1. Introduction
Thyroid tumors are the most common endocrine lesions worldwide, and their incidence has continuously increased over the last three decades [
1,
2]. Different clinical examinations, including ultrasound, CT, MRI and fine-needle aspiration cytology, have been utilized for the preoperative diagnosis of thyroid tumors. Of these examinations, pathology remains the gold standard for clinical diagnosis, and thyroid tumors arising from follicular cells can generally be categorized into malignant and benign tumors according to their pathology. However, it is occasionally extremely challenging to differentiate thyroid follicular adenoma (FTA) from follicular thyroid carcinoma (FTC). In these cases, pathomorphology is used to describe the characteristics of malignant tumors, named “atypia,” which do not appear in benign tumors. However, this morphological standard does not apply to thyroid follicular neoplasms. FTA cells can exhibit atypia, whereas FTC cells may show minimal or absent atypia. Moreover, distant metastasis of FTC sometimes occurs, such as in bones or lungs, and the morphology of metastatic tumors resembles that of FTA. Currently, FTC diagnosis is defined by capsular and/or vascular invasion; however, this standard is difficult to achieve in frozen sections or even surgically excised lesions.
As a result, in the fifth version of the World Health Organization (WHO) classification, follicular borderline lesions, which include non-invasive follicular thyroid neoplasms with papillary-like nuclear features (NIFTP), and thyroid tumors of uncertain malignant potential (UMP) have been proposed to classify follicular thyroid neoplasms with questionable capsular or vascular invasion. The prognosis of benign FTA is excellent and only requires surgical resection of the tumor. In contrast, the treatment of malignant FTC requires bilateral thyroidectomy and lymph node dissection, combination therapy with I131, and the administration of thyroxine tablets for life. Surgeons hope that a pathological diagnosis, especially an intraoperative frozen diagnosis, can provide accurate and timely diagnostic information to guide clinical operations and postoperative treatment, but discriminating between FTA and FTC is always the greatest clinical challenge for pathologists. Therefore, the proper classification of thyroid follicular diseases has become an urgent clinical need.
Mass spectrometry imaging (MSI) is an efficient method applied to tumor research, allowing the identification of different classes of molecular species in tumor surgical margin studies, tumor qualitative diagnoses, the prediction of cancer lymph node metastasis, and prognostic assessments of the neoplasm [
3,
4,
5,
6,
7]. The major advantage of MSI is the ability to combine molecular information and histomorphological images. The obtained molecules can have high tissue specificity and are closely related to clinical information, presenting clear information on molecular organization [
8,
9]. Moreover, MSI technology can provide effective auxiliary diagnostic methods for pathological histology, meaning that medical experts will not only define tissue types based on the morphological structure of the tissue, but can also include their molecular components. Metabolomic research is closely related to the occurrence and development of tumors, and changes in the expression of metabolite molecules and related metabolic pathways are important characteristics of tumor cells. Moreover, metabolomic research based on MSI is a high-throughput approach that can identify molecular information closely related to the occurrence and development of tumors.
The airflow-assisted desorption electrospray ionization (AFAIDESI)-MSI technique under ambient conditions is a high-resolution ambient molecular imaging method. Here, we employ AFAIDESI-MSI metabolomic analysis to define novel diagnostic pathways and metabolites for thyroid follicular tumors, ultimately serving as potential markers of malignant tumors with uncertain potential. Our prospective study aimed to develop and validate an AFAIDESI-MSI metabolomic analysis to define novel diagnostic pathways and metabolites to discriminate between FTA and FTC and improve the diagnosis of indeterminate cases.
3. Discussion
“Gray zone” thyroid follicular tumors, whether benign or malignant, are difficult to diagnose. Therefore, an accurate evaluation of resected benign or malignant thyroid follicular tumors after surgery is critical to avoid secondary surgery and over-diagnoses. This study is the first IMS analysis performed on UMP follicular tumors. Traditional cytologic specimens cannot be used to distinguish benign from malignant follicular neoplasms because the diagnosis of malignancy in follicular neoplasms requires tumor cells to present capsular or vascular invasion. However, both FTC and FTA are follicular thyroid lesions that cause challenging pathological issues. To date, pathologists are virtually unable to draw a clear distinction between FTC and FTA based on a frozen section examination alone. They have some overlapping morphological features; therefore, they represent a diagnostic dilemma for practicing pathologists. A well-known reliable pathological criterion for the malignant diagnosis of FTC is the presence of tumor cells invading the tumor capsule or blood vessels. However, the limited sample collection during the frozen diagnosis process makes evaluating all samples’ capsules and blood vessels difficult. Although no envelope or vascular infiltration was found in the paraffin specimen after a comprehensive evaluation, if substantial structural and morphological heterogeneity was observed, pathologists could not completely rule out the possibility of malignant tumor biology or provide clinically effective diagnosis and treatment guidance. Therefore, the recent WHO classification has proposed certain borderline lesions, for example, NIFTPs and UMP thyroid tumors. Nevertheless, proper classification of thyroid follicular diseases could be improved with better evaluation of clinical prognosis.
As a result, more reliable diagnostic markers are required. To date, there have been an increasing number of markers in the process of continuous evaluation of their diagnostic utility. For example, Kaliszewski et al. [
10] proposed that serum thyroid-stimulating hormone levels are considerably higher in patients with atypia and follicular lesions of undetermined significance. Chuang et al. [
11] suggested that IHC marker panels, including CK19, CD56, galectin-3, and some other antibody markers, can differentiate thyroid follicular neoplasms. Nevertheless, none of these biomarkers are routinely used because they lack validity.
MSI has been proven to be useful for molecular diagnoses and relies on professional imaging software. AFAIDESI-MSI is a high-resolution ambient molecular imaging method, and metabolomics research based on MSI can detect metabolic abnormalities in tumors with high throughput. In this study, MSI data combined with OPLS-DA analyses, as a powerful approach, was not only used to discriminate thyroid tumors from adjacent normal thyroid tissues, but also for tumor-specific discrimination. We can effectively mine spatial metabolite molecular information in human tumor tissue samples using high-resolution recognition and data processing methods to analyze the distribution of tumor molecules, thereby providing a better understanding of the metabolic heterogeneity of tumor tissue. The expression of phosphoric acid species and FAs was higher in the FTA group, and targeted IHC staining of the potential metabolic enzymes—including FASN and iPLAs—was performed on adjacent sections to validate our discovery. The internal validation of the classification model showed very good performance (95% CI: 62.7–84.5%, AUC = 73.6%, sensitivity = 82.1%, specificity = 60.6%).
FAs are an essential component of cell membranes, playing a major role in maintaining the basic morphology of cells and their normal physiological functions. Multiple tumors and their early lesions undergo endogenous FA biosynthesis independent of extracellular lipid levels [
12,
13,
14]. In this study, elevated FA levels in the tumor tissue were further confirmed by our MSI results.
Figure 1 shows that the FA ion intensities demonstrated an increasing trend from the normal thyroid follicular epithelium to the pathological thyroid tumor epithelia. Interestingly, the ion intensity of FAs in FTA was higher than that in FTC (
p < 0.001,
Figure 2C), indicating that FAs may predict potential diagnoses in thyroid follicular tissues. Tumor cells rapidly synthesize FAs to meet energy consumption [
15]. A multifunctional homodimeric FASN catalyzes the biosynthesis of endogenously synthesized FAs [
16,
17,
18]. Therefore, FASN is the crucial metabolic regulatory enzyme responsible for the terminal catalytic step in FA synthesis. The adjacent IHC stain showed that FASN was primarily expressed in both FTA and FTC, and that its expression was higher in FTA. Some studies have indicated that early upregulation of FASN in precursor lesions may represent an obligatory metabolic acquisition in response to the microenvironment of preinvasive lesions, which continues to occur in malignant stages. The functional and temporal linkage between the glycolytic switch and FASN-related lipogenic phenotype may represent coevolved essential components of the malignant phenotype, signifying the hallmarks of invasive cancers [
19].
Phospholipids are crucial components of the cell membrane. PC is a phospholipid that comprises the main components of cell membranes. The continuous de novo synthesis PC synthesis pathway provides tumor cells with components of synthetic cell membrane, lipid signal transduction, and energy for rapid growth of tumor cells [
20]. The MS images indicated that phospholipids were significantly upregulated in the FTA tumor region compared with the FTC epithelium (
p < 0.001,
Figure 2C). Over the years, the metabolism of phospholipids and their metabolic products in mediating cell function have received considerable attention. Emerging studies have focused on iPLAs enzymes that mediate growth and signaling in numerous cell types. They may offer unique targets for treating various pathologies whose etiology involves the generation of phospholipid signals [
21]. Some studies have demonstrated that iPLAs mediate cell growth by participating in signal transduction pathways, including epidermal growth factor receptors, mitogen-activated protein kinases, tumor suppressor protein p53, and cell cycle regulator p21 [
22,
23,
24,
25,
26]. We speculate that the upregulated expression of iPLAs in thyroid follicular tumors may be related to elevated phospholipid biosynthesis. Similarly, IHC staining displayed that iPLAs expression was considerably higher in FTA.
Subsequently, we focused on 19 patients with indeterminate pathological classifications. A prediction of benign diagnosis was obtained from 12 of the 19 patients that agreed with the follow-up, and 5 with a suspicious diagnosis were malignant. Due to the inconsistent expression intensity of FASN and iPLAs, two cases could not be classified according to the metabolite enzyme marker results. The two cases of mass spectrometry dataset predictive results were in the category of FTA. There are some explanations for this inconsistent expression, such as the heterogeneity of tumor cells or the instability that IHC antibodies may cause. As a result, this implementation of combined AFAIDESI-MSI with metabolite enzyme markers may positively affect the diagnosis classification for the “indeterminate” nodules.
4. Materials and Methods
4.1. Sample Collection
All postoperative thyroid tumor samples were collected at the Peking Union Medical College Hospital between 2014 and 2021. The Ethical Review Committee of the Peking Union Medical College Hospital approved the study protocols. All patients consented to participate in this study. Notably, none of the patients with thyroid cancer were administered preoperative treatment. All tumor tissue samples were stored at −80 °C until sectioning. Thereafter, 8-μm-thick tissue sections were used for the AFAIDESI-MSI experiments and 5-µm-thick tissue sections from adjacent slices were subjected to hematoxylin and eosin (H&E) staining to confirm the pathological diagnosis. In total, five sections were produced from each tumor tissue, four of which were used for mass spectrometry scanning using positive and negative ion modes, and one was stained with H&E before being placed on a slide box for long-term storage. Before the experiments, the slides were thawed at room temperature and then dried in a vacuum desiccator for approximately 1 h.
4.2. AFAIDESI-MSI Analysis
The AFAIDESI-MS imaging analysis data were acquired in both positive and negative ion modes, changing one parameter value in a step-by-step manner and keeping the other parameters unchanged. The strength value of the mass spectrometry data obtained under the parameter setting was investigated, and the key parameters involved in the scanning process, including a spray solvent composition set at (±) 7000 V, spray solvent flow rate of 5 μL/min, and extraction flow rate of 45 L/min, were determined. The tissue surface and background portion of the tissue were used with a computer-controlled platform in the x direction at 200 μm until the entire tissue sample was completely scanned. The mass spectra were recorded in the full scan range of m/z (mass-to-charge ratio) 100–1000.
4.3. Histopathology Analysis
All thyroid tumor specimens were fixed in 10% buffered neutral formalin and embedded in paraffin. Samples were stained with hematoxylin for 2–3 min and washed with water for 5–10 s. Thereafter, they were subjected to 1% hydrochloric acid alcohol differentiation for 1–3 s and washed with water for 1 min. Next, they were subjected to reverse blue staining with 0.2% ammonia water for 3 s and washed with water for 1 min. Finally, staining with 1% eosin solution was performed for 2 min, with subsequent washing with water for 10 s. H&E-stained slides were used for re-diagnosis by two professional pathology professors, and they were classified according to the fifth version of the WHO criteria. The two pathologists determined the tumor content, and there was no necrosis or degeneration in any tissues.
4.4. Immunohistological Staining
The expression of metabolic enzymes in thyroid tumor tissues was assessed via IHC staining using specific antibodies. Successive frozen tissue sections adjacent to the section analyzed using AFAIDESI-MSI were warmed to room temperature for 20 min. Sections were fixed in paraformaldehyde for 10 min. After washing in phosphate-buffered saline, the sections were immersed in 0.25% Triton X-100 for 15 min to make the tissue permeable, and then blocked with 1% bovine serum albumin for 30 min at room temperature. Furthermore, the sections were incubated with antibodies against iPLAs (Proteintech; Chicago, IL, USA; 22030-1-AP; 1:200) and FASN (Abcam; Toronto, ON, Canada, C; ab128870; 1:300) at 4 °C overnight, followed by rewarming at room temperature for 20 min. A PV-9000 two-step IHC kit was used according to the manufacturer’s instructions, and a DAB kit was used to detect antigen–antibody binding (Zhongshan Goldenbridge Biotechnology Ltd. Co., Beijing, China). The slides were counterstained with hematoxylin, dehydrated, mounted, and covered. FASN and iPLAs staining revealed cytoplasmic protein expression. The intensities of FASN and iPLAs were graded semi-quantitatively based on a scale comprising 0 (no staining), low expression (weak-to-moderate staining), and high expression (strong staining).
4.5. Data Processing and Statistical Analysis
The File converter function in Xcalibur was used to convert the series of raw data in .raw format, obtained via mass spectrometry scanning, into. cdf format. Thereafter, the actual length and width of the slice were inputted into the data import window of MassImager, and the raw data were imported in .cdf format in batches at once to generate the import sequence. The mass spectrometry peak detection slope was set to 100, the minimum peak intensity was set to 0, and the intensity integration method was set to total intensity. The calculation method for the mass-to-charge ratio was the weighted mass-to-charge ratio, and the in situ image of tissue slices covering the entire range of the mass-to-charge ratio (100–1000) was ultimately obtained. The endogenous metabolites that could represent the entire tissue profile were selected from the collected AFAIDESI-MSI mass spectrum profile, inputted into the ion channel corresponding to the mass–charge ratio in MassImager, and the distribution of this ion in tissue slices was obtained. Any background area outside of the tissue slice was obtained, the mass-to-charge ratio tolerance was set to 0.005, and background ions were deducted to obtain better imaging results. The H&E staining map of the tissue slices was imported; scaling, rotation, displacement, and other operations were performed in sequence; and they were then overlayed with representative ion imaging maps to accurately select the region of interest and output the average mass spectrometry map of that region for subsequent multivariate statistical analysis. Data from the metastatic and non-metastatic specimens were merged using Markview software 1.2.1 (AB Sciex, Framingham, MA, USA), and then, multivariate statistical analysis was conducted using PCA and OPLS-DA model recognition methods with SIMCA-P software 14.0.1 (Umetrics AB, Umea, Sweden). Data control analysis was conducted between the metastatic and non-metastatic groups to identify molecular information related to tumor metastasis. The extracted statistically significant mass-to-charge ratio (
m/
z) was used to image each sample, mass spectrometry imaging maps of each sample were evaluated, and spatial distribution characteristics and patterns of each molecule were discovered. After classification, the variable importance was assessed using an independent
t-test (Microsoft Office Excel 2010, Los Angeles, CA, USA). The statistical significance level was set to
p < 0.01. Lipid species were compared with LIPID MAPS (
http://www.lipidmaps.org/, accessed on 28 January 2023), Massbank (
http://www.massbank.jp/, accessed on 28 January 2023), and HMDB (
http://hmdb.ca/, accessed on 28 January 2023) databases. ROC curve analysis was performed to determine the roles of accuracy and specificity in the predictability of potential molecules. The AUC ranged from 0.5 to 1.0. Complete separation of the values based on the indicator was performed using scores above 0.75 or 75%