Light at night exposure and risk of dementia conversion from mild cognitive impairment in a Northern Italy population

Dementia is a progressive cognitive syndrome with increasing incidence in the decades. It is expected that subjects with dementia could nearly triple from 57 million in 2019 up to over 150 million in 2050 [1]. Some non-modifiable risk factors including aging, female sex, and genetic susceptibility are known to increase disease risk [2–5]. However, a pivotal role of environmental and lifestyle factors has been suggested for dementia onset and progression [5–7]. These include low education and socio-economic status, dietary habits, traumatic brain injury, and cardiovascular diseases independently from stroke [8–15].

Environmental factors linked to urbanization and climate change have been recently outlined [16, 17], especially outdoor air pollution [18–21], noise pollution [22, 23], greenness [24–27]. In addition, outdoor artificial lighting during night hours, generally referred as “light at night” (LAN), has been pointed out as possible relevant linkage within the relation between urban environment and human health [28–31].

LAN exposure already demonstrated detrimental effects in humans due to different mechanisms, mainly disruption of circadian rhythm, sleep disturbances, and altered secretion of melatonin and other hormones [32–34]. Among the investigated health outcomes, LAN exposure has been reported to increase risk of some types of cancer [35, 36], metabolic diseases [37–39], and mental disorders [40–42].

Findings from few experimental animal and laboratory studies indicates that light exposure during night cycle and disruption of circadian rhythm may affect neuron vitality and activity through several mechanisms including tau protein deposition, altered neuronal architecture and increased oxidative stress [43–50]. In addition, two recent epidemiologic studies reported a positive relation between LAN and risk of dementia [51, 52]. However, evidence is still scarce.

For these reasons, we aimed to evaluate the association between LAN exposure and dementia risk of conversion to dementia in an Italian population of subjects diagnosed with mild cognitive impairment (MCI).

Study population characteristics at baseline are reported in Table 1. Overall, we recruited 53 (males/females: 28/25) subjects with MCI with available CSF samples and with available environmental data. Median age at recruitment was 66.3 years (range 42.6–81.6 years), slightly higher in males (67.1 years) than females (64.7 years).

During an overall median follow-up of 54 months (IQR: 38–89 months), 19 subjects did not convert and MCI diagnosis was confirmed, while 34 subjects converted to any type of dementia, 26 of whom converted to AD. Among the non-converters, there was a higher number of subjects who were younger than 65 at recruitment, had lower education, and were non-smokers (Table 1). Two subjects were affected by COPD, both in those who converted. Similarly, two subjects were diagnosed with diabetes, one each in those converted and not-converted. Assessment of APOE4 genotype was available for 38 subjects, showing a higher percentage in subjects who converted to any dementia or AD. Details on other types of dementia are reported in Supplementary Table S1.

The 9-year 2014–2022 average annual VIIRS data over the study area is shown in Fig. 1. Median (IQR) level of annual LAN exposure of the 2014–2022 period is 26.4 nW/cm²/sr (IQR: 14.9–33.5) (Table 1). Levels are almost similar in males and females and with or without COPD, but slightly higher in subjects aged ≥ 65 years at recruitment, smokers, subjects with lower education (< 8 years), those without diabetes and those positive for APO4E genotype. Data are substantially comparable when considering yearly VIIRS data (Supplementary Table S2 and Figure S2). Correlation analysis showed very comparable values of VIIRS data across the considered years and with VIIRS 2014–2022 average (always > 0.95) (Supplementary Table S3 and Figures S3).

In the restricted cubic splines, there is an almost null association between LAN exposure till 30 nW/cm²/sr and baseline concentrations of biomarkers (Fig. 2). Above this level, there is a strong positive association with p-tau181 and t-tau and a slightly inverse association with amyloid Aβ₁₋₄₂. Further adjustment for mean PM₁₀ concentrations marginally affected the relations (Supplementary Figure S4).

Distribution of subjects according to increasing outdoor artificial LAN exposure and risk of conversion to any type of dementia and AD are reported in Table 2. Linear increase of LAN exposure shows positive association with HR of 1.04 (95% CI 1.01–1.07). Using the lowest tertile as reference category, LAN exposure in the second and third tertiles shows dose-response increase with HRs of 2.53 (95% CI 0.99–6.50) and 3.61 (95% CI 1.34–9.74), respectively. Similar trends are depicted by division with fixed categories. Similar increase in risk can be noted in the analysis restricted to conversion to AD, although characterized by higher imprecision due to lower number of subjects. Further adjustments for relevant confounders confirm such increase in risk, with especially stronger association after controlling for traffic-related PM₁₀ concentrations, smoking status, COPD and diabetes (Supplementary Tables S4-S7), and to a lesser extent also for APOE4 status (Supplementary Table S8). Analysis restricted to non-smokers shows similar results to the overall population (Supplementary Table S9), as it was the case after exclusion of subjects with diagnosis within 12 months (Supplementary Table S10). In the sensitivity analysis using 2014 VIIRS, we found almost identical results (Supplementary Tables S11-S18).

Spline regression analysis assessing non-linear relation is reported in Fig. 3, showing substantial linear risk for both any type of dementia and AD above 30 nW/cm²/sr of LAN exposure. The relation does not change across adjusted models further accounting for PM₁₀ concentrations and smoking status (Fig. 3). Alternative adjustment by COPD and diabetes shows similar results for any type of dementia and AD (Supplementary Figure S5). Adjustment by APO4E status yields similar results for any type of dementia, while almost null association can be noted for AD, with the risk starting to increase above 35–40 nW/cm²/sr only (Supplementary Figure S5). In the sensitivity analysis restricted to non-smokers (Supplementary Figure S6) and excluding subjects with diagnosis within 12 months (Supplementary Figures S7-S8), the relation is almost identical to the overall population, as it was the case when using 2014 VIIRS data instead of 2014–2022 average (Supplementary Figures S9-S13).

In our prospective study carried out in a Northern Italy population with MCI, we showed a positive association between outdoor artificial LAN exposure and increased risk of conversion to dementia. Such association appears to have a threshold effect above 30 nW/cm²/sr. It remained strong also after adjustment for relevant confounders including sex, age at entry, education, traffic-related air pollution and smoking history, and in sensitivity analyses performed after exclusion of subjects converted within 1 year and smokers. The association became almost linear when specifically focusing on conversion to AD.

Our findings are consistent with a previous study reporting a positive association between LAN exposure and MCI prevalence in Chinese veterans [51]. Conversely, in a study carried out in the same Italian area, we found contrasting results. Indeed, using a case-control design, LAN showed inconsistent association with early-onset dementia, while we found an almost linear positive association with late-onset dementia (LOD), although characterized by a high imprecision of the risk estimates [52].

Nonetheless, our results further support the hypothesis that either dementia onset or progression may be driven by non-genetic risk factors such as LAN, within the general landscape considering genetic and environmental risk factors and their complex interplay [7]. Interestingly, we reported consistent associations between LAN and dementia risk when assessing specific CSF biomarkers of neurodegeneration, specifically increased levels of t-tau and p-tau181 and decreased levels of amyloid Aβ₁₋₄₂, in line with previous studies [74, 75]. However, we recognize that such relation was much stronger and statistically stable for tau proteins only, being less consistent and very imprecise for amyloid Aβ₁₋₄₂. Such difference may suggest a stronger effect of LAN in the progression of neurodegeneration process specifically through alteration of tau proteins as indicated.

Our results are supported by some laboratory studies suggesting that LAN may increase neurodegeneration. In particular, dim light exposure during night hours induced disruption of circadian rhythm and sleep thus promoting accumulation of tau protein in the brains of tauopathy/AD Drosophila fly model [43]. Similarly, light exposure increases tau cleavage and neurodegeneration in the Drosophila spaghetti fly model [44]. Despite most of the attention has been devoted to circadian rhythm alteration and clock genes expression [32, 47, 50], the exact mechanisms of the effects of LAN exposure are not fully understood and several studies have indicated that they are not limited to an altered clock daytime perception. As a matter of fact, altered expression of genes related to oxidative stress (e.g., altered expression of heat shock proteins, lactate dehydrogenase, or lipid peroxidation) have been described [44, 48, 76], or altered nighttime activity and behavior as well as metabolism and hormone secretion [77–80]. Finally, some animal studies suggested that LAN negatively impacts memory and cognitive performances [76, 81–84], specifically affecting brain regions linked to learning and memory like hippocampus [76, 85]. In particular, LAN exposure may alter vascular structure and function of the hippocampus [86]. Nonetheless, other independent mechanisms may affect such relation. For instance, sleep disorders and deprivation have been suggested to dysregulate amyloid-beta clearance/deposition [87–90], thus affecting neurodegeneration independently from LAN exposure. With regards to tau protein, it is also known to be released into the extracellular space during periods of heightened excitatory neuronal activity [91]. Hence, we speculate that a prolonged elevation in neuronal activity due to LAN might enhance tau propagation and contribute to pathological processes.

An important limit of our analysis is the limited sample size affecting the statistical precision of the risk estimates and the interpretation of the results, indicating caution in the generalization to other population and the need to further assess such relation in larger investigations. In addition, despite we included major confounders and adjusted by other relevant environmental factors like smoking, traffic-related air particulate matter, some chronic disease, and partially also genetic factors, the observational study design does not entirely rule out the occurrence of residual confounding due to other factors not considered in the analysis. In particular, we did not assess greenness exposure, a factor recently associated with a non-linear U-shaped pattern with both dementia risk [24] and neuropsychiatric symptoms in subjects with dementia [27]. Although, several indicators can be used, such as living close to green spaces or satellite-based assessments like Normalized Difference Vegetation Index (NDVI) [92], several factors (especially modality and resolution) may affect their use and application in epidemiologic studies [93] and specifically hampered its assessment to this study population. Due to the study design, we could not retrieve information of indoor artificial LAN exposure as well as night shift working, although both indoor and outdoor lighting showed to affect each other and have similar intensity and spectral composition [94]. Similarly, exposure assessment using VIIRS data hampered the evaluation of possible difference of specific spectra, specifically a higher detrimental effect of blue light [47, 48].

In addition, we cannot entirely exclude risk of misclassification of exposure using residential data although we did not contact study subjects and we obtained median LAN exposure (26.4 nW/cm²/sr) almost identical with the previous investigation in the same Northern Italy province (25.8 nW/cm²/sr) [52], although lower than the Chinese study (44.4 nW/cm²/sr) [51]. Despite such difference partially limits the comparison of our results with other studies, the implementation of several categorical analyses and especially non-linear splines over the entire range of exposure levels should substantially improve risk characterization and ease the use of these findings in meta-analyses.

Some strengths of the present study should be highlighted. For exposure assessment, we used a valid method based on satellite data within a geographic information system, characterized by the highest available spatial resolution using of VIIRS imagery data [95], implementing both correlation and sensitivity analysis across different available datasets. In addition, to the best of our knowledge, this is the first prospective study assessing the relation between outdoor artificial LAN and risk of dementia conversion. The use of LAN levels from VIIRS data from either the follow-up period (2014–2022 9-year average) or at the begin at follow-up (2014), should rule out risk of misclassification and reverse causation due to change of exposure after dementia diagnosis [96].

Our findings suggest that outdoor artificial LAN may increase dementia conversion, especially above 30 nW/cm²/sr, although the limited sample size suggests caution in the interpretation of the results, and the need to further evaluate such relation in larger studies.

Below is the link to the electronic supplementary material.

Supplementary Material 1