One hundred thirty-three cognitively normal older adults participated in the study (65 ± 5.9 years; range: 54–76 years; 81 females). They were recruited from senior citizen’s associations, health-screening programs, and hospital outpatient services. All of them underwent neurological and neuropsychological assessment to discard both the presence of dementia and objective cognitive impairment. Individuals with medical conditions that affect brain structure or function (e.g., cerebrovascular disease, epilepsy, head trauma, history of neurodevelopmental disease, alcohol abuse, hydrocephalus, and/or intracranial mass) were not included in the study. Participants met the following criteria: (i) normal global cognitive status in the Mini-Mental State Examination (scores ≥ 26); (ii) normal cognitive performance in the neuropsychological tests relative to appropriate reference values for age and education level; (iii) global score of 0 (no dementia) in the Clinical Dementia Rating; (iv) functional independence as assessed by the Spanish version of the Interview for Deterioration in Daily Living Activities (Böhm et al., 1998); (v) scores ≤5 (no depression) in the short form of the Geriatric Depression Scale (Sheikh and Yesavage, 1986); and (vi) not be taking medications that affected cognition, sleep, renal or hepatic function. All participants gave informed consent to the experimental protocol approved by the Ethical Committee for Clinical Research of the Junta de Andalucía according to the principles outlined in the Declaration of Helsinki. Table 1 contains sample characteristics.

All participants received neuropsychological assessment. They were administered with the following tests: the Spanish version of the Memory Binding Test (MBT) (Gramunt et al., 2016); the short form of the Boston Naming Test (BNT); the semantic and phonological fluency tests based on the “Animal” and letter “P” naming tasks; the two forms of the Trail Making Test (TMT-A and TMT-B); and the Tower of London (TL). All neuropsychological scores were z transformed. In the case of TMT-A and TMT-B, we used the inverse z-values, while in the case of the MBT and TL we first computed a composite measure by summing the z-scores of the different scores. We then applied principal component analysis to obtain the Spearman’s “g” factor as an index of global cognitive function. This analysis was done with R Statistical Software v3.0.1 (R Foundation for Statistical Computing, Vienna, Austria) using the prcomp function. We only retained the first component (eigenvalue 3.4), which explained 53.7% of variance in the data, due to the contribution of MBT (24.2), BNT (24.2), phonological fluency (24.2), semantic fluency (3.2), TMT-A (8.6), TMT-B (9.5), and TL (6.0). The standardized residuals were used to obtain the latent variable g (factor loading MBT: 0.92; BNT: 0.92; phonological fluency: 0.92; semantic fluency: 0.33; TMT-A: 0.55; TMT-B: 0.57; TL: 0.46).

Fasting blood samples were taken at 9:00-10:00 AM in all participants to control for potential circadian effects. Briefly, venous blood samples were collected in 10 ml dipotassium ethylene diamine tetraacetic acid (EDTA) coated tubes (BD Diagnostics), and immediately centrifuged (1,989 g) at 4°C for 5 min. Supernatant plasma was aliquoted into polypropylene tubes containing 300 μl of plasma mixed with 10 μl of a protease inhibitor cocktail (cOmplete Ultra Tablets mini, Roche), and stored at –80°C until analysis. Plasma samples used in the present study were not previously thawed.

Plasma Aβ_1–42 and NfL levels were measured on the ultra-sensitive single-molecule array (Simoa) HD-1 analyzer platform (Quanterix, MA, United States) following the manufacturer’s instructions. The Aβ_1–42 Simoa 2.0 assay (Cat. No. 101664) and the NF-light Simoa assay advantage kits (Cat. No. 103186) were purchased from Quanterix. These assays measure Aβ_1–42 and NfL levels in human plasma with a detection limit of 0.044 and 0.038 pg/ml, respectively. Two quality control samples were run on each plate for each analyte. Plasma Aβ_1–42 and NfL determinations were run in duplicates, and the average of the two measurements (pg/ml) was used for statistical analysis. Samples with coefficients of variation higher than 20% were excluded.

Structural and functional magnetic resonance imaging (MRI) scans were performed on a 3T Philips Ingenia MRI scanner, equipped with a 32-channel radio-frequency (RF) receive head coil and body RF transmit coil (Philips, Best, Netherlands). The following MRI sequences were acquired in the same session: (i) 3D T1-weighted (T1w) magnetization prepared rapid gradient echo (MPRAGE) in the sagittal plane: repetition time (TR)/echo time (TE) = 2,600 ms/4.7 ms, flip angle (FA) = 9°, acquisition matrix = 384 × 384, voxel resolution = 0.65 mm³ isotropic, resulting in 282 slices without gap between adjacent slices; (ii) 3D T2w scan in the sagittal plane: TR/TE: 2,500 ms/251 ms, FA = 90°, acquisition matrix = 384 mm × 384 mm, voxel resolution = 0.65 mm³ isotropic, resulting in 282 slices without gap between adjacent slices; and (iii) T2w Fast Field Echo images using a blood-oxygen-level-dependent (BOLD) sensitive single-shot echo-planar imaging (EPI) sequence in the axial plane: TR/TE: 2,000 ms/30 ms, FA = 80°, acquisition matrix = 80 mm × 80 mm, voxel resolution = 3 mm³ isotropic, resulting in 35 slices with 1 mm of gap between adjacent slices. To allow for optimal B1 shimming, a B1 calibration scan was applied before starting the EPI sequence. We acquired 250 EPI scans preceded by four dummy volumes to allow time for equilibrium in the spin excitation. Before starting the acquisition of the EPI sequence, participants were asked to remain still and keep their eyes closed without falling sleep. Pulse and respiratory rates were recorded using the scanner’s built-in pulse oximeter placed on the left-hand index finger and a pneumatic respiratory belt strapped around the upper abdomen, respectively. Brain images were visually examined after each MRI sequence; they were repeated if artifacts were clearly identified. All participants underwent the same MRI protocol at the MRI facility of Pablo de Olavide University.

T1w scans were preprocessed using Freesurfer v6.0.¹ The Freesurfer’s pipeline included brain extraction, intensity normalization, automated tissue segmentation, generation of white matter (WM) and pial surfaces, correction of surface topology and inflation, co-registration, and projection of cortical surfaces to a sphere for establishing a surface-based coordinate system (Fischl and Dale, 2000). Pial surface misplacements and erroneous WM segmentation were manually corrected on a slice-by-slice basis by one experienced technician. The T2w image was registered to the T1w image with bbregister employing a trilinear interpolation method. bbregister is a within-subject, cross-modal registration program that uses a boundary-based cost function constrained to 6 degrees of freedom (Greve and Fischl, 2009).

We have used the T1w/T2w ratio image to indirectly estimate the relative intracortical myelin content in individual cortical surfaces. Previous studies have shown that regional variations in the T1w/T2w ratio match the myelin content derived from histologically-defined cortical areas (Glasser and Van Essen, 2011) and they further correlate with cortical myelination patterns across the lifespan (Grydeland et al., 2019). This is partially due because myelin alters the signal intensity of T1w and T2w images in opposite directions and, consequently, the T1w/T2w ratio provides both enhanced tissue contrast and sensitivity to brain myelin content (Uddin et al., 2019). Furthermore, the T1w/T2w ratio cancels radiofrequency (RF) receive field (B1−) artifacts in the absence of head motion, and corrects reasonably well for RF transmit field (B1+) effects when the participants’ head is precisely located at the isocenter of the magnetic field (Glasser et al., in press). In the present study, the latter requirements were strictly met to reduce transmit field biases and spurious results in cortical myelin maps.

The tissue fraction effect was corrected in individual T1w/T2w ratio images using PETsurfer (Greve et al., 2016), setting the point spread function estimate to zero. Next, three intermediate surfaces were generated within the cortical GM at fixed relative distances between the WM and pial surfaces (Polimeni et al., 2010). These surfaces were located at 90% (outer layer), 50% (medium layer) and 10% (inner layer) of the cortical thickness away from the WM surface. Finally, individual myelin maps obtained at different cortical depths were projected onto the average cortical surface, transformed to z-scores, and smoothed using non-linear spherical wavelet-based de-noising schemes (Bernal-Rusiel et al., 2008). All processing steps were visually checked for quality assurance. Figure 1 illustrates the analysis pipeline for intracortical myelin mapping.

Resting-state functional magnetic resonance imaging (rs-fMRI) data were preprocessed using AFNI functions,² version AFNI_20.3.01. For each participant, high-frequency spikes were eliminated (3dDespike), time-locked cardiac and respiratory motion artifacts on cerebral BOLD signals were minimized using RETROICOR (Glover et al., 2000), time differences in slice-acquisition were corrected (3dTshift), EPI scans were aligned using rigid body motion correction and selecting the first volume as reference (3dVolreg), and aligned EPI scans were co-registered to their corresponding T1w volumes (3dAllineate; cost function: lpc + ZZ).

Dynamics were removed provided that more than 5% of voxels exhibited signal intensities that deviated from the median absolute deviation of time series (3dToutcount); and/or when the Euclidean norm (enorm) threshold exceeded 0.3 mm in head motion. None of the participants showed more than 20% of artifactual dynamics after applying censoring. Simultaneous regression was further applied to minimize the impact of non-neuronal fluctuations on the rs-fMRI signal (3dTproject). Nuisance regressors included: (i) six head motion parameters (3 translational and 3 rotational) derived from the EPI scan alignment along with their first-order derivatives, (ii) time series of mean total WM/CSF signal intensity, and (iii) cardiac (measured by pulse oximeter) and respiratory fluctuations plus their derivatives to mitigate effects of extracerebral physiological artifacts on cerebral BOLD signals. No temporal band-pass filtering was applied.

Preprocessed rs-fMRI scans were projected onto the fsaverage5 cortical surface space. Seeds for FC analyses were derived from cortical regions showing significant associations between plasma measurements (i.e., Aβ_1–42 and NfL levels) and intracortical myelin content. Surface-based rs-FC seed to whole cortex maps were obtained using the Fisher’s z-transform of the corresponding Pearson’s correlation coefficients.

To estimate the sample size, we performed power analysis with the G*Power software (v3.1.9.6).³ Only two studies have previously assessed the relationship between Aβ_1–42 and intracortical myelin content, one in the AD continuum using both CSF Aβ_1–42 and amyloid PET (Luo et al., 2019) and other in cognitively normal older adults using amyloid PET (Yasuno et al., 2017). To our knowledge, research investigating associations between NfL levels and intracortical myelin content or addressing the potential moderating effect of plasma measurements (i.e., Aβ_1–42 or NfL) on the relationship between intracortical myelin content and cortical rs-FC patterns is lacking. Therefore, we computed an a priori (prospective) power analysis (fixed model, R² deviation from zero) to achieve statistical power of 80% given a 0.025 two-sided significance level and an overall Cohen’s effect size (f²) ranging from 0.05 to 0.2. Supplementary Figure 1 shows the results of this analysis for additive models evaluating the relationship between one of the plasma measurements (either Aβ_1–42 or NfL) and intracortical myelin content (3 predictors), and for interactive models assessing the moderating effect of one of the aforementioned plasma measurements on the relationship between intracortical myelin and rs-FC (5 predictors). This analysis revealed that 263 and 308 participants are required to detect an overall effect size of 0.05 for additive and interactive models, respectively. As the study sample consisted of 133 participants, interpretation of significances is conditional on the existence of overall effect sizes equal to or greater than 0.10 and 0.12 for main and interaction effects, respectively.

We first applied the frequentist approach to determine whether age was related to plasma measurements and intracortical myelin after adjusting for the effects of sex. Next, by applying the same approach, we performed vertex-wise multiple linear regression analyses to evaluate the relationship of plasma measurements (either Aβ_1–42 or NfL) with intracortical myelin content at different cortical depths (90, 50, and 10% away from the WM surface), including age and sex as covariates of no interest. To assess whether associations between plasma measurements and intracortical myelin content differed with cortical depth, we first computed the mean of the sum of all significant regions at the three cortical depths to extract the fitted values. These fitted values were entered into a mixed model with subjects as random effect to determine whether the strength of these associations varied significantly as a function of cortical depth. The effect of the moderator factor (i.e., cortical depth) was evaluated through an ANOVA that compared the null model only including the intercept with the experimental model including cortical depth as a fixed effect. If differences reached significance, post hoc comparisons were performed with the R package emmeans and multiple comparisons were adjusted using the Bonferroni correction.

Using the frequentist approach, we further evaluated whether cortical regions showing significant associations with plasma measurements also exhibited altered patterns of rs-FC. For this, we applied a vertex-wise multiple regression model that included as regressor of interest the interaction between plasma measurements (either Aβ_1–42 or NfL) and the mean intracortical myelin of the region used as seed in the rs-FC analysis. These models were adjusted by age and sex.

Vertex-wise regression analyses for intracortical myelin and rs-FC were performed using the SurfStat package.⁴ Results were corrected for multiple comparisons using a hierarchical statistical model that first controls the family-wise error rate at the level of clusters by applying random field theory over smoothed statistical maps (p_vertex < 0.001, p_cluster < 0.05), and next controls the false discovery rate at the level of vertex within each cluster (p < 0.05) over unsmoothed statistical maps (Bernal-Rusiel et al., 2010). The anatomical location of significant results was based on vertices showing the maximum statistic within each significant cluster (Desikan et al., 2006).

After inferential evidence of a main or an interaction effect, we computed the standardized measure of effect size (i.e., Cohen’s f²) to evaluate local effect size within the context of a multivariate regression model (Cohen, 1988). To establish the precision of standardized effect sizes, we computed 95% confidence intervals (CI_95%) using the normal approximated interval with bootstrapped bias and standard error (N = 10,000 bootstrap samples) with the Matlab’s bootci function.

Next, for each cortical vertex showing the maximum statistic within each significant cluster, we applied Bayesian linear regression analyses using JASP, version 0.12.2.⁵ The Bayesian approach allowed us to quantify the evidence for the alternative hypothesis and to overcome the problem of multiple comparisons resulting from performing different models for two hemispheres and three cortical depths. Bayesian linear regression analyses were based on non-informative priors such as the Jeffreys-Zellner-Siow prior with an r scale of 0.354 (Liang et al. (2008). We compared the strength of the Bayes factor for the model including all covariates of no interest (null model) with the model including the predictor of interest (experimental model) (BF₁₀). The classification scheme proposed by Lee and Wagenmakers (2013) was employed to interpret the BF₁₀. We only reported those results that reached significance according to the frequentist approach as long as the evidence in favor of the alternative hypothesis was at least moderate (BF₁₀ ≥ 3).

Finally, we evaluated whether associations of intracortical myelin content with either plasma Aβ_1–42/NfL levels or rs-FC accounted for the variability in global cognitive function. For this purpose, we first applied the Yeo-Johnson transformation to the cognitive index (i.e., the latent variable g) to reduce the detrimental effects of skewedness and heteroscedasticity in the different models (Yeo and Johnson, 2000). Next, we built four models to assess main and interaction effects. The first model only included the covariates of no interest; the second model the main effects; the third model the two-way interactions between the different predictors of interest; and the fourth model the three-way interaction. We first calculated F-tests between the different sequential models, and then compared the relative strength of the Bayes factor between the models.

Age was positively related to plasma Aβ_1–42 (F₁,₁₃₀ = 9.8, p = 0.002) and NfL levels (F₁,₁₃₀ = 28.8, p < 10^–6). On the contrary, age was negatively associated with intracortical myelin content in postcentral regions of the left hemisphere (F₁,₁₃₀ = 21.2, p < 10^–5) and superior parietal lobe of the right hemisphere (F₁,₁₃₀ = 21.3, p < 10^–5). These results are illustrated in Supplementary Figure 2.

Figure 2 displays intracortical myelin maps projected onto the average cortical surface obtained at different depths (i.e., outer, medium and inner layers) before z-transformation. As expected, T1w/T2w ratio values increased with cortical depth (Nieuwenhuys, 2013; Nave and Werner, 2014). Our T1w/T2w ratio maps differ from those reported in Glasser and Van Essen (2011), especially in some regions of the occipital cortex, whose intracortical myelin content was associated with plasma levels of Aβ_1–42 (see Figure 3). These differences may be due to the distinct approaches used to obtain the intracortical myelin maps in the two studies. Thus, we used absolute values comprising the entire range of T1w/T2w ratio values instead of a percentile scaling between 3 and 96%, as in Glasser and Van Essen (2011). Neither we adjusted the color palette to identify the transitions between adjacent areas, since we were not interested in delineating spatial gradients or myeloarchitectural features of cortical regions (Glasser and Van Essen, 2011).

Plasma levels of Aβ_1–42 were positively associated with T1w/T2w ratio values in temporo-parietal-occipital regions of both hemispheres at the three cortical depths, as illustrated in Figure 3 and reported in Table 2. The evidence in favor of the alternative hypothesis was extreme (BF₁₀ > 100) for myelin content of inner and outer layers, and moderate in the medium layer (BF₁₀ was reported in Table 2).

Regarding the association between Aβ_1–42 and intracortical myelin, mixed model ANOVAs revealed a significant main effect of cortical depth for the right hemisphere (χ² = 346.6, p < 10^–15) that was not extended to the left hemisphere (χ² = 5.7, p = 0.06). Post hoc analyses showed that the strength of positive associations in the right hemisphere decreased with cortical depth, being stronger for the outer than for the other two layers (p < 0.0001), and stronger for the medium than for the innermost layer (p < 0.0001). Figures 4A,B illustrate these results.

The regression analysis also revealed a negative association between plasma NfL levels and T1w/T2w ratio values of the right insular cortex (Figure 5 and Table 2). This result was restricted to the inner layer, where the Bayesian linear regression analysis showed a BF₁₀ of 3.4, indicating that the alternative hypothesis is about three times more likely than the null, which is classified as moderate. The mixed model ANOVA addressing the effect of cortical depth on the strength of this association revealed a significant main effect in the right insular cortex (χ² = 354.9, p < 10^–15). The association of NfL with intracortical myelin was more negative for the innermost layer than for the other two layers (p < 0.0001). Figure 4C illustrates these results.

We next performed rs-FC analysis using as seeds those regions derived from significant associations between plasma measurements (Aβ_1–42 and NfL) and myelin content (regions displayed in Figures 3, 5). The interaction between plasma Aβ_1–42 and intracortical myelin did not predict rs-FC for none of the seeds evaluated. Conversely, the multiple regression analysis performed on the rs-FC map with the right insula as seed revealed a significant two-way interaction in the left medial orbitofrontal cortex (Rmax2 = 0.13, t_max = 4.36, p_cluster = 0.006, rho = –2.22, CI_95%[–2.81 to –1.63]). Figure 6A shows the spatial location of this result. Post hoc analyses shown in Figure 6B revealed a negative relationship between FC and myelin in individuals with higher plasma levels of NfL (i.e., the group with + 1SD) and a positive relationship in those with lower NfL levels (i.e., the group with –1SD). The Bayesian analysis yielded a BF₁₀ of 304, and therefore the evidence in favor of the alternative hypothesis is considered as extreme.

The ANOVAs performed to compare the sequential models revealed no significant main or interaction effects on cognition. BF₁₀ ranged between 0.33 and 1. Accordingly, no evidence in favor of the null hypothesis was observed in either case.

Previous studies have shown that low plasma Aβ_1–42 concentrations appear in MCI/AD patients carrying the Apolipoprotein E ε4 allele (APOE4) (Kleinschmidt et al., 2016) and in cognitively normal individuals showing abnormal CSF-amyloid status and positive amyloid PET scans (Verberk et al., 2018). Moreover, plasma Aβ_1–42 levels have been negatively associated with WM hyperintensities and positively with global cognition, memory performance, hippocampal volume and cortical thickness of the temporal lobe (Llado-Saz et al., 2015; Poljak et al., 2016; Hilal et al., 2018), as well as with a steeper rate of subsequent cognitive decline (Rembach et al., 2014; Poljak et al., 2016; Verberk et al., 2020) and increased risk of AD (Chouraki et al., 2015; Hilal et al., 2018; de Wolf et al., 2020).

Our results revealed a positive relationship between plasma Aβ_1–42 levels and intracortical myelin in cognitively normal older adults, linking plasma amyloid markers to cortical myelin integrity in normal aging. The association between cerebral amyloid aggregation and myelin deficits has been mainly supported by research in mouse models of AD (Desai et al., 2010; Mitew et al., 2010; Schmued et al., 2013; Chen et al., 2021). These studies showed that myelin pathology extends across regions with Aβ deposits (Mitew et al., 2010; Schmued et al., 2013), suggesting that the absence of oligodendrocytes in the plaque core may contribute to the destabilization of neuronal networks surrounding Aβ deposits (Palop and Mucke, 2010). This interpretation has been reinforced by research showing in vivo associations between myelin alterations and amyloid pathology in preclinical AD (Dean et al., 2017; Luo et al., 2019). In particular, lower CSF Aβ_1–42 levels were found to be associated with lower myelin water fraction (MWF) in the WM of late myelinating brain regions (Dean et al., 2017), which are affected by AD lesions from the earliest stage (Braak and Del Tredici, 2015). Further evidence indicates that preclinical AD patients exhibit the lowest T1w/T2w ratio in the inferior parietal lobe, which continues to decrease with disease progression (Luo et al., 2019). Our results complement these findings, providing evidence that positive associations of intracortical myelin with plasmatic Aβ_1–42 are restricted to posterior cortical regions in normal aging. Together, these results suggest that myelin breakdown in aging and preclinical AD occurs in the reverse direction in which myelination develops, from cortical association regions to subcortical WM projection tracts, and from posterior to anterior brain regions (Bartzokis, 2004).

An alternative interpretation is that the load of WM lesions, instead of Aβ deposition in plaques, could drive the relationship between plasma Aβ_1–42 and intracortical myelin content. In line with this hypothesis, it has been reported that associations between CSF Aβ_1–42 levels and cortical atrophy are noticeable in dementia patients but absent in cognitively normal older adults, whereas correlations between CSF Aβ_1–42 and WM lesions are present in normal elderly individuals but lacked in dementia patients (Skoog et al., 2018). Further, if associations between plasma Aβ_1–42 and intracortical myelin content were a reflection of the cerebral amyloid burden, this relationship should be stronger in deeper cortical layers, where the myelin content is higher (Hughes et al., 2018). This prediction is supported by evidence indicating that Aβ staining in 5×FAD transgenic mice is more intense in deeper than in superficial cortical layers. As a result, T1 and T2 values were more drastically reduced in deeper cortical regions in transgenic mice compared to wild-type mice (Spencer et al., 2013). Moreover, analysis of the laminar distribution of transmitter receptors in the human neocortex has shown that deeper layers of temporo-occipital regions concentrate a high density of kainate receptors (Palomero-Gallagher and Zilles, 2019). Given that oligodendrocytes contribute to the control of extracellular glutamate levels (Matute et al., 2001) and oligodendrogenesis is mostly impaired in the deeper cortical layers (Orthmann-Murphy et al., 2020; Xu et al., 2021), alterations in glutamate homeostasis may result in overactivation of kainate receptors and subsequent excitotoxic oligodendroglial death, especially in the innermost layers. However, our results showed that associations between plasma Aβ_1–42 and intracortical were stronger in the outer cortical layer (at least in the right hemisphere), suggesting that plasma Aβ_1–42-myelin associations are not likely driven by cerebral amyloid burden.

It is also worth noting that T1 relaxation rate correlates with iron concentration in the brain, showing much higher iron relaxivity values in the cortex than in other brain structures (Ogg and Steen, 1998). Accordingly, high-field MRI data from postmortem brain samples has revealed a layered organization of iron content in the cortical mantle suggestive of each region’s myeloarchitecture (Fukunaga et al., 2010). Therefore, we cannot rule out that associations between plasma measurements and intracortical myelin content are partially due to iron accumulation. Future experiments should use more sophisticated in vivo approaches to separate iron and myelin content in the human cortex (Shin et al., 2021) in order to reliably determine their specific effects on aging and aging-related neurodegenerative conditions.

Neurofilament light chain is one of the scaffolding proteins of the axonal cytoskeleton with important roles in axonal and dendritic branching (Yuan et al., 2017). NfL is particularly abundant in myelinated axons and is released to the periphery upon damage of the central nervous system (Gafson et al., 2020). Although plasma NfL is not a specific AD biomarker, it has been positively associated with higher risk of cognitive decline in cognitively normal older adults, MCI and AD patients (Mattsson et al., 2017; Aschenbrenner et al., 2020; Lee et al., 2022) and higher CSF NfL concentration in the early presymptomatic stages of familial AD (Preische et al., 2019) and in all stages of sporadic AD (Mattsson et al., 2017). At baseline, it has been positively related to neocortical Aβ and tau in cognitively unimpaired and impaired older adults, respectively (Chatterjee et al., 2018; Benedet et al., 2020) and negatively to hippocampal atrophy in all stages of AD (Mattsson et al., 2017; Hu et al., 2019) and to the mean cortical thickness of the whole-brain and specific brain regions in MCI and cognitively normal older adults (Lee et al., 2022). Over time, decreasing plasma NfL levels have also been related to progressive enlargement of the lateral ventricles and decreasing brain metabolism, hippocampal volume and cortical thickness especially in MCI and AD patients (Mattsson et al., 2017; Benedet et al., 2020). Consequently, plasma NfL is considered a promising and cost-effective biomarker of axonal injury in AD and a variety of neurological conditions (Khalil et al., 2018).

High blood NfL levels may further reflect myelin breakdown, presumably due to dysfunctional oligodendrocytes, which provide trophic support to axons (Mot et al., 2018). Multiple lines of evidence indirectly support this hypothesis in humans. For instance, recent experiments have revealed associations between high serum NfL levels and MRI markers of myelin damage in multiple sclerosis (Yik et al., 2022), pediatric acquired demyelinating syndrome (Simone et al., 2021), and X-linked adrenoleukodystrophy (Weinhofer et al., 2021). Our results add further support to this hypothesis, showing a negative relationship between plasma NfL levels and intracortical myelin content of the right insular cortex, a lightly myelinated cortical region that has shown increased vulnerability in aging (Hu et al., 2014), preclinical (Cantero et al., 2020) and prodromal AD (Cantero et al., 2017).

We also showed that plasma NfL levels modulated patterns of FC between the insula and medial orbitofrontal cortex, two lightly myelinated cortical areas that appeared to be vulnerable to detrimental effects of aging (Bartzokis, 2011; Vidal-Piñeiro et al., 2016). Recent evidence has linked rs-FC patterns of insular cortex and orbitofrontal cortex to apathy (Jang et al., 2021), a multi-domain syndrome associated with poor outcomes and incident dementia in normal older adults (van Dalen et al., 2018), suggesting that impaired FC of these two regions may have unfavorable implications for aging. Tract tracing studies in primates have revealed that the anterior insula has prominent connections to the orbitofrontal cortex (Flynn et al., 1999), forming a functional unit that serves in the integration of complex autonomic, cognitive and emotional processes (Morel et al., 2013), all of them manifestly affected by aging (Salthouse, 2012). We speculate that as long as plasma NfL levels are elevated, reduced myelin content of the insular cortex may affect synchronized timing of neuronal impulses leading to abnormal functional connections with the medial orbitofrontal cortex.

Results of the present study were cross-sectionally obtained, which has strengths and weaknesses that should be acknowledged. While cross-sectional studies are valuable for establishing preliminary evidence in planning advanced studies, they are not helpful for drawing predictive conclusions. As long as the study sample is representative of the overall population, cross-sectional studies are adequate for measuring the prevalence of health outcomes, understand determinants of health, and describe features of a population (Wang and Cheng, 2020). Therefore, our results cannot establish a true cause and effect relationship between T1w/T2w ratio values and plasma levels of Aβ_1–42 and NfL, nor can they determine a temporal relationship between outcomes and risk factors. However, they could be used to generate predictive hypothesis for longitudinal studies in which T1w/T2w ratio values are used as a predictive or dependent variable.

Linking our results to AD requires caution. Brain-specific proteins are present in much lower concentrations in blood than in CSF because the blood-brain barrier prevents free passage of molecules between the CNS and blood compartments (Zetterberg and Burnham, 2019). More importantly, Aβ species are expressed in non-brain tissues and bind to a variety of blood proteins (Marcello et al., 2009), which may not reflect brain Aβ turnover/metabolism and consequently reduce the potential for monitoring Aβ pathology in the blood. Moreover, the plasma Aβ_1–42/Aβ_1–40 ratio has shown better concordance with cerebral amyloid load than plasma Aβ_1–42 alone (Janelidze et al., 2016; Schindler et al., 2019), suggesting that the Aβ_1–42/Aβ_1–40 ratio has an added value as a pre-screening biomarker in AD.