Plasma specimens collected from 50 to 90-year-old subjects with PDAC (N=39) obtained from the Herbert Irving Comprehensive Cancer Center Database Shared Resource (DBSR) and from 50 to 94-year-old control individuals without cancer purchased from Precision for Medicine (Norton, MA, USA), were used in this study. The mean age for both cohorts was 68 years old; 70% were male; 35% had metastasis at diagnosis. We collected 30 plasma samples from patients with type 2 Diabetes (DM2) and 12 plasma samples from patients with advanced PDAC (PDAC-PT) to be tested as a control cohort. For the DM2 patients, we requested blind samples with no information regarding age or sex; the only inclusion criterion was an HbA1c level above 7%. For the PDAC-PT the mean age was 71 years old, and the samples were collected from patients with advanced disease ~55 days after initial treatment. All samples were de-identified according to the Protected Health Information regulations. The study was approved by the Columbia University Institutional Review Board. Subject information is listed in Supplementary Tables S1A–C .

IgG reactivity to 93 chemical adducts was assessed using a modified version of a previously described ELISA platform (22). The ELISA panel includes: (i) forty-six synthetic 5-mer peptides, each composed of a modified amino acid flanked by four unmodified arginine, two on each side (5-mer peptides with matching unmodified residues were included in the panel as controls); (ii) 6 modified amino acids; (iii) 8 PTM compounds (e.g., small ubiquitin related modifier (SUMO) and ubiquitin); and (iv) 33 cofactors, coenzymes, and metabolites (e.g., coenzyme Q10). The panel also includes MDA–bovine serum albumin (BSA), insulin, and LPS as controls. MDA-modified BSA and MDA-lysine/arginine peptides were generated by incubating acid-hydrolyzed 1,1,3,3-tetramethoxypropane (Sigma-Aldrich, St. Louis, MO, USA) with BSA or lysine/arginine peptides. Briefly, 2 M 1,1,3,3-tetramethoxypropane was hydrolyzed in 96mMHCl for 15 min at 37°C and then neutralized with NaOH. BSA (2 mg/ml) or lysine/arginine peptides (10 mg/ml) were incubated with 0.2 M MDA for 3 hours at 37°C. Extensive dialysis against 1× phosphate-buffered saline (PBS) was performed at 4°C for 36 hours. All adducts and controls included in the panel are detailed in Supplementary Table S2 together with their sources and working concentrations.

IgG reactivity to the 93 adducts included in the panel was quantified by ELISA as follows: Corning™ Clear Polystyrene 96-Well EIA/RIA Microplates (Corning Incorporated) were coated in duplicate with adducts, compounds, and control antigens at 4°C for 20 hours. Synthetic peptides with modified amino acids, large PTM compounds, and small compounds were coated at 10 μM, 1 μM, and 1 mM, respectively ( Supplementary Table S2 ). ELISA plates were washed three times in PBS Tween (PBST) and then blocked with 3% BSA (Fisher Scientific Inc.) in PBST for 2 hours at 37°C. Plasma diluted in PBST at 1:200 for IgG was distributed to the plate and incubated for 3 hours at room temperature. Plates were then washed five times in PBST and incubated with horseradish peroxidase (HRP)–conjugated anti-human IgG affinity-purified F(ab′) ₂ fragment - Host Donkey IgG; Polyclonal; Dilution 1:4000 - (Jackson ImmunoResearch Laboratories Inc., West Grove, PA) diluted in blocking buffer for 1 hour. Plates were then washed again, and HRP activity was developed with 3,3′,5,5′-tetramethylbenzidine (TMB; Fisher Scientific Inc.). Reactivity to adducts was calculated and reported as an optimal density (OD). To determine the subclasses of plasma IgG in PDAC patients, we used anti-IgG1, anti-IgG2, anti-IgG3, or anti-IgG4 as secondary antibodies (Clone 4E3, HP6002, HP6050 and HP6025 respectively; Southern Biotech, Birmingham, AL - Dilution 1:4000).

Immunofluorescence staining was performed to detect phosphoryl-tyrosine in tissue using the antibody P-Tyr-1000 (Cell signaling- Host Rabbit mAb mix, dilution 1:100). Thirteen out of the 39 PDAC cases were evaluated including 11 tumor tissue paired with adjacent non-tumoral tissue; 2 tumor tissue without adjacent non-tumoral tissue were also available. Additionally, 4 pancreatic tissue specimens without cancer and 1 tonsil specimen were used as controls. In brief, formalin-fixed paraffin-embedded (FFPE) tissue blocks were deparaffinized with xylene and rehydrated in sequential steps with solutions containing decreasing concentrations of ethanol (100%, 95% and 70%). A DIVA Decloaker (BIOCARE MEDICAL) was used to perform antigen retrieval and slides were then blocked with 3% BSA (Fisher Scientific Inc.) for 1 hour at room temperature. Slides were subsequently incubated overnight at 4°C with anti-P-Tyr mab, followed by 3 washes with PBS 1x. Slides were then incubated for 1 hour at room temperature with a secondary anti-rabbit IgG (H+L), F(ab’)2 fragment antibody conjugated to Alexa Fluor^® 488, followed again by 3 washes with PBS 1x Lastly, slides were mounted with Vectashield Vibrance^® Antifade Mounting Medium with DAPI (H-1800) (Vector laboratories). Images were acquired with a Zeiss LSM 900 of microscope.

Flash-frozen tumor specimens and paired adjacent non-tumoral tissue from 3 PDAC cases were used to analyze the phosphoproteome by liquid chromatography/mass sprectrometry (LC-MS/MS) using Creative Proteomics services (New York, USA). Samples were first removed from -80°C and homogenized in 1 mL lysis buffer (8 M urea, 100 mM tris-HCl, pH 8.0, 1% protease inhibitor, 1% phosphatase inhibitor) by beads grinder. All cell debris were removed by centrifugation at 20,000g for 15 min. The protein concentration was then determined using a BCA assay. Approximately 1 mg protein was used for each sample. Disulfide bridges were reduced by 10 mM TCEP at 56°C for 1 h. Reduced cysteine residues were then alkylated by 20 mM iodoacetamide (IAA) in the dark at room temperature for 30 min. The solution was then subject to overnight digestion at 37°C with trypsin (Promega) using an enzyme to substrate ratio of 1:50 (w/w). After digesting, TFA was added to 1% final concentration. Any precipitate was removed by centrifugation at 1,780g for 15 min. Peptide purification was then performed at room temperature on C18 reversed-phase columns. The columns were first conditioned by 100% ACN followed by 0.1% TFA, 80% ACN and then equalized by 0.1% TFA. The acidified and cleared digest were then loaded onto column. After washing by 0.1% TFA, the peptides were eluted from the column by 0.1%TFA, 80% ACN. 10% of the eluent was used as a total proteome sample. The remaining 90% eluent was then processed to phosphopeptide enrichment using Fe-IMAC beads according to manufacturer’s protocol. Briefly, the beads were first resuspended and washed in washing buffer 3 times, and then incubated with peptide solution on a rotator for 30 min. After incubation, the beads were washed 3 times with washing buffer, and 50 μL elution buffer was added to elute the phospho-peptides from the beads. Forty μL 20% TFA was added into the eluate to acidify the eluate. The phospho-peptides were purified again and proceeded to nano LC-MS/MS analysis using an Ultimate 3000 nano UHPLC system (ThermoFisher Scientific, USA), trapping nanocolumn (PepMap C18, 100Å, 100 μm × 2 cm, 5 μm) and an analytical column (PepMap C18, 100Å, 75 μm × 50 cm, 2 μm). The full scan was performed between 300-1,650 m/z at the resolution 60,000 at 200 m/z, the automatic gain control target for the full scan was set to 3e6. The MS/MS scan was operated in Top 20 mode using the following settings: resolution 15,000 at 200 m/z; automatic gain control target 1e5; normalized collision energy at 28%.

Raw MS files were analyzed and searched against Homo sapiens protein database (UP000005640) using Maxquant (version 2.3.0.0). The parameters were set as follows: the protein modifications were carbamidomethylating (C), oxidation (M), Phospho (STY); the enzyme specificity was set to trypsin; the maximum missed cleavages were set to 2; the precursor ion mass tolerance was set to 10 ppm, and MS/MS tolerance was 0.5 Da. The analysis was conducted on paired control and tumor samples from 3 patients with PDAC. Differential Phosphorylation was computed using the msqrob2 package: raw intensities were loaded. Peptides with zero intensities were converted to NA. The raw intensities were then log2 normalized and a density plot was generated. Peptides that were only identified in one sample were dropped. The filtered data was then subjected to quantile normalization. A paired differential analysis was performed on the quantile normalized values using Limma package. Cutoff values of 0.05 for the p value and log fold-change < -0.5 or > 1 were applied for selection of a robust list of the phosphorylated differential sites. A heatmap was used to depict protein sites with excess phosphorylation in tumor cells. The quantile normalized values were scaled to make the heatmap. R stats (R version 4.3.3) was used to perform all the analyses. Packages used and reference are reported in Supplementary Table S3 . The genes corresponding to the phosphorylated proteins were annotated with EntrezIDs using Biomartr package. Pathway analysis was performed using the EntrezIDs and the clusterProfiler package to identify significantly enriched GO terms. Note: We didnt use the FDR p value. We just used the un-adjusted P value. A significance threshold (p < 0.05) was applied, and multiple testing correction methods were employed to control the false discovery rate. Enriched GO terms for biological processes, cellular components and molecular functions were visualized using the R packages reported in Supplementary Table S3 .

For feature selection, the standardized ELISA readings were subjected to random forest classification using the Boruta package. Confirmed features were then subjected to dimensionality reduction using principal component analysis from tidymodels’ step_pca function. The first three principal components were then visualized using plotly. All the packages used are reported in Supplementary Table S3 . Unpaired t test with Welch’s correction, Pearson’s correlation coefficients, violin plot and heatmaps were generated using GraphPad Prism 10.

We used an ELISA platform developed in our laboratory to assess IgG reactive to 93 common chemical adducts in the blood of patients with PDAC (N=39; age range 50-90) as well as control individuals without cancer matched for sex and age with cancer cases (N=40; age range 50-94). Importantly, all patient specimens were collected before treatment with chemotherapy. Our adduct panel includes post-translational modifications on 11 amino acids, oxidation-related modifications (e.g., malondialdehyde, MDA), advanced glycation end products (e.g., pyrraline, pentosidine) and certain co-enzymes that qualify as adducts based on their binding properties (e.g., flavin adenine dinucleotide, FAD). Figure 1A depicts the reactivity profiles toward all 93 adducts for PDAC cases and HD. As clearly apparent in this figure, reactivity to phosphoryl-threonine (Thr), -serine (Ser) and -tyrosine (Tyr) was markedly increased in cancer patients when compared to HD (red arrows). A random forest classifier built with the Boruta algorithm identified 13 adduct recognized by IgG that most efficiently discriminated between cases and controls ( Figure 1B ; Supplementary Figure S1 ). Phosphoryl-Thr, -Ser and -Tyr were the most significant among the 13 adducts. Spatial distribution based solely on reactivity to these 3 phosphorylated residues revealed two separate clusters corresponding to PDAC cases and HD ( Figure 1C ). In contrast a principal component analysis (PCA) based on reactivity to all 93 adducts did not distinguish PDAC cases from CTRL ( Supplementary Figure S2 ).

Differential serum IgG reactivity to the 3 phosphoryl adducts was next validated by ELISA using 5-mer peptides with modified amino acids as baits and unmodified amino acids as controls. As reported in Figure 2A , serum reactivity to all 3 targets was significantly elevated in PDAC cases compared to HD. Serum titration confirmed higher reactivity for PDAC patients compared to controls irrespective of the dilution factor ( Figure 2B ). Remarkably, reactivity to phosphoryl adducts was still detectable at 1:3200 - 1:6400 for several subjects, indicating high-titer antibody responses to these adducts. Results were quantified by calculating the area under the curve (AUC) for all titer curves. AUC values were significantly higher in PDAC patients than HD (p-value <0.0001, insets in Figure 2B ). To determine the subclasses of serum IgG reactive to phosphoryl adducts, we repeated the assay using secondary antibodies specific to IgG1, IgG2, IgG3 and IgG4. Results reported in Figure 2C revealed than most of the reactivity could be attributed to IgG1 and IgG3, with minimal signal detected for IgG2 and IgG4.

PDAC is often diagnosed at an advanced stage when the tumor has already spread to distant organs. This situation corresponds to M1 as part of the Tumor, Node, Metastasis (TNM) staging system. By opposition, M0, refers to absence of metastasis beyond regional lymph nodes. To assess whether the presence of metastasis, taken as a surrogate marker of tumor burden, would impact the development of antibodies to phosphoryl adducts, we compared IgG reactivity between patients diagnosed at M0 to those diagnosed at M1 stage. As shown in Figure 2D , patients with metastasis appeared to have higher reactivity to these adducts when compared to patients without metastasis, suggesting that higher tumor burden or dissemination may have enhanced antibody responses to these immunogenic targets. In addition to the differences observed between M0 and M1 patients, further analysis of IgG reactivity to phosphoryl adducts revealed significant variations among different patient groups ( Supplementary Figure S3 ). Post-treatment PDAC samples (PDAC-PT) exhibited higher IgG levels against phosphorylated adducts compared to healthy controls, suggesting that treatment did not significantly affect antibody responses to these immunogenic targets. Additionally, PDAC patients exhibited higher anti-phosphoryl adduct IgG levels when compared to patients with type 2 diabetes (T2D), suggesting that systemic metabolic alterations associated with type of diabetes, did not influence IgG responses to this modified antigens.

We then refined our predictive model to identify which of the 3 variables (i.e., IgG reactivity to phosphoryl-Ser, -Thr or -Tyr) was the most performant to discriminate between PDAC and controls. Moreover, beyond the prediction power of each individual variable, we also evaluated the performance of using a combination of two or three of these markers. As a first metric we calculated the patient-level sensitivity or true positive rate (TPR) defined as the probability of a PDAC case diagnosed. We also calculated the false positive rate (FPR or 1-specificity), defined as the probability of a wrong diagnosis of a control patient as a cancer case. Accuracy (ACC) was defined as the rate of true positive and true negative. Lastly, we calculated the area under the receiver operating curve (AUC-ROC) another critical measurement to quantify the model’s discriminatory power. Sensitivity, specificity, accuracy and AUC were calculated at an optimal cutoff point determined using the Youden index. All metric values for the 3 individual variables as well as all combinations of 2 or 3 variables are reported in Supplementary Table S4 . Figure 3A depicts the ROC for the 3 phosphoryl adducts. Overall, the models were very similar for all 3 variables and combinations thereof with respect to their discriminatory power. IgG reactivity to phosphoryl-Tyr appeared as the most performant variable with a sensitivity of 94.9% and an AUC of 81.6% but was only marginally superior to the two other variables. This observation is consistent with the fact that reactivity to phosphoryl-Ser, -Thr and -Tyr were highly correlated (pairwise Pearson correlation > 0.7).

The Sialyl Lewis antigen A or carbohydrate antigen 19-9 (CA19.9) is currently the only FDA-approved biomarker for the detection and monitoring of PDAC. We compared the predictive value of IgG reactivity to phosphoryl-Tyr to that of CA19.9 in our case-control sample. Of note, CA19.9 values at time of diagnosis were only available for 21/39 patients. Figure 3B reports the measurement for the 21 PDAC cases with the standard cutoff point of 37U/ml for CA19.9 and the previously determined cutoff point of 0.375 indicated by doted lines. As illustrated in this figure, IgG reactivity to phosphoryl-Tyr was superior to CA 19.9 in discriminating PDAC patients. Comparison between CA19.9 levels and IgG reactivity to phosphoryl-Ser and -Thr are reported in Supplementary Figure S4 .

To elicit high titer class-switched antibody responses and leave a serum “IgG footprint”, we reasoned that these adducts would have to be aberrantly expressed in tumor cells. To test this hypothesis, we assessed the presence of phosphoryl-Tyr directly in pancreatic tumor tissue. Detection of phosphoryl-tyrosine by immunofluorescence was performed in 13 cases of PDAC who had previously demonstrated elevated anti-phosphoryl-Tyr IgG levels and for whom formalin-fixed paraffin embedded tissue was available. For 11 cases, adjacent non-tumoral tissue was also available and stained as control. Figure 4 , reports two representative example showing much higher levels of phosphoryl-Tyr in the tumor area when compared to the adjacent non-tumoral area or non-cancerous pancreatic tissue. The remaining cases are reported in Supplementary Figure S5 . A summary of staining results is provided in Table 1 . On the whole, aberrant buildup of phosphoryl-Tyr in tumor cells was unmistakably observed in 13 out of 13 PDAC patients. In contrast, adjacent non-tumoral tissue either did not stain for phosphoryl-Tyr (6/11 cases) or showed minimal staining of acinar cells, benign duct cells, endothelial cells or infiltrating lymphocytes (5/11 cases).

The accumulation of phosphoryl adducts in PDAC cells revealed by immunofluorescence suggested that this cancer type may be associated with dysregulation and abnormal activation of phosphorylation pathways. We next used a mass spectrometry-based phosphoproteomics approach to analyze the phosphoproteome in 3 PDAC patients for whom paired flash-frozen PDAC and adjacent non-tumoral tissue were available. All 3 patients had previously showed elevated serum IgG levels to phosphoryl-adducts. Differential Phosphorylation was computed using the msqrob2 package as specified in the materials and methods section. A total of 10,986 sites were identified as differentially phosphorylated between tumor and non-tumoral cells in any of the 3 samples. Among these, 154 sites corresponding to 124 proteins, showed higher phosphorylation levels in tumor cells compared to adjacent non-tumoral cells in all 3 cases at a statistical significance level of 5% (p<0.05, Figure 5A ; Supplementary Table S5 ). Nineteen were confirmed at a statistical significance level of 1% (p<0.01, Figure 5B ; Supplementary Table S6 ).

A gene ontology (GO) enrichment analysis was then carried out to further characterize the 124 proteins identified as aberrantly over-phosphorylated in tumor tissue compared to adjacent non-tumoral pancreatic tissue. This analysis identified 104 biological processes, 18 cellular components and 10 molecular functions listed in Supplementary Table S7 . A visual summary of this GO analysis is presented in Supplementary Figure S6 . Key biological processes include protein localization to chromosomes, regulation of transcription by RNA polymerase I, and regulation of histone acetylation ( Supplementary Figure S6A ). Enriched cellular components indicated involvement in nuclear protein-containing complexes, pericentromeric heterochromatin, and ribonucleoprotein complexes ( Supplementary Figure S6B ). Lastly, molecular functions highlighted chromatin binding, molecular adaptor activity, and RNA binding, suggesting a role in chromatin organization and gene expression regulation ( Supplementary Figure S6C ).