GMAP Project
Here, we established the GMAP project, a complete solution for constructing high-quality datasets from tissue-based multispectral imaging at single-cell resolution.By combining powerful hardware with scientific strategies, the GMAP project solves many common problems in the generation and utilization of pathomics data, providing an example for tissue-based biomarker discovery and quantitative pathology to guide precision immunotherapy.
1 Gastric cancer cohort
1.1 Patient recruitment
After approval from the Ethics Committee of Peking University Cancer Hospital, this study included 80 GC patients, 60 of whom received anti-PD-1/PD-L1 based therapy, and the other 20 patients received chemotherapy. Patients or their legal guardians agreed to participate in this study by signing informed consent forms. None of the patients in this study had syphilis, autoimmune disease, or HIV infection. Tumor tissues for mIHC analysis were obtained from the Department of Pathology, Peking University Cancer Hospital. Pathologists confirmed adenocarcinoma as the pathological type of all tissues. Mismatch repair (MMR) status was determined by IHC analysis of the expression of four DNA mismatch repair proteins (MLH1, MSH2, MSH6 and PMS2). EBV status was identified by in situ hybridization (ISH) using a probe against Epstein-Barr encoded RNA 1 (EBER1).
1.2 RECIST guidelines
Complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD) were assessed using the Response Evaluation Criteria in Solid Tumors guideline. Patients achieving CR or PR were defined as responders, and patients with SD or PD were defined as non-responders. OS was defined as the interval time between diagnosis and death or the last follow-up; irPFS was defined as the interval time from the start of immunotherapy to PD or the last follow-up; irOS was considered the interval time between immunotherapy and death or the last follow-up.
2 Workflow of m-IHC
The expression in situ and spatial distribution of CD8, PD-1, TIM-3, LAG-3, CD4, FoxP3, CTLA-4, PD-L1, CD68, CD163, STING, CD20, CD66b, and PANCK in tumor tissues were visualized using IHC staining. Within 30 min after being excised, the tumor tissues were fixed in formalin for 24-48 h, dehydrated, and embedded in paraffin. Each paraffin block containing tumor tissues was cut to a thickness of 4 μm. Formalin-fixed, paraffin-embedded tissue slides (4 µm) were melted and dehydrated at 60 °C for 12 h. The samples were then deparaffinized and rehydrated using xylene and alcohol, respectively. Paraffin slides were placed in EDTA buffer (pH 9.0) or citrate buffer (pH 6.0), and the entire system was subjected to heat-induced antigen retrieval. The sections were blocked for 10 min with X0909 (Dako, Santa Clara, CA, USA). The manufacturers and antibodies used to label each molecule are listed in Supplementary Table 1. We pre-optimized antibody concentrations and staining order and serially incubated the slides with the primary antibodies and horseradish peroxidase-conjugated secondary antibodies. Tyramide signal amplification was performed, followed by antibody stripping and antigen retrieval by heating the slides. Cell nuclei were labeled with 4’, 6-diamidino-2-phenylindole (DAPI, Sigma-Aldrich, St. Louis MO, USA; cat. D9542). Staining quality was independently estimated by two pathologists to ensure that the sections met the requirements for further analysis.
3 Tumor regions
3.1 ROI selecting
Bright-field and fluorescence images of the whole slides were obtained using the Mantra quantitative pathology imaging system (PerkinElmer, Waltham, MA, USA) to scan the stained formalin-fixed, paraffin-embedded tissue sections. Two pathologists used the Phenochart (PerkinElmer) software to select the tumor core (TC) from the whole field. Then, the region of interest (ROI) was selected as many as possible using fixed-size stamps (930 × 700 µm; ×20 objective lens) in the TC region. Pathologists examined the quality of each ROI to ensure that all ROIs showed appropriate signal intensity and did not overlap. After ROI segmentation, the inForm image analysis software 2.4 (PerkinElmer) was used to segment and annotate cells in the multispectral images. The autofluorescence spectra of the tissues from unstained sections were recorded and used as controls.
3.2 Tissue segmentation
The acquired ROIs were analyzed with inForm for tissue-component segmentation of tumor cell and stroma regions and cell phenotyping. Each indicator was calculated in the tumor region, which was labeled with PANCK.
3.3 Cell identification
Using single-stained glass slides, the target protein labeled with a specific antibody was shown as a fluorophore, based on which a spectral library was constructed. The spectral library provided references for cell identification, including fluorophore and nucleus features (labeled with DAPI). The inForm software used a supervised method to annotate each DAPI-stained cell in the mixed fluorescence images based on the previously built spectral library.4 Interactive web server
The web server was constructed by “Shiny” packages (version 1.7.1) in R 4.1.2, and deployed on Ubuntu 18.04. The final web pages will be interactively accessed with the support of Shiny server 1.5.17.973. Data transfer between the server and client is encrypted via HTTPS certificates. No registration is required to use any function in GMAP.
4.1 Survival analysis
Survival analysis is conducted by survival package and survminer package.The interactive page is shown as follows:
[1] Tabs of survival analysis.
[2] Tabs of survival analysis for OS, irPFS or irOS. OS
was calculated in 80 patients.
Due to 60 patients in 80 patients receiving immunotherapy, irPFS
and irOS were calculated in 60 patients.
[3] Cell type.
Users can choose the cell type that they are interested in and
analyze its relationship with OS, irPFS or irOS.
[4] Cutoff percent for grouping patients.
Patients were divided into high and low groups according to the
cutoff percentage.
[5-6] Characteristics of Subtypes.
If we choose the subtype based on the clinicopathological features,
survival analysis is conducted in the
corresponding subgroup.
[7] Summary table.
The summary table provides more information on the survival curves,
and the hazard ratio is calculated by
coxph().
[8] Download button.
Click the button and the survival plot will be downloaded and named
“Survival analysis_OS.pdf”,
“Survival analysis_irPFS.pdf”, or “Survival analysis_iroS.pdf”.
4.2 Differential analysis
Differential analysis is conducted by ggpubr package and ggplot2 package.The interactive page is shown as follows:
[1] Tabs of differential analysis.
[2] Cell type.
Users can choose the cell type that they are interested in and
analyze its distribution in different subgroups.
[3] Characteristics of Subtypes.
Choose the clinicopathological characteristics to group the
patients.
[4] Statistic method.
Both parametric and nonparametric tests are conducted by
stat_compare_means(). For parametric tests,
comparisons between two groups use method = t.test,
and comparisons between three groups use
method = anova. For nonparametric tests, comparisons
between two groups use method = wilcox.test, and
comparisons between three groups use
method = kruskal.test.
[5] Density plot.
The dotted lines represent the mean.
[6] Download button.
Click the button and the density plot will be downloaded and named
“Differential analysis_density plot.pdf”.
[7] Box plot.
[8] Download button.
Click the button and the box plot will be downloaded and named
“Differential analysis_box plot.pdf”.
[9] Summary table.
The summary table provided more information on the test statistic
and the distribution of data.
4.3 Correlation analysis
Correlation analysis (Spearman correlation) is conducted by ggpubr package and ggplot2 package.The interactive page is shown as follows:
[1] Tabs of correlation analysis.
[2] Cell type on the x-axis.
[3] Cell type on the y-axis.
[4] Characteristics of Subtypes.
Choose the clinicopathological characteristics and conduct a
correlation analysis in each subgroup.
[5] Calculate 95 Confidence Intervals or
Not.
[6] Download button.
Click the button and the correlation plot will be downloaded and
named “Correlation analysis plot.pdf”.
4.4 ROC analysis
ROC analysis is conducted by pROC package(Robin et al., 2011).The interactive page is shown as follows:
[1] Tabs of ROC analysis.
[2] Cell type.
Choose the cell types which will be used to predict the outcome of
immunotherapy. When the number of cell
types is ≥ 2, comparison will be conducted by bootstrap, delong or
wenkatraman method.
[3] Comparison methods
Choose a comparison method to comparison two or more ROC
curves.
[4] Color palette.
Selected a set of colors to beautify ROC curve.
[5] Summary table.
The table provided the detailed information on the results of
comparison.
[6] Download button.
Click the button and the ROC plot will be downloaded and named “ROC
analysis plot.pdf”.
4.5 Cluster analysis
Cluster analysis is conducted by ComplexHeatmap package(Gu et al., 2016).The interactive page is shown as follows:
[1] Tabs of Cluster analysis.
[2] Cell type.
Choose the cell types (≤35) for cluster analysis.
[3] Phenotype
Select the phenotype for annotating samples in
heatmap.
[4] Number of clusters.
Set the number of row clusters (samples) or column clusters (cell
types).
[5] Order method.
The heatmap can be ordered by a phenotype or a cell type when the
number of row clusters is 0 (No row clustering).
[7] Plot button.
Click the button and a new heatmap will be plotted based on the
current setting.
[8] Download button.
Click the button and the heatmap will be downloaded and named
“Heatmap plot.pdf”.