Pancreatic ductal adenocarcinoma ubiquitination profiling reveals specific prognostic and theranostic markers

Summary

Background

Pancreatic ductal adenocarcinoma (PDAC) has been widely studied at multiomics level. However, little is known about its specific ubiquitination, a major post-translational modification (PTM). As PTMs regulate the final function of any gene, we decided to establish the ubiquitination profiles of 60 PDAC.

Methods

We used specific proteomic tools to establish the ubiquitin dependent proteome (ubiquitinome) of frozen PDXs (Patients' derived xenographs). Then, we performed bioinformatics analysis to identify the possible associations of these ubiquitination profiles with tumour phenotype, patient survival and resistance to chemotherapies. Finally, we used proximity ligation assays (PLA) to detect and quantify the ubiquitination level of one identified marker.

Findings

We identified 38 ubiquitination site profiles correlating with the transcriptomic phenotype of tumours and four had notable prognostic capabilities. Seventeen ubiquitination profiles displayed potential theranostic marker for gemcitabine, seven for 5-FU, six for oxaliplatin and thirteen for irinotecan. Using PLA, we confirmed the use of one ubiquitination profile as a drug-response marker, directly on paraffin embedded tissues, supporting the possible application of these biomarkers in the clinical setting.

Interpretation

These findings bring new and important insights on the relationship between ubiquitination levels of proteins and different molecular and clinical features of PDAC patients. Markers identified in this study could have a potential application in clinical settings to help to predict response to chemotherapies thereby allowing the personalization of treatments.

Funding

Fondation ARC (PJA 20181208270 and PGA 12021010002840_3562); INCa; Canceropôle PACA; DGOS; Amidex Foundation; Fondation de France; and INSERM.

Research in context.

Evidence before this study

Nowadays, pancreatic ductal adenocarcinoma (PDAC) remains one of the deadliest forms of cancer. While it is a very heterogeneous disease, most patients do not benefit from a personalized treatment that would warranty a better efficacy and improved quality of life. Therefore, identifying new molecular markers that could predict the tumour aggressiveness and, more importantly, the specific response to the different chemotherapies available is essential. Post-translational modifications (PTMs) of protein play key roles in regulating their functions. Alterations of PTMs actively participate in the oncogenic transformation process and they represent valuable markers that can be detected in patients' tumours. Ubiquitination is a major PTM whose deregulation is involved in most diseases, including cancer. Therefore, we decided to establish the ubiquitination profiles of a significant number of PDAC samples in order to identify specific ubiquitination profiles correlating with tumour aggressiveness and resistance to different treatments, hence serving as prognostic and theranostic markers.

Added value of this study

We identified several ubiquitination site specific prognostic and theranostic markers for the four main chemotherapeutics used against PDAC. We could confirm one of them and validated its potential use in clinical setting. These results show that exploring PTM landscapes, such as ubiquitination, has the potential of revealing new and important molecular markers to stratify patients and, importantly, to predict their chemosensitivity. Moreover, we bring a proof of concept that this kind of marker could be employed in clinical settings.

Implications of all the available evidence

This study led to the identification of new prognostic/theranostic markers as well as new molecular targets to try to tackle PDAC cells resistance mechanisms, which could not be previously identified in genomic, transcriptomic and even proteomic studies. Our findings also validate the importance of screening for diseases associated alterations of PTMs, especially ubiquitination, to identify new markers and new molecular targets.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is one of the most aggressive cancers, characterized by a median survival of six months and a five-year survival rate between 5% and 7%.¹ The main reasons for this dismal prognosis are that PDAC is often diagnosed at an advanced stage and that it is resistant to most available treatments. This situation is even more worrying since the incidence of PDAC is increasing yearly, suggesting it will be the third-commonest cause of death by cancer by 2030.² PDAC is a very heterogeneous disease and, whereas patients should benefit from personalized medicine, treatments do not consider the molecular characteristics of each tumour. Therefore, identifying new molecular markers, that could report tumour aggressiveness and specific chemotherapy response, and discovering new therapeutic options by targeting original molecular mechanisms of resistance, is of most importance.

Recently, we and others have performed multiomics studies of different preclinical models of PDAC, which allowed the stratification of this disease.3, 4, 5, 6, 7 These studies also resulted in the identification of specific transcriptomic signatures able to predict sensitivity to given chemotherapeutics8, 9, 10 which could potentially be used in the clinic as a decision-helping tool. While extremely helpful and powerful, these multiomics studies were not exclusively but mainly DNA and RNA based, which missed the importance of the biological information from proteins. Indeed, a gene usually codes for several protein isoforms (proteoforms), which may have distinct biological functions, and they are all regulated by post-translational modification(s) (PTMs) of many different kinds. In fact, there is globally a poor correlation between mRNA and protein levels,^11,12 a discrepancy also observed in PDAC,¹³ and we can reasonably suppose that this correlation is even weaker between transcripts and associated functions. Consequently, proteomic, and especially modifo-proteomic (modifomic), has the potential to reveal functionally pertinent molecular markers or signatures and, importantly, new molecular targets which could not be detected by other omics studies.

After phosphorylation, ubiquitination is the second most abundant PTM (iPTMnet v6.1) and one of the most significant as it plays a pivotal role in regulating proteostasis, via the ubiquitin proteasome system (UPS) and the autophagy-lysosome pathway (ALP),¹⁴ but also because it has the capacity to modulate the function of most proteins. Ubiquitin is a small (76 amino acids) and highly conserved protein (96% identity between human and yeast ubiquitin). It is conjugated to the target proteins via its C-terminus to the amino group of the side chain of a lysine residue of the modified protein, forming an isopeptide bond and a monoubiquitination. Of note, ubiquitin is a multi-role protein that can also form polyubiquitin chains of 8 different topologies (seven via its lysine residues, one through N-Ter to C-Ter ligation). This diversity of ubiquitination signals allows it to regulate the activation, localization and interactions of probably all cellular proteins, and is therefore involved in most cellular biological processes such as cell signalling, stress response, DNA repair and more.¹⁵ This central role played by ubiquitination is further highlighted by all alterations associated with human diseases, including cancer.¹⁶ While specific PDAC associated alterations of ubiquitination, substrates or enzymes, have been described the literature,¹⁷ no systematic profiling of PDAC has ever been done to date.

Thanks to progress in mass spectrometry, as well as tools dedicated to the enrichment of ubiquitinated substrates,^18,19 the knowledge regarding this “ubiquitin code” 1¹⁵ has been greatly improved during the past decade. One of these new technologies is based on the immune-enrichment of the so-called “diGlycine-conjugated peptides” (diGly-pept), which correspond to the remnant two C-terminal glycine residues of ubiquitin that are still conjugated to the lysine residue of the substrate protein after trypsin digestion. These diGly-pept can be easily detected and semi-quantified by mass spectrometry, providing the identity of the ubiquitinated protein, the precise ubiquitinated site and its ubiquitination level. This comprehensive approach has been successfully applied in a wide range of biological questions including cancer.²⁰ In this work, we used this strategy to generate the ubiquitination profiles of large cohort of 60 PDXs obtained from both resected and advanced (non-resected) PDACs. Profiles were analysed by bioinformatics to identify ubiquitination sites that could predict tumour aggressiveness, the patient's outcome, as well as PDAC sensitivity to the main cytotoxics used to treat this disease. Finally, we confirmed the pertinence of specific ubiquitination site profiles and validated the applicability of such markers to the clinic on PDAC paraffin-embedded tissues.

Methods

Patients' cohorts

Sixty patients with a confirmed PDAC diagnosis were included in this study. Clinical data was collected (Table S1). Tumour samples were obtained from pancreatectomy in 33 patients (55%), endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) in 18 patients (30%) and during explorative laparotomy in nine patients (15%). The retrospective validation cohort included 22 patient's PDAC (54–79 y old) from the CRO2 tissue collection (Agreement reference DC-2013-1857).²¹

Ethics

The study was approved by the Paoli-Calmettes Institute ethics committee and was registered at www.clinicaltrials.gov with registration number NCT01692873. The study conformed the ethical approval and informed consent from all participants was obtained. Participants were not compensated for participation.

Patient-derived xenograft (PDX) and patient derived organoid

All samples were xenografted in immunocompromised mice, producing PDX samples as previously described.²² Briefly, PDAC tissue was fragmented and mixed with 100 μl of Matrigel and implanted subcutaneously in a NMRI-nude mouse until the tumour reached 1 cm³ (Swiss Nude Mouse Crl: NU(lco)-Foxn1nu; Charles River Laboratories, Wilmington, MA). Animal experiments were approved by the local ethics committee and performed following the guidelines of our center (CRCM) and were approved by the “Plateforme de Stabulation et d’Expérimentation Animale” (PSEA, Scientific Park of Luminy, Marseille). All the experiments on mice were done after approval from the ethical committee for animal experimentation and French Ministry of Higher Education and Research (APAFIS# 9562-2016051914513578 version 4). The organoid model used in this work was obtained by FNA biopsy from a patient with an unresectable PDAC as previously described in Juiz et al.²³

In vivo chemoresistance tests

As described in Hoare et al.,²⁴ thirteen PDXs, that covered all different transcriptomic phenotypes, were selected from the cohort of 60 patients to perform in vivo chemoresistance tests. In vivo replicates for each condition varied between 6 and 10 Swiss Nudemice, depending on growth success. PDXs were allowed to grow until a volume of 200 mm³ was reached. At this point, mice received the treatment into the tail vein. Control mice received a solution of 0.9% NaCl, irinotecan was given every second day for a total of three 22 mg/kg administrations (Q2d × 3), gemcitabine every third day with a total of four 120 mg/kg administrations (Q3d × 4), 5-fluorouracil (5-FU) every four days with a total of two 56 mg/kg administrations (Q4d × 2) and oxaliplatin every four days with a total of two 5 mg/kg administrations (Q4d × 2). Tumour volumes were measured twice a week, over a period ranging from 0 to 200 days, with a Vernier calliper device. The tumour volume was calculated using the following formula, v = (length × width²)/2 (Table S2). Any mice exceeding a tumour volume greater than 2000 mm³ were sacrificed and excluded from the rest of the experiment for ethical reasons. Some PDX/treatment pairs did not produce chemograms and were removed from the study. There was no randomisation to allocate experimental units to control and treated groups, and other confounders were not controlled. All replicates in the control and treated groups were plotted and the areas under the curve (AUC) calculated (Figure S1a). Resistance scores of each PDXs were calculated as the ratio of average AUC of treatments divided by the average AUC of control thereby producing values from 0 (fully sensitive PDXs) to 1 (fully resistant PDXs) (Figure S1b, Table S1).

PDXs processing and diGly peptides enrichment

Liquid-nitrogen-frozen pieces of PDXs were solubilized on ice in cold lysis buffer (50 mM Tris pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% NP40, 0.1% Na-deoxycholate, + cocktail inhibitors diluted at 1/200 (Sigma-Aldrich, P8340) + 10 mM N-Ethylmaleimide. Complete dissociation was achieved using GentleMACS Octo dissociator (Miltenyi Biotec, RRID:SCR_020272) with M tubes and the program named “Protein”, then samples were returned immediately on ice. After 10 min centrifugation at 4 °C and 4.300 rpm, supernatants were transferred to new tubes. The protein concentration of these cleared lysates was estimated using Protein Assay (Bio-Rad, # 5000006). 5 mg of protein was precipitated using acetone and final protein pellet solubilized in 8 M urea, 10 mM HEPES pH 8.0 buffer. Cysteine residues were reduced with 5 mM DTT then alkylated with 14 mM chloroacetamide which was finally quenched with an additional 5 mM DTT. For the final normalization of the raw data, 15 ng of stable isotopes containing standard diGly-branched peptide (LIFAG [K (GG)]QLEDG [R (13C6; 15N4)], which correspond to the peptide of ubiquitin bearing the conjugated lysine 48, was added in each sample. Samples were first digest with 4 μg endoproteinase Lys-C 2 h at 37 °C, diluted 4 times with 10 mM HEPES pH 8.0, then digested o/n with 15 μg sequencing grade trypsin at 37 °C. Reaction was stopped by adding TFA (Trifluoroacetic acid) to a final concentration of 0.4%. Samples were centrifuged at 10.000 rpm for 10 min and pellet discarded. Peptides' mixtures were purified on Sep-Pak columns and eluted in 70% Acetonitrile, 0.5% Acetic acid, and immediately freeze in liquid nitrogen. Frozen samples were lyophilized using a speedvac, solubilized in 100 mM Ammonium Carbonate, frozen and lyophilized again, and finally solubilized in IP (Immuno-Precipitation) buffer from the UbiScan kit (Cell Signalling, #5562) containing 0.5% Triton. Purification of diGly-pepts was performed essentially following manufacturer's recommendations and final elution obtained in 0.15% TFA. Samples were immediately frozen and stored until mass spectrometry analysis.

Mass spectrometry analysis

Samples were analysed by liquid chromatography (LC)-tandem mass spectrometry (MS/MS) with an Orbitrap Fusion Lumos Tribrid Mass Spectrometer (RRID:SCR_020562) online with an Ultimate 3000RSLCnano chromatography system (Thermo Fisher Scientific, Sunnyvale, CA). Due to the size of the cohort (60 samples), only technical duplicates were performed. Peptides were concentrated and purified on a pre-column from Dionex (C18 PepMap100, 2 cm × 100 μm I.D, 100 Å pore size, 5 μm particle size) in solvent A (0.1% formic acid in 2% acetonitrile). In the second step, peptides were separated on a reverse phase LC EASY-Spray C18 column from Dionex (PepMap RSLC C18, 50 cm × 75 μm I.D, 100 Å pore size, 2 μm particle size) at 300 nL/min flow rate and 40 °C. After column equilibration using 4% of solvent B (20% water, 80% acetonitrile, 0.1% formic acid), peptides were eluted from the analytical column by a two steps linear gradient (4–20% acetonitrile/water; 0.1% formic acid for 120 min and 20–45% acetonitrile/water; 0.1% formic acid for 20 min). For peptide ionization in the EASY-Spray nanosource, spray voltage was set at 2.2 kV and the capillary temperature at 275 °C. The mass spectrometer was used in data dependent mode using advanced peak detection and switching consistently between MS and MS/MS. Time between master scans was set to 3 s. MS spectra were acquired with the Orbitrap in the range of m/z 400–1600 at a FWHM resolution of 120,000 measured at 200 m/z. AGC target was set as standard with a 50 ms maximum injection time. The precursor ions were selected, and collision induced dissociation fragmentation at 35% was performed and analysed in the ion trap using dynamic maximum injection time an AGC target at 40%. Charge state screening was enabled to include precursors with 2 and 7 charge states. Dynamic exclusion was enabled with a repeat count of 1 and a duration of 60s.

Protein identification and quantification

Relative intensity-based label-free quantification (LFQ) was processed using the MaxLFQ algorithm²⁵ from the freely available MaxQuant computational proteomics platform version 1.6.3.4.²⁶ In order to establish the ubiquitination profile of PDAC cells, spectra were searched against a Human Protein database from UniProt (accessed December 2019; 20,379 entries) and using the following settings: a maximum of two trypsin mis-cleavage was allowed, methionine oxidation and GlyGly motifs on lysine as variable modifications, and cysteine carbamidomethylation as fixed modification. To estimate the relative proportion of diGly-pepts that are human specific, those that are mouse specific, and those that are ambivalent, a three steps strategies were done. First interrogation was done as specify above at 10% FDR on the full Human database to generate a list of possible human protein database; then the same interrogation was done using Mouse database (UniProt accessed Mai 2019; 17,013 entries). Finally, two Fasta databases were generated from these interrogations and a final interrogation was done combining these databases at a FDR of 1%. In parallel, heavy labelling of arginine residue (Arg10) was used to identify and quantify the control stable isotope labelled diGly-pept (LIFAG [K (GG)]QLEDG [R (13C6; 15N4)]). The false discovery rate (FDR) at the peptide and protein levels were set to 1% and determined by searching a reverse database. For protein grouping, all proteins that could not be distinguished based on their identified peptides were assembled into a single entry according to the MaxQuant rules. Peptide intensities were extracted with Perseus program (version 1.5.1.6) from the MaxQuant environment (www.maxquant.org).

Bioinformatics analysis of MS data

Raw intensities were normalized according to values of the control stable isotope labelled diGly-pept (LIFAG [K (GG)]QLEDG [R (13C6; 15N4)]). Values corresponding to same protein and same site were merged using the average intensity value. Further, we evaluated the impact of missing values imputation on the analysis, and we found that this approach induced significant bias in the data; therefore, the downstream analyses were performed without imputation. For the next steps of analysis, a logarithmic transformation of the data (log2) was applied. To search for significant associations of ubiquitination profile to the PAMG and chemo-resistance to drugs, we performed correlation analysis based on Spearman rank correlation which quantifies the degree of linear association between the ranks the two variables. The calculation of Spearman's rho coefficients and p-values was performed using “rcorr” function of the “Hmisc” R package. Correlations with at least six observations and a p-value <0.05 were retained for next steps of analysis. Functional and enrichment analyses were performed using the lists of proteins containing the ubiquitination sites identified as significantly correlated (positively or negatively) and considered as markers. Three enrichment databases were used including KEGG (Kyoto Encyclopedia of Genes, version 103, 2022-07 release) and Reactome²⁷ pathways (version 81, 2022-06 release) along with Gene Ontology (GO, 2022-07-01 release) Biological Process (BP). These enrichments were performed using mainly “ClusterProfiler” and “ReactomePA” R packages.²⁸ Other packages and custom scripts were used for data exploration and visualization of the results based essentially on “ggplot2” R package. Survival analysis was performed using Cox proportional hazard regression for both uni- and multivariate models from “survival” R package. Time-dependent ROC curves were created using the “survivalROC” R package from censored survival data with default parameters; NNE (Nearest Neighbor Estimation) method at different time point (seven predict time). High and low groups of patients were created using a cutpoint on the continuous variable (ubiquitination level) using “surv_cutpoint” function of “survminer” R package with a minimal proportion of observations per group >40%. Survival curves Kaplan-Meier were then plotted using categorical variable of group of patients. Finally, forest-plots of multivariate analysis were created using “forestmodel” R package.

PDXs RNA-seq and PAMG calculation

The calculation of the pancreatic adenocarcinoma molecular gradient (PAMG) was performed using the RNA-seq datasets from the 60 PDXs included in this study as described previously in Nicolle et al.²⁹ Briefly, since in PDX model cancer cells are of human origin while stromal cells are of mouse origin, RNA-seq raw data were mapped on both the human (hg19) and mouse (mmu38) genomes. Quantification and establishment of expression profiles were performed using FeatureCount tool.³⁰ Normalization of data was done using upper-quartile approach. The calculation of the PAMG was done on the whole–transcriptomic profiles using the “pdacmolgrad” R package.

Proximity ligation assay and immunohistochemistry

Proximity ligation assays were performed following the Duolink ® PLA Brightfield Kit (Merk, DUO092012) Protocol. Unmasking of tissues was performed using the TRS9 protocol. Primary antibodies for PSMD2 (RRID:AB_2170472), ALDOA (RRID:AB_1078128), and for ubiquitin (RRID:AB_331292) were used at a dilution of 1/200 (PSMD2) and 1/500 (ubiquitin) for PSMD2-Ubiquitin PLA, and at 1/600 (ALDOA) and 1/300 (ubiquitin) for ALDOA-Ubiquitin PLA. For the negative control, the anti-ubiquitin primary antibody was omitted. All samples were stained in parallel. Regarding PSMD2-Ubiquitin PLA, each PDX were entirely imaged using till scanning at a magnification of 20× on a Zeiss Imager.Z2 microscope using same light and acquisition time parameters. Full images were analysed using FiJi (ImageJ, RRID:SCR_002285) software. DAB signal was extracted using colour deconvolution. Regions of interest (ROIs) with similar background were defined, applying appropriate threshold, then the number of dots with a size from 1 to 40 pixels and a circularity between 0.5 and 1 was counted in each ROI. The total number of dots was then reported to the full area used for quantification (more than 80% of the sample) and expressed as number of dots/10,000 pixel². For ALDOA-Ubiquitin PLA on tumour microarrays of patients samples (TMA) slides were entirely scanned at 40× magnification on a Nanozoomer 2.0-HT (Hamamatsu, RRID:SCR_021658). Images were then analysed with ZEN desk 3.5 software (Zeiss) using the Intellesis segmentation module trained to detect and count PLA specific dots (training and training images in Supplementary information). Only tumour compartments, delimited by a pathologist, (minus lumens when present) were considered for quantification.

Immunohistochemistry was performed using TRS9 protocol and anti-PSMD2 or anti-ALDOA antibodies at 1/200. Entire samples were imaged as previously except that a magnification of 5× was used for PDXs and 10× for TMA. Quantification of staining over full samples and their corresponding negative control (ROI covering more than 90%) was performed using FiJi, applying colour deconvolution to extract DAB signal and using a threshold set at 200 before measuring areas and mean grey values. Values from control staining were subtracted from PSMD2 or ALDOA values and results were expressed as mean intensity/pixel of the ROI surface.

Role of funders

The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the manuscript.

Results

Ubiquitin profiles from 60 PDAC xenografts and data analysis

To identify new molecular markers as well as potential new therapeutic targets for PDACs, we established and explored the ubiquitin profiles of 60 PDACs. However, because most patients (80–85%) already have advanced PDACs at diagnosis, they cannot be surgically removed, and only fine-needle aspiration (FNA)³¹ biopsies are available for these patients. This considerably limits the amount of biological material which is not sufficient to specifically enrich diGly-pept and to detect them by mass spectrometry. Hence, we first developed patient-derived xenografts (PDXs) from freshly obtained biopsies from both resected and non-resected tumours, best representative of all PDACs (Table S1). In addition to providing sufficient biological material, this model also reduces the proportion of stroma (Table S1), as observed in previous studies,³ which can represent up to 90% of the PDAC mass. Fractions of flash frozen PDXs were used for protein extraction, followed by digestion with Lys-C and trypsin, diGly-pept enrichment and analysis by LC-MS/MS, to generate a large-scale ubiquitination profiling of all patients (Fig. 1a). After normalization and filtering of the raw data (Table S3), we identified a total of 1031 diGly-pept from 634 unique proteins (Table S4 and Figure S2a). The distribution of diGly-pept according to their length was not uniform although it was Normal (Figure S2b) and generally diGly-pepts were between 10 and 25 amino acids long (Fig. 1b). Most (68.3%) of the identified proteins contained only one ubiquitination site, while a notable proportion of them contained two (18.1%), three (7.1%), or up to 14 ubiquitination sites (6.5%) (Fig. 1c). DiGly-pept intensities also displayed a Normal distribution (Figure S2c) with a mean log2 value of 31.21 (Fig. 1d). Since PDX model is a mix of human cancer cells and of mouse microenvironment, we controlled the relative proportion of diGly-pept that are human specific and those that cross with mouse and observed 47% human specific versus 53% ambiguous diGly-pept (Fig. 1e). However, the study of the relative proportion of human and mouse specific peptides within the raw data (Table S3) revealed a majority of human component with 42% human specific versus 18.8% mouse specific peptides (Figure S2d).

Fig. 1.

Ubiquitination profiling landscape in PDACs from 60 patients. (a) Illustration of experimental design used to generate the ubiquitination site profiles of 60 PDACs. Biopsies from patients with PDACs were grown in immunocompromised mice to produce PDXs. Theses PDXs were then processed to extract cellular proteins that were digested with Lys-C and trypsin proteases. DiGly-pepts corresponding to ubiquitinated sites were enriched using specific antibodies and were identified and relatively quantified by tandem mass spectrometry. (b) Distribution of diGly-pepts according to their length. (c) Distribution of the diGly-pepts according to the number ubiquitination sites identified in this study. (d) Distribution of diGly-pepts according to their global intensities. (e) Venn diagram showing the relative proportion of diGly-pepts that are human specific and those that are ambivalent human or mouse.

Ubiquitination site profiles correlation with the transcriptomic phenotype of PDAC

We recently established the pancreatic adenocarcinoma molecular gradient (PAMG), a prognostic molecular signature that is associated with the differentiation state and aggressiveness of tumours, thus allowing its grading (Figure S3).²⁹ We calculated the PAMG values of all the 60 PDXs and then studied to which extent some ubiquitination profiles could be associated with this parameter by performing correlation analyses between PAMG values and the different levels of ubiquitination at specific sites. We identified 38 ubiquitination sites for which the ubiquitination levels were significantly correlated with the PAMG (Table S5) (Fig. 2a). Out of these 38 ubiquitination sites, 15 were positively correlated, including ACLS K621, HSP90AA1 K546 and GAPDH K215, indicating that their ubiquitination level is increased in high PAMG, that is, less-aggressive tumours (Fig. 2b). Inversely, 23 ubiquitination sites were negatively correlated, such as HIST1H2AC K119/K120, CCT7 K55 and PTPRF K1376, meaning that these sites are more ubiquitinated in low-PAMG more-aggressive tumours (Fig. 2b). Note that the difference in the number of samples between selected diGly-pept is due to missing mass spectrometry data (see methods section). Hence, these ubiquitination sites could potentially be used as prognostic markers to evaluate the PDAC aggressiveness in patients. We further tested these markers with previously described molecular classifications^5,7,32 and observed that most were consistent (Figure S4). To discriminate between site specific and protein specific PAMG associated ubiquitination profiles we searched for the detection of other ubiquitination sites for each protein and found that more than half (22 out of 38) of the ubiquitination markers we identified were unique (Figure S5). For all other proteins containing two or more ubiquitination sites, we observed that some were not or inversely correlated to the main ubiquitination site. To reveal the biological functions in which these ubiquitinated proteins are involved, we searched for enriched terms using Gene Ontology (GO, Biological process), KEGG and Reactome databases. Interestingly, among them all (Table S6), we observed several enriched terms that are associated with differentiation state of pancreatic tissue, such as establishment of cell polarity, cell junction organization, the regulation of peptide secretion, cell cycle, as well as metabolic (i.e., glucose 6-phosphate, proteasome, or amino acids synthesis) and signalling pathways (i.e., interleukin or ROBO/SLIT signalling), all usually altered during the oncogenic transformation (Fig. 2c). Also, as shown by the gene-function network, some PAMG-correlated ubiquitination sites belong to key proteins, such as PSMD2 and PSMD4, involved in main biological processes and signalling pathways (Fig. 2d).

Fig. 2.

Correlation of the ubiquitination sites with transcriptomic phenotype in PDAC. (a) Volcano plot of the correlation results of all ubiquitination sites to the pancreatic adenocarcinoma molecular gradient (PAMG). Only ubiquitination sites significantly correlated to the PAMG are detailed (p-value <0.05, colour intensity). Ubiquitination sites at the right (red) have positive correlation while the one at the left (green) were negatively correlated (size of dots according to correlation coefficient). (b) Scatterplots showing examples of positively and negatively correlated ubiquitination sites with the PAMG (Figure S3) using a linear regression model and Spearman correlation (low values of PAMG are red and high values are blue). (c) Graphical representation of enrichment analysis results based on ubiquitinated proteins significantly correlated with the PAMG. Main enrichment terms are shown of different databases including Gene Ontology Biological Process (GO_PB), KEGG, and Reactome. Circle size corresponds to gene count in each enrichment term, and the colour corresponds to the adjusted p-value (p.adjust). (d) Gene-enrichment network of main enriched terms showing the functional relationship between different terms and involved genes.

Ubiquitination site profiles associated with patients' overall survival

The significant association of several ubiquitination profiles with the PAMG led us to explore the possible association between ubiquitination sites and the survival of PDAC patients. Therefore, we carried out overall survival analyses using univariate Cox regression on all ubiquitination sites that were detected in at least 70% of samples using the clinical data from the PDAC patient cohort.²⁹ As shown in Fig. 3a, we identified 23 ubiquitination sites that were significantly associated with overall survival as determined by the univariate survival analysis (log rank p-value <0.05). Out of these, 12 were associated with low survival (HR > 1) while 11 others were associated with better survival (HR < 1). To further evaluate the prognosis value of these 23 ubiquitination sites, we generated time-dependent receiver operating characteristic (ROC) curves using different survival time points. We used the ROC-associated AUC to filter out ubiquitination sites with an AUC <0.70. Thus, nine ubiquitination sites were selected and considered as potential prognostic markers (Fig. 3b and Figure S6). Furthermore, two groups of patients, high- and low-level ubiquitination, were created for all these candidates and used to perform a multivariate Cox regression analysis which also included resected and non-resected tumours (patients with resected tumours have better overall survival than non-resected ones). The results showed that, out of nine, four ubiquitination sites were still significantly associated with overall survival (p < 0.05) in the multivariate survival model. Those included PDCD6IP K501 (p = 0.002, HR = 0.12, 95% CI: 0.03, 0.46), ALDOA K330 (p = 0.02, HR = 0.28, 95% CI: 0.10, 0.80), NACA K142 (p = 0.03, HR = 0.46, 95% CI: 0.23, 0.93) and KRT17 K321 (p < 0.001, HR = 0.16, 95% CI: 0.06, 0.46) (Figure S7a). The prognosis values of these ubiquitination sites were still significant even when the resected and advanced tumours were included in the survival model (Figure S7a) as well as when covariates associated with patients (sex, age, tumour stage) were included in the model (Figure S7b). We further controlled the eventual dependency between ubiquitination levels and tumour stage (i.e., resected or not) in survival prediction using a Fisher's exact test and still observed that there was no significant relationship between tumour stage and ubiquitination level (p > 0.05) (Figure S8), confirming that the prognostic value of these four ubiquitination markers is not associated to the tumour stage. Finally, we used Kaplan-Meier method to display survival curves based on time to event data using the two groups of patients (high and low) for each ubiquitination sites. Surprisingly, for all the four ubiquitination sites, groups with high ubiquitination levels were characterized by poor clinical outcome while patients with low ubiquitination levels had improved outcome (Fig. 3c). Here as well, the difference in the number of samples between the four ubiquitin sites is the consequence of missing mass spectrometry values.

Fig. 3.

Identification of ubiquitination sites associated with survival of patients with PDAC. (a) Volcano plot showing univariate Cox regression analysis for each ubiquitination site. Only the ubiquitination sites (characterized by protein name and diGly conjugated lysine number) with significant p-value (log rank <0.05) are shown. The ubiquitination sites with a Hazard Ratio (HR) above 1 are associated with short survival while the ones with HR below 1 HR are associated with better survival. (b) Receiver operating characteristic (ROC) curves for the validation of the prognostic significance of the four selected ubiquitination sites. ROC curves were made at different survival time point and the value of the highest area under curve (AUC) is shown. (c) Kaplan-Meier curves of 4 ubiquitination sites (name of proteins and number of the ubiquitinated lysine) selected based on their p-value as well as their ROC curves shown in b. The patient cohort was divided into two groups, one with high (red) and one with low level of ubiquitination (purple).

Ubiquitination sites predicting the response to chemotherapies

As ubiquitinome profiling allowed us to identify potential ubiquitination markers of PDAC aggressiveness and potentially able to predict overall survival, we decided to search for ubiquitination site profiles that might indicate potential chemotherapy drug responses. It was shown that PDXs allow the in vivo study of the drug responses in a way that correlates well with clinical tumour responses observed in patients.³³ Hence, we randomly selected 13 PDXs with PAMG values covering all the range of this gradient (Table S7) and we established their resistance scores to the main cytotoxic drugs used to treat PDAC (i.e., gemcitabine and the 3 components of the folfirinox regimen: oxaliplatin, 5-fluorouracil, and irinotecan) as described in methods section (Figure S1). Of note, none of the patients from which the PDXs were derived underwent treatment prior to biopsy or tumour resection. We found several significantly correlated ubiquitination sites, either positively or negatively, with resistance score to 5-FU (four positive, three negative), gemcitabine (eleven positive, six negative), irinotecan (five positive, height negative) and oxaliplatin (four positive, two negative) (Fig. 4a and Table S8). Enrichment analysis using the resistance-associated ubiquitination sites of these drugs (Table S9) showed that several enriched functions and processes could be involved in resistance mechanisms such as the p53 pathway, DNA damage checkpoints, ABC transporters and apoptosis (Fig. 4b). The gene-function network highlighted the central role of some ubiquitinated proteins as well as biological functions whose alterations are often associated with resistant phenotypes (Fig. 4c). In conclusion, we have identified several ubiquitination sites which could potentially predict PDAC sensitivity to a given chemotherapy providing new opportunity for more personalized treatment in PDAC and, importantly, some of them are associated with drug resistance and represent promising molecular targets.

Fig. 4.

Ubiquitination sites associated with chemotherapeutic drug response in PDXs model. (a) Dot-plot showing all ubiquitination sites significantly correlated with the resistance score to four chemotherapeutic drugs including 5-FU, gemcitabine, irinotecan and oxaliplatin. (b) GO_BP and reactome terms enrichments results for all ubiquitination sites correlating with the different PDXs' resistance scores. Triangle and circle size corresponds to gene count in each enrichment term, and the colour corresponds to the adjusted p-value (p.adjust). (c) Networks of gene-enrichments and involved proteins for the four different drugs.

Experimental validation and potential clinical application

To confirm the ubiquitination site profiles identified by mass spectrometry, we used another technical approach on one selected theranostic marker, PSMD2, which displayed a positive correlation of its ubiquitination level with PDXs resistance to both 5-FU and irinotecan (Fig. 5a). We used PLA (proximity ligation assay) with specific antibodies to detect and quantify PSMD2 ubiquitination directly on paraffin embedded PDX samples. Positive PLA signals correspond to DAB (diaminobenzidene) dots as shown in Fig. 5b (Figure S9a). We quantified the average PSMD2 ubiquitination in all samples and could confirm a positive and significant correlation with both 5-FU (R = 0.62 and p-value = 0.07) (Fig. 5c and Table S10) and irinotecan resistant scores (R = 0.61 and p-value = 0.046) (Fig. 5d and Table S10). Since differences in PSMD2 ubiquitination levels could be due to differences in PSMD2 protein expression levels, we also quantified PSMD2 protein level, by IHC (immunohistochemistry), on the corresponding PDXs samples. As expected, there was no significant correlation between the ubiquitination levels of PSMD2 and its protein levels (Figure S9b and Table S11). We further validated the detection of PSMD2 on organoids (Figure S10a), an alternative preclinical model directly derived from patient biopsies, as well as on patient's samples (Figure S10b). As previously, a positive and specific PLA signal was detected. To validate the pertinence of establishing ubiquitinome profiles in order to detect clinically relevant markers, we used PLA to quantify ALDOA ubiquitination (identified as a prognostic marker) within the epithelial compartment of tumours, directly on paraffin embedded tissues from an independent cohort of 22 PDAC patients distributed on a tumour microarray (TMA) (Fig. 6a). We used Kaplan-Meier method to display survival curves based on time to event data using the two groups of patients with high and low ubiquitination of ALDOA (Fig. 6b and Table S12). As expected, the patient group with low ALDOA ubiquitination levels were characterized by a significant better clinical outcome compared to patients with high ubiquitination levels (p-value<0.001). This result was confirmed by univariate cox regression analysis (p = 0.002, HR = 0.13, 95% CI: 0.03, 0.48) (Fig. 6c). Since, like any PTM variation, these differences in ALDOA ubiquitination levels could result from variations in ALDOA protein levels, we quantified ALDOA protein by IHC (immunohistochemistry) on the same patient's samples (TMA). As expected, there was no correlation between ALDOA expression and ubiquitination levels (p = 0.62) (Figure S11a). Accordingly, there was also no association between ALDOA protein level and overall survival (Figure S11b and c).

Fig. 5.

Experimental validation of the ubiquitinome strategy and potential application to the clinic. (a) Scatterplots showing the positive correlation between levels of PSMD2 K27 ubiquitination determined by mass spectrometry and drug resistance score to 5-FU (n = 7) and irinotecan (n = 8) in the PDX model. (b) Typical images obtained after proximity ligation assay (PLA) detection of PSMD2 ubiquitination. Top: a PDX with low PSMD2 ubiquitination. Bottom: a PDX with high PSMD2 ubiquitination. Red arrows indicate positive PLA signals. (c) Scatterplot showing the correlation between quantified PLA signals of PSMD2 ubiquitination and the resistance score of PDXs to 5-FU (n = 9). (d) As in c with irinotecan resistance score (n = 11). R: Pearson's correlation; p: p-value of statistic test.

Fig. 6.

Validation of ALDOA ubiquitination as a prognosis marker in PDAC. (a) HES and PLA ALDOA-Ubiquitin images extracted from a tumour microarray (TMA) of 22 distinct PDAC surgical samples from an independent cohort of patient. HES stained TMA shows the delimitations of the tumour compartment (blue line) which were considered for PLA quantification. PLA pictures show a low and a high level of ALDOA ubiquitination. (b) Kaplan-Meier curves of ALDOA ubiquitination as determined by PLA quantification. The patients' cohort was divided into two groups, one with high (orange) and one with low (blue) level of ubiquitination giving the lowest p-value. (c) Forest plot evaluating the Hazard Ratios (HR) in the model of univariate analysis for ALDOA peptide ubiquitination level (low and high). A group of reference was set to HR = 1, and a HR < 1 indicates a decreased risk of the outcome (death). The mean value and quartiles of the confidence interval (CI) of 95% are shown.

Discussion

PDAC is a hard-to-treat and extremely complex disease, characterized by a vast heterogeneity. This heterogeneity calls for personalized treatments, and some molecular transcriptomic signatures are now able to predict PDAC aggressiveness, patient outcome,^7,29 and sensitivity to specific chemotherapeutic agents,^9,10,34 potentially orienting future first-line treatments. Pancreatic cancer treatment has entered a new era of personalized-medicine and the development of individualized treatment strategies is a growing field to improve patient management. This would help in better-informed decision-making improving benefit-risk ratio when treating patients with PDAC.

The ubiquitin-dependent proteome, and its disease-associated alterations, represents a promising and yet unexplored area of investigation to identify new markers and molecular mechanisms associated with PDAC. While efficient tools to study ubiquitinome from cell lines have existed for more than a decade, the ubiquitination profiling of biological tissues has recently significantly improved and allowed us to establish the ubiquitinome of PDAC samples with the final aim of identifying theranostic and prognostic markers and potentially new molecular targets.

Being able to predict the aggressiveness of a PDAC and the survival expectation of patients may be useful in future clinical practice. In this study, we identified many ubiquitination sites displaying either positive or negative correlations with PAMG (Fig. 2a and b), suggesting that these ubiquitination sites have the potential to estimate the aggressiveness of a PDAC. Exploring GO and KEGG enrichments, we noticed that these ubiquitinated proteins belong to diverse pathways that are associated with differentiation of epithelial cells such as the establishment of cell polarity, regulation of protein secretion or cell junction organization (Fig. 2), which is in accordance with PAMG definition.

Studying the potential association of ubiquitination sites with the overall survival of patients allowed the identification of four promising prognostic markers, PDCD6IP K501, ALDOA K330, NACA K142, and KRT17 K321, which discriminated two groups of patients with distinct outcomes. For all these ubiquitination sites, patients in the high ubiquitination group were characterized by poor survival outcome, independently of their resected status, sex, age or tumour stage (Figure S7b), highlighting their important clinical interest. However, since our data lack information regarding patients' treatments, a limitation of this model is that we could not include this covariate. Interestingly, regardless of their ubiquitination levels, these proteins have been previously associated with PDAC35, 36, 37 and breast cancer.³⁸ However, in our hands, the mRNA expression level of these genes was not associated with patients' survival (Figure S12) and this was also true regarding ALDOA protein level (Figure S11). Hence, our findings bring a new and valuable information to simple protein markers as their ubiquitination may be even more informative.

The next objective of this work was to identify ubiquitination sites that display correlation with drug response to the main chemotherapies used against PDAC. These could serve as predictive markers of sensitivity to specific drugs and, importantly, represent potential new molecular targets in resistant tumours. Indeed, some of these ubiquitination sites may involve important regulators of cell survival during treatment. We identified many ubiquitination sites showing positive or negative correlation with the resistance score of PDXs to gemcitabine and the three components of the folfirinox protocol: 5-FU, irinotecan and oxaliplatin. Strikingly, biological functions associated with these ubiquitination sites included apoptosis, DNA damage response and ABC transporters (responsible for drugs efflux), which are all known to be involved in drug resistance mechanisms. Hence, some drug resistance-associated ubiquitination markers identified here also represent potential molecular targets that could be used to sensitize pancreatic cancer cells to chemotherapies. Restoring their normal ubiquitination level could help to reverse resistance to anticancer drugs. Importantly, while all the ubiquitination markers identified in this study were of PDAC origin, they are not necessarily specific for PDAC. Indeed, it is probable that some of the ubiquitination profiles, and especially those associated with biological function involved in resistance mechanisms, can be found in other form of resistant cancer. Also, while very efficient in identifying and quantifying ubiquitination sites, one limitation of the “diGly” proteomic approach is that it cannot discriminate between the different types of ubiquitination (mono versus poly for example). Consequently, the ubiquitination markers we identified in this study can be of any kind. While not critical since our main objective was to identify useful ubiquitination profiles, future molecular investigations of chosen markers will reveal the exact type of ubiquitination involved. An important aim of this study was to confirm that the ubiquitination markers we identified by mass spectrometry and bioinformatics analyses can be also detected by other methods. Among all candidates, PSMD2 was of particular interest for several reasons. First, its ubiquitination at K27 was positively correlated with the resistance score of two drugs, 5-FU and irinotecan. Second, certainly due to its role as a non-ATPase regulatory subunit of the proteasome, it is involved in several biological processes that can significantly impact cancer cell survival upon treatment, such as cell death, P53 stabilization or ABC transporters (Fig. 4), and thus PSMD2 represents a new potential molecular target. Even though the role of its ubiquitination is not known, PSMD2 binds ubiquitin and ubiquitinated substrates,³⁹ favouring the degradation of important proteins such as P21 and P27⁴⁰ and hence playing a central role in oncogenic processes. Finally, PSDM2 ubiquitination has been already highlighted in other ubiquitin profiling studies^19,41 and, most importantly, we previously detected its increased ubiquitination in a PDAC cell line (MiaPaCa-2) upon treatment and acquisition of resistance.^42,43 Using PLA to detect PSMD2 ubiquitination in paraffin-embedded PDX samples, we could confirm the significant correlation between the ubiquitination of PSMD2 and the resistance scores to 5-FU and irinotecan (Fig. 5c and d). Additionally, while a majority of the ubiquitination markers we identified in this study should be of human origin (epithelial cancer cells) (Fig. 1e and Figure S2d) some might be of mouse origin (stromal cells). Though stromal markers may be as valuable as cancer cell markers,⁶ the spatial detection of these markers by PLA also makes it possible to determine their origin, as for PSMD2 and ALDOA. Additionally, we tested and validated the detection of ubiquitinated PSMD2 by PLA on organoids (Figure S10a), a technic that allows the in vitro culture of patient biopsies (even with low cellularity), as well as on a paraffin-embedded surgical biopsy from a resected PDAC (Figure S10b). Hence, it seems that any validated ubiquitination marker could be used in clinical settings and in preclinical models.

While PDXs and organoids are useful preclinical models, they do not fully replicate the biology of a patient's tumour. Hence, we verified that the ubiquitination markers we identified in this work are also transposable to patients' tumours. We chose a survival associated marker since patients are often treated with several chemotherapeutic drugs, making it impossible to know which one they were sensitive to. As evidenced by the study of ALDOA ubiquitination in an independent cohort of patients (Fig. 6), we could show that at least some of these new kind of PDAC associated markers can effectively be used directly on human samples to help the categorization of patients and orienting their therapy.

Our findings prove that ubiquitination is a valuable reservoir of potential markers and molecular targets that merits to be considered in PDAC as well as in other cancers. This work provides broad visualization on the PDAC ubiquitinome landscape and shows how ubiquitination sites can be associated with patient prognosis and chemotherapeutics responses. Finally, while ubiquitination site specific antibodies may be engineered in future, PLA could be used to detect these ubiquitination markers in a relatively simple manner, and hence would be applicable to the clinical routine as part of a precision medicine strategy, once their sensitivity and specificity clearly demonstrated.

Contributors

N.J.D. and P.S. designed research; A.E.K., N.A.F., S.A., L.C., Y.B., O.G., J.R., M.B., C.B. and P.S. performed research; A.E.K., N.A.F., S.A., L.C. and P.S. analysed data; and A.E.K., J.C., J.L.I., N.J.D. and P.S. wrote the paper. All authors read and approved the final version of the manuscript. The underlying data was verified by A.E.K, N.A.F, S.A., L.C. and P.S.

Data sharing statement

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE⁴⁴ partner repository with the dataset identifier PXD030011 (Username: reviewer_pxd030011@ebi.ac.uk, password: 02QrVujm). The remaining data are available within the Article, and Supplementary Informations.

Declaration of interests

N.A.F., J.L.I. and N.J.D. have a pending patent entitled “Simple transcriptomic signatures to determine chemosensitivity for pancreatic ductal adenocarcinoma”. All other authors declare no conflicts of interest.