Research Report: Validation and Analysis of a Prognostic Model for Oral Squamous Cell Carcinoma Constructed with TLS-Related Genes

Background and Research Motivation

Oral Squamous Cell Carcinoma (OSCC) is the most common subtype of Head and Neck Squamous Cell Carcinoma (HNSCC), exhibiting a high tendency for lymph node metastasis, especially to the cervical lymph nodes. According to the 2022 Global Cancer Observatory (GCO) report, there were approximately 380,000 new cases of OSCC, with nearly 188,000 deaths caused by this disease. GCO expects the incidence and mortality rates of this disease to continue to grow by 2040. Currently, the prognosis of OSCC patients is primarily determined by evaluating tumor size, lymph node status, and distant metastasis, using the TNM staging system. However, even among patients at the same TNM stage, there are significant differences in prognosis, reflecting the TNM staging system’s inadequacies in capturing the immune heterogeneity of OSCC.

In recent years, Tertiary Lymphoid Structures (TLS) have gained widespread attention for their immune regulatory roles in non-lymphoid tissues and positive impact on cancer patients’ prognosis. TLS are typically directly exposed to the tumor microenvironment, providing a more specific immune response than other non-infiltrative lymphocytes. In the field of OSCC, the potential of TLS as an innovative prognostic indicator has been preliminarily validated. The presence of TLS correlates positively with the overall survival (OS) and disease-free survival (DFS) of OSCC patients, with the potential to supplement the prediction of postoperative clinical outcomes. However, due to the lack of recommended molecular markers for TLS in clinical practice, its application in helping clinicians develop personalized treatment plans remains limited.

Research Source

This study was conducted by Bincan Sun and others, with the main authors affiliated with the Department of Oral and Maxillofacial Surgery and related research institutions of Xiangya Hospital of Central South University. The article was published in the 2024 issue of the journal “Cancer Cell International.”

Research Methods

1. TLS Detection and Quantification

The study first utilized a Convolutional Neural Network (CNN) to detect and quantify the distribution and area of TLS in HE-stained whole slide images of OSCC. It used a modified DeepLab V3+ CNN to detect candidate TLS regions and an active contour model to optimize their boundaries. By calculating lymphocyte counts, TLS area size, and density, candidate regions were ultimately identified as TLS. In the images, the researchers classified TLS into three types: Aggregates (AGG), Primary Follicle-Like TLS (FL1) lacking germinal centers, and Secondary Follicle-Like TLS (FL2) with germinal centers formed.

2. Screening of TLS-Related Genes (TLSRGs)

Differentially expressed gene analysis (DEGs) of 336 OSCC samples from the TCGA database was conducted to screen for genes related to TLS. Further, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) functional enrichment analyses revealed the critical roles of TLSRGs in chemokine signaling pathways, dendritic cell antigen processing, and presentation.

3. Construction of Prognostic Scoring Model and Predictive Nomogram

The study evaluated the relationship between TLSRGs and OS through univariable and multivariable Cox regression analyses and employed Lasso and stepwise regression to select CCR7, CXCR5, and CD86 as key genes to construct the risk scoring model. The risk score formula is: [ \text{Risk Score} = (0.3169 \times \text{Expression Level of CD86}) + (-1.3345 \times \text{Expression Level of CXCR5}) + (-0.2974 \times \text{Expression Level of CCR7}) ] Based on this model, patients were divided into high-risk and low-risk groups. The prognostic accuracy was analyzed through Kaplan-Meier (K-M) survival curves and ROC curve analysis. Additionally, the study incorporated other clinical variables to build a personalized prognostic predictive nomogram, further enhancing the model’s clinical application value.

4. Relationship Between TLS, Immune Cell Infiltration, and Risk Score

Utilizing the Tumor Immune Estimation Resource (Timer) tool, the study analyzed the relationship between TLS and immune cell infiltration characteristics. It was found that B cells, CD4+ T cells, CD8+ T cells, and macrophages in the TLS+ group were significantly higher than those in the TLS- group, indicating that TLS play an important immune role in the tumor microenvironment. Simultaneously, the proportion of TLS region negatively correlated with the risk score, suggesting that the presence of TLS negatively correlates with immune cell infiltration, risk score, and lymph node metastasis.

Research Results

Independent Role of TLS in OSCC Prognosis

The study confirmed the positive impact of TLS on the prognosis of OS and DFS in OSCC patients through various statistical analyses. K-M survival analysis showed that the OS and DFS of the TLS+ group were significantly better than those of the TLS- group, with the maturity and area proportion of TLS having a closer relationship with the prognosis; larger TLS areas were associated with better prognosis.

Role of TLSRGs in Risk Prediction Model

CCR7, CXCR5, and CD86, selected through Cox regression and Lasso analysis, significantly impact the prognosis of OSCC patients; the construction of the three-gene model further enhanced prediction capabilities. The critical roles of CXCR5 and CCR7 in immune cell migration, lymphoid tissue formation, and maintenance are indispensable in TLS formation and anti-tumor immunity, while the downregulation of CD86 inhibited T cell activation.

Relationship Between Risk Score and Immune Cell Infiltration

The presence of TLS in OSCC is closely related to the infiltration of B cells, CD4+ T cells, CD8+ T cells, and macrophages, which play vital roles in anti-tumor immunity. The study indicated a positive correlation between a high proportion of TLS and immune cell infiltration, reflecting the importance of TLS in immune defense. Additionally, the study found that the upregulation of CXCR5 and CCR7, and the downregulation of CD86 and risk scores, were negatively correlated with lymph node metastasis.

Innovation and Significance of the Research

1. Methodological Innovation: This study is the first to use a convolutional neural network for automated detection and quantification of TLS in OSCC samples, making TLS evaluation more efficient and precise. Additionally, the risk scoring model constructed with TLSRGs offers new possibilities for individual patient assessment.
2. Prognostic Value: The study indicated that the prognosis of TLS+ patients was significantly better than that of TLS- patients, with the area proportion of TLS having more predictive value than maturity. The establishment of the three-gene model further enhanced the sensitivity and specificity of risk prediction, providing a new reference for stratification and prognosis of OSCC patients.
3. Personalized Model Application: The risk score-based nomogram combined with clinical variables provides a basis for individualized treatment plans, potentially improving the existing staging systems and risk stratification models.

Conclusion

The OSCC prediction model and individualized nomogram based on TLSRGs established by this study provide scientific evidence for clinical application in OSCC. Compared to previous studies, this research has notable advantages: using convolutional neural networks to automatically identify TLS offers possibilities for the application of imaging analysis technologies in tumor prognosis; exploring the relationship between TLS and lymph node metastasis and immune cell infiltration, and for the first time constructing an OSCC nomogram prediction model based on TLSRGs. The distribution and area of TLS in OSCC can serve as effective prognostic markers and indicators for predicting response to immunotherapy, providing theoretical support for precision treatment of future OSCC patients.