).There is also the possibility that dark adaptation started to wane and despite the very low scotopic light levels used for the rod-activating stimulus (<0.02 cd/m 2 ), unwanted contribution from cone photoreceptors may have influenced the pRF size estimates in the control group.Unlike McKyton et al., we did not observe a reduction in the size of rod-mediated pRFs in our patient group after gene therapy.This difference could be explained by the very different stimuli used, ours were colour calibrated for peak rod sensitivity and presented at very low scotopic light levels (<0.02 cd/m), whereas McKyton et al. used black and white stimuli with a maximum luminance of 180 cd/m 2 .This level of brightness would have saturated most rod photoreceptors.
We should note that despite carefully calibrating our stimuli to silence L-and M-cone photoreceptors, we did not directly control S-cone contribution.With 3 colour channels, we were limited to controlling only 2 cone photoreceptor types as well as rod responses.Therefore, to limit S-cone contribution we chose to keep the blue (B) voltage constant and only vary the red (R) and green (G) channels.Because S-cones are relatively more sensitive to wavelengths at the blue (short) end of the visual spectrum than L-and M-cones, this shifts stimulus variations towards longer wavelengths where there is minimal S-cone sensitivity.Furthermore, given the overlap in the rod and S-cone sensitivity range, if our stimuli were poorly calibrated, we would have unwanted rod responses as well as S-cone responses when viewing the rod-equated (cone-activating) stimulus.However, given the lack of response to our rod-equated stimulus in the psychophysical task we are confident our calibration methods were at least effective for the stimulus contrast used in the fMRI experiment.
We did however find that some of our patients were able to see the highest contrast pairs presented during the psychophysics experiment prior to scanning.This suggests some unintended contribution from rod photoreceptors at these very high levels, possibly due to error in measuring and correcting the gamma function of the MR-compatible screen we used.It was reassuring that no patient was able to see the cone-contrast level chosen for use in the fMRI experiment, however, for future studies we would highly recommend using a higher contrast stimulus in the hope of detecting weak cone responses.Care will need to be taken to repeatedly ensure all measurements are as accurate as possible.
In addition to larger rod-driven pRFs in our patient group, we also found more eccentric visual field representation for any given location on the cortical surface (Fig. 4D), suggesting a subtle reorganisation of the cortical map.This pattern of rod-driven responses remained unchanged by cone-targeted gene therapy (Fig. 5D).This finding corroborates an eccentricity shift observed in three patients by (Baseler et al., 2002) who also reported that parafoveal representations covered areas of the cortex that normally encode more foveal parts of the retina primarily innervated by cones in these patients.Together these studies suggest that the rod system appears to be altered in achromatopsia, but what happens to the deprived cortical areas and the processes giving rise to the various observed shifts in visual field representations remains unclear.Our current work includes characterising these processes and how they may interact with therapeutic outcomes, important for optimising the evolving treatment regimes.

Conclusion
In this study, gene therapy in adults with achromatopsia due to the CNGB3 mutation had no measurable effect on restoration of cone-driven cortical function for 8 out of the 9 patients.However, for one individual there was a glimpse of cone-driven activity in early visual areas.Future, larger scale studies would benefit from correlating changes in retinal structure with changes further down the visual pathway.We offer suggestions on how to improve methodology and sensitivity for these future studies.

Funding
EA was co-funded by The Wellcome Trust (091593/Z/10/Z) and MeiraGTx.The Wellcome Trust and MeiraGTx funded MRI scans and

Fig. 1 .
Fig. 1.Snap shots of the three stimulus types used in fMRI experiment.All participants underwent 6 functional scan runs -2 runs for each stimulus.The order remained the same for all participants: 1st Cone-Equated, 2nd Rod-Equated, 3rd Non-Equated.Individuals were instructed to maintain fixation on the central black dot throughout, and to press a button every time it flickered white.A low-contrast radial pattern was superimposed over the entire stimulus to aid fixation stability.Neutral density filters were used to ensure each stimulus was presented within the correct luminance range (Scotopic <0.2 cd/m 2 , photopic 0.8-1.2cd/m 2 , mesopic 0.3-0.7 cd/m 2 respectively).The rotating ring/wedge checkboard covered a circular region subtending 6.5 degrees of visual angle around fixation.

Fig. 3 .
Fig. 3. Retinotopic Maps of Polar angle and Eccentricity for the Cone-Equated (rod-activating) stimulus.Individual maps for the 9 patients with achromatopsia prior to treatment.The maps have been restricted to areas V1-V4 (defined using the average map for all 3 stimulus typessee methods), surface smoothed using a kernel of 3 mm and thresholded at R 2 =0.03.The colour wheels indicate the color coding used for plotting polar angle and eccentricity.

Fig. 6 .
Fig. 6.Eccentricity maps for individual patient P07 PRE and POST treatment.A) Active voxels that survived the model fit at R 2 >0.03 are presented on the inflated surface of the right and left occipital lobe for P07.Presented voxels have been restricted to V1, increasing in number after gene therapy.B) Despite the increase in active voxels after therapy, the relationship between pRF size and eccentricity remained weak.

Table 1 Participant demographics. Individual
's gender and age at pre-treatment (baseline) scan and the time in weeks between their gene therapy injection and the 1st and 2nd post-treatment (PT) scan.The timing of the 2nd posttreatment scan was reduced for patients P06 & P07 due to decommissioning of the scanner used for this study, and P08 & P09 could not return in time.