1 INTRODUCTION

As an organism ages, changes occur at the cellular and tissue levels, and these signs of aging ultimately lead to various diseases and limit life expectancy (Jin, 2010). In general, age is considered the strongest demographic risk factor for most chronic diseases, including cancer (Latorre & Harries, 2017), but this association is not as strong for breast cancer (BC) (Bray et al., 2018).

BC, one of the most common malignant tumors, remains a serious health threat for women. However, the etiology and development of BC are complex and multifactorial. Increasing evidence has emphasized the connection between age and risk of BC; however, the underlying mechanisms remain unclear. The risk of triple-negative breast cancer (TNBC) in women who are < 40 years old has been shown to be more than two times the risk in women who are > 50 years old (Trivers et al., 2009). In addition, several prognostic studies have shown that young women with BC had worse outcomes than older women (Cancello et al., 2010; Lian et al., 2017; Santos Sda, Melo, Koifman, & Koifman, 2013; Sharma & Singh, 2017). Thus, studying the impact of age on BC and underlying age-related mechanisms has become crucial to better understand this disease.

Messenger RNAs (mRNAs) and long noncoding RNAs (lncRNAs) play important roles during development and aging. Belonging to the family of noncoding RNAs, lncRNAs have been shown to be essential in the development and differentiation of normal cells and tissues as well as in the initiation and progression of various pathogenic conditions, especially cancer (Khorkova, Hsiao, & Wahlestedt, 2015). Recent studies have suggested that there are age-dependent variations in lncRNA expression profiles (Arshi et al., 2018; Pereira Fernandes, Bitar, Jacobs, & Barry, 2018). Analyses of the molecular processes involved in some age-related phenotypes have provided compelling evidence confirming that lncRNAs are indeed key regulators in the manifestation of these phenotypes (Grammatikakis, Panda, Abdelmohsen, & Gorospe, 2014; Kour & Rath, 2016). However, most of such studies have focused on neurodegenerative diseases since neurodegenerative diseases are highly associated with increased age (Wan, Su, & Zhuo, 2017). Although there have been few studies using lncRNA and mRNA expression analyses to elucidate the links between age and cancer risk, the reported results have proven to be important. In particular, BC is affected by menstrual hormones, usually in a time-dependent manner.

In the present study, we constructed an lncRNA–mRNA model based on high-throughput data portals to explore the differences at the molecular level among different age groups of patients with BC. In addition, integrated data analysis was performed, and the results clearly showed the impact of age on BC at the molecular level. The workflow is displayed in Figure 1.

2 MATERIALS AND METHODS

3 RESULTS

3.2 Acquisition and characterization of the key mRNAs in age-stratified groups

We performed a feature analysis of the DEMs. A Venn diagram was constructed (Figure 2a), revealing 683 significant targets that overlapped among the three Affymetrix microarrays (Table_1_SuppInfo). A heatmap (Figure 2b) was constructed after clustering the top 100 overlapping DEMs with the highest |log2 FC| values using Morpheus online software. Volcano maps (Figure 2c), a heatmap (Figure_1_SuppInfo) of the ADMs were constructed using the R package. KEGG pathway and GO analyses were used to identify the enriched DEMs and ADMs (Figure 3). The key mRNAs differentially expressed in the BC and normal tissues were extensively enriched in cancer-related pathways, including cancers such as "Bladder cancer," "Prostate cancer," and "Colorectal cancer," and classical carcinogenic pathways such as "Focal adhesion," "Pathways in cancer," and "ECM-receptor interaction." ADMs were enriched in various developmental and differentiation processes, such as "ectoderm development," "keratinocyte differentiation," and "muscle organ development."

Finally, we crossed the DEMs with ADMs and obtained MT1M, MB, HBB, KLHDC7B, MMP9, and TFPI2, which were considered to be potential risk biomarkers between different age groups.

3.3 Age-stratified risk lncRNA identification

We first obtained 348 ADLs by EdgeR package screening (Table_1_SuppInfo). Subsequently, using the “WGCNA” R package, these 348 ADLs were classified via the average linkage clustering, and a total of 11 modules were identified (Figure 4a). Then, the MEs indicated that the turquoise module clearly showed the highest association with age stratification (r = 0.067, p = .03; Figure 4b). Therefore, the turquoise module with 12 ADLs was identified as an important module of lncRNAs for further analysis.

3.4 The coexpression analysis and univariate Cox regression analysis of lncRNAs and mRNAs

First, we performed coexpression analysis of 12 ADLs with six key mRNAs. p < .05 was set as the truncation criteria, and 12 significantly correlated lncRNAs were obtained (Table 1, Figure_2-7_SuppInfo). The lncRNA–mRNA coexpression pattern was visualized with Cytoscape and displayed in Figure 5. Univariate Cox regression analysis that used p < .05 as the screening criteria found three lncRNAs (AL139280.1, AP000851.1, and AC079793.1), and four mRNAs (MT1M, HBB, MMP9, and TFPI2) were closely related to the overall survival of the patients with breast cancer.

Table 1. Identification of coexpressed lncRNAs and key mRNAs

mRNAs	lncRNAs
HBB	LINC02523, AC000061.1, AC092580.2, AL139280.1, AP000851.1, LINC02474, and LINC00113
KLHDC7B	AC000061.1, AC092580.2, AP000851.1, LINC02474, LINC00479, AC079793.1, LINC01170, and LINC00113
MB	AC092580.2, AL139280.1, AP000851.1, LINC02474, AC079793.1, LINC01170, AP004247.2, and LINC00113
MMP9	LINC02523, AC092580.2, AL139280.1, LINC02474, AC079793.1, LINC01170, and LINC00113
MT1M	AC092580.2, AL139280.1, AP000851.1, LINC00479, and LINC01170
TFPI2	AL139280.1, AP000851.1, LINC02474, LINC00479, and LINC00113

• Abbreviations: lncRNAs, long noncoding RNAs; mRNAs, messenger RNAs.

3.5 Prognostic analysis of survival model

To improve the reliability of the risk biomarkers, we used the Kaplan–Meier curves to further clarify the relationship between the risk biomarkers and overall survival. We found that, altogether, two lncRNAs (AL139280.1 and AP000851.1) and three mRNAs (MT1M, HBB, and TFPI2) were significantly (p < .05) related to the overall survival on the basis of survival analysis (Figure 6a–e).

3.6 Differential expression enrichment analysis of the key targets

We aligned the TCGA data and focused on a single gene for the phenotype. We found that the differential regulation of MT1M, HBB, and TFPI2 was significantly enriched in "pyrimidine metabolism," "cell cycle," and "P53 signaling pathway" (Figure 7). This finding is also consistent with that of a previous study that showed that the three signaling pathways were closely related to cancer progression.

4 DISCUSSION

As a complex chronic systemic disease, BC seems to be associated with age (Arshi et al., 2018; Hofstatter et al., 2018). However, there have been few studies that evaluated the effect of age on BC because of the lack of a suitable molecular biomarker for age. Studies have reported that lncRNAs play key roles in complex diseases, especially in development and aging, although most lncRNAs are poorly conserved, and their expression levels are significantly lower than those of mRNAs (Fatica & Bozzoni, 2014; Mercer, Dinger, & Mattick, 2009). In our study, we identified and validated the predictive values of two lncRNAs (AL139280.1 and AP000851.1) and three mRNAs (MT1M, HBB, and TFPI2) as risk models and provided age as a molecular assessment of BC risk.

Bioinformatics has been widely used in research and helps to further characterize molecular mechanisms of disease occurrence and development. We integrated the genes in TCGA and GEO that were differentially expressed and identified six key genes as the potential key markers for age-associated risk stratification. Then, the WGCNA analysis was performed. This is a relatively new and easy-to-understand coexpression network analysis method, which can aggregate coexpressed molecules into "modules" reflecting their interactions in biological systems, and it has successfully been applied to BC research (Li et al., 2019). Therefore, in this study, the lncRNAs in the turquoise module with the highest correlation with age stratification were used for further analysis. Next, hierarchical clustering was performed to evaluate the lncRNAs coexpressed with the six key genes. The basic assumption underlying this clustering was that genes in the same group with similar expression patterns might have similar functions. Finally, the risk biomarkers were identified by univariate Cox regression analysis and survival assessment.

Based on the integrated model, the five risk biomarkers received a good survival assessment. Metallothionein (MT) is a unique cysteine-rich small protein, which plays an important role in maintaining steady-state levels and detoxification of metal ions (Albrecht et al., 2008; Coyle, Philcox, Carey, & Rofe, 2002). In addition, recent experimental evidence suggests that MTs play a key role in tumorigenesis, development, drug resistance, and metastasis by participating in a variety of biological processes (Si & Lang, 2018). As a major member of the MT family, low expression of MT1M is related to poor prognosis in large cell lung cancer, esophageal squamous cancer, and hepatocellular carcinoma (da Motta, De Bastiani, Stapenhorst, & Klamt, 2015; Ding & Lu, 2016; Oka et al., 2009). HBB is a member of the globin family, and in addition to its involvement in gas transport (Bonaventura, Henkens, Alayash, Banerjee, & Crumbliss, 2013), it exerts antitumor activity. For example, studies have shown that HBB inhibits proliferation of human neuroblastoma cells (Maman et al., 2017). Low expression of HBB has been observed in lung adenocarcinoma, and it has been shown to be associated with poor prognosis in non-small cell lung cancer (Bepler et al., 2002). In addition, HBB expression has been found to be inhibited in anaplastic thyroid cancer cell lines/tissues and oral tongue squamous cell carcinoma (Onda et al., 2005; Suresh et al., 2012). As a Kunitz serine protease inhibitor, TFPI2 can inhibit many serine proteases, including trypsin, plasmin, plasma kallikrein, and chymotrypsin (Li et al., 2016). Studies have shown that the low expression of TFPI2 in small cell lung cancer (SCLC) can promote proliferation of the cells (Cao, Guo, Yin, Li, & Zhou, 2018). In addition, TFPI2 is closely related to apoptosis and angiogenesis in cervical cancer, and its expression tends to decrease with the progression of cancer (Zhang et al., 2012).

Due to the novelty of the two lncRNAs, their disease correlation needs to be further characterized. However, their differential expression pattern indicates that age is closely related to the occurrence and prognosis of BC. In addition, we obtained their details through the UCSC database, such as sequence information, length, exon count, position, etc. (Table_2_SuppInfo). Consistent with these reports, we found differences in the performance of these three key targets under different conditions. The results established herein suggest that inhibition of the three key targets may have abnormal consequences in the young group. This observation indicates that the young group is at a higher risk of BC and may be associated with a worse prognosis.

Pathway enrichment analysis is important to understand diseases. In our study, the differentially expressed genes were enriched in pathways associated with cancer, development, and differentiation. This finding is consistent with our hypothesis and demonstrates the credibility of the data we obtained. In addition, GSEA analysis was performed with the key targets MT1M, HBB, and TFPI2, which were significantly enriched in "pyrimidine metabolism," "cell cycle," and "P53 signaling pathway". Previous studies have shown that pyrimidine metabolism is significantly associated with poor prognosis of tumors (Choi & Na, 2018). Through regulation of DNA and RNA metabolism, epigenetic modification of pyrimidine metabolism may affect cell cycle, proliferation, and differentiation, including microRNAs and noncoding RNAs (Fu & He, 2012). In general, it is believed that the correlation between abnormal metabolism and cancer involves the associations among obesity, inflammation, and cancer cell proliferation, invasion, and angiogenesis (Amiri et al., 2018; Tang, Zhou, Hooi, Jiang, & Lu, 2018). Furthermore, cell cycle can influence cancer progression through mitosis and proliferation of cancer cells (Bai et al., 2019). The p53 signaling pathway plays a significant role in cell cycle regulation, aging, metabolism, reproduction, development, and suppression of tumorigenesis (Khan et al., 2019).

Previous epidemiological and clinical studies have also provided possible factors for an effective association between age and BC. The significant decrease in postmenopausal female hormone levels was thought to lower BC risk. For example, the decrease in the incidence of BC in the early 21st century was, at least in part, related to the decline in the use of postmenopausal hormone therapy after the launch of the Women's Health Initiative trial (Ettinger, Quesenberry, Schroeder, & Friedman, 2018). In addition, older women seem to be more likely to take nonsteroidal anti-inflammatory drugs (NSAIDs) as they age because of the onset of various diseases, and NSAID exposure has been shown to be associated with a decreased risk of BC (Hung et al., 2018). A worse prognosis may be related to the type of surgery performed. Given the importance of esthetics and quality of life in societies, conservative surgery is more likely to be chosen by young women in comparison with old women. The study by Arriagada et al. (2003) showed that younger patients who received breast-conserving treatment were more likely to develop advanced BC recurrence than patients who underwent mastectomy. Although various external environmental interventions affect the body to a certain extent, molecular studies based on large real sample data still provide a reliable trend between age and BC risk.

The present study has some limitations. Identification of the two lncRNAs as BC risk biomarkers is novel. Although the evidence suggests an effective association between them and BC risk, the mechanism underlying their link with BC needs to be further characterized. In addition, although more detailed age stratification would be desirable, we divided the population into just two age groups to obtain more data for each group, thereby reducing errors. Nonetheless, our study analyzed the age stratification of BC by summarizing the existing data, and we thereby demonstrated how such data might be used to profile characteristics of the genes associated with specific risks and disorders. Evidence at the molecular level that age is a risk factor was ultimately obtained.

5 CONCLUSION

In summary, our study found that two lncRNAs (AL139280.1 and AP000851.1) and three mRNAs (MT1M, HBB, and TFPI2) were differential risk biomarkers in patients with BC in both the young and old age groups. This finding was further expanded to determine the role of age as a risk factor in the development of BC. Consistent with the results of previous studies, we found that the differences in the expression patterns of the key targets might explain the higher risk and worse outcomes in young BC patients at the molecular level.

CONFLICT OF INTERESTS

The authors declare that they have no conflict of interests.

AUTHOR CONTRIBUTIONS

C. S., X. X. conceived and designed the project; X. M., C. L., C. G., and J. W. collected and analyzed the data; C. S. and C. Z. interpreted the data; X. M., C. L., F. F., and L. W. wrote the manuscript; H. L., J. Z., and L. L. analyzed the data, and review the manuscript; H. L., Z. L., and J. L. participated in preparation of the figures and tables. All authors read and approved the final manuscript.