Final Report Summary - BIO-IRT (Biologically individualized, model-based radiotherapy on the basis of multi-parametric molecular tumour profiling)
Cancer is one of the major societal and health challenges today. Radiotherapy (RT) is one of the three main current treatment options for cancer patients. In the last years, RT has experienced a number of technological innovations and is thus today extremely flexible, fast and focused to the local tumor region only. Nevertheless, a significant number of patients still experiences treatment failures even after treatment with modern combination therapies, such as RT plus chemotherapy. Our hypothesis is that the major cause of resistance against RT lies in the biological nature of the tumor itself. Tumor tissues are extremely heterogeneous, present with chaotic vascular structures and high levels of tumor hypoxia.
The major aim of this project was to investigate key factors which drive resistance to RT on a multi-dimensional level, involving biological factors, genetic characteristics and also multi-parametric functional imaging.
Therefore, a number of different head-and-neck cancer (HNC) cell lines presenting with different radiation sensitivities was implanted into immuno-deficient mice. After a growth period of 6-8 weeks, tumors were examined using multi-parametric (mp) PET/MR imaging including dynamic FMISO PET, anatomical T2-weighted MRI, diffusion weigthed imaging (DWI) and also dynamic contrast enhanced (DCE) MRI. Following fractionated RT for two weeks, a second mp PET/MRI examination was scheduled. Then tumors were excised and conserved for further immunohistochemical and gene expression analysis.
A total of n=74 xenograft tumors were examined with sequential mp imaging in addition to gene expression analysis. For data analysis, new strategies involving machine learning methods for classification of n-dimensional data were implemented and further developed. Also for the analysis of dynamic data, new approaches from the field of data science were used, such as principle component analysis (PCA).
The results of this project have shown that mp imaging features present clusters in d-dimensional space which can predict radiation sensitivity more accurate and robust compared to single imaging features. We would show that a combination of apparent diffusion coefficients (ADC) derived from DWI combined with hypoxia information derived from dynamic FMISO PET via PCA define a metric which predicts radiation sensitivity of the different xenograft tumorus. Furthermore, hypoxia information measured with FMISO PET was shown to be replaced by a machine learning model involving different components from DWI and DCE MRI. This finding is very promising for later clinical use.
Work to combine parameters from gene expression related to a number of dedicated oncogenes relevant to HNC with mp imaging is still ongoing.
Finally, results derived from preclinical experiments are very valuable to understand fundamental processes leading to radiation resistance in HNC patients. Therefore, we aim at translating results from the pre-clinical study to a clinical scale. In a last part of the project a clinical patient study on HNC patients treated with primary chemo-RT was carried out. In the context of this study, gene expression analysis was performed on biomaterial from the tumor. Additionally patients were scanned with mp functional FMISP PET/CT or PET/MRI before and during RT. Final results of this trial are still pending as data analysis is still ongoing.
In conclusion, multi-scale biological data involving parameters from gene expression analysis and mp functional imaging seems to be very promising together with approaches from machine learning to predict individual radiation resistance of tumors. Thus, key players that drive resistance patterns can be identified as a basis for personalized RT treatment strategies in the future.
The major aim of this project was to investigate key factors which drive resistance to RT on a multi-dimensional level, involving biological factors, genetic characteristics and also multi-parametric functional imaging.
Therefore, a number of different head-and-neck cancer (HNC) cell lines presenting with different radiation sensitivities was implanted into immuno-deficient mice. After a growth period of 6-8 weeks, tumors were examined using multi-parametric (mp) PET/MR imaging including dynamic FMISO PET, anatomical T2-weighted MRI, diffusion weigthed imaging (DWI) and also dynamic contrast enhanced (DCE) MRI. Following fractionated RT for two weeks, a second mp PET/MRI examination was scheduled. Then tumors were excised and conserved for further immunohistochemical and gene expression analysis.
A total of n=74 xenograft tumors were examined with sequential mp imaging in addition to gene expression analysis. For data analysis, new strategies involving machine learning methods for classification of n-dimensional data were implemented and further developed. Also for the analysis of dynamic data, new approaches from the field of data science were used, such as principle component analysis (PCA).
The results of this project have shown that mp imaging features present clusters in d-dimensional space which can predict radiation sensitivity more accurate and robust compared to single imaging features. We would show that a combination of apparent diffusion coefficients (ADC) derived from DWI combined with hypoxia information derived from dynamic FMISO PET via PCA define a metric which predicts radiation sensitivity of the different xenograft tumorus. Furthermore, hypoxia information measured with FMISO PET was shown to be replaced by a machine learning model involving different components from DWI and DCE MRI. This finding is very promising for later clinical use.
Work to combine parameters from gene expression related to a number of dedicated oncogenes relevant to HNC with mp imaging is still ongoing.
Finally, results derived from preclinical experiments are very valuable to understand fundamental processes leading to radiation resistance in HNC patients. Therefore, we aim at translating results from the pre-clinical study to a clinical scale. In a last part of the project a clinical patient study on HNC patients treated with primary chemo-RT was carried out. In the context of this study, gene expression analysis was performed on biomaterial from the tumor. Additionally patients were scanned with mp functional FMISP PET/CT or PET/MRI before and during RT. Final results of this trial are still pending as data analysis is still ongoing.
In conclusion, multi-scale biological data involving parameters from gene expression analysis and mp functional imaging seems to be very promising together with approaches from machine learning to predict individual radiation resistance of tumors. Thus, key players that drive resistance patterns can be identified as a basis for personalized RT treatment strategies in the future.