Thread by @jonathan_nowak

Latest Wolpin-Nowak lab paper on single cell, in situ protein-based characterization of transcriptional subtypes in pancreatic cancer now out in @CR_AACR aacrjournals.org/cancerres/arti…

Led by outstanding postdoc Hannah Williams with assistance from Andressa Dias Costa (@Andress84886647), Jinming Zhang, Sri Raghavan, and Peter Winter (@peterswinter) at @DanaFarber_Hale.

We designed an 8-plex immunofluoresence assay to infer tumor transcriptional subtype using three basal and three classical markers in standard FFPE tissue sections of pancreatic cancer.

We profiled 399K cells across nearly 300 primary resected tumor and found that almost all tumors harbor a mix of basal and classical cells. Most tumors also harbor cells that co-express both basal and classical protein markers.

The location of any single tumor along a classical-basal axis is predictive of survival in a continuous fashion. Here's the MV-adjusted cubic spline curve for overall survival. Higher basal content➡️worse OS. Basal or classical designation should not be considered binary.

Co-expressor cells account for 14% of all tumor cells, exist on a basal-classical spectrum themselves, and are enriched in tumors with a mixed classical-basal phenotype.

Tumor-level subtype classifications are highly concordant with both bulk RNA sequencing and single cell RNA sequencing and can be reliably measured using the amount of tissue commonly found in biopsy specimens.

Metastatic tumor specimens are enriched in basal and co-expressor cells as compared to primary tumors.

To characterize microenvironment influence on subtype, we built a TMA of nearly 80 organoid models developed by the Aguirre lab (@andrewaguirremd) at @DanaFarber_Hale with support from the @lustgartenfdn and also analyzed using our mIF panel.

We found that organoid models derived from primary and metastatic sites show nearly equal cell subtype distributions, underscoring the importance of the anatomic niche and microenvironment in regulating subtype.

Finally, we used machine learning and graph theory to quantify the spatial arrangement of classical, basal and co-expressor cells across 17K individual tumor glands.

These analyses support a model where tumor cells can quickly switch subtype as they divide within individual glands and implicate co-expressor cells as an intermediate state that classical and cells pass through as they undergo subtype switching.

As always, special thanks to colleagues at URMC (Aram Hezel, David Linehan, Richard Dunne), Stanford (Daniel Chang, Margaret Kozak), and MD Anderson (
@ACKoongMDPhD) for contributions to the multi-institutional pancreas cancer resection cohort. This is our 7th paper together!

Also thanks to @AkoyaBio for building a flexible and powerful multispectral imaging platform that enables #spatialbiology at scale for translational cancer studies. Please see our full paper for many more beautiful images!

In terms of translational relevance, the mIF assay that be built for our paper could be run today using equipment in the @BWHPath CLIA lab space and can provide a subtype result using only a single standard FFPE slide and even with a tumor content that would preclude RNA assays.