Hyperspectral imaging

As previously detailed in Lim et al.⁷, the HSI platform was a custom-built bench ophthalmoscope that consists of a fast-switching monochromator light-source (Polychrome V, Till Photonics, Hillsboro, OR, USA), which directed light to a semi-reflective pellicle beam splitter (BP245B1, ThorLabs, Newton, NJ, USA) and then into the eye. Fourty-five percent of incident light was directed into the rodent eye via the pellicle beam splitter which is then returned from the retina through diffuse reflection and passed through the pellicle again. Hyperspectral images are influenced by the light output, as well as the pellicle reflectance and transmittance, which can create a sinusoidal artefact⁷. Light reflected from the animal’s eye is transmitted through the beam splitter and focused by a series of lenses onto the “Neo” scientific complementary metal oxide semiconductor (sCMOS) camera (Andor Technology, Belfast, UK). Anesthetised animals were placed on a 3-dimentional cradle platform (ThorLabs, Newton, NJ, United States) in front of the pellicle. Images were taken with a 500×500 pixel region of interest centred on the optic nerve (Metamorph Software, Molecular Devices, San Jose CA, USA). Image acquisition was performed in a sequence from 320 to 680 nm (1 nm increments each with a 10 nm bandwidth) to capture a stacked image (TIFF format) where each slice represents one wavelength of incident light. Each wavelength takes 100 ms to cycle through, allowing the hyperspectral system to sequence 361 wavelengths in 36.1 s (Supplementary Fig. ¹).

HSI processing and analysis

Images from each region were exported as 16-bit greyscale TIFF images, with each stack including 361 slices representing different wavelengths of incident light from 320 to 680 nm. All images were analyzed using an open-source FIJI software package³¹ to collect median reflectance data from the same region of interests across all 361 slices. In FIJI, to compensate for eye/breathing movements each slice was registered to frame 108 (427 nm) where there is a high contrast between the blood vessels and the retina using the built-in algorithm “StackReg(Translation)”³² (Fig. 1A). Briefly, this method of image registration utilises a sub-pixel algorithm that minimizes the mean square difference of intensities between the reference image and subsequent images in the stack³². The ‘translation’ method used constrains the motion correction to lateral movement (x, y) without rotating, resizing the image or changing pixel intensity. A fixed retinal region of interest (ROI, Fig. 1E) was selected by manually masking the optic nerve head (159 pixel diameter, Fig. 1B), masking the smallest unvignetted field of view across all animals (286 pixels diameter, Fig. 1B) to ensure an area of even illumination was analyzed, and masking the major blood vessels (Fig. 1C, 20-pixel thickness). The image was then cropped to minimise file size and improve processing speed (Fig. 1D).

Open in a separate window

Figure 1

Image analysis masking process (A) Retinal image at 427 nm selected where blood vessels are visible. (B) Application of an optic nerve and pupil annulus with a fixed size are applied. (C) Vessels, optic nerve head and peripapillary region within the annulus centre are masked (D) Area outside of pupil annulus is masked (E) The remaining retinal regions of interest (within yellow borders) were analysed for reflectance.

Median reflectance data were collected from the retinal ROI for each of the 361 slices using the built-in Measurestack macro³¹. Data were then down-sampled in 5 nm steps (Microsoft Excel, Microsoft) by returning every 5th wavelength, i.e. 320, 325, 330 nm etc. To correct for overall differences in luminance, reflectance profiles were “baseline corrected” by subtracting the average reflectance from 320 to 360 nm (Eq. 1, UV wavelengths where spectral sensitivity of the camera should be negligible) and normalized to the average reflectance at a long wavelength band (630–660 nm, Eq. 2) consistent with previous studies^5,7,8.

$$Baselinecorrectedreflectanc{e}_{{\lambda }_n}=reflectanc{e}_{{\lambda }_n}-Averagereflectanc{e}_{{\lambda }_{320-360\text{nm}}}$$

1

$$Normalisedreflectanc{e}_{{\lambda }_n}=\left(\frac{Baselinecorrectedreflectanc{e}_{{\lambda }_n}}{Averagereflectanc{e}_{{\lambda }_{630-660\text{nm}}}}\right)\times 100$$

2

“Residual” spectra were calculated by subtracting the average reflectance value of the normalized wild type (WT) spectrum at each wavelength, to quantify reflectance differences due to disease. A singular HSI metric was calculated as a ratio between short (450–480 nm) and long (630–660 nm) wavelengths to enable plotting of heatmaps (Matlab, version 9.6 (R2019a) The Mathworks, Inc., Natick, MA, USA) and multivariate correlation with other parameters. A 30 nm band was used to encompass an instrument pellicle-induced full oscillation in order to decrease bias induced by a singular wavelength. The 450–480 nm band was chosen given this is where the maximal PD and aging changes were observed (Figs. 2, 3, and 5) consistent with previous studies in AD^5,7,8. The 630-660 nm band was chosen as an isobestic reference given previous studies find in in vitro^5,7 and in vivo^7,8 preparations of AD, less biological scatter occurs at these longer wavelengths (>~600 nm).

Retinal tissue analysis

Retinal levels of α-synuclein were quantified using western blot analysis. As A53T are genetically modified to express human native α-synuclein, toxic phosphorylated α-synuclein (pSer129 α-synuclein) and mouse α-synuclein were tested. The methodology and A53T data has been previously reported in Tran et al.²³ (a subset n=4–6 per age/phenotype included in this study) and the same approach was employed for TauKO tissue (n=4–6 per age/phenotype). In brief, retinal tissue samples were homogenised in radioimmunoprecipitation assay (RIPA) lysis buffer, 50 µg electrophoresed on 4–12% polyacrylamide gels (NuPAGE™, Invitrogen, Thermo Fisher Scientific) and proteins transferred onto 0.45 μm PVDF membrane (Immobilon-P, Merck, Sigma-Aldrich). Prior to blocking the membrane in tris-buffered saline (TBS, Tris 20 mM, NaCl 150 mM, pH 7.6) containing 5% skim milk, Ponceau S staining solution (Thermo Fisher Scientific) was used to capture total protein and quantified using the ChemiDoc MP imaging system (BioRad, Hercules, California, USA). Membranes were incubated at room temperature with primary and secondary antibodies for 1 h each. All antibodies, human α-synuclein, 1:1000, #ab138501 [MJFR1], Abcam, Cambridge, UK; pSer129 α-synuclein 1:1000, #ab209422 [EP1536Y], Abcam; mouse α-synuclein, 1:2000, #D37A6, Dako, Glostrup, Denmark; tyrosine hydroxylase (TH), 1:1000, #AB152, Merck, Kenilworth, New Jersey, USA) were diluted in blocking buffer above. All washes were carried out using 0.01% Tween-20 (TBS-T). Enhanced chemiluminescence (ECL, Clarity Kit, BioRad) was used to detect protein bands which were visualized with the ChemiDoc MP imaging system (BioRad) and quantified by densitometry (ImageLab 6.1, BioRad). The relative protein abundance of each cell lysate was normalized to the respective lane’s automated TP measurement via ChemiDoc stain-free detection software.

Iron levels in the retina (including RPE) were quantified using inductively coupled plasma mass spectrometry approaches similar to those previously described in our group in brain tissue³³. In a subset of animals (n=3–8 per age/genotype group), retinal samples were weighed, lyophilized, and digested overnight in nitric acid (HNO₃) (65% Suprapur, Merck). The samples were then heated at 90 °C for a 20 min period. Hydrogen peroxide (H₂O₂) (30% Aristar, BDH) of equal volume to the sample was added and left for 30 min, then heated for 15 min at 70 °C. Samples were then diluted with 1% HNO₃. An Agilent 7700 series ICPMS instrument measured iron levels under routine multi-element operating conditions and a helium reaction gas cell. Standard calibration solutions used were ICP-MS-CAL2-1, ICP-MS-CAL-3 and ICP-MS-CAL-4, Accustandard. For internal control, a certified standard solution containing 200 ppb of Yttrium (Y89) was used (ICP-MS-IS-MIX1-1, Accustandard). Iron levels are expressed as microgram of metal per gram of dry tissue weight (µg/g).

Optical coherence tomography (OCT, Spectralis^®, Heidelberg Engineering, Heidelberg, Germany) was used to quantify RNFL thickness in vivo as previously reported²³. A subset of the same A53T RNFL data is analysed in the current study (n=14–26 per group) and the same analysis conducted in Tau KO (n=8–15 per group) and C57blk6J mice (n=8–9 per group). In brief, retinal volume scans (8.1×8.1×1.9 mm) centred on the optic nerve head were acquired. OCT analysis. Automatic segmentation of the RNFL was conducted using the Heidelberg Eye Explorer 2 OCT reader plugin (Heyex, Heidelberg Engineering. The average RNFL thickness in the outer 6 mm Early Treatment Diabetic Retinopathy Study [ETDRS] ring was reported^23,34,35.

Statistical comparisons

Statistical tests were performed in Prism (version 8.0.2, Graphpad Software, Inc., La Jolla, CA, USA) and Matlab (version 9.6 (R2019a) The Mathworks, Inc., Natick, MA, USA). Repeated measures two-way analysis of variance (ANOVA) was applied, using Geisser-Greenhouse correction given that sphericity was not assumed with a repeated measures design³⁶. Wavelength, age and genotype effects were examined, as well as interaction between these effects (genotype × wavelength; age × wavelength). Post hoc analysis was conducted using a Benjamini et al.^37,38 approach which controls for the false discovery rate. Significance for p values was set at α=0.05. All group data are presented as mean±95% confidence interval (CI).

Multivariate analysis was performed to determine how the HSI signal may be influenced by healthy aging, as well as key pathogenic PD factors that have shown altered reflectance using other spectral techniques and preparations, namely α-synuclein, iron and RNFL thickness. Multivariate analysis was carried out by fitting a linear model to predict the HSI ratio (450–480 nm/630–660 nm). The “fitlm” function from MATLAB R2019a was used to find a least squares solution. The multivariate predictors always included age, and one of either α-synuclein levels, iron levels, or RNFL thickness (i.e. a maximum of two predictors were included in any given model). We have not included results using more predictors combined, as sample sizes were reduced because not all of the animals had all assays done. For each model presented below, we frame the results in the context of the improvement in the fit quality, as assessed by the R² value, for the two-predictor model compared with an appropriate univariate model (e.g. with just age as the predictor).

Hyperspectral imaging in an iron overaccumulation mouse model

The hyperspectral profiles in another PD model, TauKO mice which lack the tau protein and exhibit iron overload and progressive substantia nigra cell degeneration, is illustrated in Fig. 3. The representative heat maps in Fig. 3A show that at the early stage of the disease (8-month-old) TauKO retina appear similar to age-matched WT controls however by the advanced stage of the disease (18-month-old) TauKO retina have a considerably greater reflectance ratio exhibited as warm colours compared with WT counterparts. This is in agreement with the group average reflectance profiles (Fig. 3B) and (treated–control) residual graphs (Fig. 3C), where at 8-month-old TauKO mice a significant wavelength effect was found in the average reflectance profiles (p<0.0001) but no interaction (p=0.9881) nor genotype (p=0.9643). By 18 months of age, as for the advanced A53T mice, the TauKO mice (Fig. 3Bii,Cii) exhibited greater reflectance at short wavelengths. This manifested as a significant interaction (p<0.0001), wavelength (p<0.0001) and genotype (p=0.0147) with post-hoc changes significant in some wavelengths between 450–580 nm (Supplementary Table ⁵).

Open in a separate window

Figure 3

Hyperspectral imaging in TauKO mice. (A) Representative retinal heat maps illustrate the ratio between short and long wavelengths. TauKO mice exhibit warmer colours (higher ratio) compared to WT littermates at the oldest age. For all panels WT and TauKO comparisons are shown for the (i). 8-month-old (ii). 18-month-old cohort. (B) Average reflectance profiles of TauKO (red; 8-month-old: mid red; 18-month-old: dark red) and WT (blue; 8-month-old: mid blue; 18-month-old: dark blue) mice with 95% CI (grey area). At (i) 8-month-old a wavelength effect (p<0.0001) is found and at (ii). 18-month-old an interaction effect (p<0.0001) is observed, with a zoomed-in view (inset) of the reflectance at short wavelengths (460–520 nm) (C). By subtracting the TauKO group from the WT controls the residuals (red) indicate the same as Panel B. (i) In 8-month-old mice no interaction nor genotype effect is found and (ii) by 18-month an interaction effect is observed (p<0.0001). Insert shows zoomed in changes at short wavelengths. 95% CI of WT in grey area, *significance p<0.05.

Hyperspectral imaging in healthy aging across three wild-type strains

In order to compare the effects of normal healthy age-related changes on HSI against a PD phenotypes, control animals were examined. Figure 4 illustrates 3 separate cohorts of control mice. The left column (Fig. 4Bi,Ci) illustrates the changes with healthy aging in C57blk6J mice (WT littermates of 5xFAD mice) and show that with advancing age there is reduced reflectance at short wavelengths (Fig. 4Bi, interaction p<0.0001). This is the opposite pattern to that observed in the PD models (A53T Fig. 2; TauKO Fig. 3). This manifested as a significant interaction effect in the average reflectance profile (Fig. 4Bi, p<0.0001) and treated—control residual graphs (Fig. 4Ci, p<0.0001) with post hoc significance differences between groups at shorter wavelengths largely below 625 nm (Supplementary Table ⁶).

Open in a separate window

Figure 4

Hyperspectral imaging with healthy aging. (A) Representative retinal heat maps (ratio between short and long wavelengths) show that advancing age causes a lower ratio and hence cooler colours. In Panels (B–E), the following “control” cohorts are examined (i). C57blk6J mice at young (3-month-old; light blue), mid-age (6-month-old; medium blue) and older-age (12-month-old; dark blue) (ii). WT littermates of A53T mice (B6C3H background strain) at mid- (6-month-old; medium blue) and older-age (14-month-old; dark blue) (iii). WT littermates of TauKO mice (Sv129B background strain) at mid-age (8-month-old; medium blue) and older-age (18-month-old; dark blue). (B) Average reflectance profiles show that in all three control cohorts advancing age causes an interaction effect (p<0.0001) due to decreasing reflectivity at short wavelengths and increasing reflectivity at long wavelengths, with a zoomed-in view (insets) of the reflectance at short wavelengths (460–520 nm) (C). Similarly, the residual figures (older age—younger age) illustrate the same effect (interaction, p<0.0001). Insert shows zoomed in changes at short wavelengths. *significance p<0.05.

Similarly, to further examine the effects of healthy aging, the WT littermates of A53T mice at different ages (6-month-old and 14-month-old) were compared (middle column) as were the WT littermates of TauKO mice at different ages (8-month-old and 18-month-old) (right column). A similar effect was observed in both WT groups to that found in C57blk6J mice with advancing age. More specifically, a decreased reflectance at short wavelengths was noted (WT of A53T: Fig. 4Bii, interaction p<0.0001; Fig. 4Cii p<0.0001, WT of Tau KO: Fig. 4Biii, interaction p<0.0001; Fig. 4Ciii p<0.0001. Post hoc analyses in Supplementary Tables ⁷, ⁸).

Multivariate correlations of hyperspectral imaging to examine mechanism

To investigate whether α-synuclein, iron or RNFL thinning were contributing to the hyperspectral changes found in the PD animal models these were quantified in retinal tissue. As healthy aging also caused a change in hyperspectral profile, multivariate analyses were conducted comparing HSI ratio (450–480 nm/630–660 nm) against aging and examined whether this relationship improved with the addition of α-synuclein, iron or RNFL as a co-factor. This improvement is indicated by an improvement in the R² value (warmer to cooler colours) in the continued presence of a significant p value. Table 1 and Fig. 5 show that there is a significant relationship between HSI ratio and age (Table 1A, Fig. 5A) across all groups individually and when pooled (p<0.05). When α-synuclein is added as a cofactor (Table 1B, Fig. 5B) this improves the correlation in A53T mice than just age alone (R²=0.3141) and the relationship with toxic α-synuclein is marginally more (R²=0.4302) than the combination of all α-synuclein in the retina (R²=0.4099). Similarly, with the addition of iron (Table 1C, Fig. 5C) as a co-factor the correlation in A53T mice is improved (R²=0.4251) over age alone. In the TauKO mice when α-synuclein or iron were added as a co-factor the relationship was not significant (p=0.0551 to 0.0945). When RNFL is added as a cofactor to age (Table 1D, Fig. 5D) the multivariate correlation improves when all animals are pooled together, and even more so when only WT mice are analysed (Table 1E, Fig. 5E) as evidenced by an improvement in R² in most of groups examined (cooler highlight colours).

Table 1

Multivariate correlations between HSI and age, α-synuclein, iron and RNFL.

Open in a separate window

Significant relationships (p<0.05) are highlighted in different colours according to the R² value with R²=0 to 0.1 (red), R²=0.1 to 0.2 (orange), R²=0.3 to 0.4 (yellow), R²=0.4 to 0.5 (light green), R²=>0.5 (dark green). A53T mice underwent multivariate correlation against toxic phosphorylated α-synuclein (*) and total α-synuclein (^, human native α-synuclein+toxic α-synuclein+mouse α-synuclein). TauKO mice only possess endogenous rodent α-synuclein and thus were correlated against retinal mouse α-synuclein levels (^#).

Open in a separate window

Figure 5

Multivariate correlations between HSI and age, α-synuclein, iron and RNFL. Representative images of co-factors and their method of analysis show in top row. α-synuclein assessed with western blot, from Tran et al.²³; iron assessed with mass spectrometry; RNFL quantified via OCT, from Tran et al.²³ (A). HSI ratio correlated against age (B). HSI ratio correlated against age and α-synuclein. In A53T mice toxic phosphorylated α-synuclein is plotted and in TauKO animals, mouse α-synuclein is plotted (C). HSI ratio correlated against age and iron (D). HSI ratio correlated against age and RNFL (B). HSI ratio correlated against age and RNFL in control animals only (healthy aging).

Hyperspectral imaging in Parkinson’s disease

Our HSI data suggests that in the A53T mouse model there is a wavelength dependent Parkinson's effect on reflectance. This difference presents in the late (14-month-old) disease stage as a linear decreasing trend with wavelength, resulting in increased reflectance at lower wavelengths (shorter than 585 nm) of incident light compared to WT littermates (Fig. 2Ciii, Supplementary Table ³). In comparison the 4-month-old and 6-month-old A53T data depicts an A53T reflectance profile which lies within the 95%CI of the WT littermates, suggesting that this wavelength dependent change in the A53T model presents only at advanced disease stages (Fig. 2; Supplementary Tables ¹–³). These age-related changes are in accordance with the literature and our work²³ where western blot assessment of α-synuclein in the retina (pSer129) plateaus between 8 and 18 months of age but is similar to control at 4 to 6 months of age. This is also in agreement with studies from our group and others showing that the motor phenotype and cortical α synuclein changes plateau between 8 and 16 months of age.

Similar hyperspectral effects were found in the TauKO model with an increased reflectance at short wavelengths, below 580 nm (Fig. 3, Supplementary Table ⁵). Again, the difference presented in the later disease stage (18-month-old) paralleling when motor dysfunction manifests (between 12 to 24 months²⁷), α-synuclein levels increase in the caudate putamen (15-month-old²⁸) and Nissl stereological counts and tyrosine hydroxylase stained neurons decrease in the substantia nigra (12 months²⁷). At earlier prodromal stages of the disease a hyperspectral effect is not observed (8-month-old) which again parallels the overall disease stage with a lack of motor impairment (i.e. 6 to 7-month-old^27,28) and cortical α synuclein noted at similar ages. It is worth noting that the reflectance heat maps show non-uniform reflectance across the retina. It is likely that the spatial heterogeneity may be related to the spatial distribution of the driving factor, e,g., spatial distribution of a-synuclein/iron or RNFL thinning, Further study to co-localize these heat maps with flat mounted tissue stained with α-synuclein or spatial mapping of RNFL thinning is warranted to understand the non-uniform reflectance across the retina. An additional improvement to the processing of hyperspectral images in future work would be to normalize the intensity recorded at each wavelength according to the spectral output of the source. This was not undertaken here, as the significant non-uniformity imposed on the spectrum by the pellicle beam-splitter complicates this task (especially as it is sensitive to small rotations of the beam-splitter). However, as the major aim is to compare the wild-type to the disease conditions, normalization of the spectra does not have any bearing on the conclusions of the study, which rely on making statistical comparisons between the two groups which each underwent identical processing steps.

Although the two Parkinson’s models showed similar hyperspectral profiles, healthy aging alone exhibited a distinct and opposing signature. As such given the complex interplay between these factors, to investigate the influence of PD driven pathobiological changes on HSI a multivariate analysis was conducted adding on α-synuclein or iron as a co-factor to age (Table 1). This demonstrated that in A53T mice, addition of α-synuclein improved the capacity of the model to explain the HSI ratio. This was the case for both the toxic phosphorylated form of α-synuclein and an overall measure of α-synuclein (summation of pSer129, human α-synuclein, mouse α-synuclein). Iron levels in the retina and RPE also improved the models’ predictive capacity for HSI ratio in A53T mice. Although the TauKO mice did not show marked improvements with the addition of α-synuclein or iron as a co-factor for multivariate analyses, these multivariate data sets had lower sample size (A53T n=28 to 33; TauKO n=19 to 22) and a decreased spread of data (no “young” age group represented by a narrower axis in Fig. 5) than the A53T counterparts. Thus, future investigation in larger sample sizes at a broader range of ages is warranted in TauKO mice.

This suggestion that retinal α-synuclein and iron might be the underlying biology that drive Parkinson’s related HSI is supported by the literature which has examined other forms of retinal spectral imaging in in vitro or ex vivo preparations. Oh et al.¹³ used a human neuroblastoma cell line which was treated with 1-methyl-4-phenylpyridinium (MPP+) and ferric ammonium citrate that caused iron accumulation. They imaged these cell culture preparations with a dark-field HSI system that spanned 400 to 1000 nm wavelengths. They found that imaging isolated iron within a water mixture (ferric ammonium citrate) exhibited absorption in a bell-shaped curve. These iron depositions within their cell culture model showed absorption largely occurred between~475 and 620 nm (EC₅₀ estimates) with a peak at~535 nm. In contrast, the glass plate showed no reflectance and the cellular areas without iron were much less reflectile. These findings are in agreement with the current study which shows increased reflectance across a similar wavelength band (~450–600 nm). The opposing polarity is observed between studies because Oh et al.¹³ used dark field microscopy (thus manifesting as increased absorbance) and the current study employed an ophthalmoscopic system more akin to bright field microscopy (manifesting as increased reflectance). This sign inversion from absorption to reflectance has previously been shown in hyperspectral studies of Alzheimer’s retinae.⁷

The presence of proteins in the retina have been implicated to increase light scatter in the visible spectrum range. This has been shown for amyloid-beta (implicated in AD) in cell culture⁵, within ex vivo⁵ and in vivo mouse preparations^7,8 as well as human patients⁶.This has been shown for amyloid-beta (implicated in AD) in cell culture⁵, within ex vivo⁵ and in vivo mouse preparations^7,8 as well as human patients⁶. α-synuclein (implicated in PD) is another protein which has shown increased light scattering properties. In vitro preparations of native α-synuclein exhibit Rayleigh scatter when an excitation wavelength of 350 nm was used and an emission band recorded across 300 to 400 nm Rayleigh scattering occurs when light interacts with scatterers smaller than~1/10th the wavelength of incident light and may explain the increased reflectance at shorter wavelengths within the current hyperspectral study. Mammadova et al.¹⁰ used a different spectroscopy approach (Raman spectroscopy) on retinal cross sections from A53T mice. Raman spectroscopy uses a single monochromatic wavelength (532 nm) and measures scatter when it passes through a substance at additional frequencies that are distinct from the excitation wavelength. This approach also differentiated between age-matched control and A53T mice within ex vivo samples. These studies differ in their spectrometry approach to the current study (hyperspectral imaging) but further highlight the light scattering properties of α-synuclein.

The current study shows an in vivo hyperspectral profile from PD retinae that are distinct from control. Collectively, previous literature and the current study indicate that the presence of α-synuclein and iron in the retina may influence HSI. Future interventional studies which pharmacologically decrease tissue levels of α-synuclein or iron would further aide our understanding of the factors which drive HSI.

Hyperspectral imaging with healthy aging

Three separate cohorts of wild-type mice indicate that with advancing age there is less reflectivity at shorter wavelengths, largely below 600 nm, with increasing spectral differences the shorter the wavelength (Fig. 4, Supplementary Tables ⁶–⁸). The magnitude of this reduced reflectance was more pronounced the greater the difference in age between the groups compared.

Aging in the eye can present in several different ways including vitreous liquefaction, cataracts, inflammation, oxidative stress and lipofuscin accumulation³⁹. In particular, several aging changes in the eye can reduce short wavelength reflectance. Age-related lenticular changes cause a reduction of light transmission, particularly in the shorter wavelength bands (~400–500 nm) as the yellowing of the lens promotes blue light absorption^39–41. Within HSI this would manifest as a decreased hyperspectral reflectance with worsening lens changes, particularly at shorter wavelengths.

One way to assess the effect of age-related lens changes on HSI may be to compare phakic and pseudophakic patients. However as commonly employed intraocular lenses (IOLs) have yellow-tinting, blue light and UV blocking properties this may account for a similar hyperspectral profile in 10 patients who were examined before and after cataract surgery^6. This is advantageous from a clinical implementation point of view as it would indicate that HSI may be used in patients with or without their cataracts removed. To directly answer whether the aging lens may be contributing to the decreased reflectance seen in this study would require removal of the rodent lens and replacement with a transparent IOL that does not possess any yellow tinting/blue light filters. Comparison of HSI before and after cataract surgery in human patients with transparent IOLs used may also elucidate the clinical utility here.

Previous studies in animal models and humans have shown that the RNFL is reflectile at short wavelengths^{14–16,42,43}. Addition of RNFL as a cofactor to age with both WT and PD groups did not markedly improve multivariate correlations in individual strains of mice, but an improvement in R² was observed when all groups were pooled together (Table 1A, D). In contrast, when only WT animals were analyzed, thereby examining the effect of RNFL influencing healthy aging alone, RNFL as a cofactor improved most of the correlations (Table 1A, E). One explanation for this is that when PD and control groups are combined there are multiple factors at play (including PD driven α-synuclein/iron deposition as well as age-induced RNFL thinning) whereas with advancing age alone, RNFL alterations play a more dominant role in influencing hyperspectral outcomes. To examine this a multivariate analysis which takes into account all factors would be ideal. Although not possible in the current study due to limited sample size (HSI, α-synuclein, iron, RNFL were only examined in n=5 to 8 overlapping animals) this remains an area for future investigation.

If RNFL thinning and/or lenticular changes are driving the aging changes seen (Fig. 4), given that these are commonly assessed at eyecare clinics, in the future correcting for these masking effects would further increase sensitivity of HSI to differentiate PD from control subjects.

The authors would like to thank Amelia Sedjahtera (Parkinson’s Disease Laboratory, The Florey Institute of Neuroscience and Mental Health, The University of Melbourne, Parkville, 3010, Victoria, Australia) for breeding and genotyping the A53T mice.