OpenDVP links tissue images to protein measurements, one cell neighbourhood at a time

A tumour is not a bag of identical cells. Cancer cells, immune cells and supporting tissue form neighbourhoods in which a few micrometres can change what a molecule is doing. Scientists can photograph these neighbourhoods, identify cell types and measure thousands of proteins, but connecting the three activities has required specialised software and delicate sample preparation. A Nature Methods paper published on 22 October introduces openDVP, an open-source route from a tissue image to a spatially targeted protein measurement.[1]

Its authors demonstrate the approach with archival human placenta, early lung cancer and paired primary and relapsed breast-cancer specimens. The system is not a clinical diagnosis machine. It is a research workflow that helps scientists choose which precise patch of tissue to cut out and analyse, and then relate the resulting protein abundance to the cells and tissue architecture that were visible in the image.[1]

Two ways into the same experiment

The simpler workflow, called flashDVP, is designed for researchers who can recognise relevant compartments under a microscope but do not have advanced image-analysis expertise. Images are annotated, regions are transferred to a laser microdissection system and the resulting material goes to mass spectrometry. A more capable pipeline uses AI-supported segmentation, cell classification and spatial neighbourhood analysis to guide that process.[1]

The software connects existing tools rather than replacing them with a closed platform. The authors describe using QuPath, napari, MCMICRO and established data formats including AnnData and SpatialData. That matters because a laboratory may already have expertise in one piece of the analysis, or may need to combine images and omics measurements gathered on different instruments. Standard formats make it easier to revise the computational steps independently of the underlying samples.[1][2]

The group reports that segmentation of 11 whole-slide images containing 6.34 million cells took less than two hours in an example processing run. This number describes an image-analysis stage under the reported computing conditions, not the time needed for collecting tissue, cutting samples, running the mass spectrometer and interpreting results. The latter stages remain physical experiments with their own rate limits and quality checks.[1]

A neighbourhood can be the right sampling unit

Traditional single-cell spatial proteomics aims to isolate an individual cell or a tightly defined cell class. That offers precision, but fragile or connected cells can be difficult to extract, and a single cell may yield too little material for deep protein analysis. The openDVP paper explores cellular-neighbourhood sampling: collecting a meaningful patch made up of several nearby cells that interact, rather than demanding a perfect one-cell measurement every time.[1]

In the breast-cancer comparison, neighbourhood sampling identified an average of 26% more proteins than the conventional DVP comparator in the tested tumour and immune niches. The approaches still detected many of the same proteins and produced correlated tissue signatures. The gain concerns protein coverage under that sampling comparison, rather than diagnostic accuracy.[1]

A further analysis retained 203 high-quality proteomes after filtering. Its displayed heat map contained 4,570 proteins meeting a specified completeness filter; that is not the total number of proteins identified in the experiment. Comparing neighbourhoods in primary and relapsed tumour samples revealed spatial patterns that a bulk average could obscure.[1]

Laser microdissection and mass spectrometry still require equipment and laboratory skill. The useful flexibility is in the sampling unit: sometimes the right biological question concerns a neighbourhood, and extracting every cell separately would add difficulty without improving the answer.[1]

Report an error or suggest a correction