Parkinson's disease (PD) is a progressive neurodegenerative disorder that affects motor control; however, non-motor symptoms, particularly cognitive impairment, are often the debilitating aspect of PD (Christopher and Strafella, 2013; Kehagia et al., 2010).

The concept of “brain age” has emerged to comprehend the aging brain and its associated neurodegenerative processes (Eickhoff et al., 2021; Franke and Gaser, 2019). Brain age models estimate an individual's brain age based on the brain MRI data, and the individual deviation of biological “brain age” from chronological age—the brain age difference (BAD)—is widely recognized as a marker of brain health (Cole et al., 2019). As PD shows evidence of robust brain atrophy (Xu et al., 2020), several studies have investigated brain age in this context (Beheshti et al., 2019; Eickhoff et al., 2021; Teipel et al., 2024). Recently, using the Parkinson's Progression Markers Initiative database, a study discovered that higher BAD in PD is primarily associated with global gray matter volume and with the basal forebrain, which is atrophied in both PD and Alzheimer's disease (Teipel et al., 2024). Interestingly, the magnitude of accelerated brain aging in patients with PD is constantly reported to be associated not only with motor but also with cognitive impairment (Chen et al., 2024; Eickhoff et al., 2021).

Meanwhile, the cognitive reserve (CR) theory seeks to explain the observed mismatch between the degree of brain pathology and clinical manifestations (Hindle et al., 2016). Indirect proxies are commonly used to measure CR, such as years of education, intelligence quotient, occupational attainment, and cognitive lifestyles (Hindle et al., 2017; Loftus et al., 2021; Stern, 2009). Previous studies discovered that years of education may protect against cognitive impairment associated with PD (Hindle et al., 2016; Loftus et al., 2021). Reportedly, higher educational attainment is associated with improved baseline cognitive performance across various domains in patients with PD (Gu and Xu, 2022). Therefore, education may contribute to a higher CR, providing individuals with a greater cognitive capacity and helping to buffer against the impact of neurodegeneration.

Based on the findings from studies explaining cognitive decline in PD in terms of BAD and CR, we hypothesized that higher years of education might modulate the cognitive decline associated with neurodegeneration in PD. Specifically, the study explored whether years of education moderate the relationship between BAD and multiple domains of cognitive decline in PD.

Table 1 and Figure 1 present the characteristics of the study participants and the brain age estimation process. As shown in the scatter plot in Figure 1B, brain age tends to increase with chronological age, with a Pearson's correlation of r = 0.769 (p < 0.001). The mean absolute error of brain age prediction is 6.05 ± 4.48. There was variability in antiparkinsonian medication doses, which may affect cognitive performance and limit the generalizability of the results.

Among the six cognitive domains, only the moderation model with language ability as the dependent variable showed model significance with an FDR- and the Hommel-adjusted p-value of 0.0492 (raw p = 0.0093 for language), which means that BAD, years of education, and their interaction explain a relevant portion of the variance in language ability. A sensitivity analysis using Bayesian estimation revealed a Bayes factor of 4.60, indicating moderate evidence in favor of the alternative hypothesis over the null hypothesis (Van Doorn et al., 2021). The model had an R² of 0.2001, indicating that this model accounted for approximately 20.01% of the language score variation. The remaining five domains—global cognition (R² = 0.12, Hommel-adjusted p = 0.17), attention (R² = 0.04, adjusted p = 0.51), learning and memory (R² = 0.11, adjusted p = 0.23), executive (R² = 0.15, adjusted p = 0.15), and visuospatial (R² = 0.18, adjusted p = 0.08)—did not yield significant moderation effects after multiple comparison correction.

Notably, the moderation model with language as the dependent variable had significant model coefficients (Table 2, Figure 2A). There was a significant interaction between years of education and BAD [β (95% CI) = 0.0106 (0.0023–0.0188)]. Furthermore, we examined the effect of BAD on language changes at different education levels—specifically at the 16th, 50th, and 84th percentile ranks—within the sample of 58 subjects. At the 16th percentile (having 13.96 years of education), the effect of BAD on language ability was 0.0425 and statistically significant (p = 0.0141, Figure 2B); this finding indicates that, for participants with low years of education, a high BAD was associated with a significant decrease in language scores.

This study explored whether CR moderates the interaction between neurodegeneration and the multiple domains of cognitive decline in PD, using years of education as a proxy for CR and BAD as a marker of neurodegeneration. After multiple comparison corrections, only the moderation model with language as the dependent variable was found to be significant. These results support the hypothesis that years of education moderate the relationship between BAD and language ability decline (Table 2 and Figure 2A). For individuals with fewer years of education, a higher BAD (Figure 1B, older appearing brain) was associated with lower language scores (Figure 2B, effect = −0.042). With increasing years of education, the negative association between BAD and language scores weakens and becomes non-significant at higher education years (Figure 2B, effect = 0.000). These findings provide evidence for a nuanced relationship, in which the impact of brain aging on language is not uniform but influenced by the educational background, and emphasize the benefit of achieving basic years of education. Similarly, Glatt et al. reported that having less than a high school diploma of education considerably predicts dementia in PD, although the relationship between years of education and the dementia rating scale score was small (r = 0.12) (Glatt et al., 1996). Altogether, the education's buffering effect against cognitive decline is more effective in achieving basic years of education.

Furthermore, we identified a single significant moderation model explaining language ability after multiple comparison correction. The cognitive tasks measuring language ability (i.e., BNT, FAS) focus on verbal fluency. Previous studies investigating the effect of education on verbal fluency reported that education has a positive effect on this language ability. Loftus et al. investigated the relationship between CR and cognitive decline in a large PD sample (n = 334) and reported that years of education significantly predicted verbal fluency measured by the FAS (Loftus et al., 2021). This task requires participants to generate as many words as possible within a set time, according to specific rules (Sauzéon et al., 2011). Task performance is, therefore, dependent on one's processing speed, executive functions, and vocabulary level. Accordingly, having higher years of education and vocabulary offsets the effects of cognitive decline associated with aging on letter fluency tasks (Sauzéon et al., 2011). In a study of 1,392 patients with progressive cognitive decline, Zamarian et al. reported that, although higher education is not helpful for episodic memory and executive functions at low cognitive levels, it is beneficial for retrieving words or other semantic knowledge (Zamarian et al., 2021). Therefore, the protective effect of education might be nuanced and could vary depending on the specific cognitive functions under examination. These inconsistencies highlight the interplay complexity between education and cognitive decline in the context of PD and suggest that BAD could be the key to explaining such inconsistencies and complexity.

Previous literature identified multiple gray matter regions, including the bilateral hippocampi, the right inferior temporal gyrus, and the right lateral orbitofrontal gyrus, as images feature attributable to advanced brain aging in PD (Eickhoff et al., 2021). These striato-prefrontal brain areas, driving accelerated brain aging, were demonstrated to have functional effects on cognition in the patient group (Chen et al., 2024). On the other hand, education may induce structural brain changes via neuroplasticity, making it a promising modifiable factor in PD intervention strategies (Arenaza-Urquijo et al., 2017a,b). Having a cognitive lifestyle, including longer education, is linked to neurotrophic changes in the prefrontal lobe, implying a compensatory process (Bennett et al., 2014). Longer education not only enhances the cognitive ability per se, but also compensates for brain structure that may contribute to preserving language ability, as evidenced by better verbal language ability showing greater structural covariance between left and right frontal regions (Qi et al., 2019). In addition, memory in the brain can be broadly categorized into two systems: explicit memory, which involves the conscious recall of people, places, and objects, and implicit memory, which underlies automatic skills such as driving or using correct grammar (Kandel, 2018; Squire et al., 1993). Unlike explicit memory, which depends heavily on higher cognitive regions—primarily the neocortex and hippocampus, both of which are commonly implicated in brain age models—implicit memory relies more on subcortical and sensorimotor-related structures, such as the amygdala and cerebellum (Teipel et al., 2024). Therefore, cognitive tasks that engage explicit memory and language may be more vulnerable to brain aging effects, explaining why language ability showed a significant moderation effect in this study.

There are some limitations to this study. First, our dataset of 58 PD subjects lacks a healthy comparison control group, longitudinal follow-up, and diversity in the sample (e.g., variability in education and ethnicity), as well as possible confounding effects from antiparkinsonian medications, which may inflate the model estimates and limit the generalizability of the results. Second, potential confounding factors, such as lifestyle and coexisting medical conditions, which can influence brain age and cognitive function, can make it challenging to isolate the specific effects of education on brain age and cognitive decline in PD research. Third, the sample shows a gender imbalance with more male participants, which reflects the known higher prevalence of Parkinson's disease in men (Wooten et al., 2004). However, we acknowledge that this may still limit the generalizability of our findings regarding sex-specific effects. Finally, proxies of CR that may explain cognition other than years of education were not evaluated. For example, Koerts et al. examined the relationship between CR and cognition using years of education and premorbid IQ as CR proxies and reported that premorbid IQ predicted executive function in PD, but not years of education (Koerts et al., 2011).

In conclusion, this study highlights the role of education in buffering the cognitive decline associated with accelerated brain aging in PD, particularly when referring to language abilities. Therefore, interventions promoting basic level educational achievements may help mitigate cognitive impairment in patients with PD.