M. Asaoka1,2, S. K. Patnaik1, F. Zhang1,3, T. Ishikawa2, K. Takabe1,2 1Roswell Park Cancer Institute,Surgical Oncology,Buffalo, NY, USA 2Tokyo Medical University,Department Of Breast Surgery And Oncology,Shinjuku, Tokyo, Japan 3Dartmouth Medical School,Lebanon, NH, USA
Introduction:
Lymphovascular invasion (LVI) is a significant prognostic factor in breast cancer. Multiple studies have reported that LVI is associated with increased number of tumor lymphatic vessels. However, the relationship of LVI with lymphatic-specific gene (LSG) expression is yet to be clarified. We studied which genes associate with LVI so that they may be useful as diagnostic markers.
Methods:
LVI status and mitotic scores of pre-treatment primary tumors were collated from pathology reports of 1046 breast cancer cases of The Cancer Genome Atlas (TCGA) project. Survival and other clinical data were obtained from TCGA publications. RNA sequencing counts from TCGA were normalized and log2-transformed to generate tumor gene expression data. LSGs, with ≥20x higher expression in lymphatic compared to blood endothelial cells, were identified from published studies. Clustering of cases by gene expression was examined with multi-dimensional scaling and tSNE methods. Top-scoring pair (TSP) and support vector machines (SVM) classifiers were used to quantify LVI-predictive value of LSG expression in leave-one-out cross-validation. Survival analysis was performed with Cox regression. Hallmark and Reactome sets were used for gene-set enrichment analysis. Categorical variables were compared by Fisher exact test, with P <0.05 deemed significant.
Results:
LVI was present in 242 (37.9%) of the 639 cases for which its status was noted. Patients with LVI demonstrated significantly worse outcome than those without for both 5-year progression-free (69.1% vs. 85.3%, P <0.01) and disease-specific survival (84.9% vs. 93.8%, P <0.01). Incidence of LVI was 2x higher in HER2 type compared to the other 3 subtypes (68.4% vs. 32.4%-37.8%; P = 0.04). LVI was also significantly related with late stage (I+II vs. III+IV: P <0.01; OR, 6.1), axillary lymph node metastasis (P <0.01; OR, 8.7), higher mitotic score (1+2 vs. 3: P <0.01; OR, 1.8), and higher nuclear score (1+2 vs. 3: P <0.01; OR, 2.1), but not with estrogen receptor status or tubular score (1+2 vs. 3). Of 84 LSGs, which included classical lymphatic markers like PROX1 and LYVE1, expression of only 3 was higher in LVI+ cases, and LVI+ and – cases did not cluster separately by expression of the 84 genes. LSGs also had poor predictive value for LVI in classification analyses (62% accuracy for both TSP and SVM). In global analysis, 950 genes were differentially expressed, with cell-cycle regulator gene SPDYC up-regulated the most in LVI+ tumors. Highest enrichments in the LVI+ group were observed for gene-sets related to cell proliferation.
Conclusions:
We did not observe any association of tumor lymphatic-specific gene expression with LVI. However, LVI was strongly correlated with expression of genes involved in cell proliferation. This was also reflected in association of LVI with higher mitotic scores, suggesting that highly proliferative cancer cells have a tendency to invade lymphatic vessels.