Levels of Cell Recycling Proteins in Skin Cells as Possible Blood Test Markers for Alzheimer's Disease

This study was a part of the ongoing Biobank Innovations for chronic Cerebrovascular disease With Alzheimer's disease Study (BICWALZS) and the Center for Convergence Research on Neurological Disorders [20]. Dermal fibroblast samples were obtained from the Biobank Center at Ajou University Hospital. The study cohort consisted of 9 AD patients (amyloid PET-positive) and 9 age-matched controls with non-demented mild cognitive impairment (amyloid PET-negative) [21,22]. Amyloid PET positivity was determined quantitatively using 18F-flutemetamol PET scans. Standard uptake value ratios (SUVRs) were calculated using the pons as a reference region. Global cortical 18F-flutemetamol retention was computed as the volume-weighted average SUVR from bilateral cortical regions including frontal, posterior cingulate, lateral temporal, parietal, and occipital lobes, based on the annotated anatomical labeling atlas. Participants were classified as amyloid-positive when their global cortical SUVR exceeded 0.634, a threshold established based on previous studies in the elderly Korean population [20]. To minimize potential confounding effects from vascular pathology, only subjects with White Matter Hyperintensities (WMH) scores 1 were included. WMH scores were rated visually by two psychiatrists and six neurologists. Subjects with significant psychiatric illness, other neurodegenerative disorders, or severe systemic diseases were excluded. The study protocol was approved by the Institutional Review Board of Ajou University Hospital (AJOUIRB-SUR-2021-038), and written informed consent was obtained from all participants or their legal representatives. All participants were Koreans of Eastern Asian ethnicity and were recruited from the Ajou University Hospital.

Genomic DNA was isolated from blood and single-nucleotide polymorphism (SNP) genotyping was conducted by DNA Link, Inc. using the Affymetrix Axiom KKORV1.0-96 Array (Thermo Fisher Scientific, Waltham, MA, USA) according to manufacturer's protocol. Apolipoprotein E4 (APOE4) was derived from rs429358 and rs7412.

The following primary antibodies were used: alpha-Tubulin (Abcam, Cambridge, UK; ab18251), APP-CTF (Abcam, Cambridge, UK; ab32136), Beta-site APP cleaving enzyme 1 (BACE1) (ProteinTech Group, Rosemont, IL, USA; 12807-1-AP), BCL2-associated athanogene proteins 2 (BAG2) (Bethyl, Montgomery, TX, USA; A304-751A), BCL2-associated athanogene proteins 3 (BAG3) (ProteinTech Group, Rosemont, IL, USA; 10599-1-AP), Cathepsin D (Calbiochem, San Diego, CA, USA; IM-03), Early endosome antigen 1 (EEA1) (BD Biosciences, San Jose, CA, USA; 610456), Glucose-regulated protein 78 (GRP78) (BD Biosciences, San Jose, CA, USA; 610978), Lysosome-associated membrane protein 2A (LAMP2A) (Abcam, Cambridge, UK; ab18528), Microtubule-associated protein 1 light chain 3 (LC3) (Sigma, St. Louis, MO, USA; L8918), Neighbor of BRCA1 gene 1 (NBR1) (ProteinTech Group, Rosemont, IL, USA; 16004-1-AP), Optineurin (OPTN) (Bethyl, Montgomery, TX, USA; A301-831A), PSEN2 (Bethyl, Montgomery, TX, USA; A304-342A), RAS-Associated Protein RAB7A (RAB7) (Cell Signaling Technologies, Danvers, MA, USA; 9367), Tax1-binding protein 1 (TAX1BP1) (ProteinTech Group, Rosemont, IL, USA; 14424-1-AP), Transcription factor E3 (TFE3) (Bethyl, Montgomery, TX, USA; A302-622), Toll-interacting protein (TOLLIP) (ProteinTech Group, Rosemont, IL, USA; 11315-1-AP), Ubiquitin (ProteinTech Group, Rosemont, IL, USA; 10201-2-AP).

Dermal fibroblasts were obtained from skin biopsies collected within four weeks of baseline assessment from consenting participants. Following established protocols [23], a 3 mm³ punch biopsy was performed on the upper inside arm under local anesthesia using aseptic techniques. The tissue was immediately transferred to pre-warmed culture medium and processed in the laboratory. Biopsy specimens were mechanically dissected into small pieces in a 100 mm culture dish with minimal medium to facilitate tissue attachment. After brief attachment (1 min), Dulbecco's Modified Eagle Medium (DMEM; GenDEPOT, Baker, TX, USA, CM001-050) supplemented with 20% fetal bovine serum (FBS; GenDEPOT, Baker, TX, USA, F0900-050) and 1% penicillin/streptomycin was added. Culture medium was changed every 2–3 days after the first week. Upon sufficient fibroblast outgrowth from tissue fragments, cells were trypsinized and subcultured (passage 1). Primary dermal fibroblasts were maintained at 37 °C in a humidified atmosphere containing 5% CO₂. Cells between passages 3–8 were used for all experiments to minimize potential effects of cellular senescence. Population doubling levels were monitored throughout the culture period.

Immunoblot analysis was performed as follows, based on our previously described protocols [24,25]. Fibroblasts were washed once with PBS and lysed in 1× SDS buffer (50 mM Tris-HCl pH 7.4, 2% SDS and 10% glycerol). Cell lysates were heated at 95 °C for 10 min. Equal amounts of protein were resolved by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto nitrocellulose membranes (GE Healthcare Life Science, Marlborough, MA, USA; 10600001). All samples were processed using the same protocol and reagents. Due to gel capacity and limited sample availability, some targets were assessed on separate gels. Equal sample loading was verified by Ponceau S (Biosesang, Yongin, Republic of Korea; P1031) staining. Membranes were blocked in 3% non-fat milk in Tris-buffered saline (TBS; 20 mM Tris-HCl, 137.5 mM NaCl, pH 7.6) with 0.05% Tween-20 (TBST) for 30 min and incubated with primary antibodies in TBST with 3% bovine serum albumin (BSA) at 4 °C overnight. Blots were washed three times with TBST, incubated with horseradish peroxidase-conjugated secondary antibodies (SouthernBiotech, Birmingham, AL, USA; 1031-05 and 4055-05) for 1 h, and washed three times with TBST. Clarity Western ECL Substrate (Bio-Rad Laboratories, Hercules, CA, USA) was used to detect immunoreactive proteins. Images were obtained and processed using a ChemiDoc XRS + System with Image Lab software v5.2 (Bio-Rad Laboratories, Hercules, CA, USA). Each gel included samples from both groups, and a common reference sample was included across gels to reduce between-gel variation.

Protein expression levels were quantified using densitometric analysis. Equal protein loading (8 μg per lane) was verified by Coomassie brilliant blue staining of parallel gels. Coomassie Brilliant Blue staining of a parallel gel was used solely for qualitative verification of comparable total protein loading and was not used for quantitative normalization. In addition, α-Tubulin was assessed as a loading control (Figure 1b) to confirm equivalent loading across lanes. Image Lab software v5.2 (Bio-Rad Laboratories, Hercules, CA, USA) was used for densitometric analysis of immunoblots. To minimize technical variation, control and AD samples were processed in parallel under identical conditions. For final quantification, the normalized densitometric values for each sample were expressed as relative expression levels by dividing by the mean value of the control group (n = 9). These relative expression values were used for statistical analyses and graphical presentation, with comparisons performed using the Mann–Whitney U test. Results are presented as median ± 95% confidence interval.

Autophagic flux was assessed by monitoring LC3-II accumulation in response to lysosomal inhibition. Cells were treated with either DMSO (control) or 400 nM bafilomycin A1 (Sigma, St. Louis, MO, USA; B1793) for 3 h before protein extraction. Autophagy flux was calculated as the ratio of LC3-II levels in bafilomycin A1-treated versus DMSO-treated cells. For LC3-II quantification, band intensities were first normalized to α-Tubulin within each lane, and autophagic flux was calculated as the ratio of normalized LC3-II levels in Baf. A1-treated versus DMSO-treated cells. LC3-II levels were analyzed using a two-factor mixed-effects model (GraphPad Prism 8.0), with treatment (DMSO vs. Baf. A1) as a within-subject repeated factor and group (control vs. AD) as a between-subject factor, and fibroblast ID entered as a random effect. Because the repeated factor had only two levels, sphericity assumptions were not applicable.

Lysosomal pH was evaluated using LysoTracker™ Red DND-99 (Invitrogen, Carlsbad, CA, USA; L7528). Cells were incubated with 800 nM LysoTracker for 30 min at 37 °C, then harvested by trypsinization and resuspended in PBS. Fluorescence intensity was measured using a BD FACSAria III flow cytometer, and at least 10,000 single-cell events were collected and analyzed per sample. Data analysis was performed using FACSDiva software 8.0.3 (BD Biosciences, San Jose, CA, USA).

Each biological replicate represents one subject-derived fibroblast line; assays were performed once per line (technical replicates = 1). Statistical analyses were performed using GraphPad Prism 8.0, IBM SPSS Statistics v25.0 and R software version 4.01. Normal distribution was assessed using the Shapiro–Wilk test. For demographic and clinical characteristics, categorical variables (sex and APOE4 carrier status) were analyzed using Fisher's exact test, while continuous variables (age, education duration, Clinical Dementia Rating (CDR) and CDR Sum of Boxes (CDR-SB)) were compared using Mann–Whitney U test. For protein expression analysis, group comparisons were conducted using the Mann–Whitney U test for non-parametric data. For LC3-II–based autophagic flux experiments, we analyzed normalized LC3-II levels using a two-factor mixed-effects model (GraphPad Prism 8.0), with Treatment (DMSO vs. bafilomycin A1) as a within-subject repeated factor and Group (control vs. AD) as a between-subject factor, including fibroblast ID as a random effect; the Group × Treatment interaction term was used as the primary test of differential flux responses between groups. Results are presented as median ± 95% confidence interval. Correlations between variables were evaluated using Spearman's rank correlation coefficient. For diagnostic performance assessment, receiver operating characteristic (ROC) curves were generated and area under the curve (AUC) values were calculated. Cutoff values for ROC curve analysis were determined using Youden's index, which maximizes the sum of sensitivity and specificity. Logistic regression analysis was performed to evaluate combinations of biomarkers. Pearson's chi-square test was used to evaluate categorical variables. In particular, for multivariable logistic regression models with two predictors, the events-per-variable (EPV) was well below commonly recommended thresholds, which further increases the risk of overfitting and overly optimistic estimates of model performance. Given the exploratory and hypothesis-generating nature of this pilot study, we did not apply correction for multiple comparisons across the 16 proteins analyzed. This decision reflects our aim to identify candidate biomarkers for future validation rather than establish definitive disease markers. Consequently, all reported p-values should be interpreted as descriptive statistics to guide hypothesis generation, and findings require confirmation in independent cohorts with appropriate multiple testing corrections. We acknowledge that the limited sample size (n = 18 total) constrains statistical power and increases the risk of both type I and type II errors. The diagnostic performance metrics (AUC, sensitivity, specificity) derived from ROC analysis may be subject to optimism bias due to the small sample size and lack of independent validation cohort. These preliminary performance estimates should be interpreted cautiously and require validation through cross-validation methods or independent external datasets in future studies. Statistical significance was set at p < 0.05 for descriptive purposes. All statistical tests were two-tailed.

In this exploratory case–control study, we compared 9 patients with AD confirmed by amyloid PET positivity and 9 age-matched controls with negative amyloid PET scans. To minimize the potential confounding effects of cerebrovascular pathology, we selected participants with white matter hyperintensities (WMH) scores ≤ 1. The mean age was comparable between groups (AD: 68.2 ± 5.3 years, Control: 69.8 ± 4.5 years, p = 0.395), and both groups showed a similar gender distribution with female predominance (Table 1). Education duration was similar between groups (AD: 6.8 ± 6.4 years, Control: 9.4 ± 2.9 years, p = 0.161). Notably, APOE4 carriers were more prevalent in the AD group (77.8%, 7/9) compared to controls (22.2%, 2/9; p = 0.057). As expected, AD patients exhibited significantly higher CDR and CDR-SB scores than controls (p < 0.001), confirming their clinical disease status.

To investigate potential alterations in the ALP, a key pathogenic mechanism in AD [26], we conducted a comprehensive exploratory analysis of AD- and ALP-associated proteins in patient-derived skin fibroblasts from our small cohort (n = 9 per group). The selected proteins represent key components of distinct but interconnected cellular pathways: (1) AD pathology-related proteins: APP, BACE1, PSEN2 [27,28]; (2) Autophagy machinery: BAG2, BAG3, EEA1, LC3, NBR1, OPTN, RAB7, TAX1BP1, TFE3, TOLLIP, Ubiquitin [29,30,31,32,33,34,35,36,37,38,39]; (3) Lysosomal components and regulators: LAMP2A, Cathepsin D, GRP78 [34,37,38].

Equal protein loading was confirmed by Coomassie brilliant blue staining and immunoblot of α-Tubulin (Figure 1a,b). Quantitative immunoblot analysis revealed significant alterations in two proteins (Figure 1c). Most notably, BACE1 levels were reduced by 47.2% in AD fibroblasts compared to controls (p = 0.022) (Figure 1d). Additionally, we observed a 28.5% increase in TAX1BP1 levels in AD fibroblasts (p = 0.035) (Figure 1e). TAX1BP1, which facilitates autophagic clearance of protein aggregates [36], was increased in AD fibroblasts compared with controls (p = 0.035).

Several other proteins demonstrated consistent directional changes across individual AD patient samples, though these did not achieve statistical significance in our cohort. Specifically, OPTN showed a 25% increase (p = 0.078) and BAG2 exhibited a 32% decrease (p = 0.052) in AD fibroblasts. Notably, examining individual patient data revealed that 8 out of 9 AD patients showed elevated OPTN expression relative to the control group median, and 8 out of 9 AD patients showed reduced BAG2 expression (Supplementary Materials Figure S1). Although our limited sample size (n = 18) prevented these differences from reaching statistical significance, consistent directional trends were observed across individual samples, and these preliminary observations require validation in larger independent cohorts. The remaining analyzed proteins showed no significant differences or consistent patterns between groups (Table 2 and Supplementary Materials Figure S1).

Given the crucial role of autophagy in protein quality control and its dysfunction in AD [26], we examined autophagic function in patient fibroblasts using two complementary approaches in this preliminary investigation. First, we assessed autophagic flux by measuring LC3-II accumulation in response to bafilomycin A1, a specific inhibitor of vacuolar H+-ATPase that blocks autophagosome–lysosome fusion. Autophagic flux was assessed only in lines for which paired DMSO and bafilomycin A1 measurements could be obtained. Due to limited patient-derived material/cell yield, three control lines and one AD line could not be assayed under both conditions and were excluded, leaving n = 6 (control) and n = 8 (AD). Due to limited patient-derived material/cell yield, one AD line could not be included in the LysoTracker assay, resulting in control (n = 9) and AD (n = 8) for this analysis. The mixed-effects model revealed a significant main effect of Treatment (F(1,12) = 30.33, p = 0.0001), indicating robust LC3-II accumulation upon Baf. A1 treatment across groups. The main effect of Group was not significant (F(1,12) = 0.40, p = 0.54). Importantly, the Group × Treatment interaction was not significant (F(1,12) = 2.00, p = 0.18), suggesting that the magnitude of LC3-II accumulation was comparable between control and AD fibroblasts. Notably, we found no significant difference in autophagic flux between fibroblasts from AD patients and controls (Figure 2a,b), suggesting that the baseline rate of autophagosome formation remains intact in AD patient cells.

We then evaluated lysosomal function by measuring lysosomal acidification using LysoTracker™ Red DND-99, a fluorescent dye that specifically accumulates in acidic organelles. Proper lysosomal acidification is essential for the activity of lysosomal hydrolases and efficient protein degradation [40]. Flow cytometric analysis of over 10,000 cells per sample revealed comparable LysoTracker fluorescence intensities between fibroblasts from AD patients and controls (Figure 2c and Figure S2). Consistently, mean LysoTracker fluorescence intensity did not differ significantly between groups (Control: median 3792, 95% CI 2981–4646, n = 9; AD: 3699, 95% CI 1260–5234, n = 8; Mann–Whitney U test, p = 0.870; Figure 2c and Table 2). At the current sample size, we did not detect statistically significant differences in lysosomal acidification between groups, and these negative findings should be interpreted with caution.

To gain insights into the potential interplay between AD- and ALP-associated proteins, we performed exploratory correlation analysis of protein expression levels across all samples (n = 18) (Figure 3 and Supplementary Materials Table S1). Particularly notable was the significant positive correlation between APP-CTF and BACE1 levels (Figure 3b), which aligns with the established role of BACE1 in APP processing [28]. We observed an unexpected negative correlation between APP-CTF and NBR1 (Figure 3c). While NBR1 is known to participate in selective autophagy, this finding suggests it may also play a previously unrecognized role in APP-CTF clearance.

BAG3 emerged as a central player, showing strong positive correlations with multiple autophagy regulators including OPTN and TOLLIP (Figure 3g,i). These associations reinforce the crucial function of BAG3 in selective autophagy pathways, as previously suggested by studies of protein quality control [29]. The correlation between pre-cathepsin D and mature-cathepsin D levels likely reflects normal lysosomal protease processing, indicating preserved lysosomal function in patient fibroblasts (Figure 3j).

Particularly interesting was the strong negative correlation between pre-cathepsin D and TFE3 (Figure 3m), which appears to contradict the role of TFE3 in promoting lysosomal biogenesis [31]. This unexpected finding warrants further investigation into potential feedback mechanisms regulating lysosomal function. Our analysis also revealed significant correlations involving endosomal trafficking components. EEA1, a key regulator of early endosomal dynamics, showed positive correlations with PSEN2, RAB7, and TFE3, suggesting coordinated regulation of endosomal-lysosomal trafficking in these cells (Figure 3n–p). TOLLIP exhibited multiple significant correlations with other proteins in our dataset, suggesting it may serve as an important coordinator of autophagy-related protein networks (Figure 3e,i,s). This observation extends beyond TOLLIP's known function in selective autophagy [33] and points to a potentially broader role in maintaining cellular proteostasis. Given the small sample size (n = 18), some scatter plots (e.g., in Figure 3d,k,n,p,r) contain visually prominent observations that can substantially influence Spearman's correlation coefficients and p-values. We regard these correlations as exploratory, hypothesis-generating findings, and we did not exclude any data points as statistical outliers.

To explore the potential clinical relevance of the observed protein alterations in our pilot cohort, we analyzed correlations between AD- and ALP-associated features and clinical/demographic data (Figure 4a and Supplementary Materials Table S1). This exploratory analysis revealed associations with CDR-SB scores, which provide a detailed measure of dementia severity across multiple functional domains [41]. Two proteins showed significant positive correlations with CDR-SB: GRP78 (r = 0.473, p = 0.047) (Figure 4b) and TAX1BP1 (r = 0.546, p = 0.019) (Figure 4c). However, when analyzed within individual groups, these correlations showed different patterns: in controls, GRP78 (r = 0.341, p = 0.369) and TAX1BP1 (r = 0.883, p = 0.003), and in AD patients, GRP78 (r = 0.806, p = 0.012) and TAX1BP1 (r = −0.513, p = 0.162). Given our limited sample size (n = 18), these correlation analyses should be interpreted cautiously and require validation in larger cohorts.

The correlation between GRP78 and disease severity is notable, as this ER chaperone protein plays crucial roles in both protein folding and stress-induced autophagy [38]. The observed relationship suggests that cellular stress responses may escalate with advancing clinical symptoms. Similarly, the positive correlation between TAX1BP1 and CDR-SB might reflect increasing activation of selective autophagy mechanisms as disease progresses. These findings suggest that certain stress-response pathways in peripheral tissues may parallel the progression of cognitive decline.

To assess the preliminary diagnostic potential of the observed protein alterations in this small exploratory cohort, we performed receiver operating characteristic (ROC) curve analysis (Figure 5 and Table 3). Among individual proteins, BACE1 demonstrated strong discriminatory power (AUC = 0.802, sensitivity 100%, specificity 55.6%), matching the performance of TAX1BP1 (AUC = 0.802, sensitivity 66.7%, specificity 88.9%). BAG2, while showing lower overall accuracy (AUC = 0.753), achieved notably high sensitivity (100%) at the expense of specificity (55.6%, p = 0.038) (Figure 5a).

To explore whether protein combinations might improve diagnostic accuracy, we evaluated various protein combinations using logistic regression analysis (Figure 5b). The combination of BAG2 and OPTN showed promising preliminary performance (AUC = 0.963, sensitivity 77.8%, specificity 100%), surpassing all individual markers in this dataset. Similarly, combining BACE1 and TFE3 achieved excellent discriminatory power (AUC = 0.926, sensitivity 77.8%, specificity 100%). These exploratory results suggest that multi-protein panels may provide more reliable diagnostic performance than single markers, though validation in independent cohorts is essential to confirm these findings. However, these multivariable models were derived from only 18 subjects and have a very low EPV for two predictors, so their ROC estimates are likely to be overly optimistic and prone to overfitting. Accordingly, all multivariable models in this study (e.g., BAG2 + OPTN, BACE1 + TFE3, APOE4 + BACE1) should be interpreted strictly as exploratory, hypothesis-generating tools rather than definitive diagnostic models.

Given the established role of APOE4 as a major genetic risk factor for AD [42], we investigated whether incorporating APOE4 status could further enhance diagnostic accuracy. Indeed, combining APOE4 with individual protein markers consistently improved performance, with APOE4 + BACE1 showing particularly strong results (AUC = 0.914, sensitivity 88.9%, specificity 88.9%,). Notably, the combination of APOE4 with autophagic flux measurements also demonstrated robust diagnostic potential (AUC = 0.896, sensitivity 75.0%, specificity 100%,), suggesting that functional autophagic measures may complement genetic risk assessment.

These preliminary findings suggest that fibroblast-derived protein markers, particularly when used in combination with APOE4 status, may represent promising candidates for AD diagnostic tools pending validation. The observed performance of protein combinations in this small cohort suggests that capturing multiple aspects of disease-related cellular dysfunction may be important for diagnostic accuracy, though these findings require confirmation in larger, independent datasets to assess their true diagnostic value and generalizability.