Skip to main content
NCBI home page
Frontiers in Molecular Biosciences logoLink to Frontiers in Molecular Biosciences
. 2026 Jun 23;13:1814666. doi: 10.3389/fmolb.2026.1814666

A bioinformatics pipeline for identifying clinically relevant immune and adhesion biomarkers in thyroid risk stratification

Larissa Teodoro Rabi 1,2, Karina Colombera Peres 1,3, Elisangela De Souza Teixeira 1, Alfio José Tincani 4, Natassia Elena Bufalo 1,3,5, Laura Sterian Ward 1,*
PMCID: PMC13337396  PMID: 42416487

Abstract

The substantial biological heterogeneity of thyroid cancer, particularly within follicular-patterned lesions, underscores the need for improved molecular tools for risk stratification. Genetic variability in the pathways regulating cell adhesion, immune interactions, and extracellular matrix remodeling may influence tumor behavior and the complexity of diagnosis. In this study, we conducted integrated in silico and genetic analyses to evaluate the role of polymorphisms in genes encoding immunoglobulin superfamily adhesion molecules, integrins, junctional proteins, matrix metalloproteinases, and extracellular matrix-associated proteins. Using multiple bioinformatics platforms, we screened 407,812 polymorphisms across 22 candidate genes and prioritized 133 variants with high predicted functional impact. Eight selected single-nucleotide polymorphisms were genotyped in a cohort of 648 individuals, including patients with benign thyroid nodules (n = 152), malignant thyroid nodules (n = 171), and healthy controls (n = 325). Clinical validation revealed that MADCAM1 rs3745925 significantly distinguished follicular adenoma from controls (p = 0.017, OR: 3.15; 95% CI: 1.63–6.05), goiter (p = 0.019, OR: 3.18; 95% CI: 1.49–6.85), and papillary thyroid carcinoma (p = 0.009, OR: 3.49; 95% CI: 1.73–7.09). This association remained robust after Bonferroni correction, underscoring its potential as a priority candidate. Additionally, ITGAM rs1143683, ITGAL rs2230433, and ICAM1 rs5498 were associated with tumor multifocality (p = 0.0428, p = 0.0008, and p < 0.0001, respectively). These findings demonstrate the feasibility of integrating bioinformatics-driven variant prioritization with clinical validation methods. Among the evaluated polymorphisms, MADCAM1 rs3745925 emerged as a promising auxiliary biomarker warranting further evaluation for the characterization of follicular-patterned thyroid lesions.

Keywords: bioinformatics, cell adhesion molecules, follicular thyroid lesions, in silico analysis, molecular biomarkers, single nucleotide polymorphisms, thyroid cancer

1. Introduction

Thyroid cancer (TC) is the most common endocrine malignancy worldwide, with an increasing incidence (Paz-Ibarra et al., 2024; Forma et al., 2025). Although the overall prognosis of differentiated thyroid cancer (DTC) is favorable, a significant subset of patients develops an aggressive disease characterized by local invasion, metastasis, and therapeutic resistance (Forma et al., 2025; Houten et al., 2023). This clinical heterogeneity underscores the need for better molecular stratification tools to distinguish indolent tumors from aggressive ones (Forma et al., 2025).

Genetic susceptibility significantly contributes to interindividual variability in both the risk and clinical outcomes of the disease (Tiucă et al., 2023). Single nucleotide polymorphisms (SNPs) are the most common form of human genetic variation (Saenko and Rogounovitch, 2018). Coding-region SNPs can alter protein structure and function, whereas non-coding variants may influence gene expression and RNA processing (Cargill et al., 1999). Increasing evidence suggests that polymorphisms in genes governing cell adhesion, migration, and extracellular matrix (ECM) remodeling can modulate tumor aggressiveness in DTC, making functionally relevant SNPs promising candidates for use as diagnostic and prognostic biomarkers (Saenko and Rogounovitch, 2018).

Thyroid carcinogenesis and progression are driven by dysregulated cellular processes, including adhesion, migration, and invasion, which depend on complex interactions with the tumor microenvironment (Andriescu et al., 2018; Rabi et al., 2023; Zhao et al., 2019). Cell adhesion molecules (CAMs) are central to these interactions. This heterogeneous group of transmembrane glycoproteins mediates cell-cell and cell-extracellular matrix interactions, governing tissue architecture, signaling, and cell behavior (Ley et al., 2007). Altered expression or function of these proteins can disrupt tissue organization and facilitate dissemination (Andriescu et al., 2018; Rabi et al., 2023; Smart et al., 2021; Chriqui et al., 2025; Burčík et al., 2025).

Within the CAM superfamily, immunoglobulin superfamily CAMs (IgSF-CAMs), such as ICAMs, VCAMs, PECAM-1, MAdCAM-1, and JAMs, are key regulators of leukocyte trafficking, endothelial permeability, and inflammation (Ley et al., 2007; Guerra-Espinosa et al., 2024; Singh et al., 2023). Dysregulated IgSF-CAM expression has been implicated in multiple steps of cancer progression, including invasion, metastasis, and immune evasion (von Lersner et al., 2019; Wai Wong et al., 2012). In thyroid cancer, aberrant adhesion molecule expression is linked to the loss of epithelial polarity, architectural disorganization, and invasive phenotypes, highlighting their potential as biomarkers (Rabi et al., 2023; Ma et al., 2024; Wan et al., 2022; Man et al., 2025; Kim et al., 2023; Yu et al., 2023).

Integrins constitute another major family of adhesion receptors that are critical to tumorigenesis. These heterodimeric transmembrane proteins, composed of α and β subunits, mediate bidirectional signaling between the ECM and cytoskeleton, regulating adhesion, migration, proliferation, and survival (Ley et al., 2007; Bergonzini et al., 2022; Mezu-Ndubuisi and Maheshwari, 2021; Hamidi and Ivaska, 2018). The β2 integrin subfamily (ITGB2), associated with α subunits such as ITGAL, ITGAM, and ITGAD, is primarily leukocyte-specific and mediates interactions with IgSF-CAMs in inflammatory and tumor microenvironments (Guerra-Espinosa et al., 2024). Similarly, ITGA4 contributes to lymphocyte homing and endothelial adhesion, whereas ITGA10 is involved in collagen binding and tumor-stroma interactions (Li et al., 2025). Altered integrin expression and signaling promote epithelial-mesenchymal transition, motility, and invasion in thyroid tumors (Yuan et al., 2023).

ECM remodeling, a hallmark of tumor progression, facilitates invasion, angiogenesis, and metastasis. Matrix metalloproteinases (MMPs), particularly MMP-9, degrade ECM components and are associated with aggressive features of thyroid cancer, including extrathyroidal extension and lymph node metastasis (Yuan et al., 2023; Winkler et al., 2020). MMP activity is tightly regulated by tissue inhibitors of metalloproteinases (TIMPs), and an MMP–TIMP imbalance creates a permissive microenvironment for cancer progression. Other ECM-associated proteins, such as thrombospondins, further modulate cell adhesion, angiogenesis, and tumor–microenvironmental crosstalk (Niland et al., 2022; Brassart-Pasco et al., 2020).

Given their established roles in tumor dissemination and accessibility for molecular profiling, genes encoding adhesion molecules, integrins, and extracellular matrix components are promising targets for biomarker discovery. Therefore, this study aimed to systematically evaluate the functional impact of polymorphisms in these gene families using an integrated in silico framework complemented by validation in clinical thyroid nodule specimens.

2. Materials and methods

2.1. In silico analysis

This study comprised an in silico analysis of genetic polymorphisms and their impact on DNA and/or protein structure using multiple bioinformatics tools for the functional, structural, and stability characterization of the evaluated alterations, as previously described by our group (Rabi et al., 2024; Peres et al., 2021; Nascimento et al., 2023; Viana et al., 2025; Bufalo et al., 2021; Martins et al., 2021).

Twenty-two genes were selected for analysis (Figure 1), including immunoglobulin superfamily cell adhesion molecules (ICAM1, ICAM2, ICAM3, MADCAM1, PECAM1, VCAM1, JAM3, and F11R), integrins (ITGA4, ITGA10, ITGAD, ITGAL, ITGAM, and ITGB2), extracellular matrix remodeling proteins (MMP9, TIMP1, TIMP2, and THBS1), cell surface receptors (CD44), intercellular regulatory proteins (LCN2), and secreted immune/inflammatory mediators (NTAN1 and RASSF5).

FIGURE 1.

Panel A displays a protein-protein interaction network with nodes labeled by gene names and interconnected by multicolored lines. Panel B categorizes genes into functional groups: Ig superfamily CAMs, integrins, ECM regulators, cell-surface receptor, intracellular regulatory proteins, and secreted mediators, each with associated gene names and simplified gray or colored molecular icons for visual representation.

Open in a new tab

Protein–protein interaction (PPI) network and functional classification of adhesion and ECM-related proteins. (A) PPI network highlighting interactions among CAMs, integrins, ECM remodeling regulators, and associated proteins (String Tool). (B) Functional grouping of genes evaluated in this study.

SNPs data were obtained from the Single Nucleotide Polymorphism Database (dbSNP) (Phan et al., 2025) of the National Center for Biotechnology Information (NCBI). For each gene, information was collected on SNP identifier (rsID), chromosomal position, reference, and polymorphic alleles of the gene. Structural and functional information of the proteins encoded by the genes of interest was obtained from the Universal Protein Resource Database (UniProt) (UniProt Consortium, 2025), including primary amino acid sequence (FASTA), functional domains, subcellular localization, biological function, and three-dimensional structures modeled on the AlphaFold (Fleming et al., 2025) platform.

2.2. Analysis of SNPs in non-coding regions

SNPs in non-coding regions (Table 1), including promoters, enhancers, 5′and 3′UTRs, and intronic sites, were analyzed using PredictSNP 2.0 (Bendl et al., 2016), which integrates multiple predictive algorithms: Combined Annotation Dependent Depletion (CADD) (Kircher et al., 2014), deep learning annotation of genetic variants (DANN) (Quang et al., 2015), functional analysis through hidden Markov models-multiple kernel learning (FATHMM-MKL) (Shihab et al., 2015), FunSeq2 (Fu et al., 2014), and genome-wide annotation of variants (GWAVA) (Ritchie et al., 2014).

TABLE 1.

Computational tools and databases employed for in silico functional prediction and structural analysis of genetic variants.

ID Access link Purpose/Prediction Methodology
dbSNP https://www.ncbi.nlm.nih.gov/snp/ Public archive for genetic variation and SNPs Curated database integrating submissions from researchers and large-scale sequencing projects
UniProt https://www.uniprot.org/ Protein sequence and functional information database Curated database combining Swiss-Prot and TrEMBL
AlphaFold https://alphafold.ebi.ac.uk/ Provides highly accurate protein structure predictions Deep learning system using NN
PredictSNP2.0 https://loschmidt.chemi.muni.cz/predictsnp2/ Consensus classifier for predicting pathogenicity of variants in non-coding and coding regions Meta-predictor
CADD Scores deleteriousness of SNPs Integrates multiple annotations using SVM
DANN Pathogenic effects of SNPs in non-coding regions Deep neural network
FATHMM-MLK Functional consequences Combined hidden Markov models with machine learning
FunSeq2 Prioritizes functional variants in non-coding regions Integrates genomic context and functional annotations using weighted scoring system
GWAVA Predicts impact of non-coding genetic variants Random forest classifier
PredictSNP1.0 https://loschmidt.chemi.muni.cz/predictsnp1/ Consensus classifier for missense variant pathogenicity prediction Meta-predictor
MAPP Impact of amino acid substitutions on protein function Physicochemical variation analysis in multiple sequence alignments
PhD-SNP Disease-associated missense mutations SVM
PolyPhen-1 Functional effects of amino acid substitutions Sequence-based and structure-based features with naive bayes classifier
PolyPhen-2 Impact of amino acid substitutions on protein structure and function Machine learning
SIFT Functional effects of amino acid substitutions Sequence homology-based approach using normalized probabilities from sequence alignments
PANTHER Functional effects of amino acid substitutions Phylogenetic-based approach combined with hidden markov models
SNAP Functional effects of amino acid substitutions NN trained on protein sequence and evolutionary information
DynaMut2.0 http://biosig.unimelb.edu.au/dynamut2/ Impact of amino acid substitutions on protein stability and dynamics Graph-based signatures integrating structural flexibility
MuPRO http://mupro.proteomics.ics.uci.edu/ Impact of amino acid substitutions on protein stability SVM
iStable http://predictor.nchu.edu.tw/iStable/ Impact of amino acid substitutions on protein stability Meta-predictor
NetSurf https://services.healthtech.dtu.dk/services/NetSurfP-3.0/ Surface accessibility and secondary structure of proteins Deep NN
ConSurf https://consurf.tau.ac.il/ Evolutionary conservation analysis Phylogenetic analysis of homologous sequences
NetPhos3.1 https://services.healthtech.dtu.dk/services/NetPhos-3.1/ Serine, threonine, and tyrosine phosphorylation sites NN trained on experimentally verified phosphorylation sites
Musite http://musite.net/ PTM sites including phosphorylation SVM with disorder and evolutionary information
ProP1.0 https://services.healthtech.dtu.dk/services/ProP-1.0/ Arginine and lysine propeptide cleavage sites NN trained on cleavage site sequences
NetNGlyc1.0 https://services.healthtech.dtu.dk/services/NetNGlyc-1.0/ Predicts N-glycosylation sites in proteins Artificial NN trained on experimentally verified glycosylation sites
NetOGlyc1.0 https://services.healthtech.dtu.dk/services/NetOGlyc-4.0/ O-glycosylation sites in mammalian proteins NN using sequence context and surface accessibility
GPS6.0 http://gps.biocuckoo.cn/ Kinase-specific phosphorylation sites Group-based scoring algorithm with kinase-specific motifs and evolutionary information
DeepLoc2.1 https://services.healthtech.dtu.dk/services/DeepLoc-2.1/ Subcellular localization of eukaryotic proteins DL using attention mechanisms on protein sequences
DeepTMHMM https://dtu.biolib.com/DeepTMHMM Transmembrane topology and signal peptides DL combining convolutional NN and hidden markov models
Open in a new tab

SNPs, Single Nucleotide Polymorphisms; NN, Neural Network; SVM, Support Vector Machine; PTM, Post Translational Modifications and DL, Deep Learning.

2.3. Analysis of missense SNPs

Missense SNPs (Table 1), which result in amino acid substitutions in the protein sequence, were evaluated using the PredictSNP 2.0 platform (Bendl et al., 2016) for DNA impacts and subjected to pathogenicity and functional impact predictive analyses using a robust set of platforms. The PredictSNP 1.0 platform (Bendl et al., 2014) was used to assess the deleterious potential of missense variants by integrating six predictive tools based on different algorithmic principles: MAPP (Multivariate Analysis of Protein Polymorphism) (Stone and Sidow, 2005); PhD-SNP (Predictor of human Deleterious Single Nucleotide Polymorphisms) (Capriotti et al., 2006); PolyPhen-1 (Ramensky et al., 2002); PolyPhen-2 (Adzhubei et al., 2010); SIFT (Sorting Intolerant From Tolerant) (Ng and Henikoff, 2003); PANTHER (Brunham et al., 2005) and SNAP (Screening for Non-Acceptable Polymorphisms) (Bromberg and Rost, 2007).

2.4. Protein stability analysis

DynaMut 2.0 (Rodrigues et al., 2021), MuPRO (Protein Stability Prediction upon Mutation) (Cheng et al., 2006) and iStable (Chen et al., 2013) platforms were used to classify mutations as stabilizing or destabilizing with associated confidence values (Table 1).

2.5. Analysis of structural and functional protein characteristics

Evolutionary conservation analysis was performed using the NetSurf (Høie et al., 2022) and ConSurf (Glaser et al., 2003) servers. In addition, phosphorylation sites on serine, threonine, and tyrosine were predicted using NetPhos 3.1 (Blom et al., 1999; Blom et al., 2004) and Musite (Wang et al., 2020; Wang et al., 2019; Wang et al., 2017). Propeptide Cleavage Sites 1.0 (Duckert et al., 2004) was used to identify the proteolytic cleavage sites. N-glycosylation sites were predicted using NetNGlyc 1.0 (Gupta and Brunak, 2002), and O-glycosylation sites on serine and threonine were predicted using NetOGlyc 4.0.0.13 (Steentoft et al., 2013). GPS 6.0 (Group-based Prediction System) (Chen et al., 2023) predicted various post-translational modifications, including acetylation, methylation, ubiquitination, and sumoylation, and identified variants that potentially alter the modification profiles. Subcellular localization was predicted using DeepLoc 2.1 (Nielsen, 2025), and membrane topology and transmembrane domains were assessed using DeepTMHMM (Hallgren et al., 2022) (Table 1).

2.6. Validation in thyroid nodules

This study was conducted in accordance with the ethical principles for human research and was approved by the Research Ethics Committee of the Faculty of Medical Sciences, University of Campinas (CEP-FCM/UNICAMP) under protocol CAAE 70994623.9.0000.5404. Biological material was obtained from the Biorepository of the Laboratory of Cancer Molecular Genetics (GEMOCA) at FCM/UNICAMP, comprising previously extracted genomic DNA samples from peripheral blood.

Eight polymorphisms were selected for validation: rs1143676 (ITGA4); rs2230433 (ITGAL), rs1143683 (ITGAM), rs1799969 (ICAM1), rs5498 (ICAM1), rs3745925 (MADCAM1), rs17576 (MMP9) and rs17577 (MMP9). The selection of these specific variants was based on three main criteria: (i) a Minor Allele Frequency (MAF) > 0.1 in the Brazilian population, verified through the Archive of Brazilian Online Mutations (ABRAOM) repository, to ensure adequate statistical power for clinical validation in our specific cohort; (ii) existing scientific literature supporting their functional roles in related pathways; and (iii) deleterious consensus scores across our bioinformatics pipeline.

2.7. Study population

The study population comprised three groups: patients with benign nodules (n = 152), patients with malignant tumors (n = 171), and healthy controls (n = 325). The distribution of histopathological diagnoses is shown in Figure 2. The benign nodule group had a mean age of 48.3 ± 14.1 years and mean nodule size of 2.13 ± 1.19 cm. This group was predominantly female (89.5%; n = 136). The diagnoses consisted of goiter (68.4%; n = 104) and follicular adenoma (FA; 31.6%; n = 48) (Figure 2A). Patients in the malignant group had a mean age of 42.0 ± 12.6 years and a mean tumor size of 1.77 ± 1.05 cm, with 81.2% (n = 138) of patients being women. The histopathological features included multifocality (31.5%; n = 52), associated chronic lymphocytic thyroiditis (40.4%; n = 46), invasion (37.1%; n = 56), and metastasis at diagnosis (25.4%; n = 30). The majority of cases were papillary thyroid carcinoma (PTC; 84.7%; n = 144), with the remainder being follicular thyroid carcinoma (FTC; 15.3%; n = 26) (Figure 2B). Among the PTC cases, the classic variant was the most frequent (CVPTC; 82.9%; n = 119), followed by follicular (FVPTC; 13.9%; n = 20), sclerosing (SVPTC; 2.1%; n = 3), and oxyphilic variants (OVPTC; 1.4%; n = 2) (Figure 2C). The control group (n = 325) consisted of age- and sex-matched individuals with no history of thyroid disease, with a mean age of 42.8 ± 11.9 years, and 84.9% (n = 276) of the participants were women.

FIGURE 2.

Panel A shows a pie chart with goiter at sixty-eight point four percent and follicular adenoma at thirty-one point six percent. Panel B displays a pie chart with papillary thyroid carcinoma at eighty-four point seven percent and follicular thyroid carcinoma at fifteen point three percent. Panel C presents a pie chart with classic variant papillary thyroid carcinoma at eighty-two point six percent, follicular variant at thirteen point nine percent, solid variant at two point one percent, and other variant at one point four percent.

Open in a new tab

Distribution of histopathological diagnoses in the benign and malignant groups. (A) Benign lesions: goiter (n = 104) and FA (n = 48). (B) Malignant tumors: PTC (n = 144) and FTC (n = 26). (C) PTC variants: CVPTC (n = 119), FVPTC (n = 20), OVPTC (n = 2), and SVPTC (n = 3).

2.8. Genotyping methodology

Genotype determination was performed using TaqMan® SNP Genotyping Assays (Applied Biosystems™) on a 7300 Real-Time PCR System (Applied Biosystems™, Foster City, United States). Commercially available pre-validated assays (TaqMan® Pre-Designed SNP Genotyping Assays, Applied Biosystems™) were used to ensure specificity and reproducibility (ITGA4, rs1143676, C___1276216_1_; ITGAL, rs2230433, C__11789692_10; ITGAM, rs1143683, C__11388882_20; ICAM1, rs1799969, C___8726331_10 and rs5498, C___8726337_40; MADCAM1, rs3745925, C__27472568_10 and MMP9, rs17576, C__11655953_10 and rs17577, C__11655948_1_).

2.9. DNA quantification and quality control

Genomic DNA was quantified by spectrophotometry using an Epoch Microplate Spectrophotometer (BioTek Instruments) and Gen5 software. Concentrations were determined by measuring absorbance at 260 nm, and purity was verified using the A260/A280 and A260/A230 ratios within the acceptable ranges (1.8–2.0 and >1.8, respectively). The samples were then diluted to a standardized concentration of 10 ng/μL with ultrapure water and quantified. Only samples with confirmed concentrations between 5 and 15 ng/μL were processed further; those outside this range were adjusted and re-evaluated until they met this criterion.

2.10. PCR amplification

Each amplification reaction was performed in a final volume of 7 μL, containing 20 ng of genomic DNA, 3.5 μL TaqMan™ Genotyping Master Mix (2X, final concentration 1X), 0.175 μL TaqMan-specific assay (40X, containing primers and fluorescent probes, final concentration 1X), and 1.325 μL ultrapure water (Milli-Q). Negative controls containing ultrapure water instead of DNA were included in all runs to monitor contamination and to validate reaction specificity. The thermal cycling conditions were as follows: initial polymerase activation at 95 °C for 10 min, followed by 40 cycles of denaturation at 92 °C for 15 s and annealing/extension at 60 °C for 90 s. Allelic discrimination and genotype determination were performed using 7500 Software version 2.3 (Applied Biosystems™, Foster City, United States), following the manufacturer’s recommendations. Individuals were classified into three genotypic categories based on their characteristic fluorescence patterns: homozygous for the reference allele (wild-type), heterozygous, and homozygous for the polymorphic allele.

2.11. Statistical analysis

Descriptive statistics (mean, standard deviation, median, and range) were used to summarize the sample profiles. Between-group comparisons for continuous and categorical variables were performed using the Mann-Whitney U test and chi-square or Fisher’s exact test, respectively. A two-sided p-value <0.05 was considered statistically significant for all tests, and all p-values were corrected using the Bonferroni method, when appropriate.

3. Results

3.1. Bioinformatics analysis

The SNPs registered for twenty-one candidate genes were systematically evaluated in this study. Initial screening identified 407,812 polymorphisms reported in the dbSNP platform, of which 133 SNPs and 207 nucleotide substitutions were selected as potential targets for future validation in TC based on predefined selection criteria. Comprehensive details of the screening process and the selected variants are presented in Table 2. The most functionally relevant variants are discussed herein, with an emphasis on those that demonstrate the highest pathogenic potential.

TABLE 2.

Summary of in silico analysis: evaluated and potential target SNPs identified across twenty-two candidate genes.

Gene Total reported in dbSNP MAF>0.1 After duplicate and merged removal SNPs (N alt) Potential targets Missense variants MAF>0.1
N SNPs (N alt) rsID N SNPs (N alt) rsID
ICAM1 8819 104 21 (41) 1 (2) rs923366 1 (2) rs5498
ICAM2 9040 61 19 (48) 3 (6) rs4968683, rs4968605, rs16947352 0 -
ICAM3 4809 62 25 (42) 1 (1) rs7250300 2 (3) rs2304237
rs7258015
VCAM1 9397 71 21 (36) 2 (4) rs3917010, rs2392221 0 -
MADCAM1 7109 59 33 (47) 1 (3) rs758503 1 (2) rs3745925
PECAM1 28,421 115 38 (80) 3 (6) rs9902260, rs1470453, rs12941698 0 -
JAM3 44,370 656 183 (310) 20 (32) rs621698, rs659196, rs610382, rs7936421, rs595102, rs11223686, rs1143809, rs470966, rs12287552, rs470846, rs470417, rs647837, rs470456, rs1143811, rs637517, rs7101690, rs11223703, rs11223709, rs2276266, rs610829 0 -
F11R 12,276 182 54 (89) 8 (14) rs28365979, rs28365980, rs28365978, rs2481085, rs790055, rs11807930, rs2988726, rs2481084 0 -
ITGA4 39,162 530 175 (308) 15 (21) rs155102, rs17290693, rs3770138, rs2305586, rs62191453, rs3770109, rs113276800, rs4667308, rs16867433, rs3770108, rs2305588, rs3770116, rs6748657, rs16867438, rs10490691 2 (5) rs1143676
rs7562325
ITGA10 10,920 36 13 (10) 0 - 0 -
ITGAD 16,534 103 30 (62) 1 (3) rs8050500 0 -
ITGAL 24,144 254 64 (127) 7 (14) rs11150589, rs12716977, rs11574938, rs12598978, rs11574941, rs1557672, rs4341759 1 (1) rs2230433
ITGAM 33,448 366 137 (228) 10 (14) rs35314490, rs9940397, rs4889647, rs9939679, rs140615691, rs9928749, rs7198598, rs11865746, rs11864503, rs9933520 2 (2) rs1143678
rs1143683
ITGB2 22,629 398 140 (243) 6 (7) rs2280965, rs2838727, rs59498405, rs2072703, rs2072702, rs7281781 0 -
MMP9 5803 97 22 (35) 6 (8) rs3918262, rs9509, rs2274755, rs20544, rs3918240, rs3918253, rs17576, rs2250889 3 (8) rs17576
rs17577
rs2250889
TIMP1 2381 20 7 (13) 1 (2) rs6609534 1 (2) rs4898
TIMP2 33,003 489 259 (338) 6 (9) rs7222198, rs4789936, rs12452379, rs8179096, rs72849591, rs9912418 0 -
THBS1 9653 79 20 (35) 3 (3) rs1478605, rs2664139, rs7170682 2 (5) rs2228261
rs2292305
CD44 39,602 515 156 (276) 23 (37) rs10836343, rs12365962, rs11033008, rs11033010, rs353624, rs353620, rs353644, rs353629, rs353628, rs3794110, rs11033022, rs10836340, rs4756199, rs12362973, rs353558, rs2553808, rs1981154, rs353612, rs353627, rs3794109, rs2295756, rs10734436, rs74938975 2 (6) rs1071695
rs9666607
LCN2 3296 8 6 (11) 0 - 0 -
RASSF5 32,526 319 101 (172) 14 (18) rs12124627, rs4845112, rs67223538, rs72755019, rs7530746, rs11119043, rs11119083, rs11589, rs71633591, rs871782, rs3813976, rs4844540, rs6660116, rs7527917 1 (2) rs4845112
NTAN1 10,470 123 37 (70) 3 (3) rs1135999, rs11075253, rs2280017 2 (5) rs1135999
rs1136001
TOTAL 407,812 4647 1561 (2621) 133 (207) 20 (43)
Open in a new tab

dbSNP, SNP databank NCBI; MAF, Minor Allele Frequency; N Alt, Nucleotide alteration.

3.2. Immunoglobulin superfamily cell adhesion molecules (ICAM1, ICAM2, ICAM3, MADCAM1, PECAM1, VCAM1, JAM3 and F11R)

Analysis of ICAM1, ICAM2, and ICAM3 revealed recurrent but gene-specific patterns of predicted functional impact. In ICAM1, rs923366 showed moderate deleterious potential (33.3%, n = 2; FunSeq and GWAVA) and predicted regulatory effects. The missense variant rs5498 (A>G at position 10285007; K469E) showed limited consensus pathogenicity (PredictSNP2.0, 16.7%, n = 1; GWAVA, 100% neutral by PredictSNP1.0, Table 3) but was localized to an exposed (NetSurf and ConSurf), proteolytic cleavage site (ProP Score = 0.098), and harbored post-translational modification sites including ubiquitination (0.294), sumoylation (0.045), N6-acetyllysine (0.205), methyllysine (0.081), and hydroxylysine (0.031) (Musite). In addition, reduced protein stability (MuPRO ΔΔG = −0.6109; iStable = 0.5103) and the addition of an interaction with GLY405, which is absent in the wild-type structure, promote charge alterations that potentially disrupt molecular contacts (HOPE).

TABLE 3.

Pathogenicity prediction and protein stability analysis of missense variants using PredictSNP1.0 and MuPRO.

Gene rsID AA change PredictSNP MAPP PhD-SNP PolyPhen-1 PolyPhen-2 SIFT SNAP nsSNPAnalyzer Panther MuPRO
ICAM1 rs5498 K469E N N N N N N N U N −0,61
ICAM3 rs7258015 R115G N N N N N N N U N −1,21
ICAM3 rs2304237 D143G N N N N N N N U N −1,17
ICAM3 rs2304237 D143A N D N N N D N U N −0,70
MADCAM1 rs3745925 P300H N N N D D N D U N −0,72
MADCAM1 rs3745925 P300R D D N D D N D U N −0,54
ITGA4 rs1143676 R878Q N N N N N N N U N −1.35
ITGA4 rs1143676 R878L N N D N N D N U N 0.01
ITGA4 rs7562325 H961Q N N N N N N N U N −0.90
ITGAL rs2230433 R791T N N N N N N N U N −0.51
ITGAM rs1143678 P1147S N N N N N N N U U −0.13
ITGAM rs1143683 A858V N D N N N N N U U −0.57
MMP9 rs17576 Q279P D U D D D N N U U −1.94
MMP9 rs17576 Q279R N U N N N N D U U −1.62
MMP9 rs17576 Q279L N N N N N N D U U −1.03
MMP9 rs17577 R668Q N N N N N N N U N −0.53
MMP9 rs17577 R668P N D D N N N D U N −0.73
MMP9 rs2250889 R574Q N N N N N D N U N −0.52
MMP9 rs2250889 R574P N N N N N N N U N −0.88
MMP9 rs2250889 R574L N N N N N N N U N 0.11
TIMP1 rs4898 F124L N D D N N N N U U −0.49
THBS1 rs2228261 N470K N N D N N D D U N −0.75
THBS1 rs2292305 T523P N N N N N N N U N −1.13
THBS1 rs2292305 T523A N N N N N N N U N −0.76
CD44 rs1071695 H85Q N D D N N N D U N −0.79
CD44 rs9666607 K417T N N N N N N N U N −0.32
CD44 rs9666607 K417R N N N N N N N U N −0.21
CD44 rs9666607 K417I D U D D N N D U N 0.13
RASSF5 rs4845112 R277W D D D D D D D U D −0.67
NTAN1 rs1135999 S287A N D N N N N N U N −0.87
NTAN1 rs1135999 S287P N N N D N N N U N −1.00
NTAN1 rs1136001 H283Y N N N N N N N U N −0.84
NTAN1 rs1136001 H283D N N N N N N N U N −1.08
NTAN1 rs1136001 H283A N N N N N N N U N −1.18
Open in a new tab

AA, amino acid; N, neutral; D, Deleterious and U, unknown.

In ICAM2, three variants (rs4968683: A>C, A>G, and A>T; rs4968605: A>C and A>T; and rs16947352: C>T, C>A, and C>G) demonstrated a higher deleterious consensus (50%, n = 3, PredictSNP2.0, CADD, and DANN) across the PredictSNP2.0 consensus, suggesting potential structural or regulatory relevance of these variants. Similarly, in ICAM3, rs7250300 (C>A) exhibited moderate deleterious potential (50%, n = 3; PredictSNP2.0, CADD and DANN). Structural evaluation of rs2304237 (T>C; D143A or T>G; D143G) revealed that the D143A substitution was classified as 22.2% deleterious (n = 2; MAPP, SIFT) by PredictSNP1.0 (Table 3). Both demonstrated reduced stability (D143A ΔΔG = −0.703 and D143G ΔΔG = −1.171, MuPRO) and loss of residue interactions (ARG115), despite partial neutrality classification (Dynamut2.0). An additional ICAM3 missense variant, rs7258015 (T>C, R115G), despite a partial neutrality classification (PredictSNP2.0, 16,7% n = 1 GWAVA and 100% neutral PredictSNP1.0, Table 3), further demonstrated destabilization (MuPRO ΔΔG = −1.21; iStable 0.5669) and disruption of multiple residue contacts in an exposed and functionally annotated region (ALA45, TYR89, and ASP143, Dynamut2.0). Position 115 was characterized as an exposed region (NetSurf, ConSurf), cleavage site (ProP 0.035), transmembrane region (DeepLoc 0.9700), and methylarginine modification site (Musite 0.053).

In MADCAM1, rs758503 showed moderate deleterious potential (33.3%, n = 2; FunSeq, GWAVA). Missense substitutions at position 300 (rs3745925; C>A,G; P300 H/R) were inconsistently classified across predictors (PredictSNP2.0 100% neutral and 37,5% deleterious n = 3; PolyPhen-1, PolyPhen-2, SNAP in PredictSNP1.0, Table 3) but were uniformly associated with reduced protein stability (MuPRO ΔΔG = −0.7155; iStable 0.5625). Position 300 is located in an exposed region (NetSurf, ConSurf), classified as transmembrane (DeepLoc 0.9670), and harbors a hydroxyproline modification site (Musite 0.153).

In PECAM1, rs9902260 (C>A, C>G, C>T) and rs1470453 (T>A) exhibited a high deleterious consensus (66.7%, n = 4; PredictSNP2.0, CADD, FATHMM and FunSeq2), whereas rs12941698 (T>A, T>C) showed a moderate predicted impact (50%, n = 3; PredictSNP2.0, CADD, DANN). VCAM1 harbored some of the most pathogenic variants identified, particularly rs3917010 (A/G), classified as 100% deleterious, along with rs2392221 (C>A), which showed a high deleterious consensus (66.7%, n = 4; PredictSNP2.0, CADD, FATHMM, and FunSeq2).

All evaluated JAM3 polymorphisms demonstrated uniformly high pathogenic potential. Variants rs621698 (G>A), rs659196 (T>A, T>C, T>G), rs610382 (T>A, T>C), and rs7936421 (T>C) were consistently classified as ≥83.3% deleterious (n = 5; PredictSNP2.0, CADD, DANN, FATHMM, and FunSeq2), indicating a substantial predicted functional or regulatory effect. Among all the analyzed genes, F11R exhibited the highest pathogenic burden. The rs28365979 (C>T) variant was classified as 100% deleterious by PredictSNP2.0, whereas multiple additional substitutions (rs28365980, rs28365978, rs2481085, rs790055, rs11807930, rs2988726, and rs2481084) consistently demonstrated ≥83.3% (n = 5) deleterious consensus, indicating widespread susceptibility to functional disruption across this locus.

3.3. Integrins (ITGA4, ITGA10, ITGAD, ITGAL, ITGAM and ITGB2)

Analysis of ITGA4 revealed 15 non-coding variants (rs155102, rs17290693, rs3770138, rs2305586, rs62191453, rs3770109, rs113276800, rs4667308, rs16867433, rs3770108, rs2305588, rs3770116, rs6748657, rs16867438 and rs10490691) demonstrating a high deleterious consensus, suggesting widespread regulatory susceptibility. Two missense substitutions (rs1143676, R878Q and R878L; and rs7562325, H961Q) showed variable pathogenesis. Structural evaluation revealed that R878Q (100% neutral by PredictSNP1 - Table 3) reduced protein stability (MuPRO ΔΔG = −0.8957) and altered critical residue interactions (Glu930; Dynamut). R878L was considered 22.2% (n = 2, PhdSNP and SIFT, Table 3) deleterious with increased stability (MuPRO ΔΔG = 0.0093). The variant rs7562325 (H961Q; 100% neutral in PredictSNP1.0, Table 3) demonstrated destabilization effects (MuPRO ΔΔG = −0.8434) and disrupted residue contacts essential for integrin-mediated cell adhesion (Val959). In contrast, ITGA10 showed no variants that met the threshold criteria (≥50% deleterious). ITGAD harbored a single non-coding variant (rs8050500, T>A,C,G) with high deleterious potential (83.3%, n = 5, PredictSNP2.0, CADD, DANN, FUNSEQ, and GWAVA), suggesting a regulatory impact on integrin αD expression.

ITGAL demonstrated seven non-coding variants (rs11150589, rs12716977, rs11574938, rs12598978, rs11574941, rs1557672, and rs4341759) that were consistently classified as deleterious by multiple prediction platforms. The missense substitution rs2230433 (R791T, 100% neutral by PredictSNP1.0, Table 3) was localized to a functionally critical domain and reduced protein stability (MuPRO ΔΔG = −0.51).

Analysis of ITGAM revealed ten non-coding variants (rs35314490, rs9940397, rs4889647, rs9939679, rs140615691, rs9928749, rs7198598, rs11865746, rs11864503 and rs9933520) demonstrating a high deleterious consensus across the PredictSNP2.0 consensus. Two missense substitutions (rs1143678 P1147S, and rs1143683 A878V) were identified with variable pathogenicity predictions (P1147S 100% neutral and A878V 16, deleterious only by MAPP; Table 3) and both were associated with reduced protein stability (MuPRO ΔΔG = −0.13 and ΔΔG = −0.57, respectively). Similarly, ITGB2 harbored six non-coding variants (rs2280965, rs2838727, rs59498405, rs2072703, rs2072702, and rs7281781) with strong predicted regulatory impacts, indicating a potential disruption of integrin β2 expression and susceptibility to leukocyte adhesion deficiency.

3.4. Extracellular matrix remodeling proteins (MMP9, TIMP1, TIMP2 and THBS1)

Analysis of MMP9 revealed eight non-coding variants (rs3918262, rs9509, rs2274755, rs20544, rs3918240, rs3918253, rs17576 and rs2250889) demonstrating a highly deleterious consensus across PredictSNP2.0. Three missense substitutions (rs17576, rs17577, and rs2250889) were identified, with variable pathogenicity predictions and distinct structural effects. rs17576 exhibited the most severe predicted functional consequences in the present study. The Q279P substitution was classified as deleterious by four PredictSNP1.0 tools (PredictSNP, PhD-SNP, PolyPhen-1, and PolyPhen-2; Table 3) with substantial destabilization (MuPRO ΔΔG = −1.94). Alternative substitutions at this position (MuPRO Q279R: ΔΔG = −1.62; Q279L: ΔΔG = −1.03) were also classified as deleterious by SNAP, indicating the critical structural sensitivity of this residue. In contrast, rs17577 (R668Q) was uniformly classified as neutral with modest destabilization (MuPRO ΔΔG = −0.53), while the alternative R668P substitution showed a deleterious classification by three tools (MAPP, PhD-SNP, SNAP - Table 3) with moderate destabilization (MuPRO ΔΔG = −0.73). The variant rs2250889 exhibited a predominantly neutral classification. Both R574P and R574L were classified as 100% neutral (Table 3) with minimal stability changes (ΔΔG = −0.88 and 0.11, respectively, according to MuPRO), whereas R574Q showed a deleterious classification solely by SIFT (MuPRO ΔΔG = −0.52).

TIMP1 harbored one non-coding variant (rs6609534) with moderate (50%) deleterious potential and one missense substitution (rs4898, F124L) demonstrating inconsistent classification (100% neutral by PredictSNP2.0 and MAPP and PhDSNP classified as deleterious; Table 3), but a potential impact on metalloproteinase inhibitory function with reduced stability (MuPRO ΔΔG = −0.49). In contrast, TIMP2 exhibited the highest regulatory burden among the analyzed genes, with six non-coding variants (rs7222198, rs4789936, rs12452379, rs8179096, rs72849591, and rs9912418) consistently classified as deleterious, indicating a widespread susceptibility to regulatory disruption.

THBS1 demonstrated three non-coding variants (rs1478605, rs2664139, and rs7170682) with a strong predicted impact and two missense variants (rs2228261 and rs2292305) with variable classifications. Structural analysis revealed that rs2228261(N470K) localizes to a calcium-binding domain, reduces protein stability (MuPRO ΔΔG = −0,75), and alters key residue interactions involved in extracellular matrix binding (PhDSNP, SIFT, and SNAP classified as deleterious; Table 3). Similarly, rs2292305 showed destabilization (MuPRO T523P ΔΔG = −1,13 and T523A ΔΔG = −0,76), despite a partially neutral classification across certain predictors (100% neutral by PredictSNP1; Table 3).

3.5. Cell surface receptors (CD44)

Analysis of CD44 revealed extensive predicted functional impact across 22 non-coding variants (rs10836343, rs12365962, rs11033008, rs11033010, rs353624, rs353620, rs353644, rs353629, rs353628, rs3794110, rs11033022, rs10836340, rs4756199, rs12362973, rs353558, rs2553808, rs1981154, rs353612, rs353627, rs3794109, rs2295756, rs10734436, rs74938975) consistently classified as deleterious across PredictSNP2.0. This widespread regulatory susceptibility suggests multiple potential disruption sites that affect CD44 expression and alternative splicing. Two missense substitutions (rs1071695 H85Q and rs9666607 K417T, K417R, and K417I) were identified with variable pathogenicity. rs1071695 (H85Q) was classified as deleterious by three PredictSNP1.0 tools (MAPP, PhD-SNP, and SIFT) with moderate destabilization (MuPRO ΔΔG = −0.79). rs9666607 demonstrated position-specific sensitivity to residue 417. Although K417T (MuPRO ΔΔG = −0.32) and K417R (MuPRO ΔΔG = −0.21) were both classified as 100% neutral with modest destabilization, the alternative substitution K417I was classified as deleterious by five tools (PredictSNP, PhD-SNP, PolyPhen-1, PolyPhen-2, and SNAP; Table 3) and paradoxically showed increased stability (MuPRO ΔΔG = 0.13).

3.6. Intercellular regulatory proteins (LCN2)

Analysis of LCN2 revealed no variants that met the established threshold. None of the evaluated SNPs demonstrated a high or moderate deleterious consensus across PredictSNP2.0, indicating an insufficient predicted functional impact to warrant further validation, according to the applied methodology. Consequently, no LCN2 polymorphisms were selected as candidates for experimental validation.

3.7. Secreted immune/inflammatory mediators (NTAN1 and RASSF5)

In NTAN1, regulatory variants (rs11075253 C>A, rs1135999 A>C, and rs2280017 C>T) showed moderate deleterious potential (50%, n = 3; CADD, DANN, FunSeq2). Among the missense variants, the rs1136001 polymorphism (G/A, G/C, or G/T at position 15038117) caused H283Y, H283D, or H283A substitutions, respectively, and was considered 100% neutral by both PredictSNP2.0 and PredictSNP1.0 (Table 3). Despite the neutral classification, all three substitutions enabled interaction with PHE286, which is absent in the wild-type structure, and promoted reduced protein stability (MuPRO ΔΔG = −0.8407, −1.0756, and-1.1843, respectively). Position 283 is located in an exposed region (NetSurf, ConSurf) but is not associated with posttranslational modification. The rs1135999 variant (A>C) causes S287A substitution, classified as 50.0% deleterious (n = 3; CADD, DANN, FunSeq2) by PredictSNP2.0, and was considered deleterious by MAPP (Table 3), with decreased stability (MuPRO ΔΔG = −0.8679). The alternative A>G variant causes S287P substitution, classified as 33.3% deleterious (n = 2; CADD, FunSeq2) by PredictSNP2.0 and was considered deleterious by PolyPhen-1 (Table 3), also with decreased stability (MuPRO ΔΔG = −0.9974). Position 287 is characterized as an exposed region (NetSurf, ConSurf), protein kinase C phosphorylation site (NetPhos 0.877), and an O-glycosylation site (NetOGlyc 0.2733).

Analysis of RASSF5 revealed 14 non-coding variants (rs12124627, rs4845112, rs67223538, rs72755019, rs7530746, rs11119043, rs11119083, rs11589, rs71633591, rs871782, rs3813976, rs4844540, rs6660116 and rs7527917), demonstrating a high deleterious consensus, suggesting widespread regulatory susceptibility across this tumor suppressor locus. Notably, rs4845112 was classified as 100% deleterious by PredictSNP1.0 (Table 3), with substantial protein destabilization (MuPRO ΔΔG = −0.67).

3.8. Validation of selected polymorphisms in thyroid nodules

The analysis identified subtype-specific genetic signals with potential utility in the molecular characterization of thyroid lesions, particularly follicular-patterned tumors. Although no polymorphism broadly distinguished malignant from benign states (Tables 4, 5), histological stratification revealed relevant discriminators (Tables 4, 5), suggesting that the evaluated variants may contribute to a more refined assessment of thyroid nodular diseases.

TABLE 4.

Allelic and genotypic distribution of validated SNPs across histological subtypes: goiter, FA, PTC, and FTC.

Gene and rsID Allele or genotype Benign (B) Malignant (M) Controls (C) p-value HWEp
G FA B PTC FTC M
ITGA4 rs1143676 A 157 70 227 197 36 233 443 B vs. C = 0.2352*
M vs. C = 1.0000*
B vs. M = 0.2907*
 G vs. C p = 0.802*; G vs. FA p = 1.0000*; G vs. PTC p = 0.882*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000* 0.06
G 51 26 77 91 16 107 199
AA 60 26 86 70 13 83 158 B vs. C = 0.6588 M vs. C = 1.0000
B vs. M = 0.7218		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000*
AG 37 18 55 57 10 67 127
GG 7 4 11 17 3 20 36
ITGAL rs2230433 G 150 66 216 192 42 234 433 B vs. C = 0.6957*
M vs. C = 1.0000*
B vs. M = 1.0000*		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 0.507*; FTC vs. C p = 0.444* 0.489
C 58 30 88 96 10 106 213
GG 55 24 79 65 17 82 145 B vs. C = 1.0000
M vs. C = 1.0000
B vs. M = 1.0000		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000*
GC 40 18 58 62 8 70 143
CC 9 6 15 17 1 18 35
ITGAM rs1143683 C 166 74 240 240 44 284 519 B vs. C = 1.0000*
M vs. C = 0.6867*
B vs. M = 0.4677*		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000* 0.106
T 42 22 64 48 8 56 127
CC 66 27 93 101 20 121 214 B vs. C = 0.5607 M vs. C = 1.0000
B vs. M = 0.3153		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 0.285*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000*
CT 34 20 54 38 4 42 91
TT 4 1 5 5 2 7 18
ICAM1 rs1799969 G 195 88 283 266 46 312 606 B vs. C = 1.0000*
M vs. C = 0.5418*
B vs. M = 1.0000*		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000* 0.969
A 13 8 21 22 6 28 38
GG 91 40 131 123 20 143 286 B vs. C = 0.8436*
M vs. C = 0.2496*
B vs. M = 1.0000*		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 0.973*
GA 13 8 21 20 6 26 34
AA 0 0 0 1 0 1 2
ICAM1 rs5498 A 123 60 183 163 24 187 352 B vs. C = 1.0000*
M vs. C = 0.2490*
B vs. M = 0.6048*		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 0.590*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 0.400* 0.814
G 85 36 121 125 28 153 226
AA 38 17 55 45 3 48 119 B vs. C = 1.0000
M vs. C = 0.3453
B vs. M = 0.9294		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 0.381*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 0.754*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 0.233*
AG 47 26 73 73 18 91 114
GG 19 5 24 26 5 31 56
MADCAM1 rs3745925 C 167 59 226 231 36 267 497 B vs. C = 0.9603*
M vs. C = 1.0000*
B vs. M = 0.6663*		 G vs. C p = 1.0000*; G vs. FA p = 0.016*; G vs. PTC p = 1.0000*; G vs. FTC p = 0.936*; FA vs. C p = 0.027*; FA vs. PTC p = 0.007*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 0.661*; FTC vs. C p = 1.0000* 0.726
A 41 35 76 55 16 71 141
CC 66 16 82 93 14 107 196 B vs. C = 0.6723 M vs. C = 1.0000 B vs. M = 0.5880 G vs. C p = 1.0000*; G vs. FA p = 0.019*; G vs. PTC p = 1.0000*; G vs. FTC p = 0.408*; FA vs. C p = 0.017*; FA vs. PTC p = 0.009*; FA vs. FTC p = 0.905*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 0.434*; FTC vs. C p = 1.0000*
CA 35 27 62 45 8 53 105
AA 3 4 7 5 4 9 18
MMP9 rs17576 A 126 59 185 182 33 215 434 B vs. C = 0.4629*
M vs. C = 1.0000*
B vs. M = 1.0000*		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000* 0.403
G 72 35 107 96 15 111 202
AA 42 19 61 61 12 73 148 B vs. C = 0.7938 M vs. C = 1.0000
B vs. M = 1.0000		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000*
AG 42 21 63 60 9 69 138
GG 15 7 22 18 3 21 32
MMP9 rs17577 G 178 80 258 248 48 296 569 B vs. C = 1.0000*M vs. C = 1.0000*B vs. M = 1.0000* G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000* 0.508
A 28 10 38 34 4 38 73
GG 78 35 113 107 22 129 252 B vs. C = 1.0000
M vs. C = 1.0000
B vs. M = 0.5376		 G vs. C p = 1.0000*; G vs. FA p = 1.0000*; G vs. PTC p = 1.0000*; G vs. FTC p = 1.0000*; FA vs. C p = 1.0000*; FA vs. PTC p = 1.0000*; FA vs. FTC p = 1.0000*; PTC vs. C p = 1.0000*; PTC vs. FTC p = 1.0000*; FTC vs. C p = 1.0000*
GA 22 10 32 34 4 38 65
AA 3 0 3 0 0 0 4
Open in a new tab

Chi-square test and *Fisher’s exact test. Abbreviations - G, goiter; FA, follicular adenoma; PTC, papillary thyroid cancer; FTC, follicular thyroid cancer; B, benign; M, malignant; C, control; vs, versus; HWEp, Hardy Weinberg equilibrium p-value. All p-values were corrected using the Bonferroni correction.

TABLE 5.

Clinical and pathological characteristics of thyroid nodule (benign and malignant).

ITGA4 - rs1143676 ITGAL - rs2230433 ITGAM - 1143683 ICAM1 - rs1799969
Group Characteristics AA AG GG p-value GG GC CC p-value CC CT TT p-value GG GA AA p-value
Control Gender Male 24 21 3 0,473 26 21 2 0,1904 38 10 1 0,1619 42 6 1 0,6294
Female 134 106 33 119 122 33 176 81 17 224 28 1
Benign Gender Male 7 6 3 0,1493 11 5 0 0,228 9 6 1 0,7875* 13 3 0 0,4658
Female 79 49 8 68 53 15 84 48 4 118 18 0
Malignant Gender Male 15 13 4 0,9688 14 15 4 0,7742 25 6 1 0,6288 30 2 0 0,1139*
Female 68 54 16 68 56 14 96 36 6 113 24 1
Multifocality Yes 30 18 4 0,3435 32 14 12 0,0008 33 14 5 0,0428 45 7 0 >0,9999*
No 52 47 14 49 52 6 86 25 2 97 15 1
Capsule Yes 26 14 2 0,0994 20 19 12 0,0014 32 7 3 0,2688 41 1 0 0,0122*
No 40 40 12 47 33 3 68 22 2 75 16 1
CLT Yes 23 17 6 0,8889 20 21 10 0,1851 28 17 1 0,2015 39 7 0 0,5873*
No 34 27 7 29 29 5 50 15 3 60 7 1
Vascular and lymphatic invasion Yes 29 23 4 0,3626 26 21 9 0,2443 41 14 1 0,4408 48 8 0 0,8066*
No 43 38 14 49 39 7 66 23 6 83 11 1
Metastasis at diagnosis Yes 17 9 4 0,4417 15 11 4 0,5904 23 7 0 0,3175 25 5 0 0,5490*
No 41 38 9 35 41 12 58 25 5 77 10 1
Control Gender Male 22 20 7 0,4754 29 16 3 0,9755 22 18 6 0,6979 39 10 0 >0,9999*
Female 97 124 49 167 89 15 126 120 26 213 55 4
Benign Gender Male 7 8 1 0,5148 3 12 0 0,0060* 7 6 3 0,8559 14 1 1 0,3605*
Female 48 65 23 79 50 7 54 57 19 99 31 2
Malignant Gender Male 10 17 5 0,8714 20 10 2 0,9667 20 8 4 0,0602 27 5 0 0,3535*
Female 38 74 26 87 43 7 53 61 17 102 33 0
Multifocality Yes 13 58 10 <0.0001 32 16 4 0,5189 23 20 3 0,2721 42 9 0 0,3184*
No 35 29 20 71 37 4 49 45 18 82 29 0
Capsule Yes 14 24 4 0,5658 27 12 3 0,7407 19 18 4 0,7604 33 8 0 >0,9999*
No 30 47 15 57 30 4 44 33 11 72 19 0
CLT Yes 13 21 12 0,4062 35 8 2 0,1271 15 23 5 0,0663 33 12 0 0,4884*
No 24 33 11 42 24 2 36 21 10 53 13 0
Vascular and lymphatic invasion Yes 21 27 8 0,2730 39 13 3 0,1533 27 22 5 0,4677 47 9 0 0,1586*
No 24 53 18 53 37 5 38 39 14 67 25 0
Metastasis at diagnosis Yes 9 16 5 0,8671 23 5 2 0,0841 16 10 3 0,5548 23 7 0 0,8010
No 30 42 16 53 32 2 37 37 11 67 18 0
Open in a new tab

CLT, chronic lymphocyte thyroiditis.

Bold values indicates p < 0.05.

*Fisher's exact test.

The MADCAM1 rs3745925 variant emerged as the most informative marker, significantly distinguishing follicular adenoma from controls (p = 0.017, OR: 3.15; 95% CI: 1.631–6.053), goiter (p = 0.019, OR: 3.18; 95% CI: 1.49–6.85), and PTC (p = 0.009, OR: 3.49, 95% CI: 1.728–7.092), underscoring its potential value as an auxiliary biomarker in the differential characterization of thyroid nodular lesions. Additionally, ITGAM rs1143683, ITGAL rs2230433, and ICAM1 rs5498 were associated with tumor multifocality (p = 0.0428, p = 0.0008, and p < 0.0001, respectively).

4. Discussion

This integrated in silico and clinical study provides insights into the potential contribution of variations in adhesion- and immune-related genes to the molecular characterization of thyroid nodular diseases. Across the immune adhesion and extracellular matrix remodeling networks analyzed, bioinformatic prioritization revealed a clear gradient of predicted functional disruption, with many variants concentrated in regulatory regions rather than being limited to protein-coding changes. Several loci showed recurrent deleterious predictions across independent algorithms, suggesting that disease susceptibility in this dataset is likely shaped by both transcriptional/splicing regulation and structural protein effects.

This was particularly evident for ICAM2, JAM3, F11R, ITGA4, ITGAM, ITGB2, TIMP2, RASSF5, and CD44, which carried multiple non-coding variants with a consistent deleterious consensus, supporting the idea that altered gene regulation may be a major driver of biological impact. Simultaneously, selected missense substitutions displayed functionally meaningful consequences, including reduced stability, altered residue interactions, exposure at functionally relevant sites, and disruption of cleavage or modification motifs. In this context, the integrative use of consensus predictors, stability estimators, conservation analysis, and structure-based annotations is essential to distinguish variants with likely biological relevance from those appearing neutral in isolated prediction tools.

Among the coding changes, the most compelling signals were observed in ICAM1, ICAM3, MADCAM1, PECAM1, VCAM1, MMP9, THBS1, NTAN1, and F11R, where the cumulative burden of deleterious substitutions was the highest. Variants such as MMP9 rs17576, THBS1 rs2228261 and rs2292305, ICAM3 rs2304237 and rs7258015, and CD44 rs1071695 and rs9666607 illustrate how bioinformatics can reveal functionally plausible candidates, even when prediction outputs are not uniformly concordant. Importantly, several substitutions classified as neutral by some algorithms still showed marked destabilization or loss of residue contacts, underscoring the value of integrating multiple complementary approaches rather than relying on any single method alone.

By systematically screening 407,812 polymorphisms across 22 candidate genes, we prioritized 133 variants with high predicted functional impact, eight of which were selected for clinical validation in a cohort of 648 individuals. Among these, MADCAM1 rs3745925 emerged as the only polymorphism that showed a consistent and statistically significant clinical signal. This variant distinguished follicular adenoma from controls (p = 0.017, OR: 3.15; 95% CI: 1.631–6.053), goiter (p = 0.019, OR: 3.18; 95% CI: 1.49–6.85), and papillary thyroid carcinoma (p = 0.009, OR: 3.49; 95% CI: 1.728–7.092). Notably, this association remained robust after Bonferroni correction for multiple comparisons, supporting its potential value as an auxiliary marker for the differential characterization of follicular-patterned thyroid lesions. MAdCAM-1 regulates lymphocyte homing to mucosal tissues (Keir et al., 2021; Iizuka et al., 2000; Kuhbandner et al., 2019), and although its role in thyroid biology is poorly defined, our results suggest that genetic variation at this locus may influence follicular cell proliferation and early tumorigenesis in thyroid tissues.

Additionally, ITGAM rs1143683, ITGAL rs2230433, and ICAM1 rs5498 were associated with tumor multifocality (p = 0.0428, p = 0.0008, and p < 0.0001, respectively); however, these associations did not survive Bonferroni correction and should therefore be interpreted as exploratory. In contrast, variants in ITGA4, ITGAL, and MMP9 were not significantly associated with this cohort. This does not discount their biological relevance in thyroid cancer progression but suggests that the effects of these common polymorphisms may be modest or context-dependent, requiring larger cohorts or integrative multi-omics approaches for detection.

Several limitations should be considered when interpreting the findings. First, the sample size of certain histological subgroups, particularly FTC, was modest and may have limited statistical power. Second, as a single-center study, external validation in independent populations is required to strengthen generalizability. Third, functional assays are required to elucidate the mechanistic consequences of the identified variants and explore their predicted impacts on protein structure, stability, and post-translational modifications. Fourth, given the exploratory nature of this study, the observed associations should be viewed as hypothesis-generating only. Finally, while we selected eight SNPs for clinical validation based on a multi-step prioritization strategy, other potentially relevant polymorphisms from the 133 prioritized variants may also merit further investigation.

Despite these limitations, this study had several strengths. The integrated approach combining systematic bioinformatics prioritization with clinical validation in a well-characterized cohort of 648 individuals provides a model for discovering candidate genetic markers of thyroid nodular disease. The identification of MADCAM1 rs3745925 as a variant that distinguishes follicular adenoma from goiter, PTC, and controls represents a specific, testable finding that can be directly examined in future studies. Moreover, the associations observed for ITGAM rs1143683, ITGAL rs2230433, and ICAM1 rs5498 with tumor multifocality generate new hypotheses regarding the role of adhesion and immune pathways in thyroid tumor pathobiology. Collectively, this study establishes a foundation for larger association studies and functional experiments aimed at clarifying the biological and clinical relevance of these genetic variants.

5. Conclusion

This exploratory study identified specific germline polymorphisms in cell adhesion- and immune-related genes associated with distinct clinicopathological features of thyroid nodular diseases. Notably, the association of MADCAM1 rs3745925 with follicular adenoma and follicular-patterned lesions remained robust after stringent correction for multiple comparisons, positioning this variant as a priority candidate for future studies. The additional associations of ITGAM rs1143683, ITGAL rs2230433, and ICAM1 rs5498 with tumor multifocality, although not surviving Bonferroni correction, offer preliminary hypotheses that warrant further investigation. Collectively, these findings support the involvement of adhesion and immune pathways in thyroid tumor biology and demonstrate the feasibility of integrating computational variant prioritization with targeted clinical validation of the results. Independent replication in larger, multicenter cohorts and functional studies is needed to clarify the biological and clinical relevance of these variants.

Acknowledgements

The authors thank the statisticians of the School of Medical Sciences of University of Campinas (UNICAMP). We would also like to thank Dr. José Barreto Carvalheira and Sandra Brambilla from the Molecular Oncology Laboratory (UNICAMP) for allowing us to use their laboratory facilities.

Funding Statement

The author(s) declared that financial support was not received for this work and/or its publication.

Footnotes

Edited by: Matteo Becatti, University of Firenze, Italy

Reviewed by: Carina Mucciolo Melo, Faculdade de Medicina do ABC, Brazil

Weihong Niu, Henan Provincial People’s Hospital, China

Data availability statement

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.

Ethics statement

The studies involving humans were approved by Research Ethics Committee of the Faculty of Medical Sciences, University of Campinas (CEP-FCM/UNICAMP), under protocol CAAE 70994623.9.0000.5404. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

Author contributions

LR: Writing – original draft, Investigation, Conceptualization, Formal Analysis, Visualization, Resources, Software, Validation, Writing – review and editing, Data curation, Methodology. KP: Data curation, Software, Methodology, Writing – review and editing, Investigation. ET: Validation, Methodology, Writing – review and editing, Investigation. AT: Investigation, Validation, Writing – review and editing. NB: Conceptualization, Writing – review and editing, Supervision, Validation, Formal Analysis. LW: Supervision, Writing – review and editing, Data curation, Conceptualization, Writing – original draft, Formal Analysis.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmolb.2026.1814666/full#supplementary-material

Table1.doc (490.5KB, doc)

References

  1. Adzhubei I. A., Schmidt S., Peshkin L., Ramensky V. E., Gerasimova A., Bork P., et al. (2010). A method and server for predicting damaging missense mutations. Nat. Methods 7 (4), 248–249. 10.1038/nmeth0410-248 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Andriescu E. C., Căruntu I. D., Giuşcă S. E., Lozneanu L., Ciobanu Apostol D. G. (2018). Prognostic significance of cell-adhesion molecules in histological variants of papillary thyroid carcinoma. Rom. J. Morphol. Embryol. 59 (3), 721–727. [PubMed] [Google Scholar]
  3. Bendl J., Stourac J., Salanda O., Pavelka A., Wieben E. D., Zendulka J., et al. (2014). PredictSNP: robust and accurate consensus classifier for prediction of disease-related mutations. PLoS Comput. Biol. 10 (1), e1003440. 10.1371/journal.pcbi.1003440 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bendl J., Musil M., Štourač J., Zendulka J., Damborský J., Brezovský J. (2016). PredictSNP2: a unified platform for accurately evaluating SNP effects by exploiting the different characteristics of variants in distinct genomic regions. PLoS Comput. Biol. 12 (5), e1004962. 10.1371/journal.pcbi.1004962 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Bergonzini C., Kroese K., Zweemer A. J. M., Danen E. H. J. (2022). Targeting integrins for cancer therapy - disappointments and opportunities. Front. Cell Dev. Biol. 10, 863850. 10.3389/fcell.2022.863850 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Blom N., Gammeltoft S., Brunak S. (1999). Sequence and structure-based prediction of eukaryotic protein phosphorylation sites. J. Mol. Biol. 294 (5), 1351–1362. 10.1006/jmbi.1999.3310 [DOI] [PubMed] [Google Scholar]
  7. Blom N., Sicheritz-Pontén T., Gupta R., Gammeltoft S., Brunak S. (2004). Prediction of post-translational glycosylation and phosphorylation of proteins from the amino acid sequence. Proteomics 4 (6), 1633–1649. 10.1002/pmic.200300771 [DOI] [PubMed] [Google Scholar]
  8. Brassart-Pasco S., Brézillon S., Brassart B., Ramont L., Oudart J. B., Monboisse J. C. (2020). Tumor microenvironment: extracellular matrix alterations influence tumor progression. Front. Oncol. 10, 397. 10.3389/fonc.2020.00397 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Bromberg Y., Rost B. (2007). SNAP: predict effect of non-synonymous polymorphisms on function. Nucleic Acids Res. 35 (11), 3823–3835. 10.1093/nar/gkm238 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Brunham L. R., Singaraja R. R., Pape T. D., Kejariwal A., Thomas P. D., Hayden M. R. (2005). Accurate prediction of the functional significance of single nucleotide polymorphisms and mutations in the ABCA1 gene. PLoS Genet. 1 (6), e83. 10.1371/journal.pgen.0010083 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Bufalo N. E., Dos Santos R. B., Rocha A. G., Teodoro L., Romaldini J. H., Ward L. S. (2021). Polymorphisms of the genes CTLA4, PTPN22, CD40, and PPARG and their roles in graves' disease: susceptibility and clinical features. Endocrine 71 (1), 104–112. 10.1007/s12020-020-02337-x [DOI] [PubMed] [Google Scholar]
  12. Burčík D., Macko J., Podrojková N., Demeterová J., Stano M., Oriňak A. (2025). Role of cell adhesion in cancer metastasis formation: a review. ACS Omega 10 (6), 5193–5213. 10.1021/acsomega.4c08140 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Capriotti E., Calabrese R., Casadio R. (2006). Predicting the insurgence of human genetic diseases associated to single point protein mutations with support vector machines and evolutionary information. Bioinformatics 22 (22), 2729–2734. 10.1093/bioinformatics/btl423 [DOI] [PubMed] [Google Scholar]
  14. Cargill M., Altshuler D., Ireland J., Sklar P., Ardlie K., Patil N., et al. (1999). Characterization of single-nucleotide polymorphisms in coding regions of human genes. Nat. Genet. 22 (3), 231–238. 10.1038/10290 [DOI] [PubMed] [Google Scholar]
  15. Chen C. W., Lin J., Chu Y. W. (2013). iStable: Off-The-Shelf predictor integration for predicting protein stability changes. BMC Bioinforma. 14 (Suppl. 2), S5. 10.1186/1471-2105-14-S2-S5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Chen M., Zhang W., Gou Y., Xu D., Wei Y., Liu D., et al. (2023). GPS 6.0: an updated server for prediction of kinase-specific phosphorylation sites in proteins. Nucleic Acids Res. 51 (W1), W243. 10.1093/nar/gkad383 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Cheng J., Randall A., Baldi P. (2006). Prediction of protein stability changes for single-site mutations using support vector machines. Proteins 62 (4), 1125–1132. 10.1002/prot.20810 [DOI] [PubMed] [Google Scholar]
  18. Chriqui L. E., Cavin S., Perentes J. Y. (2025). Dual implication of endothelial adhesion molecules in tumor progression and cancer immunity. Cell Adh Migr. 19 (1), 2472308. 10.1080/19336918.2025.2472308 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Duckert P., Brunak S., Blom N. (2004). Prediction of proprotein convertase cleavage sites. Protein Eng. Des. Sel. 17 (1), 107–112. 10.1093/protein/gzh013 [DOI] [PubMed] [Google Scholar]
  20. Fleming J., Magana P., Nair S., Tsenkov M., Bertoni D., Pidruchna I., et al. (2025). AlphaFold protein structure database and 3D-Beacons: new data and capabilities. J. Mol. Biol. 437 (15), 168967. 10.1016/j.jmb.2025.168967 [DOI] [PubMed] [Google Scholar]
  21. Forma A., Kłodnicka K., Pająk W., Flieger J., Teresińska B., Januszewski J., et al. (2025). Thyroid cancer: epidemiology, classification, risk factors, diagnostic and prognostic markers, and current treatment strategies. Int. J. Mol. Sci. 26 (11), 5173. 10.3390/ijms26115173 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Fu Y., Liu Z., Lou S., Bedford J., Mu X. J., Yip K. Y., et al. (2014). FunSeq2: a framework for prioritizing noncoding regulatory variants in cancer. Genome Biol. 15 (10), 480. 10.1186/s13059-014-0480-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Glaser F., Pupko T., Paz I., Bell R. E., Bechor-Shental D., Martz E., et al. (2003). ConSurf: identification of functional regions in proteins by surface-mapping of phylogenetic information. Bioinformatics 19 (1), 163–164. 10.1093/bioinformatics/19.1.163 [DOI] [PubMed] [Google Scholar]
  24. Guerra-Espinosa C., Jiménez-Fernández M., Sánchez-Madrid F., Serrador J. M. (2024). ICAMs in immunity, intercellular adhesion and communication. Cells 13 (4), 339. 10.3390/cells13040339 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Gupta R., Brunak S. (2002). Prediction of glycosylation across the human proteome and the correlation to protein function. Pac Symp. Biocomput, 310–322. [PubMed] [Google Scholar]
  26. Hallgren J., Tsirigos K. D., Pedersen M. D., Almagro Armenteros J. J., Marcatili P., Nielsen H., et al. (2022). DeepTMHMM predicts alpha and beta transmembrane proteins using deep neural networks. bioRxiv. 10.1101/2022.04.08.487609 [DOI] [Google Scholar]
  27. Hamidi H., Ivaska J. (2018). Every step of the way: integrins in cancer progression and metastasis. Nat. Rev. Cancer 18 (9), 533–548. 10.1038/s41568-018-0038-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Høie M. H., Kiehl E. N., Petersen B., Nielsen M., Winther O., Nielsen H., et al. (2022). NetSurfP-3.0: accurate and fast prediction of protein structural features by protein language models and deep learning. Nucleic Acids Res. 50 (W1), W510. 10.1093/nar/gkac439 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Houten P. V., Netea-Maier R. T., Smit J. W. (2023). Differentiated thyroid carcinoma: an update. Best. Pract. Res. Clin. Endocrinol. Metab. 37 (1), 101687. 10.1016/j.beem.2022.101687 [DOI] [PubMed] [Google Scholar]
  30. Iizuka T., Tanaka T., Suematsu M., Miura S., Watanabe T., Koike R., et al. (2000). Stage-specific expression of mucosal addressin cell adhesion Molecule-1 during embryogenesis in Rats1. J. Immunol. 164 (5), 2463–2471. 10.4049/jimmunol.164.5.2463 [DOI] [PubMed] [Google Scholar]
  31. Keir M. E., Fuh F., Ichikawa R., Acres M., Hackney J. A., Hulme G., et al. (2021). Regulation and role of αE integrin and gut homing integrins in migration and retention of intestinal lymphocytes during inflammatory bowel disease. J. Immunol. 207 (9), 2245–2254. 10.4049/jimmunol.2100220 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Kim S. J., Lee S. E., Kim Y. I., Nam-Goong I. S., Jung H. W., Kim E. S. (2023). Papillary thyroid cancer with hashimoto's thyroiditis attenuates the tumour aggressiveness through the up-regulation of E-cadherin and TGF-β expression. Clin. Exp. Med. 23 (3), 833–840. 10.1007/s10238-022-00857-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Kircher M., Witten D. M., Jain P., O'Roak B. J., Cooper G. M., Shendure J. (2014). A general framework for estimating the relative pathogenicity of human genetic variants. Nat. Genet. 46 (3), 310–315. 10.1038/ng.2892 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Kuhbandner K., Hammer A., Haase S., Terbrack E., Hoffmann A., Schippers A., et al. (2019). MAdCAM-1-Mediated intestinal lymphocyte homing is critical for the development of active experimental autoimmune encephalomyelitis. Front. Immunol. 10–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Ley K., Laudanna C., Cybulsky M. I., Nourshargh S. (2007). Getting to the site of inflammation: the leukocyte adhesion Cascade updated. Nat. Rev. Immunol. 7 (9), 678–689. 10.1038/nri2156 [DOI] [PubMed] [Google Scholar]
  36. Li Y., Peng S., Xu J., Liu W., Luo Q. (2025). Integrin signaling in tumor biology: mechanisms of intercellular crosstalk and emerging targeted therapies. PeerJ 13, e19328. 10.7717/peerj.19328 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Ma J. J., Xiang C., Zhu H. Q., Bai B. L., Wang P., Zhao G. A. (2024). Expression and prognosis analysis of integrin subunit α3 (ITGA3) in papillary thyroid cancer. Heliyon 10 (1), e23163. 10.1016/j.heliyon.2023.e23163 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Man J., Zhang Z., Cui T., Shi T. (2025). METTL3-mediated m6A modification of TACSTD2 mRNA inhibits papillary thyroid cancer progression by activating the TNF signaling and blocking epithelial-mesenchymal transition. Sci. Rep. 16 (1), 857. 10.1038/s41598-025-30448-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Martins M. B., de Assis Batista F., Bufalo N. E., Peres K. C., Meneghetti M., da Assumpção L. V. M., et al. (2021). Polymorphisms of IL-4 and IL-4R are associated to some demographic characteristics of differentiated thyroid cancer patients but are not determinants of risk in the Brazilian population. Endocrine 72 (2), 470–478. 10.1007/s12020-020-02486-z [DOI] [PubMed] [Google Scholar]
  40. Mezu-Ndubuisi O. J., Maheshwari A. (2021). The role of integrins in inflammation and angiogenesis. Pediatr. Res. 89 (7), 1619–1626. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Nascimento M., Teixeira E. S., Dal' Bó I. F., Peres K. C., Rabi L. T., Cury A. N., et al. (2023). NR3C1 rs6198 variant may be involved in the relationship of graves' disease with stressful events. Biomedicines 11 (4). 10.3390/biomedicines11041155 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Ng P. C., Henikoff S. (2003). SIFT: predicting amino acid changes that affect protein function. Nucleic Acids Res. 31 (13), 3812–3814. 10.1093/nar/gkg509 [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Nielsen H. (2025). Practical applications of language models in protein sorting prediction: signalp 6.0, DeepLoc 2.1, and DeepLocPro 1.0. Methods Mol. Biol. 2941, 153–175. [DOI] [PubMed] [Google Scholar]
  44. Niland S., Riscanevo A. X., Eble J. A. (2022). Matrix metalloproteinases shape the tumor microenvironment in cancer progression. Int. J. Mol. Sci. 23 (1), 146. 10.3390/ijms23010146 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Paz-Ibarra J., Concepción-Zavaleta M. J., Quiroz-Aldave J. E. (2024). Environmental factors related to the origin and evolution of differentiated thyroid cancer: a narrative review. Expert Rev. Endocrinol. Metab. 19 (6), 469–477. 10.1080/17446651.2024.2377687 [DOI] [PubMed] [Google Scholar]
  46. Peres K. C., Teodoro L., Amaral L. H. P., Teixeira E. S., Barreto I. S., de Freitas L. L. L., et al. (2021). Clinical utility of TGFB1 and its receptors (TGFBR1 and TGFBR2) in thyroid nodules: evaluation based on single nucleotide polymorphisms and mRNA analysis. Arch. Endocrinol. Metab. 65 (2), 172–184. 10.20945/2359-3997000000330 [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Phan L., Zhang H., Wang Q., Villamarin R., Hefferon T., Ramanathan A., et al. (2025). The evolution of dbSNP: 25 years of impact in genomic research. Nucleic Acids Res. 53 (D1), D925. 10.1093/nar/gkae977 [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Quang D., Chen Y., Xie X. (2015). DANN: a deep learning approach for annotating the pathogenicity of genetic variants. Bioinformatics 31 (5), 761–763. 10.1093/bioinformatics/btu703 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Rabi L. T., Peres K. C., Nascimento M., de Souza Teixeira E., de Freitas L. L. L., Barreto I. S., et al. (2023). Investigation of the clinical utility of adhesion molecules in the management of thyroid nodules. Sci. Rep. 13 (1), 4069. 10.1038/s41598-023-31302-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Rabi L. T., Valente D. Z., de Souza Teixeira E., Peres K. C., de Oliveira Almeida M., Bufalo N. E., et al. (2024). Potential new cancer biomarkers revealed by quantum chemistry associated with bioinformatics in the study of selectin polymorphisms. Heliyon 10 (7), e28830. 10.1016/j.heliyon.2024.e28830 [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Ramensky V., Bork P., Sunyaev S. (2002). Human non-synonymous SNPs: server and survey. Nucleic Acids Res. 30 (17), 3894–3900. 10.1093/nar/gkf493 [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Ritchie G. R., Dunham I., Zeggini E., Flicek P. (2014). Functional annotation of noncoding sequence variants. Nat. Methods 11 (3), 294–296. 10.1038/nmeth.2832 [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Rodrigues C. H. M., Pires D. E. V., Ascher D. B. (2021). DynaMut2: assessing changes in stability and flexibility upon single and multiple point missense mutations. Protein Sci. 30 (1), 60–69. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Saenko V. A., Rogounovitch T. I. (2018). Genetic polymorphism predisposing to differentiated thyroid cancer: a review of major findings of the Genome-Wide association studies. Endocrinol. Metab. Seoul. 33 (2), 164–174. 10.3803/EnM.2018.33.2.164 [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Shihab H. A., Rogers M. F., Gough J., Mort M., Cooper D. N., Day I. N., et al. (2015). An integrative approach to predicting the functional effects of non-coding and coding sequence variation. Bioinformatics 31 (10), 1536–1543. 10.1093/bioinformatics/btv009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Singh V., Kaur R., Kumari P., Pasricha C., Singh R. (2023). ICAM-1 and VCAM-1: gatekeepers in various inflammatory and cardiovascular disorders. Clin. Chim. Acta 548, 117487. 10.1016/j.cca.2023.117487 [DOI] [PubMed] [Google Scholar]
  57. Smart J. A., Oleksak J. E., Hartsough E. J. (2021). Cell adhesion molecules in plasticity and metastasis. Mol. Cancer Res. 19 (1), 25–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Steentoft C., Vakhrushev S. Y., Joshi H. J., Kong Y., Vester-Christensen M. B., Schjoldager K. T., et al. (2013). Precision mapping of the human O-GalNAc glycoproteome through SimpleCell technology. Embo J. 32 (10), 1478–1488. 10.1038/emboj.2013.79 [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Stone E. A., Sidow A. (2005). Physicochemical constraint violation by missense substitutions mediates impairment of protein function and disease severity. Genome Res. 15 (7), 978–986. 10.1101/gr.3804205 [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Tiucă R. A., Tiucă O. M., Paşcanu I. M. (2023). The role of genetic polymorphisms in differentiated thyroid cancer: a 2023 update. Biomedicines 11 (4), 1075. 10.3390/biomedicines11041075 [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. UniProt Consortium (2025). UniProt: the universal protein knowledgebase in 2025. Nucleic Acids Res. 53 (D1), D609-D617. 10.1093/nar/gkae1010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Viana T. R. X., Ratti R. P., Rabi L. T. (2025). Genetic variations in BDNF: therapeutic potential and influence on neuroregeneration in neurodegenerative diseases. Silico Pharmacol. 13 (3), 206. 10.1007/s40203-025-00497-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. von Lersner A., Droesen L., Zijlstra A. (2019). Modulation of cell adhesion and migration through regulation of the immunoglobulin superfamily member ALCAM/CD166. Clin. and Exp. Metastasis 36 (2), 87–95. 10.1007/s10585-019-09957-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Wai Wong C., Dye D. E., Coombe D. R. (2012). The role of immunoglobulin superfamily cell adhesion molecules in cancer metastasis. Int. J. Cell Biol. 2012 (1), 340296. 10.1155/2012/340296 [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Wan J., Lv J., Wang C., Zhang L. (2022). RPS27 selectively regulates the expression and alternative splicing of inflammatory and immune response genes in thyroid cancer cells. Adv. Clin. Exp. Med. 31 (8), 889–901. 10.17219/acem/147271 [DOI] [PubMed] [Google Scholar]
  66. Wang D., Zeng S., Xu C., Qiu W., Liang Y., Joshi T., et al. (2017). MusiteDeep: a deep-learning framework for general and kinase-specific phosphorylation site prediction. Bioinformatics 33 (24), 3909–3916. 10.1093/bioinformatics/btx496 [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Wang D., Liang Y., Xu D. (2019). Capsule network for protein post-translational modification site prediction. Bioinformatics 35 (14), 2386–2394. 10.1093/bioinformatics/bty977 [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Wang D., Liu D., Yuchi J., He F., Jiang Y., Cai S., et al. (2020). MusiteDeep: a deep-learning based webserver for protein post-translational modification site prediction and visualization. Nucleic Acids Res. 48 (W1), W140–W146. 10.1093/nar/gkaa275 [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Winkler J., Abisoye-Ogunniyan A., Metcalf K. J., Werb Z. (2020). Concepts of extracellular matrix remodelling in tumour progression and metastasis. Nat. Commun. 11 (1), 5120. 10.1038/s41467-020-18794-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Yu Y., Guo X., Chai J., Han Z., Ji Y., Sun J., et al. (2023). Identification of key immune genes related to lymphatic metastasis of papillary thyroid cancer via bioinformatics analysis and experimental validation. Front. Oncol. 13, 1181325. 10.3389/fonc.2023.1181325 [DOI] [PMC free article] [PubMed] [Google Scholar]
  71. Yuan Z., Li Y., Zhang S., Wang X., Dou H., Yu X., et al. (2023). Extracellular matrix remodeling in tumor progression and immune escape: from mechanisms to treatments. Mol. Cancer 22 (1), 48. 10.1186/s12943-023-01744-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. Zhao K., Yang H., Kang H., Wu A. (2019). Identification of key genes in thyroid cancer microenvironment. Med. Sci. Monit. 25, 9602–9608. 10.12659/MSM.918519 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Table1.doc (490.5KB, doc)

Data Availability Statement

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.

Articles from Frontiers in Molecular Biosciences are provided here courtesy of Frontiers Media SA

RESOURCES