Introduction

Thyroid cancer is the most common endocrine tumor, with the ninth-highest cancer incidence worldwide and the third-most common cancer in women in China1,2. Thyroid fine needle aspiration (FNA) biopsy is the primary method to evaluate the malignancy of thyroid nodules, but approximately one-third of indeterminate thyroid nodules still undergo molecular testing for diagnosis3. Although the prognosis of papillary thyroid cancer (PTC) is generally favorable, up to 20% of patients may have a recurrence risk4,5. Recent data suggest that specific molecular profiles, such as the co-existence of BRAF with other oncogenic mutations (such as TERT and PIK3CA), may serve as less favorable outcomes of thyroid cancer6,7. The American Thyroid Association (ATA) risk stratification system categorizes the co-existence of BRAF with TERT mutations into high levels8. Therefore, the clinical precision management of thyroid cancer requires comprehensive analyses utilizing multiple mutation profiling.

Current clinical approaches for detecting molecular profiles primarily include real-time quantitative polymerase chain reaction (qRT-PCR), droplet digital PCR (ddPCR), Sanger sequencing, and next-generation sequencing (NGS)9,10,11. These technologies are widely used in clinical practice but have shortcomings. For instance, qRT-PCR and ddPCR methods are rapid and cost-effective tools that struggle to detect multiple mutations effectively. In addition, Sanger sequencing has an LOD of 10%, rendering it inadequate for identifying low-abundance mutations12. NGS often provides broad coverage of genes or hotspot of up to several hundred genes with improved sensitivity, but limitations include the high cost and long turnaround time of approximately ten days, which may be incompatible with the demands of timely clinical decision-making13. Therefore, an adaptable, rapid, and multiplex detection method is highly demanded for screening somatic variants for thyroid cancer.

The MassARRAY platform is highly automated, cost-effective, and suitable for high-throughput analysis of somatic oncogenic mutations14. The ddNTPs corresponding to the mutant and/or wild-type allele are extended during single base extension, generating allele-specific extension products15. Therefore, ddNTPs are crucial component in the single base extension process. Although the MALDI-TOF MS method has significant advantages in detecting multiple mutations, the method suffers from drawbacks, such as low sensitivity and high noise signal for somatic mutations. When using the Single Base Extension (SBE)-MALDI-TOF method, many wild-type products may be generated, which can adversely impact the detection of mutant products. Recently, this method has been successfully used to detect somatic mutations16,17. However, the sensitivity of the assay is only about 5%, which is insufficient for a low number of tumor cells. Several studies have focused on improving sensitivity, such as UltraSEEK™, which achieves a detection limit as low as 0.1%. However, its operational complexity and reliance on magnetic bead capture make the process time-consuming18,19,20. modified SBE enables efficient discrimination of low-abundance mutant alleles through innovative chemical modifications, optimized enzymatic reaction designs, and signal amplification strategies. Nonetheless, it may promote mismatch incorporation, potentially compromising assay fidelity21. There are also reports indicating that enrichment techniques can be integrated with mass spectrometry (MS), as exemplified by COLD-PCR. However, the inherent complexity of COLD-PCR, such as the requirement for multiple PCR cycles, can render the combined workflow more cumbersome and require greater technical expertise22.

Herein, we present a new multiplex variant detection technology that integrates nucleotide enrichment (NE)-assisted specific identification of variant alleles with MALDI-TOF MS, termed NE-MS, enabling the simultaneous identification of 26 thyroid cancer somatic variants. The NE-MS can dramatically enhance sensitivity and applicability in two ways: (ⅰ) In the single base extension reaction, the wild-type matched mass-modified ddNTPs are removed from the ddNTP mix, and only the mutant allele can be preferentially extended, which significantly increases the detection sensitivity of the variant allele; (ⅱ) The results are analyzed automatically using a Z-score based on median absolute deviation. Next, this strategy is applied to screen the thyroid cancer molecular profiling in FNA and formalin-fixed paraffin-embedded (FFPE) tissues. With its advantages of high sensitivity and high throughput, the NE-MS assay exhibited a robust performance when evaluated for diagnostic performance and clinical prognosis of thyroid cancer.

Results

Design of the NE-MS assay

The NE-MS assay employs PCR to amplify target regions of interest. After removal of unincorporated dNTPs, a sequence-specific primer is extended by a single nucleotide using ddNTPs. The ddNTP mixture for single-base extension is formulated by excluding the mass-modified ddNTP corresponding to the wild-type nucleotide (e.g., wild-type T), thus retaining only ddNTPs complementary to potential mutant bases (A, C, G). This customized mixture is termed NE ddNTPs. This selective extension ensures that only mutant alleles are amplified, thereby dramatically improving the detection sensitivity of low-abundance mutations. The schematic diagram is illustrated in Fig. 1.

Fig. 1
Fig. 1
Full size image

Schematic illustration of the NE-MS and regular MS assays.The signal intensity (X) of each assay was normalized to the linear fit of the internal controls. Z-scores were calculated for each assay using the median (M) and median absolute deviation (MAD) values pre-established from wild-type reference samples. Regular MS assays refer to the current integrated Single Base Extension (SBE)-MALDI-TOF technology used for detecting tumor somatic mutations. MS, Mass Spectrometry.

The regular MS procedure indicates that the concentration of each ddNTP is equal, referred to as regular ddNTPs. When single-base extension occurs, both mutant and wild-type alleles are extended and produce different molecular weights. The abundance of wild-type extension products can significantly interfere with the detection of mutant products. It is challenging to find low-abundance mutations when judging the results from the original mass spectrogram. In this study, the mutant allele is much more extended using NE ddNTPs in the single-base extension reaction. In addition, without the interference of wild-type allele peak signal, the sensitivity of mutant allele detection is significantly increased. Given that each target exhibits a characteristic signal distribution and background noise profile, it is challenging to ascertain the variants based on the signal peaks. To address this, we implemented an automated Z-score (median absolute deviation-based) calculation to define cutoff values from raw data.

To confirm successful sample amplification, we designed amplicon control assays to detect the wild-type sequence of each amplicon in the panel, serving as an internal quality control for both the sample and experimental workflow. After the post-PCR step, a set of normalization controls is added to each reaction for data normalization and mutation calling (Supplementary Fig. S1).

Feasibility of the NE-MS assay

The composition of ddNTPs plays a crucial role in the NE-MS assay in specifically recognizing the somatic variants. We performed a comparative analysis of two different ddNTPs (NE ddNTPs and regular ddNTPs) using the reference standards of NRAS Q61K and PIK3CA H1047R with a 5% variant allele frequency (VAF). As shown in Fig. 2a and b, the signal peak of NRAS Q61K and PIK3CA H1047R mutations could be correctly distinguished from the original result using NE ddNTPs. In contrast, mutant signals were difficult to identify in the regular ddNTP group, as most extension products corresponded to wild-type alleles. To gain deeper insights into the feasibility of the NE-MS assay, After Z-score conversion, the signal intensity in the NE ddNTP group was significantly higher than in the regular ddNTP group; notably, the relative signal for PIK3CA H1047R was 16.4-fold greater (Fig. 2c).

Fig. 2
Fig. 2
Full size image

Comparative analysis of the NE-MS assay performance during single-base extension reactions using NE ddNTPs versus regular ddNTPs. Mass spectra of reference standards of (NRAS Q61K, (a) PIK3CA H1047R, (b) with a 5% VAF. (c) Z-score for NRAS Q61K and PIK3CA H1047R assays, and the N/R ratio (right axis) of NE ddNTPs and regular ddNTPs based on the Z-score. Error bars represent the standard deviation from three independent replicates. N/R, NE ddNTPs / regular ddNTPs.

Performance analysis of the NE-MS assay for somatic mutations detection

To further verify the analytical performance of NE-MS, we evaluated its limit of detection (LOD), we serially diluted reference gDNA containing BRAF V600E, NRAS Q61K, TERT C228T, and PIK3CA H1047R mutations into wild-type (WT) gDNA, resulting in variant allele frequencies (VAFs) of 5%, 2.5%, 1.25%, 0.63%, 0.32%, and 0.16%. To directly evaluate the capacity of the NE-MS assay to detect somatic mutation, we compared the regular MS method with the NE-MS method. Raw mass spectra were presented in Fig. 3a and d. A continuous decrease in peak intensity was observed when the mutant allele frequency decreased from 5% to 0.16%. As depicted in Fig. 3e and h, using the NE-MS method, the lowest observable signals for BRAF V600E and TERT C228T mutations were 0.63%. Meanwhile, the NRAS Q61K and PIK3CA H1047R mutations had detectable signals at 0.32%. In contrast, the regular MS method showed the lowest observable signals for the BRAF V600E, NRAS Q61K, and TERT C228T mutations at 1.25%, 2.5%, and 5%, respectively. Notably, the PIK3CA H1047R mutation showed almost no significant signal peaks across any of these gradients.

Fig. 3
Fig. 3
Full size image

Comparison of the analytical sensitivity for somatic mutations detection using NE-MS and regular MS assays. The raw mass spectrograms (a–d), mutant allele peak average intensity (e–h), and Z-score distribution (i–l) of BRAF V600E, NRAS Q61K, TERT C228T, and PIK3CA H1047R mutations with the NE-MS and regular MS methods. The horizontal red dashed lines represent Z-score thresholds of 7.

Our detection method, which utilized an optimized Z-score threshold of 7, enabled sensitive detection of as low as 0.32% (Fig. 3i and l). Specifically, compared to the regular MS method, for the BRAF V600E mutation, NE-MS achieved a 4-fold lower LOD (0.63%) compared to the regular MS method (2.5%). For the NRAS Q61K mutation, the LOD was reduced from from greater than 2.5% to 0.32%, representing an 8-fold improvement. Similarly, the LOD for the TERT C228T mutation was reduced from 5% to 0.63%, an 8-fold improvement. Most significantly, for the PIK3CA H1047R mutation, the LOD was determined to be 1.25% level. However, the regular method failed to detect the mutation at VAF as low as 5%. This decreased LOD indicates that the NE-MS assay exhibits significantly improved sensitivity and reliable, which is crucial for detecting somatic mutations.

The assay was initially developed to enable multiplex detection of somatic mutations in thyroid cancer. The qRT-PCR method is typically used to detect the BRAF V600E mutation in thyroid FNA samples to aid diagnosis, and the NGS method evaluates multiplex somatic mutations for prognosis assessment in surgical FFPE samples. To further assess the reliability and accuracy of the assay, a comparison was conducted using the qRT-PCR method (56 with the BRAF V600E mutation and 80 without the mutation) on FNA samples from 136 thyroid nodule patients and the NGS method for somatic hotspot mutations on FFPE samples from 50 cancer patients. The NE-MS assay demonstrated 100% sensitivity and specificity, with a kappa value of 1.00 when compared with qRT-PCR and NGS (Table 1). A heatmap of these 50 FFPE samples s (Supplementary Fig. S2) depicted the consistency of multiple mutation detection between the NE-MS and NGS methods.

Table 1 Comparison of the NE-MS assay with qRT-PCR and NGS methods for the detection of thyroid cancer-related somatic mutations.
Full size table

NE-MS-based highly sensitive profiling somatic mutations in thyroid FNA samples

The BRAF V600E mutation often occurs at low abundance in clinical FNA samples, which are frequently of poor quality. To further demonstrate the applicability of the system for detecting low abundance BRAF mutation in FNA samples, the ddPCR method verified the feasibility. We observed an intriguing phenomenon: some clinical samples had low peak intensity on the mass spectrum (Supplementary Fig. S3a) and relatively low mass height on the call cluster plot, which was analyzed by the “height” method (Supplementary Fig. S3b). Somatic Variant Report (SVR) software still judged the results as mutant. We verified the accuracy of this method in detecting low abundance using the ddPCR method. We analyzed 40 clinical FNA specimens exhibiting low peak intensity and mass height using both the NE-MS assay and a commercial ddPCR assay. The concordance between the novel NE-MS assay and ddPCR for BRAF V600E mutation was 40/40 (100%) (Supplementary Table S1). To clarify the precision of NE-MS, including inter-run and inter-operator variability for low-VAF BRAF V600E mutations, three technicians (labeled A-C) performed testing using clinical FNA samples harboring low-VAF BRAF V600E mutations across 15 independent replicates (Table 2). The inter-run concordance rate for detection of the BRAF V600E mutation was 100%.

Table 2 Inter-run and inter-operator reproducibility of for BRAF V600E mutations. “+” represents BRAF V600E mutation.
Full size table

Using the multiplex NE-MS assay, we analyzed 466 FNA samples for 26 somatic hotspot mutations. Prior to evaluating the clinical diagnostic value of thyroid cancer of the proposed system for thyroid cancer diagnosis, we assessed the performance of the NE-MS model for somatic mutations detection. The receiver operating characteristic (ROC) curve of the system was shown in Fig. 4a. The system evaluated the accuracy performance of BRAF mutation with a high area under the curve (AUC) of 0.99. At a Z-score threshold of 7, the assay achieved a true positive rate (TPR) of 0.99 and a false positive rate (FPR) of 0.00. While the sensitivity of NE-MS is not as high as that of the most sensitive ddPCR assays currently available, the accuracy of NE-MS has been significantly improved through optimization of the analytical workflow. Furthermore, we examined the percentage of BRAF alone, RAS alone, and multiple mutations (≥ 2 mutations) across each Bethesda grade (Supplementary Fig. S3c). BRAF mutation was present in all Bethesda grades, with a notably high prevalence in Bethesda V (74.03%) and VI (89.01%). The highest occurrence of RAS mutations was observed in Bethesda IV (26.09%), while multiple mutations were predominantly found in Bethesda grades IV through VI.

Fig. 4
Fig. 4
Full size image

Clinical applicability of the NE-MS assay for thyroid FNA samples. (a) Receiver operating characteristic (ROC) curve showing the true positive rate (TPR) and false positive rate (FPR) of the NE-MS assay for BRAF V600E mutation detection. (b) ROC curves for thyroid cancer detection comparing the NE-MS assay, qRT-PCR, and cytological analysis. AUC, the area under the ROC curve.

Of the 466 FNA samples mentioned above, 211 thyroid nodules underwent surgical resection, followed by histopathological examination and mutation profiling (Supplementary Fig. S4). The diagnosis models for thyroid cancer were constructed using the three methods, including NE-MS for multiple mutations, qRT-PCR testing for BRAF V600E mutation, and cytology analyses for pathological diagnosis. As illustrated in Fig. 4b, the ROC curve for cytology analysis yielded an AUC of 64.29% (95% CI: 48.58% − 80.01%). In contrast, qRT-PCR testing showed a favorable performance with an AUC of 74.38% (95% CI: 54.14% − 94.62%), while NE-MS assay had a more favorable performance with an AUC of 92.40% (95% CI: 88.37% − 96.44%). These results indicated that the NE-MS assay might be a more dependable tool for thyroid cancer diagnosis and clinical decision-making.

Clinical significance for NE-MS-based highly sensitive profiling somatic mutations in thyroid cancer

We analyzed 1,003 thyroid cancer FFPE samples using the NE-MS assay (Supplementary Fig. S5). A total of 826 (82.35%) cases BRAF mutations, 7 (0.70%) RAS mutations (NRAS, HRAS, and KRAS), 4 (0.4%) RET mutations, and 1 (0.1%) TERT mutation alone were identified. Additionally, 36 (3.59%) patients had multiple mutations (2.79% with BRAF and TERT, 0.50% with BRAF and PIK3CA, 0.20% with BRAF and NRAS, and 0.10% with BRAF, TERT, and PIK3CA). Among cases with BRAF mutations, 68.36% were diagnosed with PTC, 31.52% with papillary thyroid microcarcinoma (PTMC), and 0.12% with MTC.

Table 3 The relationship between clinicopathologic parameters and multiple mutations in thyroid cancer. * P < 0.05; *** P < 0.001; **** P < 0.0001. RAI, radioactive iodine.
Full size table

Several studies have suggested that the co-existence of BRAF mutation and other mutations is associated with an increased risk of malignancy in thyroid cancer8. Furthermore, detecting multiple mutations is clinically recommended in thyroid cancer patients23,24,25. We compared the clinicopathological features of thyroid cancer cases with BRAF mutations alone, multiple mutations, and no mutations (Table 3). The data showed that multiple mutations were more frequently observed in age ≥ 55 years (p <0.0001), male gender (p < 0.05), tumor size > 1 cm (p < 0.05), distant metastasis (p < 0.05), TNM stage Ⅱ (p < 0.0001) and more adjuvant RAI therapy (p < 0.05) than BRAF mutation alone. Compared to those without mutations, multiple mutations were also associated with older age (p <0.0001), male gender (p < 0.05), and higher tumor stage (p < 0.01). Furthermore, we found 4 out of 10 MTC cases harboring the RET M918T mutation (Table S2). Notably, all patients with this specific mutation were female and classified as TNM stage ≥ II. In addition, one of these patients had developed lung metastases, further highlighting the aggressive nature of RET-mutated MTC. These findings thus highlight the potential clinical value of detecting multiple somatic mutations in thyroid cancer for prognostic assessment.

Discussion

The global incidence of thyroid cancer has increased significantly over the past 30 years, especially among young people. This increasing incidence is thought to be partially attributed to overdiagnosis26. Accurate diagnosis and optimal management of thyroid cancer remain unmet clinical needs. Multiple mutation detection is helpful for thyroid cancer diagnosis. Several multiplex molecular testing techniques for thyroid cancer have been reported. Representative projects include the Afirma gene expression classification profile, ThyGenX, and Thyroseq27. However, the aforementioned techniques have higher personnel and equipment requirements, more time and economic costs, and cannot meet the clinical requirements. MALDI-TOF MS is widely used to detect single nucleotide polymorphisms (SNPs). However, its limited sensitivity renders it unsuitable for routine somatic mutations detection17,28.

The detection limit of MALDI-TOF MS technology is affected by both multiplex PCR amplification and multiplex primer extension reactions29. To enhance the analytical sensitivity of this platform, herein, we implemented a novel approach NE-MS in which the wild-type matched mass-modified nucleotide was excluded from the ddNTPs mixture during the extension reaction, and extension probes selectively target mutant products, thereby enriching mutant alleles for downstream analysis and minimizing false-positive signals. In this study, peak intensities were converted into quantifiable Z-score values, and results were automatically analyzed using predefined cutoff thresholds, eliminating the time inefficiency associated with manual interpretation. Furthermore, we explored the application of the NE-MS assay for somatic mutations profiling, achieving an 8-fold lower LOD compared to the regular MS assay. We achieved a high analytical sensitivity of ≤ 0.32% for detecting somatic mutations, including NRAS Q61K and TERT C228T.

Compared with NGS, this method offers a shorter processing time (approximately 6 h), a simpler operational procedure, and a more accessible analytical workflow, making it better suited for clinical applications, particularly in diagnostic settings involving preoperative biopsies. Furthermore, the overall cost is reduced by approximately two-thirds relative to NGS30. Notably, NE-MS differs from UltraSEEK™ in that it omits the biotin purification step, thereby improving operational efficiency and reducing turnaround time. Compared with the SBE-MALDI-TOF platform, the NE-MS system incorporates large-scale amplification at the initial stage, enabling preservation of limited clinical samples-a critical advantage in settings where sample availability is constrained. At the same time, unlike SBE-MALDI-TOF, it does not amplify both wild-type and mutant alleles simultaneously, thereby enhancing detection sensitivity. Finally, its significant innovation lies in its application to the differential diagnosis of benign and malignant thyroid lesions and the prediction of disease prognosis, addressing unmet clinical needs in thyroid cancer management.

We developed this assay to enable the simultaneous detection of 26 somatic hotspot mutations that cover genes recommended by the National Comprehensive Cancer Network (NCCN) guidelines for thyroid cancer diagnosis and prognosis, such as BRAF, RAS, TERT, PIK3CA, and RET mutations31. For the first time, our research develops a highly sensitive NE-MS assay for thyroid cancer somatic mutations, providing a comprehensive view of the somatic mutations landscape in thyroid cancer. This study detected both FNA and FFPE samples, as both sample types are commonly used for thyroid molecular analysis. We compared three diagnostic methods for thyroid cancer, and the NE-MS assay demonstrated the best diagnostic performance. Notably, although RAS mutations have limited diagnostic utility in thyroid cancer, they are a significant risk factor for the transformation of adenomas into adenocarcinomas32,33. Therefore, it should also be used as a key screening indicator for thyroid cancer.

When applying NE-MS to detect BRAF mutations in FFPE samples, we identified BRAF mutations in 82.35% of cases, 68.36% of which were diagnosed with PTC, consistent with prior reports34. Certainly, we also observed that the rate of TERT mutations was lower compared to values reported in the literature. The low detection rate of TERT mutations is primarily attributed to strict sample quality requirements. Therefore, for reliable TERT mutation detection, we recommend a minimum DNA input of 100–150 ng and an A260/280 ratio of 1.8–2.0. Some studies suggested that the co-existence of BRAF mutation with other genes is an essential factor leading to poor prognosis8,35. Consistent with prior studies, our findings demonstrate that multiple mutations are associated with aggressive clinicopathological features in thyroid cancer36. However, BRAF mutation alone did not correlate significantly with aggressive clinicopathologic characteristics37,38. In addition, our study identified 4 cases of RET M918T mutation for high-grade MTC. For patients with a family history of hereditary MTC, prospective RET mutation screening in family members can identify carriers before the onset of clinical symptoms39. It is crucial to note that RET M918T mutation is currently used for high-risk staging of MTC. Hence, the highly somatic mutations profile is also the basis for targeted drug therapy, including selpercatinib and pralsetinib40.

Despite the excellent performance of the NE-MS assay, there are some limitations. Firstly, the assay requires a higher input of DNA (50 ng) for poor-quality samples, which adds a DNA enrichment step to the process. In addition, there are a limited number of patients in the analysis of clinicopathological parameters due to incomplete information on patients’ stages and RAI therapy. Future studies are necessary to follow up data on these patients and increase the number of patients with clinicopathological information, further establishing its utility in clinical practice.

In summary, we developed a mutant allele-selective ddNTP enrichment (NE)-assisted MALDI-TOF MS assay (NE-MS) for the detection of multiple somatic mutations in thyroid cancer, enabling comprehensive profiling of the mutation landscape. Additionally, with the advantages of highly sensitive and throughput of the NE-MS analysis, it exhibited a robust performance when evaluated for diagnostic performance and clinical prognosis of thyroid cancer. Given its lower cost, simple workflow, and high sensitivity, NE-MS is a practical and ideal adjunct to pre- and postoperative pathological evaluation and clinical decision-making. Based on the principle of nucleotide enrichment coupled with MALDI-TOF MS, this approach is not limited to thyroid cancer but can also be applied to detect somatic mutations across multiple cancer types.

Methods

Patients and samples

466 FNA samples and 1003 FFPE samples were prospectively collected from June 2022 to October 2023 at the biobank of the First Affiliated Hospital of Chongqing Medical University (Chongqing, China), respectively. FNA samples were performed by interventional radiologists using ultrasound-guided FNA. FFPE samples were obtained from thyroid nodule patients who underwent surgery at the Department of Breast and Thyroid Surgery. All samples utilized in this study were derived from clinical patients. Patient demographic information, cytology reports, and pathology reports were collected from the Hospital Information Management System. The clinicopathological and demographic characteristics of the subjects were detailed in Supplemental Table S3. For a more thorough analysis of the relationship between clinicopathological parameters and multiple mutations in FFPE samples from thyroid cancer patients, the study enrollment process and flow diagram were illustrated in Supplemental Fig. S6. The inclusion criteria for participants were as follows: (1) Patients aged 18 years or olde; (2) Clinical samples primarily consisting of thyroid cancer, including both newly diagnosed and recurrent cases; (3) Availability of complete and traceable sample information, including sample number, sex, age, and clinical background data such as pathological findings. The exclusion criteria were as follows: (1) Samples with incomplete or untraceable case records; (2) Only RAS, TERT or RET mutations; (3) Patients with duplicate records due to left and right sides. Tumor-node-metastasis (TNM) staging was classified according to the 8th edition of the American Joint Committee on Cancer (AJCC) staging system36. All patients provided signed informed consent for sample donation. This study was approved by the Ethics Committee of our hospital (No. K2024-088-01) in accordance with the Declaration of Helsinki.

DNA isolation and quantification

Genomic DNA was extracted from FNA and FFPE samples using the UPure Tissue DNA Kit (Biokeystone, Sichuan) and UPure FFPE Tissue DNA Kit (Biokeystone, Sichuan), respectively, according to the manufacturer’s instructions. The total DNA amount should be no less than 100 ng. DNA purity was assessed by measuring the A260/230 and A260/280 absorbance ratios using a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific, USA). The 260/230 ratio is recommended to be approximately between 2.0 and 2.2, and the 260/280 ratio approximately between 1.8 and 2.0. Prior to testing, DNA should be diluted or concentrated as necessary using a Concentrator plus (Eppendorf, Germany) to achieve a final loading concentration of 50 ng/µL.

Assay design

This assay comprises a two-round PCR process, as shown in Supplemental Fig. S7. Each NE-MS assay consists of two PCR primers and a single-base extension primer. All amplicons are designed to be less than 150 bp long to ensure successful amplification. The first PCR primers consist of a specific sequence at the 3’ end and a universal adaptor (ACGTTGGATG) at the 5’ end. The second round of PCR is a single-base extension using four wells containing 26 somatic hotspot mutations in seven genes for thyroid cancer (Supplemental Table S4). All primers were obtained from the ABA clearSEEK universal reagent kit (Cat:30000589SG). Primers were designed using the Agena Cx customer portal (https://support.agenabio.com/) and synthesized by Agena Bioscience Inc.

NE-MS workflow

Multiplex PCR was performed in a 10 µL reaction volume, containing 1 µL of genomic DNA (50 ng/µL) and 9 µL of a master mix composed of 1× PCR buffer supplemented with 2 mmol/L MgCl2, 100 µmol/L dUTP/dNTPs mix, 2 U Taq polymerase, and 100 nmol/L of each PCR primer mix. The reaction parameters were as follows: 1 cycle of 98˚C for 3 min, followed by 10 cycles of 98˚C for 30 s, 68˚C for 30 s, and 72˚C for 1 min with a decrease of one degree per cycle, and 35 cycles of 98˚C for 30 s, 56˚C for 30 s, and 72˚C for 1 min, one cycle of 70˚C for 5 min. Amplified products (10 µL) were treated with 4 µL of 0.073 U/µL shrimp alkaline phosphatase (SAP) in 0.24× SAP buffer, resulting in a final volume of 14 µL for 40 min at 37˚C, followed by denaturation for 5 min at 85 °C.

Single-base extension reactions were performed by adding 2 µL of extension master mix (0.222× extension buffer, 1× extension enzyme, 0.94 µL extension probes, and 0.2 µL ddNTPs mix) to 7 µL of the purified PCR products. The reaction parameters were as follows: initial incubation at 95 °C for 30 s, followed by 40 cycles of 94 °C for 5 s with five nested cycles of 52 °C for 5 s and 80 °C for 5 s, and a final incubation at 72 °C for 3 min.

To each well, we added 0.63 µL of normalization control (Cat: 06169, Agena Bioscience) and 40.37 µL of HPLC-grade water. Extension products were then automatically processed using the Chip Prep Module Dx (CPM) of the MassARRAY® Dx Analyzer 4 (MA4) system (Agena Bioscience, San Diego, California, USA). After desalting the reaction products with 13 µL of MassARRAY® Clean Resin, samples were dispensed onto SpectroCHIP® CPM-96 chips with an initial dispense volume of 1200 nL. Ionization was achieved via a nitrogen laser within the MassARRAY® system (wavelength: 337 nm), with a maximum cumulative laser irradiation set at 30 shots, aiming at the acquisition of 12 (minimum) to 15 (maximum) spectra of acceptable quality. Raw data were analyzed using MassARRAY® TYPER software (v5.0.10), which automatically generated genotype peak area reports.

Allele calling and Z-score analysis

The variant allele frequency (VAF) was returned by the automated SVR software v 1.0.5 (Agena Bioscience), and allele calls with a peak area ≥5 were considered. 20 known wild-type FNA samples were used for baseline optimization, and an external baseline data file was generated (Supplementary Table S5). This baseline dataset was used as a historical reference for mutation detection in all subsequent analyses. Mutations were detected using a robust Z-score (median absolute deviation-based Z-score). Each assay’s intensity (X) was normalized to the linear fit of the internal controls. The Z-score was calculated for each assay using the median (M) and the median absolute deviation (MAD) values established for known wild-type samples41. By default, the analysis software marks a Z-score ≥ 7 as a mutation and reports it accordingly. The Z-score was calculated using the following formula:

$$Z - score{\text{ }} = \frac{{X - M}}{{MAD}}$$

qRT-PCR assay

BRAF V600E mutation was detected via the qRT-PCR method. According to the manufacturer’s protocol, the thermocycling conditions of PCR were as follows: 1 cycle of 37˚C for 10 min and 95˚C for 5 min, followed by 40 cycles of 95˚C for 15 s, 60˚C for 60 s. The reagents were provided by YZY Medical Science and Technology (Cat: YZYMT-003-A). ΔCT = Ct (BRAF target gene assay) - Ct (internal control assay); ΔCT < 9 was considered a positive qRT-PCR result.

NGS analysis

DNA was analyzed with the capture-based targeted NGS panel. The gene panel detects 520 cancer-related genes (Burning Rock Biotech Ltd., RS03F-12-1, Guangzhou, China) in BRAF, KRAS, NRAS, HRAS, TERT, PIK3CA and RET. Library preparation and quality control followed the same standardized pipeline for all samples. The samples were sequenced on a Miniseq sequencer (Illumina, San Diego, CA) with paired-end reads. NGS data were analyzed using Burning Rock Biotech.

ddPCR assay

All ddPCR experiments were performed on the Droplet Digital PCR system (Targeting One, Beijing, China) using a commercially available human BRAF V600E gene mutation kit (Cat: 12231; Targeting One, Beijing, China). Experiments were performed according to the manufacturer’s instructions. The PCR conditions were one cycle of 95˚C for 10 min, followed by 50 cycles of 94˚C for 30 s, and 60˚C for 1 min. ddPCR was used as the gold standard for validating reference standard dilutions and confirming low-abundance BRAF mutations.

Limit of detection (LOD)

The BRAF V600E, NRAS Q61K, TERT C228T, and PIK3CA H1047R mutations reference standards were purchased from Genewell (Cat:2315, 3470; Shenzhen, China) with a 5% VAF. Dilutions tested included the following: 5%, 2.5%, 1.25%, 0.63%, 0.32%, and 0.16%, with the wild-type being used as a negative control. The experiments were repeated three times independently.

Statistical analyses

Statistical analyses were performed using GraphPad Prism 8.0.1 (GraphPad Software, San Diego, CA), R version 4.3.2, and MedCalc software version 22.007 (Ostend, Belgium). P < 0.05 was considered statistically significant. All statistically significant results were noted in the tables. The performance of the NE-MS assay was evaluated using receiver operating characteristic (ROC) curves and a 95% confidence interval (CI). Results were coded as mutation-positive (1) or mutation-negative (0) for the calculation of sensitivity and specificity.