Study On The Homogeneity And Heterogeneity Of PD-L1 (Ventana SP142) Expression In Breast Cancer

Breast cancer is one of the most common malignant tumors in women,and its incidence is increasing year by year worldwide,accounting for 7% to10% of all malignant tumors.As immunotherapy has achieved good results in malignant tumors,immunotherapy for breast cancer has gradually emerged and has become a hot research topic.This research is divided into two parts.The first part analyzes the consistency of the pathologist’s interpretation of PD-L1(Ventana SP142)IC,and proposes an artificial intelligence-assisted(AI-assisted)model based on deep learning for the PD-L1 IC score.The second part discusses the heterogeneous expression of breast cancer PD-L1(Ventana SP142)from multiple perspectives of space and time.Analyze the relationship between PD-L1 expression in needle biopsy specimens and surgical specimens.Find the difference in PD-L1 expression between metastatic breast cancer and primary tumor.The degree of PD-L1 expression loss of paraffin sections at different storage temperatures at different time nodes has also been studied and explored.Solve the problem of actual clinical testing,so as to better screen patients who may benefit from immunotherapy.Part one AI Assisted PD-L1(Ventana SP142)Expression Scoring in Breast CancerObjective: PD-1/PD-L1 is an important biomarker,which can identify patients suitable for immunotherapy.However,the consistency of tumor infiltrating immune cell(IC)score of PD-L1 expression is low in clinical practice.Therefore,we need a repeatable and quantifiable PD-L1 IC scoring method for breast cancer.Method: In this study,we propose a deep learning-based artificial intelligence-assisted(AI-assisted)model for PD-L1 IC scoring.From 100 cases of invasive breast cancer PD-L1(Ventana sp142)immunohistochemical(IHC)Whole slide images(WSI)sliding windows with no overlap were scanned through the tumor regions and generated 4,246 image patches.After that,images with manually identified necrosis area,cancer in situ,and normal areas were excluded,and 2,395 image patches remained.109 image patches were randomly proposed from the 2,395 image patches.Three rounds of ring studies(RSs)involving 31 pathologists from 10 hospitals were carried out.The purpose of ring study 1(RS1)was to evaluate the scoring concordance across pathologists in the current clinical practice.After a 2-week washout period,ring study 2(RS2)was carried out,with the purpose of evaluating intra-and inter-observer concordance.After another 2 weeks of washout period,ring study 3(RS3)was conducted.Here,the emphasis was on evaluating the role of AI assistance in PD-L1 scoring.The gold standard for the categorical PD-L1(SP142)IC scores of the test images were provided through consensus reading from three experienced pathologists who received formal training from Roche Diagnostics and practiced PD-L1(SP142)expression scoring in their routine clinical work.Assess PD-L1 expression with continuity score.In the consistency analysis,Kappa test was used for categorical variables,and ICC was used for continuous variables.The accuracy evaluation was represented by several metrics,including accuracy,area under the curve(AUC),and weighted F1 score.Results:1.The AI 2-category score accuracy was 0.963(105/109 images),with AUC of 0.888 and 2-class weighted F1 score of 0.962.For the 4-category score,accuracy was 0.752(82/109 images),with AUC of 0.797,and 4-class weighted F1 score of 0.764.2.With the assistance of AI,the scoring concordance across pathologists was boosted to excellent in RS3(0.950,95% confidence interval(CI):0.936-0.962)from moderate in RS1(0.674,95% CI: 0.614-0.735)and RS2(0.736,95% CI: 0.683-0.789).3.For the 2-category scoring,AI assistance gave a significant accuracy boost(p<0.001)in RS3(0.959,95% CI: 0.953-0.964),which improved 4.2%from RS2.For 4-category scoring,the average scoring accuracy also had a significant improvement of 14.7% from RS2 to RS3 through AI-assistance.4.The AI results were generally accepted by pathologists with 61% “fully accepted” and 91% “almost accepted”.Part two Heterogeneity of PD-L1(Ventana SP142)expression in breast cancerObjective: Due to gene variation,disease progression and other reasons,tumor has a certain degree of heterogeneity,which brings a challenge for clinicians to choose PD-L1 detection time for breast cancer patients.Therefore,we analyzed the relationship between PD-L1 expression in biopsy specimens and surgical specimens.Objective to explore the difference of PD-L1 expression between metastatic breast cancer and primary tumor.Explore the loss of PD-L1 expression in paraffin sections at different storage temperatures at different time nodes.Method:Simulated puncture from 124 surgical specimens,5 tissues were penetrated in each case,and the relationship between PD-L1 expression and overall slices was analyzed.298 cases of primary breast cancer and its matched metastases were stained with PD-L1 to explore the differences in expression.71 cases of PD-L1 positive invasive breast cancer specimens were included,all of which were sliced serially at different temperatures(room temperature conditions(20-25℃),4℃,-20℃ and-80℃)and different time nodes(1 week,2 Weeks,3 weeks,4 weeks,2 months,3 months,and 6months)PD-L1 staining was performed,and the expression of PD-L1 was compared with the corresponding fresh slices.Assess PD-L1 expression with continuity score.Tumor interstitial infiltrating immune cells positive≥1% are considered positive.In the consistency analysis,Kappa test was used for categorical variables,and ICC was used for continuous variables.The area under the ROC curve(AUC)was used to analyze the number of punctured tissues that were more consistent with the expression of PD-L1 in surgical specimens.Mc Nemar test was used to analyze the difference of PD-L1 between primary and metastatic lesions.Results:1.The expression of PD-L1 was heterogeneous in 48.4% of patients.The agreement between the maximum expression of PD-L1 in the 5 punctured tissues and the expression of surgical specimens was 80.6%.When the number of punctured tissues increased from 1 to 5,the κ value of PD-L1 expression and PD-L1 expression in surgical specimens gradually increased from 0.165 to 0.615,and κ was 0.499 for 3,reaching a moderate consistency;AUC gradually increased from 0.610 to 0.819,the AUC of 3 punctured tissues reached 0.8.2.In the study of the heterogeneity of PD-L1 expression between primary breast cancer and matched different metastatic sites,94 cases of lymph node metastasis and 204 cases of distant metastasis were included,including 75 cases of liver,47 cases of lung,30 cases of bone,and 29 cases of chest wall,10 cases of brain,5 cases of ovary,3 cases of stomach,2 cases of muscle tissue,1 case of adrenal gland,1 case of small intestine,1 case of eyeball.The overall conversion rate of PD-L1 expression in primary breast cancer and paired metastatic breast cancer was 36.6%,and the difference was statistically significant(P=0.045).In each metastasis site,the conversion rate of the expression of PD-L1 in lung metastasis was 46.8%,and the difference was statistically significant(P=0.001).There was no statistically significant difference between the other metastatic sites and the changes of the primary tumor(P>0.05).3.With the extension of the storage time of paraffin sections,the PD-L1 antigen is gradually lost,and the positive rate gradually decreases.The expression of PD-L1 in slices stored at room temperature for 1 and 2 weeks is more consistent than that of fresh slices(ICC≥0.9).When stored under refrigeration at 4℃ or-20℃,the expression of PD-L1 in slices stored for 1week,2 weeks,3 weeks,and 4 weeks is more consistent with the expression of PD-L1 in fresh slices(ICC≥0.9).The expression of PD-L1 in slices stored at-80℃ for 1 week,2 weeks,and 3 weeks was more consistent than that of fresh slices(ICC≥0.9).Conclusions:1.The proposed AI assisted method can help pathologists at all levels to improve the PD-L1 assay(SP142)IC assessment in breast cancer in terms of both accuracy and concordance.2.The AI tool provides a scheme to standardize the PD-L1 IC scoring in clinical practice.3.The expression of PD-L1(SP142)in the three breast cancer punctured tissues has a moderate correlation with the PD-L1 expression of the corresponding surgical specimens.When using punctured tissues for PD-L1 detection,it is recommended to detect at least three punctured tissues.4.There is a difference in the expression of PD-L1 between primary breast cancer and paired metastatic breast cancer.When new metastases appear in patients,the expression of PD-L1 in metastases needs to be reassessed.5.Long-term storage of paraffin sections will lead to a decrease in the expression level of PD-L1.It is recommended to store them under refrigeration at 4℃ or-20℃,and the storage time within one month.