A potential multimodal biomarker – cognitive signature associated with the conversion from subjective cognitive decline to mild cognitive impairment

Abstract

INTRODUCTION

The disruption of key mechanisms involved in amyloid beta (Aβ) clearance during the early stages of dementia may contribute to the progression of cognitive decline toward irreversible brain damage. In this study, we investigated multiple immune‐related pathways implicated in the management and clearance of Aβ within circulating extracellular vesicles (cEVs) and serum from individuals with subjective cognitive decline (SCD) who later progressed to mild cognitive impairment (MCI).

METHODS

A cytokine panel and the levels of Aβ_1–42 were quantified in both cEVs and serum from a longitudinally followed cohort of elderly with SCD, using mesoscale and Luminex technologies. We investigated associations with Aβ burden, cognitive performance, APOE ε4 allele status, and the likelihood of conversion to MCI.

RESULTS

In SCD patients, the concentrations of Aβ_1–42 and macrophage‐colony stimulating factor (M‐CSF) were higher, respectively, in cEVs and serum. No difference was observed for fraktaline, interleukin (IL)‐4, IL‐13, interferon gamma (IFN‐γ), and sCD40L in either cEVs or serum between SCD and control patients. Based on receiver operating characteristic curve analysis, regression modeling, and correlations with cognitive performance, M‐CSF levels in serum effectively distinguished individuals with SCD who converted to MCI from those who remained stable. Interestingly, combining M‐CSF and cEVs Aβ_1–42 with the Rey Auditory Verbal Learning Test (RAVLT) cognitive scores provided an excellent classification for SCD converted to MCI up to 2 years prior to clinical diagnosis.

DISCUSSION

Our findings support the potential value of integrating serum M‐CSF levels with RAVLT performance and cEVs Aβ_1–42 concentrations into a multimodal biomarker panel for longitudinal monitoring of progressive neurocognitive impairment.

Highlights

• We investigated early immune‐related biomarkers of amyloid beta (Aβ) clearance in individuals with SCD who later develop MCI.

• The multimodal model combining serum M‐CSF, Aβ_1–42 in circulating extracellular vesicles (cEVs), and the RAVLT scores exhibits high predictive accuracy up to 2 years before MCI diagnosis.

• M‐CSF may represent a promising therapeutic signaling target to mitigate Aβ damage.

1. BACKGROUND

Subjective cognitive decline (SCD) is increasingly recognized as an early clinical manifestation of preclinical Alzheimer's disease (AD), preceding objective impairment. ¹ Amyloid beta (Aβ) deposition is a consistent hallmark in this at‐risk population, ² although the mechanisms driving progression to dementia remain unclear. Among Aβ isoforms, Aβ_1–42 is considered the most neurotoxic ³ due to its high hydrophobicity, propensity to misfold into oligomers and fibrils, ⁴ and distinct clearance pathways involving microglia and macrophages. ⁵ , ⁶

In early AD, microglia contribute to immunosurveillance and neuroprotection by clearing Aβ. ⁷ , ⁸ This activity is regulated by signaling molecules such as the macrophage colony‐stimulating factor (M‐CSF), which promotes microglial proliferation and differentiation, Aβ phagocytosis, ⁹ , ¹⁰ and cytokines including fractalkine, interferon‐γ (IFN‐γ), interleukin‐4 (IL‐4), IL‐13, and the inflammatory mediator soluble CD40 ligand (sCD40L) that modulate microglial activation in response to Aβ load and promote the clearance of amyloid plaques. ¹¹ , ¹² , ¹³ , ¹⁴ Although microglia are recruited to amyloid plaques, they ultimately fail to prevent their accumulation. Consequently, brain‐derived Aβ may be exported to the periphery, where peripheral phagocytes and their regulatory factors may contribute to Aβ clearance. ⁶ , ¹⁵ Accordingly, the expression of regulatory factors controlling peripheral macrophage activity such as fractalkine or some chemokines plays a critical role and may directly reflect pathological processes occurring in the brain.

Aβ can bind to or be packaged into circulating extracellular vesicles (cEVs) and may be cleared from the bloodstream through interactions with macrophages. ¹⁶ , ¹⁷ EVs are lipid‐based nanoparticles that transport diverse molecular cargo and regulate key physiological and pathological processes. ¹⁸ We previously showed that cEVs could bind Aβ. ¹⁶ The cEVs–Aβ complexes could be cleared by macrophages via phosphatidylserine receptors embedded on their membrane. ¹⁷ cEVs may therefore contribute to Aβ clearance while reflecting macrophage activity, as they can also secrete EVs. ¹⁹

Within this framework, we have shown that certain inflammatory and neurotrophic factors in neuronal‐derived EVs are differentially regulated in response to stress‐related treatments. ²⁰ Inflammatory markers are also commonly released within cEVs, and some have been found to be three‐ to 10‐fold more concentrated in cEVs than in plasma itself. ²¹ Importantly, EV‐associated cytokines are biologically active and capable of inducing functional responses in recipient cells. ²¹ However, the specific profile of some inflammatory markers involved in the microglial activity in cEVs from SCD patients converted to AD remains to be established.

cEVs transport stable, enriched, cell‐derived cargo that may reflect biological processes not captured by soluble markers. We hypothesize that cEVs‐associated biomarkers offer complementary and more sensitive signatures of disease‐related changes, improving both mechanistic insight and diagnostic potential.

Using samples of SCD participants at baseline from the Consortium for the Early Identification of Alzheimer's Disease‐Quebec (CIMA‐Q) cohort, we identified markers in cEVs and serum that reflect the spectrum of macrophage activity and could predict future cognitive decline in synergy with Aβ.

2. METHODS

2.1. Participants and data source

The CIMA‐Q is a longitudinal multi‐center cohort study designed to investigate the progression of AD and to identify early biomarkers as well as potential novel therapeutic targets. ²² Control participants were selected according to stringent inclusion criteria. ²² Inclusion and exclusion criteria, the details of the blood collection, and the clinical evaluation are summarized in Table S1.

2.2. Isolation and quality control of extracellular vesicles

cEVs were isolated from serum using established protocols previously validated by our group. ¹⁶ , ²⁰ , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ The final cEVs suspension was stored at −80°C until downstream analyses. To evaluate potential contamination with serum‐derived lipoproteins, levels of apolipoprotein A1 (ApoA1) were measured in the cEVs preparations using a quantitative enzyme‐linked immunosorbent assay (ELISA; DuoSet ELISA, DY3664‐05, R&D Systems, Minneapolis, MN, USA).

2.3. Characterization of extracellular vesicles

2.3.1. Transmission electron microscopy (TEM)

cEVs morphology was assessed by TEM. Briefly, isolated cEVs were fixed in 2% paraformaldehyde, applied to Formvar‐carbon–coated grids, negatively stained with 2% uranyl acetate, air‐dried, and imaged using a Hitachi H‐7100 microscope at 75 kV with 15,000 to 40,000 × magnification.

2.3.2. Nanoparticle tracking analysis (NTA)

cEVs size distribution and concentration were characterized by NTA using the NanoSight NS300. Detection threshold and camera level were set at 5 and 14, respectively, and three 60‐s videos per sample were recorded. Videos were analyzed with NanoSight NTA 3.2 software using automatic tracking for consistent and reproducible measurements.

2.3.3. Capillary Western blot analysis

All samples were standardized to 1 mg/mL protein, denatured at 95°C for 5 min, and separated using the capillary Western blot Jess system (ProteinSimple, CA, USA) with the 12‐ to 230‐kDa module (375 V, 25 min) according to the manufacturer's instructions. Proteins were incubated for 30 min with primary antibodies against TSG101 (Novus Biologicals LLC, Catalog No.: NB200–112, 1:10), CD63 (R&D Systems; Catalog No.: MAB50482, 1:20), ALIX (R&D Systems; Catalog No.: MAB50482, 1:20), and calnexin (Novus Biologicals LLC; Catalog No.: NB100–1965, 1:25), followed by washing and incubation with species‐specific secondary antibodies. Fluorescent detection was performed using streptavidin‐HRP, and signals were acquired and analyzed with Compass software (version 5.0.1, ProteinSimple).

2.4. Meso scale analysis

The levels of Aβ_1–42 were determined in serum and cEVs samples (50 µL) using R‐PLEX ELISA kit (Meso Scale Diagnostics [MSD], USA). Aβ_1–42 concentrations were determined based on standard curves generated and analyzed using MSD DISCOVERY WORKBENCH® software. All assays were performed in accordance with the manufacturer's guidelines. Limit detection sensibility for Aβ_1–42 is described in Table S2 in supporting information.

2.5. Luminex assay

Multiplex quantification of selected cytokines including, M‐CSF, fraktaline, IFN‐γ, IL‐4, IL‐13 and sCD40L, was performed using Luminex xMAP technology. The assay was conducted on the Luminex™ 200 system (Luminex Corporation, TX, USA). Biomarker levels were simultaneously quantified in 100 µL of either serum or cEVs samples using the Human Focused Discovery Assay (MilliporeSigma, MA, USA) following the manufacturer's protocol. The lower limits of detection for all analytes are reported in Table S2 in the Supporting Information.

RESEARCH IN CONTEXT

1. Systematic review: The authors conducted a literature review using PubMed. Publications describing immune‐related mechanisms involved in Aβ clearance with a particular focus on the preclinical stage of AD are cited throughout the manuscript. The contribution of cEVs to Aβ removal and their emerging value as diagnostic tools is also explored.

2. Interpretation: Our findings demonstrate that the integration of serum M‐CSF, cEVs Aβ_1–42 levels, and RAVLT performance enhances the predictive value for identifying individuals with SCD who are at risk of converting to MCI. These results underscore the potential utility of this combined biomarker–cognitive panel for the early monitoring of cognitive decline.

3. Future directions: Future investigations should comprehensively explore the immune response elements associated with Aβ burden to improve the prediction of early signs and manifestations of cognitive impairment.

2.6. Statistical analysis

All statistical analyses were performed using SPSS software version 20.0 and GraphPad Prism version 10.0. Normality of data distribution was assessed using the Shapiro–Wilk test. Statistical analyses of clinical and biochemical parameters were conducted using Student's t‐test and one‐way analysis of variance (ANOVA), followed by Tukey post hoc tests for normally distributed data. For non‐normally distributed data, the non‐parametric Mann–Whitney U test and Kruskal–Wallis test were applied, followed by Dunn's post hoc test. For the study‐specific markers, analysis of covariance (ANCOVA) was conducted to account for potential confounding factors. All models were adjusted for key demographic covariates, including age and sex, and post hoc pairwise comparisons were performed using Bonferroni correction. Correlation analyses were performed using partial correlation adjusted for age and sex. Multiple regression models and receiver operating characteristic (ROC) analysis were performed, and corresponding areas under the curve (AUC) were calculated accordingly. ROC analyses were employed to assess relative discriminatory capacity rather than predictive accuracy or clinical utility. These analyses were conducted in an exploratory approach. AUC values indicate discriminatory ability and should not be interpreted as validated predictive accuracy. Given the small sample size, all statistical analyses were performed within an exploratory and hypothesis‐generating strategy. Differences were considered statistically significant at p < 0.05.

3. RESULTS

3.1. Cohort characteristics

Biochemical, clinical, and demographic data of the study population are summarized in Table 1. The sex distribution and mean age were comparable between the groups. Biochemical parameters, including lipid, hepatic, and renal profiles, showed no significant differences between the control and SCD groups. Similarly, no differences were observed in complete blood count or cognitive assessments using the Montreal Cognitive Assessment (MoCA), the Digit Symbol Substitution Test (DSST), alpha span, the Benton Judgment of Line Orientation Test (BORB), the Boston Naming Test (BNT), and Rey Auditory Verbal Learning Test (RAVLT) between both groups (Figure S1A–F in Supporting Information).

TABLE 1.

Clinical and biochemical characteristics of study participants.

Characteristics	Controls (n = 14)	SCD (n = 23)	p value
Clinical profile
Age (years)	74.6 ± 1.8	72.4 ± 0.9	0.237
Sex ratio (Male/Female)	3/11	6/17	0.757
Genotype (APOE ε4−/APOE ε4+)	14/0	17/6 *	0.038
MoCA score	28.4 ± 0.3	27.7 ± 0.3	0.176
Glycemia (mmol/L)	5.1 ± 0.5	5.3 ± 0.1	0.514
HbA1c (%)	5.1 ± 0.4	5.4 ± 0.1	0.406
Lipid and hepatic profiles
Total cholesterol (mmol/L)	4.5 ± 0.4	4.9 ± 0.2	0.321
Triglycerides (mmol/L)	1.2 ± 0.2	1.1 ± 0.1	0.876
LDL‐C (mmol/L)	2.5 ± 0.3	2.8 ± 0.1	0.325
HDL‐C (mmol/L)	1.5 ± 0.1	1.6 ± 0.1	0.483
Phosphatase alkaline (UI/L)	55.8 ± 6.2	55.3 ± 3.6	0.939
Renal profile
Creatinine (umol/L)	76.6 ± 7.9	75.5 ± 2.8	0.876
Urea (mmol/L)	5.8 ± 0.6	5.3 ± 0.4	0.415
Sodium (mmol/L)	130.1 ± 10.1	139.8 ± 0.5	0.218
Potassium (mmol/L)	3.9 ± 0.32	4.2 ± 0.3	0.505
Calcium (mmol/L)	2.2 ± 0.2	2.3 ± 0.1	0.274
Vitamins and inflammation
Vitamin B12 (pmol/L)	250.3 ± 37.5	310.8 ± 26.9	0.190
CRP (mg/L)	2.4 ± 0.6	1.4 ± 0.2	0.051
Complete blood count (CBC)
White blood cell	4.8 ± 0.5	5.2 ± 0.3	0.463
Red blood cell	4.0 ± 0.3	4.5 ± 0.07	0.089
Neutrophiles (10e9/L)	2.8 ± 0.3	3.0 ± 0.2	0.533
Lymphocytes (10e9/L)	1.4 ± 0.1	1.6 ± 0.1	0.252
Monocytes (10e9/L)	0.4 ± 0.05	0.4 ± 0.02	0.970
Basophiles (10e9/L)	0.02 ± 0.0	0.04 ± 0.0	0.232

Values are given as mean ± standard error mean.

Abbreviations: APOE, apolipoprotein E; HbA1c, glycated hemoglobin; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; MoCA, Montreal Cognitive Assessment; CRP, C‐Reactive protein; SCD, subjective cognitive decline; (−), ε4 non‐carrier; (+), ε4 carrier.

Student's t‐test was used for statistical analysis with *, p < .05 compared to controls.

3.2. cEVs characterization

TEM images revealed the presence of a clearly distinguishable lipid bilayer, characteristic of vesicular structures (Figure 1A). According to NTA, cEVs exhibited a size distribution ranging from approximately 50 to 300 nm in diameter, with the highest particle concentration observed around 100 nm (Figure 1B). ApoA1 levels were markedly lower in isolated cEVs compared to unprocessed serum, confirming effective separation (Figure 1C). Immunoblotting confirmed cEVs markers TSG101, ALIX, and CD63, while calnexin was detected only in SK‐N‐SH lysates, confirming isolate purity (Figure 1D).

FIGURE 1.

Characterization and imaging of serum‐derived cEVs. (A) cEVs images as acquired by transmission electron microscopy; bar represents 100 nm. (B) Concentration and size distribution of cEVs examined by NTA using Nanosight‐NS300. (C) ApoA1 concentration in serum and cEVs. (D) Immunoblot detection of EV protein markers and non‐associated proteins. The parametric Student's t‐test was used for statistical analysis of ApoA1; ***p < 0.001. ALIX, ALG‐2‐interacting protein X; ApoA1, apolipoprotein A1; CD63, cluster of differentiation 63; cEVs, circulating extracellular vesicles; CTR, controls; nm, nanometer; NTA, nanoparticle tracking analysis; SCD, subjective cognitive decline; TSG101, tumor susceptibility gene 101.

3.3. Inflammatory and Aβ_1–42 analysis in serum and cEVs

The levels of Aβ_1–42 in cEVs, but not in serum, were significantly higher in SCD participants compared to controls (p = 0.016) (Figure 2A,B, Table S3). The discriminative ability of cEVs‐associated Aβ_1–42 was moderate, as reflected by the ROC curve (AUC = 0.77; p = 0.006) (Figure 2E, Table S4 in Supporting Information). Conversely, M‐CSF levels were significantly reduced in cEVs from SCD participants relative to controls (p = 0.039) (Figure 2C,D, Table S3), with serum M‐CSF demonstrating moderate diagnostic accuracy, slightly lower than that observed for cEVs‐associated Aβ_1–42 (AUC = 0.75; p = 0.012) (Figure 2H, Table S4 in supporting information). However, no significant differences were observed between SCD and controls for the levels of fraktaline, IL‐4, IL‐13, IFN‐γ, and sCD40L in either cEVs or serum (Table S5 in Supporting Information). Notably, the concentrations of these markers were between two and 50 times higher in cEVs than in serum, except for sCD40L (Table S5 in Supporting Information). Interestingly, in SCD patients, the cEVs/serum ratio of these markers was reduced, except for sCD40L. Partial correlation analysis showed a strong positive correlation between M‐CSF and Aβ_1–42 levels in either cEVs or serum across all participants, suggesting a potential link between peripheral and vesicle‐associated biomarker dynamics (Table S6 in Supporting Information). Interestingly, correlation analyses yielded a significant association between serum M‐CSF levels and cognitive performance scores of DSST and RAVLT tests. In contrast, no such relationship was observed for Aβ_1–42 levels (Table S6 in Supporting Information).

FIGURE 2.

Levels of studied markers in EVs and serum. (A) EV Aβ_1‐42. (B) Serum Aβ_1‐42. (C) EV M‐CSF. (D) Serum M‐CSF. (E) EVs Aβ_1‐42 ROC curve of CTR versus SCD. (F) Serum Aβ_1‐42 ROC curve of CTR versus SCD. (G) EV M‐CSF curve of CTR versus SCD. (H) Serum M‐CSF curve of CTR versus SCD (N = 37, including 14 CTR and 23 SCD). Statistical analysis was performed using analysis of covariance models adjusted for age and sex; *p < 0.05 versus controls. Aβ_1‐42, amyloid beta peptide 1–42; AUC, area under the curve; CI, confidence interval; CTR, controls; EV, extracellular vesicle; M‐CSF, macrophage colony‐stimulating factor; ROC, receiver operating characteristic; SCD, subjective cognitive decline.

3.4. Subpopulation analysis

3.4.1. SCD⁺ and SCD⁻ stratification

To better characterize the heterogeneity within the SCD group, participants were stratified into two subgroups: SCD⁺ and SCD⁻. The same panel of biomarkers was re‐evaluated within the stratified SCD subgroups (Table S7 in Supporting Information). Notably, cEVs Aβ_1–42 levels were higher in SCD⁺ individuals compared to SCD⁻ participants and controls (p = 0.027 and p = 0.002, respectively) (Figure S2A, Table S3) with a good discriminatory capacity (AUC = 0.84, p = 0.002) (Figure S2E, Table S4). Interestingly, cEVs Aβ_1–42 levels were also able to distinguish between SCD⁺ and SCD⁻ subgroups (Figure S2G, Table S4), suggesting its potential as a stratification marker within the SCD population. By contrast, M‐CSF concentrations in both cEVs and serum remained comparable between SCD⁺, SCD⁻, and control participants (Figure S2C,D). In addition, cognitive tests failed to distinguish between any of the groups, indicating limited diagnostic relevance in this context (Figure S3A–F in Supporting Information).

3.4.2. APOE ε4 allele approach

To assess the impact of the APOE ε4 allele on serum and vesicular biomarker concentrations, the SCD⁺ subgroup was further stratified into ε4 carriers (SCD⁺ APOE ε4⁺) and ε4 non‐carriers (SCD⁺ APOE ε4⁻) (Table S8 in Supporting Information). Our results show that individuals in the SCD⁺ APOE ε4⁺ group exhibited significantly elevated levels of Aβ_1–42 in both serum and cEVs compared to the other groups (p = 0.0001 vs cEVs controls; p = 0.002 vs. cEVs SCD⁺ APOE ε4⁻; p = 0.01 vs serum controls; p = 0.03 vs serum SCD⁺ APOE ε4⁻) (Figure S4A,B, Table S3). Importantly, cEVs Aβ_1–42 demonstrated superior discriminatory performance (AUC = 1, p = 0.0005) relative to its serum counterpart (AUC = 0.86, p = 0.04) (Figure S4E,F, Table S4). Moreover, cEVs Aβ_1–42 concentration was statistically higher in SCD APOE ε4⁺ carriers than non‐carriers, further supporting the genotype‐related differences observed in peripheral biomarker expression (Figure S4A,B,G). However, M‐CSF levels in either cEVs or serum remained consistent with or without APOE ε4 (Figure S4C,D). Similarly, cognitive test scores did not differ significantly between SCD APOE ε4 carriers, non‐carriers, and controls (Figure S5A–F in Supporting Information).

3.4.3. Longitudinal approach

Prospective follow‐up assessments revealed that, within the SCD group, eight individuals progressed to MCI, while 15 remained cognitively stable. Based on this clinical outcome, participants were subsequently stratified into two groups: SCD converters (those who progressed to MCI) and SCD non‐converters (those who remained stable) (Table S9 in Supporting Information). Our data demonstrated that cEVs Aβ_1–42 levels were significantly elevated in SCD converted to MCI compared to controls (p = 0.01) (Figure 3A, Table S3), with an AUC of 0.9 (p = 0.0013), indicating good discriminatory capacity (Figure 3E, Table S4). In addition, serum M‐CSF levels were markedly higher in SCD converters relative to controls (p = 0.043) (Figure 4D, Table S3) and demonstrated good discriminatory performance, although with lower efficiency than cEVs Aβ_1–42, as reflected by an AUC of 0.84 (p = 0.007) (Figure 4F, Table S4). Serum M‐CSF levels also effectively distinguished between SCD converters and non‐converters, achieving a higher discriminatory performance with an AUC of 0.86 (p = 0.0051), compared to an AUC of 0.79 for cEVs Aβ_1–42 (p = 0.019) (Figure 4A,C, Table S4). In contrast, no significant differences were observed in serum Aβ_1–42 or cEVs M‐CSF levels between the two groups (Figure 3B,C). Interestingly, the RAVLT score (total recognition score) differed significantly between SCD converters and controls, as demonstrated by both standard analyses (p = 0.001) (Figure S6F, Table S3) and ROC curve results (AUC = 0.89, p = 0.0016) (Figure S6G, Table S4). In addition, the RAVLT score was different between SCD converters and non‐converters (p = 0.007) and achieved a stronger discriminatory performance than cEVs Aβ_1–42 and serum M‐CSF to discriminate between these two groups (AUC = 0.9, p = 0.0013) (Figure 4B, Table S4).

FIGURE 3.

EVs and serum biomarker profiles based on a longitudinal approach. (A) EVs Aβ_1‐42. (B) Serum Aβ_1‐42. (C) EV M‐CSF. (D) Serum M‐CSF. (E) EV Aβ_1‐42 ROC curve of CTR versus SCD converted. (F) Serum M‐CSF ROC curve of CTR versus SCD converted (N = 37, including 14 CTR, 14 SCD non‐converted and nine SCD converted). Statistical analysis was performed using analysis of covariance models adjusted for age and sex; *p < 0.05 and **p < 0.01. Aβ_1‐42, amyloid beta peptide 1–42; AUC, area under the curve; CI, confidence interval; CTR, controls; EV, extracellular vesicle; M‐CSF, macrophage colony‐stimulating factor; ROC, receiver operating characteristic; SCD, subjective cognitive decline.

FIGURE 4.

Diagnostic performance of studied markers following longitudinal stratification. (A) EV Aβ_1‐42 ROC curve of SCD non‐converted versus SCD converted. (B) RAVLT score ROC curve of SCD non‐converted versus SCD converted. (C) Serum M‐CSF ROC curve of SCD non‐converted versus SCD converted. (D) Multimodal panel ROC curve of SCD non‐converted versus SCD converted. Aβ_1‐42, amyloid beta peptide 1–42; AUC, area under the curve; CI, confidence interval; M‐CSF, macrophage colony‐stimulating factor; RAVLT, Rey Auditory Verbal Learning Test; ROC, receiver operating characteristic; SCD, subjective cognitive decline.

Based on these findings, a multiple regression model was developed incorporating serum M‐CSF concentrations, cEVs Aβ_1–42 levels, and RAVLT scores. In exploratory analyses, the biomarker–cognitive panel showed excellent discriminatory capacity between SCD at high risk of MCI conversion and those who remained stable (AUC = 1, p < 0.001) (Figure 4D, Table S4).

4. DISCUSSION

The implementation of preventive strategies at the initial stages of cognitive decline may contribute to slowing neurodegenerative processes and conferring clinical benefit. Accordingly, effective detection of cognitive impairment prior to the onset of clinical symptoms is of paramount importance. In this context, SCD is increasingly recognized as the earliest behaviorally measurable stage along the AD continuum and thus represents a valuable window for intervention and disease monitoring. The current study aimed to identify predictive biomarkers capable of stratifying individuals with SCD who are at increased risk of progressing to MCI, leveraging a well‐characterized longitudinal cohort.

Our study revealed that Aβ_1–42 levels in cEVs were already elevated in SCD participants compared to controls, with a pronounced increase observed in SCD⁺ APOE ε4 carriers. Our findings are consistent with a recent systematic review and meta‐analysis of 46 longitudinal studies that identified strong predictors of progression from SCD to objective impairment, including elevated Aβ deposition and APOE ε4 variant. ²⁸ Interestingly, Aβ_1–42 levels in cEVs demonstrated greater classification accuracy than serum, with improved discriminatory capacity. As demonstrated in our previous works, it may be attributed to the dual functional role of cEVs – their ability to internalize Aβ, as well as to sequester oligomeric forms of this neurotoxic peptide. ¹⁶ , ²⁵ Our findings also align with the known impact of APOE ε4 on Aβ metabolism, where carriers show exacerbated Aβ accumulation largely driven by impaired microglial autophagy and reduced receptor‐mediated clearance in both brain and peripheral compartments. ²⁹ , ³⁰ , ³¹

Therefore, we focused on the mechanisms involved in Aβ regulation during the early SCD stage, particularly the activation of the microglial defense system. In the later stages of AD, inflammation is a prominent feature caused by sustained microglial activation. ³² Although the role of neuroinflammation in symptomatic AD has been well documented, the inflammatory landscape during SCD remains poorly defined. Accordingly, we investigated the regulation of several cytokines selected for their known roles in reinforcing glial cell capacity to respond to Aβ‐induced insults. Our work demonstrated that elevated serum M‐CSF levels and their concordant decrease in cEVs could be used to differentiate SCD participants from controls. Activation of the M‐CSF signaling pathway in macrophages has been shown to exert protective and beneficial effects, driving toward an M2 polarization state characterized by anti‐inflammatory and tissue repair functions. ³³ In transgenic mouse models, systemic administration of M‑CSF reduced plaque density and accelerated Aβ phagocytosis. ³⁴ Mechanistically, upregulation of the M‑CSF receptor on microglia augments their phagocytic capacity toward fibrillar Aβ. ³⁵ Clinically, cerebrospinal fluid (CSF) and plasma M‑CSF levels were higher in AD patients compared to age‐matched MCI and controls. ¹⁰ Our finding of elevated serum M‐CSF in SCD patients aligns with results from a recent study. ³⁶ Its upregulation reflects an early macrophage response to the gradual accumulation of Aβ, potentially mitigating neuronal damage and thereby helping to preserve normal brain function.

Aside from M‐CSF, the remaining cytokine panel (fractalkine, IL‐4, IL‐13, IFN‐γ, and sCD40L) were largely comparable across groups. These results suggest that, despite the mechanistic rationale linking these cytokines to Aβ removal and clearance, systemic and cEVs‐associated cytokine changes may be specific rather than uniform cytokine shifts, at least at the SCD stage.

The observed positive correlation between M‐CSF in the serum and Aβ_1–42 levels confirms their interconnection and suggests that M‐CSF may serve as a sensitive indicator of early Aβ buildup. Accordingly, M‐CSF is likely mobilized as a frontline defense mechanism by macrophages to tackle Aβ emerging pathology. Moreover, serum M‐CSF levels showed a significant correlation with scores from the DSST and RAVLT cognitive tests, whereas Aβ_1–42 levels failed to establish such association. The poor performance of peripheral Aβ on the studied executive‐function assessments suggests a limited relationship with cognition at this stage. In contrast, serum M‐CSF effectively captured early cognitive changes, highlighting its potential as a sensitive biomarker during the initial phases of disease progression.

Considering that M‐CSF appears to track disease progression during the preclinical and clinical AD stages, we further evaluated its diagnostic utility, along with Aβ_1–42, in predicting conversion from SCD to MCI. Our findings from the 2‐year follow‐up assessment suggest that serum M‐CSF exhibits strong potential as a stratification biomarker for SCD‐to‐MCI conversion, demonstrating a discriminatory capacity comparable to that of cEVs Aβ_1–42. In this longitudinal approach, RAVLT scores also indicated risk of MCI conversion. The RAVLT primarily evaluates the capacity to consolidate, retrieve, and retain verbal information. ³⁷ It is sensitive to subtle changes in memory function, making it useful in the early detection of cognitive decline and MCI. ³⁷

Collectively, these data highlight the utility of combining serum M‐CSF, cEVs Aβ_1–42, and RAVLT scores into a multimodal approach to monitoring the disease progression. A recent study demonstrated that stratifying SCD individuals by combinations of plasma biomarkers and neuroimaging findings improved prediction of subsequent cognitive decline. ³⁸ Our data showed that the combined model provided an excellent discrimination of SCD participants converted to MCI compared to SCD non‐converted (AUC = 0.93, p < 0.0001), outperforming any single biomarker alone. Our results support the idea that multimodal approaches are more predictive than any single measure and justify the use of integrated biomarker‐cognitive models for early cognitive decline detection and progression.

Finally, it is noteworthy that cEVs M‐CSF levels failed to differentiate between SCD participants and healthy controls across various stratification approaches. This observation is not unique to our study, as previous reports demonstrated that some serum or plasma biomarkers exhibited superior diagnostic accuracy and clinical robustness compared to EV‐derived markers. ³⁹ , ⁴⁰ This may be explained by the heterogeneity of the SCD population and/or by the fact that biomarker changes within cEVs are often more subtle than the pronounced shifts typically observed in serum during inflammation or disease progression. In fact, some biomarkers may be abundantly present in serum due to processes such as active secretion or immune activation, yet may not be selectively packaged into EVs, which can maintain relatively constant levels regardless of the disease state.

Despite the strengths of our study, several limitations should be considered. Although our longitudinal cohort is well characterized, the modest sample size can impact model stability and predictive accuracy, especially in stratified analyses; for this reason, all results should be considered preliminary and must be confirmed in larger cohorts. Analyses focusing on SCD converters were performed to provide initial exploratory insights into potential biomarker trajectories. The present dataset does not support formal longitudinal modeling approaches, including mixed effects, which will be addressed in future studies using extended follow‐up cohorts. Sex differences influence cognitive and biomarker measures in aging and AD and may affect generalizability. ⁴¹ Although sex was included as a covariate, the uneven sex ratio could impact subgroup stability in small samples, though it reflects the female predominance observed in the AD population.

Finally, as the biomarkers studied are largely immune‐related, future studies using microglia‐specific EVs could improve cellular specificity and strengthen biological interpretation. While mechanistic interpretations are supported by prior literature, experimental studies are required to directly validate these pathways. Although M‐CSF, Aβ_1–42, and RAVLT show promising predictive potential, multimodal approaches including imaging and CSF biomarkers are therefore recommended to improve diagnostic specificity.

5. CONCLUSIONS

This study highlights serum M‐CSF as both a potential avenue for future research to enhance Aβ clearance and a biomarker for disease staging, particularly in predicting MCI conversion. Early immunological changes may define therapeutic windows before irreversible neurodegeneration. Integrating M‐CSF with cognitive measures, such as in the M‐CSF–Aβ_1–42–RAVLT model, could improve longitudinal monitoring in preclinical AD; however, its diagnostic utility in clinical stages warrants further comprehensive investigation.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest and no competing interests. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

Written informed consent was provided by all participants prior to study participation, and approval was obtained from the Institutional Review Board.

Supporting information

Supporting File 1: trc270240‐sup‐0001‐tableS1.docx

Supporting File 2: trc270240‐sup‐0002‐tableS2.docx

Supporting File 3: trc270240‐sup‐0003‐tableS3.docx

Supporting File 4: trc270240‐sup‐0004‐tableS4.docx

Supporting File 5: trc270240‐sup‐0005‐tableS5.docx

Supporting File 6: trc270240‐sup‐0006‐tableS6.docx

Supporting File 7: trc270240‐sup‐0007‐tableS7.docx

Supporting File 8: trc270240‐sup‐0008‐tableS8.docx

Supporting File 9: trc270240‐sup‐0009‐tableS9.docx

Supporting File 10: trc270240‐sup‐0010‐figureS1.docx

Supporting File 11: trc270240‐sup‐0011‐figureS2.docx

Supporting File 12: trc270240‐sup‐0012‐figureS3.docx

Supporting File 13: trc270240‐sup‐0013‐figureS4.docx

Supporting File 14: trc270240‐sup‐0014‐figureS5.docx

Supporting File 15: trc270240‐sup‐0015‐figureS6.docx

Supporting File 16: trc270240‐sup‐0016‐SuppMat.pdf