Immune Response in Triple-Negative Breast Cancer: Machine Learning-based Insights from Histology and -Omics

Breast cancer is the most common type of cancer in women worldwide. Triple-negative breast cancer (TNBC) is a subtype of breast cancer that lacks the expression of the oestrogen receptor, progesterone receptor, and amplification of HER2. This subtype is the most aggressive subtype and a heterogeneous subgroup of breast cancer, making up around 10-15% of the cases, often affecting younger patients and presenting with a higher risk of relapse. One factor that seems to positively impact patient outcomes in this heterogeneous subtype is the presence of an active immune response. This thesis focuses on improving the understanding of immune infiltration in TNBC and deriving approaches for better prognostication and treatment prediction. For this, we developed an automated image analysis pipeline and applied it to (among others) six immunohistochemical markers of immune cells. Digital cell counts extracted from five tissue microarray blocks showed how immune cells are often co-expressed in the tumour micro-environment. Furthermore, we demonstrate how a combination of immune status with DNA repair deficiency status can improve prognostication in (chemotherapy-treated) patients. In addition, we created a stand-alone classifier for gene expression data, based on the immunomodulatory subtype, that reflects immune response in TNBC. Our classifier was borderline non-significant in the stratification of neoadjuvant- treated chemotherapy patients and could stratify adjuvant chemotherapy-treated patients into subgroups of better or worse prognosis. Lastly, we studied the spatial heterogeneity of the tumour immune microenvironment in TNBC and its connection to molecular and genomic subtypes. We connected ecosystems in the tumour micro-environment with patterns of immune infiltration and could show TNBC specific differences. In conclusion, this thesis advances the quantitative and integrative study of the tumour (immune) micro-environment in TNBC, offering new approaches and insights to bridge the gap between immune infiltration, molecular heterogeneity, and clinical outcome.

Breast cancer is the most common type of cancer for women all around the world, but not all breast cancers are the same. Triple-negative breast cancer (TNBC) is a less common form of breast cancer, affecting about 10–15% of women with breast cancer. What makes TNBC unique is that it lacks the (overexpression of) certain receptors or proteins that most breast cancers use to grow and spread. Unfortunately, this also means that TNBC is harder to treat because many targeted treatments are ineffective. As a result, chemotherapy is one of the few available options, making it critical to understand other factors that influence TNBC outcomes.
One important factor is the immune response. Recent research shows that the activity of the immune system can give us clues about the chances of a patient surviving TNBC, regardless of whether they receive chemotherapy. The immune system, which fights infections and diseases, has different types of cells that can work together to attack cancer. These cells can enter the tumour and the area around it, known as the ‘tumour micro-environment’. Think of the tumour micro-environment as a city. Some parts of the city have many police officers (immune cells) working to keep criminals (cancer cells) under control. But not all police are equally good at their job, and some may even work against the system! Additionally, some cities (tumours) have very few or no police at all, allowing crime (cancer) to take over. Understanding how these ‘police patrols’ work—and why they fail in some cases—is key to finding better treatments for TNBC.
In this thesis, we focused on how the immune system interacts with TNBC and how this affects patient outcomes. First, we developed a computer tool to analyse images of TNBC tissue automatically. This allowed us to identify patterns, such as when one type of immune cell is present in high numbers, another type is often present in high numbers too. Based on these observations, we divided patients into two groups: those with ‘high immune activity’ and those with ‘low immune activity’. Next, we combined these immune activity groups with a DNA-based classification that examines how well a tumour can repair its DNA. By combining these two pieces of information, we created four patient subgroups. Interestingly, we found that patients with high immune activity and faulty DNA repair tended to have better outcomes.
We also developed a tool that uses gene activity data to measure the immune response in TNBC. While this tool wasn’t statistically perfect in predicting outcomes for patients who had chemotherapy before surgery, it was able to separate patients who had chemotherapy after surgery into better- and worse-performing groups. This shows promise for its use in identifying which patients might respond better to certain treatments.
Finally, we examined how immune cells are distributed within tumours and linked this spatial information to specific TNBC subtypes based on molecular and DNA changes. This helped us better understand how the immune system interacts with the unique biology of different TNBC subtypes.
Taken together, this thesis added some small pieces to our understanding of the immune system’s role in TNBC. By analysing the immune response in a quantitative way and linking it to patient outcomes and tumour characteristics, we have taken important steps toward improving treatments for this challenging cancer.