Invasion status stratifies the composition of the human pancreatic cancer perineural niche

Problem Statement
Pancreatic ductal adenocarcinoma (PDAC) is characterized by near-universal innervation and a high prevalence of perineural invasion (PNI), where cancer cells invade the nerve sheath. PNI is a strong correlate of metastatic burden, recurrence, and poor patient outcomes. Despite this clinical significance, the cellular and molecular composition of the heterocellular perineural niche remains largely undefined. Existing studies have established that nerves interact with tumor cells and stromal components, but the specific cell states and signaling mechanisms distinguishing invaded nerve neighborhoods from non-invaded ones are unclear. Furthermore, standard computational workflows for imaging-based spatial transcriptomics (imST) often rely on cell segmentation, which can lead to misassigned transcripts and incomplete cellular representation, particularly in complex, dense tissues like PDAC.

Methodology
The authors performed a multi-omics characterization of resected PDAC patient tissues stratified by PNI status (high vs. low). The study integrated two primary data modalities:

1. Single-Nucleus RNA Sequencing (snRNA-seq): Performed on 11 fresh-frozen surgical samples (8 PNI-high, 3 PNI-low) to obtain transcriptome-wide gene expression profiles while minimizing dissociation-induced stress responses.
2. Imaging-Based Spatial Transcriptomics (imST): Performed on 11 matched samples (8 PNI-high, 3 PNI-low) using the 10x Genomics Xenium platform with a custom 480-gene panel targeting tumor, tumor microenvironment (TME), and nerve-associated cells.

To overcome computational bottlenecks in imST analysis, the authors employed Nicheverse, a novel vision-centric artificial intelligence model based on discrete representation learning. Unlike standard workflows, Nicheverse bypasses cell segmentation entirely. It utilizes a hierarchical vector quantized autoencoder to learn two discrete codebooks: a cell codebook for annotating cell states based on transcriptional programs and a neighborhood codebook incorporating compositional information from the surrounding tissue context. This allows for the assignment of every nucleus to a cell-type code regardless of transcript count.

Nerve neighborhoods (NNs) were defined as the nerve structure (identified via Schwann cell clusters) plus a 250µm radial expansion. These neighborhoods were manually annotated by pathologists as either "invaded" (malignant cells encircling or infiltrating the sheath) or "non-invaded." The study utilized rigorous statistical frameworks, including linear mixed models (LMM) to account for patient-level heterogeneity and differential abundance testing (Milo) for snRNA-seq data.

Key Contributions and Results

• High-Resolution Atlas of the Perineural Niche: The study generated a high-fidelity spatial and single-cell atlas defining 44 distinct cell types and 16 spatial niches within the PDAC microenvironment. This included the identification of rare cell populations such as specific Schwann cell subsets (myelinating, non-myelinating, fibroblast-like) and distinct Cancer-Associated Fibroblast (CAF) states.
• Stratification by Invasion Status: The analysis revealed a stark dichotomy between invaded and non-invaded nerve neighborhoods:
  • Invaded Nerves: Enriched for classical malignant cells, myofibroblastic CAFs (myCAFs), and lipid-associated tumor-associated macrophages (TAMs) characterized by ALOX5AP, OLR1, and MARCO expression. These areas showed depletion of T and B lymphocytes.
  • Non-Invaded Nerves: Characterized by inflammatory CAFs (iCAFs), CD163-high TAMs, and significant infiltration of B and T lymphocytes (including mature B cells, CD4+ naïve/effector T cells, and CD8+ memory T cells).
• Transcriptional and Spatial Remodeling:
  • Gene Expression: Invaded nerves upregulated genes associated with epithelial identity (KRT16, EPCAM), mucin production, and invasion (FAM83A, PADI1). They also showed upregulation of neuroactive factors Neuromedin-U (NMU) and Brain-Derived Neurotrophic Factor (BDNF), alongside extracellular matrix (ECM) remodeling genes (MMP1, MMP11, COL11A1). Conversely, non-invaded nerves expressed quiescent stromal programs and markers of nerve support (IGF1).
  • Lipid Metabolism: Invaded nerves exhibited a shift toward lipid scavenging and synthesis (OLR1, DHCR24), while non-invaded nerves favored lipid efflux (ABCA6, ABCA8).
  • AP-1 Activity: Both malignant cells and CAFs in PNI-high tissues showed elevated AP-1 transcription factor activity, linking neurotrophic receptor expression (e.g., NTRK2, NGFR) and invasion programs.
• Cell-Cell Communication: Ligand-receptor analysis (LIANA) identified a PNI-high signaling network dominated by VEGF signaling, cell-matrix adhesion (laminin-integrin interactions), and lipid transfer between TAMs, CAFs, and malignant cells. Specifically, a BDNF-NTRK2 axis between classical tumor cells and CAFs was implicated.
• Niche Heterogeneity: The study identified distinct spatial niches, such as a "classical PDAC PNI with myCAF rim" and a "classical progenitor PNI front," suggesting that the severity and nature of PNI vary even within PNI-high samples, with some nerves showing circumferential tumor involvement that excludes immune cells.

Significance
The paper claims that this work provides the first deeply phenotyped spatial and single-cell atlas of the human PDAC perineural niche, enabling a high-resolution comparison of invaded versus non-invaded nerve neighborhoods. The primary significance lies in the finding that these data raise the possibility that PNI, rather than innervation itself, is immune-suppressive in this setting, a hypothesis that motivates further functional studies. The data suggests that the presence of cancer cells invading the nerve sheath actively excludes lymphocytes and recruits immunosuppressive, lipid-metabolizing macrophages and myCAFs.

By defining the specific cellular states (e.g., lipid-associated TAMs, neurotropic CAFs) and molecular programs (e.g., AP-1 regulons, lipid processing, neurotrophic signaling) associated with invasion, the study nominates these processes as potential therapeutic targets. The authors posit that understanding the mechanisms by which PNI remodels the perineural niche to exclude immune cells could inform strategies to improve therapeutic responses in pancreatic cancer. The study also highlights the utility of segmentation-free AI models like Nicheverse for resolving complex tissue architectures in spatial transcriptomics.