Circulating miRNAs have attracted much interest in cancer diagnosis and prognosis as they are small and very stable in circulation and might have essential functions. For these reasons, they are good candidates as molecular biomarkers capable of selecting patients at higher risk of recurrence. However, very conflicting results regarding their role have been reported, leaving doubts about their real value as prognostic markers. The heterogeneous results are mainly due to different aspects: the panel of miRNAs analyzed (small panel of miRNAs or miRNome analysis), the methodology used for miRNA evaluation and normalization, the statistical approach applied, the study endpoint, the specimen used for the analysis (plasma or serum) and methods of collection. Moreover, “circulating miRNAs” can be different whether we consider EV-derived miRNAs or the overall circulating miRNAs content, as it is well established that EV-derived miRNAs could exert more specific functions in the context of lung cancer, with respect to those released in the circulating compartment [14, 15]. Furthermore, despite numerous studies that have analyzed miRNAs as potential prognostic factors, it is not clear the added value of miRNAs with respect to the established prognostic factors used in the clinic, mainly the stage of the disease.

To address some of these aspects, we performed a prospective multicentre study to clarify the role of circulating miRNAs as prognostic biomarkers in ES-NSCLC, focusing on the added value of the miRNAs with respect to the known clinicopathological prognostic parameters.

From the model built on the miRNA data, we found that 3 CF-miRNAs and 21 EV-miRNAs were associated with DFS without considering the integration of known clinical prognostic factors. Most of them exhibited a negative regression coefficient, meaning that their higher expression values are associated with a better prognosis, pointing out a tumor-suppressor function of the specific miRNAs.

Concerning the CF-miRNAs, miR-135a-5p, and miR-877-3p were overexpressed in patients with a lower risk of relapse or death, which aligns with their reported biological function. It was shown that miR-877 overexpression repressed NSCLC cell growth by targeting tartrate-resistant acid phosphatase (TRAP), also known as acid phosphatase 5 (ACP5), and inhibiting the PI3K/AKT pathway [16]. Concerning miR-135a-5p, it has been shown in several studies that its expression decreased in different solid tumors such as glioma [17], gallbladder [18], colorectal cancer [19], and also lung cancer [20]. Conversely, we observed that CF miR-29c-3p upregulation was associated with worse DFS. This observation goes along with its potential role as a tumor promoter, recently shown in ovarian cancer models [21].

On the other hand, we found 21 EV-derived miRNAs significantly associated with patient prognosis. Of these, 12 (miR-127-3p, miR-1277-3p, miR-136-3p, miR-181a-2-3p, miR-18a-5p, miR-323b-3p, miR-3615, miR-532-3p, miR-589-5p, miR-628-5p, miR-6852-5p and miR-99b-5p) might act as tumor suppressors, with an association between high levels and a better prognosis. The demonstrated biological role of most of these miRNAs aligns with our results. In particular, it has been shown that miR-181a-2-3p is downregulated in NSCLC tissue compared with control samples and that its low expression in tumor tissue and plasma is associated with longer progression-free survival [22]. Interestingly, plasma EV-derived miR-3615, combined in a model with other 3 miRNAs, was demonstrated to be able to distinguish early-stage lung adenocarcinoma patients with respect to healthy donors, suggesting a potential role as a noninvasive biomarker for the early detection [23]. A tumor suppressor activity was also demonstrated for miR-532-3p that, by targeting FOXP3, can inhibit the growth of NSCLC cell models [24] and for miR-589-5p, which, since it usually inhibits Histone deacetylases 5 (HDAC5), when downregulated leads to an overexpression of HDAC5 and therefore to an increase in cancer cell proliferation [25]. Concerning miR-18a-5p, its complex role in cancer has been demonstrated, as it can act both as an oncogene and a suppressor [26]. Among the 9 EV-derived miRNAs whose overexpression resulted associated with shorter DFS (miR-182-5p, miR-192-5p, miR-32-5p, miR-328-3p, miR-339-3p, miR-361-3p, miR-370-3p, miR-5187-5p, and miR-9-5p), suggesting their role as oncomiR, their demonstrated biological role in the literature is in part in line and in part in contrast with our results. In particular, a role as oncomiR has been confirmed for miR-182-5p [27], miR-328-3p [28], miR-339-3p [29], miR-361-3p [30], and miR-9-5p [31]. Conversely, contrasting results are reported regarding miR-192-5p [32], miR-370-3p [33], and miR-32-5p [34], as from literature results, they seem to have a role as tumor suppressor miRNAs.

Cell-free miRNAs can be released and uptaken by cells through vesicle trafficking and protein carrier mechanisms, and they are able to function as gene expression regulators in cell to-cell communication mechanisms under normal and pathological conditions, such as cancer [35]. However, we have to consider that miRNAs are released into the bloodstream from different cell types, not only from tumor cells [36], and this still unclear aspect leaves some open questions regarding the real role that these miRNAs found in the blood circulation may have. Moreover, miRNAs can target several different genes and as a consequence can affect different pathways. Taking this into consideration, a pathway enrichment analysis was performed to reach a global overview of the different pathways on which the significant miRNAs are involved.

Interestingly, enrichment of TGF-beta/SMAD and NOTCH pathways was observed, which was even more evident when considering only EV-derived miRNAs. In particular, 5 miRNAs (miR-125a-5p, miR-370-3p, miR-628-5p, miR-381-3p, miR323a-3p), which evidence of interaction with the TGF-beta pathway is present in the literature [37,38,39,40,41], were found to have higher expression levels in patients with a longer DFS in our cohort.

TGF-beta/SMAD and NOTCH are well-known prognostic pathways in NSCLC [42,43,44,45]. In particular, a specific association between TGF-beta expression and risk of relapse in ES-NSCLC has been shown [46]. With regard to NOTCH, a previous study showed that specific polymorphisms of the gene are associated with survival rates in ES-NSCLC [47]. Moreover, an in vitro study demonstrated that the NOTCH signaling significantly affects the growth and the malignant phenotype of both colorectal and lung models [48]. Interestingly, an enrichment of the PI3K pathway was also observed in EV-derived miRNAs. The PI3K pathway involvement in the prognostic risk determination of ES-NSCLC was already shown in previous studies [49, 50], in accordance with our results. Overall, we found an evident enrichment of TGF-beta/SMAD, NOTCH, and PI3K pathways in EV-derived miRNAs.

This reinforce the already demonstrated role of EVs as components with specific functional activities in the regulation of cell growth, which consequently can have important prognostic and predictive roles in response to therapies [51]. It has been shown that both tumor and immune cells can release specific EVs containing components, mainly miRNAs, with specific functions able to regulate cancer-specific processes such as epithelial-mesenchymal transition [52, 53], neovascularization [54, 55], anti-tumor immune cell function [56, 57]. Hence, we can conclude that EV-derived miRNAs seem to have more specific functions with respect to CF-miRNAs, which could be derived by EV itself but also from other, more non-specific sources, such as necrotic or apoptotic cells.

To evaluate the added predictive value of the miRNAs to basic prognostic factors, statistical models combining these and disease stage, age, and sex were specified and then compared with a model including only the demographic and clinical factors. In the combined models (one for CF- and one for EV-miRNAs), two CF-miRNAs (miR-29c-3p and miR-877-3p) and five EV-miRNAs (miR-181a-2-3p, miR-182-5p, miR-192-5p, miR-532-3p and miR-589-5p) remained associated with DFS and we did not observe significantly higher predictive performance compared to the models including the clinical factors alone. However, when we derived the combined model considering the data from both types of miRNAs, a significant increase in the prognostic accuracy was found. In particular, 3 CF-miRNAs (miR-135a-5p, miR-29c-3p, miR-877-3p) and one EV-miRNA (miR-192-5p) gave a substantial contribution in addition to the clinical factors.

Our study has several strengths and limitations. Among the primary strengths are i) the prospective collection of biological specimens and clinical data, ii) the analysis of the miRNome and differentiation of miR origins, iii) Moreover, despite the relatively small sample size and the lack of an external validation cohort, we were still able to preliminarily evaluate the performance of the fitted models, as well as, through the approach proposed by Simon et al. [12], to investigate the added predictive value of the miRNAs as compared to basic non-biological prognostic factors, in particular the stage of disease, something that is not often reported in the published studies.

As mentioned above, one limitation of this study is the modest sample size and the limited follow-up time (the median was about 26 months), especially given the high proportion of stage I cancers in our cohort (64.3%). Both these factors limited the possibility of stratified analyses and their statistical power, leaving several questions unanswered, for example, whether different miRNAs are involved in the prognosis prediction by disease characteristics (e.g., stage and histotype) or if their prognostic effect may vary within strata. However, enrollment is ongoing in another funded study on ES-NSCLC carried out by some of the RESTING centers, which will permit us to validate our data. Another limitation is relating to the uncertain source of release of the miRNAs found in circulation, which leaves some open questions regarding their functional role in cancer development. Moreover, we are conscious that the different methodologies available for EV isolation could maintain a portion of small lipoproteins, that could interfere with the subsequent analyses. However, we used an EV purification method that relies on the specific binding of EV surface proteins to negatively charged silicon membranes, exploiting differences in pH and isoelectric points compared to other proteins like histones and lipoproteins in biological fluids. This approach allows for the selective adsorption and subsequent elution of EV while effectively excluding most lipoprotein contaminants, as demonstrated in existing literature [58]. Given that there are no gold standard procedures for EV isolation, some open questions remain about the optimal methodology to be used. However, our approach seems to be reproducible and robust with some potentiality to be included in the clinical practice in the future.

In summary, our study highlights potential miRNAs able to predict the risk of relapse after surgery in patients with ES-NSCLC. When considered separately, CF- and EV-miRNAs could not significantly improve the prognostic performance of the model with known clinical prognostic factors. However, when considering them together, a significant improvement was observed. Thus, the identified circulating miRNAs could represent a non-invasive approach that could permit clinicians to have further information to decide the most appropriate patient’s management.