English

• Lung cancer is the leading cause of cancer-related deaths world-wide [1]. It is well accepted that the complete resection of tumor is associated with an improved prognosis for lung cancer [2]. Usually, lung cancer surgical margin is assessed intraoperatively by histo-logical evaluation of multiple frozen sections [3]. Histological eval-uation process involves tissue freezing, cryosection, hematoxylin and eosin (H & E) staining, and microscopic examination, this typ-ically takes more than 30 min. Due to the histological examina-tion is a subjective procedure, the results are highly dependent on the skill of technicians and pathologists as well as the qual-ity of frozen sections [3-6]. Fluorescence imaging can be utilized for intraoperative determination of tumor margins with the help of tumor-targeted fluorescence nanoprobes. Nevertheless, conven-tional fluorescence probes frequently encounter challenges related to poor signal-to-background ratios [7]. Therefore, it is urgent to develop alternative methods for the more accurate, reliable, and real-time intraoperative assessment of tumor margins during lung cancer surgeries. This would be of great benefit in improving sur-gical outcomes and prognosis, as well as for reducing the risk of postoperative recurrence in lung cancer patients.

  In alternative to the standard histopathology methods for the evaluation of tumor margins, molecular analysis driven by mass spectrometry-based techniques has gained increasing attention in tumor margins definition. This is largely because molecular analy-sis, with high accuracy and sensitivity, provides more information than that may be unavailable in histopathology [8]. The molecular analysis of tumor margins by ambient mass spectrometry (AMS) opens a new perspective to incorporate cancer-specific biomarkers into clinical decision-making for improving the accuracy, specificity and spatial resolution of cancer diagnosis, with the advantages in real-time and direct profiling of molecular information from var-ious biological samples [6,9,10]. For example, desorption electro-spray ionization mass spectrometry (DESI-MS) has been employed for the determination of the boundaries between healthy and neo-plastic tissues of human brain tumors and glioma based on lipid patterns [11,12]. DESI-MS imaging has been employed for the direct molecular assessment of surgical margins of gastric cancer [9] and human medulloblastoma cancers [13]. In addition, the in vivo application of rapid evaporative ionization mass spectrometry (REIMS) has been demonstrated in the detection of the tumor mar-gins in brain, liver, lung, breast, or colorectal tumors [14]. More re-cently system, MasSpec Pen [6] has been demonstrated suitable for tumor margin assessment during pancreatic cancer surgery [15]. These results indicate the huge potential of AMS for the determina-tion of cancer margins. However, DESI-MS mainly obtained surface molecular information, while a large amount of heat is involved in the operation/ionization sampling process of REIMS. Also, the degradation of biological analytes is difficult to avoid, which might cause some information to be lost and eventually introduce un-certainties into the results. MasSpec Pen device must be placed in tight contact with the tissue for analysis. Possible contamination of the device with tissue debris on the analytical results limits its use for cancer research [10,16].

  Internal extractive electrospray ionization mass spectrometry (iEESI-MS) [17-20] is an alternative AMS method for the accu-rate evaluation of lung cancer molecular margin. Like DESI-MS and REIMS, iEESI-MS technology enables direct analysis of tissue sam-ples. Unlike DESI-MS and REIMS, where tissues are typically ana-lyzed online by continually scanning position of the sampling spot, in iEESI-MS every tissue spot is sampled and analyzed separately. Given the necessity for precise control over tissue section thickness in this study, we sliced the tissue sections to a thickness of 1 mm and mounted them onto the slide. Typically, when tissue sample thickness is not a critical factor, sample preparation steps such as fixing and slicing may be unnecessary. The overall workflow of cell injection, tissue sample collecting, sample grouping, and data pro-cessing is shown in Fig. 1A. To ensure the reproducibility of our work, we meticulously defined the capacity of the sample chamber, which in turn determined the precise amount of sample loaded into it. As shown in Fig. 1B, the tissue sample was directly loaded into the sampler chamber by applying a single punch with the sampler onto the tissue. Then the sample chamber loaded with the tissue samples were assembled with an iEESI source. The position-ing was meticulously controlled, with sampling conducted row by row and column by column. These procedures ensure that each tissue spot undergoes sampling and analysis individually. Such a me-thodical approach not only maintains uniformity but also enhances the reproducibility of iEESI-MS analysis. The process of iEESI-MS could be briefly described as follows: the extraction solution of CH₃OH/H₂O (v/v, 30/70) charged with +5.0 kV was infused into the tissue samples at 2.0 μL/min through a syringe (250 μL, Hamil-ton, GR, Switzerland) controlled by a syringe pump (Harvard, MA, USA). This process selectively extracts analytes distributed within the tissue samples, producing gas ions for mass spectrometry pro-filing. The iEESI-MS approach has been successfully applied for the rapid discrimination of tissue samples from human esophageal squamous cell carcinoma [21], lung cancer [22,23], and liver cancer [24]. While in the earlier works only cancer and normal datasets were used for differentiation analysis. In the present study, we in-corporated the data from tissue samples in the tumor margin area (M) to delineate the possible locations of tumor molecular margins. Additionally, we tracked the recurrence of mice following tumor removal based on the conventional pathological method. Our find-ings revealed that mouse, with ca. 1–2 mm wider molecular tumor margins than pathological tumor margins, experienced postoper-ative recurrence three weeks after tumor removal based on the pathological margins. These results underscore the great potential of iEESI-MS in improving the accuracy of tumor margin determina-tion during lung cancer surgery. Furthermore, our study identified two potential biomarkers, creatine and taurine, which showed pos-itive correlations with lung cancer.

  Figure 1

  

  Figure 1.  Experimental workflow for tracing the molecular tumor margin using iEESI-MS. (A) Schematic diagram of operation procedures of cell injection, tissue sample collection, sample grouping, and data processing, (B) disposable iEESI ion source for tissues analysis.

  DownLoad: Full-Size Img PowerPoint

  Note that normal tissues used in this study were collected from mice with lung cancers (outside the tumor) rather than from healthy mice without lung cancer. This strategy was employed to delineate the potential molecular margins of lung cancer. The cur-rent experiment included a total of 660 tissue samples. The time required for one tissue section is size-dependent. For example, if there are around 50 tissue spots within one tissue section, the whole time is around 30 min. The details of materials and method refer to the Supporting information.

  The m/z range of 700–900 was used for molecular differentia-tion as it contains the characteristic band of phospholipids. Phos-pholipids are integral to a multitude of biological functions includ-ing cell signaling, cell-cell recognition, immune response, energy metabolism, and malignant transformation of cells [25,26]. Conse-quently, investigating phospholipids is of great interest in reveal-ing the essence of life activities. Earlier studies revealed that phos-pholipids profiles differ between cancerous tissues and normal tis-sues in lung cancer [22,26] and liver cancer [24]. Fig. S1 (Sup-porting information) shows the mass spectrum of normal tissues (Fig. S1A) and cancerous tissues (Fig. S1B) of mice obtained on LTQ mass spectrometer. Partial least-squares discriminant analysis (PLS-DA) is chosen over other discriminant analysis methods due to its ability to handle multicollinearity, flexibility, dimensionality-reduction, and its proven success in modeling high-dimensional datasets for disease classification in medical diagnosis [27,28]. As shown in Fig. S2A (Supporting information), normal tissue samples are clearly separated from primary tumor tissue samples. For PLS-DA model, the values of R²X, R²Y and Q²Y were 0.77, 0.68 and 0.64, respectively. Validation with 200 random permutation tests of PLS-DA model generated intercepts R² = 0.19 and Q² = −0.14, in-dicating that the model was not overfitted (Fig. S2B in Supporting information). The iEESI-MS data corresponding to 24 tissue sam-ples in tumor margin area (numbers from 1A to 6D in Fig. S3A in Supporting information) were processed by the PLS-DA differential model (Fig. S2A) to assess lung cancer molecular tumor margin. As shown in Fig. S3B (Supporting information), the PLS-DA plot indi-cated that 8 tissue samples (Nos. 1A, 2A, 3A, 4A, 5A, 6A, 1B, and 2B) were assigned to normal tissue area, this was consistent with the results of histological analysis. Nevertheless, 3 tissue samples (Nos. 3B, 4B, and 5B) were assigned to cancerous tissue by PLS-DA. The results might suggest ca. 1–2 mm larger tumor size com-pared to the results of histological analysis. Additionally, multi-step screening processes including S-plot, and variable influence on pro-jection (VIP) value revealed that m/z 757 [PC(32:0) + Na]⁺, and m/z 783 [PC(36:4) + H]⁺ have the highest contribution to the molecular differentiation of lung cancer (Fig. S4 in Supporting information).

  To increase the reliability of experimental results obtained on LTQ at Nanchang, China, we performed cross-validation experi-ments on Orbitrap Fusion Tribrid mass spectrometer at Changchun, China. Thirty-seven normal tissue samples and 87 primary tumor tissues samples were used to build the molecular differentiation model using PLS-DA. As shown in Fig. 2A, the primary tumor and normal tissue samples could be completely separated. Validation with 200 random permutation tests of PLS-DA model (Fig. 2B) generated intercepts R² = 0.45 and Q² = −0.14, indicating that the model was not overfitted, which could be used for the molecu-lar margin determination. Therefore, the 18 iEESI-MS data points corresponding to 18 tissue points in tumor margin area (numbers from 1B to 8C in Fig. 2C) were then processed by the PLS-DA differ-ential model (Fig. 2A) to trace possible lung cancer molecular mar-gins. It could be observed from Fig. 2D that all data were attributed to the cancer group by PLS-DA model. Interestingly, 5 tissue points, including 1B, 1C, 6D, 8B, and 8C, were diagnosed as normal tis-sue samples via histological analysis. This molecular analysis re-sult suggests ca. 1–2 mm larger tumor size compared to the results of histological analysis. It is worth to be mentioned that mouse, with ca. 1–2 mm wider molecular tumor margins than pathological tumor margins, experienced postoperative recurrence three weeks after tumor removal according to the pathological margin, this further indicated that tumor removal based on molecular tumor margins might be helpful for the lower risk of tumor recurrence. Note that some single 1 mm square tissue points used for iEESI-MS might cross the boundary of a cancer tissue. In other words, both normal and caner tissues could exist in some tissue points. Thus, final analysis results may be biased if only depending on the percentage of cancer tissue over normal tissues.

  Figure 2

  

  Figure 2.  (A) Score plot of PLS-DA derived from iEESI-MS data of primary tumor (PT, red squares) and normal tissue samples (N, blue squares). (B) Validation of results with 200 permutation tests of PLS-DA model. (C) H & E stains of tissue sections. (Green dotted line indicated pathological tumor margin, and 18 numbers (from 1B to 8C) indicate the sampling location of 18 margin tissue samples). (D) Score plot of PLS-DA derived from iEESI-MS data of primary tumor (PT, red squares), normal (N, blue squares), and margin tissue samples (M, green squares).

  DownLoad: Full-Size Img PowerPoint

  Herein we also compared the molecular difference among nor-mal, primary tumor, and recurrent tumor tissue samples. As shown in Fig. S5 (Supporting information), a variety of ions were detected from normal (Fig. S5A), primary tumor (Fig. S5B) and recurrent tumor tissues (Fig. S5C). The predominant peaks at mass range of m/z 100–1000 were identified as m/z 148.0040 [Taurine + H]⁺, m/z 162.1128 [L-Carnitine + H]⁺, m/z 741.5335 [SM(34:1) + K]⁺, m/z 756.5530 [PC(34:3) + H]⁺, m/z 772.5280 [PC(32:0) + H]⁺, m/z 782.5690 [PC(36:4) + H]⁺. Chemical assignment for the 22 confidently identified species from the tissue sample using iEESI-MS/MS is shown in Table S1 (Supporting information). Here, the incon-sistency in positive and negative deviations of the results in Ta-ble S1 can be attributed to different mass spectrometry measure-ments. The predominant compound assignments were based on high-resolution MS data, collision-induced dissociation (CID) ex-periments, as well as human metabolome database. Note that the MS profiles recorded from the normal tissues (Fig. S5A) were sig-nificantly different from those recorded using the primary tumor (Fig. S5B) and recurrent tumor tissues (Fig. S5C). For example, many peaks with −44 Da mass difference were detected at mass range of m/z 500–1000 in the mass spectra of normal tissues, while peaks with −44 Da mass difference were detected at mass range of m/z 400–600 in the mass spectra of recurrent tumor tissues. For primary tumor tissues, it mainly detected lipids species at mass range of m/z 700–900, including m/z 741.5335 [SM(34:1) + K]⁺, m/z 756.5530 [PC(34:3) + H]⁺, m/z 772.5280 [PC(32:0) + H]⁺, and m/z 782.5690 [PC(36:4) + H]⁺. Although both primary tumor and recur-rent tumor tissues detected many lipids species at the mass range of m/z 700–900, subtle differences exist in the relative abundance.

  Previous studies confirmed that the molecular information ob-tained by iEESI-MS is diagnostic and predictive of disease state [22,24], hence the iEESI-MS data obtained from 174 tissue sam-ples (including 37 normal tissue samples, 87 primary tissue sam-ples, and 50 recurrent tumor tissue samples) were subjected to PLS-DA for differentiation of molecular differences among differ-ent types of samples. As shown in Fig. 3, the score plots of PLS-DA among recurrent tumor and primary tumor tissues (Fig. 3A), as well as recurrent tumor and normal tissues (Fig. 3B) exhib-ited a clear separation. In addition, leave one out cross-validation (LOOCV) revealed that the PLS-DA models among recurrent tumor and primary tumor tissues (Fig. 3C), as well as recurrent tumor and normal tissues (Fig. 3D) had robust cross validation scores (R² value = 0.98, Q² value > 0.93, accuracy > 0.98), suggesting that the established PLS-DA models had a good explanatory abil-ity and prediction ability. Due to the potential biomarkers con-tributed to the recurrence of lung cancer are very helpful for of-fering new therapeutic approaches for limiting tumor recurrence. Through the t-test, 1180 and 1126 features, with P < 0.05 and a false discovery rate (FDR) < 0.05, were identified between recur-rent tumor and normal tissues (RT + N), as well as between re-current tumor and primary tumor tissues (RT + PT), respectively. Venn diagram (Fig. S6 in Supporting information) showed that 558 identified features were shared between RT + N group and RT + PT group. Twenty-one out of 558 confidently identified metabolites (Table S1) were performed enrichment analysis (Fig. 3E), results indicated that these metabolites mainly involved in 13 metabolic pathways, such as spermidine and spermine biosynthesis, argi-nine and proline metabolism, biotin metabolism, glycine and serine metabolism, and taurine and hypotaurine metabolism. Two altered metabolic pathways, specifically arginine and proline metabolism, as well as taurine and hypotaurine metabolism, were revealed in pathway analysis of data derived from mouse inoculated subcuta-neously with A549 lung cancer cells using air-flow assisted desorp-tion electrospray ionization mass spectrometry imaging (AFA-DESI-MSI) [29]. This underscores the potential of iEESI-MS technology in molecular diagnosis of lung cancer. Note that since l-carnitine in Table S1 showed no significant difference between RT + N group and RT + PT group, it is excluded when performing enrichment analysis.

  Figure 3

  

  Figure 3.  Score plot of PLS-DA of (A) recurrent tumor and primary tumor tissues, (B) recurrent tumor and normal tissues. LOOCV of PLS-DA model of (C) recurrent tumor and primary tumor tissues, and (D) recurrent tumor and normal tissues (A, R², and Q² indicated accuracy, the explained variance, and the predictive capability of the model, respectively). (E) Metabolites enrichment analysis of 21 confidently identi-fied metabolites between recurrent tumor and primary tumor tissues as well as recurrent tumor and normal tissues. Note: RT, PT and N indicated recurrent tumor, primary tumor, and normal tissues, respectively.

  DownLoad: Full-Size Img PowerPoint

  Furthermore, using the Network Explorer module from Metabo-Analyst 5.0, we created metabolite-disease interaction networks to identify connections that cross pathway boundaries among 21 sig-nificantly changed metabolites. As shown in Fig. 4A, creatine and taurine showed positive correlations with lung cancer. Boxplots of creatine (Fig. 4B) and taurine (Fig. 4C) showed upregulated levels in the recurrent tumor tissues compared to that in the normal and primary tumor tissues.

  Figure 4

  

  Figure 4.  (A) Metabolite-disease interaction network analysis of shared 21 confidently identified metabolites within recurrent tumor and normal tissue group, as well as recurrent tumor and primary tumor tissue group) (Enriched terms are represented as nodes, and the node size represents the significance for each term. Circles rep-resent one metabolite and squares an associated disease). Boxplot of creatine (B) and taurine (C) among normal (N), primary tumor (PT) and recurrent tumor (RT) tissue samples. Box plots demarcate the median line, the 25^th and 75^th percentile (box), and 1.5 times the interquartile range (whiskers). Significant difference was determined by a two-tailed t-test (^* P < 0.05, ^** P < 0.01, and ^{* * *} P < 0.001).

  DownLoad: Full-Size Img PowerPoint

  Tumor surgery implies the removal of an apparently non-tumorous tissue around the tumor to reduce recurrence chances [30]. From this regard, accurate identification of resection margins plays a critical role in lowering cancer recurrence risk. Our results indicated that iEESI-MS combined with PLS-DA molecular differ-entiation model revealed that there was a potential molecular tu-mor margin ca. 1–2 mm wider than histological tumor margin. This suggests that iEESI-MS analysis may be more sensitive to the oc-currence of cancer in margin areas. Especially, consistent results between LTQ mass spectrometers at Nanchang, China and Orbitrap Fusion Tribrid mass spectrometer at Changchun, China further pro-vided solid support that molecular tumor margins were wider ca. 1–2 mm compared to the conventional histological tumor margins. Taken together, our results further highlighted that the develop-ment of novel molecular diagnosis methods plays a critical role in lowering the risk recurrence of lung cancer.

  Moreover, metabolite-disease interaction networks analysis highlighted that creatine and taurine showed positive correlations with lung cancer. Aberrant levels of metabolites are associated with cancer progression, which could potentially be used for di-agnosis or indicators for therapeutic evaluation [31]. Boxplot in Fig. 4 showed that creatine and taurine were upregulated in the recurrent tumor tissues compared to that in the normal and pri-mary tumor tissues. Creatine is a nitrogen-containing organic acid and can be converted into phosphocreatine to provide energy for muscle and nerve tissues [31]. Compared the expression levels of creatine among normal, primary tumor, and recurrent tumor tis-sues, creatine showed an up-regulated level both in primary tumor and recurrent tumor tissues, this is consistent with previous report that the mean concentration of creatine significantly increased in lung cancer patients compared with non-diseased controls in urine and serum samples [32]. Taurine is a non-protein amino acid, and influences various cellular functions, such as osmoregulation, an-tioxidation, and ion movement [33]. Previous study found that lung cancer patients with high serum taurine levels generally responded to PD-1 blockade antibody therapy, suggesting that taurine could serve as a potential therapeutic agent for lung cancer patients [34]. Taken together, these findings in respect to aberrant levels of cre-atine and taurine will aid in the elucidation of the mechanisms of lung cancer pathogenesis. Furthermore, we are continuing to ex-plore these differential features revealed by PLS-DA, with the aim of discovering more potential biomarkers and performing further validation of their specific role in lung cancer pathology. Based on the previous successful application of iEESI-MS in quantitative de-termination of bulk molecular concentrations of β -agonists in pork tissue samples [20] and amino acids in tissues for the assessment of lung cancer [23], we are going to utilize this method to quan-titatively and specifically identify biomarkers associated with lung cancer to enhance its accuracy and specificity in molecular diagno-sis of lung cancer margins.

  In conclusion, the proof-of-concept data demonstrate that the iEESI-MS technology coupled with PLS-DA could be used to deter-mine molecular tumor margins via mice with lung cancer through metabolites profiling with ca. 1 mm spatial resolution. The result indicate that potential molecular tumor margins are ca. 1–2 mm wider than histological tumor margins, highlighting the great po-tential of iEESI-MS approach for the objective determination of lung cancer margin with higher accuracy and sensitivity compared to the conventional histological analysis. However, major efforts should be focused on spatial resolution and potential misalign-ment of the histological image and annotation with the iEESI pix-els in the following work, with aim at increasing the accuracy of tumor margins determination during lung cancer surgery and low-ering the cancer recurrence risk. Further validation studies (such as comparison analysis between frozen sections and iEESI-MS) are required to determine the accuracy of iEESI-MS for intraoperative oncological margin assessment in the future work.

  Ethical statement

  All animal treatment and experiments were conducted accord-ing to the ethical guidelines for laboratory animal use and ap-proved by the Medical Research Ethics Committee of the Second Affiliated Hospital of Nanchang University.

  Declaration of competing interest

  The authors declare that they have no known competing finan-cial interests or personal relationships that could have appeared to influence the work reported in this paper.

  CRediT authorship contribution statement

  Haiyan Lu: Data curation, Writing – original draft. Jiayue Ye: Data curation, Formal analysis, Methodology, Writing – original draft. Yiping Wei: Formal analysis, Methodology, Writing – review & editing. Hua Zhang: Data curation, Methodology. Konstantin Chingin: Investigation, Methodology, Writing – review & editing. Vladimir Frankevich: Methodology, Writing – review & editing. Huanwen Chen: Conceptualization, Funding acquisition, Project administration, Supervision, Writing – review & editing.

  Acknowledgments

  This work was supported by Jiangxi Provincial International Sci-ence and Technology Cooperation Project (Nos. 20203BDH80W010 and 20232BBH80012), the National Natural Science Foundation of China (Nos. 82160410 and 81860379), Foundation of Jiangxi Provin-cial Department of Science and Technology (No. 20212ACB206018), and Key Research and Development Program of Jiangxi Province (No. 20223BBG71009).

  Supplementary materials

  Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.cclet.2024.110077.

  
  
• 1. [1]

     H. Sung, J. Ferlay, R.L. Siegel, et al., CA Cancer J. Clin. 71 (2021) 209–249. doi: 10.3322/caac.21660

  2. [2]

     P. Goldstraw, K. Chansky, J. Crowley, et al., J. Thorac. Oncol. 11 (2016) 39–51.

  3. [3]

     J.D. Predina, J. Keating, N. Patel, et al., J. Surg. Oncol. 113 (2016) 264–269. doi: 10.1002/jso.24130

  4. [4]

     V. Pirro, R.S. Llor, A.K. Jarmusch, et al., Analyst 142 (2017) 4058–4066.

  5. [5]

     K. Ashizawa, K. Yoshimura, H. Johno, et al., Oral Oncol. 75 (2017) 111–119.

  6. [6]

     J.L. Zhang, J. Rector, J.Q. Lin, et al., Sci. Transl. Med. 9 (2017) eaan3968.

  7. [7]

     J. Xin, S. Han, M. Zheng, et al., Chin. Chem. Lett. 35 (2024) 109165.

  8. [8]

     H.M. Brown, V. Pirro, R.G. Cooks, Clin. Chem. 64 (2018) 628–630. doi: 10.1373/clinchem.2017.281923

  9. [9]

     L.S. Eberlin, R.J. Tibshirani, J. Zhang, et al., Proc. Natl. Acad. Sci. U. S. A. 111 (2014) 2436–2441. doi: 10.1073/pnas.1400274111

  10. [10]

      H. Lu, H. Zhang, Y. Wei, et al., Analyst 145 (2020) 313–320. doi: 10.1039/c9an01365b

  11. [11]

      L.S. Eberlin, I. Norton, D. Orringer, et al., Proc. Natl. Acad. Sci. U. S. A. 110 (2013) 1611–1616. doi: 10.1073/pnas.1215687110

  12. [12]

      V. Pirro, C.M. Alfaro, A.K. Jarmusch, et al., Proc. Natl. Acad. Sci. U. S. A. 114 (2017) 6700–6705. doi: 10.1073/pnas.1706459114

  13. [13]

      L. Katz, M. Woolman, F. Talbot, et al., Anal. Chem. 92 (2020) 6349–6357. doi: 10.1021/acs.analchem.9b05340

  14. [14]

      J. Balog, L. Sasi-Szabo, J. Kinross, et al., Sci. Transl. Med. 5 (2013) 194ra93.

  15. [15]

      M.E. King, J. Zhang, J.Q. Lin, et al., Proc. Natl. Acad. Sci. U. S. A. 118 (2021) e2104411118.

  16. [16]

      P. Saudemont, J. Quanico, Y.M. Robin, et al., Cancer Cell 34 (2018) 840–851.

  17. [17]

      H. Zhang, K. Chingin, L. Zhu, et al., Anal. Chem. 87 (2015) 2878–2883. doi: 10.1021/ac504371z

  18. [18]

      H. Zhang, L. Zhu, L. Luo, et al., J. Agric. Food Chem. 61 (2013) 10691–10698. doi: 10.1021/jf4032469

  19. [19]

      H. Zhang, H. Gu, F. Yan, et al., Sci. Rep. 3 (2013) 2495.

  20. [20]

      J. Xu, S. Xu, Y. Xiao, et al., Anal. Chem. 89 (2017) 11252–11258. doi: 10.1021/acs.analchem.7b00517

  21. [21]

      J. Zhang, J. Xu, Y. Ouyang, et al., Sci. Rep. 7 (2017) 3738.

  22. [22]

      H. Lu, H. Zhang, K. Chingin, et al., Anal. Chem. 91 (2019) 10532–10540. doi: 10.1021/acs.analchem.9b01507

  23. [23]

      H. Lu, Y. Li, H. Zhang, et al., Talanta 233 (2021) 122544.

  24. [24]

      H. Lu, H. Zhang, Y. Xiao, et al., Analyst 145 (2020) 6470–6477. doi: 10.1039/d0an01080d

  25. [25]

      H. Zhang, K. Chingin, J. Li, et al., Anal. Chem. 90 (2018) 12101–12107. doi: 10.1021/acs.analchem.8b03022

  26. [26]

      H.-Y. Lu, J.-Y. Zhang, W. Zhou, et al., Chin. J. Anal. Chem. 44 (2016) 329–334.

  27. [27]

      L.C. Lee, C.Y. Liong, A.A. Jemain, Analyst 143 (2018) 3526–3539. doi: 10.1039/c8an00599k

  28. [28]

      D. Ruiz-Perez, H. Guan, P. Madhivanan, et al., BMC Bioinf. 21 (2020) 2.

  29. [29]

      J. Huang, S. Gao, K. Wang, et al., Chin. Chem. Lett. 34 (2023) 107865.

  30. [30]

      F.G. Almeida, P.F. de Aquino, S.R. Chalub, et al., J. Proteomics 154 (2017) 59–68.

  31. [31]

      L. Zhang, P. Bu, Trends Cell Biol. 32 (2022) 380–390.

  32. [32]

      D.P. Patel, G.T. Pauly, T. Tada, et al., J. Pharm. Biomed. Anal. 191 (2020) 113596.

  33. [33]

      T. Qaradakhi, L.K. Gadanec, K.R. McSweeney, et al., Nutrients 12 (2020) 2847. doi: 10.3390/nu12092847

  34. [34]

      Y. Ping, J. Shan, Y. Liu, et al., Cancer Immunol. Immunother. 75 (2022) 1015–1027.

• 

  Figure 1  Experimental workflow for tracing the molecular tumor margin using iEESI-MS. (A) Schematic diagram of operation procedures of cell injection, tissue sample collection, sample grouping, and data processing, (B) disposable iEESI ion source for tissues analysis.

  下载: 全尺寸图片 幻灯片
  

  Figure 2  (A) Score plot of PLS-DA derived from iEESI-MS data of primary tumor (PT, red squares) and normal tissue samples (N, blue squares). (B) Validation of results with 200 permutation tests of PLS-DA model. (C) H & E stains of tissue sections. (Green dotted line indicated pathological tumor margin, and 18 numbers (from 1B to 8C) indicate the sampling location of 18 margin tissue samples). (D) Score plot of PLS-DA derived from iEESI-MS data of primary tumor (PT, red squares), normal (N, blue squares), and margin tissue samples (M, green squares).

  下载: 全尺寸图片 幻灯片
  

  Figure 3  Score plot of PLS-DA of (A) recurrent tumor and primary tumor tissues, (B) recurrent tumor and normal tissues. LOOCV of PLS-DA model of (C) recurrent tumor and primary tumor tissues, and (D) recurrent tumor and normal tissues (A, R², and Q² indicated accuracy, the explained variance, and the predictive capability of the model, respectively). (E) Metabolites enrichment analysis of 21 confidently identi-fied metabolites between recurrent tumor and primary tumor tissues as well as recurrent tumor and normal tissues. Note: RT, PT and N indicated recurrent tumor, primary tumor, and normal tissues, respectively.

  下载: 全尺寸图片 幻灯片
  

  Figure 4  (A) Metabolite-disease interaction network analysis of shared 21 confidently identified metabolites within recurrent tumor and normal tissue group, as well as recurrent tumor and primary tumor tissue group) (Enriched terms are represented as nodes, and the node size represents the significance for each term. Circles rep-resent one metabolite and squares an associated disease). Boxplot of creatine (B) and taurine (C) among normal (N), primary tumor (PT) and recurrent tumor (RT) tissue samples. Box plots demarcate the median line, the 25^th and 75^th percentile (box), and 1.5 times the interquartile range (whiskers). Significant difference was determined by a two-tailed t-test (^* P < 0.05, ^** P < 0.01, and ^{* * *} P < 0.001).

  下载: 全尺寸图片 幻灯片
•