Peripheral GFAP and NfL as early biomarkers for dementia: longitudinal insights from the UK Biobank

14 May 2024

BMC Medicine

Share on XTweet

Peripheral GFAP and NfL as early biomarkers for dementia longitudinal
insights from the UK Biobank

Background Peripheral glial fibrillary acidic protein (GFAP) and neurofilament light chain (NfL) are sensitive markers of neuroinflammation and neuronal damage. Previous studies with highly selected participants have shown that peripheral GFAP and NfL lev

Demographics

The study design and primary outcomes are shown in Fig. 1. A total of 48,542 participants from the UK Biobank cohort has complete GFAP and NfL measurements assessed using Olink’s assay from blood samples obtained at recruitment (mean age: 56.8 ± 8.21 years) (average follow-up duration: 13.18 ± 2.42 years). Demographic details of the participants are presented in Table 1 and Additional file 1: Table S4. We identified 1360 all-cause dementia cases, of which 1312 were diagnosed after recruitment. Participants with dementia diagnosis before recruitment are relative younger than those diagnosed after recruitment (Additional file 1: Fig. S1). The mean duration to dementia diagnosis was 8.35 ± 4.09 years, with the longest duration being 15.21 years. The average age of a dementia diagnosis made after recruitment was 73.39 ± 6.16 years.

Table 1 Baseline demographic characteristics of participants enrolled in UK Biobank with GFAP and NfL expressions

Full size table

The distribution of peripheral GFAP and NfL within our study cohort is detailed in Additional file 1: Fig. S2 and Additional file 1: Table S5-S6. Aligning with previous findings, we observed that age, BMI, and racial background significantly influenced GFAP and NfL expression levels, and these levels varied in the presence of certain medical conditions including diabetes, hypertension, cerebrovascular disorders, demyelination, neurodegeneration, organic brain diseases, and mental health disorders. Intriguingly, our analysis revealed that individuals exhibiting higher levels of GFAP and NfL (falling within the highest quartile) were more likely to be carriers of the APOE*E4 allele.

In consideration of the observation that healthier individuals are more likely to complete repeated measurements, we categorized participants based on their follow-up protein measurement data (Additional file 1: Table S7). Notably, a vast majority, 97.68% (47,418 out of 48,542), had only baseline measurements of GFAP and NfL. Participants with subsequent protein measurements tended to be younger (mean age 50.2 years compared to 57.0 years, P P P

Higher levels of peripheral GFAP and NfL associated with cognition and dementia risk

Figure 2 delineates the relationship between peripheral GFAP and NfL levels and cognitive measurements, with demographic characteristics presented in Additional file 1: Table S4. The influence of GFAP and NfL on various cognitive dimensions is evident. Elevated GFAP levels were associated with poor fluid intelligence (model 1, estimate = − 0.017, 95% CI = − 0.027 to − 0.006), prolonged reaction times (model 1, estimate = 3.23, 95% CI = 1.12 to 5.34), and suboptimal pair-matching results (model 1, estimate = 0.01, 95% CI = 0.001 to 0.020). Elevated NfL levels significantly correlate with decreased numeric memory (model 3, estimate = − 0.015, 95% CI = − 0.028 to − 0.006) and prolonged reaction time (model 3, estimate = 3.57, 95% CI = 1.15 to 5.99). Using a global z-score for combining different domains of cognition, significant negative associations were found for both GFAP (model 3, estimate = − 0.021, 95% CI = − 0.04 to − 0.002) and NfL (model 3, estimate = − 0.026, 95% CI = − 0.045 to − 0.006) (Additional file 1: Fig. S3-S5).

To find the cutoff values for GFAP and NfL associated with an elevated risk of dementia, we grouped their expression by quartiles and found that interval 4 of both GFAP and NfL significantly increased the risk of developing dementia (Additional file 1: Table S8), and the cutoff values at this point were 0.363 for GFAP and 0.353 for NfL. Participants with GFAP levels exceeding 0.363 faced a significantly increased hazard of developing all-cause dementia (model 3, hazard ratio (HR) = 2.25; 95% CI = 1.96 to 2.58) (Fig. 3A). High GFAP levels were also associated with an increased risk of developing other types of dementia (HR = 3.02, 2.21, and 3.05 for ADRD, VD, and FTD, respectively) (Fig. 3A). Similarly, significant associations are noted for high expression of NfL (> 0.353) with the hazard of developing all cause-dementia, ADRD, and VD (HR = 1.98, 2.08, and 2.07, respectively) (Fig. 3B). High expression of NfL showed more risk in developing FTD (model 3, HR = 4.23; 95% CI = 2.32 to 7.72) (Fig. 3B). Taken together, higher peripheral GFAP and NfL levels are independent risk factors for cognitive decline and dementia in dementia-free participants.

Early elevation of peripheral GFAP and NfL levels shared genetic factors with AD

The participants were divided into three groups based on their dementia diagnosis: never (no dementia record in the cohort, N = 47,182), potential (dementia diagnosis made after recruitment, N = 1,360), and existing (already diagnosed with dementia at recruitment, N = 48). We noted significant elevations in GFAP and NfL NPX levels when comparing participants in the never group to those in the existing group; importantly, participants in the potential group demonstrated GFAP and NfL NPX levels similar to those in the existing group (Additional file 1: Fig. S6). After matching for age and sex using the propensity score matching method at a ratio of 1:10, significant differences remained (Additional file 1: Fig. S7). In accordance with the results shown in Fig. 2, cognitive decline was observed in the potential group (Additional file 1: Fig. S8).

Previous studies have suggested that the increase in GFAP in the pre-clinical phase of AD is independent of classical AD pathology [39, 40]. We therefore used two genetic tools for AD, the AD-GRS and the number of APOE*E4 alleles carrying as independent variables, to explore the association between genetic of dementia and GFAP or NfL. We found a significant association between the AD-GRS and GFAP (estimate = 0.03, 95% CI = 0.025 to 0.034) and NfL (estimate = 0.011, 95% CI = 0.007 to 0.015) (Fig. 4A). We also found a significant association between the number of APOE*E4 alleles and GFAP (estimate = 0.064, 95% CI = 0.055 to 0.074) and NfL (estimate = 0.015, 95% CI = 0.007 to 0.024) (Fig. 4B). For proteins implicated in the pathogenesis of AD and dementia, including amyloid beta precursor protein binding family B member 1 (APBB1IP), amyloid beta precursor like protein 1 (APLP1), amyloid beta precursor protein (APP), and microtubule associated protein tau (MAPT), only trends without statistical significance were observed in relation to AD-GRS. Similar results were found when using APOE*E4 alleles as genetic tools, except for APLP1 which showed a negative association. Chitinase 3 Like 1 (CHI3L1), previously reported to be a differentially expressed protein in AD and dementia, also showed no association with the genetic tools for AD. In addition, no associations were observed between AD-GRS or APOE*E4 alleles and several housekeeping proteins used as controls since they are often stably expressed in different physiological or pathological conditions. Taken together, these results suggest that early neuroinflammation and neuronal damage, reflected by elevated peripheral GFAP and NfL expression, may be genetically determined in patients at high risk of AD.

Association between the longitudinal changes of peripheral GFAP and NfL with dementia progression

For participants who later manifested dementia, the time to diagnosis was significantly associated with GFAP and NfL levels, but this may be a false positive result due to the large sample size (Additional file 1: Fig. S9). We then used a backward time scale to observe the trajectory of GFAP and NfL levels prior to the onset of dementia (Additional file 1: Fig. S10-S11). In line with previous findings, peripheral GFAP and NfL are increased with advancing age. Pearson correlation analyses also reveal positive correlation between GFAP (r = 0.377, P r = 0.501, P 1: Table S9). By setting the time of dementia diagnosis (for those with dementia) or time of end of follow-up (for those without dementia) as 0-time mark, we discerned notable disparities in the expression levels of GFAP and NfL up to 15 years before diagnosis.

It should be noted that the 97.7% (47,418 of 48,542) of participants in this study had a single protein measurement (Additional file 1: Table S7). The remain 1124 participants with repeated protein measurements were healthier in respect of the risk factors for dementia, including younger age, smaller BMI, higher education level, fewer smoking status, and less comorbidities, and they presented superior cognitive performances. This subset provided a unique opportunity to examine the relationship between the accumulation rates of peripheral GFAP and NfL and the rate of cognitive decline preceding dementia and early cognitive impairment. We found accelerated annual change in GFAP was significantly associated with more rapid cognitive decline (model 3, estimate = − 0.101, 95% CI = 0.163 to − 0.040), and the interaction between GFAP expression and follow-up time was not statistically significant. However, we did not observe associations between accumulation of NfL with global cognitive decline (Additional file 1: Table S10-S11).

These results suggest that peripheral GFAP and NfL levels, as well as the accumulation rate of GFAP, are associated with the progression of dementia, further emphasizing the potential of early interventions targeting neuroinflammation in dementia.

High predictive value by using baseline GFAP and NfL NPX for dementia

We evaluated the predictive value of baseline GFAP and NfL levels for dementia in dementia-free participants with leave-one-region out validation, as shown in Fig. 5 and Additional file 1: Table S12. Using GFAP and NfL alone achieved AUC in predicting dementia of 0.781 to 0.816, with corresponding C-Index of 0.792 to 0.829. Compared to using age alone as the predictor for dementia, incorporating GFAP and NfL improved the NRI for predicting all-cause dementia, ADRD, and VD (NRI = 0.012 to 0.088). Furthermore, the predictive values were higher when using GFAP and NfL alone compared to CAIDE model. When adding GFAP and NfL to CAIDE and DRSm models, the predictive value significantly improved in predicting all-cause dementia and ADRD (NRI = 0.128 to 0.173). The best model for predicting all-cause dementia, ADRD, and VD was DRSm combined with GFAP and NfL, with corresponding AUC of 0.867, 0.892, and 0.892 respectively, and corresponding C-Index of 0.871, 0.904, and 0.910. The combination of GFAP and NfL with established models did not improve the efficacy in predicting FTD (NRI = − 0.0002).

Sensitivity analysis

We performed subgroup analyses for age, sex, BMI, self-reported racial background, AD-GRS, and number of APOE*E4 alleles for incident all-cause dementia (Figs. 6A and 7A), and we found the results were consistent with our primary findings (Fig. 3) (except for NfL that did not show significance in non-White participants), indicating the increment of both GFAP and NfL across different subgroups suggests the risk of incident dementia. Interestingly, upon stratification by AD-GRS and number of APOE*E4 alleles, we noticed that GFAP demonstrated stronger correlation in the high-risk participants, while NfL showed stronger correlation in the low-risk participants.

When examining subgroups with distinct comorbidities, we observed that elevated GFAP levels heightened the risk of dementia among individuals with organic brain disorders. Although similar trends were seen in those with neurodegenerative disorders, demyelination, and cerebrovascular disorders, the associations did not reach statistical significance (Fig. 6B). Conversely, increased NfL levels were significantly associated with a higher risk of dementia in cases of demyelination and cerebrovascular disorders (Fig. 7B). These patterns may be attributed to the non-specific nature of GFAP and NfL, which broadly indicate neuroinflammatory and neurodegenerative changes. Nonetheless, they appear to mirror the shared pathological shifts accompanying dementia comorbid with certain neurological conditions. We then strictly included dementia-free participants who did not have comorbidities that were mentioned in our model 3 (Figs. 6C and 7C), and we found the results of these analyses were consistent with our primary findings.

In conducting a competing risk analysis, where non-dementia deaths were treated as competing events, we observed a slightly reduced HR for dementia in relation to GFAP, while an increase in HR was noted for NfL (Additional file 1: Table S13). Acknowledging that missing data related to BMI, lifestyle factors, and genetic risks could potentially skew our results, we carried out imputations for these variables and re-analyzed our dataset. The re-evaluation confirmed the robustness of the dementia risks associated with GFAP and NfL (Additional file 1: Table S14).

Furthermore, we refined our mixed model by integrating follow-up durations as random slopes, which reinforced the finding that higher rates of GFAP accumulation were significantly linked to accelerated cognitive decline (Additional file 1: Table S10).

Finally, we stratified our participants into two age groups (1: Table S15-S16). Overall, we found that GFAP and NfL provided higher predictive values for dementia in participants 1: Table S16).