Spatial proteomics gets an open-source path from whole-slide image to mass spec, with neighborhood sampling about 100 times faster

Spatial proteomics now has an open-source end-to-end route that runs from a whole-slide image to samples measurable by mass spectrometry, without proprietary software. DVP image analysis had depended on commercial software, and no nonproprietary open-source workflow supported scalable image analysis, laser-microdissection experiment design, multimodal integration and interoperability across imaging and omics platforms. openDVP was applied to three archival tissue studies, human placenta, early stage lung adenocarcinoma and primary versus locally relapsed triple-negative breast cancer, covering whole-slide and tissue-microarray images from conventional widefield or confocal microscopy.

The pipeline splits image analysis, cell labeling, cutting-contour export and proteomic data integration into replaceable modules that exchange standard file formats. Image processing builds on MCMICRO and SOPA, visualization and quality control use QuPath and napari, and integration uses the scverse AnnData and SpatialData formats. Segmenting 11 whole-slide images totaling 6.34 million cells took under 2 hours, and contours exported through the QuPath-to-LMD web application were validated for single-cell and region alignment on a Leica LMD7 at 20x and 40x magnification.

Neighborhood-guided sampling adds a third route between single-cell cutting and cell pooling, replacing 100 single-cell cuts with one contour spanning about 100 phenotyped cells. Original DVP pooled phenotypically coherent cell populations for coverage, while single-cell DVP maximized spatial resolution at the expense of depth and ease of use. On directly adjacent cells of identical phenotype in primary and relapsed breast cancer, cnDVP identified an average 26% more proteins than DVP with roughly 100-fold faster tissue collection, whereas only 3 of 50 single immune cells retained more than 400 proteins after filtering.

The demonstrated scope is three archival human tissue studies: first-trimester placenta, early stage lung adenocarcinoma, and primary versus locally relapsed triple-negative breast cancer, with whole-slide and tissue-microarray images spanning H&E, immunohistochemistry, immunofluorescence and multiplex immunofluorescence. The fast-track route targets users without image-analysis expertise who want raw images converted into exploratory spatial proteomics results; the extended route targets users needing single-cell spatial analysis and complex phenotype-to-proteotype associations. In its current implementation, neighborhood-guided sampling is most suitable for homotypic or spatially coherent niches.

Neighborhood-guided sampling generally gives lower biological resolution than DVP or single-cell DVP; DVP and single-cell DVP remain necessary when rare phenotypes, heterogeneity within a niche or direct cell-to-cell variation are the main questions. Performance of all workflows depends on tissue quality, staining fidelity, imaging performance, segmentation or annotation accuracy and laser-microdissection efficiency, each of which can introduce technical variability, and the available biological information is set by the imaging modality and marker panel.