Immune cellular patterns of distribution affect outcomes of patients with non-small cell lung cancer

Abstract

Studying the cellular geographic distribution in non-small cell lung cancer is essential to understand the roles of cell populations in this type of tumor. In this study, we characterize the spatial cellular distribution of immune cell populations using 23 makers placed in five multiplex immunofluorescence panels and their associations with clinicopathologic variables and outcomes. Our results demonstrate two cellular distribution patterns-an unmixed pattern mostly related to immunoprotective cells and a mixed pattern mostly related to immunosuppressive cells. Distance analysis shows that T-cells expressing immune checkpoints are closer to malignant cells than other cells. Combining the cellular distribution patterns with cellular distances, we can identify four groups related to inflamed and not-inflamed tumors. Cellular distribution patterns and distance are associated with survival in univariate and multivariable analyses. Spatial distribution is a tool to better understand the tumor microenvironment, predict outcomes, and may can help select therapeutic interventions.

Conflict of interest statement

E.R.P. is pathology consultant of the Nucleai LTD. C.H. reports research funding to institution from Sanofi, Dragonfly, BTG, Trisalus, Iovance, and Avenge; scientific advisory board member of Briacell with stock options; personal fees from Nanobiotix and speaker fees/honorarium from SWOG and SITC outside the scope of the submitted work. J.V. H. has received research support from AstraZeneca, Bayer, GlaxoSmithKline, and Spectrum; participated in advisory committees for AstraZeneca, Boehringer Ingelheim, Exelixis, Genentech, GlaxoSmithKline, Guardant Health, Hengrui, Lilly, Novartis, Specrtum, EMD Serono, and Synta; and received royalties and/or licensing fees from Spectrum. D.G. has served on scientific advisory committees for AstraZeneca, GlaxoSmithKline, Sanofi, Eli Lilly and Janssen and has received research support from Janssen, Takeda, Ribon Therapeutics, Astellas and AstraZeneca. I.I.W. has provided consulting or advisory roles for AstraZeneca/MedImmune, Asuragen, Bayer, Bristol Myers Squibb, Genentech/Roche, GlaxoSmithKline, Guardant Health, HTG Molecular Diagnostics, Merck, MSD Oncology, OncoCyte, Novartis, Flame Inc, and Pfizer; has received grants and personal fees from Asuragen, Genentech/Roche, Bristol Myers Squibb, AstraZeneca/MedImmune, HTG Molecular, Merck, and Guardant Health; has received personal fees from GlaxoSmithKline and Oncocyte, Daiichi-Sankyo, Roche, AstraZeneca, Pfizer and Bayer; has received research funding to his institution from 4D Molecular Therapeutics, Adaptimmune, Adaptive Biotechnologies, Akoya Biosciences, Amgen, Bayer, EMD Serono, Genentech, Guardant Health, HTG Molecular Diagnostics, Iovance Biotherapeutics, Johnson & Johnson, Karus Therapeutics, MedImmune, Merck, Novartis, OncoPlex Diagnostics, Pfizer, Silicon Biosystems, Takeda, and Novartis. The other authors declare no competing interests.

© 2023. The Author(s).

Figures

Fig. 1. Representative examples of multispectral images from non-small cell lung cancer tissue microarray specimens, with their chord diagram of the markers.

Composite spectral mixing images from multiplex immunofluorescence (mIF; 20× magnification, scale bars represent 50 µm on each image) is shown for (A) panel 1: cytokeratin (CK), CD3, CD8, PD-1, PD-L1, and CD68; (B) panel 2: CK, CD3, CD8, CD45RO, granzyme B (GZB), and FOXP3; (C) panel 3: CK, CD3, PD-L1, B7-H3, B7-H4, IDO-1; and VISTA; (D) panel 4: CK, CD3, ICOS, LAG3, OX40, TIM3, and CD20; (E) panel 5: CK, Arg-1, CD11b, CD14, CD33, CD66b, and CD68. F Chord diagram visualization showing the diversity of inter-relationships between markers’ co-expression, including all the markers in the five mIF panels. Data from 225 samples was used. Experiments and quantifications related to the presented results were conducted once. The images were generated using Vectra-Polaris 1.0.13 scanner system and InForm 2.4.8 image analysis software (Akoya Biosciences). The chord diagram was generated using all tumor cores from all mIF panels by R studio software version 3.6.1. (Source data is provided as a source data file).

Fig. 2. Representative examples of multispectral images and uniform manifold approximation and projection (UMAP) to identify cell types from non-small cell lung cancer tissue.

Composite spectral mixing images from multiplex immunofluorescence (mIF; 20× magnification, scale bars represent 50 µm on each image) showing colored marker co-expression for (A) panel 1, (C) panel 2, (E) panel 3, (G) panel 4, and (I) panel 5. Color-coded UMAP visualizations show cell types identified by mIF panels: (B) 13 major cell types identified in panel 1, (D) 14 major cell types identified in panel 2, (F) 27 major cell types identified in panel 3, (H) 25 major cell types identified in panel 4, and (J) 12 major cell types identified in panel 5. Data from 225 samples was used. Experiments and quantifications related to the presented results were conducted once. mIF images were generated using Vectra-Polaris 1.0.13 scanner system and InForm 2.4.8 image analysis software (Akoya Biosciences), and UMAP visualizations were generated using the markers from each mIF panel by Python v.3.8.9. (Source data is provided as a source data file).

Fig. 3. Nearest neighbor distance G function and theoretical Poisson curve score graphs showing different cellular patterns of distance from cytokeratin + cells (malignant cells) to CD3 + T-cells, and heat map representing distance patterns by histologic type.

A, B Representative example of the scoring system across tissue specimens and threshold to be considered part of the mixed (A) or unmixed (B) pattern. Graphs represent the scoring system (left), point pattern distributions related to the major T-cell population (middle), and G function and theoretical Poisson curve area (right). Composite spectral mixing images from multiplex immunofluorescence (mIF; 20× magnification, scale bars represent 50 µm on each image) showing a representative image of mixed pattern of macrophages PD-L1 expression (C) and PD-1+ antigen experience T-cells (D), in the bottom inside detail of the pattern (mIF; 40× magnification, scale bars represent 20 µm on each image). Unmixed pattern of CD8 + cytotoxic T-cells (E) and in their bottom inside detail of the pattern (mIF; 40× magnification, scale bars represent 20 µm on each image). F Model interaction based on the G function and theoretical Poisson curve score shows two groups of interaction between the most critical cell phenotypes observed and malignant cells. Cell phenotypes with a score of −10 to 10 were characterized as having a mixed/heterogeneous pattern indicating more interaction with malignant cells, and cell phenotypes with a score of >10 were characterized as having an unmixed/clustering pattern indicating less interaction with malignant cells. G Median distance heat map representing the 27 most common tumor-associated immune cells near malignant cells (CK + ) across the multiplex immunofluorescence panels in adenocarcinomas (ADCs = 142 samples) and squamous cell carcinomas (SCCs = 83), data from 225 samples was used. Experiments and quantifications related to the presented results were conducted once. Graphs and heat map were generated using R studio software version 3.6.1. mIF images were generated using Vectra-Polaris 1.0.13 scanner system and InForm 2.4.8 image analysis software (Akoya Biosciences). (Source data is provided as a source data file).

Fig. 4. Violin plots showing associations between patterns of immune cell distribution and clinicopathologic features.

Significant cellular distribution scoring patterns between immune cells and malignant cells are shown by (A) smoker status, (B) tumor size, and (C) KRAS mutation status for lung adenocarcinoma (n = 142) specimens. D Significant cellular distribution scoring patterns between malignant cells and immune cells are shown by tumor size for lung squamous cell carcinoma (n = 83) specimens. Violin plots showing the median bar value, lower adjacent value and outside points. Kruskal–Wallis test was used in A–D comparisons between groups. Data from 225 samples was used. Graphs were generated using GraphPad Prism v.9.0.0 using the 26 cell phenotypes distribution patterns and the relevant clinicopathologic information using un-adjusted P-values. (Source data is provided as a source data file).

Fig. 5. Violin plots showing the significant associations between distances of immune cell populations from malignant cells and clinicopathologic features.

Significant differences in distance between malignant cells and immune cells are shown by (A) smoker status, (B) tumor size, (C) final stage, and (D) mutation status in lung adenocarcinoma (n = 142) specimens. E Significant differences in distances between malignant cells and immune cells by tumor size are shown for lung squamous cell carcinoma (n = 83) specimens. Violin plots showing the median bar value, lower adjacent value and outside points. Kruskal-Wallis test was used in A to E comparisons between groups. Data from 225 samples was used. Graphs were generated using GraphPad Prism v.9.0.0 using the 128 measurements from malignant cells to different tumor-associated immune cells (TAICs) and between important TAICs and the relevant clinicopathologic information using un-adjusted P-values. (Source data is provided as a source data file).

Fig. 6. Kaplan–Meier analysis of overall survival (OS) by cellular patterns of distribution and distance from malignant cells to various immune cell subpopulations.

A Composite spectral mixing images from a detail of multiplex immunofluorescence (mIF; 40× magnification scale bars represent 50 µm on each image) and illustration of the two different patterns of distribution, mixed and unmixed. B Composite spectral mixing images from a detail of multiplex immunofluorescence (mIF; 40× magnification, scale bars represent 50 µm on each image) and illustration of the distance metrics from malignant cells to different immune cells. (C–G) Kaplan-Meier OS curves. Red lines indicate mixed pattern or close (≤median) distances between malignant cells and various cell phenotypes, and blue lines indicate unmixed pattern or long (>median) distances between malignant cells and various cell phenotypes. C Patients with CD66b + granulocytes (PMNs) with a mixed pattern had better OS than those with an unmixed pattern in lung adenocarcinoma specimens. Close (≤median) distances from malignant cells to (D) CD3 + CD8 + cytotoxic T-cells, (E) CD3 + CD8 + GZB + activated cytotoxic T-cells, and (F) CD68 + macrophages and long (>median) distances from malignant cells to B7-H3 + T-cells were associated with better OS in lung adenocarcinoma specimens using un-adjusted P-values. Data from 225 samples was used. Experiments and quantifications related to the presented results were conducted once. The images were generated using the Vectra-Polaris 1.0.13 scanner system and InForm 2.4.8 image analysis software (Akoya Biosciences), and Kaplan–Meier curves and log-rank test were used and generated by the R studio software version 3.6.0. using the distribution patterns and distances from the 26-cell phenotypes with un-adjusted P-values. (Source data is provided as a source data file).