A wide range of pancreatic cancer (PC) phenotypes causes difficulties in PC research regarding its diagnostics, treatment, and prognosis, due to ambiguous, not-subtype-specific results. PC shows a variety of phenotypes, regarding its histology (morphology), genetic alterations, and epigenetic modifications (EMs). The subtyping appreciates different aspects of the PC picture, consequently, molecular (transcriptomics), clinical, or histological subtypes were distinguished 1. Studies revealed different biology of some of them, resulting in different prognoses and treatment efficiency 1–7.
Although it might be hard to correlate and link together different systems of PC subtyping, some attempts have been made 1,8. Nevertheless, important practical aspects entitle PC histological subtypes to be ahead of others. The routine pathological practice, which establishes the details of PC diagnosis used in clinical management, exploits histopathological subtyping. It is partially caused by the cost-effectiveness of routine diagnostics which currently does not allow for extended molecular testing of each case. The antibody-based assays for histone and DNA modifications are being developed 9, however, caution must be taken when introducing new techniques and proper standardized procedures should be implemented before 10. Transcriptomic subtypes distinguished by Collison et al 2, Moffit et al 3, or Bailey et al 4 were being highly exploited in PC research 1,11. Currently, however, they are not reflected in pathological diagnosis or clinical patient management procedures. Recognizing those subtypes is largely based on expensive RNA profiling, even though other approaches were proposed 11. Moreover, only a few molecular subtypes were distinguished, whereas observed morphological and biological presentations of PC range much wider (over thirteen WHO and four non-WHO subtypes 5,12). Experience learned from the classification of central nervous system tumors based solely on the profile of DNA methylation taught us the potential of EMs in practical tumor subtyping 5,13. It showed the multitude of these tumors and revealed new entities, that had not been recognized before 13.
PC emerges in a well-documented fashion, originating as pancreatic acinar cells, through acinar-to-ductal metaplasia, or as pancreatic duct epithelial cells. The initial process of carcinogenesis, driven by genetic alterations is followed by subsequent modifications (i.e., mutations in KRAS, TP53, CDKN2A, or SMAD4), depicted as dysplastic changes in flat lesions called pancreatic intraepithelial neoplasia (PanIN) or cystic lesions, such as the most prevailing intraductal papillary mucinous neoplasm (IPMN) 5. However, genetics alone does not sufficiently reflect a variety of PC subtypes 14.
The emerging role of epigenetic alterations in PC has been recognized. Recently, we linked some of the most common PC histological subtypes with variabilities in DNA methylation 15. Moreover, recent studies on targeted therapy approaches focus on that particular aspect of cancer evolution 16. Indeed, epigenetic alterations are excellent targets for PC therapeutics because they are reversible 16. Notably, the change in PC epigenome is very dynamic, not only in highly proliferating cancer cells but also as a reaction to the tumor environment and its host showing a modus operandi that even justifies a travesty of a “mafia” within the “society of the body” 17. The tumor-stroma crosstalk, another highly exploited subject of PC research, impacts the tumor epigenome 18. This drives cancer spreading (meaning the ability to invade and metastasize), a logical consequence of the theory of evolution and its natural selection mechanisms 19.
Recent advances in cancer research reveal three main dimensions of transcriptomic machinery. First is the genome, thus the nucleotide sequence, which is prone to rearrangements, duplications, deletions, and other alterations 20. In the past, it was thought that DNA sequence solely determines basic transcription into proteins. However, at least two other overlying mechanisms alter the amino acid sequences in the resulting proteinogram. Local DNA three-dimensional (DNA conformation) structure directly affects the binding of different proteins involved in the DNA transcription process 21,22. Finally, DNA and histone modifiers affect chromatin modeling, thus blocking or favoring the transcription of certain DNA sequences 23.
DNA appears in different secondary structures, known as conformations. The most prevailing is the right-handed double helix B-DNA, which builds up the DNA backbone 24. Other double helix conformations (non-B) include right-handed A-DNA and left-handed Z-DNA. Additional DNA structures include DNA bubbles, cruciform, three-stranded H-DNA, four-stranded G-quadruplexes, and others 24,25. Cancer research has been exploiting Z-DNA in the context of new therapeutic approaches 26. Recently, the Z-form of DNA was shown to be involved in cancer’s ability to suppress the apoptosis determined by the immune response – an escape mechanism of cancer cells from the so-called immune checkpoint blockage (ICB) therapeutics 27. Proteins containing the Z-DNA-specific domains (called Zα domain) such as the adenosine deaminase acting on RNA 1 (ADAR1) or Z-DNA binding protein 1 (ZBP1) 28–31 selectively bind to the Z-DNA form. Activation of ZBP1 leads to autoinflammation and necroptotic cell death, although it is negatively regulated by ADAR1 32–34. Cancer cells use ADAR1 to induce immune silencing and prevent termination, thus bypassing the ICB therapeutic effect 28. Recognition of Z-DNA distribution in PC might provide insight into the cancer-induced ADAR1 activation.
EMs are posttranslational processes crucially important in PC biology. The constant interplay between EMs leads to a variable gene expression, which is dynamically regulated as a reaction to the environment or the cellular requirements (i.e., highly proliferating cancer cells) 16. EMs include DNA and histone modifications. DNA methylation (DNA-m) most commonly occurs in areas of cytosine-guanine dinucleotides (so-called CpG islands) and controls the transcription of specific genes, increasing or blocking it. On the other hand, histone modifications locally alter the state of chromatin, causing its transformation into dense heterochromatin or loosened euchromatin. The latter allows DNA transcription and gene expression. EMs are driven by enzymes that add or remove methyl or acetyl groups to CpG islands or histone amino acids. These are grouped into so-called “writers”, which include DNA methyltransferases, histone lysine methyltransferases, and histone acetyltransferases. Another group called “erasers” contains DNA demethylases, histone lysine deacetylases (HDAC), histone lysine demethylases (KDM), and protein arginine methyltransferases. Consequently, histone acetylation leads to activating the transcription, whereas histone methylation results depend on the site of residue and the degree of modification 16.
Blocking some of the enzymes responsible for EMs in PC (such as specific HDACs or KDMs) are considered potential targets currently in clinical trials. This emerging role of epigenetics profiling in cancer therapy was excellently reviewed in 14 or 35. Another possible therapeutic approach involves blocking ADAR1 or activating ZBP1, to support the ICBs effect and re-enable cancer cells’ apoptosis determined by the immune response 26,28. Nevertheless, to successfully utilize these drugs in PC therapy, researchers must properly address the PC heterogeneity and its subtyping, because of suspected variable response to them. Indeed most trials of PC therapy to date ended with unsatisfactory results 35.
Here we aimed to spatially recognize EMs and DNA conformations in PC concerning its histological subtypes. To do that we employed an innovative approach relying on Raman hyperspectral mapping (RHM) combined with advanced machine learning techniques, including unsupervised autoencoders (AE) and convolutional neural networks (CNN). RHM is a method of unlabelled molecular imaging based on Raman spectroscopy, that allows studying cancer tissues with submicrometric resolution, visualizing cellular components up to even individual chromosomes 36,37. Adjacent pixels of spectral measurements are merged to form a so-called hyperspectral image. The information collected with RHM contains molecular interactions in the studied sample in an all-in-one manner, which authorizes the conclusion of the sample molecular structure and, thus the contents of DNA or proteins, such as histones, and their secondary structures. Nevertheless, the required know-how of obtaining high-quality RHM measurements and complex analysis of the resulting data, until recently, prevented the usage of RHM in high-class cancer tissue explorations. Only a few reports are available describing DNA-m using Raman and infrared spectroscopy techniques 38,39 and some attempts have been made to detect histone acetylation 40,41. These studies though, primarily examined isolated DNA strands or individual cells rather than complex tissue samples. Recently, our group reported on DNA-m patterns among some PC histological subtypes obtained with RHM 15.
In this study, PC tissues of six histologically different subtypes were compared to benign control pancreatic ductal tissues (CTRL). Specifically, we evaluated conventional ductal adenocarcinoma (cPDAC), adenocarcinoma derived from intraductal papillary mucinous neoplasm (IPMC), predominantly foamy-glands carcinoma (FG), predominantly large duct type carcinoma (PLD), and squamous differentiated adenocarcinoma (SD). These are the most common subtypes of ductal adenocarcinoma, which is the most common form of PC 12. Furthermore, we added for the comparison the ampulla of Vater adenocarcinoma (AVAC), which is a cancer entity not classified as the form of PC in WHO, nevertheless, it develops in the pancreatic main duct opening into the duodenum (the ampulla of Vater). The diagnosis of AVAC vs PC is frequently histopathologically challenging, because of similarities in their morphologies and immunoreactivity 15. Thus practically, both PC and AVAC are considered while diagnosing a patient with a pancreatic tumor.6,7
Above all, we spatially show and semi-quantify the heterogeneity of EM in PC. Especially, we recognized and analyzed the distribution of DNA-m, methylated lysine (Lys-m), acetylated lysine (Lys-a), and methylated arginine (Arg-m). Additionally, we complete these findings with the results of Z-DNA and B-DNA conformation distributions. Consequently, we identify FG and SD PC subtypes as potentially less vulnerable to epigenetic regulator-targeting therapies.