Alzheimer’s disease (AD) is a complex neurodegenerative disorder whose risk is influenced by many factors, including age, genes, exposures, health issues, and others (Yashin et al. 2018; Ukraintseva et al., 2024; Akushevich et al., 2023; Livingston et al., 2024). A ‘multi-hit’ hypothesis of AD implies that some critical amount of risk factors (‘hits’) is required to trigger AD in an individual (Gong et al., 2018; Patrick et al., 2019; Steele et al., 2022). The APOE4 is the strongest genetic risk factor for AD. Infections and traffic-related air pollution (TRAP) have recently emerged as important exposure-related risk factors that may act together with APOE4 to promote AD-related pathology (Popov et al., 2024; Rajendrakumar et al., 2025).

Infections have been associated with AD in many studies, though exact mechanism is debated, as multiple pathways may be involved (Vigasova et al., 2021; Fülöp et al., 2020; Lotz et al., 2021; Whitson et al., 2022; Ecarnot et al., 2023; Ukraintseva et al., 2024). The roles of neuroinflammation, inflammasome signaling, microglia, self-reactive T cells, gut-brain axis, and amyloid beta as a potential antimicrobial protein, have been in major focus of recent research (Whitson et al., 2022; Ecarnot et al., 2023). Herpesviruses as drivers of AD have also been broadly discussed (Itzhaki, 2017; Wainberg et al., 2021). Our recent study compared associations of very different infectious diseases (bacterial, viral, and fungal) with AD in a large pseudorandomized sample of older adults, and found that they all are associated with AD, while different vaccines were protective. This indicates that compromised immunity could play a central role in the associations between infections and AD rather than a specific pathogen (Ukraintseva et al., 2024).

A growing body of research also points to a connection between exposure to air pollution and neurodegenerative disorders, including AD, though mechanisms are not fully understood (Yuan et al., 2023; Parra et al., 2022; Tham and Schikowski, 2021; Finch, 2023; Franz et al., 2023). Various pollutants are present in the air, and some may pose risks to human health. For example, inhalable particulate matter (PM) and nitrogen dioxide (NO₂) have been intensively studied in this regard (Craig et al., 2008; Akimoto, 2003; Dominski et al., 2021). A recent analysis of UK Biobank (UKB) data found that higher exposure to PM2.5 (median particle with diameter ≤ 2.5 μm) and NO₂ is associated with multimorbidity in a dose-dependent manner (Ronaldson et al., 2022). The PM, NO₂, and volatile organic compounds (VOCs) are common components of the traffic-related air pollution (TRAP). These and other types of air pollution (such as ozone, sulfur oxides, carbon monoxide, and lead), might be harmful to the central nervous system (CNS) and promote neuroinflammation and neurodegeneration (Costa et al., 2020; Cheng et al., 2016; Calderón-Garcidueñas et al., 2002, 2015; Hogan et al., 2015; Spangenberg and Green, 2017). A review of epidemiological and experimental studies of the role of PM in neurodegeneration emphasized a link between chronic exposure to PM and onsets of cognitive deficits, dementia, and AD (You et al., 2022). A meta-analysis of 14 studies concluded that PM2.5 is a risk factor for dementia, with more limited support for nitrogen oxides, though the authors stressed that these results should be interpreted with caution (Wilker et al., 2023). Higher exposure to NO₂ itself was associated with lower cortical thickness of brain regions relevant to AD (Crous-Bou et al., 2020). Another study that used the UKB data (Li et al., 2023) reported an association between residential distance to major roads and dementia that was mediated by TRAP, mainly NO₂. It has been associated with a decline in cognitive function and progression of mild cognitive impairment (MCI) to AD (Jack Jr et al., 2000; Henneman et al., 2009; Qu et al., 2023). Thus, the role of exposure to TRAP in neurodegeneration has been broadly supported by recent research.

Here, we investigate how a history of infections and chronic exposure to TRAP—separately and in combination—may influence the risk of AD and other dementias (AD+) in carriers and non-carriers of APOE4, the strongest genetic risk factor for AD. Our goal is to better understand the interplay between these major risk factors within the multifactorial mechanism of AD development.

The data was adjusted to align age means in the age distributions in the groups DNMR_Infs, DNMR_noInfs, noDNMR_Infs, and noDNMR_noInfs (Supplementary Tables 5–10). After adjusting data sets, the difference between mean ages in groups was not significant for males, females and both males and females. This ensured that the observed effect in risk of exposure to TRAP and history of infections on AD+ was not due to potentially younger mean age in the groups. Note that only seven female subjects were in the DNMR_Infs group (Table 1). So, we preferred to mostly utilize a data set including both females and males as being more statistically reliable.

Using pairwise group comparisons (Table 2 and Figure 1A), we found that UKB participants aged 60–75 years with a history of infections and high exposure to TRAP, for men and women combined, had a 164% higher risk of AD+, as compared to individuals of the same age without either risk factor (RR = 2.64 (DNMR_Infs/noDNMR_noInfs), 95% CI [1.79, 3.88]). Separately, infections without TRAP increased the risk of AD+ by 54% (RR = 1.54 (noDNMR_Infs/noDNMR_noInfs), 95% CI [1.32, 1.78]). The impact of TRAP without infections on AD+ was not significant. In non-carriers of APOE4 (Figure 1C, Supplementary Table 4) with both a history of infections and exposure to TRAP, the relative risk of AD+ was 4.49 (95% CI [2.68, 7.50], risk ratio (DNMR_Infs/noDNMR_noInfs)) compared to subjects without either risk factor. Infections alone (without TRAP) influenced the risk of AD+ less substantially (RR = 2.04 (noDNMR_Infs/noDNMR_noInfs), 95% CI [1.62, 2.55]). The impact of TRAP without infections on AD+ was not significant. In APOE4 carriers (Figure 1B, Supplementary Table 3), the association between a history of infections and exposure to TRAP with the risk of AD+ was not significant. Table 2 and Supplementary Tables 3, 4 show results of comparisons between all groups.

The logistic regression analysis confirmed and supplemented the result obtained by means of the groups pairwise comparisons analysis for participants aged 60–75 (Table 3). In particular, going from dnmr = 1, infs = 1 to dnmr = 0, infs = 0 corresponds to the comparison between DNMR_Infs and noDNMR_noInfs groups. Let us denote by A the right side of the logistic regression equation using regression coefficients in Table 3: A = −15.489 + 0.171*Age + 0.429*infs + 0.558*dnmr*infs. Then the expression for the risk is as follows: risk = 1/(1 + exp (−A)) (Supplementary material, Analytic approach 2.2). Taking into account that the mean age in DNMR_Infs group meanAge1 = 63.56 years and the mean age in noDNMR_noInfs group meanAge2 = 63.59 years (Supplementary Tables 5, 6) and setting Age = meanAge1 or Age = meanAge2, after calculations, we have the risk values in groups DNMR_Infs and noDNMR_noInfs and the relative risk respectively: risk1 = 0.0257, risk2 = 0.0098, and RR = risk1/risk2 = 2.6265. With a good approximation, these values are close to the corresponding values 0.0267, 0.0101, and 2.6395 presented in Table 2. For APOE4 carriers and APOE4 non-carriers cases, the results for the logistic regression analysis are presented in Supplementary Tables 12–15 (more details about regression coefficients are shown in Supplementary material 2 tables).

Note also that the risk difference and the relative risk in groups DNMR_Infs and noDNMR_noInfs increased, respectively, by 92% and by 42% when going from the reference model without interaction (when main factors were independent) to the model with interaction between main factors (Supplementary material, Reference regression model, Supplementary Table 16).

The regression model (Table 3) was assessed for confounding by the covariates related to education, smoking, and the Townsend index (Townsend Deprivation Index). The Townsend index has been the favoured deprivation measure among UK health authorities (Measuring Deprivation, 2002). The following variables were added: education = 1 (for subjects with College or University degree), education = 0 (for subjects without College or University degree), smoker = 1 (if the subject was a smoker), smoker = 0 (if the subject was a non-smoker). For Townsend index, five quintiles were considered: Quintile 1, 0–20%, it represents the least deprived 20% of the population, also known as the most affluent group; Quintile 2, 20–40%, it represents the second-least deprived group; Quintile 3, 40–60%, it represents the middle group in terms of deprivation. Quintile 4, 60–80%, it represents the second-most deprived group; Quintile 5, 80–100%, it represents the most deprived 20% of the population. Four dummy variables tsi1, tsi2, tsi3, and tsi4, with each variable representing one of the Quintiles 2–5 compared to the Quintile 1. The regression model was then run using these four dummy variables as predictors, allowing us to analyze the effect of each level on the risk of AD.

The results are presented in Table 3 in ‘Models with relevant covariates’ section. Note that adding education, smoking, tsi1, tsi2, tsi3, and tsi4 variables to a regression model, only slightly changed the coefficients of the initial variables of interest. It suggests that the added variables were not significantly related to the terms of our interest Age, infs, dnmr, infs*dnmr and the regression model (Best model, females and males, 60–75 in Table 3) has been already capturing the essential information.

To further analyze sensitivity and sustainability in our analysis, we applied bias correction technique (Firth, 1993) and the Synthetic Minority Over-sampling Technique (SMOTE) oversampling technique (Chawla et al., 2002). In particular, both techniques are highly effective for handling rare events and balancing data in logistic regression models. The results of the bias correction analysis showed that the coefficients for terms Age, infs, dnmr the logistic regression changed about 5% when applying the bias correction methods (Supplementary material 3, Bias correction for handling rare events in logistic regression). It supported a good level of sensitivity and sustainability in our analysis.

Applying the SMOTE techniques showed how sensitive and sustainable our model and data were to relatively small changes in initial conditions. (Supplementary material 3, Estimate logistic regression coefficients using SMOTE). It allowed to successfully check the absence of potential tangible irregularities in the model and data. Therefore, the SMOTE results also supported a good level of sensitivity and sustainability in our analysis.

Thus, adding covariates, utilizing the bias correction methods, and researching the model behavior due to small perturbations showed that our analysis was robust. Also, the main logistic regression assumptions and the model fit were analyzed and confirmed in Supplementary material 4.

Our study found that a history of infections (regardless of type) is associated with a 54% higher risk of AD+ in UKB participants without high exposure to TRAP. This broadly aligns with our previous findings of the associations between infections and AD risk in the Health and Retirement Study data linked to Medicare records (Ukraintseva et al., 2024). In that research, we demonstrated that infectious diseases of diverse origins (viral, bacterial, fungal) were all associated with increased AD risk (ranging from 16 to 42%). One potential explanation is that weakened immunity may play a more critical role in AD development than any particular microbe, by rendering individuals vulnerable to a broad spectrum of pathogens, which may in turn increase the burden of damage contributing to neurodegeneration. Another possibility is that each infection elicits inflammatory responses that may drive the risk of AD (Whitson et al., 2022).

Exposure to TRAP alone did not significantly impact AD+ risk in our analysis. However, among individuals with both high TRAP exposure and a history of infections, the risk of AD+ was approximately 70% higher than in those with a history of infections alone, suggesting a potential synergy between infections and TRAP. This finding aligns with the ‘multi-hit’ hypothesis of AD, which suggests that the presence of multiple risk factors (‘hits’), especially with a synergy between them, is required for AD to progress to its clinical manifestation (Gong et al., 2018; Patrick et al., 2019; Steele et al., 2022; Lathika Rajendrakumar et al., 2025). Several biological mechanisms could be responsible for the synergy between infections and TRAP: (i) Chronic exposure to TRAP may compromise the integrity of the blood–brain barrier (BBB) and increase the brain’s permeability to pathogens and immune cells activated by them. This can promote neuroinflammation and increase damage burden contributing to neurodegeneration (Perry et al., 2003; Patabendige and Janigro, 2023; Calderón-Garcidueñas et al., 2002, 2015). (ii) Infections can trigger immune response and neuroinflammation via cytokine release (Adamu et al., 2024). Exposure to TRAP could amplify this by activating microglia and astrocytes, prolonging inflammation, and accelerating neuronal damage (Block and Calderón-Garcidueñas, 2009). (iii) Exposure to TRAP, especially to PM2.5 and nitrogen oxides, can generate reactive oxygen species (ROS) that may damage mitochondria (Calderón-Garcidueñas et al., 2002, 2015; Mussalo et al., 2024). This may lead to a deficiency of energy required for a proper response to infections. (iv) Infections may stimulate Aβ production as a part of an antimicrobial defense (Soscia et al., 2010). Exposure to TRAP may further enhance Aβ production, e.g., through increased lipid oxidation (Cacciottolo et al., 2020). These and other possible mechanisms deserve further exploration and confirmation in biomedical research.

A notable finding of our study is that AD+ risk in participants with both a history of infections and high exposure to TRAP, compared to those without either factor, was substantially (349%) higher in non-carriers of APOE4, but it became non-significant in APOE4 carriers (Figure 1). One possible explanation could be that APOE4 is the strongest AD risk factor (besides age), whose AD-promoting effects may mask and outweigh those of the other risk factors. Indeed, the APOE4 has been linked to numerous AD-promoting features. It can directly contribute to cholesterol transport deficiency in the brain, resulting in poorer myelin synthesis and axon maintenance by oligodendrocytes (Blanchard et al., 2022). This could make neurons more vulnerable to damaging exposures. APOE4 may compromise BBB integrity (Montagne et al., 2020) and reduce Aβ clearance across the BBB, as well as by microglia, and promote a pro-inflammatory microglial phenotype (Liu et al., 2013; Shi et al., 2017). APOE4 can also stimulate ROS production and mitochondrial inefficiency, among its many other effects that may facilitate neurodegeneration (Mahley and Huang, 2009; Mahley, 2023).

We recognize several study limitations. In our analysis, we evaluated regression models using the Akaike information criterion. One should note that there is no universal procedure by which one can determine the “best model.” We applied the AIC approach calculating goodness-of-fit and model variability in order to select the most parsimonious regression model (Burnham and Anderson, 2002; Anderson, 2008; Burnham et al., 2011). The AIC approach only gave some rationales behind our analysis but it was neither the main argument, nor the only technical mean for uniquely inferring the shape of regression that we obtained in this study. Another potential limitation could be that the formal statistical association evaluated from regression analysis may not imply actual causality, which should be further studied using causal inference approaches. We also acknowledge that our results reflect association rather than causation, consistent with the observational nature of UKB data. There also remains a possibility that the observed relationship is partially attributable to unmeasured confounders. No occupational exposures (such as industrial pollutants or pathogens) were available for this paper. No primary care data was available too. So, severe infections may be over-represented. An additional limitation is that in the UKB, the participant’s residence distance to the nearest major road (DNMR) was collected only once at the time of entry, so the residential mobility during the follow-up may potentially influence results of the analysis. One more limitation is that the majority of UK Biobank participants (around 94.6%) are white born in the UK. In our study, taking ethnicities into account would result in some groups with only tens of subjects, which would not be enough for statistical analysis. So, ethnicity was not taken into account in our study. Because the number of AD+ events among APOE4 carriers is relatively limited, we recognize reduced statistical power in this group, which may limit biological interpretation. Due to sample size limitations, additional covariates available in UKB were not incorporated into the analysis. Moreover, while the use of a 50-meter distance to a major road is well supported by previous studies (e.g., Chen et al., 2017), it may still be useful to explore alternative DNMR thresholds (e.g., 75 or 100 meters) in future work. Finally, the UK Biobank is a volunteer-based study and may be prone to a volunteer bias (i.e., participants may be healthier and wealthier than general population), so the results of our analyses might not represent the entire UK population. Generalization of our findings to a more diverse population would require calculating sample weights to correct for the potential healthy volunteer bias. The existing UKB sample weights were, however, not available and applicable to our study because we worked with a selected subsample of the UKB participants. Calculating such weights for this sample is beyond the scope of this short communication.

One should also note that DNMR, which was used as a proxy for a chronically high exposure to TRAP and as an explanatory variable in our analysis, is an indicator of aggregated exposure to all road-related pollutants, not only to those specifically found in car exhaust fumes. Some of the pollutants that are not from car exhaust could be relevant to AD. E.g., a higher intensity traffic has been associated with the higher concentration of airborne fungi in urban air environments. Examples include Alternaria and Cladosporium species which may cause infection and inflammation potentially contributing to neurodegeneration (Muafa et al., 2024; Phuna and Madhavan, 2022; Alonso et al., 2017). The role of exposure to airborne fungi in AD pathology deserves separate investigation, especially in the light of our resent finding suggesting that the impact of recurrent fungal infections on AD risk can be even larger than that of other types of infections, including bacterial and viral (Ukraintseva et al., 2024). Other road-related pollutants, such as noise (The Lancet Regional Health-Europe, 2023), light pollution (Chepesiuk, 2009; Wyse et al., 2011; Aubrecht et al., 2013), and electromagnetic fields (Ahlbom and Feychting, 2003; Kıvrak et al., 2017) may also be relevant to health risks. For instance, noise is currently considered a health problem for citizens of the European Union (European Commission, 2023).

This study, which involved UK Biobank participants aged 60–75 years, found that chronically high exposure to TRAP significantly exacerbates the detrimental effects of infectious diseases on the risk of AD and other dementias in aging individuals. This finding aligns with the ‘multi-hit’ hypothesis of AD, which implies that the presence of multiple risk factors (‘hits’) is required for AD to progress to clinical onset. The largest relative increase in AD+ risk was seen in participants with a history of infections and exposure to TRAP, who were non-carriers of APOE4 variant. In presence of APOE4, the increase in AD+ risk caused by infections and exposure to TRAP became non-significant. One potential explanation for this observation is that APOE4, aside from age, is the strongest known risk factor for AD, so its AD-promoting effects outweigh and mask those of other risk factors. Since our results reflect association rather than causation, consistent with the observational nature of UKB data, they require confirmation in future research.