Background
Pancreatic ductal adenocarcinoma (PDAC) remains one of the least understood malignancies to date, resulting in lack of targeted and immune-based approaches. Though molecular characterization and subtyping into two major subtypes [
1‐
3], classical and basal-like/quasi-mesenchymal, have brought transcriptome-based patient stratification into focus, the underlying biology of PDAC with regard to progression and therapy resistance remains largely elusive [
4]. Desmoplastic stroma that surrounds and modulates the phenotype of cancer and immune cells within the tumor microenvironment is a hallmark of PDAC and likely a direct modulator of disease progression [
5]. The stroma, consisting mainly of extracellular matrix (ECM), blood vessels and ECM-producing fibroblasts, creates a setting that communicates with cancer and immune cells and modulates tumor growth, metastasis and drug resistance. Identifying molecules differentially expressed in the stroma and cancer cell compartment may allow better understanding of cancer–stroma communication and development of technologies for compartment-specific targeting and stromal modulation.
Glycans are sugar moieties, oligosaccharides or polysaccharides, that when attached to proteins and lipids create glycoconjugates. Glycans present very diverse structures, have shorter or longer, branched or linear sugar chains and are additional modified by sulfatation, phosphorylation, etc. Depending on how the sugars are attached to the protein/lipid molecule, glycoconjugates are clustered in four general groups: (i)
N-linked glycoproteins where sugar motifs are attached to asparagine, (ii)
O-linked glycoproteins where sugars are attached to serine/threonine, (iii) heavily glycosylated proteoglycans with one or more glycosaminoglycans (GAG) attached and (iv) glycolipids. It is estimated that 50–70% of all human proteins are post-translationally glycosylated and correct glycosylation is essential for proper cellular localization and function of the protein or lipid. Glycosylation fundamentally influences protein trafficking, stability and forming, and many proteins and lipids involved in essential physiological processes such as cell–cell communication, cell–ECM communication, protein folding and signal transduction are glycosylated. It has been shown that in tumors, including PDAC, altered glycosylation is involved in tumor-related processes of cell adhesion, proliferation, invasion, metastasis and angiogenesis [
6,
7]. Typical cancer-associated glycosylation changes include sialylation, fucosylation,
O-glycan truncation and
N- and
O-linked glycan branching [
7], all present in PDAC as well. For example, interaction of tumor-associated sialyl Lewis antigen (SLe
x) with endothelial adhesion molecules selectins allows extravasation of cancer cells and metastatic spread [
8]. In PDAC, abundance of SLe
x predicts development of liver metastasis [
9]. Truncated
O-glycans have been detected on EGFR in pancreatic cancer [
10], increased
N-glycosylation and branching on integrins and ECM proteins [
11], and increased fucosylation is detected in serum of PDAC patients [
12]. Furthermore, the main glycan synthesis pathway, the hexosamine biosynthesis and consecutive
O-glycosylation are strongly upregulated in highly resistant hypoxic PDAC cells [
13].
Tumor stroma supports the cancer progression in multiple ways. Growth factors, adhesion molecules, nutrients are all produced by the stromal cells and locally provided to the proliferating cancer cell. There is whole plethora of interactions observed between glycosylated stromal proteins and cancer cells, all serving the function of cancer progression, invasion and metastasis (reviewed in [
14]). Stromal ECM provides a settling niche for cancer and immune cells, and next to collagen fibers, glycosylated proteoglycans are the main component of tumor ECM. Proteoglycans are comprised of one or several glycosaminoglycans (GAG) with repeats of chondroitin sulfate (CS), heparan sulfate (HS) and dermatan sulfate (DS) covalently attached to protein core. Proteoglycans are mainly produced by the cancer-associated fibroblasts (CAFs) in the stroma and play a role in intercellular and ECM interactions via activation of receptor tyrosine kinases (EGFR, FGFR, IGF1R, INSR) on the cancer cells [
14] regulating the proliferation and survival cascaded. Furthermore, stromal proteoglycans communicate with cancer cell membrane integrins and regulate cells motility and invading potential [
14].
Due to their cancer-specific expression and function, proteins with altered glycosylation are highly attractive as potential biomarkers, targets for therapeutic agents and targeted drug delivery [
15]. The diversity of glycosylated proteins and lipids and their functions in cancer opens these perspectives. However, which glycans are typically enriched in cancer cells and which specifically in the stromal compartment, respectively, has not been addressed yet. PDAC is one of the most stroma-rich solid tumors where desmoplastic reaction consumes 50–80% of the tumor tissue and can even be stratified to “normal” and “activated” stroma with later having worse prognosis [
16]. Defining glycans that are differentially expressed in cancer cell or stromal compartment may improve our understanding of the stroma–cancer cell interaction and indeed allow better, compartment targeted therapeutical approaches.
In this work, we report on high mass resolution matrix-assisted laser desorption/ionization Fourier-transform ion cyclotron resonance mass spectrometric imaging (MALDI-FT-ICR-MSI)-based detection of prominent differences in native glycan fragments distribution among cancer and stromal compartments in human PDAC FFPE tissue samples. Previously performed glycan mass spectrometry studies on PDAC tissue used digestion methods for release of
N-glycan structures from the protein core [
17]. Enzymatic degradation of tissue increases measurability but hinders the differentiation among naturally occurring glycans and those previously bound [
18]. Our MALDI-FT-ICR-MSI is performed on FFPE PDAC samples without previous tissues digestion with PNGase F or other biochemical process allowing detection of naturally occurring native glycan fragments, products of tissue-specific proteoglycan/GAG degradation. We classify differentially expressed glycans in cancer cells and the tumor stroma, which may have prognostic and functional relevance in PDAC progression. This glycan map may thus serve as a reference for further glycan-based diagnostic and targeting approaches.
Methods
Tissue microarray (TMA) preparation
Tissue microarrays (TMAs) were constructed using formalin-fixed paraffin embedded resected Pancreatic Ductal Adenocarcinoma (PDAC) specimens from 109 patients. The study was approved by the ethics committee of the Technical University of Munich, Germany (documents no. 1926/2007 and 126/2016S). Written informed consent was obtained from every patient. Representative tumor areas were identified by an experienced pathologist, and three cores with 1 mm diameter from each sample were transferred into the microarray.
MALDI-FT-ICR MSI analysis and data processing
The FFPE samples were cut into 3 μm sections on a paraffin microtome (HM 355S, Microm, Thermo scientific), mounted onto the ITO-coated glass slides. The FFPE sections were incubated at 60 °C for 1 h, deparaffinized in xylene (2 × 8 min) and dried on a hot plate at 37 °C. The matrix solution consisted of 10 mg/ml 9-aminoacridine hydrochloride monohydrate (9-AA) (Sigma-Aldrich, Germany) in water/methanol 30:70 (v/v). SunCollect™ automatic sprayer (Sunchrom, Friedrichsdorf, Germany) was used for matrix application. The flow rates were 10, 20, 30 and 40 μl/min, respectively, for the first four layers. The other four layers were performed at 40 μl/min. The MALDI-MSI measurement was performed on a Bruker Solarix 7T Fourier-transform ion cyclotron resonance mass spectrometer (FT-ICR-MS) (Bruker Daltonik, Bremen, Germany) in negative ion mode using 50 laser shots per spot at a frequency of 500 Hz. The MALDI-MSI data were acquired over a mass range of m/z 50–1000 with 50 μm lateral resolution. The acquired data from the tissue samples were underwent spectra processing in FlexImaging v. 4.0 (Bruker Daltonics, Germany). MALDI-MSI data were normalized to the root mean square of all data points. Cancer cell and stroma regions were annotated as regions of interest (ROIs). The average spectral data of ROIs were then exported to peak list as.csv files from FlexImaging software.
MATLAB
® R2014b (v.7.10.0, Mathworks, Inc., Natick, MA) was used as MALDI spectral preprocessing tool for the subsequent data bioinformatics analysis. The exported.csv files including all mass spectra from cancer cell and stroma regions are loaded into MATLAB
® R2014b and underwent resampling, smoothing and baseline subtraction to lower the data dimensionality and to remove the noise-level peaks and artefacts. Peak picking was performed using an adapted version of the LIMPIC algorithm [
19] with
m/
z 0.0005 minimal peak width. The signal-to-noise and intensity threshold were set to 2 and 0.01%, respectively. Isotopes were automatically identified and excluded.
Kaplan–Meier analysis and statistical analysis
In order to determine the prognostic power for each identified glycan, the individual patient glycan fragment abundances were used to split the cohort into good and poor survivor groups by the application of intensity cutoffs, which were optimized to the clinical endpoint. Cutoff-optimized survival analyses were performed as previously described [
18,
21] using a Kaplan–Meier Fitter and log-rank test. Cutoff optimized in this context means that the threshold for low and high abundance of a compound was chosen such that the
p value in the resulting Kaplan–Meier curve is minimal. Cutoff points and patients survival table are given in Additional file
2: Table S1. Overall survival rates were calculated using the Kaplan–Meier method and included 95% confidence interval estimates. Survival curves were tested with the log-rank
χ2 value and Cox proportional hazards regression analysis. Univariate and multivariate analysis was performed using Cox proportional hazards regression models. In a Cox regression, the coefficients of predicate variables relate to hazard and the hazard ratio (HR) of a predicate variable is given by the exponent of its coefficient. All statistical analyses were performed within the R statistical environment (R Foundation for Statistical Computing, Vienna, Austria), in which
p values ≤ 0.05 were considered statistically significant.
Through factor analysis, a smaller set of analyzed variables could be group into “factors” with common characteristics [
22]. Factor analysis provides a way of extracting underlying “factors” or “factor constructs” which accounts for the inter-correlations among the variables [
23,
24]. It serves as a good procedure to identify the factors that summarize the “good” or “poor” prognostic glycan fragments identified in MALDI-MSI analysis. Previous studies have attempted to apply similar analysis for identifying latent factors and to use these factors to stratify patients for survival analysis [
25]. In this study, the analyses were done using factor analysis function (“fa”) from R “Psych” package. Default factoring method (“minres”) and rotating method (“oblimin”) were applied. To explore the optimal factor numbers to extract, fa.parallel function was applied for the “screen plot” which helps determining the factor numbers to extract. Factor analysis generates factor scores for each unraveled factor. As a summarization of the variation of a set of related original variables, factor score represents the variation of extracted “factor” variable among samples and can be used in subsequent analyses [
26]. In this study, factor scores are applied in Kaplan–Meier analysis and multivariate survival analysis.
Discussion
Considering the late diagnosis, high intrinsic and acquired resistance to chemotherapies and the resulting high mortality in PDAC, identification of early disease biomarkers and novel targets are a priority task to improve clinical outcome. Recent studies suggest that the heavily developed tumor stroma in PDAC is actively participating in PDAC biology and influences the clinical outcome [
16]. Particularly cancer-associated fibroblasts (CAFs) not only produce the ECM but actively support inflammatory and migratory actions of cancer and immune cells [
27]. Knowledge about molecules specifically expressed on stromal and cancer cell compartment may allow development of strategies for disruption of cancer cell–stroma synergy and provide clinical improvement. Protein and lipid glycosylation is prominently altered in cancer [
7]. Carbohydrate antigen CA19-9, the only clinically relevant PDAC biomarker used for therapy follow-up, is a tetrasaccharide cleaved from membrane glycoproteins of epithelial cancer cells and released into the blood. Very recently, a prominent study with targeted CA19-9 overexpression in pancreas of genetically engineered mice suggested not only a biomarker but also a functional role of tissue CA19-9 in progression of pancreatitis to PDAC and highlighted CA19-9 as a therapeutic target [
28].
To date, most of glycan research performed in PDAC focuses on serum biomarkers and specificities in PDAC tissue glycosylation patterns have been rarely addressed. Latest work showed that C1GALT1 controlled glycosylation of cell surface integrins in PDAC cells leads to increased invasiveness and poor patient survival [
29] and a very recent MALDI-IMS study performed on formalin-fixed human PDAC FFPE specimens reports on distinct
N-glycan populations presented in healthy pancreas and PDAC tissue [
17] with high sialylation, poly-LacNAc extensions and fucosylation of high mass
N-glycans specifically detected in PDAC. Due to high abundance and functional significance of stroma in PDAC, we postulated that different glycoproteins, GAGs and proteoglycans may be present in the two compartments, cancer cells and stroma, and their biological degradation may generate different glycan fragments. Identification of those fragments may help in addressing the biological differences and functionality of the two compartments.
We used a MALDI-MSI approach on a collection of 109 PDAC patient FFPE samples where the respective stromal and cancer cell compartment was annotated and separately analyzed. Differently than in mentioned recent studies where tissue digestion was used for release on
N-glycan bound sugars [
17], we used an MALDI-FT-ICR MSI with high mass resolution and accuracy that does not demand tissue pre-treatment. This method was previously optimized by our group [
18,
30] and allows detection of naturally occurring unbound, native glycans in FFPE fixed tissues. We identified 8 and 18 glycan fragments uniquely expressed in only stromal or only cancer cell compartment in PDAC, respectively. In both compartments, specific glycans predicting worse or better patient survival were identified, suggesting functional implications of the identified glycans in cancer. It is, however, still difficult to interpret the origin of the identified glycan fragments as they can be biodegradation products of multiple complex proteoglycans and glycosaminoglycans (GAG). Hyaluronan and chondroitin sulfate are GAGs that show 12- and 22-fold increase, respectively, in PDAC in comparison with normal pancreas [
31]. Hydrolytic cleavage of hyaluronan yields HexA-HexNAc and HexNAc-HexA-HexNAc fragments [
18]. High intensity of those fragments in stroma and tumor, respectively, predicted worse survival in our study, going along with the current view that hyaluronan maintains high interstitial pressure in PDAC stroma, thus hampers the delivery of drugs and imposes lower therapy response and worse prognosis [
32]. Enzymatic digestion of stromal hyaluronan with hyaluronidases improved delivery of chemotherapeutic drug Gemcitabine into the murine PDAC and prolonged survival [
32,
33]. It should also be noted that hyaluronan and chondroitin masses cannot be distinguished at the full mass mode used here. Thus, the mass identified as hyaluronan may as well be chondroitin and must be carefully interpreted.
Chondroitin sulfate (CS) is not only very abundant GAG in the PDAC but also shows alterations in the sulfatation pattern of disaccharide chains with dramatically less abundant 4-sulfated disaccharides and more abundant non-sulfated disaccharides than in normal pancreas [
31]. Hydrolytic cleavage of CS yields HexS, HexNAcS, HexA–HexNAc, HexA-HexNAcS and HexNAc-HexA-HexNAc fragment signals [
18], and these fragments have different predictive values in PDAC tumor and stromal regions. High abundance of HexS in stroma predicts better survival, while HexNAcS and HexAHexNAcS abundance predicts worse survival. In the cancer cell compartment, chondroitin abundance predicts poor survival. At this point, it is difficult to explain those differences, but they might indicate differential activities of sulfatases and sulfotransferases that are responsible for addition and removal of sulfate groups from saccharides creating differently abundant sulfated and non-sulfated glycan fragments in cancer cell and stromal compartment.
It is also important mentioning that we detected Kdo glycan fragments in the cancer cell compartment. Kdo glycan fragments are parts of lipopolysaccharide (LPS) in the outer cell membrane of gram negative bacteria. Recent publications support the existence of intratumoral microbiome in PDAC and even suggest active influence of this microbiome on tumor growth and immune infiltration finally resulting in patient survival differences [
34]. The Kdo fragments identified here may originate from the gut microbiota since the gut is in the direct vicinity of the pancreas and there may be surgical cross-contamination between gut and PDAC during biopsy but may also originate from the tumor-specific microbiota. It is tempting to speculate that MALDI-based identification of Kdo fragments may be used for defining the specific intratumoral microbiota. This, however, remains to be addressed in the future studies.
Our multivariate analysis incorporating UICC classification demonstrated that distinct glycan patterns are independent prognostic factors for PDAC patients. Thus, the identified factors and glycan groups may potentially serve clinical prediction purposes. Glycosylated proteins located on the extracellular side of the plasma membrane coordinate the interplay between the cancer cells and the ECM in the tumor microenvironment [
14]. Considering the limited number of stroma/cancer cells differentiating glycans determined in our study, these glycans might serve as starting point for further functional analysis.
Declarations
Competing interests
WW has attended Advisory Boards and served as speaker for Roche, MSD, BMS, AstraZeneca, Pfizer, Merck, Lilly, Boehringer, Novartis, Takeda, Bayer, Amgen, Astellas, Eisai, Illumina, Siemens, Agilent, ADC, GSK and Molecular Health. WW receives research funding from Roche, MSD, BMS and AstraZeneca, all outside the submitted work; JTS reports the following disclosures: Bristol Myers Squibb, Celgene, Roche (Research Funding); AstraZeneca, Bayer, Bristol Myers Squibb, Celgene, Immunocore, Novartis, Roche, Shire (Consulting or advisory role); AstraZeneca, Aurikamed, Baxalta, Bristol Myers Squibb, Celgene, Falk Foundation, iomedico, Immunocore, Novartis, Roche, Shire (honoraria); minor equity in iTheranostics and Pharma15 (< 3%) and member of the Board of Directors for Pharma15, all outside the submitted work. KS is member of the advisory board of TRIMT and received research funding from Roche, all outside the submitted work. KS has a patent filed for a radiopharmaceutical compound. MS served as speaker and scientific advisor for Roche, MSD and BioNTech.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.