The data obtained from The Cancer Genome Atlas (TCGA) (https://cancergenome.nih.gov/, accessed on 01/04/2024) and Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/, accessed on 01/04/2024) repositories for this study included RNA-seq data and detailed clinical information. The training cohort included 177 PAAD patients from TCGA, and the validation cohort included 65 PAAD patients from the GSE62452 dataset. Pan-cancer datasets were downloaded from TCGA.

According to previously published studies [11–13], we conducted our research using a comprehensive set of 332 genes involved in lactylation (Table S1).

We applied consensus clustering analysis to categorize PAAD into distinct LacClusters, and LacClusters-enhanced subgroups according to prognosis-associated lactylation-related genes and prognosis-related differentially expressed genes (DEGs) between the two LacClusters, respectively.

Prognosis-related DEGs were analyzed using least absolute shrinkage and selection operator (LASSO) regression analysis and multiple Cox regression analysis. Finally, three genetic factors crucial to the development of the prognostic model were identified, namely, peptidylprolyl isomerase A (PPIA), LRRC1, and aryl hydrocarbon receptor nuclear translocator 2 (ARNT2). The lactylation index of three genes (LacI-3) was calculated using the following formula: LacI-3 = PPIA expression × (0.007376) + LRRC1 expression × (0.033445) + ARNT2 expression × (−0.033908). According to the median of LacI-3 score, patients from the training and validation sets were categorized into LacI-3^high and LacI-3^low groups.

The R package “GSEA” (version 4.0.1) was used for enrichment analysis to explore the heterogeneity of glycolysis-related pathways. The gene sets “HALLMARK_GLYCOLYSIS. gmt” and “REACTOME_GLYCOLYSIS. gmt” were the reference gene sets downloaded from GSEA (https://www.gsea-msigdb.org/gsea/msigdb/human/search.jsp, accessed on 01/04/2024). Lactylation scores were calculated based on the 332 previously identified lactylation-related genes using single-sample GSEA (ssGSEA).

The R package “pRRophetic” was utilized for conducting the drug sensitivity analysis. Differential analysis and correlation analysis of drug sensitivity were performed in the LacI-3^high and LacI-3^low groups.

The mutation annotation format (MAF) was created utilizing the “maftools” package, and somatic mutations in PAAD were plotted. The TMB was assessed for every patient with PAAD in the TCGA dataset.

xCell was used to calculate the immune cell scores of PAAD patients [14]. The infiltration levels of 29 immune cells were estimated using ssGSEA [15,16]. The immune scores, ESTIMATE scores, and stromal scores were calculated using the ESTIMATE algorithm [17]. The scores of 22 immune cell types in PAAD were calculated using CIBERSORT [18].

The HPDE6-C7 cell line derived from normal human pancreatic ductal epithelial cells was obtained from Jennio Biotech Co., Ltd. (Guangzhou, China). AsPC-1 and PANC-1 cell lines derived from human PAAD were acquired from the Chinese Academy of Sciences Cell Bank (Shanghai, China). The cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco, C11995500BT, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS) (Gibco, A3160902, Carlsbad, CA, USA) and 1% penicillin-streptomycin solution (Beyotime Biotechnology, C0222, Shanghai, China). The cells were incubated at 37°C and 5% CO₂ with saturating humidity in a cell culture incubator (Thermo Scientific, 411, Wilmington, DE, USA). We used L-(+)-lactic acid (MACKLIN, L812422, Shanghai, China) or the lactate dehydrogenase inhibitor galloflavin (MCE, HY-W040118, New Jersey, USA) to regulate lactylation levels in cells. The validity of each cell line was verified using short tandem repeat (STR) profiling.

PANC-1 and AsPC-1 cells were plated in a 6-well plate at 60%–70% confluence and transfected with LRRC1-specific short interfering RNAs (siRNAs) or control siRNA purchased from Tsingke Biotechnology (Beijing, China) accompanied by Lipofectamine™ RNAiMAX (Thermo Fisher Scientific, 13778150, Waltham, MA, USA). The sequences of the hLRRC1 siRNAs used were as follows: si1 sense, 5′-GCUUGGACUUAGUGAUAAUTT-3′; si1, antisense 5′-AUUAUCACUAAGUCCAAGCTT-3′; si2 sense, 5′-GGAACUGAGAGAGAAUCUUTT-3′; and si2, antisense 5′-AAGAUUCUCUCAGUUCCTT-3′.

Total RNA was isolated from HPDE6-C7, AsPC-1, and PANC-1 cells using RNAiso Plus (TaKaRa, 9108, Otsu, Shiga, Japan) and then quantified and reverse transcribed (TaKaRa, RR047A, Otsu, Shiga, Japan). Next, LRRC1 and GAPDH were analyzed using a SYBR PrimeScript PCR Kit II (TaKaRa, RR820A, Otsu, Shiga, Japan) and quantified using a Bio-Rad CFX96 detection system (Bio-Rad, 1855195, Hercules, CA, USA). The following primers were utilized: hLRRC1 forward, 5′-CAGACTAACTCGGATACCTGCAG-3′; hLRRC1 reverse, 5′-CTGGTTGTCAGATAGCCACAGAG-3′; hGAPDH forward, 5′-CAATGACCCCTTCATTGACC-3′; and hGAPDH reverse, 5′-GACAAGCTTCCCGTTCTCAG-3′. Using GAPDH as an internal control, the quantification of gene expression was determined using the comparative threshold (Ct) cycle approach.

The cells were disrupted using RIPA buffer (Beyotime Biotechnology, P0013B, Shanghai, China), and the total protein concentration was measured using a BCA kit (Thermo Fisher Scientific, Richmond, 23225, NY, USA). Equivalent quantities of protein were separated using 10% SDS‒PAGE, transferred onto polyvinylidene fluoride (PVDF) membranes (Bio-Rad, 1620177, Hercules, CA, USA) and incubated with anti-KLA (1:1000; PTM BIO, PTM1401RM, Hangzhou, China) and anti-β-actin (1:10000; ABclonal, AC026, Wuhan, China) antibodies at 4°C overnight. Subsequently, the membranes were exposed to secondary antibodies were conjugated to HRP (1:3000; Cell Signaling Technology, 7074S, Danvers, MA, USA), and the bands were visualized with SuperSignal^TM West Femto (Thermo Fisher Scientific, 34096, Waltham, MA, USA) and a FUSION FX VILBER LOURMAT6 (VILBER, FX6-XT, Paris, France). β-Actin was used as an internal reference in this study.

The cells (1 × 10⁶) were stained, washed, and resuspended using an apoptosis kit (Beyotime Biotechnology, C1062L, Shanghai, China) following the instructions outlined by the manufacturer. Finally, the cells were detected on a flow cytometer (BD, FACSAria II, Franklin Lakes, NJ, USA). The data were analyzed utilizing Flow Jo (version 7.6.1, Stamford, USA).

AsPC-1 and PANC-1 cells treated with lactic acid or galloflavin were sorted into the high-lactylation group and the low-lactylation group, respectively. Different concentrations of tamoxifen (Selleck, S1238, Houston, Texas, USA) or doxorubicin (Selleck, E2516, Houston, Texas, USA) was added to a 96-well plate containing three groups of cells, with 10⁴ cells per well. After 24 h, cell counting kit 8 (CCK8) reagent (Beyotime Biotechnology, C0037, Shanghai, China) was added to the plate (10 μL CCK8 + 90 μL DMEM). The plate was subsequently incubated in the dark for 2 h, after which the optical density (OD) was measured at 450 nm (Thermo Fisher Scientific, VLB000D0, Waltham, MA, USA). Cytotoxicity (%) was calculated by comparing the OD of the sample and control. Each sample was tested in three replicates.

After LRRC1 knockdown, AsPC-1 and PANC-1 cells were sorted into the siLRRC1-1 group and the siLRRC1-2 group, respectively. Three groups of cells were seeded into 96-well plates, with each well containing 2000 cells. CCK8 assays were performed daily for 7 consecutive days. Three replicates were used for each sample. Proliferation (%) was calculated by comparing the OD of the sample and control.

The cell culture insert (pore size 8 μm, Corning, 353097, NY, USA) was placed into a 24-well plate. In the upper chamber of the cell culture insert, a total of 8 × 10⁴ cells were seeded with 200 µL of FBS-free DMEM. Next, 600 µL of DMEM supplemented with 20% FBS was added to the 24-well plate. After an incubation period of 24 or 36 h, the cells were fixed and subjected to crystal violet staining (Beyotime Biotechnology, C0121, Shanghai, China). Images of the cells that had invaded/migrated were taken, and the number of cells was quantified using ImageJ (version 1.52, National Institutes of Health, Bethesda, MD, USA). The bottom of the cell culture insert was covered with Matrigel (Biocoat, 354277, Commonwealth of Pennsylvania, USA) in the invasion experiment.

Individual cell suspensions were created in DMEM containing 10% FBS. Four thousand cells/well were inoculated in each group of 6-well plates and gently shaken. The cells were incubated for approximately 7–14 days (culture termination when visible clones appeared). Next, the cells were treated with fixative and subjected to crystal violet staining (Beyotime Biotechnology, C0121, Shanghai, China). Images of the cells were captured, and the quantity of cells was assessed utilizing ImageJ (version 1.52, National Institutes of Health, Bethesda, MD, USA).

Six-well plates were used to culture the cells. Within 24 h, the cell density reached 90%. Scratches on the cells were generated using a pipette tip, and images of the cells at the scratch were taken at 0, 24, and 36 h. The migration area was quantified using ImageJ (version 1.52, National Institutes of Health, Bethesda, MD, USA). The cell migration rate (wound healing rate) was calculated as follows: (initial scratch area – scratch area at a certain moment)/initial scratch area × 100%.

Quantitative data analysis and differential expression analysis were conducted using ImageJ (version 1.52, National Institutes of Health, Bethesda, MD, USA) and GraphPad Prism (version 8.0, La Jolla, CA, USA). The data obtained from flow cytometry analysis were processed using Flow Jo software (version 7.6.1, Stamford, USA). For normally distributed data with uniform variance, the comparison between two groups was conducted using a two-tailed independent-sample t test, while comparison among multiple groups was carried out using one-way ANOVA. All experiments were independently replicated three times. A significance threshold of p < 0.05 was considered to indicate statistical significance. Survival curves were estimated using the Kaplan‒Meier method, and the divergence between curves was evaluated using the log-rank test. The median LacI-3 concentration was determined as the cutoff. For differential expression analysis, the “limma” (version 3.20.9) package was utilized. Cox regression analysis was utilized for both single-variable and multiple-variable analysis, with factors exhibiting a p value less than 0.05 in the single-variable analysis being incorporated into the multiple-variable analysis. “*” indicates the p value in the article. *p < 0.05; **p < 0.01; ***p < 0.001; and ****p < 0.0001.

Previous research revealed 332 lactylation-related genes [11–13] (Table S1). Using univariate Cox regression analysis, 103 genes related to lactylation with prognostic significance were identified. The forest plot shows the top 20 genes with the lowest p values (Fig. S1A and Table S2). To visualize the somatic mutations in genes associated with lactylation, a waterfall plot was generated, and Kirsten rat sarcoma viral oncogene homolog (KRAS), which is among the most frequently occurring oncogenes in humans, was shown to have the highest somatic mutation rate (Fig. S1B). The degree of amplification and deletion of the chromosome copy number variation (CNV) is shown in Fig. S1C, which exhibited the most significant variation in terms of gene expression among the top 20 genes. The position of the CNV is shown in Fig. S1D.

The relationships among lactylation, patient prognosis, and the TIME landscape were explored. Using consensus clustering analysis, PAAD patients (TCGA, n = 177) were divided into 2 LacClusters according to the expression of 103 prognostic lactylation-related genes (Fig. 1A). The results of principal component analysis (PCA) demonstrated effective stratification (Fig. 1B). The survival analysis results indicated that PAAD patients in LacCluster A exhibited a favorable prognosis, whereas patients in LacCluster B had a poor prognosis (p = 0.038; Fig. 1C). Most lactylation-related genes were significantly upregulated in LacCluster B, indicating that lactylation was relatively active. In contrast, LacCluster A exhibited decreased lactylation. A heatmap was constructed to visualize the expression levels of the top 20 lactylation-related genes, which indicated significant differential expression in the different LacClusters and clinicopathological features (Fig. 1D). In conclusion, lactylation was correlated with the prognosis and various clinicopathological characteristics of PAAD patients.

Although 332 genes are associated with lactylation [11–13], there are likely many genes related to lactylation that have not been discovered because research on lactylation is still in its infancy. Therefore, we conducted differential expression analysis between the 2 LacClusters to identify additional potential genes related to lactylation. A total of 2167 DEGs (p < 0.001, |log₂FC| > 1.5) were identified between the 2 LacClusters. And then a total of 632 DEGs related to prognosis were detected (Table S3). Then, 177 patients were divided into 2 LacCluster-enhanced cohorts according to the 632 prognosis-related DEGs (Fig. 2A). PCA showed good stratification results (Fig. 2B). Compared with LacCluster-enhanced B patients, LacCluster-enhanced A patients exhibited a significant increase in survival (p = 0.037; Fig. 2C). A comparison of the ROCs of the LacCluster and LacCluster-enhanced cohorts revealed that the LacCluster-enhanced cohort had an advantage in predicting the 3- and 5-year overall survival (OS) of PAAD (LacCluster: LacCluster-enhanced; 1-year: 0.627 vs. 0.620; 3-year: 0.550 vs. 0.573; 5-year: 0.488 vs. 0.532) (Fig. S2A,B). A heatmap was constructed to visually represent the expression of the top 20 prognostic DEGs with significant differences in the different LacCluster-enhanced cohorts and clinicopathological features (Fig. 2D).

We additionally examined the levels of immune checkpoint molecules and the presence of infiltrating immune cells within different enhanced LacClusters. Interestingly, most immune checkpoint genes, such as IDO1, CTLA4, TIGIT, PDCD1, HAVCR2, LAG3, ICOS, CD27, ADORA2A, TNFRSF8, CD200R1, BTLA, KIR3DL1, CD48, IDO2, CD200, BTNL2, CD40LG, CD28, and TNFSF14, were highly expressed in LacCluster-enhanced A samples, indicating that these patients may benefit from the corresponding monoclonal antibody immunotherapy (Fig. 2E,F). Moreover, many pathways related to immune-related functions, including T-cell coinhibition, T-cell costimulation, checkpoint, inflation promotion, cytotoxic activity, type II interferon (IFN) response, cytokine and cytokine receptor (CCR), and human leukocyte antigen (HLA), were highly activated in the LacCluster-enhanced A samples, which exhibited high levels of antigen presentation and antitumor immunity (Fig. 2F). Furthermore, there was an increase in the majority of infiltrating immune cells in the LacCluster-enhanced A group, which was opposite to the trend observed for lactylation; these cells included various dendritic cells (DCs) (aDCs, cDCs, and DCs), various B cells (B cells, naïve B cells, and memory B cells), various T cells (naïve CD8 T cells, CD8 T cells, memory CD4 T cells, naïve CD4 T cells, and effector memory CD4 T cells), macrophages (M1 and M2), mast cells, and monocytes (Fig. 2F,G). These findings suggested that lactylation was greater in “immune cold tumors” and indicated an unfavorable prognosis.

To improve the clinical application of our model, we performed LASSO regression analysis on 632 prognosis-related DEGs, which were lactylation-related genes, in 177 PAAD patients whose prognosis was the output signal. A prognostic prediction model containing three lactylation-related genes, named LacI-3, was established as follows: LacI-3 = PPIA expression × (0.007376) + LRRC1 expression × (0.033445) + ARNT2 expression × (−0.033908) (Fig. S3A,B). Cox regression analysis indicated that PPIA independently contributed to prognosis (Fig. S3C). Next, we conducted internal and external validation of this model. For internal validation, the TCGA dataset was split into test 1 and test 2 sets at a ratio of 4:6. The findings indicated that the model had prognostic significance in both sets (Fig. S4A,B) and could predict OS (Fig. S4C–F).

For external validation, we obtained GSE62452 from the GEO database. Survival analysis indicated that, compared with those in the LacI-3^high subgroup in TCGA (training) cohort (Fig. 3A) and GEO (validation) cohort (Fig. 3B), the survival advantage in the LacI-3^low subgroup was significant (p = 0.001 and p = 0.046, respectively). TCGA cohort demonstrated areas under the ROC curves (AUCs) of 0.716 for 1 year, 0.673 for 3 years, and 0.749 for 5 years (Fig. 3C). The AUCs for the GSE62452 cohort at 1 year, 3 years, and 5 years were 0.559, 0.700, and 0.730, respectively (Fig. 3D). ROC curve analysis revealed the outstanding performance of LacI-3 in predicting the OS of PAAD patients. PCA revealed identifiable differences between the LacI-3^high and LacI-3^low groups in the cohorts used for training and validation, respectively (Fig. S5A,B). In addition, the LacI-3^high group had a shorter progression-free survival (PFS) compared with the LacI-3^low group, indicating that the LacI-3^high group was more prone to tumor progression (Fig. 3E). To assess the predictive reliability of the LacI-3 score and clinicopathological characteristics, we plotted the 1-year ROC curve and calculated the AUC. LacI-3 had the highest AUC, which was 0.716 (Fig. 3F). To explore the potential of the LacI-3 level as an independent prognostic indicator, we initially performed univariate Cox regression analysis, which revealed that the LacI-3 level, age, and N stage were significant prognostic factors (Fig. S6A). Furthermore, multivariate Cox analysis indicated that age, N stage, and LacI-3 were independent prognostic factors (Fig. S6B). These findings suggest that the LacI-3 could serve as a predictive indicator for the survival prognosis of patients with PAAD.

The prognostic efficacy of the three prediction models was compared (Fig. 3G). The LacCluster, LacCluster-enhanced, and LacI-3 clusters showed strong consistency. Compared with those of the LacCluster B (Fig. 3H) and LacCluster-enhanced B (Fig. 3I) cohorts, the LacI-3 levels of the LacCluster A and LacCluster-enhanced A patients were significantly lower (p < 0.001). To further explore the functional attributes of LacI-3, we conducted GSEA of the LacI-3^high and LacI-3^low groups. The results indicated that the LacI-3^high subgroup exhibited enrichment in pathways related to glycolysis, such as the HALLMARK_GLYCOLYSIS (Fig. 3J) and REACTOME_GLYCOLYSIS (Fig. 3K) pathways, which confirmed the positive correlation between LacI-3 and lactylation. Finally, we calculated the lactylation score (lacscore) using ssGSEA of TCGA-PAAD samples, and the analysis results indicated that the LacI-3^high subgroup exhibited a greater lacscore, confirming the positive correlation between LacI-3 and lactylation (p < 0.001; Fig. 3L).

In conclusion, the above results suggested that the LacI-3 could serve as a prognostic factor independent of other variables for PAAD.

The greater the number of mutations, the greater the number of neoantigen types with immunogenicity and the greater the chance of triggering T-cell responses [19]. A positive correlation was identified between the LacI-3 level and the TMB using spearman correlation analysis (R = 0.49, p < 0.001; Fig. 4A). The TMB in the LacI-3^high subgroup was greater than that in the LacI-3^low subgroup (p < 0.001; Fig. 4B). Patients exhibiting high TMB experienced worse outcomes than those with low TMB (p = 0.008; Fig. 4C). After the LacI-3 level was integrated with the TMB, analysis of survival outcomes demonstrated that patients in the TMB^highLacI-3^high subgroup had the worst prognosis. Patients in the TMB^lowLacI-3^low, TMB^highLacI-3^low subgroups had better prognoses. Unexpectedly, patients in the TMB^lowLacI-3^high subgroup had the best prognosis, possibly due to the small sample size (only four patients in the TMB^lowLacI-3^high subgroup) (p < 0.001; Fig. 4D). Moreover, compared with patients in the LacI-3^low subgroup, patients in the LacI-3^high subgroup had a significantly greater frequency of somatic mutations, especially in KRAS (81% vs. 40%), TP53 (68% vs. 43%), SMAD family member 4 (SMAD4) (24% vs. 19%), cyclin-dependent kinase inhibitor 2A (CDKN2A) (26% vs. 9%), and titin (TTN) (14% vs. 9%; Fig. 4E,F). In summary, high TMBs, high LacI-3 values, and high frequencies of somatic mutations indicated poor prognoses in PAAD patients.

ESTIMATE immune infiltration analysis revealed significant correlations between low LacI-3 expression and high immune, stromal, and ESTIMATE scores, indicating an increase in the abundance of infiltrating immune cells (Fig. 5A). To delve deeper into the association between LacI-3 and the TIME, an examination was conducted to scrutinize variances in immune checkpoints and immune cells. The LacI-3^low subgroup exhibited increased expression of immune checkpoint genes, including TIGIT, PDCD1, HAVCR2, ICOS, BTLA, CTLA4, and IDO2; the above results revealed the potential benefits of immune checkpoint inhibitor treatment (Fig. 5B). Moreover, the LacI-3^low group exhibited active signal transduction and antitumor effects, including TIL, HLA and T-cell coinhibitory effects, T-cell costimulation, cytolytic activity, and a type II IFN response (Fig. 5C). In addition, the results of the xCell algorithm showed that various immune cells related to immune-mediated killing, such as various DCs, B cells, T cells, macrophages, mast cells, and monocytes, were highly abundant in the LacI-3^low group (Figs. 5C,D). CIBERSORT immune infiltration analysis revealed that plasma cells, resting DCs, monocytes, activated memory CD4 T cells, and CD8 T cells were negatively correlated with LacI-3 (Fig. 5E). In summary, the LacI-3^low subgroup is characterized by a “hot” immune phenotype associated with high levels of antitumor immune cell infiltration, indicating a good response to various treatment options, especially immunotherapy.

To explore the significance of this scoring system in guiding the clinical treatment of PAAD patients, we performed a drug sensitivity assessment to determine the half-maximal inhibitory concentration (IC50) of 251 chemotherapy drugs in the LacI-3^high and LacI-3^low groups. There were 121 drugs with differences in IC50 between the LacI-3^high and LacI-3^low groups. Patients in the LacI-3^high subgroup exhibited sensitivity to 40 drugs, while patients in the LacI-3^low subgroup exhibited sensitivity to 81 drugs (Table S4). Various drugs commonly used in clinical tumor treatment, such as tamoxifen (Fig. 6A) and doxorubicin (Fig. 6B), were included. Tamoxifen showed greater efficacy in patients with low LacI-3, whereas doxorubicin demonstrated greater effectiveness in patients with high LacI-3. We treated AsPC-1 and PANC-1 cells with lactic acid to increase lactylation and treated them with the lactate dehydrogenase inhibitor galloflavin to reduce lactylation. WB analysis revealed that the control group exhibited moderate levels of lactylation, the lactic acid-treated group displayed high levels of lactylation, and the galloflavin-treated group demonstrated low levels of lactylation, confirming that we successfully constructed three groups of cells (Fig. 6C). The IC50 values of tamoxifen and doxorubicin were compared among the three groups. The IC50 of tamoxifen in the low-lactylation group was the lowest (AsPC-1: 12.42 μM, PANC-1: 7.33 μM), and that in the high-lactylation group was the highest (AsPC-1: 34.42 μM, PANC-1: 18.98 μM). However, the IC50 of doxorubicin in the low-lactylation group was the highest (AsPC-1: 27.17 μM, PANC-1: 19.52 μM), and that in the high-lactylation group was the lowest (AsPC-1: 10.48 μM, PANC-1: 5.29 μM) (Fig. 6D). The results of the apoptosis experiments indicated that the low-lactylation group increased the percentage of apoptotic cells after tamoxifen treatment, whereas the high-lactylation group increased the percentage of apoptotic cells after doxorubicin treatment (Fig. 6E,F). In conclusion, these findings offer a basis for categorizing PAAD patients based on treatment stratification.

In the scoring system, we analyzed the expression of hub genes and observed that LRRC1 was significantly overexpressed in PAAD tumor tissues compared with adjacent tissues. This finding was validated at the protein level using immunohistochemistry (IHC) data downloaded from the Human Protein Atlas (HPA) website (https://www.proteinatlas.org/, accessed on 01/04/2024) (Figs. S7B–D). PPIA and LRRC1 expression predicted poor patient prognosis (Fig. S8A,B). ARNT2 expression predicted a good prognosis (Fig. S8C). The ROC curve indicated that LRRC1 had AUC values above 0.6 at 1 year, 3 years, and 5 years; thus, LRRC1 plays a crucial role in predicting the outcome of patients with PAAD (Fig. S8D–F). In addition, increased expression of PPIA and LRRC1 was related to increased mortality (Fig. S8G) and decreased disease-specific survival (DSS) (Fig. S8H) and PFS (Fig. S8I). Nevertheless, no notable differences in the mRNA or protein levels of the prognostic genes PPIA and ARNT2 were detected between cancer tissues and adjacent tissues; thus, these genes had low predictive value for prognosis (Figs. S7A,C,D, S8D–F).

In addition, PPIA is a known lactylation-related gene, whereas the LRRC1 and ARNT2 genes identified in this study have not previously been shown to be related to lactylation. The 177 patients with PAAD were grouped into two categories based on the median of LRRC1. According to the results of the differential expression analysis, 226 genes associated with lactylation exhibited a high level of expression in the high-LRRC1 subgroup (Fig. S9A).

Previous research has shown that the E1A binding protein P300 (EP300) is a lactylase (writer) and that histone deacetylase 1-3 (HDAC1-3) and sirtuin1-3 (SIRT1-3) are delacetyltransferases (erasers) [11,12]. Correlation analysis revealed positive correlations between LRRC1 and EP300, HDAC1-3, and SIRT1 (Fig. S9B). After analyzing the data provided, it can be inferred that the LRRC1 gene may be associated with lactylation, indicating the importance of conducting additional research on the role of LRRC1.

Due to limited research on LRRC1, we conducted functional phenotype experiments on LRRC1 in vitro. First, LRRC1 mRNA expression was assessed in both normal pancreatic cells and PAAD cells using RT‒qPCR. The findings showed that LRRC1 expression in AsPC-1 cells was 11-fold greater than that in HPDE6-C7 cells and that in PANC-1 cells was 2-fold greater than that in HPDE6-C7 cells (Fig. 7A). Next, LRRC1 was knocked down by more than 50% in AsPC-1 (Fig. 7B) and PANC-1 (Fig. S10A) cells using siRNA, and the knockdown efficiency was determined with RT‒qPCR. To investigate the role of LRRC1 in PAAD, we conducted experiments on cell proliferation, invasion, migration, and apoptosis. Proliferation experiments showed that knocking down LRRC1 significantly inhibited AsPC-1 (Fig. 7C) and PANC-1 cell proliferation (Fig. S10B). The colony formation experiment showed that knocking down LRRC1 significantly inhibited the colony formation of AsPC-1 (Fig. 7D) and PANC-1 cells (Fig. S10C). Wound healing and transwell experiments showed that knocking down LRRC1 reduced AsPC-1 (Figs. 7E and 7F) and PANC-1 cell migration and invasion (Fig. S10D,E). The flow cytometry results using PI/Annexin V-FITC indicated that knocking down LRRC1 led to an increase in both early and late apoptosis in AsPC-1 (Fig. 7G) and PANC-1 cells (Fig. S10F). In summary, our findings confirmed that LRRC1 enhances PAAD cell invasion, migration, proliferation, and colony formation while suppressing apoptosis in vitro.

To demonstrate the clinical value of LacI-3, we conducted a pan-cancer analysis. Prognostic analysis was also conducted for 32 types of tumors, excluding PAAD, utilizing data downloaded from TCGA (Fig. 8, Fig. S11). The results showed that LacI-3 was predictive of OS in 12 types of cancers (p < 0.05), such as breast invasive carcinoma (BRCA), lung adenocarcinoma (LUAD), rectum adenocarcinoma (READ), sarcoma (SARC), mesothelioma (MESO), thymoma (THYM), kidney clear cell carcinoma (KIRC), colon adenocarcinoma (COAD), uterine corpus endometrial carcinoma (UCEC), liver hepatocellular carcinoma (LIHC), uveal melanoma (UVM), and lung squamous cell carcinoma (LUSC) (Fig. 8). A high level of LacI-3 indicated a poor prognosis in BRCA, SARC, UCEC, LIHC, and LUAD patients but indicated a good prognosis in COAD, READ, MESO, THYM, KIRC, UVM, and LUSC patients. These findings indicated the potential for lactylation to exert varying effects on distinct types of tumors. The xCell algorithm was used for the 12 types of cancers mentioned above, and the results showed that for COAD, READ, LIHC, LUAD, and LUSC patients, LacI-3 was negatively associated with immune cell infiltration, whereas LacI-3 was positively associated with immune cell infiltration in UCEC patients (Fig. S12). In summary, LacI-3 can predict prognosis and immune infiltration in cancer patients.