How can high-throughput spatial proteomics be achieved from cellular to subcellular scales?

Why micro-FFPE samples lose protein before LC-MS

Formalin-fixed paraffin-embedded (FFPE) tissue is the paper's starting material because it preserves morphology and is paired with clinical annotation, but the formaldehyde crosslinks that hold the tissue together must be reversed at temperatures typically above 90 °C before enzymes can digest the proteins. In a microscale region, the reaction volume is tiny, so the droplet can evaporate before decrosslinking finishes. The paper names three consequences: tissue can be physically lost, digestion fails and peptide recovery drops, and evaporation varies between samples so reproducibility suffers.

The authors also frame the existing alternatives as a trade-off rather than a solved problem. Continuous buffer replacement and multi-step protocols achieve coverage but depend on specialized instrumentation or add SpeedVac drying, repeated decrosslinking cycles, and long digestion, all of which cut throughput. Earlier hanging-droplet designs reduced evaporation but processed samples individually. The gap the paper targets is therefore not sensitivity alone but sensitivity combined with parallel handling of many laser-capture microdissection (LMD) regions.

How the hanging droplet holds reaction volume

The workflow uses the EVO96 proteoCHIP, a 96-well device whose Teflon-based capillaries form pillars. When the chip is placed face-down over a standard 96-well plate and heated, droplets remain suspended from those pillars. Droplet stability depends on surface tension balancing gravity, and evaporation is suppressed because solvent vapor saturates the confined headspace, so net evaporation is near zero. The paper reports minimal volume loss of about 1 µL during decrosslinking and digestion.

This design is what connects the chemistry to the throughput claim. Because each pillar is a separate reaction chamber in a 96-well format, lysis, decrosslinking, and digestion happen in the same droplet without transferring the sample between vessels, which is where LMD workflows typically lose material. The authors report that the full path from tissue lysis to LC-MS-ready tryptic peptides on Evotips takes about two hours, roughly 30% less preparation time than reference methods, and that up to 192 samples can be handled in parallel with standard equipment.

Buffer and temperature choices that set proteome depth

The paper optimized the droplet chemistry rather than accepting default buffers. Comparing ammonium bicarbonate (ABC) with triethylammonium bicarbonate (TEAB) on 20,000 µm² breast cancer sections at 5 µm thickness, TEAB at the optimal 25 mM produced about 12.5% more protein groups and 25% more peptide identifications. The authors suggest TEAB's lower volatility may keep local ionic conditions more stable during long incubation in a small droplet, an interpretation rather than a measured mechanism.

Adding the detergent DDM did not help overall. Although 0.02% DDM slightly increased the share of membrane and extracellular proteins, total identifications were consistently higher with TEAB alone, and increasing DDM concentration steadily reduced protein groups and peptides. The authors attribute this to competing effects: DDM can limit surface adsorption and stabilize membrane proteins, but it also changes droplet surface tension and may destabilize the droplet during heating.

Enzyme and heating conditions were tuned on the same 20,000 µm², 10 µm-thick sections. A Trypsin/Lys-C concentration of 5 ng/µL maximized identifications, and raising it further changed the missed-cleavage rate by only about 1–3%. Digestion for 60 minutes gave the highest mean protein groups and modified peptides, 40 minutes was best for decrosslinking, and 90 °C improved recovery and replicate reproducibility over 60 °C. These are optimization results on defined section sizes, not guarantees for other inputs.

Coverage scales with dissected area and cell number

The central performance result is a titration across excised areas from 1,000 µm² to 50,000 µm² at 10 µm thickness, corresponding to roughly 5 to 250 cells, analyzed with the 80 samples-per-day (SPD) gradient. At the highest input the workflow identified about 5,500 protein groups and about 50,000 modified peptides; at the lowest input it identified about 2,500 protein groups and about 13,000 modified peptides, which is 42% of the proteins seen at the highest load.

Reproducibility needs to be read alongside that drop. Smaller regions showed greater biological heterogeneity and lower reproducibility, yet low-input samples still shared about 85% of identified protein groups and had Pearson correlations of about 0.9 in quantitative profiles. Signal intensity scaled with input and saturated above roughly 50 cells. Gene Ontology enrichment showed core functional processes remained consistent even where fewer proteins were identified, so a smaller region costs depth more than it costs the broad biological signature.

Trading gradient speed against proteome depth

The paper separates two acquisition speeds. The 80 SPD method identified about 15% more protein groups and about 25% more peptides than the 120 SPD method, and the two shared about 80% of identified protein groups while covering similar top Gene Ontology biological process terms. The 80 SPD runs also detected lower-intensity protein groups and had more data points per peak, 6 versus 4. Even so, the 120 SPD method retained key breast cancer proteins including PARP1, TOP2A, ERBB2, CDH1, and COL1A1 in the three highest intensity quantiles.

Benchmarks against published workflows come with an explicit caveat: the sampled tissues differed, so the comparison tests performance at comparable input and gradient length rather than identical samples. With the 80 SPD gradient, the method identified 27% more protein groups and 38% more peptides than openDVP for 8,000 µm², 5 µm samples. Against scDVP at a similar single-cell-equivalent volume of 2,000 µm³, it identified about 46% more peptides with a comparable number of protein groups; the authors note scDVP found slightly more low-intensity protein groups, which they attribute to its longer 40 SPD gradient and single compensation-voltage FAIMS acquisition.

What breast cancer tissue reveals at region scale

The biological test used 20,000 µm² cancer and stroma regions from 10 µm-thick triple-negative breast cancer FFPE sections. The method identified about 5,500 protein groups and 50,000 modified sequences from cancerous regions and about 4,500 protein groups and 33,000 modified sequences from stromal regions, with good overlap between samples. Across subtypes, ERBB2 was more abundant in cancer cells than stroma in HER2-positive and luminal-B samples, and PR was more abundant in cancer cells in luminal-A and luminal-B samples, while no differentially regulated markers appeared in the triple-negative samples.

The authors also describe compartment-specific programs: tumor regions enriched for RNA processing, proliferative signaling, and canonical cancer markers, stroma enriched for extracellular matrix remodeling, immune activation, and fibrotic repair, and triple-negative stroma showing MYC pathway activation. These are the paper's interpretations of its own measurements. The important scope limit sits here: each subtype was represented by samples from a single patient, so the authors state the observations cannot yet be generalized and may partly reflect inter-patient variability rather than subtype-specific biology.

Subcellular reach and where it stops

The smallest regions tested were nuclear-enriched areas of about 60 µm², from which the workflow identified about 1,800 protein groups. The authors describe this as approaching near-organelle resolution and as bridging the sensitivity gap between single-cell proteomics and tissue-level spatial resolution without highly specialized instrumentation. That claim rests on nuclear-enriched dissections, so it should be read as subcellular-scale sampling of enriched structures rather than routine profiling of arbitrary organelles.

The practical priority follows from the titration rather than from the headline number. If the question needs deep coverage of a tissue region, the 50,000 µm², ~250-cell end of the range delivers more than 5,500 protein groups; if the question needs many small regions processed in parallel, the workflow still returns about 2,500 protein groups at 1,000 µm² and about 1,800 at 60 µm² nuclear regions, at the cost of lower reproducibility in the smallest samples. The 80 versus 120 SPD choice is the second lever: 120 SPD keeps roughly 80% of protein groups and the key breast cancer markers in half the analysis time, while 80 SPD buys additional depth and lower-intensity proteins.

Sources used in this answer