An analysis was conducted to determine the required sample size, considering several factors such as the effect size, significance level, power, number of groups, and covariates. Based on this analysis, a sample size of 90 participants would be sufficient to conduct an ANCOVA with a power of 80%. To enhance the robustness of our analysis and account for potential dropouts, we decided to include an additional 30 subjects. As a result, the final study sample size consisted of 120 participants, all of whom were between the ages of 60 and 75.

The study included 61 aMCI cases referred from Thailand’s King Chulalongkorn Memorial Hospital, as well as 60 healthy controls. The two groups were selected from the same geographical region, specifically Bangkok, Thailand. Subjects with aMCI were recruited from various departments at King Chulalongkorn Memorial Hospital, including the Geriatric Clinic, the Cognitive Fitness Center unit, the Geriatric Psychiatry clinic, and the Neuroscience Center. The control group included a variety of individuals, such as senior Red Cross volunteers, attendees of health check-up clinics, members of neighborhood senior organizations, and healthy elderly caregivers of aMCI patients who visited the Dementia Clinic.

An efficient screening process was conducted to identify eligible individuals for enrollment. Patients diagnosed with aMCI underwent assessment and met the criteria established by Petersen.^[26] Additionally, they obtained a Clinical Dementia Rating (CDR) score of 0.5. These criteria encompass the identification of both subjective and objective memory impairments, the exclusion of dementia, and alterations in activities of daily living (ADL).

Individuals with neurological conditions such as Parkinson’s disease, Alzheimer’s disease, multiple sclerosis, epilepsy, or stroke were deemed ineligible for participation in the study for both groups. In addition, the study excluded patients who had been diagnosed with various neuropsychiatric disorders such as substance use disorders, chronic fatigue syndrome, bipolar disorder, MDD, post-traumatic stress disorder, schizophrenia, obsessive-compulsive disorder, autism spectrum disorders, and generalized anxiety disorder. Individuals with chronic kidney disease, cancer, metabolic syndrome, chronic inflammatory bowel disease, HIV infection, hepatitis, or chronic obstructive pulmonary disease were not eligible to participate. People with communication difficulties, visual impairments, hearing loss, physical conditions affecting their ability to sit or stand, or mobility issues were also not included.

The study received approval from the Institutional Review Board committee of the Faculty of Medicine, Chulalongkorn University, and the ethical committee of King Chulalongkorn Memorial Hospital (886/64). Prior to any data collection, all participants provided written informed consent.

This study utilized a case-control design, consisting of 60 control participants and 61 individuals with aMCI.^[9,27] We conducted a comprehensive interview to collect demographic and clinical information, utilizing a range of questionnaires. The Mini International Neuropsychiatric Interview (M.I.N.I.)^[28] was utilized for the purpose of establishing axis-1 diagnoses, specifically MDD. To assess the extent of neurocognitive impairments, we employed the Montreal Cognitive Assessment (MoCA) scale^[29] and the Thai version of the Mini-Mental State Examination (MMSE)^[30]. Additionally, we utilized the modified CDR score.^[9,31] The qMCI score, which serves as the primary principal component, was derived from the scores on the MoCA, MMSE, and the modified CDR score.^[9]

For the assessment of distress, depression, and anxiety symptoms, a variety of standardized scales were utilized. These included the Perceived Stress Scale (PSS) developed by Wongpakaran and Wongpakaran^[32]; the State-Trait Anxiety Inventory (STAI) developed by Spielberger et al.^[33]; the Thai Geriatric Depression Scale (TGDS) developed by Yesavage et al.^[34] and translated into the Thai version by Pongwarin^[35]; the depression (HADS-D) and anxiety (HADS-A) subscales of the Hospital Anxiety and Depression Scale developed by Nilchaikovit^[36]; and the neuroticism trait score from the Five Factor Model standardized psychometric pool of items (IPIP-NEO) developed by Yomaboot and Cooper.^[37] We organized the study group into smaller groups based on a HADS-D score of ≥8, resulting in two subgroups: one with higher depressive symptoms (n=26) and one with lower depressive symptoms (n=95). In a study conducted by Tran-Chi et al.^[9], a distress symptom of old age (DSOA) dimension was constructed through principal component analysis (PCA) on stress-affective symptoms. This dimension was derived as the first principal component extracted from various subscales, including PSS, STAI, TGDS, HADS-D, HADS-A, and the neuroticism score. In previous research, it was found that there was no significant association between the DSOA and qMCI dimensional scores.

In addition, a specific group of individuals with cognitive impairments was identified through cluster analysis. This advanced machine learning technique effectively removed subjects who displayed symptoms associated with the DSOA dimension in aMCI subjects, as demonstrated in the study by Tran-Chi et al.^[9] This two-step cluster analysis indicated that the study sample based on the aMCI diagnostic criteria is overinclusive because part of the individuals with DSOA symptoms were incorrectly classified as aMCI. As such, the newly formed MCI subgroup (denoted as mCoDy, or mild cognitive dysfunction) removed subjects with higher DSOA scores, thereby generating a more homogenous group comprising people with only episodic and semantic memory disorders.^[9] Therefore, a more specific subset of aMCI, referred to as mCoDy, was established. In this study, we will discuss the findings from a group of individuals with aMCI (n=61) as well as a more specific subgroup called mCoDy (n=52), which is more restricted and homogeneous. Furthermore, we employed the established Thai translation of the Adverse Childhood Experiences (ACEs) Questionnaire^[38] to evaluate five dimensions of ACE, including emotional, physical, and sexual abuse, as well as emotional and physical neglect. The Negative Event (Hassle) Scale, developed by Maybery and Neale^[39], was utilized to assess negative life events (NLEs). The Alcohol, Smoking, and Substance Involvement Screening Test (ASSIST)^[40] was utilized to identify and eliminate individuals who engaged in substance use, including tobacco.

An analysis of variance was employed to investigate the variations in continuous variables among various diagnostic groups. Associations between the diagnosis of aMCI and the biomarkers were evaluated using multivariate and univariate GLM analysis. We implemented false discovery rate (FDR) p-correction on the biomarkers, which included multiple comparisons. In addition, we examined the effects of the interaction pattern between the diagnosis of aMCI x HADS-D groups. The chi-square test or Fisher’s exact probability test was used to compare nominal variables across distinct categories. Stepwise automatic binary logistic regression analysis was performed with inclusion and exclusion p-values set at 0.05 and 0.06, respectively, to delineate the most relevant biomarkers (with demographic, ACEs, and NLEs data) of the diagnosis of aMCI (dependent variable with controls as reference group). We employed manual and automatic multiple regression analyses to determine the most accurate predictors of quantitative scores (e.g., qMCI, MMSE, DSOA) using biomarkers and demographic data. We also utilized a forward stepwise automatic regression method with inclusion and exclusion p-values set at 0.05 and 0.06, respectively. The final regression models included standardized coefficients, t-statistics, and exact p-values for each explanatory variable. Additionally, the total variance was represented by effect size measures such as R² or partial eta squared and F statistics (along with their corresponding p-values). We employed both the White and modified Breusch-Pagan tests to evaluate the heteroskedasticity of the errors. An analysis was conducted to examine collinearity and multicollinearity. Various measures, such as tolerance (with a cut-off value of less than 0.25), the variance inflation factor (with a cut-off value of greater than 4), the condition index, and variance proportions from the collinearity diagnostics table, were utilized to assess the presence of heteroskedasticity. Furthermore, the data underwent z-score transformation to enhance their interpretability, and we generated z-unit-weighted composite scores that accurately depict distinct immune profiles. The data distributions were subjected, where necessary, to various normalization processes, such as logarithmic or rank-based transformations. In each investigation mentioned, a two-tailed design was used to determine statistical significance at an α value of 0.05. The software used for analysis was IBM Windows SPSS version 29.

The a priori sample size was calculated using G*Power 3.1.9.4. The primary outcome analysis was a multiple regression analysis with the qMCI score as the dependent variable and biomarker data as the input variables. The minimum sample size required was 78, based on an f value of 0.179 (corresponding to approximately 15% of the explained variance), 5 explanatory variables, an alpha value of 0.05, and a power of 0.8.

Table 1 displays the socio-demographic and clinical features of aMCI and control individuals in the current study. The MoCA and MMSE scores were significantly lower in aMCI subjects than in controls. The PSS score was slightly increased in aMCI, although there were no significant differences in the HADS depression and anxiety, STAI, and neuroticism scores among aMCI and control individuals. Emotional neglect and negative life events scores did not significantly differ between aMCI subjects and controls.

Table 2 shows the results of the O&NS measurements. The -SH groups were significantly higher in aMCI participants than in controls, while MDA was significantly lower in aMCI than in controls. There were no significant differences in folate, AOPP, and GSH (all results of GLM analysis with age, sex, BMI, and the drug status as covariates) and ApoE genotypes among the study groups. No significant interactions between the aMCI and HADS-D groups were found. The differences shown in Table 2 remained significant after FDR p correction (at p<0.0125).

We conducted a thorough investigation of the potential impact of the medicines and supplements that the participants were consuming on their oxidative stress profile. There were no significant differences in the use of antihypertensive drugs between the control group and the group with aMCI (43/17 versus 35/26, p=0.129). Similarly, there were no significant differences in the use of antidiabetic agents (54/6 versus 51/10, p=0.422), cholesterol lowering drugs (24/36 versus 25/36), vitamins (26/34 versus 31/30, p=0.409), fish oil (55/5 versus 51/10, p=0.270), and calcium supplements (40/20 versus 46/15, p=0.321). We have analyzed the potential impact of these medicines and supplements on the biomarkers of oxidative stress using multivariate and univariate GLM analysis, as presented in Supplementary Table 1 in the Supplement (see after References). The multivariate GLM analysis revealed a significant impact of cholesterol-lowering drugs; the univariate GLM analysis indicated that this impact was specifically observed in the -SH groups, which were significantly reduced. This observed effect remained statistically significant after applying the FDR p-correction at p=0.04.

Table 2 also shows the measurements of TC, FC, triglycerides, LDL-C, ApoB, LCAT, and the Castelli index in aMCI patients and control subjects. The results of GLM analyses, which were adjusted for age, sex, BMI, and drug status (as covariates), showed significant interaction patterns between the aMCI and HADS-D groups. We found significant interaction patterns for triglycerides, HDL, ApoA, ApoB, ApoB/ApoA ratio, RCT, and CRI-I. All differences in the lipid markers shown in Table 2 remained significant following FDR p-correction (at p<0.0282). Supplementary Figures 1 through 7 illustrate the interaction patterns. While for people without depressive symptoms there were no significant differences in HDL, ApoA, and RCT between aMCI and control subjects, these three markers were lower in aMCI subjects with depressive symptoms compared to those without. While in people without depressive symptoms there were no significant differences in ApoB, ApoB/ApoA ratio, and Castelli risk index 1 between aMCI and control subjects, these markers were higher in subjects with higher HADS-D scores compared with their counterparts. HDL, ApoA, and the RCT index were significantly lower in aMCI than in controls, whilst the ApoB/ApoA ratio was significantly higher in aMCI than in controls.

We investigated the effect of the treatment on the participants’ lipid profiles, and the findings are presented in Supplementary Table 2. The multivariate GLM analysis did not reveal any statistically significant effects on the lipid biomarkers. However, antidiabetic medications showed notable inhibitory effects on TC, LDL-C, ApoB, the ApoB/ApoA ratio, and free cholesterol based on univariate GLM. The FDR p-correction revealed that the impacts on TC, LDL-C, and ApoB remained statistically significant with a p-value of 0.0266. Regardless, we have calculated the residualized values for all biomarkers and assessed if utilizing these adjusted data will alter any outcomes. Nevertheless, there were no notable disparities observed when comparing the utilization of the unprocessed data with the residualized values.

Table 3 shows the results of a binary logistic regression analysis when the aMCI diagnosis is the dependent variable with the healthy control group as the reference group. Table 3, model #1, shows that aMCI was predicted by 4 variables, namely -SH groups (positively), education, MDA, and ApoB (all inversely), with a Nagelkerke value of 0.519 and an accuracy of 78.5%. We also examined whether the same variables would predict the restricted study group of mCoDy versus controls. Table 3, model #2, shows that mCoDy was predicted by the same variables with a Nagelkerke value of 0.456 and an accuracy of 79.2%. Model #3 shows that two variables predicted the subgroup with increased HADS-D scores, specifically AOPP and NELs health+money, with an accuracy of 79.5%.

Table 4 shows the results of multiple regression analyses with qMCI, MoCA, MMSE, DSAO, and HADS-D scores as dependent variables. Model #1 shows that 26.9% of the variance in the qMCI score was explained by the regression on education and MDA (both inversely) and the -SH groups and age (both positively). Model #2 shows that 24.9% of the variance in the MoCA score was explained by the same four variables. We found that 13.9% of the variance in the MMSE score was explained by two variables, namely education and MDA (both positively). Finally, we also regressed the DSOA and HADS-D scores on all biomarkers, psychological data (ACEs and NLEs), and demographic data. Not one of the biomarker data was significantly associated with the DSOA or HADS-D scores. For example, model #4 shows that NLEs health+money, education, and ACE neglect were associated with the score DSAO and explained 31.6% of its variance. Model #5 shows that NLEs health+money, and ACEs neglect were related to the HADS-D score and explained 16.4% of the variance.

Our study discovered that individuals with aMCI had lower MDA concentrations compared to the control group. Additionally, there were no notable variations in AOPP and antioxidants (GSH and folic acid) between the two groups. These findings challenge the initial hypothesis we had. According to a previous meta-analysis conducted by Nantachai, Vasupanrajit^[16], it was found that individuals with MCI had elevated levels of MDA and lipid hydroperoxides (LOOH) compared to the control group. The latter study examined elevated levels of other oxidative stress biomarkers, such as homocysteine and carbonyls, alongside reduced levels of antioxidant biomarkers, including vitamins, GSH, and GSH peroxidase.^[16]

One explanation for the differences is that subjects with MDD were not always excluded from the previous studies. MDD is known to be associated with higher levels of oxidative stress, such as increased MDA and AOPP levels, as well as lower levels of antioxidants like GSH and GSH peroxidase.^[25] Therefore, research on aMCI that includes patients with MDD may not provide a clear understanding of the true levels of oxidative stress and antioxidant biomarkers in aMCI.

MDA is formed through the degradation of arachidonic acid and other polyunsaturated fatty acids.^[44] It exhibits enhanced chemical stability and membrane permeability in comparison to reactive oxygen species (ROS) and demonstrates lower toxicity than other aldehyde compounds such as 4-hydroxy-2-nonenal (4-HNE) and methylglyoxal. In addition, several papers have proposed that MDA might offer protective effects, indicating that reduced MDA levels could potentially contribute to the pathophysiology of certain conditions. In fact, MDA can have a range of effects on the body, from potentially causing harm to offering protection.

MDA and 4-HNE function as signaling molecules that can stimulate gene expression and promote cell survival. However, they can also inhibit gene expression and contribute to cell death, as observed in studies conducted by Ayala et al.^[45] A recent study found that MDA acts as a signaling messenger, specifically influencing islet glucose-stimulated insulin secretion (GSIS) through the Wnt pathway. According to a study by Wang et al.^[46], it was discovered that MDA levels of 5 and 10 µM have a moderate effect on enhancing islet GSIS. These levels were also found to increase the ATP/ADP ratio and cytosolic calcium levels, as well as influence gene expression and the production of proteins/activity associated with key regulators of GSIS.

The response of aldehyde dehydrogenase, which is activated by oxidative stress in mammals, is of utmost importance in determining the consequences of both immediate and prolonged elevations in aldehydes, such as MDA.^[47] The outcome is contingent upon several factors, including the extent and duration of MDA elevations, as well as the reaction of ALDH to oxidative stress. In a study conducted by Ayala et al.^[45], it was shown that lower levels of MDA could potentially interfere with important redox mechanisms in the body. Moreover, in the case of human beings, MDA represents a distinct epitope linked to oxidation, which is present on lipoproteins and cells that are undergoing apoptosis.^[48] This elicits an intrinsic immune response facilitated by IgM antibodies, which plays a crucial role in upholding homeostasis.^[48,49]

It is interesting to observe that within the plant kingdom, there is a captivating occurrence where short-lasting small increases in MDA levels can initiate a signaling process. This signaling process is of utmost importance in the regulation of cellular redox balance, the activation of the antioxidant system, and the plant’s ability to survive in challenging environments.^[50] Continuous increases in MDA, on the other hand, lead to genotoxicity, a decrease in membrane fluidity, and various detrimental effects.^[50]

Based on this observation, it is possible that the decreased MDA concentrations in individuals with aMCI may be linked to a reduced ability of redox homeostatic processes.

Our study also unexpectedly revealed an elevation, rather than a reduction, in -SH groups among individuals diagnosed with aMCI. This association is consistent with a study that found higher thiol concentrations in MCI participants than in controls.^[51] Interestingly, elevated levels of thiol protein groups in individuals with depression are linked to reduced efficiency in declarative memory, working memory, and verbal fluency as well.^[52]

None of our subjects took particular supplements known to elevate thiol groups (known as thiol supplements), such as lipoic acid, bromelain, glucosamine, selenium, cysteine, glutathione, and N-acetyl cysteine.^[53] Therefore, the increased levels of -SH groups in aMCI are not induced by intake of such supplements. The question is whether increased thiol accumulation may have a pathophysiological role, and if this were the case, why could this predispose individuals to develop aMCI? Enhanced -SH aggregation might coincide with heightened levels of S-prenylation^[54] and increased glutathionylated protein levels^[55]. Furthermore, an elevation in S-nitrosylation due to increased nitric oxide levels could also occur.^[54] The latter mechanisms may exert deleterious effects, potentially contributing to age-related processes and neurodegeneration, including AD.^[55] It is important to emphasize that the prolonged administration of GSH or NAC to both young and aged animals may disrupt overall gene expression, suppress SKN-1-mediated transcription, and hasten the aging process.^[56] On the contrary, restricting the intake of dietary thiols may prolong lifespan.^[56] Homocysteine, a thiol compound, constitutes a part of the overall plasma thiol pool, which comprises low molecular weight thiols, cysteine, glutathione, and cysteinylglycine.^[57] Elevated levels of homocysteine are linked to neurocognitive deficits in older individuals and AD.^[58] Finally, high thiol antioxidant amounts could lead to increased oxidative damage, misfolded proteins, endoplasmic reticulum (ER) stress, and accelerated aging.^[56,59].

Our lipid profile results were more in line with the a priori hypothesis; there were lower HDL-C and ApoA concentrations and a higher ApoB/ApoA ratio in aMCI participants compared to controls. Collectively, these findings indicate lowered antiatherogenic but increased atherogenesis processes. A prior meta-analysis examining correlations between lipid profiles and aMCI revealed that elevated TC and LDL-C levels were associated with increased cognitive impairment.^[14] There was a trend towards increased TC levels in aMCI, whilst there were no significant changes in LDL-C, TG, and HDL-C. Plasma TG levels were found to be higher in MCI than control subjects in another case-control investigations.^[10,11,60] Metabolic syndrome biomarkers (TG/HDL-C ratio and lowered ApoA1) correlated with a swifter deterioration of cognitive functions in subjects with aMCI.^[13] Meta-analyses studies performed in AD showed that this illness is accompanied by increased TC and LDL-C^[14], and lowered serum/plasma ApoA1.^[11,12]

In addition to the observed associations between the lipid profiles and aMCI diagnosis, this study also found that the co-occurrence of aMCI and subclinical depressive symptoms was characterized by lowered HDL, ApoA, a reverse cholesterol transport index, and increased ApoB, ApoB/ApoA ratio, and a Castelli risk index 1. Increased ApoB/ApoA and Castelli risk index ratios are important predictors of increased risk of cardiovascular disease.^[61] Thus, subclinical depression already is sufficient in increasing atherogenicity when coupled with aMCI. Previous literature has mentioned that individuals diagnosed with MDD were found to exhibit significantly higher atherogenicity indices and a lowered reverse cholesterol transport (including HDL-C and ApoA1 levels) compared to control subjects.^[24,62-64] Moreover, Morelli et al.^[65] found that, in MDD, elevated atherosclerosis and insulin resistance were largely influenced by reactive oxygen and nitrogen species.

According to past research, atherogenicity is considered a risk factor for cognitive deficits. and elevated atherogenic lipoproteins have been associated with impaired cognition.^[66] Additionally, low global cognitive performance, memory, and executive function have also been reported in those with atherogenic dyslipidemia (characterized by elevated triglyceride levels, increased LDL concentrations, and reduced HDL concentrations).^[67] The relationship between dyslipidemia and cognitive deficits may be due to the role that dyslipidemia plays in the formation of atherosclerotic plaques, leading to ischemic injuries in the brain.^[21]