Tumor cell program tied to lymphatic responses in aggressive thyroid cancer

An epithelial cell state defined by an extracellular matrix program, together with a molecular axis involving SERPINA1, SOX4 and FN1, has been linked to lymphatic endothelial responses in papillary thyroid carcinoma, the most common endocrine malignancy. The study, published in Frontiers in Cell and Developmental Biology, combined single-cell RNA sequencing, spatial transcriptomics, bulk tumor cohorts and laboratory experiments to trace how a specific malignant epithelial state relates to the lymphatic microenvironment that shapes lymph node metastasis.

Lymph node metastasis complicates risk assessment and treatment planning in papillary thyroid carcinoma. Although most patients have a favorable prognosis, some tumors invade locally, spread to regional lymph nodes or recur, and cervical nodal involvement is common at diagnosis. Routine assessment can miss occult nodal disease and micrometastases, and ultrasound findings can be inconclusive in small or anatomically constrained nodes, motivating research into molecular features that could complement imaging.

The research team integrated four public single-cell datasets into an atlas of 221,187 cells from 46 thyroid specimens. Clustering resolved seven major cell populations and, within the epithelial compartment, eleven transcriptionally distinct subpopulations designated THY1 through THY11. One of these, THY5, was characterized by increased expression of SERPINA1, HLA-DRA and CD74 alongside SFTPB and CXCL14, and its associated genes were enriched in collagen-containing extracellular matrix, cell adhesion, antigen processing and presentation, and T-cell activation pathways. Hallmark analysis identified significant enrichment of TNFA signaling via NF-κB among THY5 upregulated genes, along with inflammatory response, interferon responses, epithelial–mesenchymal transition and angiogenesis gene sets.

THY5 showed features of an aggressive, partially dedifferentiated state. It carried a relatively high BRAF-related transcriptional score and an intermediate thyroid differentiation score, below those of the normal-associated THY1 and THY10 populations, and pseudotime ordering placed it in later regions of the reconstructed trajectory. In pooled summaries across tissue groups, THY5 was scarce in normal thyroid and more abundant in aggressive tumors, which the authors defined as pathological T stage beyond T3b, locally invasive disease or the diffuse sclerosing variant. Enrichment analysis using observed-to-expected cell ratios gave THY5 values of 0.021, 0.961 and 1.819 in normal, non-aggressive and aggressive tissues respectively. However, the authors were careful to note a key limitation: at the individual sample level, THY5 abundance did not differ significantly between tumor groups, with a two-sided Wilcoxon rank-sum test yielding P = 0.396, and the three largest contributing samples accounted for roughly 85 percent of all THY5 cells in each group.

To connect the single-cell findings to clinical data, the team used Bayesian deconvolution with BayesPrism to estimate THY5 abundance in bulk transcriptomic cohorts from The Cancer Genome Atlas and twelve Gene Expression Omnibus cohorts comprising 859 samples. Estimated THY5 abundance was higher in tumors than in normal tissues, higher in lymph-node-positive than node-negative tumors, and generally greater at later overall and T stages, although not all pairwise stage comparisons were significant. In TCGA, BRAF was the most frequently mutated gene, and its mutation frequency was 85 percent in the THY5-high group versus 34 percent in the THY5-low group, while NRAS and HRAS frequencies fell in the THY5-high group. The THY5-high group also showed higher cancer-associated fibroblast, myeloid-derived suppressor cell and immune exclusion scores, and drug-response predictions differed, with lower predicted values for trametinib but higher values for dabrafenib and sorafenib.

From these bulk analyses the researchers built an exploratory 11-gene classifier for lymph node status using weighted gene co-expression network analysis, differential expression and machine learning, with backward stepwise regression followed by gradient boosting machine selected from 121 algorithm combinations. The genes were SCEL, B3GNT3, FN1, CHI3L1, PLCD3, CST6, TIMP1, ADORA1, PRSS23, SERPINA1 and CDH6. Areas under the receiver operating characteristic curve were 0.856 in the training cohort, 0.810 in the TCGA internal evaluation set, 0.768 in GSE138042 and 0.800 in GSE129562, for a mean of 0.793 across the three evaluation cohorts. Decision curve analysis suggested estimated net benefit over treat-all and treat-none strategies across selected threshold ranges. The authors emphasized that the evaluation cohorts contributed to candidate screening and model selection, so performance estimates may be optimistic, and the eight-sample GSE129562 cohort in particular requires independent validation of a prespecified model and preprocessing pipeline.

Communication and spatial analyses then implicated FN1 as a bridge between the epithelial state and lymphatic endothelial cells. NicheNet and CellChat analyses predicted FN1-related interactions between THY5 cells and LECs, including FN1 with SDC4, CD44 and integrin complexes, and in networks built separately for aggressive and non-aggressive cells, FN1 interactions with integrins αVβ1 and α5β1 met the displayed significance threshold only in the aggressive group. Spatial transcriptomics from 13 sections of normal thyroid, papillary thyroid carcinoma and locally advanced disease showed FN1 signaling ranked first within the extracellular matrix–receptor category in all tumor samples, whereas collagen, laminin and thrombospondin pathways led in normal samples. Co-occurrence analysis with ISCHIA found THY5–LEC co-detection exceeding random expectation in four composition clusters across selected tumor sections, although the authors noted that spot-level resolution cannot establish direct epithelial–LEC contact.

Laboratory experiments then tested functional relationships among SERPINA1, SOX4 and FN1 in BCPAP and TPC-1 papillary thyroid carcinoma cell lines. SERPINA1 knockdown reduced SOX4 and FN1 protein expression, SOX4 knockdown reduced FN1 protein and mRNA, and chromatin immunoprecipitation followed by quantitative PCR showed significant enrichment of the assayed FN1 promoter region in anti-SOX4 immunoprecipitates relative to an IgG control. When SOX4 was re-expressed in SERPINA1-depleted cells, FN1 protein rose from 1.33 ± 0.21 to 3.95 ± 0.21 arbitrary units in BCPAP cells and from 0.21 ± 0.16 to 1.87 ± 0.18 arbitrary units in TPC-1 cells, with unadjusted P < 0.001 for both comparisons. The authors cautioned that these molecular rescue experiments do not establish a direct SERPINA1–SOX4–FN1 regulatory sequence, that cellular phenotypes under the SOX4 re-expression combination were not assessed, and that reporter assays and binding-site mutagenesis are needed to confirm SOX4’s regulatory function.

The lymphatic experiments used conditioned medium and Transwell co-culture of tumor cells with lymphatic endothelial cells. Tumor cell conditioned medium and co-culture increased FLT4 and PROX1 protein in LECs, with total NF-κB p65 increasing under both conditions and phosphorylated NF-κB p65 rising under co-culture, alongside increased tube formation, migration and wound healing. When tumor cells carried SERPINA1 knockdown, recipient LECs showed reduced FLT4, PROX1, total and phosphorylated NF-κB p65, and reduced functional readouts; FN1 re-expression in those tumor cells significantly increased all three functional measures relative to knockdown alone, though the combined-treatment group remained below the FN1-overexpression-only group. The molecular receptors mediating these LEC responses remain undefined in the study. Separately, an immunoprecipitation–mass spectrometry screen nominated OGT as a SERPINA1-associated protein, and OGT perturbation in the cell lines altered global O-GlcNAc levels together with SERPINA1 and FN1 protein abundance, although direct OGT-mediated modification of SERPINA1 requires site-level biochemical validation.

In vivo, subcutaneous xenografts derived from SERPINA1-knockdown BCPAP or TPC-1 cells showed significantly smaller maximum diameters and lower weights than controls, with weaker SERPINA1 and FN1 staining by immunohistochemistry, and human tissue microarray images showed stronger SERPINA1 and FN1 staining in tumors than in adjacent non-tumor tissue. The authors stressed that the subcutaneous model assessed tumor growth without examining the thyroid microenvironment, lymphangiogenesis or lymph node metastasis, and that models with lymphatic and nodal endpoints are needed to assess relevance to metastasis. Taken together, the findings support functional links among SERPINA1-associated SOX4–FN1 expression, malignant epithelial behavior and in vitro lymphatic endothelial responses, and they position the exploratory 11-gene classifier as a framework for molecularly informed risk assessment that now requires validation in independent clinical cohorts.

Subject of Research: Molecular programs linking epithelial states and lymphatic endothelial responses in papillary thyroid carcinoma

Article Title: A SERPINA1-associated SOX4–FN1 epithelial program linked to lymphatic endothelial responses in aggressive papillary thyroid carcinoma

Article References: Zhong, Y., Zhang, X., & Yang, Q. (2026). A SERPINA1-associated SOX4–FN1 epithelial program linked to lymphatic endothelial responses in aggressive papillary thyroid carcinoma. Frontiers in Cell and Developmental Biology, 14, 1929991. https://doi.org/10.3389/fcell.2026.1929991

Image Credits: AI Generated

DOI: 10.3389/fcell.2026.1929991

Keywords: papillary thyroid carcinoma, lymph node metastasis, SERPINA1, SOX4, FN1, lymphatic endothelial cells, single-cell RNA sequencing, spatial transcriptomics, tumor microenvironment, extracellular matrix, OGT, xenograft