Limited changes in the CNS immune microenvironment in patients with breast cancer metastasis and capturing these changes using machine learning.

Abstract

1092 Background: Metastasis of breast cancer to the central nervous system (CNS) is common, especially in triple negative and HER2-positive tumors. The CNS is considered immune specialized and likely the brain and brain-border immune microenvironment creates a sanctuary site for breast cancer CNS metastasis. A better understanding of the immune microenvironment may allow for better utilization of immunotherapy to treat CNS metastasis. Toward this goal, we evaluated the cellular transcriptomic profile of cerebrospinal fluid (CSF) cells and compared between patients with documented metastatic tumor by cell-free DNA (cfDNA) testing of the CSF fluid (cfCSF-Pos) and patients without evidence of cfDNA metastasis (cfCSF-Neg) (Charifa et al., https://doi.org/10.1016/j.jlb.2024.100281). Methods: RNA was extracted from the CSF cells of 63 cfCSF-Pos patients and 93 cfCSF-Neg patients. The RNA was sequenced and quantified using a targeted RNA panel of 1600 genes by next generation sequencing (NGS). We used two thirds of the samples for training a machine learning (ML) system and one third for testing. The ML system uses Bayesian statistics with k-fold cross-validation (with k = 12) to first rank the top biomarkers distinguishing CSF-Pos from CSF-Neg samples. Then Random Forest was used to distinguish between the two classes using the top ranked biomarkers. Results: cfCSF-Neg contains mainly T-cells with median CD2:CD22 RNA ratio of 70.07 (range 0.01-14820). This was not significantly different (p = 0.19) from cfCSF-Pos cases (median: 41.31, range: 0.4-10172). The ratio of CD4:CD8A in cfCSF-Neg (median: 4.89, range: 0.47-2522) was also not significantly different (p = 0.31) from that in cfCSF-Pos cases (median: 5.0, range: 0.29-48). While significant variation in the levels of T- and B-cells is noted within each group, there was no significant difference in the individual cell population (T-cells, B-cells, plasma cells, natural killer cells, neutrophils, monocytes, or dendritic cells) overall after adjusting for multiple testing. Despite this lack of difference in cell populations, the testing set showed that cfCSF-Pos patients can be readily distinguished from cfCSF-Neg patients with AUC of 0.886 (CI: 0.797-0.976) using 30 top genes selected by ML. Except for KRT8 and KRT19, the majority of the top genes selected by the ML algorithm suggests modulation of T-cell activation including but not limited to TBX21, CD3D, CD5, IKZF3, NFATC2, and INPP5D. Conclusions: The data suggest the CNS remains immunologically specialized with modulating the adaptive immune response in the setting of breast cancer metastasis. Significant modulation in the CSF T-cells suggests selective targeting with immunotherapy may prove beneficial with the potential for active monitoring using this combination CSF cellular and cfDNA approach.