Extensive research efforts have focused on identifying novel biomarkers for challenging thyroid lesions, with the aim to improve diagnostic accuracy and define useful prognostic indicators to guide patient management [19]. The introduction of innovative molecular markers able to fill the existing accuracy gap of the combined ultrasound-FNAC approach would more precisely identify patients who are truly candidates for invasive therapeutic procedures. Follicular-patterned thyroid lesions represent a particularly challenging group of neoplasms, encompassing entities with distinct behaviors and different malignant potential. Among these, NIFTPs have been recently “downgraded” as indolent neoplasms [20], although their biology is still far from being fully understood. Moreover, the presence of “atypical” cases, often bearing RAS mutations, further complicates efforts to distinguish NIFTPs from the other follicular-patterned lesions [21].

MALDI-MSI has emerged as a promising tool for biomarker discovery, enabling the identification of the specific molecular signals associated with the analyzed lesions [22]. Given the considerable amount of information deriving from MALDI-MSI analysis, we built an LDA-supervised approach for feature selection and classification based on proteomic data obtained from a group of challenging thyroid lesions (i.e., FAs, NIFTPs, FVPTCs, and PTCs). Among these, FAs were significantly larger at diagnosis compared to the other neoplasms, reflecting their indolent course and the subsequent late diagnosis. As expected, NRAS mutations were significantly more frequent in follicular-patterned lesions (i.e., NIFTP and FVPTC) compared to classic PTCs. Our approach enabled the identification of specific molecular markers that correspond to follicular-patterned lesions, including NIFTPs and FVPTCs, while also providing valuable insights into the complex behavioral trajectories of NIFTPs with respect to morphologically similar neoplasms. This, in turn, could be useful for discerning whether such behavior more closely resembles that of the malignant class (PTC, FVPTC) as opposed to the benign category (FA). Specifically, the similar downregulation of DDX42 in NIFTPs and FVPTCs/PTCs compared to FAs may suggest that the former lesions share similar alterations in pre-mRNA processing. Splicing abnormalities could thus represent important mechanisms underlying the development and progression of these neoplasms, similar to what has been observed in other tumor types [23]. On the other hand, the comparable distribution of the Histone H4 signal between NIFTPs and FAs and its underexpression in FVPTCs/PTCs may indicate a higher burden of chromatin structural aberrations in the latter category of lesions, which is a hallmark of a malignant behavior. Indeed, downregulation of conventional histone proteins may reflect the increased expression of modified histone variants, which have been identified as key contributors to tumor progression and invasiveness in several malignancies, including breast and colorectal cancer [24].

Apart from their pathobiological significance, the identified signals may be valuable for differentiating between these challenging thyroid lesions in routine pathological practice. Indeed, previous work by our group has shown a significant downregulation of the DDX42 and Histone H4 proteins in RAS-mutant compared to RAS-wild type NIFTPs from a separate cohort of patients, encouraging further validation of these markers as ancillary tools to assist in the differential diagnosis of borderline thyroid neoplasms [25]. This would require assessing the expression of the identified markers through immunohistochemistry in a separate and, possibly, larger sample of FAs, PTCs, FVPTCs, and NIFTPs. Specifically, the commercial availability of antibodies for the immunoenzymatic detection of the DDX42 and Histone H4 proteins may warrant the recruitment of additional subjects to establish a validation set of sufficient size for statistically robust analyses.

Finally, the comparison of wtNRAS vs. mNRAS cases among follicular-patterned lesions (i.e., FVPTC and NIFTP) enabled the identification of molecular signals associated with NRAS Q61R positivity, potentially providing insights into the distinct pathogenic pathways through which these lesions develop. Among the differentially expressed signals, we were able to putatively identify the Complement C4-B and PDIA1 proteins, which we found to be overexpressed in wtNRAS compared to mNRAS cases. Altered expression of both of these proteins has already been implicated in the biology of thyroid neoplasms: PTC cells have been shown to escape host immunosurveillance through the synthesis of anti-idiotypic immunoglobulin G (IgG) molecules that bind to and neutralize host IgGs directed against tumor antigens. The subsequent formation of immune complexes leads to the activation of the classical complement pathway, explaining the observed co-localization of IgGs and complement proteins (including C4) in PTC tissues [26]. Similarly, members of the protein disulfide isomerase family, to which PDIA1 belongs, have been identified as pivotal factors in promoting the progression and dissemination of several cancer types, including PTC [27]. PDIs, which are redox proteins bearing thioredoxin-like domains [28], have pleiotropic roles in regulating thyroid hormone receptor signaling, misfolded protein degradation, and progression through the cell cycle [29]. Therefore, aberrant PDIA1 expression may contribute to tumor development and progression by disrupting intracellular protein homeostasis, as well as by altering the transcriptional responses elicited by the binding of thyroid hormones to their receptors. In addition, reduced expression of another member of the PDI family, namely PDIA3, has been found to correlate with lymph node metastasis and poor prognosis in PTC patients [30]. Overall, these observations support the hypothesis that mNRAS and wtNRAS lesions are intrinsically distinct entities that rely on different pathways for tumor growth and immune evasion. Whether these biological differences lead to a distinct clinical behavior of mNRAS compared to wtNRAS tumors, regardless of histology, is still uncertain, warranting prospective clinical studies with large patient cohorts.

The wealth of discriminative features identified in our study further confirms the remarkable potential of MALDI-MSI to streamline biomarker discovery in the field of thyroid pathology, particularly when machine learning algorithms are applied to support data analysis. However, the implementation of this tool into the routine diagnostic practice is currently hampered by its relatively high costs, limited availability, and technical constraints regarding sample preparation for proteomic analyses. Thus, future advancements to improve the accessibility and scalability of this methodology are highly anticipated. This may be achieved by defining a restricted panel of relevant markers and testing their expression through multiplex spatial imaging approaches, such as MALDI-HiPLEX-IHC, which allows to precisely map the distribution of proteins of interest in FFPE tissues [31]. As shown by our group using clear cell renal cell carcinoma models, the integration of MALDI-HiPLEX-IHC with untargeted spatial proteomics enables the spatial localization of antigens of interest as well as the characterization of the tumor immune microenvironment, requiring only a single FFPE tissue section [32]. Based on these promising results, we are planning to extend the applicability of our workflow also to challenging thyroid lesions.

Although promising, our preliminary application of the LDA model to distinguish between histologically similar thyroid lesions has some limitations: due to the small number of subjects enrolled in our study, we decided to analyze the full patient cohort without splitting our dataset, as this would have resulted in underpowered training folds. For similar reasons, our proteomic analyses were performed considering I-FVPTCs and iE-FVPTCs as a single neoplastic category (i.e., FVPTC). Thus, the full generalizability of our model still requires further validation with external cohorts.

Moreover, the main focus of our study was to investigate putative biomarkers to facilitate the differential diagnostic process of borderline thyroid lesions. Whether our methodology can also identify markers correlated with relevant histopathological features (e.g., the presence of capsular and/or lymphovascular invasion) remains to be explored. Nevertheless, since the identified proteins may aid in the distinction between NIFTPs and histologically similar invasive neoplasms, they might serve as indirect indicators of tumor invasiveness irrespective of histology. Again, these preliminary findings warrant further validation studies with larger patient cohorts.

Finally, our analyses were carried out on a relatively limited set of neoplasms, so additional efforts are needed to extend the applicability of our approach to other challenging entities.