The murine Lewis lung carcinoma (LLC) cells and B16 melanoma cells were purchased from the National Collection of Authenticated Cell Cultures (Shanghai, China). Cells were cultured in RPMI Medium 1640 (Gibco), which contained 10% fetal bovine serum (Gibco) and 1% penicillin/streptomycin (Hyclone). Cells were incubated at 37℃ in 5% CO2.
Tumor models and treatments
Female C57BL/6 N mice between six to eight weeks old were purchased from Zhejiang Vital River Laboratory Animal Technology Co. Ltd (Jiaxing, Zhejiang, China) and were maintained under specific pathogen-free (SPF) conditions, with the temperature of 20 ~ 25℃, humidity of 40% ~ 70%, and the light time set from 8:00 to 20:00.
To establish tumor models, 1 M LLC or 0.5 M B16 cells were inoculated subcutaneously into the right flank of mice. Treatment was started when average tumor volume reached 200mm3 (10 days after inoculation, defined as D0). Mice were randomized to receive irradiation at D0 (RT group) or 200 μg anti-PD-L1(10F.9G2, BioXcell) antibody i.p. since D7, every 3 days for 5 times (ICI group). Three mice per group per interested time point were used in each of the three independent experiments. All mice experiments were ethically approved by the Institutional Animal Care and Use Committee of Shanghai Chest Hospital.
Mice were anesthetized and positioned in the prone orientation. A CT scan was acquired and treatment planning was performed using Monte Carlo algorithm. Dose-volume histogram evaluation showed the target volume receiving 95% ~ 105% of the dose. Radiation was delivered using a 6MV X-ray linear accelerator at a dose rate of 600 cGy/min.
For RT group, micro-PET/CT was performed on D0, D3, D7, D14 and D21. While for ICI group, micro-PET/CT was performed on D0, D7 (before anti-PD-L1 treatment) and D22 (after anti-PD-L1 treatment). Mice were injected intraperitoneally with 5.55 MBq 18F-FDG in 0.1 mL saline. After an hour, micro-PET/CT images were acquired using an Inveon micro-PET/CT scanner (Siemens Preclinical Solution). During scanning period, mice were maintained under 2% isoflurane anesthesia and were placed in the prone position on the bed of the scanner. CT scans were performed first (Voltage: 80 kV, Current: 500μA, Exposure 300 ms, Projections: 120, effective pixel size: 112.85 μm). Then, PET scans were performed 10 min later, OSEM3D/SP-MAP algorithm (2 OSEM iterations, 18 MAP iterations, target resolution 1.5 mm and matrix size: 128 × 128) without filtering and smoothing was used to reconstruct PET images. The PET images were converted into SUV units by normalizing the activity concentration to the dosage of injected 18F-FDG and mouse body weight. The micro-PET and CT images were generated separately and then fused using Inveon Research Workplace (Siemens Preclinical Solution).
RNA extraction and microarray analysis
RNA was collected from LLC tumors either before irradiation (D0) or at day 3, 7, 14, 21 post-irradiation (D3, D7, D14 and D21). Total RNA was quantified by the NanoDrop ND-2000 (Thermo Scientific) and the RNA integrity was assessed using Bioanalyzer 2100 (Agilent). The sample labeling, microarray hybridization and washing were performed based on the manufacturer's standard protocols. In brief, total RNA was transcribed to double strand cDNA, then synthesized into cRNA and labeled with Cyanine-3-CTP. The labeled cRNAs were hybridized onto the microarray. After washing, the arrays were scanned by the Agilent Scanner G2505C (Agilent). The Agilent Mouse Gene Expression 4*44 K v2 Microarray (DesignID: 026,655) was used in this experiment and data analysis was conducted by OE Biotechnology Co. Ltd (Shanghai, China).
The abundance of tumor-infiltrating immune cell subsets was estimated by a deconvolution approach with online analytical tool CIBERSORTx (Cell type identification by estimating relative subsets of known RNA transcripts; https://cibersortx.stanford.edu) [26, 27]. The CIBERSORT LM22 matrix, consisting of 547 genes, transforms gene-expression data of microarray analysis into relative fractions of immune cells phenotypes. CIBERSORT implements Monte Carlo sampling to generate an empirical P-value for the deconvolution. Only cases with a P-value < 0.05, which indicated a reliable estimation of immune cell infiltration, were used for further analysis.
Tumor samples were disposed with Tumor Dissociation Kit (Miltenyi Biotec). Then single-cell suspensions from tumors were lysed to remove red blood cells using RBC Lysis/Fixtion Solution (BioLegend) and filtered through a 70 μm MACS SmartStrainer (Miltenyi Biotec). After that, the single-cell suspensions were incubated with fluorescently labeled antibodies (BioLegend) against CD45, CD3, CD4, CD8 and PD-1 at 4℃ for 45 min. For intracellular staining, cells were fixed and permeabilized, and then incubated with TIM-3 (BioLegend), TOX (eBioscience), TCF-1 (BD) at 4℃ for 6 h. Multicolor flow cytometry analysis was conducted with BD Fortessa flow cytometer. Expression level of PD-1 on CD8+ T cells was gated by its low controls of CD8+ T cells from spleen of untreated mice (PD-1low), high controls of cells stained by both anti-TOX and anti-PD-1(PD-1hi) and between them (PD-1int)  (Details are provided in the Supplementary Methods section). The proportion of early Tex (CD8+PD-1int) and terminal Tex (CD8+PD-1hi) in CD8+ T cells, the ratio of early to terminal Tex (E/T), and mean fluorescence intensity (MFI) of PD-1 expression on CD8+ T cells were further analyzed using FlowJo software (Treestar).
Tumors were manually delineated on PET/CT images using MIM Maestro Version 7.1.4 (MIM Software), excluding adjacent organs, bone structure and large vessels. Radiomics features were extracted with Imaging biomarker explorer (IBEX) , 57 from CT images and 61 from PET images, including first-order features, shape features, and textural features derived from Gray Level Cooccurence Matrix (GLCM) and Gray Level Run Length Matrix (GLRLM) (Table S1). Radiomics features would undergo a two-step selection. First, the feature reproducibility under varying scanning conditions was assessed by test–retest analysis [29, 30], using RIDER Lung CT and RIDER Phantom PET-CT datasets from The Cancer Imaging Archive (TCIA). Concordance correlation coefficient (CCC) was calculated and features with good robustness would be selected for further analysis. Second, with the ICI group as the training set, least absolute shrinkage and selection operator (LASSO) logistic regression was used to select radiomic features significantly correlated with characteristics of T cell exhaustion, such as the infiltrating proportion of early Tex and terminal Tex among CD8+ T cells, E/T, or MFI of PD-1 on CD8+ T cells. Then, significant features were further collected to establish a binary logistic regression model for identifying the level of T cell exhaustion (median proportion of terminal Tex as cutoff). Finally, validation of the model was performed with RT group as external validation set. Concordance index (C-index) and calibration plot were used to evaluate the discriminability and calibration of this model. The goodness-of-fit of model was assessed by Brier score.
Data are presented as the mean \(\pm\) SEM. For comparing different time groups, one-way analysis of variance with Bonferroni corrections was applied with two-sided P-values (#P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001). Statistical analyses were performed using R software 4.1.2 and GraphPad Prism 8.4.3.
The data generated in this study are available within the article and its supplementary data files.