RNA sequencing normalised counts were prepared for linear modelling using voomWithQualityWeights transformation (Law et al., 2014) and then a statistical model was fitted using lmFit (Ritchie et al., 2015) and moderated t-statistics were computed using ebayes function. Tables of fold change and p-value estimates were generated with top Table function and genes with p-value<0.05 and absolute fold changes >0.5 were deemed differentially expressed. Principal component analysis (PCA) and hierarchical clustering (HC) were used as methods for sample clustering and computations were performed using prcomp and hclust R functions (Team RC, 2019). Gene-wise scaling was implemented prior to sample clustering and Euclidean distance (calculated with dist R function) and complete linkage were used as clustering distance and agglomeration method respectively. Statistical testing of sample grouping on PCA was performed using the envfit function from the vegan R package using 10^6 sample permutations. Modular gene co-expression analysis and gene set enrichment analysis were carried out using CEMiTool R package and MSigDB C2 canonical pathway as reference set (Subramanian et al., 2005; Russo et al., 2018). Unpaired Student’s t-tests, one-way analysis of variance (ANOVA), or two-way ANOVA were performed using Prism V7.0. p-values < 0.05 were deemed statistically significant. All data was expressed as the mean ± SEM. Normality and homoscedasticity assumptions for parametric tests were visually verified by inspecting distributions of standardised residuals and the relationship between the predictor variables and the square root of standardised residuals. Kruskal–Wallis and Wilcoxon rank sum tests were employed where parametric tests were deemed unsuitable.

Exercise has well known and documented systemic benefits, including to the CNS and in protection against neurodegenerative diseases (reviewed in Chu-Tan et al. (2022). Therefore, to investigate the effects of exercise on retinal protection against degeneration, mice either with or without (sedentary) free-access to running wheels for 14 or 28 days (14 days in supplementary data), were subjected to PD-induced retinal degeneration. Note that for the 14-day experiments, mixtures of male and female mice were used due to experimental constraints. Therefore, this may not be directly comparable to the 28-day data, where only females were used to control for sex-associated effects. We thus focus on the 28-day cohorts, with the 14-day experiment placed in the supplementary data and alluded to. Following experimental paradigms, retinas were analysed and compared between groups for functional, morphological and molecular changes (Figure 1A). To verify exercise outputs, and correlate running data to measures of retinal protection; recorded measures of distance (km), time (min), number of runs, max speed (m/s) and, average speed (m/s) in hour bins over a period of 28 days were analysed.

Mice initiated physical activity immediately after being provided with access to the running wheel and continued to utilise the running wheel until the cessation of the 28-day exercise period (Figure 1B). After deriving daily totals for distance, time, and number of runs and deriving cumulative sums over the 28 days period we report that, mice on average ran a total distance of 316 km (166—428, SEM 32.3), utilised the running wheel for an average of 128 h (70—128, SEM 677) and performed on average a total of 4,670 individual runs (4,048—5,799, SEM 220) (Figure 1B). All three exercise parameters were accrued over the 28 days in a consistent linear fashion with an almost invariant slope (Figure 1B). To understand the exercise behaviour of mice in further detail, daily averages were next computed for distance, time, number of runs as well as mean and maximum speed. Mice on average, ran 10.2 km (5.54—14.2, SE 1.84), 257 min (137—323, SE 24.3), and 155 runs (134—192, SE7.9) runs per day, with a max speed of 0.598 m/s (0.417—0.679, SE 0.035), and average speed of 0.060 m/s (0.033—0.083, SE 0.0068) (Figure 1C). Intraday exercise activity as measured by running distance, time, number of runs and maximum speed occurred cyclically and coincided with the 12 h light/dark housing conditions with most of the exercise activity being performed during the nocturnal phase (Figure 1D).

Next, we compared the daily exercise output of mice with access to running wheels to that of mice performing spontaneous in-cage activity without running wheels as reported by Pernold et al. (2019). The metrics chosen for this comparison were daily distance and daily mean speed measured using both CST and video tracking by Pernold et al. (2019). In the absence of a running wheel, mice covered a distance of 0.2 km per day at an average speed of 0.0024 m/s (Figure 1E) which represents approximately 50-fold decrease in activity level when compared to the 10.2 km daily distance and 0.060 m/s daily mean speed registered on the running wheel system (Figure 1E).

Taken together, tracking of the exercise output over a 28-day period demonstrates that the use of a running wheel setup is a reliable method for increasing physical activity in mice, orders of magnitude over the level of activity performed as a result of spontaneous in-cage movement.

To investigate the effect of exercise on retinal protection against degeneration, mice (n = 8–9) were subjected to 5 days of photo-oxidative damage immediately after 30 days of exercise then retinal function and morphology were assessed using ERG, OCT and fundoscopy. Relative to sedentary controls, exercised mice had significantly improved retinal function for both a-wave (photoreceptor function) and b-wave (second order neuron) (p < 0.05, Figures 2A–C). Fundus/OCT imaging (Figures 2D–G), further demonstrate the protective effects of exercise on the retina during degeneration. Areas of free of degenerative lesions (defines as retinal regions presenting with of hyperreflective puncta, hyperpigmentation and severe ONL loss shown by dashed areas in Figure 2D) were larger in exercised mice (39.7% ± 6.84%) compared to sedentary controls (17.4% ± 2.42%) (Figure 2E, p < 0.05). Thickness measurements carried out two-disc diameters superior to the optic nerve head indicated that both the ONL (Figure 2E) and whole retinal thickness (WR) (Figure 2F) were significantly (p < 0.05) larger in the exercise animals across five distinct eccentricities. Exercised mice had a WR of 116.2 ± 2.08 μm and ONL thickness of 26.7 ± 1.73 μm while the same two measurements were 104.8 ± 1.34 μm and 12.4 ± 0.99 μm for sedentary mice respectively. In the 14-day cohort, no significant differences were observed in either ERG or ONL thickness (Supplementary Figure S1).

Given the partial protection in retinal function and preservation of retinal morphology observed in exercised mice, (Figure 2), key features of retinal degeneration (photoreceptor cell death, and presence and activation of microglia/macrophage immune cells) were measured and compared between sedentary and exercised mice following PD. Compared to sedentary controls, mice with access to voluntary exercise had significantly reduced levels of photoreceptor cell death (p < 0.05, Figures 3A,B), as measured by a decreased number of TUNEL⁺ cells across the ONL, and in the superior ONL (p < 0.05, Figure 3C), and an increased number of photoreceptor rows (p < 0.05, Figure 3D). Further, as a measure of retinal inflammation, IBA1⁺ microglial/macrophage cells were counted and morphologically characterised in the ONL. In exercised mice, IBA1⁺ cells were significantly reduced in the ONL compared to respective sedentary controls (p < 0.05, Figures 3E,F), notably with significant reductions in the superior retina (p < 0.05, Figure 3G), and in both ramified and amoeboid IBA1⁺ morphologies (p < 0.05, Figure 3H). This was true also for the 14-day cohort where photoreceptor cell death and inflammation as measured by IBA1 was significantly reduced in the exercised mice (Supplementary Figure S2). Collectively these results demonstrate the protective effect of voluntary exercise against retinal degeneration within the 28-day time frame, with functional preservation attributed to reduced levels of photoreceptor cell death and retinal inflammation in exercised mice.

To investigate the molecular pathways underpinning the observed retinal protection, bulk RNA sequencing was performed on retinal lysates from exercised and sedentary mice exposed to PD and compared to dim sedentary mice. Following read count normalisation (Figure 4A), a two-dimensional principal component analysis was employed to visualise global differences between the expression of retinal mRNA profiles from exercised and sedentary mice (Figure 4B). Using envfit analysis and permutation-based statistical testing, we determined that exercise and sedentary mice are arranged in two distinct clusters on the PCA space (p < 0.05), thus suggesting that when compared to sedentary mice, exercise induces consistent gene expression changes in mice exposed to PD (Figure 4B). Next, we sought to determine if global gene expression patterns in exercised mice shifts the RNA profile of a heathy (non-PD) mouse retina. To achieve, a subset of genes (differentially expressed between exercised and sedentary mice, log₂fold change (FC) > 0.5, p < 0.05) were identified (Figure 4C; orange dots), then using the expression values of this subset of genes, the Euclidean distances (ED) of both exercise and sedentary mice to dim controls were calculated. Hierarchical clustering of EDs computed based on differentially expressed genes, indicated that 7/9 mice had a gene expression profile more similar to a heathy retina than sedentary controls (Figure 4D). Besides clustering, the magnitude of EDs was statistically tested (Figure 4E) and revealed that the exercise shortens the separation between heathy controls and PD-retinas (ED_{dim-sedentary} = 24.5 ± 0.47, ED_dim-active = 20.2 ± 0.21, p < 0.05). Taken together these results suggest that exercise partially shifts the transcriptome of the PD retina towards an expression profile that aligns more closely to a dim-reared, “healthy” retina.

As the retinal gene expression profile from exercised mice was found to be trending towards that of healthy control (dim) retinas, to identify the specific gene and gene pathways modulated in response to voluntary exercise and responsible for this profile shift, modular gene co-expression analysis was performed in combination with gene set enrichment analysis. Following modularisation of normalised gene counts, six gene expression modules were identified, with modules (M) M2, M3, and M4 showing significantly positive enrichment, and modules M1, M5 and M6 showing significantly decreased enrichment in exercised mice compared to sedentary controls (p < 0.05, Figure 5A). Of the six modules (Supplementary Table S2), M5 and M6 (Figure 5B) were further explored because they contained signalling pathways and immune mediators central to the pathogenesis of retinal degeneration. Pathways associated M5 included microglial and leukocyte cell migration and phagocytosis, complement activation and inflammasome activation (p < 0.05, Figure 5C). Consistent with a downregulation of in the enrichment of pathways in M5, key mediators of retinal inflammation such as C3, C1q, Cx3CR1 and Cd74 were observed to be downregulated in exercised animals (Figure 5C) relative to sedentary controls. Further, pathways significantly associated with M6 genes, and known to play key roles in retinal degeneration included extracellular matrix modulation, integrity and organisation, and oxidative stress responses (p < 0.05, Figure 5D).

Taken together, these results support a role for exercise-mediated retinal protection against degeneration through a combination of influencing inflammation and ECM modulation and show a transcriptomic shift towards a healthy phenotype.

We have provided for the first time a foundational picture of the major molecular pathways underlying exercise-mediated retinal protection and identified two major themes that had overall transcriptomic shifts towards a healthy phenotype: inflammation, and extracellular matrix integrity.

Over-activation of the immune response within the CNS is a hallmark feature of neurodegeneration and has been established to influence the progression of retinal degenerations, including AMD (Ambati et al., 2013; Kauppinen et al., 2016). Chronic and progressive inflammation in retinal degenerations is primarily mediated by the innate immune system, and largely controlled by both the resident retinal microglia as well as recruited peripheral macrophages (Amor et al., 2010; Ambati et al., 2013; Kauppinen et al., 2016). Reducing pathological levels of inflammation may therefore be a key therapeutic strategy in slowing the progression of retinal degenerations, making the anti-inflammatory properties of exercise a considerable therapeutic interest. There is a strong link between exercise and the inflammatory system in the human body including in the CNS (Nieman and Wentz, 2019). Exercise is known to modulate key inflammatory mechanisms, mitigating and regulating systemic inflammation in neurological diseases such as multiple sclerosis (Dalgas et al., 2012), systemic lupus erythematous (Perandini et al., 2012), Alzheimer’s and Parkinson’s diseases (Fuller et al., 2020), and preclinical models of retinal degeneration (Crowston et al., 2017; Sellers et al., 2019; Zhang et al., 2019; Dantis Pereira de Campos et al., 2020). However, these anti-neuroinflammatory effects, whilst observed, have not been well-substantiated or often investigated past single target identification.

Results from this study support the anti-inflammatory effects of exercise, with reduced infiltration and activation of IBA1⁺ microglia/macrophages in the retina of exercised mice. in particular, reduced level of microglia/macrophages were detected in the ONL of the central/superior retina—an area well-characterised to be the primary site of damage during light-induced degeneration (Natoli et al., 2016). In support of these observations, pathway analysis of RNA sequencing results indicated decreased expression of microglial/macrophage markers, including Cd68, Cd74 and Cx3Cr1—all known to have increased expression and drive pathological inflammatory processes in AMD (Ambati et al., 2013). Further, bioinformatic analyses indicated that the complement system, which has been previously identified as a major instigator of inflammatory dysregulation in AMD (Edwards et al., 2005; Hageman et al., 2005; Haines et al., 2005; Klein et al., 2005), was less active in exercised animals. Notably, complement pathway genes C1qa, C1qb, C1qc, C3 and Serping1 were all significantly downregulated in the retinas of exercised mice. In addition, the Nlrp3 pathway also appeared to be less active in exercised animals. In AMD, NLRP3-dependent cleavage and activation of pro-inflammatory cytokine IL-1β (Halle et al., 2008) has been shown to drastically increase excessive inflammation in the retina, (Wooff et al., 2019), and is identified to play a key role in disease pathogenesis causing progressive photoreceptor cell death. Taken together, these results highlight potential pathways modulated in response to exercise that could be key to dampening the observed retinal immune response and providing protection against degeneration.

In addition to inflammatory modulation, key findings from this work identified that extracellular matrix integrity pathways were also significantly modulated in response to exercise. The extracellular matrix, acellular protein-rich scaffold, consisting of proteoglycans and glycoproteins, plays an important homeostatic role in maintaining retinal health and integrity functioning in both structural and mechanical support (Al-Ubaidi et al., 2013). In retinal degenerations including AMD, extracellular matrix degradation and remodelling have been strongly associated with disease pathogenesis (Al-Ubaidi et al., 2013). Results from our study show that in exercised mice extracellular matrix genes such as Itgam (Cd11b), a key mediator of monocyte adhesion (Solovjov et al., 2005), were significantly downregulated compared to sedentary mice. As integrin-mediated adhesion to the ECM plays a key role in infiltration of immune cells during damage or degeneration (Baiula et al., 2021; Mrugacz et al., 2021), modulations in this pathway may lead to reduced immune cell presence in the retina hereby haltering damaging and pathological immune cascades. Our results strongly support this hypothesis, with an observed reduction in recruited IBA1⁺ cells and microglial/macrophage markers, as well as a downregulation in several notable immune pathways known to be propagated by these immune cells and involved in AMD disease progression (Rutar et al., 2015; Jiao et al., 2018). Additionally, Tissue Inhibitor of Metalloproteinases 3 (Timp3) which was also found to be downregulated in exercised mice, has been heavily implicated as a potential candidate in AMD pathogenesis, with elevated levels seen to contribute to the thickening of Bruch’s membrane commonly seen in AMD patients (Kamei and Hollyfield, 1999). The diverse roles of Timp3 in inflammation, angiogenesis, and extracellular matrix turnover/degradation, have made it a closely looked at molecule for complex degenerative diseases including Alzheimer’s disease and AMD (Dewing et al., 2019).

These novel results provide further insight into the molecular pathways that are involved in providing neuronal protection to the retina following exercise. It must be noted that unlike in previous studies (Poo, 2001; Binder and Scharfman, 2004), we did not observe a significant change in brain-derived neurotrophic factor (BDNF) which has been a heavily postulated to be a crucial molecule in exercise-mediated neuronal protection (Poo, 2001; Binder and Scharfman, 2004). However, the sensitive nature of Bdnf deems it difficult to detect changes without conducting high-sensitivity experiments on its effect on specific pathways such as TrkB (Allen et al., 2018) or detecting their protein rather than RNA levels. Further, the protective effects of exercise are pleiotropic with the same gene influencing different phenotypic traits and expression patterns differing based on various forms of exercise (Côté et al., 2011; Keeler et al., 2012). Therefore, we cannot rule out a possibility of Bdnf playing a role in the protection that we observed despite no significant changes to it or its primary receptor.

To date, the majority of animal studies investigating the effect of exercise in the retina have used forced exercise models where an often unpleasant or painful stimulus is used to ensure the animals complete a defined amount of physical activity (Ji et al., 2013; Kim et al., 2013; Chrysostomou et al., 2014; Lawson et al., 2014; Chrysostomou et al., 2016; Allen et al., 2018; Mees et al., 2019; Dantis Pereira de Campos et al., 2020). While it is understandable that studies using forced-exercise models use this method in an attempt to ensure equal and consistent levels of desired exercise, output measures from our study revealed that mice would voluntarily and consistently utilize the running wheels, running large distances of 10.2 km/day on average, mitigating any need to force this seemingly natural behavior onto them. Further, given the stress and anxiety-like behaviours known to be inflicted in these forced exercise models (Leasure and Jones, 2008), forced-exercise methods may potential confound the true effects of exercise. Studies have in fact demonstrated not only significant differences in brain behaviours, but also differences in the intensity and duration of activity under forced and voluntary exercise animal paradigms, despite attempts at standardization (Leasure and Jones, 2008). Even when distance was held constant, voluntary exercisers reached much higher speeds but for shorter time periods (43.7 m/min in voluntary exercise animals and 15.5 m/min in forced exercise animals) (Leasure and Jones, 2008). Due to the much higher speeds in the voluntary exercise groups, to effectively standardise the distance; forced exercise animals had to run for far-longer periods of time, creating inherent differences within the exercise paradigms (Leasure and Jones, 2008), and making the results difficult to compare and decipher. Further, stress caused by such models may disturb the inflammatory balance of the retina and impact on output measures as a result (Malan et al., 2020). As the use and output measures of voluntary models of exercise more closely mimic what may be prescribed in a clinical setting to human patients, it appears more intuitive to use voluntary models such as the one in this study to truly understand the molecular mechanisms and pathways underlying exercise-induced neuroprotection.

It should also be noted that our supplementary results demonstrated that there at 14 days of exercise, the protective effect seen at 28 days is not present in the functional results and only in histology did we observe protection against photo-oxidative damage. Whilst interesting and potentially indicative of a certain amount of exercise needed to provide the full suite of protection, these results cannot be directly compared due to differences in sex and age of animals for the experiments.

It remains unclear precisely how systemic effects from exercise can target the retina and confer protection against degeneration. It is likely however that tissue crosstalk is occurring to communicate the message of exercise to the rest of the body. Landmark studies have in fact identified that exercise-mediated protection may be facilitated, in part, by cellular communication via extracellular vesicles (EV) (Frühbeis et al., 2015; Whitham et al., 2018). EV (including exosomes and microvesicles) are nanosized cell-to-cell delivery vesicles which are released by nearly all cell types in the body and function to induce a biological response in recipient cells (Hessvik and Llorente, 2018) via the transfer of molecular cargo, including proteins, lipids and non-coding RNA, such as miRNA (Hessvik and Llorente, 2018). Studies have shown that following exhaustive cycle and treadmill exercise, a significant increase in EV numbers was found in the circulation immediately post exercise (Frühbeis et al., 2015; Whitham et al., 2018; Brahmer et al., 2019). Further, beneficial molecular cargo including proteins and miRNA carried within exercise-EV, have been shown to be responsible for exerting systemic neuroprotective effects, including promoting advantageous metabolic changes (Castaño et al., 2020), increasing skeletal muscle fibre growth and recovery (Wu et al., 2022), and modulating immune responses in the CNS (Frühbeis et al., 2015; Whitham et al., 2018; Vechetti et al., 2021). These findings largely support a role for EV/EV cargo in mediating exercise-induced systemic protection, suggesting that investigations into the role of EV in mediating retinal protection during exercise warrant further investigations, and should be avenues for future exploration.