Spatial Phenotyping

"Spatial phenotyping" is a general term from spatial biology, not a StrataQuest engine; the Engines Reference has no chapter by that name. The idea has two steps: give each cell measurements of its surroundings, then classify cells on those measurements the way you would on marker intensity. StrataQuest has documented tools for both.

1. Define the cell types. Gates and cutoffs in diagrams define the phenotypesLoading.... Every spatial measurement is taken between these populations, so their accuracy limits everything after.
2. Count neighbors. Phenotype InteractionsLoading..., used in a layer's Measurements, gives each primary-phenotype cell Count in Range, Mean Range, Minimum Range, Maximum Range and Closest ID in Range for one secondary phenotype.
3. Place cells relative to structures. Proximity AreasLoading... divides the tissue into bands between consecutive distances from areas of interest, and Assign classes to objectsLoading... gives each cell the index of its band. The quick action "Add Proximity areas quantification layers" builds the histogram, one gate per band and one layer per band for you. For a continuous value, a distance mapLoading... can be the Shades image of Standard MeasurementsLoading....
4. Map crowding. Density of EventsLoading... turns a set of events into a grayscale image whose high values mark many events nearby, weighting closer pixels more through a Gaussian model.
5. Gate. Plot the spatial measurements on scattergrams or histograms and gate them into classes, as the User Manual does with interaction counts.

Two limits are worth knowing. The neighbor measurements are counts for one pair of phenotypes, not shares of all neighbors; a share needs one Phenotype Interactions per pair and a ratio built in Derived MeasurementsLoading.... And none of these tools tests whether an arrangement differs from chance.

Neighborhoods in spatial biology. In multiplex imaging research generally, a common approach describes each cell by the cell-type composition of its neighbors, then clusters those compositions into recurring "cellular neighborhoods". Schürch and colleagues (2020) used it to map the invasive front of colorectal cancer. StrataQuest's measurements are counts per phenotype pair rather than composition vectors, so that kind of clustering would run on data exported from Raw DataLoading... (as .xlsx, .csv or GeoJSON).

Attraction or crowding. Cell positions form what statisticians call a spatial point process. A count of eight T cells near a tumor cell can mean the two types attract each other, or only that both are packed into the same dense region. Spatial statistics separates the two by comparing observed counts with counts from random arrangements, for example by shuffling cell-type labels or with Ripley's cross-K function. The StrataQuest tools give the observed counts that such a test starts from.

Scale. Direct contact happens within about one cell diameter, roughly 10 to 20 µm for many cells in tissue. Secreted signals reach farther, over distances that vary with the molecule. Regions such as an invasive margin are often defined as bands hundreds of micrometers wide, so one distance rarely answers every question.

Tumor immune patterns. Chen and Mellman (2017) grouped tumors by where their immune cells sit: inflamed, with T cells among the tumor cells; excluded, with T cells held in the surrounding stroma; and desert, with few T cells anywhere. These are spatial definitions, and in some studies they have been associated with response to checkpoint inhibitors.

There is no spatial phenotyping engine, so the relevant settings are those of the tools above.

None of these is a single "neighborhood radius". Phenotype Interactions takes an interval, which can be a disk (minimum 0) or a ring, and Proximity Areas takes a list of band edges. Record the distances you chose next to every spatial class you report.

Source: StrataQuest 8.1 Engines Reference, ch. 4 (pp. 12–13), ch. 12 (pp. 400–402), ch. 13 (p. 403), ch. 34 (pp. 483–494), ch. 36 (pp. 500–503), ch. 41 (p. 538); User Manual, Quick Actions (p. 154), §12.1 (p. 166), §12.6.5 (p. 199).

Suppose you are characterizing the tumor-immune interface in a PD-L1 study.

1. Gate the phenotypes: CK+ tumor cells and CD3+CD8+ cytotoxic T cells.
2. Add Phenotype Interactions in Measurements with CK+ tumor cells as the Primary Phenotype and CD8+ T cells as the Secondary Phenotype, Minimum Distance 0 and Maximum Distance 30 µm. Select Count in Range and Minimum Range.
3. Plot the tumor cells' Count in Range on a histogramLoading.... Gate tumor cells with no CD8+ T cell within 30 µm, and those with one or more.
4. Add a second instance with the phenotypes swapped, so each T cell gets its count of nearby tumor cells. Gate T cells with many tumor neighbors and T cells with none. The names you give these gatesLoading..., such as "engaged" and "bystander", record your rule.
5. For position, run Proximity AreasLoading... from the tumor region with Distance Values of 50, 100 and 200 µm, then use "Add Proximity areas quantification layers" and count CD8+ T cells in each band.

The result shows where T cells meet tumor cells and where they stop short. A rule such as "more than 30% of neighbors are CD8+" needs a count for every neighbor type: one Phenotype Interactions per pair, then a ratio in Derived Measurements.