The study included individuals aged 20-40 years who were non-smokers and reported no drug use during the study period. Patients with AD (n=36; mean age ± SD, 23.3±6.5 years) and healthy controls (n=36; mean age ± SD, 21.9±4.3 years) were recruited from the student and staff population at Wenzhou-Kean University (Wenzhou, China) between September 2023 and May 2024 through voluntary responses to campus-wide announcements and personal referrals. Inclusion criteria for the patients with AD included having a clear medical record of the disease. Healthy controls had no history of AD, psoriasis or other skin conditions, and did not have a family history of AD. All participants (n=72) were included in phenotypic analyses. Reverse transcription-quantitative PCR (RT-qPCR) analysis was performed in four representative participants (2 controls and 2 patients with AD) to assess expression stability over time. Protein analysis by immunoblotting was performed on samples from 14 healthy controls and 26 patients with AD. Participant allocation across molecular analyses is summarized in Fig. S1.

At the time of buccal cell collection, a thorough examination was conducted to confirm that participants did not have any oral mucosal diseases, such as oral lesions or ulcers. Participants thoroughly rinsed their mouths with water and buccal cells were collected by gently chewing the inner cheek with 15 ml of sterile water, followed by spitting into a 50-ml tube. Centrifugation was performed using a Sorvall^™ ST 16 Centrifuge (Thermo Fisher Scientific, Inc.) equipped with a F15-6x100y Fixed Angle Rotor (Thermo Fisher Scientific, Inc.). To ensure complete sedimentation, samples were centrifuged at 10,956 x g (10,000 rpm) for 10 min at 4˚C. The pellets were then transferred to 1.5-ml tubes and centrifuged again at 21,100 x g (14,800 rpm) for 10 min at 4˚C using a FRESCO 21 centrifuge (Thermo Fisher Scientific, Inc.). It is noteworthy that mouth rinsing with a commercial oral rinse solution, as opposed to water, resulted in increased viscosity, thereby exacerbating the challenge of separating cell sediments from the supernatant. Pellet viscosity was assessed visually based on pellet edge clarity and compaction following centrifugation at multiple speeds (986, 3,944, 7,800 and 21,100 x g). Palmar hyperlinearity was evaluated using categorical visual scoring. No formal blinding procedures or objective rheological measurements were performed.

Total RNA was isolated using Trizol^™ (Thermo Fisher Scientific, Inc.) according to the manufacturer's recommendations. Following RNA extraction, DNase I (RNase-free; cat. no. D7076; Beyotime Biotechnology) treatment was performed to eliminate potential genomic contamination following the manufacturer's recommendations. Reverse transcription was performed using 1 µg of total RNA with the BeyoRT^™ III First Strand cDNA Synthesis Kit (cat. no. D7178; Beyotime Biotechnology) according to the manufacturer's protocol. RT-qPCR was conducted using SYBR Green qPCR Mix on a QuantStudio^™ 6 Flex system (Thermo Fisher Scientific, Inc.). The thermocycling conditions were as follows: Initial denaturation at 95˚C for 2 min, followed by 40 cycles of 95˚C for 15 sec and 60˚C for 1 min. A melting curve analysis was performed to verify amplicon specificity. Relative gene expression was calculated using the 2^-ΔΔCq method (18) and normalized to GAPDH as the internal reference gene.

To ensure reliable expression analysis, three primer sets were tested targeting the 5'-terminal region (PrimerBank 60097901c1, 5'-TGAAGCCTATGACACCACTGA-3' and 5'-TCCCCTACGCTTTCTTGTCCT-3'), 3'-terminal region (5'-TGAGGGCACTGAAAGGCAAA-3' and 5'-TGGCCACATAAACCTGGGTC-3'), and highly conserved internal repeat region (5'-CAGACACACARTCAGTGTCAG-3' and 5'-CAYGAATGGTGTCCTGACC-3') of FLG. However, only the 5' region primer set met the requirements for RT-qPCR analysis. FLG gene expression levels were normalized to GAPDH (PrimerBank 378404907c1, 5'-GGAGCGAGATCCCTCCAAAAT-3' and 5'-GGCTGTTGTCATACTTCTCATGG-3'). Gene expression was assessed in 4 representative participants (2 controls and 2 patients with AD). However, FLG mRNA levels showed substantial inter- and intra-individual variability across different collection time points (Fig. S2), with Cq values exhibiting poor reproducibility. Given this high variability and the exploratory nature of this analysis, qPCR was not extended to the full cohort, and subsequent investigations focused on protein-level characterization. Biological replicates (batch 1-batch 4) were collected over different months (September, October, November and December 2023) to assess expression stability.

For protein extraction, various buffer formulations with modified concentrations of detergents such as SDS, NP-40 and sodium deoxycholate were tested. Despite these efforts, none of the attempts resulted in satisfactory results. Direct cell lysis with SDS loading buffer was found to yield the highest quality results, with successful FLG protein detection in all samples tested. Briefly, buccal cells were collected, and the supernatant was carefully removed. For every 100 mg of cell pellet, 1 ml of 2X SDS loading buffer [25 mM Tris (pH 6.8), 2.5% SDS, 2% glycerol, 4% β-mercaptoethanol and 0.001% bromophenol blue] was added. The pellet was resuspended, boiled for 10 min, cooled on ice, and subsequently centrifuged at 21,100 x g (14,800 rpm) for 10 min at 4˚C.

For western blotting, 10 µl (for FLG detection) or 15 µl (for mucin detection) per lane of protein samples obtained from the previous steps were applied to 12% gels for FLG, 10% gels for mucin (MUC)1, MUC4, MUC7 and MUC16, and 7.5% gels for MUC5B and MUC19, and underwent SDS-PAGE. Proteins were transferred onto nitrocellulose membranes (Pall Corporation). Membranes were blocked with 5% BSA in Tris-buffered saline containing 0.1% Tween-20 (TBST) at room temperature for 1 h and washed with TBST (three times for 5 min each) between all incubation steps.

FLG was detected using two widely used antibodies: mouse anti-FLG antibody AKH1 (1:750, cat. no. sc-66192, Santa Cruz) and rabbit anti-FLG antibody Poly19058 (1:750; cat. no. 905804; BioLegend, Inc., raised against residues 24-39 of FLG). Membranes were first probed with AKH1 in blocking buffer overnight at 4˚C, followed by the appropriate IRDye-conjugated secondary antibody at room temperature for 1 h, and visualized using the ODYSSEY CLx imaging system (LI-COR Biosciences). The same membranes were then probed with Poly19058 overnight at 4˚C, followed by the appropriate secondary antibody at room temperature for 1 h, and visualized again. This sequential probing strategy took advantage of the dual-channel capacity of the Odyssey system. To confirm protein loading, membranes were then stripped with NewBlot^™ IR Stripping Buffer (cat. no. 928-40028; LI-COR Biosciences) according to the manufacturer's protocol. Complete signal removal was confirmed by re-scanning prior to reprobing. The stripped membranes were blocked again with 5% BSA in TBST at room temperature for 1 h and probed with rabbit anti-keratin 14 (KRT14) antibody (1:1,000; cat. no. A19039; ABclonal Biotech Co., Ltd.; KRT14) overnight at 4˚C, followed by goat anti-rabbit IRDye 800CW secondary antibody at room temperature for 1 h. KRT14 was selected as a loading control for FLG and mucin blots due to its stable expression as a structural protein in epithelial cells and as it is well separated from potential FLG monomer fragments. This selection follows established practice in the FLG field (6,19) and has been used in subsequent studies of FLG expression in epithelial tissues (20,21). For mucin detection, membranes transferred from 10% SDS-PAGE gels were probed sequentially with rabbit anti-MUC4 antibody (1:750; cat. no. A3438; ABclonal Biotech Co., Ltd.), rabbit anti-MUC7 antibody (1:1 000; cat. no. 26544-1-AP; Proteintech Group, Inc.), and, after stripping, mouse anti-MUC1 antibody (1:250; cat. no. sc-7313; Santa Cruz Biotechnology, Inc.) and rabbit anti-MUC16 antibody (1:500; cat. no. A4666; ABclonal Biotech Co., Ltd.). Membranes transferred from 7.5% SDS-PAGE gels were probed with mouse anti-MUC5B antibody (1:100; cat. no. sc-21768; Santa Cruz Biotechnology, Inc.) and rabbit anti-MUC19 antibody (1:500; cat. no. bs-17905R; BIOSS) under identical conditions as aforementioned. Secondary antibodies used were goat anti-mouse IRDye 680RD (cat. no. 926-68070; LI-COR Biosciences) and goat anti-rabbit IRDye 800CW (cat. no. 926-32211; LI-COR Biosciences), both at 1:15,000 dilution. Membranes were washed with TBST three times for 5 min after each antibody incubation.

Data were analyzed using GraphPad Prism (v8.0; Dotmatics). For categorical data in Fig. 1D, comparisons between healthy controls and patients with AD were performed using the two-tailed Fisher's exact test. P<0.05 was considered to indicate a statistically significant difference. No formal statistical comparisons were performed on the immunoblotting data (Figs. 2, S3 and S4) and the results are presented descriptively.

In the present study, 36 patients with AD and 36 healthy controls, aged 20-40 years, were recruited. Initial exploratory analyses, including phenotype observation, gene expression analysis and protein profiling, were performed in 2 healthy controls (H1 and H2) and 2 patients with AD (AD1 and AD2), with AD2 reporting a family history of AD.

PH, characterized by prominent palmar creases, is frequently associated with FLG deficiency (7,8). Among the representative participants, H2 and AD1 showed no evident PH, H1 exhibited mild PH, and AD2 displayed a pronounced PH phenotype (Fig. 1A and B). Notably, during buccal cell collection, a marked difference in pellet properties was observed between groups. Samples from the patients with AD consistently exhibited increased viscosity and poorly defined pellet boundaries, whereas control samples formed compact pellets with sharp edges (Fig. 1B and C). Even at high centrifugal force, AD-derived pellets remained diffuse and difficult to compact. Across the full cohort, mild PH was observed in 11% of healthy controls compared with 33% of patients with AD (Fig. 1D). Similarly, increased pellet viscosity was more frequently observed in AD samples at all tested centrifugation conditions.

To investigate potential contributors to this phenotype, mucin expression was assessed; however, no marked or consistent differences were observed between the AD and control groups (Fig. S3), suggesting that mucins are unlikely to account for the increased viscosity.

FLG gene expression was evaluated in a subset of participants (H1, H2, AD1, and AD2) across four independent collections over different months (Fig. S2). Expression levels were detectable in all samples but exhibited substantial intra- and inter-individual variability over time. Given the limited sample size and exploratory nature of this analysis, these data are presented descriptively to illustrate variability rather than for statistical inference. Subsequent analyses therefore focused on protein-level characterization.

To investigate FLG protein expression, two widely cited antibodies were utilized for FLG protein detection: AKH1 (mouse monoclonal) and Poly19058 (rabbit polyclonal, raised against residues 24-39 of filaggrin) (6,22-25). Across all participants, signals corresponding to putative FLG-derived fragments were consistently detected. By contrast, distinct differences in FLG degradation profiles were observed between patients with AD and healthy controls (Fig. 2). Notably, low-molecular-weight fragments, particularly around ~8 kDa, showed clear differences between groups and were consistently detected by both antibodies. Additional variability was observed in intermediate fragments (20-40 kDa), suggesting altered processing or degradation of FLG in AD samples.

Analysis of an expanded cohort (Fig. S4; Table SI) confirmed that FLG degradation profiles differed between patients with AD and healthy controls. Notably, the putative terminal ≤10 kDa fragment was less frequently detected in AD samples (19% with AKH1 and 4% with Poly19058) than in the healthy control samples (50% with both antibodies), while greater diversity in the 20-40 kDa region was observed in a subset of AD samples. Together, these findings indicate that while overall FLG protein expression remains stable in buccal cells, its degradation profile is altered in AD.