This retrospective cohort study was conducted in accordance with the Declaration of Helsinki. This study was reviewed and approved by the Institutional Review Board of Jinshan Hospital (JIEC 2023-S85). Written informed consent was obtained from all participants prior to enrollment for publication of any potentially identifiable data or images.

From October 2011 to October 2023, the skin tumor sample library was searched. The pathological and clinical information of adult patients undergoing surgical resection for skin tumors at Jinshan Hospital was retrieved in the study. Inclusion criteria were: 1) Histopathologically confirmed SCC; 2) Age ≥18 years. Exclusion criteria included: 1) Presence of systemic infection; 2) Incomplete clinical data; 3) Loss to follow-up. The patients were followed for up to 1 year to monitor tumor recurrence and metastasis. Finally, 34 patients with SCC were enrolled.

Demographic and clinical data (age, gender, tumor location) were collected for all participants. During surgery, both the SCC tumor and control tissue (healthy tissue ≥1 cm from the tumor margin) were excised for histopathological examination and subsequent lipidomic analysis. Tissue samples were frozen and stored at −80°C until analysis.

Isopropanol was used for initial sample preparation, while chloroform and methanol were part of the internal standard mixture. The lipid internal standards included species within the classes of cholesteryl ester (CE(18:1), CE(20:4), CE(22:6)), diacylglycerol (DAG(16:0/16:0), DAG(16:0/18:1), DAG(18:0/18:1), DAG(18:0/22:6)), lysophosphatidylcholine (LPC(16:0)), lysophosphatidylethanolamine (LPE(18:0)), phosphatidylcholine (PC(14:0/14:0)), phosphatidylethanolamine (PE(18:0/18:0), PE(18:0/22:6)), phosphatidylinositols (PI(16:0/16:0)), sphingomyelin (SM(16:0), SM(18:1), SM(24:1)), and triacylglycerol (TAG((16:0/18:1/18:1), TAG((16:0/18:1/22:6), TAG((18:0/18:1/18:2), TAG((18:1/18:1/18:2)). The mobile phases for LC-MS analysis included ammonium acetate and acetonitrile, with nitrogen gas aiding the electrospray ionization process. All reagents were sourced from Sigma-Aldrich and Thermo Fisher Scientific.

Tissue samples were prepared by adding 350 µL isopropanol and 9 µL internal standard mixture (A cold chloroform:methanol = 2:1 solution containing 1 μM LM6002), pre-cooled at 4°C, then mixed for 1 min. After 10 min of incubation at 24°C, samples were stored overnight at −20°C, followed by centrifugation at 12,000 g for 20 min. The supernatant was collected for LC-MS analysis. A targeted lipidomics approach was applied to screen the lipids (Dickinson et al., 2020).

Lipidomic analysis was performed using an AB SCIEX QTRAP 5500 system equipped with a Waters Acquity UPLC BEH HILIC VanGuard pre-column. Mobile phases were composed of acetonitrile-water mixtures with ammonium acetate. Electrospray ionization (ESI) in both positive and negative ion modes was employed for data acquisition.

The samples were injected into a Pre-column (2.1 mm × 5 mm, 1.7 μm, Waters Acquity UPLC BEH HILIC VanGuard). The phase A contained 95% acetonitrile (acetonitrile:water, 95:5) and 10 mmol/L of ammonium acetate. The phase B contained 50% acetonitrile (acetonitrile:water, 50:50) and 10 mmol/L of ammonium acetate. Ammonium hydroxide was added to phase B until its pH was equal to the pH of phase A. The flow rate was set to 0.5 mL/min. The elution gradient was as follows: Phase B started at 0.1%, raised to 20% in 10 min, then raised to 98% in the next 10–11 min, and remained for 2 min, and then returned to the initial 0.1%, which maintained until 16 min. Nitrogen was used as the solvent in the electrospray ionization process. Injection volumes were set to 2 μL for positive mode (ESI+) and 5 μL for negative mode (ESI-). The parameters were set as follows: curtain gas was 35 psi, GS1 was 50 psi, GS2 was 60 psi, ion spray voltage was 5,500 V, declustering potential was 80 V, entrance energy was 10 V, and collision energy collision energy was 50 V.

Raw LC-MS data were processed in Analyst software (version 1.7, SCIEX, MA, United States) for peak extraction and alignment. The data, including specific m/z values and fragmentation patterns, were analyzed to determine the composition and positional specificity of the fatty acids in phospholipids. MultiQuant software (AB SCIEX) facilitated lipid identification and quantification, referencing the LIPID MAPS database. The assignment to the sn1 position and the sn2 position is based on the known fragmentation behavior of molecules. Lipids detected in at least 80% of samples in either group were included, with missing values replaced by the median. Data normalization was performed using Pareto scaling prior to performing principal component analysis (PCA) and sparse partial least squares discriminant analysis (sPLS-DA) for data visualization and identification of discriminatory lipid profiles. Pooled QC samples were prepared by combining aliquots from all individual samples to create a representative mix. These pooled QC samples were used to monitor the performance of the LC-MS system throughout the analysis.

Key lipids were identified using stringent selection criteria to ensure their significance and reliability. These criteria included a variable influence on projection (VIP) score greater than 1, indicating the importance of the variable in the model, a false discovery rate (FDR) less than 0.05, ensuring statistical significance, and an area under the receiver operating characteristic curve (AUC) greater than 0.8, reflecting strong discriminative power. To validate these findings, the same criteria were applied to a publicly available lipidomics dataset from a study by Dickinson et al., which compared lipid profiles between 9 oral SCC tumor tissues and 10 healthy tissues using LC-MS (Züllig et al., 2020). Lipids that consistently met these criteria in both our dataset and Dickinson et al.'s dataset were defined as the key lipids associated with SCC tumor tissues, demonstrating their robust association with the disease.

Key differentially expressed lipids were first queried against the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (https://www.kegg.jp/) to predict potential metabolic pathways affected by altered lipid levels in SCC. MetaboAnalyst (https://www.metaboanalyst.ca/) was used to identify significantly altered metabolic pathways associated with SCC lipid metabolism.

Statistical analyses were conducted using R (V4.3.2) software. Data were assessed for normality and homogeneity of variance, and either paired t-tests or Kruskal-Wallis H tests were employed for group comparisons based on the results. The Benjamini-Hochberg procedure was utilized for multiple comparison corrections, and a p-value <0.05 was considered statistically significant. Results are presented as mean ± standard deviation (SD).

The study enrolled 34 patients with SCC. Their average age was 72 ± 11.5 years (range: 47–92 years), with no statistically significant difference between males (n = 18; 71 ± 9.4 years; 56–84 years) and females (n = 16; 73 ± 13.0 years; 47–92 years). During the follow-up period, no patients experienced recurrence or metastasis, highlighting the potential effectiveness of the employed therapeutic approach. Detailed clinical characteristics are presented in Table 1, with a representative case illustrated in Figure 1.

Analysis of the primary group revealed a rich lipidomic landscape, with 781 lipids identified initially. After applying stringent data quality criteria, 599 lipids were retained for further analysis. PCA plots indicated differential lipid regulation between groups and QC samples (Supplementary Figure S1). Utilizing sPLS-DA, a powerful statistical technique for dimensionality reduction and group discrimination, these lipids were effectively separated into distinct clusters (Figure 2A). The high predictive power of the model was confirmed by the R2X, R2Y, and Q2 values of 0.46, 0.91, and 0.85, respectively, indicating good model fitness and accurate classification capabilities.

A meticulous selection process identified 65 key lipids based on pre-defined criteria. These key lipids were deemed to be the most informative and reliable markers for discriminating between SCC and control samples.

Similarly, the validation group yielded valuable insights into the lipidomic profile of SCC. From the initial identification of 1370 lipids, 430 were deemed suitable for further analysis. Applying sPLS-DA, the model effectively separated the samples into distinct groups, as demonstrated by the score plot in Figure 2B. While the R2X, R2Y and Q2 values were 0.19, 0.76, and 0.62, respectively.

Loading plots for both datasets (Figure 3) provide detailed information about the contributions of individual lipids to the observed group separation. These plots allow for a deeper understanding of the underlying biological mechanisms and the specific lipid species driving the observed differences between SCC and control samples.

Furthermore, the prediction background plot (Figure 4) provides a visual representation of how individual samples were classified by the sPLS-DA model in both groups. This visual representation further confirms the effectiveness of the model in accurately discriminating between SCC and control samples.

A comparison of the key lipids identified in the primary and validation groups revealed a high degree of consistency. Eight key lipids consistently demonstrated significant upregulation in SCC tumor tissues compared to control tissues, highlighting their robust association with the disease. These lipids were PI(16:0/22:4), PI(18:1/20:4), PE(16:0/20:4), PE(16:0/22:5), PE(16:0/22:6), PE(18:1/20:3), PC(18:1/20:2), and PC(18:2/20:2). This finding provided evidence for the potential role of these specific lipids in SCC pathogenesis and their potential utility as biomarkers for diagnosis, prognosis, or therapeutic targeting (Table 2).

To gain deeper insights into the biological implications of the observed lipidomic changes, pathway enrichment analysis was conducted. This analysis revealed a significant involvement of glycerophospholipid metabolic pathways in SCC. Specifically, alterations were identified in diacylglycerophosphocholines metabolism, glycerophosphoethanolamines metabolism, glycerophosphoinositols metabolism, and diacylglycerophosphoinositols metabolism (Figure 5).