Braxton and colleagues used 3D histological reconstruction and multi-region genomic analysis to understand the microanatomy and model the progression of pancreatic intraepithelial neoplasias (PanINs), precursors of pancreatic ductal adenocarcinoma (PDAC). Their study revealed the unexpected prevalence, spatial distribution, and genetic heterogeneity of PanINs, illuminating their developmental trajectories toward PDAC.

Advanced pancreatic ductal adenocarcinoma (PDAC) is notorious for its resistance to treatment, with limited benefit provided by chemotherapy, immunotherapy, and inhibitors of mutant KRAS, the hallmark PDAC oncogene.1 Elucidating the early disease stages is needed to advance early detection and intervention strategies, ultimately enhancing outcomes. The primary precursors of PDAC, pancreatic intraepithelial neoplasias (PanINs), are microscopic lesions associated with the pancreatic ducts, typically harboring KRAS mutations. Clonal and morphological analysis of different lesions within the same pancreas indicates that PanINs undergo multi-step histological and genetic evolution to invasive PDAC.2,3 However, PanINs are detectable only incidentally in surgical specimens. Hence, serial monitoring is impossible, limiting definitive understanding of their evolutionary path. In addition, since PanINs spread three-dimensionally along the pancreatic duct, tissue slides incompletely capture the overall presentation, making it difficult to profile the entirety of distinct lesions and to assess their total burden and spatial relationship to one another.

To overcome these hurdles, Braxton and colleagues developed a novel framework to reconstruct the spatial architecture of PanINs within the pancreatic parenchyma at single-cell resolution from serially sectioned H&E slides of large tissues (Fig. 1).4 They integrated these data with phylogenic clonal analysis based on mutational profiles. Their approach defined the size, shape, and number of PanINs in the non-cancerous pancreas and established the genetic relationship between different regions within individual PanINs and across distinct PanINs, thereby providing insights into PDAC initiation.

Fig. 1: Framework for understanding the spatial evolutionary trajectories of PanIN.
figure 1

Machine learning-based cell type identification through serial tissue sections (left panel) allows for the 3D reconstruction of the micro-architecture of pancreatic histology, providing a deeper understanding of PanIN, including its spatial distribution and prevalence (middle panel). Integrating this with fine geographic mutational mapping, both sequence-based and in situ hybridization-based, further enables the phylogenetic analysis of PanIN cells (right panel). This approach elucidates the genetic relationships within and between different regions of individual PanINs, offering insights into PDAC initiation. ID identification, 3D three-dimensional, PanIN pancreatic intraepithelial neoplasia, ISH in situ hybridization.

The team analyzed 38 grossly normal pancreas samples from patients undergoing resections for PDAC or other neoplasms of the pancreas or bile duct. Remarkably, a median of 30% of ductal cells were neoplastic (range 0.2%–75%), and the burden of PanINs per pancreas was ~1000, with each PanIN consisting of ~3515 cells. Genomic analysis of 37 PanINs (from 8 patients) demonstrated that 97% harbored activating KRAS mutations. Individual PanINs were monoclonal, or occasionally, appeared to reflect a merging of separate lesions. Different KRAS variants and passenger gene mutation profiles were not shared between discontinuous PanINs, suggesting that, despite their very large numbers, each PanIN arose independently. The great majority of the PanINs had entirely low-grade histology, with a small proportion containing regions with high-grade dysplasia. Thus, from a field of multiple incidental clones, dominant clones emerge with low frequency.

The authors also found evidence that the processes driving changes in genomic integrity were distinct between PanIN and PDAC. The types of acquired mutations observed in PanIN (e.g., C-T alterations emerging from the process of cytidine deamination) are considered ‘clock-like’ signatures, because they arise predictably with numbers of cell divisions and aging. By contrast, PDAC exhibited mutational signatures and chromosomal copy number changes that were indicative of reactive oxygen species, activity of the APOBEC enzyme, and other types of DNA damage. Finally, the frequency of different KRAS hotspot mutations in PanINs (G12D > G12V > G12R > G12C) matched that observed in PDAC. This suggests that distinct fitness advantages of KRAS alleles5 are selected for early in tumorigenesis and/or do not change during subsequent molecular and histopathological progression.

These findings have multiple implications. First, the large number of lesions across each sample is striking considering the relatively low overall incidence of PDAC (~1/60 lifetime risk),6 indicating that despite the presence of activating KRAS mutations, individual PanINs have limited propensity to progression. This is corroborated by the detection of PanINs in the healthy pancreas from individuals in their twenties.7 The abundance of KRAS mutant cells and spatial heterogeneity of PanINs underscore the difficulties in developing biomarkers for early detection, with monitoring KRAS mutations in liquid biopsies seeming insufficient. The potential use of other indicators of disease progression such as the aberrant induction of transposable elements (TEs) may provide a more robust approach for gauging disease onset.8

The data raise additional questions relating to disease progression. Are there features present in rare subsets of low-grade lesions making them prone advancing to higher grades or are stochastic processes involved? Against the backdrop of abundant low-grade PanINs, the scarcity, but large size, of high-grade PanINs suggests that they undergo rapid progression to PDAC. What causes the switch in mutational mechanism, and could this switch explain the apparently rapid progression of advanced lesions? Candidates for driving genomic instability include telomere shortening9 and activation of TEs due to global hypomethylation, which can result from cumulative cell divisions.10

We should also consider conclusions emerging from genetically engineered mouse models (GEMMs) of PDAC, which typically involve activation of a single hotspot Kras mutation in a large proportion of pancreatic cells, in contrast to the focal and heterogeneous nature of human PanINs. While acinar-to-ductal metaplasia (ADM) is established as the primary origin of PanINs in GEMMs,2 human PanINs appeared to arise with the ductal system. Spatial reconstruction of precursors in GEMMs could reveal the extent to which mouse ADM/PanIN recapitulates human PanIN.

Another notable observation highlighted by the authors was that the extremely high burden of precursors (hundreds of PanINs), contrasted with other well-studied multifocal precancers, such as colonic polyps (~5 adenomas per colon) or melanocytic nevi (10–60 per individual). These differences might suggest organ-specific processes of field cancerization. Considering the interconnected nature of the pancreas through pancreatic ducts, inflammatory signals resulting from ongoing acinar damage could readily propagate throughout the entire organ. Understanding the mechanism of this pancreas-specific precancer burden may provide insights into both tumor prevention and maintenance of a healthy pancreas, potentially safeguarding from conditions such as diabetes or pancreatitis.

One note of caution raised by the authors is that all apparently normal pancreas tissues analyzed were from patients with pancreato-biliary tumors, which could have affected PanIN phenotypes. The application of the approaches developed by the authors to study pancreatic tissue from patients without tumors and of different ages promises to provide further insights into the inception and progression of PDAC. Precise understanding of PanIN biology may lead to groundbreaking advances in combating this dreaded disease.