Metastatic disease is the primary cause of death in patients with cancer; thus, identifying new therapeutic targets to stop the spread of cancer is needed. Triple-negative breast cancer (TNBC) is a highly metastatic subtype with the highest death rate in breast cancer and limited therapeutic options. Myeloid-Derived Suppressor Cells (MDSCs) are a heterogeneous and poorly defined group of tumour-infiltrating myeloid progenitors that can promote cancer progression and metastasis. MDSCs have an immunosuppressive function, however, MDSCs are indistinguishable from normal neutrophils and monocytes using traditional cell surface markers. Recently, it has been suggested that MDSCs contribute to the formation of the pre-metastatic niche but the molecular drivers that trigger MDSC differentiation towards the oncogenic and pro-metastatic phenotype remain poorly understood. Here, we have generated a comprehensive single-cell MDSC atlas of breast cancer in pre-isolated monocytic and polymorphonuclear MDSC populations from metastatic lungs from a highly aggressive TNBC model as well as a non-metastatic TNBC model as control. To generate this atlas, we have integrated single-cell information on the epigenome (ATAC-seq) and transcriptome (BD Rhapsody WTA). Our data shows that polymorphonuclear MDSCs (PMN-MDSCs) are more heterogeneous than monocytic MDSCs (M-MDSCs), and interestingly, we identify a specific PMN-MDSC subpopulation almost exclusive of the metastatic model. These metastasis-associated MDSCs are epigenetically distinct and can be subdivided into three subtypes depending on their differentiation status, the pre-metastatic MDSC, which were already present in the lung pre-metastatic niche and the intermediate and terminal MDSCs which were exclusive to the lungs at the time of metastasis. This suggests that MDSC recruitment is a prerequisite for metastasis, but further MDSC differentiation is needed for metastasis development. In conclusion, exploiting our high-resolution transcriptomic and epigenomic single-cell data we have generated a hierarchical atlas of MDSCs and identified oncogenic master regulators in metastatic preclinical models.