In this study on circulating miRNAs in breast cancer, we found models able to differentiate controls from BC cases and controls from different types of BC cases, namely cases detected by screening, cases which are disease-free in the follow-up and cases that are not disease-free in the follow-up. Although there is some degree of overlapping between the different models, it is remarkable that their calibration (i.e., their ability to discriminate between cases and controls) increases with the severity of the cancer, as shown by their areas under the ROC curve: 0.6327 to distinguish between controls and cases diagnosed by screening, 0.7345 to differentiate between controls and disease-free cases in the follow-up and 0.8216 to distinguish between controls and cases with active disease.
A total of eleven miRNAs were selected in our four models. Three miRNAs appear as downregulated (miR-101-3p, miR-186-5p and miR-29a-3p) and eight as upregulated in cases (miR-423-3p, miR-139-5p, miR-324-5p, miR-1299, miR-142-3p, miR-1307-3p, miR-331-3p and miR-21-3p).
miRNAs downregulated in breast cancer
miR-101-3p is consistently downregulated in our four models. miR-101-3p has been described as downregulated in women with BC [
28,
29]. It promotes BC cell apoptosis by targeting JAK2 (Janus kinase 2) [
30] and inhibits BC growth by targeting CXCR7 (CXC chemokine receptor 7) [
28] and STMN1 (Stathmin1) [
29]. Harati et al. [
31,
32] observe that the miR-101-3p is downregulated in metastatic breast cancer cells in comparison with less invasive cells due to the COX-2 (cyclooxygenase-2) induction. Liu et al. [
33] consider that the miR-101-3p inhibits the expression of AMPK (AMP-activated protein kinase) in triple negatives breast cancer, whose dysfunction has been linked to breast cancer; while Zhao et al. [
34] reflect that the overexpression of this miRNA could induce changes in the macrophages, increasing cellular proliferation and migration.
miR-186-5p appears as downregulated in our models comparing controls with all BC cases and with disease-free cases, in agreement with Giussani et al. [
20]. This miRNA seem to inhibit CXCL13 (C-X-C motif chemokine ligand 13) and is associated with tumor staging and size [
35]. Another way of action was raised by Hamurcu et al. [
36]. They contemplate that the FOXM1 (Forkhead Box 1), which is upregulated in breast cancer cells, exerts its oncogenic effects acting over the miRNA expression. In this work, one of the miRNAs with altered expression is miR-186-5p whose upregulation is associated with the development and progression of breast cancer [
36]
miR-29a-3p only appears downregulated in the model comparing controls with all BC cases. Previous results on miR-29a-3p are contradictory. While Wu et al. (2019) found a tumorigenesis role via downregulation of the histone H4K20 trimethylation, Wu et al. [
37] and Li et al. [
38] found it was downregulated in BC. In addition, some authors [
39,
40] indicate that when the miRNA is sponged by a circRNA such as ACAP2 (circACAP2) [
39] or PVT1 (Pvt1 oncogene) [
40], cellular invasion, proliferation or migration increased.
miRNAs upregulated in breast cancer
miR-423-3p is upregulated in three out of four models of ours: controls vs. all BC cases, controls vs. disease-free cases and controls vs. non-disease-free cases in the follow-up. Consistent with these results, Murria et al. [
41] found that the miRNA hyperexpression is associated with estrogen or progesterone receptor positive breast cancers. In addition, the same authors [
42] found that this miRNA is part of a signature, together another nine (being miR-423-3p the best differentiated), that allows discriminated hereditary and non-hereditary breast cancers. It has been experimentally observed that miR-423-3p promotes cell proliferation in BC cell lines, and its silencing leads to a decrease in cell proliferation [
43]. Consistent with these results, the same authors [
41] found that the miRNA hyperexpression is associated to estrogen or progesterone receptor positive breast cancers. However, it shows a lower expression in triple negative breast cancers [
41]. No reference against our results was found.
Contrary to our results, miR-139-5p had previously found downregulated in BC [
22,
44]. We have found no other article in agreement with our results and have no explanation for this disagreement.
Furthermore, miR-324 is upregulated when compared controls vs. all BC cases and in the comparison of controls vs. screening. In the bibliography, miR-324-5p was found upregulated in BC cases in Giusani et al. [
20], Kuo et al. [
45], Hong et al. [
46], Lou et al. [
47], and Turashvili et al. [
48]. All of them have demonstrated that its upregulation is associated with worse prognosis, especially in triple negative breast cancer cancers [
46‐
48]. Lou et al. [
47] proposed a possible mechanism for this miRNA. They analyzed the GPX3 (Glutathione peroxidase 3) in BC and found that its low expression increased cell proliferation and this could be due to the release of miR-324-5p inhibition.
miR-1299 inhibits tumor cell proliferation, invasion and metastasis [
49] and, so, it was found downregulated by Liu et al. [
50]. This result concurs with its role in other cancers and contradicts our result which shows it as upregulated in BC. In fact, Sant et al. [
51] propose that the ciRS-7 sponge the miR-1299 in triple negative breast cancer cells, leading to increase the migration and invasion cells. In the same way, Zhang et al. [
52] conclude that the circ-UBR1 sponge also the miR-1299, being able to inhibit the apoptosis and facilitating the proliferation cell and metastasis.
Several authors have reported that miR-142-3p is downregulated in BC and exerts a protective role via inhibiting BC cell invasiveness [
53] or targeting HMGA2 (high mobility group AT-hook 2) and inducing apoptosis [
54]. These results contradict our finding of miR-142-3p as upregulated in BC. However, some authors support our results: Jusoh et al. [
55] found that this miRNA was upregulated in breast cancer patients as compared to the miRNA expression of healthy subjects. In addition, Naseri et al. [
56] consider that this miRNA is upregulated in many types of breast cancer resulting in the hyperproliferation of cancer cells in vitro and mammary glands in vivo.
In our results, hsa-miR-1307-3p was significantly upregulated in non-disease-free survival patients compared to controls. In the bibliography, Han et al. [
57] found that the upregulation of this miRNA correlates with a poor prognosis (lower survival rate) given that this miRNA seems to stimulate cell proliferation. Shimomura et al. [
58], comparing patients with breast cancer and non-breast cancer serums, conclude that a combination of five miRNA (miR-1246, miR-1307-3p, miR-4634, miR-6861-5p and miR-6875-5p) is able to detect breast cancer. Its possible mechanism has been proposed by Han et al. and Shimomura et al. who consider that the miR-1307-3p contributes to BC development and progression by targeting SMYD4 (SET and MYND domain containing 4) [
57,
58]
miR-331 was overexpressed in women in metastatic BC, not only when comparing with healthy controls, but also when comparing to women with non-invasive luminal-A BC [
59]. Likewise, miR-331 was overexpressed in BC with lymph node metastasis, higher TNM stage and poor prognosis [
60]. These publications are consistent with our results. In addition, Pane et al. [
61], using omic data integration and machine learning, anticipated that five miRNAs (mir-323a-3p, mir-323b-3p, mir-331-3p, mir-381-3p, and mir-1301-3p) could target in
EGFR (epidermal growth factor receptor) family to develop breast cancer in the patients (among other tumors).
In our results, miR-21-3p was significantly upregulated in non-disease-free survival patients compared to controls. This is consistent with Amirfallah et al. [
62], who found that its upregulation is associated with metastasis and a short disease-free survival. In addition, they found that the overexpression of this miRNA is associated with a poor prognosis. Ouyang et al. [
63] also support the results. They identified 5 upregulated miRNAs (miR-155-5p, miR-21-3p, miR-181a-5p, miR-181b-5p, and miR-183-5p) when comparing the miRNAs profile expression between triple negative breast cancer and normal breast tissues. Aure et al. [
64] also observed that the overexpression of three miRNAs associated with copy number gain (miR-21-3p, miR-148b-3p and miR-151a-5p) increases proliferation of breast cancer cell lines. Regarding its mechanism, some authors consider that miR-21 promotes cell proliferation and suppression of apoptosis by targeting SMAD7 (SMAD—Mothers Against decapentaplegic homolog- family member 7), PDCD4 (programmed cell death 4) and PTEN (phosphatase and tensin homolog) [
65], eventually leading to increased proliferation and invasiveness of some BC [
66].
As shown in both the background and the discussion sections, results on miRNA role in BC are far from homogeneous. While the role of some miRNAs (namely, miR-21, miR-101-3p, miR-186-3p, miR-331, miR-423-3p, miR-1307-3p) appears to be coherent across the literature, results on others (miR-29a-3p, miR139-5p, miR-1299 miR-142-3p) are contradictory and no clear conclusion could be reached. A similar statement could be made regarding combinations of miRNAs in models/signatures: miRNAs selected vary from model to model, making the results unreliable. For instance, only one out of five miRNAs included in the model by Shimomura et al. [
58] was selected in any of our models (miR-1307-3p); Kahraman et al. (2018) [
67] developed a model with seven miRNAs, but only one of them (miR-101-3p) was selected in ours; and Giussani et al. [
20] obtained signatures using five miRNAs, but none was selected in our analysis. By-the-way, signatures developed by Shimomura et al. [
58] Kahraman et al. [
67] and Giussani et al. [
20] do not share any miRNA with each other [
20,
58,
67].
Explanations for this result variability would include [
1] differences in statistical or lab procedures; in this regard, to select miRNAs on their crude statistical significance or using methods such as stepwise regression, which is known to inflate alpha error, could even involuntarily lead to p-hacking or cherry picking. [
2] Random variability -somehow associated with the frequently small sample sizes-; and [
3] true biological variability, which could be associated with diversity in the genetic background in patients studied in different countries or continents or to biological differences according to the intrinsic subtype of BC included in each study.
Our study has some limitations. Firstly, the selection of miRNAs for the validation phase was only partially based on the screening phase results, but also on previously published studies. When doing it, the authors chose miRNAs associated with BC in most recent studies (i.e., published in 2020 and 2021), but at the end the selection has some degree of subjectivity. In this way, the selection of miRNAs using their p-value in the screening phase could have led to missing some miRNAs that could have been associated with BC cases in the multivariate setting. Secondly, although beginning with a cohort of 1738 BC women, the final sample size was relatively small; this is especially true for the group of women with active disease in the follow-up, which was strongly limited out of the progressive improvement in diagnosing and treating BC. Thirdly, the discriminative power of our models is moderate as shown in areas under the ROC curve ranging 0.637 to 0.783. The study has also some strengths. Firstly, women included in the analysis were diagnosed in 10 different Spanish provinces and 23 Spanish hospitals, which guarantees some clinical variability. Secondly, our models were obtained using regression with penalization. This method (LASSO) allows for selecting parsimonious models (i.e., models with few regressors) while controlling the alpha error and avoiding the intervention of the researchers in selecting the finally included miRNAs. Moreover, LASSO is considered to outperform regression methods (e.g., stepwise) that select variables using the criticized p-value. Thirdly, we have a variety of cases (diagnosed by screening, disease-free in the follow-up and with active disease in the follow-up), which allows us to develop different models for diverse types of cases.