Many studies have investigated the relationship between estrogen and NSCLC [
16‐
21]. Several studies have shown that different estrogen reactivities might affect the prognosis of NSCLC patients [
42,
43]. In our previous study, we showed that estrogen is a pro-tumor factor for NSCLC [
22]. However, models for accurately predicting the effects of estrogen reactivity on the OS of NSCLC patients are absent. The screening of prognostic biomarkers based on bioinformatics methods has been widely performed in studies on lung cancer. In this study, we used hallmark gene sets as the reference gene sets, and the samples in the TCGA-LUAD cohort were divided into the high-estrogen reactivity group and the low-estrogen reactivity group according to their estrogen reactivity scores based on the results of the GSVA. A risk signature was constructed based on DEGs between the groups. The results of the univariable and multivariable Cox regression analyses confirmed this risk signature as an independent prognostic factor in patients with LUAD, and this signature was validated using the GSE31210 dataset. We also showed the tumor-promoting effects of estrogen in vivo using an orthotopic mouse model of LUAD. Finally, we constructed a nomogram based on the risk signature and some clinical characteristics to predict the one-year, three-year, and five-year OS of the LUAD patients. The results showed that our nomogram was similar to the observed scenario.
Many immune cells were found to infiltrate tumor tissue, and these immune cells were involved in tumor metastasis, drug resistance, and immune escape [
41]. Therefore, we analyzed the differences in the immune cell infiltration status between the groups with different estrogen reactivities. We found that the abundance of M2 macrophages increased while the abundance of M1 macrophages decreased in the high-estrogen reactivity group. In the tumor microenvironment, M1 macrophages have an antitumor effect, while M2 macrophages can promote immunosuppression [
44]. These results suggested that estrogen might strongly influence tumor immune escape. We also studied the expression levels of five immune checkpoints, which were closely related to antitumor immunity [
45]. Patients with high expression levels of these checkpoints respond better to immunotherapy. Our results showed that their expression levels were higher in the low-estrogen reactivity group, which suggested that immunotherapy had a greater effect on the patients in this group.
By performing differential expression analysis, we obtained 795 downregulated genes and 291 upregulated genes in the high-estrogen reactivity group. Among them, four key prognostic genes (LINGO2, DKK1, IGFBP1, and GTF2H4) were identified and used for constructing a risk signature. The results of the K-M survival analysis showed that our risk signature had excellent prognostic value. LINGO2 (Leucine Rich Repeat And Ig Domain Containing 2) was found by Carim-Todd et al. to be expressed in the early developmental stages of the central nervous system and also in the limbic system and cerebral cortex of adult tissues [
46]. Studies on LINGO2 are limited and are mostly related to non-neoplastic diseases, such as Parkinson’s disease [
47,
48]. Only one study on the molecular mechanism in gastric cancer found that it influences the progression of gastric cancer by altering gastric cancer initiation, stem cells, and cell motility tumorigenesis [
49]. DKK1 (Dickkopf WNT Signaling Pathway Inhibitor 1) is a secreted protein that antagonizes the Wnt/b-catenin pathway. It regulates bone formation and affects the development and progression of bone metastases [
50]. Several studies have indicated its role in the development, progression, and metastasis of tumors, including pancreatic ductal adenocarcinoma, breast cancer, ovarian cancer, cervical cancer, and endometrial cancer [
51]. In deficient mismatch repair colorectal cancer, DKK1 can also attenuate the efficacy of immunotherapy by suppressing CD8 + T cells [
52]. Based on the findings of studies on DKK1 in cellular and animal models, several clinical trials have been initiated to evaluate the safety and efficacy of anti-DKK-1 neutralizing antibodies in cancer [
53]. One study used a panel of four genes (including DKK1) to predict the OS of LUAD [
54]. The regulatory relationship between DKK1 and estrogen was also investigated. By preventing an increase in DKK1 levels, low physiological levels of E2 protect the hippocampal CA1 region against global cerebral ischemia [
55]. IGFBP1 (Insulin Like Growth Factor Binding Protein 1) is the most prevalent IGFBP found in amniotic fluid and is typically expressed in the placenta, endometrium, and liver in a tissue-specific manner. After being secreted, IGFBP functions by interacting with IGFs [
56]. Several studies have investigated its role as a biomarker in tumors such as gastrointestinal tumors and prostate cancer [
57,
58]. IGFBP1 is specifically expressed in ovarian clear-cell adenocarcinoma [
59]. In breast cancer cells, 4-OHT suppresses IGF-1 signaling due to the accumulation of extracellular IGFBP1, which is mediated by GPER1 and CREB [
60]. The transcription factor II H (TFIIH) component GTF2H4 (also known as p52) is involved in nucleotide excision repair [
61]. Studies on its role in tumors are limited. Overall survival was found to be strongly correlated with GTF2H4 SNPs in lung cancer [
62]. Estrogen regulation of IGFBP1 and DKK1 has been reported in previous studies. In breast cancer cells, estrogen regulates IGFBP1 expression via GPER1 [
59]. The expression of DKK1 in CD4 + and CD8 + T cells was increased in ovariectomized mice. No literature has reported the correlation between LINGO2 and GTF2H4 and estrogen [
63]. Our findings suggested that the estrogen signaling pathway might affect the progression and prognosis of LUAD by regulating the expression of these four genes.
In this study, we divided LUAD patients into the high-estrogen reactivity group and the low-estrogen reactivity group, which had clinically important prognostic significance. Based on this grouping, a risk signature and a nomogram were constructed, which could effectively predict the prognosis of LUAD patients. In another study, we showed the pro-cancer effects of estrogen using a subcutaneous tumor model. In this study, we confirmed the effects using an orthotopic mouse model and obtained more reliable results. However, our study had some limitations that should be addressed in subsequent studies. First, our study was based on data collected from public databases, and choosing the median as a threshold to binarize variables might not be the best solution. Thus, our findings need to be validated by conducting large prospective clinical trials. Second, we analyzed the differences in immune cell infiltration between different estrogen reactivity groups, but these differences need to be verified experimentally, which we aim to perform in our next study. Third, the ROC curve and AUC values of the prognostic signature in the validation set are not ideal (Supplement Fig.
1A, B). It might be brought on by variations in clinical characteristics such as patient counts, tumor stages, and smoking histories. Finally, information on the effects of three of the four key prognostic genes on LUAD (i.e., except DKK1) investigated in this study is limited. Hence, further cell and animal experiments need to be performed to elucidate the functions of these genes.