Pancreatic ductal adenocarcinoma (PDAC) is the most common malignancy of the pancreas and has one of the highest mortality rates of any cancer type with a 5‐year survival rate of <5%. Recent studies of PDAC have provided several transcriptomic classifications based on separate analyses of individual patient cohorts. There is a need to provide a unified transcriptomic PDAC classification driven by therapeutically relevant biologic rationale to inform future treatment strategies. Here, we used an integrative meta‐analysis of 353 patients from four different studies to derive a PDAC classification based on immunologic parameters. This consensus clustering approach indicated transcriptomic signatures based on immune infiltrate classified as adaptive, innate and immune‐exclusion subtypes. This reveals the existence of microenvironmental interpatient heterogeneity within PDAC and could serve to drive novel therapeutic strategies in PDAC including immune modulation approaches to treating this disease.

Gene expression states influence the three-dimensional conformation of the genome through poorly understood mechanisms. Here, we investigate the conformation of the murine HoxB locus, a gene-dense genomic region containing closely spaced genes with distinct activation states in mouse embryonic stem (ES) cells. To predict possible folding scenarios, we performed computer simulations of polymer models informed with different chromatin occupancy features, which define promoter activation states or CTCF binding sites. Single cell imaging of the locus folding was performed to test model predictions. While CTCF occupancy alone fails to predict the in vivo folding at genomic length scale of 10 kb, we found that homotypic interactions between active and Polycomb-repressed promoters co-occurring in the same DNA fibre fully explain the HoxB folding patterns imaged in single cells. We identify state-dependent promoter interactions as major drivers of chromatin folding in gene-dense regions.The link between gene expression states and long-range chromatin contacts between different transcription units is a major open question in our understanding of gene regulation (Fig. 1a). Imaging approaches suggest that active transcription units cluster in sites of transcription, called transcription factories 1,2 , whereas Polycomb-repressed genes are found to associate with Polycomb bodies or poised transcription factories [3][4][5][6] . Genes in the same metabolic pathways can co-localise with each other, especially in specialized cell types, although at low frequency across cell populations 7,8 . Although the molecular mechanisms that underlie promoter co-associations and their functional purpose remain unclear, they suggest that gene activation states may help to establish cell-type specific chromatin folding configurations that partition active from Polycomb-repressed chromatin domains. CTCF binding has important roles in the formation of chromatin loops and enhancer-promoter interactions [9][10][11][12][13][14] , and has been proposed to organise chromatin domains through loop extrusion mechanisms in gene-poor areas 15 and to help insulate spreading of active marks into Polycomb repressed domains 16 . However, CTCF . CC-BY-NC-ND 4.0 International license peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not . http://dx.doi.org/10.1101/111773 doi: bioRxiv preprint first posted online Feb. 28, 2017; 3 contribution to the large scale folding of more complex, gene-dense regions remains unknown in particular regions populated with active and Polycomb-repressed loci. residues. Polycomb-repressor complexes can induce chromatin folding independently of catalytic activity [30][31][32] . The co-associations of active genes or Polycomb-repressed genes have so far been studied separately, without assessing whether one type of association might have predominant contributions to chromatin folding at specific loci. Mouse ES cells lack laminTo further understand the und...