Association between transthyretin levels and Alzheimer's disease: A meta-analysis

Abstract

Background

Transthyretin (TTR) has been implicated in amyloid-β (Aβ) binding, cerebrospinal fluid homeostasis, and neurodegenerative processes in Alzheimer's disease (AD), but clinical evidence regarding its levels in AD and mild cognitive impairment (MCI) remains inconsistent.

Objective

This meta-analysis aimed to clarify the association between blood/cerebrospinal fluid (CSF) TTR levels and AD/MCI.

Methods

Following the PRISMA guidelines, this meta-analysis systematically identified and included 18 eligible studies from databases including PubMed, Web of Science, Embase, and Cochrane Library.

Results

Random-effects meta-analysis showed lower TTR levels in AD versus healthy controls (HC) (Hedges’ g = −0.77, 95% CI: −1.50, −0.04; p = 0.04), with no significant differences in MCI versus HC. Subgroup analyses confirmed a significant reduction in TTR levels across multiple dimensions: 1) sample types (serum); 2) population characteristics (age difference < 5 years and sex difference ≥ 10%); and 3) detection methods (immunoassay). Multivariate meta-regression with region, sample source, detection technology, and female ratio difference as covariates, accounted for 36.81% of the between-study heterogeneity. Sensitivity analysis verified result robustness, and publication bias was resolved after excluding an outlier study.

Conclusions

Lower TTR levels are significantly associated with AD. This finding suggests that TTR warrants further investigation as a candidate biomarker for AD, although prospective studies are needed to validate its clinical utility.

Introduction

Alzheimer's disease (AD), as a common neurodegenerative disorder, is featured by progressive memory decline and overall cognitive dysfunction. As the most common cause of dementia, AD is responsible for approximately 70% of all dementia cases worldwide, ¹ imposing a growing global burden with aging. According to estimates, about 57.4 million individuals globally had dementia in 2019, while this figure is projected to rise to 152.8 million by 2050. ² Although biomarker research for AD over the past decades has been dominated by amyloid-β (Aβ), tau, neurodegeneration, and inflammatory markers, no reliable and specific biomarkers have yet been successfully translated into clinical practice to enable early and accurate diagnosis of the disease. ³ Proteins involved in cerebrospinal fluid homeostasis, proteostasis, and peripheral metabolic regulation may provide complementary information. Given that understanding the underlying factors of AD is essential for developing effective interventions, there is an urgent need to identify potential biomarker candidates capable of reducing disease risk or delaying its onset.

Early diagnosis plays a pivotal role in improving AD prognosis, with biomarkers as the core support. The 2024 NIA-AA criteria have upgraded the ATN framework to ATNIVS. On the basis of the original three core biomarkers: A (amyloid deposition), T (tau pathology), and N (neurodegeneration), the new framework adds three supplementary biomarkers: I (inflammation), V (vascular brain injury), and S (α-synuclein), while emphasizing the application of blood-based biomarkers.^4,5 This upgrade promotes the transformation of AD diagnosis toward a multi-dimensional assessment paradigm and imposes higher standards for novel biomarkers. Thus, identifying high-quality biomarkers is vital for early screening, disease monitoring and optimizing therapeutic strategies in AD.

Transthyretin (TTR), a homo-tetrameric protein secreted within cerebrospinal fluid (CSF) and plasma, has emerged as a promising candidate in AD research due to its multifaceted biological functions relevant to neurodegeneration. First, TTR can directly bind to Aβ, inhibit its aggregation into neurotoxic fibrils, and promote Aβ clearance through the blood-brain barrier (BBB).^6–8 Second, TTR is primarily produced by the liver and the choroid plexus, thereby linking peripheral and CSF compartments. ⁹ Third, beyond its role in transporting thyroxine and retinol-binding protein, TTR is also involved in nutritional and inflammatory status; consequently, TTR may serve both as a mechanistic participant in AD pathogenesis and as a reflection of disease state. ¹⁰ These mechanistic insights suggest that TTR may serve as a potential biomarker for AD. However, clinical evidence regarding the association between TTR levels and AD, as well as its prodromal stage, mild cognitive impairment (MCI), remains inconsistent: several studies have reported that peripheral or CSF TTR levels are markedly decreased among AD patients compared to cognitively normal individuals,^11–13 while other studies have failed to detect such a difference.^14–16 Similarly, findings in MCI populations are conflicting—some research indicates that MCI patients have lower TTR levels than cognitively normal controls, ¹² whereas other studies have reported contradictory findings. ¹⁷

Therefore, the present meta-analysis aims to systematically evaluate whether AD or MCI is associated with alterations in TTR levels in blood and CSF, so as to provide evidence-based conclusions regarding the role of TTR in AD and clarify its potential as a biomarker for disease identification. Our results will help comprehensively understand TTR's role in AD and offer valuable guidance for future research directions in AD prevention, early diagnosis, and treatment.

Methods

Standard protocol approvals, registrations

The present meta-analysis was carried out following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines and had been registered at PROSPERO, with the registration number CRD420251218755.

Search strategy

Two researchers systematically identified English-language studies in several databases, including “PubMed”, “Web of Science”, “Embase”, and “Cochrane Library”. Each database corresponded to its respective search strategy, and the detailed search strategies were provided in the supplementary materials. The search was conducted until September 30, 2025.

Eligibility criteria

Only English studies were enrolled. There were no constraints on the publication year or healthcare setting type. Furthermore, studies below were enrolled: (1) investigations involving human participants; (2) research exploring the relation of CSF/blood TTR levels with AD/MCI incidence; (3) studies involving case group (AD/MCI patients) and control group (cognitively normal subjects); (4) application of established clinical criteria for AD/MCI diagnosis; (5) available TTR data.

Studies below were eliminated: (1) lack of adequate TTR concentration data; (2) solely involving an individual subject group; (3) duplicate publications of the same study; (4) non-human studies, conference abstracts, reviews, meta-analyses, case reports, commentaries or studies addressing irrelevant topics.

Data extraction

For title and abstract screening, a 5% sample was evaluated by three reviewers (B.Q. Li, F.Y. Li, and L.Z. Xu), discussed to reach consensus, and then the remaining results were assessed independently by each reviewer. Two researchers (B.Q. Li and F.Y. Li) reviewed full-texts and extracted data, with any discrepancies resolved through discussion among all authors.

Data collected were publication year, first author, country of participants, diagnostic criteria, sample sizes in patient/control groups, mean age and sex distribution of each group, apolipoprotein E (APOE) ε4 carriers frequency, dementia severity (Mini-Mental State Examination, MMSE), sample type (CSF, serum, plasma), detection technology (immunoassays or mass spectrometry (MS)), TTR concentrations and mean and standard deviation (SD) or median with minimum and maximum values or median with interquartile range (IQR) were recorded from the included studies. If studies stratified case or control groups into subgroups, the data of every subgroup were pooled as $n=n_1+n_2$ , $\overline{x}=\frac{n_1{\overline{x}}_1+n_2{\overline{x}}_2}{n_1+n_2}$ and $SD=\sqrt{\frac{(n_1-1)S{D}_1^2+(n_2-1)S{D}_2^2+\frac{n_1{n}_2}{n_1+n_2}({\overline{x}}_1^2+{\overline{x}}_2^2-2{\overline{x}}_1{\overline{x}}_2)}{n_1+n_2-1}}$ .

For studies reporting only medians and IQRs rather than means and SDs without minimum/maximum values, medians were deemed to be equivalent to means, and SDs were obtained by dividing IQR by 1.35. ¹⁸ Alternatively, if minimum/maximum values were provided, means and SDs were estimated following the method described by Hozo et al. ¹⁹

Risk of bias and study quality evaluation

Two researchers (B.Q. Li and F.Y. Li) assessed risk of bias and study quality. To ensure standardized evaluation, they used the Revised Risk of Bias Assessment Tool for Nonrandomized Studies (RoBANS 2), encompassing 7 critical domains: target group comparability, target group selection, confounding factors assessment, exposure measurement, assessor blinding, selective outcome reporting, and insufficient outcome data. ²⁰ Detailed findings from this assessment are shown in Supplemental Table 1.

Statistical analysis

For meta-analysis, the association between TTR levels and AD/MCI incidence was assessed by computing the pooled standardized mean difference (SMD) with corresponding 95% confidence intervals (CIs). Since the included studies reported TTR concentrations using a wide variety of units, SMD was chosen as the effect size measure to ensure comparability across studies. Specifically, Hedges’ g was used as the effect size metric, as it corrects for small sample bias and provides a more consistent estimate across studies compared to other effect size measures such as Cohen's d.

Given the anticipated heterogeneity across studies, a random-effects model was employed. Between-study variance (τ²) was estimated using the restricted maximum likelihood (REML) method. To provide more conservative confidence intervals that account for the uncertainty in the between-study variance estimate, the Hartung-Knapp (HK) adjustment was applied. The statistical significance of the pooled effect size was assessed using a Z test, with a significance level of p < 0.05 (two-tailed).

Inter-study heterogeneities were evaluated by Q test and I² statistic, with p < 0.1 upon Q test and I² > 50% suggesting obvious heterogeneity, so the random effects model should be applied to combine the data from the included studies. All meta-analytic procedures, including forest plot generation, were performed in STATA 17.0 (Stata Corporation, College Station, TX, USA).

To examine the potential impacts of age difference, sample type, and study region, subgroup analyses were carried out. Subgroup analyses were conducted based on the factors below: age difference (< 5, ≥ 5 years), sample type (CSF, plasma, serum), study region (Europe, Asia, etc.) as well as detection technology (immunoassays, MS). The age difference categories (<5 versus ≥5 years) were chosen because a 5-year interval represents a clinically meaningful period in cognitive aging research, ²¹ during which cognitive decline and biomarker changes—including TTR levels—progress measurably, and this threshold has been commonly adopted in previous meta-analyses to define adequate age matching between case and control groups. The female ratio difference categories (<10% versus ≥10%) were chosen because a ≥ 10% disparity in sex distribution between comparison groups is considered a meaningful imbalance in epidemiological studies, particularly given the higher prevalence of AD in females and the established modulatory effects of estrogen on TTR expression and neuroprotective function. This criterion also aligns with standard practice in meta-analyses of sex-related biomarkers, where 10% is widely used as a pragmatic indicator of clinically relevant sex distribution imbalance. R software (version 4.3.0) was adopted for subgroup analysis using random effects models.

Meta-regression with restricted maximum likelihood estimation (REML) was performed for identifying candidate covariates associated with between-study heterogeneity. In this meta-regression, effect size was designated as the dependent variable, and age difference (< 5, ≥ 5 years), female ratio difference (< 10%, ≥ 10%), sample source (CSF, plasma, serum), study region (Europe, Asia, etc.), and detection technology (immunoassays or MS) were defined as independent variables to explore the heterogeneity sources.

Sensitivity analysis was also performed for assessing whether our finding was stable, involving sequentially removing each individual study. The Begg's and Egger's - test was utilized to assess publication bias, and p < 0.05 indicated obvious publication bias. Besides, funnel plots were constructed for visual validation. Statistical analyses, involving, meta-analysis, meta-regression, subgroup analyses, as well as bias evaluation, were conducted in R software (version 4.3.0). p < 0.05 (two-tailed) suggested statistical significance across analyses.

Results

Study screening and features

From primary database search, 354 studies were obtained in PubMed, 809 in Web of Science, 1148 in Embase and 14 in Cochrane Library. Following systematic de-duplication and screening of titles and abstracts, 154 articles underwent full-text evaluation. After application of predefined eligibility criteria, 36 studies met the inclusion criteria, whereas 18 studies investigating TTR and AD were excluded. Figure 1 presents study selection flowchart.

Figure 1.

Flowchart of meta-analysis for AD. AD: Alzheimer's disease; TTR: transthyretin.

Data from the 18 studies included in this meta-analysis were sourced either from published literature or obtained directly from the authors when necessary. Key characteristics of the included studies were summarized in Table 1 and Supplemental Table 2. In one study where both CSF and blood samples were derived from the same cohort, ²² only CSF data were utilized to maintain statistical independence. The total sample comprised 1550 participants, including 727 AD cases, 253 MCI cases, and 570 healthy controls (HC). For patients across studies, their average age was 64–84.4 years for AD patients, 67.6–76 years for MCI patients, and 61–83.1 years for HC. The female ratios across diverse studies were 40–77.78 for AD patients, 41.10–76.19 for MCI patients, and 32.65–100.00 for HC.

Table 1.

General characteristics of including studies and participants.

Study	Sample size			Mean age			Gender (Male)			Mean TTR (±SD) concentrations				Diagnosis criteria
First author	AD	MCI	HC	AD	MCI	HC	AD	MCI	HC	AD	MCI	HC	unit
Elovaara et al., 1986 23	22	-	16	73.3	-	68.2	-	-	-	216 ± 47.4	-	256 ± 42.4	mg/L	criteria proposed by Roth in 1955
Davidsson et al., 1997 22	26	-	18	73.1	-	64.7	10	-	6	17.51 ± 2.95	-	18.17 ± 2.31	mg/L	NINCDS-ADRDA
Serot et al., 1997 24	40	-	41	74.2	-	76	13	-	13	17.49 ± 2.02	-	20.02 ± 2.45	mg/L	NINCDS-ADRDA
Merched et al., 1998 25	20	-	10	84.4	-	83.1	5	-	3	22.21 ± 6.30	-	34.52 ± 13.03	mg/L	NINCDS-ADRDA
Puchades et al., 2003 26	5	-	5	81	-	62	3	-	2	21,425 ±  12,812	-	68,603 ±  27,957	-	NINCDS-ADRDA
Gloeckner et al., 2008 27	23	-	19	64	-	64	11	-	7	15 ± 3	-	19 ± 2	mg/L	NINCDS-ADRDA
Hansson et al., 2009 28	35	-	29	75.3	-	73	12	-	0	17 ± 8	-	20 ± 3	mg/L	NINCDS-ADRDA
Schultz et al., 2010 14	59	-	13	69	-	70	25	-	5	20.0 ± 3.0	-	21.8 ± 4.1	mg/L	NINCDS-ADRDA
Han et al., 2011 11	111	-	90	73.9	-	62.62	41	-	59	585.64 ± 150.08	-	806.53 ± 180.04	mg/L	NINCDS-ADRDA
Ribeiro et al., 2012 12	56	55	41	73.6	71.6	74.2	26	24	18	207.1 ± 8.4	201.9 ± 7.6	237.3 ± 7.5	mg/L	NINCDS-ADRDA
Velayudhan et al., 2012 29	90	-	50	81.4	-	80.8	20	-	12	63.9 ± 0.1	-	82.3 ± 0.1	mg/L	NINCDS-ADRDA
Uchida et al., 2015 13	42	51	49	73.9	71.2	69.8	10	21	33	357.0 ± 122.1	445.0 ± 144.8	520.5 ± 144.8	mg/mL	DSM-IV-TR
Wang et al., 2016 30	4	4	4	75.7	76	74	2	2	1	2.96*108 ± 108	1.34*109 ± 8*108	5.55*108 ± 108	Mean peak area	NIA-AA
Heywood et al., 2018 31	13	-	15	72.1	-	61	5	-	5	12.13 ± 2.21	-	5.03 ± 1.92	pmol/100 µL	NINCDS-ADRDA
Cheng et al., 2018 32	98	-	101	78.76	-	78.33	34	-	35	457.61 ± 113.94	-	411.31 ± 128.43	ng/mL	NINCDS-ADRDA
Liu et al., 2019 33	4	49	10	77	67.6	62.6	1	21	2	23.4 ± 3.9	25.1 ± 5.9	22.2 ± 4.1	mg/dL	DSM-IV-TR
Tien et al., 2019 17	62	73	40	77.4	75.6	74.8	30	43	16	328.6 ± 66.5	360.2 ± 81.9	324.6 ± 61.4	ug/mL	NIA-AA
Eldem et al., 2022 34	17	21	19	72	73	64	8	5	6	0.519 ± 0.44	0.49 ± 0.41	0.99 ± 0.65	normalized [a.u.]	Swiss Memory Clinics

AD: Alzheimer disease; MCI: mild cognitive impairment; HC: healthy control; SD: standard deviation; NINCDS-ADRDA: National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer's Disease and Related Disorders Association; DSM-IV-TR: Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision; NIA-AA: National Institute on Aging-Alzheimer's Association.

Association between TTR levels and AD/MCI during comparisons

Our meta-analysis explored the association of TTR levels with AD incidence, based on data from 18 articles involving 727 AD patients alongside 570 HC. Velayudhan et al. (2012) showed an extreme SMD of −182.99, which markedly deviated from the range of other studies (SMD ranging from −4.65 to 3.35) (Supplemental Figure 1). Including this study resulted in substantial heterogeneity (I² = 96.6%, tau² = 1613.87) and a non-significant pooled SMD from the random-effects model (Hedges’ g = −10.23, 95% CI: −30.91 to 10.45, p = 0.31), which was primarily driven by this single outlier. The effect size was considered biologically implausible and exceeded the predefined outlier threshold (|SMD| > 5). Therefore, this study was excluded from the primary analysis.

After exclusion, the heterogeneity decreased substantially (I² = 93.2%, tau² = 1.739), and the decreased TTR levels were significantly related to an increased AD incidence. By adopting a random-effects model, the combined pooled Hedges’ g was −0.77 (95% CI: −1.50, −0.04; p = 0.04; I² = 93.2%) (Figure 2). Given the profound heterogeneity (I² = 93.2%), we interpret this pooled estimate as an average trend across diverse study settings rather than a precise biological constant. Notably, 13 out of 17 studies (76.5%) reported directionally consistent findings (lower TTR levels in AD patients), which represents the primary evidence supporting an association between reduced TTR and AD. The quantitative summary should be viewed with caution, and the conclusion relies mainly on this observed directional consistency. For MCI, only six studies contributed to the comparison with HC (253 MCI patients versus 570 HC). TTR showed a tendency to decrease, but the difference was not statistically significant [Hedges’ g = −0.68, 95% CI (−2.87, 1.52), p = 0.46; I² = 96.6%]. Given the limited number of contributing studies and the lack of statistical significance, these null findings should not be overinterpreted as definitive evidence that TTR is not altered in MCI. Instead, the current evidence base is insufficient to draw meaningful conclusions regarding TTR levels in the MCI stage.

Figure 2.

Meta-analysis using a random-effects model among AD/MCI/HC groups. AD: Alzheimer's disease; CI: confidence interval; HC: healthy controls; MCI: mild cognitive impairment; SMD: standardized mean difference.

Subgroup analysis

Of the 17 studies included in the analysis, six had an age difference ≥ 5 years, 10 had an age difference <5 years, and one study did not provide age information. Subgroup stratification by age difference demonstrated that TTR levels were closely related to AD in “age difference <5 years” subgroup [Hedges’ g = −1.13, 95% CI (−2.00, −0.26), p = 0.02; I² = 94.4%], with no significant association observed in “age difference ≥ 5 years” subgroup (Figure 3).

Figure 3.

Age difference subgroup analysis using a random-effects model between AD group and HC group. AD: Alzheimer's disease; HC: healthy controls; CI: confidence interval; SMD: standardized mean difference.

When conducting subgroup analyses by sample source (CSF, plasma or serum), we included nine studies with CSF data, four studies with serum data, and four studies with plasma data. The results of subgroup analysis demonstrated that TTR was significantly reduced in the AD group only for serum sample type [Hedges’ g = −1.00, 95% CI (−1.92, −0.08), p = 0.04; I² = 62.0%], not for cerebrospinal fluid and plasma samples. (Supplemental Figure 2). Furthermore, subgroup analyses stratified by country of participants were conducted, which divided studies in European, Asian, and American groups. For America, only one study was included. TTR levels were not related to AD in both European subgroup and Asia (Supplemental Figure 3).

Out of the 17 comparisons, samples from 13 studies were detected using immunoassay method, while only four using MS method. The “MS” subgroup was analyzed, revealing that TTR was not associated with AD [Hedges’ g = −0.36, 95% CI (−4.48, 3.77), p = 0.80; I² = 93.3%]. Nonetheless, the “Immunoassay” subgroup analysis indicated that TTR was significantly related to AD [Hedges’ g = −0.90, 95% CI (−1.53, −0.26), p = 0.01; I² = 93.6%] (Supplemental Figure 4). Among those 17 articles enrolled for meta-analysis, 15 studies included sex information. Such articles were classified as two groups by female ratio difference in AD versus HC: sex difference <10% (nine studies) and sex difference ≥10% (six studies). TTR levels were related to AD in “sex difference ≥10%” subgroup [Hedges’ g = −1.09, 95% CI (−1.66, −0.52), p = 0.004; I² = 50.5%]. Nonetheless, the “sex difference <10%” subgroup exhibited no relation [Hedges’ g = −0.35, 95% CI (−1.76, 1.06), p = 0.58; I² = 95.6%] (Supplemental Figure 5).

Meta-regression analyses

To explore potential sources of heterogeneity, we conducted a multivariate meta-regression with region, sample source, detection technology, and female ratio difference as covariates. The model accounted for 36.81% of the between-study heterogeneity (R² = 36.81%). The overall test of moderators was not significant (F = 1.91, p = 0.185), indicating that these variables collectively did not significantly predict effect size. At the individual variable level, only Asian region (versus Americas) was significantly associated with effect size (estimate = 7.06, p = 0.015), while other variables showed no statistical significance (all p > 0.05). Significant residual heterogeneity remained (QE = 59.66, p < 0.001, I² = 93.53%), suggesting that other unmeasured factors contribute to the observed heterogeneity (Supplemental Table 3).

Publication bias

To evaluate the potential for publication bias, we conducted Egger's regression test and Begg's rank correlation test. The Egger's test did not reveal significant funnel plot asymmetry (t = −0.65, p = 0.528). The Begg's test also showed no evidence of publication bias (z = 0.00, p = 1.000). Taken together, these results indicate that publication bias is unlikely to have substantially affected the pooled estimates (Figure 4).

Figure 4.

Begg's funnel plot for publication bias test of TTR level with AD risk. AD: Alzheimer disease; TTR: transthyretin.

Sensitivity analysis

We conducted sensitivity analysis for exploring how every included article affected pooled SMD through eliminating one dataset each time. No alterations were detected in combined SMD or result significance (Supplemental Figure 6), which verifies the resilience of the results to study-screening process.

Discussion

In this meta-analysis, 18 eligible studies were systematically synthesized, and we found that lower TTR levels were markedly related to a higher AD incidence, supporting an inverse association between TTR levels and AD risk. For MCI, the limited evidence (six studies) precludes definitive conclusions, and the non-significant results reflect insufficient power rather than evidence of no association; a declining tendency without statistical significance was observed in MCI versus HC. We found that serum exhibited significant associations. The subgroups with age difference < 5 years and sex ratio difference ≥ 10% exhibited significant associations. For detection methods, a significant association was observed in the immunoassay subgroup.

Most studies document decreased TTR concentrations in the serum,^11,13 plasma,^12,29 or CSF^28,30 of AD patients compared to age-matched healthy controls. We further found that age differences < 5 years can significantly reveal the association, while it lacks statistical significance when the age difference ≥ 5 years. Aging is closely associated with decreased TTR levels—with advancing age, the stability of TTR's tetrameric structure declines, ¹⁶ impairing TTR's physiological functions of binding Aβ and promoting Aβ clearance. ⁸ Therefore, when conducting TTR-related research on AD or MCI, it is crucial to prioritize strict age matching among cohorts.

The significant association between lower TTR levels and AD risk in the “sex difference ≥10%” subgroup suggests that sex may modulate the relationship between TTR and AD pathology. Estrogen enhances the neuroprotective effect of TTR, ³⁵ while decreased estrogen levels attenuate this protective effect, leading to more pronounced TTR reduction in female AD patients. ³⁶ This sex difference suggests that TTR may be more meaningful for the diagnosis of female AD, and it also highlights the importance of considering sex factors and conducting sex-stratified research in AD diagnosis and treatment. ³⁷

In the MS-based subgroup (n = 4 studies, 107 samples), the lack of association between TTR and AD may reflect limited statistical power due to small sample size. In contrast, the immunoassay subgroup (n = 13 studies, 1303 participants) showed a significant association. This highlights the importance of adequate sample size in MS studies to mitigate technical variability and ensure reliable detection of protein biomarkers like TTR. The divergent results across detection platforms also underscore that assay standardization is a prerequisite for comparing TTR findings across different studies.

TTR is a tetrameric protein predominantly secreted into plasma and CSF and is traditionally recognized as a carrier of thyroxine and retinol. Meanwhile, its critical relevance to AD pathogenesis has been robustly established—serving as the primary amyloid-binding protein in CSF and exerting anti-amyloidogenic and neuroprotective effects within the central nervous system. ³⁸ Several studies have proposed mechanistic explanations for TTR alterations during AD pathogenesis,^39,40 including that TTR regulates Aβ aggregation by preferentially binding monomeric Aβ, thereby depleting the pool of aggregation-competent monomers and destabilizing fibril formation. ⁴¹ This effect critically depends on the dissociation of TTR tetramers into monomers, which exposes Aβ-binding sites that are inaccessible in the tetrameric state.^42,43 TTR monomers further direct Aβ aggregation toward non-toxic, protease-sensitive amorphous aggregates instead of toxic oligomers and fibrils. ⁴² Additionally, TTR can bind to Aβ oligomers, suppress the primary and secondary nucleation processes of Aβ fibrillogenesis, and limit aggregate proliferation and toxicity. ⁷ Therefore, reduced TTR abundance or impaired TTR stability may weaken an endogenous anti-amyloid buffering mechanism.

In addition to the well-established role of TTR in Aβ metabolism, emerging evidence suggests a potential bidirectional interaction between AD and wild-type transthyretin amyloidosis (ATTRwt), a disorder characterized by systemic deposition of misfolded TTR tetramers. Both AD and ATTRwt are highly prevalent in the aging population, ⁴⁴ and their co-occurrence has raised interest in the crosstalk between systemic and compartmentalized amyloidogenesis. ⁴⁵ Post-mortem studies have detected TTR-derived amyloid deposits in cerebral vessels and leptomeninges of elderly individuals, ⁴⁶ and patients with hereditary ATTR may exhibit cognitive impairment that mimics AD. ⁴⁷ However, while TTR is neuroprotective against Aβ, misfolded TTR aggregation could theoretically exacerbate proteostatic stress. Whether ATTRwt modifies AD risk or progression, or whether AD-related changes in TTR stability predispose to systemic TTR aggregation, remains unresolved. Future studies are needed to clarify this interaction. Furthermore, TTR is also influenced by systemic nutrition, liver function, inflammation, aging and thyroid-retinol transport. Thus, a lower TTR level in AD may reflect a mixture of central nervous system changes, systemic metabolic state, and disease-related secondary effects.

The multivariate meta-regression analysis found that study region (specifically Asian region versus Americas) significantly moderated the effect size, while sample type did not reach statistical significance (all other covariates p > 0.05). This indicates that geographic context may influence the association between TTR levels and AD, conforming to previous results from the study based on UK Biobank participants. ⁴⁸ The observed inconsistencies in TTR levels across AD populations may be partially explained by ethnic-specific factors. While most studies reported reduced TTR in AD patients, elevated TTR levels were noted in certain Asian cohorts. Such discrepancies could arise from variations in genetic backgrounds (e.g., APOE allele frequencies), ⁴⁹ environmental exposures (e.g., dietary habits), ⁵⁰ or lifestyle. ⁵¹ The regional variation in TTR–AD associations also suggests that population-specific reference ranges may be needed if TTR is to be developed as a clinical biomarker. Moreover, the interaction between TTR and APOE ε4 (a major genetic risk factor related to late-onset AD, which accelerates Aβ aggregation and disrupts clearance pathways ⁵² ) warrants further investigation. TTR competitively binds Aβ and inhibits its neurotoxicity, and the presence of APOE ε4 may attenuate TTR's effects or alter its expression patterns. However, some included studies failed to stratify participants by APOE genotype, leaving potential confounding effects unaddressed, highlighting the need for future investigations to integrate genetic profiling to clarify these interactions.

The reliability of our findings was reinforced by the absence of publication bias and robust sensitivity analyses. The extreme outlier study by Velayudhan et al. was excluded because its plasma TTR data had an implausibly small standard deviation, which artificially inflated the effect size and heterogeneity. After excluding this study, the overall meta-analysis showed a significant reduction in TTR levels in AD patients. Nevertheless, subgroup analyses focusing solely on plasma or cerebrospinal fluid samples both yielded no significant differences, indicating that the overall significance was driven by serum rather than by plasma or CSF.

This meta-analysis demonstrates notable strengths, including a comprehensive literature search, PRISMA-compliant protocols, rigorous inclusion criteria, and advanced statistical modeling. However, key methodological constraints must be highlighted. First, substantial between-study heterogeneity was observed (I² = 93.2%), with meta-regression explaining only 36.81% of this variability, leaving 63.19% of the variance unaccounted for. This high heterogeneity is the core limitation of our meta-analysis and severely restricts the precision and generalizability of the pooled estimates. Residual heterogeneity likely stems from several unmeasured confounders: (1) disease severity differences — MMSE scores varied considerably across studies, and TTR levels may decline with disease progression; (2) APOE ε4 genotype distribution — as the strongest genetic risk factor for AD, APOE ε4 status strongly influences TTR expression, yet most primary studies did not report genotype-stratified data; (3) TTR post-translational modifications — oxidized or glycated TTR exhibits altered Aβ affinity compared to native TTR, and different assays may capture different TTR isoforms^53,54; (4) assay calibration differences — substantial batch-to-batch variation exists across ELISA platforms and kits; and (5) pre-analytical variables — differences in sample storage conditions, freeze-thaw cycles, and hemolysis can all affect measured TTR stability and concentration. (6) measured subgroup factors — as demonstrated in our subgroup analyses, differences in sample source, detection technology, study population, and baseline characteristics each contributed to the observed heterogeneity to varying degrees. For instance, serum samples yielded significant associations, whereas CSF and plasma did not; immunoassays showed significance while MS-based studies did not; and studies with better age/sex matching revealed clearer TTR-AD associations. These findings collectively suggest that the TTR-AD relationship is highly context-dependent, underscoring the need for careful matching and standardized detection protocols in future studies to minimize heterogeneity and improve cross-study comparability. (7) Several included studies demonstrated unclear or high risk of bias in important methodological domains such as confounding and outcome assessment. These quality issues may have potential impacts on the pooled effect estimates and the overall interpretation of findings. However, our meta-analysis did not identify any heterogeneity attributable to differences in study methodological quality. Nevertheless, the presence of some methodological concerns underscores the need for future primary studies to adopt rigorous confounding control strategies, standardized protocols for sample handling and TTR quantification, and blinded outcome assessment, in accordance with established reporting guidelines such as STROBE. Future studies should systematically collect and report these variables to enable more robust heterogeneity assessment and meta-regression adjustment. Second, TTR showed a decreasing trend in MCI, but the difference was not statistically significant, likely due to the small number of studies, high heterogeneity, and confidence intervals crossing zero. Therefore, we should not interpret this as evidence of no difference, but rather consider the findings inconclusive, necessitating large-scale longitudinal studies to clarify the value of TTR in MCI. Third, most included studies were cross-sectional, which fundamentally limits causal inference. It is impossible to determine whether reduced TTR levels are a cause, consequence, or merely a correlate of AD pathology. For example, reduced CSF TTR could reflect choroid plexus dysfunction secondary to neurodegeneration rather than a primary protective mechanism. ⁵⁵ Longitudinal studies should be carried out to clarify whether reduced TTR levels precede AD onset as a potential risk factor or emerge as a secondary consequence of disease progression. Fourth, the included studies reported TTR in diverse units (mg/L, mg/mL, ng/mL, μg/mL, pmol/100μL, and arbitrary normalized units [a.u.]). We employed SMD to enable statistical pooling across these scales, as SMD is dimensionless and eliminates unit dependency. Sensitivity analysis excluding the study reporting arbitrary units ³⁴ confirmed result robustness. However, we acknowledge that cross-unit back-calculation was not performed to verify absolute concentration equivalence, as such conversions assume identical assay calibration curves, which is biologically invalid across different antibody platforms. Future meta-analyses should prioritize studies using certified reference materials to enable both statistical pooling and clinical threshold determination.

Conclusion

This meta-analysis provides evidence that reduced TTR levels are significantly associated with an increased risk of AD, supporting an inverse association between TTR levels and AD pathogenesis. Subgroup analyses reveal that only decreased serum TTR is significantly related to AD, suggesting that serum TTR merits further validation as a candidate blood-based biomarker for AD, whereas no significant association is observed for cerebrospinal fluid or plasma TTR. However, given the observational nature of most included studies, the substantial between-study heterogeneity (I²=93.2%), and the limited evidence for MCI, these findings should be considered exploratory and hypothesis-generating rather than definitive. The persistent heterogeneity is partly attributed to uncontrolled confounding factors (e.g., APOE genotype stratification), indicating that the mechanism of TTR in AD is complex. Future studies should focus on standardized TTR detection protocols, APOE-stratified analyses, and large-scale longitudinal cohort studies to clarify the dynamic role of TTR in the pathogenesis of AD and to determine whether TTR can be validated as a clinically useful biomarker for AD.

Data availability statement

The research contributions reported herein are included in the article and Supplemental Material, with additional queries to be addressed to the corresponding author.