Use of murine models to identify tumor immune components that correlate with response to immunotherapy in breast cancer

Background: The heterogeneity of the breast tumor microenvironment (TME) may contribute to the lack of durable responses to immune checkpoint blockade (ICB), however, mouse models to test this are currently lacking. Proper choice and use of pre-clinical models are necessary for rigorous, pre-clinical studies to rapidly move laboratory findings into the clinic to treat patients. Methods: To elucidate how tumor latency and the heterogeneity of the TME contributes to ICB resistance, we performed comprehensive characterization of the TME using quantitative flow-cytometry and RNA expression analysis (NanoString) utilizing three distinct breast cancer models, all derived from the same autochthonous model. Tumor cells were obtained from the commonly used MMTV-PyMT murine breast cancer model and 1E6, 1E5 or 1E4 cells were either immediately injected into the mammary fat pad of FVB/NJ wild type mice or frozen (i.e. the tumor inoculation derived from the MMTV-PyMT tumors were never cultured). We then correlated the immunophenotyping to the efficacy observed from ICB. Results: These studies reveal that the number of cells used to generate syngeneic tumors significantly influences tumor latency, the infiltrating leukocyte population and response to ICB. The 4 models had vastly different TMEs which correlated to responses to ICB. Compared to the autochthonous model, all three syngeneic models had significantly more tumor infiltrating lymphocytes (TILs; CD3 + , CD4 + , and CD8 + ) and higher proportions of PD-L1 positive myeloid cells, whereas the MMTV-PyMT model had the highest frequency of myeloid cells out of total leukocytes. Increased TILs correlated with response to anti-PD-L1 and anti-CTLA-4 therapy; but tumor cell PD-L1expression and T-cell PD-1 expression did not. Conclusions: Here we have identified ICB-sensitive and resistant breast cancer models, generated from the same tumor cell inoculum. These models represent an opportunity to further interrogate the TME for breast cancer treatment and provide novel insights into therapeutic combinations and response to ICB. We believe this work serves as an important resource for the field to inform proper mouse model selection for pre-clinical studies.


Background
The success of immune checkpoint blockade (ICB) in a variety of human cancer types has stimulated interest in its use for the treatment of breast cancer. Current therapy for breast cancer is guided by the molecular pathology of the tumor. Breast cancer is often driven by overactive hormone signaling (estrogen and/or progesterone receptors; ER, PR) or amplification of growth factor response (HER2) and therefore treated with endocrine therapy or HER2-targeted agents; alternatively patients can also be treated with general therapies such as chemotherapy and/or radiation. [1][2][3][4] After initial treatment for early stage disease, approximately 30% of women will eventually develop recurrent advanced or metastatic disease. 5 Almost all who develop metastatic breast cancer will succumb to the disease, highlighting the need for more effective strategies. 6 ICB aims to target T-cell inhibitory molecules using antibodies against cytotoxic T-lymphocyte associated protein 4 (CTLA-4) and programmed cell death protein-1 (PD-1) as well as its ligand (PD-L1). ICB works by reinvigorating anti-tumor immune responses by inhibiting negative interactions between T-cells and antigen presenting cells (APCs) or tumor cells in several cancer types. 7,8 In 2011, the first ICB agent, ipilimumab, a human monoclonal antibody targeting CTLA-4, was approved by the FDA for treatment of metastatic melanoma based on significant improvement in overall survival in a randomized, double-blinded Phase III study. 9 Importantly, ipilimumab has doubled 10-year survival for metastatic melanoma compared with historical data. [10][11][12] Antibodies targeting PD-1 and PD-L1 have also shown a durable clinical response in melanoma, as well as renal cell carcinoma, non-small cell lung cancer, and bladder cancer. [13][14][15][16][17][18][19][20] To date, such responses have led to FDA approval of seven ICB therapies for the treatment of more than 15 different types of cancer. 21 Because their effector pathways are distinct, there are reasons to believe that the combination of CTLA-4 and PD-1/PD-L1 therapy can provide an enhanced response. 22,23 While ICB has shown promising results in a subset of patients with other cancer types, responses in breast cancer has demonstrated only minimal responses. 24 Breast cancer has a lower mutational burden compared to other types of cancer which may explain the lack of efficacy in response to ICB. 25,26 Despite this generalization, triple negative breast cancer (TNBC) has demonstrated some benefit from ICB therapy, albeit not achieving the response rates demonstrated in melanoma and lung cancer. While pembrolizumab (anti-PD-1) showed promising activity as a single agent against advanced or metastatic TNBC in the KEYNOTE-012 (NCT01848834) and KEYNOTE-086 (NCT02447003) clinical trials, in the randomized, Phase III KEYNOTE-119 (NCT02555657) clinical trial there was no improvement in overall survival compared to single-agent chemotherapy in metastatic TNBC. 27 Benefit from ICB therapy has been observed in patients treated in the first line setting and/or in patients whose tumors or immune cells express PD-L1. [28][29][30] For example, in the PCD4989g (NCT01375842) clinical trial, evaluating Atezolizumab as a single agent, expression of PD-L1 on 1% or greater of immune infiltrating cells was associated with a 12% ORR compared to 0% when there was no expression of PD-L1, and high levels of immune cell infiltration (greater than 10%) was independently associated with higher overall response rate (ORR) and overall survival (OS). 29 Importantly, the first FDA approval of ICB in breast cancer came from the Phase III IMpassion130 (NCT02425891) clinical trial that demonstrated atezolizumab in combination with nab-paclitaxel showed significant extension of median disease-free overall survival compared to nab-paclitaxel alone, from 15.5 to 25 months in patients with 1% or more PD-L1 positive immune cells, where there was no benefit in PD-L1 negative tumors. 31,32 The use of PD-L1 as a biomarker for ICB has been rigorously investigated but has raised concerns including poor agreement between different antibodies as well as scoring between pathologists. 33 To date, there are 9 FDA approvals for the use of ICB based on a specific PD-L1 threshold and companion diagnostic, with variable thresholds both within and across tumor types using several different assays, including approvals at the following PD-L1 positive percentage thresholds: 1, 5, and 50%. In a recent meta-analysis that examined all approvals of ICB as of April 2019, PD-L1 was predictive in 28.9% of those approvals, and was either not predictive (53.3%) or not tested (17.8%) in the remaining approvals. 21 Other predictors of response to ICB include the presence of immune cells, as immune cells within HER2 + and TNBC tumors have been shown to correlate with better response to HER2-targeted therapy and chemotherapy, respectively. 34 However, immune cell infiltration has been reported to differ among each subtype of breast cancer. 35 Further work to characterize the TME of breast tumors will provide opportunity for ICB therapy in these patients.
Mouse models have been instrumental in understanding the molecular mechanisms of oncogenesis and metastasis. Being able to translate in vivo pre-clinical findings to patients depends largely on how accurately the mouse model replicates histological markers, biochemical pathways, and genetic aberrations observed in the same human tumor type. 36 In light of the advances in immunotherapy, it is now also necessary to meticulously characterize the TME of pre-clinical mouse models. Therefore, here we have generated and characterized three models of breast cancer, derived from a commonly used pre-clinical autochthonous model of breast cancer, in which the polyoma middle T (PyMT) oncogene is driven by the mouse mammary tumor virus (MMTV)-LTR. The MMTV-PyMT model is representative of human breast carcinomas; where several of the same signaling transduction pathways that are commonly disrupted in human breast cancer patients are seen in the MMTV-PyMT model such as Src family, Ras and PI3K kinase pathways. 37,38 In addition, both innate and adaptive immune cells infiltrate into the tumor during tumorigenesis. 39,40 Macrophages have been shown to play a key role in the development of these tumors, in which CCL2 recruits inflammatory monocytes to facilitate breast tumor growth and metastasis. 41 Additionally, it's been shown that the phenotype is mediated through IL-4 expressing CD4 + T-cells. 40 A spectrum of macrophage phenotypes have been recognized, ranging from classically activated macrophages ("M1"-like) that are effective in clearing intracellular pathogens and can recruit cytotoxic T lymphocytes to activate adaptive immune responses 42 ; to alternatively activated macrophages ("M2"-like), which function to help with parasite clearance, exhibit tissue remodeling capabilities and promote tumor progression by recruiting T regulatory and Th2 T-cell subsets lacking cytotoxic functions. 43 Tumor associated macrophages (TAMs) likely exhibit features of both M1-and M2-like macrophages but in general, exhibit an M2-like phenotype and promote tumor progression and metastasis by secreting factors that regulate angiogenesis and that recruit tumor suppressive cells such as T regulatory (Treg) cells. 44,45 In line with the lack of clinical efficacy of ICB in breast cancer, several studies have shown that MMTV-PyMT mice are resistant to ICB monotherapy. 39,46,47 With respect to ICB therapy in the clinical care of breast cancer patients, it is currently unclear if immune cell infiltration, including type, number and/or phenotype, correlates to responses. Understanding what factors are critical for ICB efficacy in breast cancer will allow careful patient selection and/or catalyze clinical development of novel therapies to convert nonresponders to responders, deepen responses that do occur, surmount acquired resistance to immunotherapy and identify biomarkers that can more accurately predict durable response.
Therefore, here we generate three distinct syngeneic models derived from the MMTV-PyMT model and provide deep immuno-phenotypic analysis of biomarkers and mechanisms that correlate with efficacy to ICB therapy.

Animal husbandry
All experiments used either virgin female FVB/NJ mice or virgin female FVB/N autochthonous mice carrying the polyoma middle T (PyMT) transgene under the control of the mammary tumor virus (MMTV) promoter were used. The FVB/NJ mice were purchased from Jackson laboratory (001800). All mice were maintained within the Dana-Farer Cancer Center (DFCI) and all experiments were conducted under The Institutional Animal Care and Use Committee (IACUC).

Generation of syngeneic models
Late stage MMTV-PyMT tumors were harvested and tumor suspension was either immediately injected into recipient FVB/NJ wild-type mice or frozen for subsequent experiments. The tumor suspension was never cultured. Each experiment was performed with 3 different batches of cells harvested from MMTV-PyMT mice. FVBN/J mice were inoculated with one million (1E6), one hundred thousand (1E5), or ten thousand (1E4) cells in the 4 th mammary fat pad to generate syngeneic models.

Tumor digestion
Tumors were extracted and minced, and subsequently blended using the gentleMACS  3 . Tumors from mammary fat pad numbers 5 and 10 were excluded from the analysis. The sum of the volumes for the MMTV-PyMT autochthonous tumors (1-4 and 5-9) were used and indicated as "total tumor burden". The syngeneic mice that had tumors that measured 80-100 mm 3 were enrolled into an experiment. At the indicated time points animals were euthanized in a CO2 chamber before performing a cardiac perfusion with normal saline. Lungs and tumors were removed for analysis.

Flow cytometry
Tumors were digested as described above, and single cells were re-suspended in a buffer

RNA isolation
When syngeneic mouse tumors reached 100 mm 3 , tissue samples were snap frozen for later processing. Samples were also collected from autochthonous mice with total tumor burden in the range of 300-600 mm 3 49 Cell type scores were calculated using the average log2 normalized expression of each cell type's marker genes. The cell type abundance scoring is modified from other reports 50 where strict cell type gene correlation-driven QC p-values were determined based on data that passed QC. The cell scoring employed within the manuscript makes assumptions that result in the cell typing data being best described as trends.

Results
Tumor infiltrating leukocyte populations are significantly altered in the 4 pre-clinical murine breast cancer models.
Three distinct pre-clinical models of breast cancer were generated by isolating tumor cells Our group previously described that converting pro-tumor macrophages to an anti-tumor phenotype induced reduction of primary and metastatic tumors, indicating that the myeloid cell population is a major contributor to disease progression. 39 The phenotype of tumor macrophages has previously been shown to correspond to drug sensitivity and disease outcome in this model, therefore we further investigated macrophage phenotype across the different models 39,40 . Markers suggesting that the 1E6 and 1E5 models had more suppressive myeloid cells, in line with their higher frequencies of PD-L1 + myeloid cells.

The breast cancer models have distinct tumor immune transcriptional profiles.
To interrogate mRNA transcripts expressed by cells in each of the distinct tumors, Responders and non-responders also clustered in the pathway analysis, where the inflammation pathway revealed significant differences between the responders and non-responders (Supplemental Figures 8d,e).

Discussion
There is a critical need for biomarkers to predict response to ICB in breast cancer, and mouse models are currently lacking. Given the substantial heterogeneity of the TME, conclusions based on specific mouse models might limit generalizations, especially regarding detailed characterization of molecular signaling mechanisms. Here we exploited a syngeneic mouse model to make two major findings. First, the initiating conditions of the tumor (in this case, the number of cells in the inoculum) can dramatically alter the tumor immune microenvironment.
Second, we found that these differences in the TME were closely related to quality of response to ICB. In this study we used cells derived from tumors that spontaneously arise in the MMTV-PyMT murine model of breast cancer to generate three distinct syngeneic models using 1E6, 1E5 or 1E4 cells injected into the mammary fat pad of wild-type FVB/NJ mice. Our findings are the first to report a detailed characterization of difference in the TME as a variable of the number of cells injected to generate syngeneic tumors. Importantly, we find that only the 1E6 and 1E5 models responded to ICB, whereas the 1E4 and MMTV-PyMT model are resistant. Our data suggest that protection from the inhibitory effects of Tregs and presence of high numbers of Tcells and macrophages paired with enhanced antigen processing capabilities, as seen in the syngeneic models sensitive to anti-CTLA4 and anti-PD-L1 therapy, might be a setting where ICB may have great therapeutic potential for breast cancer. We found that the 1E6 model had the highest absolute number of T-cells and myeloid cells, and was responsive to ICB. Additionally, we found when comparing ICB-sensitive models (1E6 and 1E5) to the ICB-resistant models (1E4 and autochthonous), there is a significant difference in gene signatures related to antigen presentation and innate immune cell response. These data support our hypothesis that in addition to T-cells, macrophages and other myeloid cells are required to play a critical role in initiating an anti-tumor immune response.
Macrophages play an essential role in T-cell activation by presenting antigen and providing activating and stimulatory cytokines essential for T-cell function. 54 In addition, macrophages can mediate antibody dependent cellular toxicity of cancer cells 55 as well as eliminate cancer cells through FcR-mediated phagocytosis. 56 However, TAMs can also dampen effector T-cell function by producing IL-10 that in turn increase their own PD-L1 expression and suppresses cytotoxic T-cell responses. 57 Interestingly, we found that the ICB-sensitive models had the highest absolute number of CD11b + myeloid and F480 + macrophages. These myeloid cells were more suppressive than those found in the ICB-resistant tumor models; indicated by a higher proportion of myeloid cells expressing PD-L1 + (Fig. 4f), as well as a lower ratio of M1:M2 macrophages that suggested more M2-like macrophages (Supl. Fig. 3). In line with these observations, we found that transcripts levels related to Ccl2 and its receptor were higher in the ICB-sensitive models. CCL2 is a cytokine largely known for its involvement in the recruitment of CCR2+ monocytes from the bone marrow to other sites in the body where they differentiate into macrophages 58,59 . Additionally, CCL2 has been shown to recruit monocytes and macrophages to breast tumors and to facilitate breast cancer metastasis. 60,61 The CCL2/CCR2 axis may represent an unique opportunity for anti-cancer therapy and work in this area is already being explored. 62,63 The combination of CCL2 antagonism with anti-PD-1 has demonstrated efficacy in some mouse models. 64 It is currently unclear if the absolute number of myeloid cells within the TME or the proportion of myeloid cells of total CD45 + immune cells is a more important factor for the efficacy of ICB. The data here suggest that the former is a stronger predictor of response, and that the phenotype might not be as critical since the ratio of M1:M2 macrophages was higher in the ICB-  (Fig. 3a,g), but did not find a pattern of ICB response related to CTL numbers or proportions, as the autochthonous model had the highest frequency but was resistant to ICB. We did observe a correlation between baseline PD-L1 expression of myeloid cells (Fig.   4f) but not tumor cells (Fig. 4g) and response to ICB. This is an important observation seeing as inclusion criteria for some ICB treatment and/or clinical trials can include assessment of PD-L1 expression (NCT03258788, NCT02536794). We also noted by NanoString gene expression analysis the 1E6 tumors had elevated levels of both CD274 (PD-L1) and CTLA-4; which corresponds with the response to anti-CTLA-4 and anti-PD-L1 monotherapy (Fig. 5e). A limitation to this work is that the TME was not assessed after ICB, which may reveal additional changes that correlate with response to therapy.

Conclusion
The evasion of immune surveillance is a challenge in breast ICB therapy that warrants further investigation. Mechanistic understanding of how the TME promotes tumor progression will be critical to understand which cell populations play the most influential role in promoting an immune escape. Overall, we found that by immunophenotyping the three syngeneic models compared to the autochthonous model provided valuable insights for sensitivity to anti-CTLA4 and anti-PD-L1 therapy. We've uncovered that innate immunity and antigen processing may play a vital role in determining response to checkpoint blockade. Further work to characterize the signals within the TME that promote immune evasion will be vital to advancing checkpoint blockade therapy for the treatment of breast cancer. The four models of murine breast cancer presented here, of which 2 are ICB-sensitive and 2 are ICB-resistant, represent a unique opportunity to further interrogate biomarkers of response to ICB.

Consent for publication
n/a

Availability of supporting data
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.