Traditionally, the size of breast cancer at diagnosis is seen as a key determinant of clinical outcome. However, some aggressive subtypes challenge this notion despite being small (≤ 1 cm) [6]. In certain subtypes, tumor size, lymph-node status, and prognosis may be uncoupled due to a disproportionate number of metastatic cancer cells relative to tumor size [5]. Therefore, understanding the underlying mechanisms of distant metastasis in small-size tumors is an important clinical issue for appropriate treatment decisions. We used machine learning Random Survival Forest and WGCNA techniques to identify nine prognostic genes (ABHD11, DDX39A, G3BP2, GOLM1, IL1R1, MMP11, PIK3R1, SNRPB2, and VAV3) that were predictive of DMFS. Their functions were related to “cell adhesion” (IL1R1 and MMP11), “cell–cell signaling” (ABHD11 and PIK3R1), “cellular metabolic process” (G3BP2, GOLM1 and SNRPB2), “cell cycle phase” (DDX39A), and “not specific” (VAV3) from the results of WGCNA. Patients with higher risk scores had a three-to-fourfold increased risk of developing distant metastasis. When ER status was added, the risk score had good prediction efficacy, with AUCs of 0.75 and 0.79. Furthermore, the risk score reflected clinical characteristics such as lower age, poor Grading, ER-, PR-, and a higher proportion of luminal A/B.

In this study, we employed nomogram models to simplify and visualize prognostic genes in the prediction of DMFS shown in Fig. 4C, D. By obtaining the normalized gene expression of these nine genes, we can determine the individual corresponding scores by referring to the “Points” scales and summing them to derive the “Total Points.” For example, using the nomogram model in Fig. 4D, a hypothetical patient with the following values: DDX39A (12.2 mapping to 100 points), VAV3 (11 mapping to 10 points), GOLM1 (5 mapping to 0 points), MMP11 (11.5 mapping to 20 points), G3BP2 (12.5 mapping to 99 points), PI3KR1 (10.8 mapping to 10 points), ABHD11 (7.5 mapping to 0 points), IL1R1 (8 mapping to 0 points), and SNRPB2 (12.2 mapping to 0 points) would accumulate a total of 239 (100 + 10 + 0 + 20 + 99 + 10 + 0 + 0 + 0) points. These total points correspond to an estimated 3-year survival probability of approximately 73%, a 5-year survival probability of around 59%, and a 7-year survival probability of about 57%.

Previous molecular studies identified genetic markers involved in the prognosis of small breast tumors. For instance, the expression of stromal type IV, an extracellular matrix protein, in small invasive breast cancers has been linked to a higher risk of developing distant metastasis and poorer survival outcomes. It is possible that stromal type IV collagen can promote metastasis formation by supporting cancer cell survival and tumor progression, and high levels of type IV collagen in the metastases appear to be beneficial for metastatic growth [8]. The expression of P-cadherin has been found to be highly predictive of a poor prognosis in small, node-negative breast cancers. P-cadherin has an important role in maintaining the structural integrity of the epithelium [9]. These studies highlighted the dysregulation of the extracellular matrix in the progression of small breast tumors, consistent with our findings on IL1R1 and MMP11, which are involved in “cell adhesion”. In addition, significant correlation of IL1R1 with MMP11 expression was found and involved in breakdown of extracellular matrix, tissue remodeling, and metastasis [10]. Breast tumors infiltrated by MMP-11+ mononuclear inflammatory cells are more likely to metastasize, have high levels of interleukin (IL)-1, IL-5, IL-6, IL-17, interferon (IFN), and NFB, and an increased CD68+/(CD3+CD20+) cell ratio at the invasive front. These factors are implicated in the crosstalk between tumors and their inflammatory microenvironment [11]. MMP11 expression in mononuclear inflammatory cells was associated with shorter relapse-free survival and overall survival [12].

Early in tumorigenesis, IL-1R1 signaling suppresses mammary tumor cell proliferation and inhibits breast cancer outgrowth and pulmonary metastasis. In breast cancer, IL-1-mediated IL-1R1 signaling is tumor-suppressive [13]. Patients treated with anti-estrogen therapy have increased IL1R1 expression, which predicts treatment failure [14]. In addition, inhibition of IL-1 signaling with the anti-IL1β antibody or the IL1R antagonist inhibits bone metastasis in pre-clinical models of breast cancer [15]. Colorectal cancer patients who did not respond to Cetuximab blockage had higher levels of IL1R1 than responsive subjects, and high levels of IL1R1 are predictive of survival [16].

In this study, we identified PIK3R1-IL1R1-MMP11-GOLM1-VAV3-EGFR protein–protein interaction network connected by Epidermal Growth Factor Receptor (EGFR). EGFR and its downstream pathway regulate epithelial–mesenchymal transition, migration, and tumor invasion and that high EGFR expression is an independent predictor of poor prognosis in inflammatory breast cancer. Targeting EGFR enhances the chemosensitivity of tumor cells by rewiring apoptotic signaling networks in Triple-Negative Breast Cancer [17]. Therefore, this genetic network may play crucial role in triggering small-size breast tumor metastases.

GOLM1 has been identified as a potential target for cancer therapy, because it is overexpressed in many solid tumors, promotes tumor growth and metastasis, and leads to poor survival [18]. GOLM1 could promote breast cancer cell aggressiveness by regulating matrix metalloproteinase-13 (MMP13) [19]. Knocking down GOLM1 expression further increased the epigallocatechin gallate (a natural migration-inhibiting substance) treatment effect in breast cancer cells [18]. What’s more, GOLM1 also acts as a positive regulator of Programmed Cell Death Ligand 1 (PD-L1) expression via the EGFR/Signal Transducer and Activator Of Transcription 3 (STAT3) signaling pathway in the human hepatocellular carcinoma [20].

VAV3, a GEF for Rho family GTPases, belongs to the VAV protein family [21]. It is a downstream signal transducer of EGFR/HER2 and could bind to several partners, including PI3K, modulates cell morphology, and induces cell transformation [22]. High nuclear VAV3 expression in tumor cells was associated with poorer endocrine therapy response [23]. It complexes with ERα and together enhance ERα-mediated signaling axis, participating in breast cancer development and/or progression [24]. The depletion of VAV3 reduced the viability of cell models of acquired endocrine therapy resistance [23].

In this study, we unrevealed that G3BP2, PIK3R1, and SNRPB2 were co-regulated by hsa-miR-302a-5p. The MiR-302 family exerts antitumor effects in several cancers [25]. MiR-302a, -b, -c, and -d were found to cooperatively inhibit BCRP expression to increase the drug sensitivity of breast cancer cells [26]. Dysregulation of the phosphoinositide 3-kinase (PI3K) pathway contributes to the development and progression of tumors. PIK3R1 underexpression is an independent prognostic marker in breast cancers [27]. Silencing PIK3R1 enhanced the sensitivity of breast cancer cell lines to rapamycin [28], implicating a negative role of PIK3R1 in PI3K pathway activation. Both PIK3R1 and EGFR were involved in anti-cancer drug effects. Schisandrin A (SchA), a good anti-cancer drug, significantly down-regulated EGFR, PIK3R1, and MMP9 but up-regulated cleaved-caspase 3, thus inhibiting the migration and promoting the apoptosis of MDA-MB-231 cells [29].

G3BP2 (G3BP Stress Granule Assembly Factor 2) regulates breast tumor initiation by stabilizing squamous cell carcinoma antigen recognized by T cells 3 (SART3) mRNA. The loss of G3BP2 inhibits breast tumor initiation, possibly lead to improved cancer treatments [30]. Cell-cycle checkpoint regulator MK2 or G3BP2 inactivation sensitizes cisplatin-resistant TNBC cell lines to cisplatin [31]. Suppression of G3BP2 inhibits the immune checkpoint molecule PD‐L1 due to mRNA degradation [32].

SNRPB2 (Small Nuclear Ribonucleoprotein Polypeptide B2) is the encoding gene of protein U2 small nuclear ribonucleoprotein B, one component of spliceosome. While primarily studied in hepatocellular carcinoma [33], SNRPB2 is a novel gene of interest in breast cancer. Many oncogenic insults deregulate RNA splicing, often leading to hypersensitivity of tumors to spliceosome-targeted therapies (STTs). Mis-spliced RNA is a molecular trigger for tumor killing through viral mimicry. STTs cause widespread cytoplasmic accumulation of mis-spliced mRNAs, many forming double-stranded structures in MYC-driven triple-negative breast cancer [34].

DDX39 encodes protein DExD-Box Helicase 39A that unwinds double-stranded RNA in an ATP-dependent manner. It is involved in transcription, splicing, ribosome biogenesis, RNA export, RNA editing, RNA decay, translation, and the protection and maintenance of telomeres. Increased DDX39 mRNA expression was associated with poor outcomes in ER-positive breast cancers. Inhibiting DDX39 could enhance the sensitivity of MCF-7 to doxorubicin [35]. In hepatocellular carcinoma (HCC), DDX39 knockdown inhibited HCC cell migration, invasion, growth, and metastasis by activating the Wnt/β-catenin pathway [36].

ABHD11 (Abhydrolase Domain Containing 11) is a protein-coding gene. ABHD11 antisense RNA 1 (ABHD11-AS1) is highly expressed in many cancers. Several studies have highlighted the clinical importance of ABHD11-AS1 in cancer prognosis, diagnosis, stage prediction, and treatment response. The ABHD11-AS1 has been shown to cause cancer by sponging various microRNAs (miRNAs), altering signaling pathways such as PI3K/Akt, epigenetic mechanisms, and N6-methyladenosine (m6A) RNA modification [37]. ABHD11 and Esterase D could predict the development of distant metastases and the presence of aggressive lung adenocarcinomas [38].

In the present study, although prognostic genes were validated in the independent dataset, further testing on clinical data is warranted. HER2 status, Ki67, and lymphovascular invasion were not included for analysis due to too much missing data. It should be included in the future work. In addition, the current mRNA expression data require experimental studies to validate the findings and elucidate the mechanisms.