Background: Racial disparities in breast cancer incidence and clinical outcome have become increasingly recognized. In particular, African American (AA) patients have higher mortality rates than other racial groups. Research into these racial disparities has primarily focused on socioeconomic determinants, with some studies exploring potential molecular determinants. Biologically, few differences in genetics or transcriptomics have been found to be due to race. Understanding racial differences in breast cancer development as they relate to patient outcomes requires a comprehensive approach, integrating tumor intrinsic and extrinsic factors. We have developed a platform that combines patient-specific imaging and -omics data to elucidate how biological differences are integrated to cause racial differences in tumor behavior and disparities in patient outcome.
Methods: Imaging and transcriptomic data from over 500 patients were integrated within the SimBioSys TumorScope biophysical modeling software to understand how racial differences in tumor biology are coordinated to determine tumor behavior. Specifically, we used artificial intelligence to automatically segment tumors from patient MRIs, and construct 3D models of patient tumors and the surrounding microenvironment. We then used these segmentations to identify radiological differences between AA and Caucasian patients. Simultaneously, we used patient-specific mathematical models of tumor metabolism constructed from RNA-seq data to determine metabolic differences between tumors in AA and Caucasian patients.
Results: We profiled structural differences in 266 tumors (181 Caucasian and 85 AA) and metabolic differences in 552 tumors (422 Caucasian and 130 AA). In our patient cohort, AA tumors were smaller (AA/Caucasian volume = 0.573, p = 0.004), more spherical (AA/Caucasian sphericity = 1.12, p = 0.0046), had lower spatial variation in tumor stiffness (AA/Caucasian Contrast GLRLM Gray Level Non-uniformity = 0.63, p = 0.0004), and had higher vascular perfusion than Caucasian tumors (AA/Caucasian vascular shell density = 1.43, p = 0.0015). In addition, AA tumors demonstrated increased levels in several features which may contribute to a decreased response to therapy, including a higher spatial heterogeneity in drug delivery and secretion (AA/Caucasian WIS std = 2.2, p = 5.9×10 -9, WOS std = 11.5, p < 10 -10) and higher levels of adipose surrounding the tumor (AA/Caucasian adipose shell density = 1.13, p = 0.0024). Metabolically, we found that tumors from AA patients have increased sphingolipid metabolism (AA/Caucasian fold change (FC) = 1.4, p = 0.04) and pyruvate metabolism (FC = 1.03, p = 0.03); and decreased starch and sucrose metabolism (FC = 0.89, p = 0.03), fatty acid metabolism (average FC = 0.83, p = 0.04), glycerolipid metabolism (FC = 0.70, p = 0.025), and acylglyceride metabolism (FC = 0.69, p = 0.025). Furthermore, we found that these metabolic features integrate in such a way that tumors from AA patients had an increased specific growth rate (FC = 1.52, p = 0.014), compared to Caucasian patients (which remained when correcting for PAM50 subtype).
Conclusion: Taken together, these results indicate that tumors of AA patients have a more aggressive phenotype (high growth rate) and potentially a lower response to chemotherapeutic drugs (high spatial heterogeneity of drug delivery and secretion and high surrounding adipose). These features likely work together to contribute to the disproportionately poor survival rates repeatedly observed in AA breast cancer patients. By characterizing racial-specific tumor intrinsic and extrinsic biological factor and their potential interactions, TumorScope simulation platform can assist in finding strategies that may improve clinical management, life-expectancy, and overall outcomes of racial minorities with breast cancer.
