Clinical functional proteomics of intercellular signalling in pancreatic cancer

a, Chemical structure of in-house synthesized biotin-hydrazide for glycoprotein labelling and enrichment. b, Labelling and enrichment efficiency of the biotin-hydrazide probe. Mouse pancreas proteins were extracted and labelled with 2 mM probe for 1 h (lane 2). Lysate without biotin-hydrazide labelling was subjected for streptavidin pull down as control (lane 3). Labelled proteins were pulled down by streptavidin beads (lane 4). Supernatant was collected from the streptavidin pull down (lane 5). c, Comparison of biotin-hydrazide probe-based method with conventional hydrazide bead-based method. Either biotin-hydrazide probe or hydrazide beads were incubated with 500 μg lysate for different times. After on-bead trypsin digestion to remove non-glycopeptides, glycopeptides were released from the streptavidin beads or hydrazide beads by PNGase F for LC–MS/MS analysis. Data are mean of two technical replicates. d,e, Subcellular ___location of proteins identified by glycopeptides (d) and sequence coverage comparison of 1,313 S–PM proteins commonly identified by glycopeptides and non-glycopeptides (e) using 6 pairs of tumour and NT samples. f, The comparison of analysis by non-glycopeptides and glycopeptides in terms of identification and quantification performance of S–PM proteins. The left panel is the number of S–PM proteins identified by glycopeptides and non-glycopeptides. Data are mean ± s.d. of 6 pairs of tumour and NT samples. The middle panel is the sum peptide intensities of all identified S–PM proteins from these 12 tissue samples. The right panel is the CV of 1,313 overlapped S–PM proteins quantified with LFQ intensities by glycopeptides and non-glycopeptides. g, The percentage of secreted (S), plasma membrane (PM), and other transmembrane (TM) proteins, non-S/PM/TM proteins, and nonspecifically labelled proteins reported by Matthews et al.⁷⁰ to total proteins in terms of protein number (left panel) and sum LFQ intensity (right panel) as identified by glycopeptides and non-glycopeptides. h, The percentage of proteins nonspecifically absorbed on streptavidin beads to total proteins identified by glycoprotein enrichment. The experimental procedure for identification of nonspecifically absorbed proteins was the same as glycoprotein enrichment, except that no biotin-hydrazide probe was added. Proteins identified only in the streptavidin pull down sample without biotin-hydrazide probe or having 2-fold higher LFQ intensity as compared with glycoprotein enrichment were defined as nonspecifically absorbed proteins. i,j, Comparison of quantification performance of S–PM proteins by global proteomic method (Unenriched) and N-glycoproteomic method (Enriched) using 6 pairs of tumour and NT samples. S–PM proteins quantified by Unenriched and Enriched were compared in terms of protein number and intensities (i), and CV of 1,546 S–PM proteins commonly quantified by Unenriched and Enrich (j). k,l, Correlation of the average tumour/NT ratios of commonly quantified S–PM proteins between the two methods. Bio-replicates denote correlation of the 6 pairs of tumour and NT samples; correlations of glycoproteins identified with different number of N-glycosites are independently displayed (k); technical replicates denote correlation of 1 pair of tumour and NT samples for three technical replicates (l).

a, Pearson’s correlation coefficients of LFQ intensities between any two samples from the same group of normal, NT, CP, or tumour tissues, respectively. b, Classification of S–PM proteins according to their roles in intercellular communications. The bar graph shows the top 20 intercellular communication roles. c, Clustering of top 50 GOMF terms of significantly changed S–PM proteins between Tumour and NT, related to Fig. 1g. d, Comparison of top 20 GOMF terms of significantly altered S–PM proteins between Tumour versus Normal and Tumour versus CP. P values are from hypergeometric distribution (c, d).

a, Cell type deconvolution of 29 tumour samples using our bulk S–PM proteome in Fig. 1b and published scRNA-seq dataset of PDAC (Methods). Box plots are as defined in Fig. 2c. b, Representative images of IHC-guided and hematoxylin staining-based LCM of PCC and stromal regions from tumour tissue section. The stroma region was stained by IHC targeting PDGFRB, a marker of stroma cells. The PCC region was stained by IHC targeting KRT19, a marker of epithelial cells. c, The number of S–PM proteins identified in each of the tissue and cell line samples. d, Pearson’s correlation coefficients of LFQ intensities between any two biological replicates from the same region of PCCs or stroma. e, Comparison of proteome depth of our LCM-based proteomic dataset with a recently published LCM-based proteomic dataset of PDAC⁸. The raw files of the published dataset were downloaded and processed using the same workflow and criteria as our dataset. f, Comparison of the S–PM proteins identified in spatial proteomic dataset with genes of S–PM proteins from the transcriptomic dataset summarized from spatial or scRNA-seq analysis of PDAC^18,71,72. All the identified genes of S–PM proteins from the three references were combined as the transcriptome dataset. g, Cell-type annotation of the S–PM proteins identified from N-glycoproteomic profiling of 29 bulk tumour tissues. h, TSA staining of 5 PCC-specific and tumour-upregulated PM proteins, including the aspartate/glutamate transporter SLC1A3, the PKD1 calcium channels, the TMPRSS4 protease, the CDON Hedgehog signalling regulator, and the LRIG2 protease regulator. KRT19 and PDGFRB were marker of PCCs and stroma, respectively. Box plots are as defined in Fig. 2c. The pearson’s r values indicates the colocalization potential of target proteins with KRT19 or PDGFRB (Methods). i,j, The number of ligands, receptors, and ligand–receptor pairs annotated from secreted ligands and plasma membrane receptors of Extended Data Fig. 2b by an in-house ligand–receptor database (see Methods) (h), and display of the 1,724 ligand–receptor pairs (j). The pairs of three classes of most enriched GOMF terms are shown as examples, with representative proteins displayed. k, Dot plot showing GOMF terms of the 1,724 ligand–receptor pairs. Top 20 GOMF terms according to pair number were annotated and terms in red were presented in j. l, Protein expression trend of the 1,724 pairs according to LFQ intensity ratio distribution of tumour and normal tissues. One dot represents one pair. The X and Y axes are the log₂ transformed LFQ ratio between Tumour and Normal of ligand and receptor, respectively. The pie chart shows the percentage of pairs in each section to total pairs. m, The number of spatially resolved and cell type-specific ligands and receptors (top panel) and their predicated signalling flow (bottom panel) in 29 tumours. Data are mean ± s.d. P values are from two-tailed unpaired Student’s t-test (e,m), or hypergeometric distribution (k).

a, Pearson’s correlation coefficients of LFQ intensities between any two samples from the same group of NT, 3 weeks, 5 weeks, or 7 weeks of KPC mice, respectively. b, Consensus-clustering analysis of S–PM proteins differentially expressed between NT and different tumour stages of the KPC mouse model. c, The number of differentially expressed S–PM proteins in each stage of tumour tissues as compared with NT, respectively. d, Overlap and expression trend consistency of differentially expressed proteins between human PDAC and KPC mouse model. e, Relative expression levels of representative PDAC biomarkers, including Lif¹⁴, Thbs2⁷³, Timp1⁷⁴, Gpc1⁷⁵, Msln⁶⁰, and Lgals3⁷⁶, in different stage tumour tissues of KPC mice and in human pancreatic tissues. The number of human and mouse tissue samples in each group is indicated in Figs. 1a and 3b, respectively. f, Top 5 GOMF terms of S–PM proteins in the three clusters. P values are calculated by hypergeometric distribution. g, Validation of TNFRSF11B and NPTX1 in PDAC plasma samples. Human plasma samples from training and validation cohorts were tested by ELISA kits. h, Annotation of ligand–receptor pairs in 29 bulk human tumour tissues on the basis of the indicated tumour progression trends. Data are mean ± s.d. All pairs refer to the pairs shown in Extended Data Fig. 3j in which the expression of at least ligand or receptor was significantly changed in human Tumour versus Normal in N-glycoproteomic dataset. Box plots are as defined in Fig. 2c (e, tissue results of g). Box plots of plasma results in g: centre line, median; boxes, interquartile range; whiskers, minimum to maximum. P values are from two-tailed unpaired Student’s t-test.

a, Clinical tissue samples used for N-glycoproteomic analysis, photo-pTyr-scaffold, and pTyr peptide enrichment. b, Overlap of pTyr writers, readers, and erasers identified by photo-pTyr-scaffold approach and pTyr peptide enrichment approach. c, Summary of all the pTyr writers, readers, and erasers identified by two approaches in Fig. 4b. The rectangle indicates proteins identified by the photo-pTyr-scaffold approach and the circle indicates the pTyr sites identified by pTyr peptide enrichment approach. Proteins are classified according to their molecular functions. d, Expression levels and pTyr sites of top ranked pTyr writers, erasers and readers in Fig. 4c. Box plots are as defined in Fig. 2c. P values are from two-tailed unpaired Student’s t-test. e, The overlapped GOBP terms of proteins identified by our multidimensional proteomics (S–PM and pTyr machinery proteins identified in tumour tissues by N-glycoproteomic profiling and pTyr-mediated complexes profiling) with the core signalling pathways summarized from global genomic analysis of PDAC^77,78. The core signalling pathways summarized in these two references were combined. P values are from hypergeometric distribution. f, The four categories of activated ligand–receptor–downstream signalling axes in which the expression of both receptor and downstream protein were significantly changed in tumours in the pTyr signalling dataset (circles labelled with P). g, Spatial and temporal annotation of the 148 pairs in Fig. 4d. h, The activated paracrine signalling pairs from stromal cells to PCCs in tumour and their downstream proteins screened out from g.

a, Expression levels of PDGFRA, PDGFRB, and PDGFC in the multidimensional proteomic datasets shown in Figs. 1–4. The number of samples for protein expression, cell type, tumour progression, and pTyr activation is indicated in Fig. 1a, Fig. 2a, Fig. 3b, and Extended Data Fig. 5a, respectively. Box plots are as defined in Fig. 2c. P values are from two-tailed unpaired Student’s t-test. b, Validation of PDGFR activation by PANC1 CM or recombinant PDGF-BB and inhibition by the PDGFR inhibitor crenolanib in HPaSteCs. Cells were starved in FBS-free medium for 12 h with or without PDGFR inhibitor crenolanib, and then stimulated with PDGF-BB or PANC1 CM for 5 min before cell lysis for WB analysis. c, Validation of signalling proteins downstream of PDGFRA and PDGFRB. HPaSteCs were transfected with siRNAs targeting PDGFRA or PDGFRB in normal medium for 24 h, followed by in FBS-free medium for 24 h, and then stimulated with PANC1 CM for 5 min before cell lysis for WB analysis. d, Validation of PTPN11 as an upstream regulator of ERK signalling but not AKT signalling. Cells were starved in FBS-free medium for 12 h with or without PDGFR inhibitor crenolanib or PTPN11 inhibitor SHP099 with indicated concentrations, and then stimulated with PDGF-BB for 5 min before cell lysis for WB analysis. e, Secretome analysis of HPaSteCs after PDGF-BB stimulation and PDGFR or PTPN11 inhibition (n = 3 biological replicates). HPaSteCs were treated with 100 ng ml⁻¹ PDGF-BB, 0.5 μM crenolanib or 10 μM SHP099 for 24 h, respectively. f, Investigation of the signalling axis regulating LIF expression upon PDGF stimulation. Cells were treated with PDGFR inhibitor crenolanib (0.5 μM), JAK inhibitor ruxolitinib (1 μM), PI3K inhibitor LY294002 (10 μM), PTPN11 inhibitor SHP099 (10 μM), MEK inhibitor mirdametinib (10 μM), or ERK inhibitor ravoxertinib (10 μM) for 24 h before cell lysis for total RNA extraction and measurement of LIF mRNA levels by real-time qPCR. g, Effect of PTPN11 knockdown on LIF protein secretion from HPaSteCs with or without PDGF-BB stimulation. h, Effect of inhibitors targeting the PDGFR–PTPN11–ERK signalling axis on LIF promoter activity. HPaSteCs were stimulated with PDGF-BB with or without an inhibitor targeting PDGFR, PTPN11 or ERK. LIF promoter activity was measured by a dual-luciferase reporter assay system. i, Validation of FOS as a LIF gene transcription factor after PDGF-BB stimulation (100 ng ml⁻¹, 24 h) as determined by ChIP assay. P1 to P4 indicate four promoter regions of LIF, the result shows that P2 region is the binding site of FOS. Due to similar molecular weight of proteins, samples were run on separate gels, with β-actin as sample processing control (b–d). All the bar graphs are mean ± s.d. of n = 3 (b–d,f,g,i) or n = 4 (h) biological replicates; P values are two-tailed unpaired Student’s t-test (a,f–i) or one-tailed paired Student’s t-test (b–d).

Source Data

a, Schematic representation of detail criteria for identification of shed PM proteins in tumours. In step 1, the unique peptide ratio of ECD/ICD for PM proteins (Ratio_PM) identified in the PM proteome by the N-glycoproteomic analysis of cell pellets (data presented in Extended Data Fig. 3c), the same ratio for PM proteins (Ratio_S) identified by the secretome analysis (data presented in Extended Data Fig. 3c), and the same ratio for PM proteins (Ratio_S–PM) identified in the S–PM proteome by the N-glycoproteomic analysis of 29 PDAC tumour tissues (data presented in Fig. 1b) were calculated after data filtering according to the indicated criteria, respectively. The strict cutoff ensures the reliability and reproducibility of quantification performance in each dataset. In step 2, to avoid false positive calculation of shedding proteins due to the intrinsic difference of sequence length of ECD and ICD, the Ratio_S were first normalized to (divided by) their corresponding Ratio_PM to generate the Index_training values so as to identify shed PM proteins in the secretome of 8 cell lines. Then, 45 shed PM proteins were identified in cell lines. By using the same normalization step and indicated filtering criteria, 22 shed proteins were finally identified and validated in tumour tissues. b, Molecular functions of the 22 shed proteins identified in tumour tissues. c, Expression level of MMP1, MMP9, MMP11, MMP15, ADAM10, and ADAM17 in tumour and normal tissues by N-glycoproteomic profiling. d, Knockdown efficiency of each MMP by siRNA in PANC1 cells in Fig. 5e. Cells were transfected with siRNAs in normal medium for 24 h, followed by in FBS-free medium for 24 h before that WCLs were collected for total RNA extraction. The mRNA levels of MMPs were measured by real-time qRT–PCR. e, Statistic analysis of WB result in Fig. 5f, showing sAXL levels in PANC1 CM after knockdown of different MMPs. Cells were transfected with siRNA targeting MMP1, MMP9, MMP11, or MMP15 in normal medium for 24 h, followed by in FBS-free medium for 24 h before that CM and WCL were collected for WB analysis. f, Knockdown efficiency of each MMP by siRNA in MIA PaCa2 cells. g, Global profiling of shedding substrates of MMPs in MIA PaCa2 cells. The heatmap shows shedding level of 17 PM proteins identified as shed proteins in cell lines and tumour tissues. AXL levels are shown as an example. h, Validation of AXL shedding upon MMP knockdown by siRNAs in MIA PaCa2 cells. Cells were treated with siRNAs by following the same procedure as PANC1. sAXL in CM and WCL was detected by WB, β-actin was run on the same gel as loading control, and the bar graph shows band intensities of sAXL in CM of the WB image. Box plots are as defined in Fig. 2c (c,g). All the bar graphs are mean ± s.d. of n = 4 (d,f) or n = 3 (e,h) biological replicates. P values are from two-tailed unpaired Student’s t-test (c,d,f,g) or one-tailed paired Student’s t-test (e,h).

Source Data

a, Expression level of TAM family members AXL, MERTK and TYRO3 in multi-dimensional proteomic datasets. The number of samples for protein expression, tumour progression, and pTyr activation is indicated in Fig. 1a, Fig. 3b, and Extended Data Fig. 5a, respectively. Box plots are as defined in Fig. 2c. P values are from two-tailed unpaired Student’s t-test. b, The linearity of PRM quantification for stable isotope labelled heavy peptides of ECD and ICD of AXL, and GAS6. Different amount of heavy peptides were spiked into 100 ng of PANC1 peptides before PRM-MS analysis. c-e, Representative PRM-MS transition peaks of ICD (c) and ECD (d) of AXL, and GAS6 (e) from tumour samples. The transition peaks of stable isotope labelled heavy peptide spiked into the sample are shown on the right. The dark arrow indicates the retention time and mass error for the highest transition peak. f, IHC image of lymph node metastasis. Tumour cells metastasized into the lymph node were marked with IHC staining targeting KRT19. Ten lymph nodes were stained and representative image is presented. g–i, Classification of tumours according to lymph node metastasis status and relative levels of sAXL and GAS6 in PCC and stroma regions of 50 primary tumour samples (g,h), and in 41 samples showing consistent trend between PCCs and stroma (namely, the absolute amount of sAXL is higher or lower than GAS6 for both PCC and stromal regions isolated from the same tumour). Displayed are averages of PCC and stromal regions quantified from the same tumour for both sAXL and GAS6 (i). j, Boxplot of sAXL, GAS6, and CA19-9 levels in human plasma samples from the training cohort and validation cohort of normal controls and PDAC patients. Centre line, median; boxes, interquartile range; whiskers, minimum to maximum. The results from the training cohort and the validation cohort were combined for ROC curve analysis in Fig. 6d.

Source Data

a, Inhibition of AXL shedding by the broad-spectrum MMP inhibitor BB-94. β-actin was run on the same gel as loading control. The bar graph is statistical analysis of sAXL level in PANC1 CM. b, Inhibition of AXL shedding in primary KPCP tumour cells by the broad-spectrum MMP inhibitor BB-94. Cells were maintained in FBS-free medium containing 10 μm BB-94 for 24 h, and secretome in CM was profiled by LC-MS/MS analysis. The LFQ intensity of AXL was extracted for comparison between control and BB-94 treatment. c, Statistic analysis of the WB results in Fig. 6e, showing relative band intensities of p-AKT after normalization to total AKT. d, PRM analysis of two AXL pTyr peptides after enrichment of phosphopeptides. The treatment of PANC1 cells was the same as in Fig. 6e. The left panel of each pTyr peptide is representative transition peaks. The dark arrow indicates the retention time and mass error for the highest transition peak. e, Schematic workflow for testing synergistic effect between BB-94 and R428 on PDOs. f, Histological characterization of PDAC tumour tissues and corresponding PDOs and PDOs-derived xenografts. PDOs and PDOs-derived xenografts formed typical glandular tubular structures similar to the corresponding patient tumours, and had similar expression levels of the epithelial cell marker KRT19 and SOX9, and cell proliferation marker Ki67. g,h, Solo or combined drug treatments on xenograft tumours generated by one case of PDO (DAC-71) with significant synergistic response (g) between BB-94 and R428 (statistics are presented in Fig. 6g). i–k, Validation of synergistic effect on xenograft models generated by one case of PDO (DAC-18) with no synergistic response (i). Data are mean + s.d. of n = 8 xenograft tumours per group (j,k). l, Orthotopic tumours of PDAC under different drug treatment. Orthotopic model was constructed by injecting KPCP primary cancer cells into pancreata of NSG mice. After tumour formation, mice were daily administrated with indicated drugs for 14 days. The dark red tissues were spleens, which were removed before measuring tumour weights. Tumour weights are presented as mean ± s.d. of n = 9 mice per condition. m, Correlation of MMP1 and MMP11 mRNA expression levels in 66 pancreatic PDO lines with drug combination response of BB-94 and R428. Box plots are as defined in Fig. 2c. n,o, Colony formation assay of MIA PaCa2 cells with stable knockdown of MMP1 or MMP11 by shRNAs. Cells were treated with DMSO as control or 0.5 μM R428 for testing of inhibition efficiency (n). Knockdown efficiency was measured by real-time qRT–PCR (o). p, Transwell migration assays of MIA PaCa2 cells under shedding inhibition and/or AXL inhibition. Shown are representative crystal violet-stained images of transwell migration assays of MIA PaCa2 cells under different treatments. Quantification of cell migration determined by counting the number of migratory cells per field and relative numbers of migratory cells were calculated by normalizing to control cells. q, Expression of EMT markers in cancer cells upon inhibition of shedding by BB-94. Due to similar molecular weight of proteins, samples were run on separate gels, with β-actin as sample processing control. All the bar graphs are mean ± s.d. of n = 3 biological replicates (the data points in p are relative cell numbers from 2 random views of each biological replicates). P values are from two-tailed unpaired Student’s t-test (b,d,k,l,m,n,o,p) or one-tailed paired Student’s t-test (a,c,q).

Source Data