The spread of interferon-γ in melanomas is highly spatially confined, driving nongenetic variability in tumor cells

Abstract

Interferon-γ (IFNγ) is a critical antitumor cytokine that has varied effects on different cell types. The global effect of IFNγ in the tumor depends on which cells it acts upon and the spatial extent of its spread. Reported measurements of IFNγ spread vary dramatically in different contexts, ranging from nearest-neighbor signaling to perfusion throughout the entire tumor. Here, we apply theoretical considerations to experiments both in vitro and in vivo to study the spread of IFNγ in melanomas. We observe spatially confined niches of IFNγ signaling in 3-D mouse melanoma cultures and human tumors that generate cellular heterogeneity in gene expression and alter the susceptibility of affected cells to T cell killing. Widespread IFNγ signaling only occurs when niches overlap due to high local densities of IFNγ-producing T cells. We measured length scales of ~30 to 40 μm for IFNγ spread in B16 mouse melanoma cultures and human primary cutaneous melanoma. Our results are consistent with IFNγ spread being governed by a simple diffusion-consumption model and offer insight into how the spatial organization of T cells contributes to intratumor heterogeneity in inflammatory signaling, gene expression, and immune-mediated clearance. Solid tumors are often viewed as collections of diverse cellular "neighborhoods": Our work provides a general explanation for such nongenetic cellular variability due to confinement in the spread of immune mediators.

Keywords: cytokine signaling; interferon-γ; melanoma; quantitative biology; tumor-infiltrating lymphocytes.

Fig. 1.

TIL density varies dramatically within individual tumors and across different tumors. (A) Image reconstructions from melanoma lesions stained and imaged using CyCIF and obtained from three different sites of one patient. Single cells were first segmented using Hoechst (nuclear) staining, and tumor cells were assigned by thresholding on the level of S100A staining. T cells were identified by thresholding on levels of CD3, CD4, or CD8 and absence of FoxP3. Infiltrated T cells (TILs) were further selected by removing all T cells not within at least 25 μm from at least 10 S100A+ tumor cells. Images were then converted to point clouds where tumor cells are represented by gray points and TILs are red. (B) Local relative TIL density in a melanoma lesion from an additional tumor site. Local TIL density was computed by scanning over the image with a 100-μm bandwidth Gaussian kernel density estimator. (C) Images demonstrating variable TIL density from each of the red highlighted regions in (B). (D) Probability distribution function of local TIL density depicted in (B). (E) Cumulative distribution function of local TIL density depicted in (B). (F) Theoretical interparticle distance between TILs in a dense-packed 3-D environment as a function of TIL density. We assume randomly scattered TILs and that all cells are 10 μm in diameter.

Fig. 2.

Dense, 3-D culture conditions generate cell-to-cell heterogeneity in the response to IFNγ. (A) Diagram of the basic geometry of conventional well plates and our clusterwell plates. Clusterwells culture cells in 3-D and at high density compared to well plates. (B) Experiment diagram for results shown in (C and D). B16 cells were cultured in either conventional 96-well plates or in clusters and stimulated with either 0.1, 1, or 10 nM of IFNγ for 60 m, before staining for pSTAT1 and performing flow cytometry. (C) Histograms of pSTAT1 levels in IFNγ-stimulated B16 cells. (D) Percentages of pSTAT1+ cells from (C). (E) Experimental diagram for results shown in (F and G). T cells were stimulated with PMA and Ionomycin for 4 h before being cocultured at varying densities with B16 cells in clusterwells for 1 h. Cells were then stained for pSTAT1 and measured by flow cytometry. (F) Histograms of pSTAT1 levels in B16 cells cocultured with activated T cells. (G) Percentages of pSTAT1+ cells from (F).

Fig. 3.

Imaging reveals niches of IFNγ signaling surrounding activated T cells and enables direct measurement of the signaling length scale. (A) Schematic demonstrating the principle of the PlaneView technique that enables dense, 3-D cell culture and convenient 2-D imaging. (B) Experimental diagram of B16-OT-1 T cell PlaneView coculture experiment. (C) Zoomed-out view of clusters of pSTAT1 staining in B16 cells after coculture with OT-1 T cells. (D) Close-up image of a cluster of pSTAT1 staining in B16 cells surrounding one OT-1 T cell. (E) Example quantifications in the decay in pSTAT1 staining surrounding OT-1 T cells. Data are shown as points with error bars, and solid lines depict fits from our theoretical diffusion–consumption equation. (F) Theory depicting expected pSTAT1 decay profiles given a defined length scale of 3 cell diameters and varying the IFNγ flux (a.k.a. IFNγ production rate by T cells).

Fig. 4.

Gradients of IFNγ translate into variability in antigen presentation and susceptibility to T cell–mediated killing. (A) Representations of 1-D slices of 3-D simulations. For simulations, cells were embedded in a 3-D cubic lattice, and a varying fraction of randomly-selected cells were assigned as cytokine producers. Individual producers generated a spherically symmetric cytokine niche around themselves, corresponding to a constant IFNγ flux of 50 mol/s per T cell, and ${\lambda }_{niche}$ = 30 μm, as measured in our experiments. Red squares represent the centroid of a cytokine producer. Squares are not always present in the center of a cytokine response niche because, in many cases, the producer centroid lies outside of the 1-D slice. (B) Histograms of simulated IFNγ response of cells interspersed randomly with varied densities of IFNγ-producer cells. (C) Simulation results for the fraction of cells responding to IFNγ as a function of varied IFNγ producer density. (D) Simulation results for the right-tailness (>2 SDs above the population mean) of cellular response to IFNγ as a function of varied IFNγ producer density. (E) Histograms of MHC-I levels on B16 cells cocultured with activated T cells. T cells were stimulated with PMA and Ionomycin for 4 h before being cocultured at varying densities with B16 cells overnight in clusterwells. Cells were then stained for MHC-I (H2-K^b) and measured by flow cytometry. (F) Quantification of the fraction of MHC-I+ cells from (E). (G) Quantification of the right-tailness (>2 SDs above the population mean) of MHC-I levels from (E). (H) Experiment diagram for results shown in (I). T cells were stimulated with PMA and Ionomycin for 4 h before being cocultured at varying densities with B16 cells in clusterwells overnight. Cells were then collected from clusterwells and cocultured with Pmel-1 CD8+ T cells and IFNγ neutralizing antibodies. Pmel-1–mediated killing was quantified by flow cytometry analysis of DAPI incorporation by B16s. (I) Quantification of the fraction of surviving B16s as a function of the density of cocultured IFNγ-producing T cells.

Fig. 5.

Human melanomas exhibit spatially confined niches of IFNγ signaling that surround TILs. (A) Clustergram of Moran’s I values computed from human primary cutaneous melanoma. Moran’s I is a global measure of spatial correlation between pairs of stained markers in adjacent cells. Computed values were then clustered hierarchically, and groups of correlated markers were annotated manually. (B) Images of regions of human tumor depicting the spatial pattern of IFNγ signaling through pSTAT1 and IRF1 staining, surrounding CD8+ T cells. (C–E) Probability distribution functions for pSTAT1 (C), IRF1 (D), and PD-L1 (E) in TILs and Sox10+ melanocytes. (F–H) Decay in the mean normalized intensities of pSTAT1 (F), IRF1 (G), and PD-L1 (H) as a function of distance from TILs. Data are shown as colored and gray points with error bars, and the solid colored lines depict the fits from our theoretical diffusion-consumption equation.