| Breast cancer is the malignant tumor with the highest incidence worldwide,posing a serious threat to women’s health and lives.As a heterogeneous mixture of cells,the Tumor microenvironment（TME）of breast cancer exhibits inter-individual differences in the gene expression patterns of cell subpopulations,making patients behave differently in terms of drug resistance and treatment response,which present significant challenges for the determination of personalized treatment plans.Therefore,an in-depth understanding of the TME cellular compositions and inter-individual heterogeneity is of great significance for achieving personalized treatment and improving clinical outcomes.Related studies have estimated the cellular components of TME through bulk gene expression data.However,existing unsupervised deconvolution-based methods can only estimate the averaged gene expression profiles of cellular components that are shared in a population,while the marker gene-based methods can only estimate the absolute abundance of immune and stromal cells in the TME.Neither of them can reflect the specific gene expression patterns of TME cellular components in individual samples,leading to difficult to evaluate the heterogeneous state of TME in individuals.In addition,some studies have shown that radiomics has certain advantages in evaluating tumor heterogeneity and can provide reference information for tumor prognosis prediction,but radiomics is essentially data-driven and cannot explain the underlying biological mechanisms.In order to solve the above problems,this study proposes a radiogenomic analysis based on sample-wise Convex Analysis of individualized mixture（sw CAM）,which can estimate the cellular components of TME and the sample-wise variational expression profiles by decomposing the bulk gene expression data,to explore the association between individual heterogeneity of TME and prognosis.This study,meanwhile,mines imaging biomarkers with prognostic significance by establishing the relationship between cellular compositions and tumor imaging to promote the clinical precision and non-invasive diagnosis and treatment of breast cancer.The specific research content includes the following three parts:（1）Analysis of TME cell components and individual heterogeneity based on decomposition of bulk gene expression data.Bulk gene expression data were decomposed by Convex Analysis of Mixtures（CAM）,and by selecting model parameters,the averaged gene expression profile and ratio matrices of cell subpopulations in a population were obtained.Combined with survival data,meanwhile,the key cell subpopulations affecting the development and growth of breast tumors were identified.Then,based on the averaged subtype expression profiles and ratio matrices of cell subpopulations,sw CAM was used to obtain the sample-wise variational gene expression profiles of cell subpopulations.In this study,seven cell subpopulations were obtained in the optimal decomposition model,from which two key cell subpopulations affecting the development and growth of breast tumors were identified(P values were 1.64×10-3 and 1.44×10-3,respectively),and sw CAM reveal the individual heterogeneity of TME by decomposing bulk gene data.（2）Analysis of the prognostic biomarkers based on individual heterogeneity of TME.Characteristic functional modules that differentially expressed in different survival outcomes were detected based on the sample-wise variational gene expression profiles of key cell subpopulations,and the biological functions of key cell subpopulations were inferred through enrichment pathway analysis of co-expressed gene clusters in the characteristic modules.Then,according to the module eigengene in the characteristic modules,breast cancer patients were divided into subtypes based on the individual heterogeneity of TME.And by comparing the survival differences between different subtypes,the specific mechanisms of key cell subsets affecting tumor development and evolution were analyzed.In this study,two key cell subpopulations were found to be enriched in cancer-related pathways,and the differences in key cell subpopulations among individuals could significantly divide patients into two types with different survival outcomes(p=4.59×10-6),the independent validation set also confirmed the prognostic value of subtypes(p=3.37×10-2).The results show that individual heterogeneity of TME is closely related to the survival rate of breast cancer.（3）Analysis of radiogenomic method and prognosis based on sample-wise cell subpopulation differences.In experimental datasets that simultaneously possess imaging data and genetic information,the bulk gene expression data was used to estimate the sample-wise variational gene expression profiles of key cell subpopulations,and the co-expression model and clustering model established in the previous experiments were used to obtain patient subtype labels.By combining radiomic features,the prognostic prediction model was established to obtain the correlation between genomic features and tumor radiomic features.In this study,the area under the receiver operating characteristic curve（AUC）was used to evaluate the classification performance of the prognostic prediction model,and the AUC of the model based on the optimal feature subset was 0.840.The prognostic value of the radiogenomic label was also confirmed by two independent imaging datasets(P values were 3.66×10-2 and 3.96×10-3,respectively).This paper focused on the individualized analysis of the breast TME,and explored the relationship between individual heterogeneity of TME and prognosis based on the sample-wise variational gene expression profiles of cell subpopulations.The sample-wise cell subpopulation differences were found to be biomarkers for the development and evolution of breast cancer.Two of the TME cell subpopulations that have the potential to become prognostic biomarkers could provide effective guidance for the determination of personalized treatment plans.Additionally,the relationship between cell subpopulation features and imaging features was explored based on the radiogenomic method,which provided a non-invasive,repeatable prognosis prediction of breast cancer. |
PDF Full Text Request