Identification strategy of wild and cultivated Astragali Radix based on REIMS combined with two-dimensional LC-MS

Astragali Radix (AR), the dried root of Astragalus membranaceus (Fisch.) Bge. var. monghoicus (Bge.) Hsiao (AMM) or Astragalus membranaceus (Fisch.) Bge. (AM), is a widely used traditional Chinese medicine containing saponins, flavonoids, organic acids, and polysaccharides with diverse bioactivities. Market demand and declining wild resources have led to mixing of wild AR (WAR) and cultivated AR (CAR), and prior studies indicate chemical differences between WAR and CAR that may alter clinical efficacy. Thus, a rapid and effective method to distinguish WAR from CAR is needed. Conventional spectroscopy, chromatography, and MS approaches can authenticate botanicals but often require laborious preparation and may offer limited coverage for complex herbal matrices. REIMS provides real-time, ambient ionization MS with minimal preparation, and 2DLC can improve chromatographic separation for comprehensive profiling. This study aimed to rapidly and accurately distinguish WAR and CAR by integrating REIMS for real-time phenotyping with 2DLC-MS for comprehensive chemical characterization and QTRAP-MS for semi-quantitative analysis.

Recent advances in spectroscopy (Raman, IR, UV–Vis) and MS-based methods (HPLC, TLC, LC-MS, GC-MS) have enhanced plant authentication, with LC-MS offering rapid, information-rich detection and being widely used to identify easily confused plants. Targeted and untargeted MS strategies are common in plant metabolomics workflows. However, these methods often require cumbersome preparation and long runs, and may not fully characterize complex herbal matrices. REIMS, introduced in 2013, enables ambient, real-time molecular fingerprinting without chromatography, and has been applied to tissue typing and food quality/safety. Two-dimensional LC (e.g., HILIC coupled with RPLC) increases orthogonality and peak capacity, beneficial for high-throughput metabolomics. QTRAP-MS provides multiple scan modes (e.g., MIM, MRM, precursor/neutral loss scans and enhanced scans) that facilitate targeted and pseudo-targeted analyses. Prior AR studies include diagnostic ion-guided UPLC-QTOF/MS for species tracing and multidimensional LC/HRMS revealing extensive metabolite coverage across AR species. Given the complexity of herbal metabolites, integrating complementary MS platforms can provide systematic chemical characterization and robust differentiation.

Study design: An integrated workflow combined REIMS for rapid classification, 2DLC-MS for qualitative profiling, and QTRAP-MS for screening and semi-quantitation of differential markers between WAR and CAR. Samples: 40 authenticated batches (20 WAR, 20 CAR) were collected and authenticated (including microscopic confirmation). For REIMS modeling, each sample was cut 10 times (technical replicates). Eight additional unknown market samples (U1–U8) were used for real-time identification. Sample preparation: - REIMS: Powdered AR (1.0 g) was wetted with 1.5 mL water to a viscous state and placed in a cylindrical, foil-wrapped container for iKnife cutting. A pooled QC (from all batches) optimized REIMS conditions. - 2DLC-MS: 500 mg powder extracted with 10 mL 70% methanol (v/v), ultrasonicated 1 h, centrifuged (10,000 rpm, 10 min), supernatant filtered (0.22 µm PTFE). A pooled QC monitored system performance. Solutions stored at 4 °C. REIMS instrumentation and conditions: REIMS ion source coupled to a Waters Xevo G2-XS Q-TOF MS. Monopolar electrosurgical knife (PS01-63H) and Erbe VIO 50 C generator. Aerosols aspirated via PTFE tube and venturi gas jet pump. Negative ion mode, m/z 50–1200; lock-mass leucine enkephalin (200 pg/µL, 0.1% formic acid–50% acetonitrile/water) at 2 µL/min. Cutting parameters: cutting time 1 s, length 1 cm, 10 replicates per sample with 2–3 s delays. Key optimizations using QC: cone voltage 40 V (max signal at 40 V), heater bias 55 V (max TIC at 55 V). Mass drift corrected to m/z 554.3. REIMS data processing and modeling: 400 spectra (20 WAR, 20 CAR; 10 cuts each) yielded multispectral fingerprints. Preliminary assignment of 36 compounds (organic acids, flavonoids, saponins, others) by accurate mass and literature. Multivariate analysis used LiveID: PCA for unsupervised variance exploration, followed by PCA-LDA for supervised classification with reduced overfitting risk. Model validation via stratified 5-fold cross-validation; performance metrics included correct/incorrect classifications and outliers. Real-time recognition: The trained model was applied to unknowns U1–U8. LiveID provided decision and confidence. A rule required at least 3 of 10 cuts with 100% match for positive identification. 2DLC-MS qualitative profiling: A Nexera X2-based 2DLC system (HILIC in 1D for high-polar compounds; heart-cuts to RPLC in 2D for mid/low polarity) expanded metabolite coverage and resolution across polarities. Twenty WAR and twenty CAR batches were profiled. Compound annotation employed molecular formula prediction, online databases, and Insight Explore “assign” functionality; retention times, accurate masses, and fragments compiled (Table S2). 2DLC-MS multivariate statistics: Peak tables from negative ion mode were analyzed in MetaboAnalyst 6.0. PCA assessed overall variance and QC stability; supervised modeling with OPLS-DA and 100-permutation testing evaluated class separation and model robustness. Volcano plots aided selection of differential ions. QTRAP-MS screening and semi-quantitation: Potential markers from 2DLC-MS statistics were further screened in QTRAP-MS using MIM-IDA-EPI workflows. A UHPLC-QTRAP-MS semi-quantitative method was established for 45 differential compounds to compare relative abundance between WAR and CAR.

- REIMS optimization: Cone voltage 40 V and heater bias 55 V provided maximal ion intensity. - REIMS chemical features: 36 compounds preliminarily annotated across m/z 100–500 and 600–1000, including organic acids (e.g., citric acid m/z 191.0186), sugars (sucrose m/z 341.1078), fatty acids (linoleic acid m/z 279.2319), flavonoids/isoflavones (liquiritigenin m/z 255.0652, formononetin m/z 267.0652, naringenin m/z 271.0601), and multiple saponins (e.g., astragaloside III m/z 783.4536, astragaloside II m/z 825.4642, astragaloside VI m/z 945.5065, agroastragaloside IV m/z 987.5170). Distinct spectral differences between WAR and CAR were evident, notably in m/z 100–500 and 600–1000 regions. - REIMS classification performance: PCA showed tight clustering and clear separation of WAR and CAR. PCA-LDA achieved robust discrimination; stratified 5-fold validation over 314 spectra yielded three outliers and 99.04% correct classification, indicating high stability and generalization. - Real-time identification: Eight unknowns were tested online; U1–U4 classified as WAR and U5–U8 as CAR with >99% confidence per sample, meeting the predefined matching criterion. - 2DLC-MS coverage: 124 compounds characterized across cohorts (40 flavonoids, 29 organic acids, 20 saponins, 35 others), including species detected by REIMS, demonstrating broad chemical coverage and improved separation via HILIC–RPLC orthogonality. - 2DLC-MS statistics: PCA/OPLS-DA indicated clear group separation between WAR and CAR; 100-permutation testing supported model robustness. Volcano plots highlighted differential ions. - QTRAP-MS semi-quantitation: A UHPLC-QTRAP-MS method quantified 45 differential compounds; most exhibited higher relative abundance in WAR than CAR, supporting REIMS-based differentiation. - Overall: The integrated REIMS + 2DLC-MS + QTRAP workflow provided rapid, preparation-free classification with molecular-level validation and marker semi-quantification, reliably distinguishing WAR from CAR.

The study addresses the need for rapid and accurate authentication of wild versus cultivated Astragali Radix. REIMS enabled near-instant, preparation-free acquisition of molecular fingerprints that captured compositional differences, achieving high-accuracy classification with robust cross-validation and successful real-time identification of market samples. The orthogonal 2DLC-MS profiling confirmed substantial chemical divergence between WAR and CAR across multiple metabolite classes and provided extensive coverage (124 annotated compounds). Subsequent QTRAP-MS screening and semi-quantitative analysis of 45 differential markers corroborated that most featured higher relative levels in WAR, offering a chemical basis for REIMS-based phenotypic separation. Collectively, the findings validate an integrated strategy that balances speed (REIMS) with depth (2DLC-MS) and targeted verification (QTRAP-MS), enhancing confidence in TCM discrimination and supporting quality control and authenticity testing.

This work establishes an integrated identification strategy for Astragali Radix that combines REIMS for rapid, real-time classification with comprehensive 2DLC-MS profiling and QTRAP-MS-based semi-quantitation of differential markers. The REIMS model achieved 99.04% validation accuracy and confidently identified unknown market samples, while 2DLC-MS annotated 124 compounds and QTRAP-MS semi-quantified 45 markers, most at higher relative levels in WAR. The approach provides a fast, minimally destructive, and accurate solution for distinguishing WAR from CAR and offers a reference framework for TCM analysis and authentication. Future work could expand sample diversity and geographic coverage, integrate additional ionization modes and higher-resolution MS/MS, and refine quantitative calibration to further standardize marker-based differentiation.