Development and Validation of Ferroptosis-Related lncRNAs as Prognosis and Diagnosis Biomarkers for Breast Cancer

Breast cancer (BC) is one of the most common malignancies affecting women. Ferroptosis is a novel cancer treatment option. The present study is aimed to identify suitable ferroptosis-related lncRNAs to predict and diagnose BC. Differential expression and Cox regression analyses were used to screen suitable prognostic biomarkers and construct a suitable risk model. We identified four ferroptosis-related differentially expressed lncRNAs (FR-DELs) (LINC01152, AC004585.1, MAPT-IT1, and AC026401.3), which were independently correlated with the overall survival of BC patients. The area under the curve value of the prognostic model using those four biomarkers was over 0.60 in all three groups. The sensitivity and specificity of the diagnostic model using those four biomarkers were 86.89% and 86.73%, respectively. Our present study indicated that these four FR-DELs (LINC01152, AC004585.1, MAPT-IT1, and AC026401.3) could be prognostic biomarkers for BC, although clinical validation studies are required.


Introduction
With more than 2 million new cases (11.7% of all new cancer cases) and 68 thousand deaths (6.9% of all cancer deaths) in 2020 globally, breast cancer (BC) is one of the most common malignancies in women [1]. With the development of early screening and anticancer strategies, the treatment effect of BC has improved dramatically [2]. However, the recurrence rates of BC remain high [3,4]. Additionally, evidence has indicated that BC's prognosis is affected by many factors, such as age, tumor size, grade, lymph node involvement, and lymphatic vascular infiltration [5,6]. The prognosis of BC remains a complex problem, even though many BC prognostic biomarkers have been discovered in ESMO Clinical Practice Guidelines for diagnosis, treatment, and follow-up [5][6][7][8][9][10][11][12][13]. Constructing a new risk assessment model to predict the prognosis of BC patients is necessary.
Long noncoding RNAs (lncRNA) are a class of transcripts with 200 nucleotides in length, which generally do not have protein-coding potential. Previous evidence demonstrated that lncRNAs play an essential role in various biological processes, including the proliferation, apoptosis, migration, and invasion of cancer cells [35][36][37][38]. Many lncRNAs are closely related to the survival status and could be prognostic biomarkers for several cancers [39][40][41]. Therefore, we aimed to identify suitable ferroptosis-related lncRNAs as prognostic biomarkers to predict the outcome of BC.

Survival Analyses and Principal Component Analyses.
After regrouping the samples by the median value of each gene, we performed univariate and multivariate Cox regression analyses using survival, survminer, and regparallel packages in R software (3.6.1). We carry out the principal component analyses (PCA) in R software (3.6.1).

Development and Validation of Prognosis Biomarkers.
To screen for suitable prognosis biomarkers and verify them, we randomly divided the BC patients into training and validation groups (Table 1). We found that 13 FR-DLEs ( Figure 1 Subsequently, we performed similar analyses in the validation and entire groups. These results were shown in Figures 2(e)-2(l). PCA analyses indicated that these BC patients with high-risk values could well be distinguished from the BC patients with low-risk values using these four FR-DELs (Figures 2(d), 2(h), and 2(l)).

Independent Prognostic Factors of Overall Survival.
To know whether the risk model could be used independently, we performed univariate and multivariate Cox regression analyses for different clinical features and the risk model. We found that age, pathological TNM, pathological Stage, and risk model were correlated with the OS, whereas age, pathological M, and risk model were correlated with the OS independently (Figure 3(a)). In the validation group, the age, pathological NM, pathological Stage, and risk model were correlated with the OS by univariate Cox regression analyses. In contrast, age and pathological M were correlated with the OS by multivariate Cox regression analyses (Figure 3(b)). In the entire group, the age, pathological NM, pathological Stage, and risk model were correlated with the OS by univariate Cox regression analyses, whereas the age, pathological M, pathological Stage, and risk model were correlated with the OS by multivariate Cox regression analyses (Figure 3(c)). Compared with pathological TNM and Stage, these results suggested that these four FR-DELs were more accurate as measured by the ROC analyses ( Figure 3(d)-3(f)). The area under curve (AUC) value of the prognostic model using those four biomarkers was over 0.60 in all three groups. Subsequently, we performed ROC curves analyses at 1, 3, 5, and 10-year in the entire group (Figure 3(g)).   (Figure 4(k)).

Correlation Analyses with the Immunity.
Previous studies demonstrated that ferroptosis and the immune system could regulate each other to achieve their antitumor effect. We used the ESTIMATE package to evaluate the immune status. We found that the ESTIMATE score and stromal score were significantly decreased while the immune score and tumor purity were significantly increased in the patients with BC (Supplementary Figures 2(a)-2(d)). In the patients with BC with a high-risk value, the ESTIMATE score, immune scores, and stromal scores were significantly decreased while the tumor purity was significantly increased (Figure 5 We also carried out the difference analyses for the infiltration of immune cells and immune factors of each BC patient. We found that 84 immune cells and factors differ between normal and BC patients. Of the 84, we found that 55 immune cells and factors differ significantly (Figures 6(a)-6(g)). We introduced the correlation analyses for the risk value and these 55 immune cells and factors. We found that four immune cells and factors were significantly correlated with the risk model in the entire group.

Construction of a Diagnostic
Model. X-ray mammography is the golden standard for diagnosing BC patients, which has some drawbacks, such as unsuitability for people under 40, for people with high gland density, can be done no more than twice a year, and high cost [45]. Therefore, to know its role in diagnosing patients with BC, we performed a stepwise logistic regression analysis for these four FR-DELs and constructed a diagnosis model according to previous studies. The sensitivity and specificity of the diagnosis model were 86.89% and 86.73%, respectively (Table 2). We also

Discussions
With the development of early screening and anticancer strategies, the treatment of BC has improved dramatically. However, the recurrence rates of BC remain high. Therefore, it is necessary to construct a new risk prediction model to predict the outcome. Ferroptosis is a novel type of cell death, which could be a novel anticancer therapeutic strategy. In the present study, we identified four FR-DELs (LINC01152, AC004585.1, MAPT-IT1, and AC026401.3) correlated with the OS of BC.
The ROC curve showed that the AUC value of the risk model constructed by these four FR-DELs was over 0.6, which indicated that these four FR-DELs (LINC01152, AC004585.1, MAPT-IT1, and AC026401.3) could be prognosis biomarkers for BC in the outcome prediction. Chen et al. found that LINC01152 is up-expressed in HBVpositive hepatocellular carcinoma (HCC) [46]. LINC01152 could promote HCC cell proliferation and tumor formation in nude mice [46]. Wu et al. found that LINC01152 was upregulated in glioblastoma multiforme [47]. LINC01152 could promote the progression of GBM by upregulating MAML2 [47]. But in the present study, we found that LINC01152 was downregulated, which differed from previous studies. The patients with BC with low expression of LINC01152 displayed worse OS. In the present study, we found that the expression of AC004585.1 was significantly increased in BC and significantly decreased in BC with a high-risk value. By the univariate and multivariate Cox regression analyses, we found that AC004585.1 was correlated with the OS of BC. Our present study further increased the possibility of AC004585.1 as a prognosis biomarker for BC [48,49]. The expression of MAPT-IT1 was significantly increased in BC and significantly decreased in BC patients with a high-risk value. In the present study, we found that MAPT-IT1 was correlated with OS of BC significantly. The HR ratio of less than 1.0 was consistent with the previous study [50]. There have been few studies on the function of MAPT-IT1, but our results further confirmed the possibility of MAPT-IT1 as a prognosis biomarker for BC. The expression of AC026401.3 was significantly increased in BC. We found AC026401.3 was correlated with the OS of BC. The HR ratio was less than 1.0. The patients with high expression of AC026401.3 displayed better OS. Our present study indicated that AC026401.3 could be a prognosis biomarker for BC. Previous studies also demonstrated that AC026401.3 was correlated with the OS and could be a prognosis biomarker for several cancers, including renal adenocarcinoma, colon adenocarcinoma, and ovarian cancer. These results indicated that AC026401.3 might play an essential role in developing a variety of cancers.
Although imaging diagnoses have high sensitivity and specificity, these methods also have some major drawbacks, such as being unsuitable for people under 40, for people with high gland density, and can be done no more than twice a year [45]. Although the sensitivity and specificity of this model were lower than those of imaging diagnosis, molecular biotechnology examination can overcome these shortcomings. LncRNA can regulate the cell cycle of tumor cells and various cell signaling pathways of cancer cell proliferation, apoptosis, migration, and invasion of cancer cells [35][36][37][38], which could be prognostic and diagnostic biomarkers for several cancers [39][40][41]. Such as Ye et al. found that the AUC value of myocardial infarction associated transcript as a diagnostic biomarker was 0.86, and the sensitivity and specificity were 87.80% and 75.61%, respectively [51]. El-Ashmawy et al. found that lncRNA-ATB had a high AUC value (AUC: 0.844, p < 0:01) for early diagnosis of breast cancer in patients with stage I-II [52]. In our present study, we found that the AUC value of the diagnostic model reached 0.927. The sensitivity and specificity of this diagnostic model were 86.89% and 86.73%, respectively. These results suggest that this model may be a better diagnostic model for breast cancer, although further studies are needed to verify it. Additionally, although the sensitivity and specificity of this model were lower than those of imaging diagnosis, it can overcome these shortcomings.

Conclusions
In the present study, we found four FR-DELs (LINC01152, AC004585.1, MAPT-IT1, and AC026401.3) that could be used as prognosis and diagnosis biomarkers to predict the outcome of BC. Our study provides a new perspective on the prognosis and diagnosis of BC, although there were several limitations, especially due to the lack of clinical validation.

Data Availability
The data that support the findings of this study are openly available in TCGA at https://portal.gdc.cancer.gov/.

Conflicts of Interest
The authors declare no competing interests.

Authors' Contributions
Zhi-Yong Yao and Chaoqun Xing contributed equally to this work.