Skip to main content

Development of human pancreatic cancer avatars as a model for dynamic immune landscape profiling and personalised therapy

Cite this dataset

Willenbrock, Frances; O'Neill, Eric (2024). Development of human pancreatic cancer avatars as a model for dynamic immune landscape profiling and personalised therapy [Dataset]. Dryad.


Pancreatic ductal adenocarcinoma (PDAC) is the most common form of pancreatic cancer, a disease with dismal overall survival. Advances in treatment are hindered by a lack of preclinical models. Here we show how a personalised organotypic ‘avatar’ created from resected tissue, allows spatial and temporal reporting on a complete in situ tumour microenvironment, and mirrors clinical responses. Our perfusion culture method extends tumour slice viability, maintaining stable tumour content, metabolism, stromal composition, and immune cell populations for 12 days. Using multiplexed immunofluorescence and spatial transcriptomics, we identify immune neighbourhoods and potential for immunotherapy. We employed avatars to assess the impact of a pre-clinically validated metabolic therapy and show recovery of stromal and immune phenotypes and tumour re-differentiation. To determine clinical relevance, we monitored avatar response to gemcitabine treatment and identified a patient avatar-predicable response from clinical follow-up. Thus, avatars provide valuable information for the syngeneic testing of novel therapeutics and a truly personalised therapeutic assessment platform for patients.

README: Development of human pancreatic cancer avatars as a model for dynamic immune landscape profiling and personalised therapy

Description of the data and file structure

Spatial Transcriptomics files

Spatial transcriptomic datasets are Excel spreadsheets of gene counts from RNAseq analysis which have been normalised using Q3 normalisation. Cellular areas are designated tumours, stroma, or immune by staining for pan-cytokeratin, alpha-smooth actin and DC45, as described in the methods. Column names are the sample names and row names are the gene names. 

The file 'control_ascorbicAcid_metformin_sample_ID_timepoint.xlsx' gives a key to interpret the sample names in the file 'control_ascorbicAcid_metformin_GeoMX' where

  • SegmentLabel = segment designation (tumour, stroma, immune)
  • SegmentDisplayName = the sample name given in 'control_ascorbicAcid_metformin_GeoMX'. 'C' or 'T' at the start of the name represents control and treated samples. The name ends with the segment designation. The rest gives no relevant information.
  • Timepoint = length of time for which the samples have been exposed to metformin and ascorbic acid

The file 'SampleStabilityGeomx_Q3_Norm' is performed on samples which have been designated tumour or stroma only and have not been treated. The data gives the normalised gene counts of samples which have been maintained in the perfusion system for different lengths of time.

  • data is normalised gene counts (normalised using Q3 normalisation) with row names giving gene names in column 1
  • row 1 gives the sample name consisting of: segment designation = tumour or stroma; culture condition = baseline (day 0), perf (perfused) or static (not perfused); DX = number of days for which the section has been maintained (5, 7, 9 or 12); 3-digit identifier
  • row 2 gives the length of time for which the avatar has been maintained in culture in days.

Immune Cell Proximity files

Immune cell proximity study data files are in 9 zipped folders. Each folder corresponds to one of three samples (PDCA 10, 12, 13).

PDCA 10 consists of 2 folders:

  1. PDCA10baseline_proximity-immuneCells_ImmuneCells
  2. PDCA10proximityToTumour_ObjectData

PDCA 12 consists of 5 folders:

  1. PDCA12baseline_proximity-immuneCells_ImmuneCells
  2. PDCA12perfusedDay12_proximity-immuneCells_ImmuneCells
  3. PDCA12perfusedDay12b_proximity-immuneCells_ImmuneCells
  4. PDCA12proximityToTumour_ObjectData
  5. PDCA12staticDay12_proximity-immuneCells_ImmuneCells

PDCA 13 consists of 2 folders:

  1. PDCA13baseline_proximity-immuneCells_ImmuneCells
  2. PDCA13proximityToTumour_ObjectData

Within each folder are 20 CSV files giving distances between two types of cell, given in the name for column H: eg CD4 within 100μm of Macrophage means that in this file the distance of each CD4 cell from all neighbouring macrophages is listed.            

Folder naming convention considerations:

  • The 'proximity_immuneCells_immuneCells' files give data on the distances between each type of immune cell analysed and other proximal immune cells or tumour cells.
  • The 'ObjectData' files give the number of tumour cells within areas of tissue that are predominantly tumour cells ie tumour infiltration.

Within each zipped file is a series of data files for each of the 3 samples at timepoints Day0 (baseline) or Day12. The cell type and proximal cell of interest are given in the column headers.

File naming convention considerations:

  • PTxx = sample PDCAxx
  • BASE = baseline (i.e., day 0)
  • STAT = static culture
  • PERF = perfused culture
  • the rest of the filename gives no important information


  • Image location = lab drive identifier. Ignore this
  • Analysis region = lab identifier for the location of tissue slice on the slide
  • Algorithm name = algorithm as defined by HALO software
  • Object id = ID given to the cell pair
  • Cell ID (column E) = ID given to immune cell of interest (macrophage/CD4+ Tcell/CD8+ Tcell/ Treg/B cell)
  • Cell ID (column F)  = ID given to the target cell (macrophage/CD4+ Tcell/CD8+ Tcell/ Treg/B cell/tumour cell)
  • Distance (μm) = distance between the two identified cells in μm
  • macrophage/CD4+ Tcell/CD8+ Tcell/ Treg/B cell within 100um of macrophage/CD4+ Tcell/CD8+ Tcell/ Treg/B cell/tumour cell = '0' if distance  > 100 μm, '1' if distance < 100 μm
  • XMin, XMax, YMin, YMax = coordinates of cell of interest. not used in this analysis

Sharing/Access information





Acquisition of tissue and blood samples

The study (REC number 19/A056) was approved for the collection of tumour and healthy pancreatic tissue by the Oxford Radcliffe BioBank. The collection of the specimens was supported by the Oxford Centre for Histopathology Research. All patients recruited to the study provided written consent confirming voluntary participation and permission for tissue donation for research. Biopsy punch samples (5mm diameter) were obtained by the pathologist at the John Radcliffe Hospital following surgery provided that clear surgical margins could be determined.

Sectioning and culture of live tumour slices

Samples were transported on ice in unsupplemented low-glucose DMEM media before suspension in agarose scaffolds. Following a manual wash in LG DMEM media, biopsy punches were suspended in 4% low-gelling-temperature agarose and cooled. The agarose scaffold structure was generated by melting the solution and suspending the slice in a small embedding mould. 250 µm sections were generated using the Leica VT1200 vibratome (blade angle +21°C, speed 1.5 mms-1, amplitude 2 mm) in a bath of ice-cold PBS and transported in LG DMEM media on ice.

Alvetex perfusion plates (REPROCELL) were used to conduct dynamic perfusion experiments, with syringe pumps maintaining a constant flow of 10 µLmin-1. The apparatus was assembled within a sterile tissue culture hood. To construct the circuit, 60 mL syringes were connected to silicone tubing and flushed with 70% ethanol before washing in unsupplemented LG DMEM media. Alvetex 12-well tissue culture inserts were activated in 70% ethanol for 2 minutes before washing in LG DMEM media. Long-term culture of tissue slices was conducted using LG DMEM complete pancreatic medium (see reagents table) at +37°C and 5% CO2.

Treatment of slices in perfusion culture

All ex vivo treatment of avatars was commenced on day 0 (day of tumour acquisition). Systemic chemotherapy was prepared in DMSO. Drug delivery commenced after the avatars had been created and subsequently placed in the perfusion plate, via a inflow circuit comprising a syringe and tubing, attached to the perfusion plate and connected to the perfusion pump. After the intended time period for drug delivery, the infusion was stopped, and the perfusion plate removed from the incubator and placed within a tissue culture hood. The inflow circuit was removed and the inflow channel on the perfusion plate temporarily occluded using a 2ml syringe. An inflow circuit was then created, and the syringe filled with culturing media. The inflow circuit was primed with media (from the attached syringe) to remove any air bubbles present in the tubing prior to attachment to the perfusion plate to ensure that there was a constant column of media from the syringe and the tubing to the perfusion plate. For treatment with metformin and ascorbic acid, 20 mM metformin and 100 mM ascorbic acid was perfused for 5 days, with freshly made-up solutions applied each day. Gemcitabine concentration was perfused for one hour at a concentration of 250 mM.

GeoMX spatial transcriptomic analysis

Spatial transcriptomic analysis was provided by NanoString Technologies, Inc. through their GeoMX® DSP Technology Access Program Grant. Formalin-fixed paraffin-embedded (FFPE) samples were provided cut at 5 µm thickness and mounted on negatively-charged slides. Samples were sent to the NanoString Technology Access Program labs for analysis using the GeoMX Digital Spatial Profiler. Slides were incubated with oligonucleotide-antibody conjugates with photocleavable linkers. UV light was then used to selectively release oligonucleotide barcodes which were read and quantified through sequencing. Four immunofluorescent markers were used for morphology definition, facilitating region of interest (ROI) selection. A total of 68 ROIs were selected across treated and control samples. The Whole Transcriptome Atlas was used for sample profiling, with genes mapped to barcodes using an in-house algorithm produced by NanoString Technologies, Inc., generating spatially resolved transcriptional data. Quality control involved assessing raw read threshold, percent-aligned reads, and sequencing saturation. A quantification limit was set based on a negative probe signal (mean + two standard deviations). Reads were filtered based on their expression in >5% AOIs and counts were normalised to account for differences in AOI size and cellularity.

Multiplexed immunofluorescence and HALO analysis

The multiplex IF staining of avatars was performed in collaboration with the Oxford Translational Histopathology Lab. Slides were stained using the Leica BOND RXm autostainer machine (Leica, Microsystems) whilst following the OPAL™ protocol 28 (AKOYA Biosciences). A total of 6 staining cycles were subsequently performed.  The following primary antibody-opal fluorophore pairing was used: CD4 – Opal 520, CD8 – Opal 570, CD20 – Opal 480, Foxp3 – Opal 620, CD68 – Opal 690, pan Cytokerratin – Opal 780. In adherence to the manufacturer's instructions, the primary antibodies were incubated for one hour and subsequently detected with the BOND™ 4 Polymer Refine Detection System (DS9800, Leica Biosystems). The DAB step was replaced by the opal fluorophores which consisted of a 10 minute incubation with no haematoxylin step. The antigen retrieval step was performed with Epitope Retrieval (ER) Solution 2 (AR9640, Leica 8 Biosystems) for 20 minutes at 100 °C. This was performed prior to the application of each primary antibody. VECTASHIELD® Vibrance™ Antifade Mounting Medium with DAPI 10 (H-1800-10, Vector Laboratories.) was used to mount each slide. The AKOYA Biosciences Vectra® Polaris™ was used in order to obtain whole slide scans and multispectral images (MSI). A batch analysis of every MSIs was performed using the inForm 2.4.8 software. The final step consisted of fusing the multiple batched analysed MSIs on the HALO (Indica Labs) software. This resulted in the creation of a spectrally unmixed reconstructed whole image of the avatar. Analysis of the multiplex IF was performed on the Indica Labs HALO® (version 3.0.311.407). This software allows for deconvolution of the image by selecting individual fluorophores for analysis. The initial step was teaching the software a Random Forest Classifier module in which the images were segmented into tumour and stromal regions. Manual annotation of the slides was performed to exclude areas with staining artefacts. Cell detection and subsequent phenotyping was performed using the Indica Labs - HighPlex FL v3.1.0 (fluorescent images). Individual cells were defined by their expression of the specific markers: Tumour (DAPI+ panCytokeratin+), CD4 helper (DAPI+CD4+), CD8 cytotoxic (DAPI+CD8+), regulatory T-cell (DAPI+CD4+Foxp3+), B cells (DAPI+CD20+) and Macrophages (DAPI+CD68+). 


Cancer Research UK

Medical Research Council