The Primary Tumor Immune Microenviornment Status Predicts the Response to Immunotherapy and Overall Survival in Breast Cancer

The tumor immune microenvironment (TIME) of breast cancer is a known source of tumor heterogeneity and it has been increasingly recognized as having a role in the course of disease. In the present study, we used a computational approach to dissect the landscape of TIME states among TCGA breast cancer patients. Our central hypothesis is that the pre-existing TIME states represent a dimension which is informative about the prognosis and the response to immunotherapy. In order to test this hypothesis, we first classified breast cancer patients according to their primary TIME status. Next, we describe a TIME-based classification with prognostic value for overall survival among the TCGA patients. We further demonstrated that absolute quantification of mast cells, M0 macrophages, CD8 T cells and neutrophils were predictive of overall survival. In order to identify the TIME states which, predict response to immune checkpoint blockade, we performed a similar analysis of 11 different mouse models of primary invasive breast carcinoma that were subsequently treated with immune checkpoint inhibitor (ICI) therapy. These analyses revealed that the TIME content of M1 macrophages, monocytes and resting dendritic cells were predictive of sensitivity to ICI therapy. Taken together, these results indicate that (1) the landscape of human primary TIME states is diverse and can identify patients with more or less aggressive disease and (2) that pre-existing TIME states may be able to identify patients, of all molecular subtypes of breast cancer, who are good candidates for ICI therapy.


62
ICI therapies block the inactivation of the anti-tumor immune response by the tumor itself, 63 thus promoting immune-mediated cell killing of the tumor. Therefore, the primary tumor immune 64 microenvironment (TIME) may have a role in determining the effect of ICI therapies in breast 65 cancer by establishing a permissive or suppressive microenvironment for the immune system 66 thereby adding to or detracting from the effect of ICI therapy, respectively.

67
Recent efforts to identify biomarkers for and mechanisms of resistance to ICI therapy have 68 focused primarily on genetic or tumor intrinsic modes including the mutational burdon of the 69 tumor[6] and the expression of ICI target molecules and immune modulating genes including PDL-70 1 itself [7][8][9]. Relatively little, however, is known about the exact cellular composition of the 71 primary TIME of breast cancer in general and which TIME states specifically are predictive of ICI 72 response. Early work in identifying the TIME determinants of response to ICI therapy has 73 demonstrated the importance of the tumor lymphocyte (TIL) and macrophage abundance [10][11][12] 74 However, with recent the development of single cell high-throughput sequencing, studies have 75 begun to dissect the breast cancer TIME at the cell type compositional level [13,14]. Despite this 76 technological advancement, high cost and computational constraints remain and it is largely 77 infeasible to perform these experiments at the scale required to achieve the statistical power to 78 characterize the complete landscape of TIME statuses present among large, heterogenous 79 cohorts of breast cancer patients and associate trends in this landscape with clinical outcomes.

80
In the present study, we apply CIBERSORT [15], a computational approach to infer the 81 abundance of specific cell types from bulk RNA-seq data, to 922 individual samples of human 82 primary breast cancer from TCGA [16]. Using this approach, we were able to interrogate the 83 trends in the TIME statuses of these patients that are associated with prognosis of the disease.
84 In addition, we were able to apply a similar approach to mouse models of metastatic breast cancer 85 to identify TIME states that are predictive of objective response to ICI therapy in these models.     157 Interestingly, patients with any detectable TIME content of neutrophils (p=0.0056) and eosinophils 158 (p=0.0056) had significantly worse prognosis compared to those without.

159
Based on the predictive capacity of lymphocytes and granulocytes in the primary breast 160 TIME, we designed a stronger classifier by aggregating the individual cell content information for 161 each of these individual cell types -which function as weak classifiers in our model. Patients with 162 naïve B cell content and CD8 T cell content higher than the 25th percentile for all patients and 163 undetectable neutrophil or eosinophil content were assigned to class 1 (n = 131). Patients with 164 the inverse TIME profiles were assigned to class 3 (n = 144), and patients that failed to fit into 166 As expected, average CD8 T Cell and naïve B cell content were significantly higher in 167 class 1 patients compared to class 3 (p < 2.2e-16, Figure 1c). Class 2 patients tended to have 168 greater numbers of CD8 T cells and naïve B cells compared to those in class 3 (p < 2.2e-16), but 169 significantly lower than those in class 1 (p < 2.2e-16). Neutrophil and Eosinophil content was low 170 in both class 1 and class 2 patients but only significantly higher in class 3 compared to both 171 classes 1 and 2 (p = 8.08e-10 and p = 0.0011, respectively).

172
Consistent with the hypothesis that primary TIME status of primary tumors is predictive of 173 outcomes, our primary TIME classification strategy was able to identify patients significantly 174 different overall survival (p < 0.0001, Figure 1d). Moreover, this predictive capacity of the primary 175 TIME status remained statistically significant after controlling for molecular subtype (Figure 1e).

186
Given the capacity of the human primary TIME to predict prognosis in the context of 187 conventional therapeutic strategies, we next sought to test whether the primary TIME was 188 informative for the response to immune checkpoint inhibition (ICI) therapy (Figure 2a).
189 CIBERSORT analysis was performed on publicly available RNA-seq data obtained from a panel 190 of 11 different mouse models of triple negative breast cancer, which were subsequently treated 218 Consistent with other reports, our analysis revealed that TIME content of lymphocytes was 219 predictive of the objective response to ICI treatment (supplemental figure 1). Specifically, plasma 220 cell, CD8 T cell and CD4 memory T cell contents were significantly associated with response.

234
The ICI Response Score of the primary TIME was significantly higher among animals that

264
Specifically, the relationship between tumor lymphocyte content and overall survival are 265 supported by our data, however we additionally identify that high neutrophil and eosinophil content 266 confers a worse overall prognosis. Our novel primary TIME-based classification strategy 267 incorporates these findings and demonstrates that the TIME state of the primary tumor indeed 268 has prognostic value, independently of the molecular subtype of the tumor.

269
Furthermore, in order to use the primary TIME status to identify those patients who could 270 benefit from immune checkpoint inhibition therapy, we developed an ICI Response Scored based 271 on primary TIME status. A subset of patients of all molecular subtypes and primary TIME classes 272 were identified with high ICI Response Scores, suggesting that future ICI therapy preclinical and 273 clinical trials would benefit from stratification methods which consider the primary TIME prior to 274 enrolment.

275
Taken together, these findings demonstrate that pre-existing TIME states are relevant to 276 both the prognosis of breast cancer patients and to the choice of therapy. Future studies should 277 further dissect these TIME states by flow cytometric and/or single cell RNA-sequencing 278 approaches in prospective studies.