CD39+PD-1+CD8+ T cells mediate metastatic dormancy in breast cancer

Some breast tumors metastasize aggressively whereas others remain dormant for years. The mechanism governing metastatic dormancy remains largely unknown. Through high-parametric single-cell mapping in mice, we identify a discrete population of CD39+PD-1+CD8+ T cells in primary tumors and in dormant metastasis, which is hardly found in aggressively metastasizing tumors. Using blocking antibodies, we find that dormancy depends on TNFα and IFNγ. Immunotherapy reduces the number of dormant cancer cells in the lungs. Adoptive transfer of purified CD39+PD-1+CD8+ T cells prevents metastatic outgrowth. In human breast cancer, the frequency of CD39+PD-1+CD8+ but not total CD8+ T cells correlates with delayed metastatic relapse after resection (disease-free survival), thus underlining the biological relevance of CD39+PD-1+CD8+ T cells for controlling experimental and human breast cancer. Thus, we suggest that a primary breast tumor could prime a systemic, CD39+PD-1+CD8+ T cell response that favors metastatic dormancy in the lungs.

Metastasis is the major cause of death in patients with breast cancer. The metastatic behavior of this disease is heterogeneous regarding affected organs as well as timing1,2. Whereas some patients develop metastasis shortly after or even before diagnosis, others show metastatic lesions only years or decades after removal of the primary tumor1,3. In fact, ~20% of disease-free patients will relapse within 7 years after resection, and patients that appear to be cured from breast cancer have a higher mortality than the rest of the population even 20 years after surgery4,5. Such late relapses are thought to result from disseminated cancer cells (DCCs) that reached different organs but remained dormant for several years6. Indeed, the presence of DCCs in the bone marrow of breast cancer patients loosely correlates with the development of metastatic relapse7,8.

The cancer cell-intrinsic and -extrinsic mechanisms governing metastatic dormancy are poorly understood. Some cancer cell-intrinsic factors have been associated with dormancy, such as inhibition of the PI3K-AKT pathway9, activation of p38 or triggering of an endoplasmic reticulum stress response10. In addition, the microenvironment in the (pre)metastatic organ may induce dormancy11 by different signals such as an endothelial cell-derived thrombospondin-112 or TGF-β2 produced in the bone marrow13. Only few studies addressed the influence of cancer cell-extrinsic factors, such as innate14,15 and adaptive immunity16,17 on dormancy. Although infiltration of the primary tumor by T cells was shown to correlate with a good prognosis18, it remains unclear whether this correlation is explained by cytotoxic T cells that eliminate cancer cells or by T cells that prevent the outgrowth of cancer cells and induce dormancy19,20.

Here we sought to understand the mechanism that governs metastatic dormancy of DCCs and discovered that the primary tumor primes a systemic, CD8+ T cell response that prevents metastatic outgrowth of DCCs in the lungs. The protective T cells expressed markers that are generally associated with activation and exhaustion. Promoting such a response may provide a rationale for the development of future immunotherapies that aim to prevent metastatic relapse.

Here, we show that CD39+PD-1+CD8+ T cells mediates metastatic dormancy in a preclinical model for breast cancer and correlate with increased disease-free survival post-resection in breast cancer patients.

Disseminated 4T07 breast cancer cells are dormant

We used two preclinical models for breast cancer that were originally derived from the same spontaneous tumor in BALB/c mice, 4T1 and 4T0721 (Fig. 1a). 4T1 orthotopic breast cancer produces macro-metastasis in different organs22,23, whereas 4T07 is essentially non-metastatic. Some studies, however, showed the presence of disseminated 4T07 cells in distal organs21,24.

Orthotopically injected 4T1 and 4T07 cells showed comparable breast cancer growth and incidence in immunocompetent BALB/c mice, although 4T1 tumors grew faster at late stages (Supplementary Fig. 1a and b). But, whereas 4T1 breast cancer readily induced aggressively growing and macroscopically visible lung metastasis, 4T07 did not (Supplementary Fig. 1c and d). Despite absence of macro-metastasis, we found disseminated 4T07 cells in the lungs (Supplementary Fig. 1e) of all mice, suggesting that 4T07 DCCs seed the lungs but fail to grow out. Presence of dormant DCCs in the lungs of all mice bearing a 4T07-mCh primary tumor was confirmed and quantified by immunofluorescence (Fig. 1b–d, and Supplementary Fig. 1f). Disseminated 4T07 cells were present as single cells and were Ki67-negative, as has been previously described for dormant DCCs15,17. The fact that we found non-proliferating, disseminated 4T07 21 days after resection of the primary tumor (Fig. 1e–g) further underscores the dormant state of 4T07 DCCs.

Breast cancer induces CD8+ T cells resulting in dormancy

We first investigated whether cancer cell-intrinsic features influenced the outgrowth of DCCs in the lungs by injecting 4T1 or 4T07 breast cancer cells intravenously (i.v.) into naïve mice (Fig. 2a). Under these experimental conditions, both cell lines comparably formed macro-metastases (Fig. 2b), suggesting that both cell lines can progressively grow in the lungs.

The myeloid cell compartment was shown to undergo cancer-induced changes that promote metastasis by creating a pre-metastatic niche in the 4T1 model25,26. Specifically, 4T1 primary tumors induce systemic neutrophilia27 and splenomegaly26, as well as accumulation of inflammatory monocytes28, eosinophils, neutrophils27 and alveolar macrophages29 in the pre-metastatic lung. We therefore compared 4T1- and 4T07-induced changes on the myeloid compartment in blood and lungs and observed no differences regarding splenomegaly (Supplementary Fig. 2a), or neutrophilia (Supplementary Fig. 2b). The lungs of mice with 4T1 or 4T07 breast cancer were similar concerning the presence of inflammatory monocytes (Supplementary Fig. 2c), neutrophils (Supplementary Fig. 2d), eosinophils (Supplementary Fig. 2e) and alveolar macrophages (Supplementary Fig. 2f) as determined by flow cytometry (Supplementary Fig. 2g). We saw the previously reported accumulation of myeloid cells in the lungs during breast cancer progression27 in 4T1 and 4T07 alike (Supplementary Fig. 2). Thus, the difference in metastatic behavior of 4T1 and 4T07 breast cancer cannot be explained by systemic changes in the myeloid compartment.

Orthotopic 4T07 tumors induced progressively growing lung metastases in T cell-deficient BALB/c Foxn1nu/nu mice, whereas T cell-deficiency hardly influenced metastatic behavior of 4T1 tumors (Fig. 2c–e). To compare lung metastases in both strains, we analyzed wild type and Foxn1nu/nu mice at the same time point after injection and added an additional group of wild type mice in which breast tumors were allowed to progress until they matched the size in Foxn1nu/nu mice (WT late) (Fig. 2d and e). Thus, metastatic dormancy of disseminated 4T07 breast cancer cells completely depends on T cells. The fact that 4T1 cells are intrinsically more metastatic than 4T07 cells in immunodeficient mice reflects (a combination of) traits that are different between 4T1 and 4T07 cells, some of which are unrelated to T-cells. In fact, we think that the many differences between 4T1 and 4T07 cells preclude appropriate and conclusive comparison in vivo.

To study whether CD8+ T cells are responsible for metastatic dormancy, we depleted CD8+ T cells from mice followed by orthotopic injection of 4T07 cells and subsequent analysis of lung metastatic load by IVIS (Fig. 2f). While the growth of primary tumors was unaffected by CD8-depletion, disseminated 4T07-LZ cells grew out to macro-metastases in the absence of CD8+ T cells (Fig. 2g and h), suggesting that primary 4T07 breast cancer induces CD8+ T cell-dependent immunity.

To test the hypothesis that CD8+ T cells are essential for metastatic dormancy, we orthotopically injected untagged 4T07 cells (or PBS as control) followed by an i.v. challenge with luciferase-tagged 4T07-LZ cells 11 days later (Fig. 2i). If the primary tumor had induced protective immunity, we would expect a reduction of lung metastatic load. Because we measured lung metastatic load by bioluminescence, we specifically quantified the i.v. injected, luciferase-tagged 4T07 cells. The presence of a primary 4T07 breast tumor prevented metastatic outgrowth of i.v. injected 4T07-LZ cells (Fig. 2i–k) but did not influence the amount of seeding as measured 0.5 and 3 h after i.v. injection (Supplementary Fig. 3a and b). At the endpoint, i.v. injected cells were present as disseminated, non-cycling single 4T07 cells in the lungs (Supplementary Fig. 3c and d). Resection of the primary tumor before i.v. challenge did not interfere with dormancy. Specifically, i.v. injected 4T07-mCh cells readily induced macroscopically visible lung metastasis in control mice (PBS in breast and mock surgery; Supplementary Fig. 3e, f), whereas only single, non-proliferating 4T07 cells were detected in the lungs of 4T07 tumor-bearing mice, independent of resection (Supplementary Fig. 3e, g). We confirmed these results in a similar experimental set-up using 4T07-LZ cells and bioluminescence as read out (Supplementary Fig. 4a, b). Depletion of CD8+ cells before i.v. challenge enabled metastatic outgrowth (Fig. 2l and m), confirming the dependence of dormancy on CD8+ T cells. In contrast, 4T07 nor 4T1 breast cancer protected against experimental 4T1 lung metastasis, although 4T07-induced systemic immunity reduced the metastatic load resulting from i.v. injected 4T1-LZ cells (Supplementary Fig. 4c and d). Although the lung metastatic load resulting from i.v. injected 4T1 cells is reduced in 4T07-bearing mice, the lesions are still progressive (i.e., multicellular and Ki67+) and not dormant (Supplementary Fig. 4a–d). Furthermore, we think that the number of adoptively transferred CD39+PD-1+CD8+ T-cells is simply insufficient to control 4T1 cells that have an intrinsically higher metastatic potential.

Together, these data suggest that 4T07 breast cancer induces a systemic, protective immune response that mediates dormancy of DCCs in the lungs.

CD39+PD-1+CD8+ T cells mediate metastatic dormancy

First, we analyzed the presence of different immune cells in primary 4T1 and 4T07 tumors (Supplementary Fig. 5a). We found no differences in the number of eosinophils (Supplementary Fig. 5b), neutrophils (Supplementary Fig. 5c), NK cells (Supplementary Fig. 5d) or inflammatory monocytes (Supplementary Fig. 5e), but 4T1 tumors contained a higher number of macrophages than 4T07 tumors (Supplementary Fig. 5f).

Since we found that 4T07 metastatic dormancy depends on CD8+ T cells, we characterized these cells in 4T1 and 4T07 primary tumors by high-dimensional flow cytometry. First, we gated on T cells (Supplementary Fig. 6a), visualized all the markers in our 23-parameter panel using a two-dimensional t-stochastic neighbor embedding (tSNE) projection30 (Fig. 3a) and clustered the cells using FlowSOM31,32 (Fig. 3a and b, upper panels and lower left panel). Based on the median marker intensities observed in the clusters (Fig. 3b, lower right panel), we annotated the major cell populations (CD8+ T cells, CD4+ T cells, and CD4+ CD25+regulatory T cells) and determined their frequency and the position of each population in the tSNE projection. In addition, not only the number of CD39+PD-1+CD8+ T cells was higher in 4T07 than in 4T1 tumors, but also their proportion of CD8+ T cells. The frequency of regulatory T cells was similar in 4T1 and 4T07 tumors, while conventional CD4+ cells were more abundant in 4T1 tumors compared to 4T07. In contrast, 4T07 primary tumors contained a higher proportion of CD8+ T cells (Fig. 3b, lower left panel).

Of note, within the CD8 compartment (Fig. 3c), we observed that 4T07 breast cancer accumulated CD39+PD-1+CD8+ T cells (Fig. 3d).

To substantiate that enrichment of CD39+PD-1+CD8+ T cells is the key immunological difference between 4T07 and 4T1 breast cancers, we analyzed our high-dimensional flow cytometry data using the unbiased representation-learning algorithm CellCnn33,34. In agreement with FlowSOM analysis, CellCnn detected a population with high expression of CD39, PD-1, LAG3, and Tim-3 (Fig. 3e and Supplementary Fig. 6a and b). Indeed, the two markers that best defined the population and showed the biggest differential abundance in terms of the Kolmogorov–Smirnov two-sample test between the whole-cell population and the selected cell subsets (higher KS value) were PD-1 and CD39 (Fig. 3e). In addition, CD39+PD-1+CD8+ T cells expressed more of LAG-3 and Tim-3 (Fig. 3f), which characterizes this population as effector cells that experienced recent T cell receptor engagement35,36 but may also indicate exhaustion37,38. This population was around three times more abundant in 4T07 compared to 4T1 tumors (Fig. 3g and h, p < 0.001). Therefore, we analyzed the functionality of CD39+PD-1+CD8+ T cells sorted from 4T07 tumors and confirmed the production of IFNγ and TNFα (Supplementary Fig. 6c). In addition, CD39+PD-1+CD8+ T cells displayed cytotoxicity toward 4T07 as well as 4T1 cells in vitro (Supplementary Fig. 6d). We found increased presence of such CD39+PD-1+CD8+ T cells already at d 10 in primary 4T07 (and not 4T1) breast cancers (Supplementary Fig. 7a–d). Together, these results suggest that CD39+PD-1+CD8+ T cells in 4T07 tumors possess protective effector functions.

Importantly, CD39+PD-1+CD8+ T cells are found in the lungs of mice with primary 4T07 breast cancer (Fig. 3i). A proportion of this population expresses CD103, hence resembling tissue-resident memory T cells39,40 (Fig. 3j). Our observation that CD39+PD-1+CD8+ T cells did not express CD103 in the primary tumor (Fig. 3f) is in line with recent data showing that tissue-resident-like memory cells in human breast cancer are CD103-negative41. Thus, accumulation of CD39+PD-1+CD8+ T cells is the main immunological difference between dormant and metastatic tumors.

Because the CD8+ cells specifically emerging in 4T07 express PD-1, we wondered whether blocking PD-1 may improve their presumed protective effect. Indeed, immunotherapy with anti-PD-1 reduce the number of disseminated 4T07 cells in the lungs approximately twofold (Fig. 4a and b), suggesting that the PD-1+ subset of CD8+ T cells comprises protective effector cells. To investigate whether the CD39+PD-1+CD8+ T cell subset indeed controls metastatic outgrowth and therefore mediates metastatic dormancy, we sorted this population from 4T07 breast cancer and adoptively transferred them to naïve BALB/c mice followed by an i.v. injection of 4T07-LZ cells (Fig. 4c and d). Adoptive transfer of CD39+PD-1+CD8+ T cells prevented 4T07 lung metastasis, whereas injection of PBS or CD8+ T cells lacking CD39+PD-1+ cells (referred to as other CD8+ T cells) did not (Fig. 4e and f). In these experiments, we did not visualize the dormant cells, thus leaving some room for the possibility that the i.v. injected cells actually never reached the lungs when CD39+PD-1+CD8+T-cells were transferred. For this reason, we performed the same experiment and visualized disseminated 4T07-mCh cells in the lungs. We observed metastatic outgrowth in all lungs of mice that did not receive CD39+PD-1+CD8+ T-cells; in contrast, in mice that received CD39+PD-1+CD8+ T-cells, we found single, non-proliferating and thus dormant 4T07-mCh cells in the lungs (Supplementary Fig. 8). Thus, our results suggest that CD39+PD-1+CD8+ T cells are necessary and sufficient to control disseminated 4T07 breast cancer cells.

The presence of non-proliferating disseminated 4T07 cells in the lungs suggests that CD8+ T cells mediate dormancy by cell cycle arrest. Therefore, we tested whether CD39+PD-1+CD8+T cell-derived IFNγ and TNFα can induce senescence42 by exposing 4T07 cells to those cytokines in vitro. We observed a significant increase in the number of senescent cells as measured by expression of senescence-associated β-galactosidase activity (Supplementary Fig. 9a and b). To confirm the relevance of IFNα/TNFα-induced senescence induction in vivo, we blocked both cytokines in mice with 4T07-mCh breast cancer and analyzed the status of disseminated 4T07 cells in the lungs by immunofluorescence (Fig. 4g). Blocking IFNγ and TNFα allowed the metastatic outbreak of disseminated cells as illustrated by the presence of clusters of proliferating 4T07 cells (Fig. 4h). We observed such metastatic outbreaks in the lungs of 4 out of 8 mice that were treated with anti-IFNγ and anti-TNFα. In the lungs of isotype-treated mice, we found no metastatic outbreaks; instead, we observed single, non-proliferating disseminated cells in the lungs of all mice analyzed. Thus, the mere presence of multicellular metastatic 4T07 nodules in the lungs after blocking IFNγ and TNFα shows that induction of dormancy depends on these two cytokines.

Thus, 4T07 primary tumors prime systemic CD8+ T cells that induce cell cycle arrest of DCCs by IFNγ and TNFα and thereby prevent overt metastatic disease.

Effector function and exhaustion in CD39+PD-1+CD8+ T cells

To understand why this population controls DCCs, we sorted CD39+PD-1+CD8+ and other CD8+ T cells from 4T07 breast cancer and compared their transcriptome (Fig. 5a and b and Supplementary Fig. 10a). CD39+PD-1+CD8+ cells show overrepresentation of transcripts associated with effector function including Prf1, Fasl, Gzmf, Gzme, Gzmd, Gzmc, Gzmb, Gzmk, and Ifng (Fig. 5c). This was confirmed by gene set enrichment analysis (GSEA)43 showing that the transcriptome of the sorted CD39+PD-1+CD8+ cells shows strong similarities with human tissue-resident CD8+ T cells (Fig. 5d), which were recently shown to correlate with improved survival in triple negative breast cancer19. This is in line with their unique capacity cells to prevent metastatic progression. Furthermore, the fact that dormancy is never observed in 4T1 breast cancer is explained by the low number of such protective T cells in 4T1-bearing hosts (Fig. 3) and not by their transcriptional signature, since CD39+PD-1+CD8+ cells from 4T1 and 4T07 tumors did not show many differences (Supplementary Fig. 10b–d). The transcriptome of CD39+PD-1+CD8+ cells was also enriched in markers of activation and exhaustion such as Vsir (VISTA), Tnfrsf18 (GITR), Tigit, Cd224 (2B4), Tnfrsf9 (4-1BB), Tnfrsf4(OX-40), Icos, Lag3, Ctla4, and Havcr2 (Tim-3). This is consistent with their overexpression of the transcription factor TOX (Fig. 5c) which leads to expression of exhaustion markers and persistence of effector function during chronic antigen stimulation44.

Taken together, CD39+PD-1+CD8+ cells have a unique transcriptomic signature defined by the co-expression of immune checkpoint molecules and effector proteins.

CD39+PD-1+CD8+ T cells correlate with survival in human breast cancer

To investigate the clinical relevance of our preclinical findings, we analyzed primary breast cancer tissues from a cohort of 54 patients, who had a metastatic relapse after primary tumor resection (Supplementary Table 1) for the presence of CD8, CD39, PD-1, and epithelial cells using 5-plex immunofluorescence (Fig. 6a). Our cohort shows the typical survival curves for the individual subtypes, suggesting that it is representative (Supplementary Fig. 11a). Patients with a high density of intra-tumoral CD39+PD-1+CD8+ T cells had a significantly longer disease-free survival after surgery than patients with a low density of such cells (Fig. 6b). The density of extra-tumoral CD39+PD-1+CD8+ T cells did not correlate with disease-free survival (Supplementary Fig. 11b), whereas to density of CD39+PD-1+CD8+ T cells independent of their location did (Supplementary Fig. 11c). In addition, we observed similar data when focusing on luminal A and B patients (Supplementary Fig. 11d and e). The density of intra-tumoral CD39+PD-1+CD8+ T cells was not an independent variable according to multivariate Cox regression analysis (Supplementary Table 1). Importantly, the density of intra-tumoral CD8+ T cells showed no correlation with disease-free survival (Fig. 6c), strongly supporting our preclinical observations, namely that that disseminated cancer cells are controlled by this specific subset of CD39+PD-1+CD8+ T cells.

We discovered that breast cancer can prime a systemic, CD8+ T cell response that mediates metastatic dormancy in the lungs and consequently, protects against clinical metastatic outgrowth. We thus describe an unexpected trait of primary tumors, in addition to their well-known metastasis-promoting capacity25,26,45,46. Using high-dimensional single-cell profiling of primary tumors, we identified the protective CD8+ T cell population as PD-1+ and CD39+ cells, suggesting recent cognate interaction and tumor-specificity20,47. In line with our preclinical data, we found that a high density of intra-tumoral CD39+PD-1+CD8+ T cells significantly correlated with disease-free survival after resection in breast cancer patients. Even in patients with luminal A or B subtypes, where the tumor tissue is generally less infiltrated by CD8+ T cells48, the presence of CD39+PD-1+CD8+ T cells correlated with disease-free survival after primary tumor resection. As in our preclinical model, the heterogeneous population containing all CD8+ T cells did not correlate with disease-free survival, strongly supporting the notion that CD39+PD-1+CD8+ T cells comprise a population of cells that is uniquely equipped to control disseminated cancer cells.

The role of CD39 in cancer is rather complex. On the one hand, there is evidence that CD39 expression marks tumor-specific T-cells and presence of such cells correlates with a better prognosis or immune-mediated control47. On the other hand, CD39 is constitutively expressed by some cancer cells and suppressive cell types in the tumor microenvironment (Tregs and myeloid cells) and catalyzes pro-inflammatory extracellular ATP, resulting in impaired CD8+ T cell-dependent tumor control49,50. The same papers described a reduction in tumor-associated macrophages and monocytes as well as a better T cell function (and tumor control) after blocking the enzymatic activity of CD39.

We observed that CD39+PD-1+CD8+ T cells express effector molecules but also markers that are typically associated with exhaustion. A similar population was recently described in human breast cancer19,51,52,53. Exhausted T cells are thought to have a reduced effector function and loss of proliferative capacity54,55. Nonetheless, exhausted T cells in breast cancer may be more functional than in melanoma56.

In addition to local immunity, the induction of systemic immune responses by immunotherapy was shown to be essential for its efficacy57, suggesting that protective CD8+T cells are recruited to the tumor. Recirculating CD8+ T cells can also enter tissues, where they develop into CD103+ tissue-resident memory cells and may protect tissues against disseminating tumor cells by the induction of cell death or cell cycle arrest58. Indeed, we detected CD39+PD-1+CD8+ T cells in the primary tumor and in the lungs and found that a proportion expressed CD103 as a marker of tissue residency. It was recently shown that the presence of intra-tumoral CD103-expressing CD8+ T cells correlate with survival in human breast cancer41.

As mentioned above, metastatic dormancy may result from two different states59: Disseminated single cells that are quiescent60, or micro-metastases that remain stable through an equilibrium between proliferation and death61. We found that disseminated 4T07 cells were present in the lungs as single non-cycling cells, suggesting dormancy through cell cycle arrest. Our observation that dormancy essentially depended on CD8+ T cells uncovers a novel mechanism underlying dormancy. We found that protective CD39+PD-1+CD8+ T cells produced TNFα and IFNγ, which were described to induce irreversible senescence in the context of Th1-mediated protection against cancer42; the same cytokines prevented proliferation of disseminated 4T07 breast cancer cells in the lungs.

It may be that environmental cues induce cancer stem cell-like properties in disseminated cancer cells62, which is in line with the low cycling frequency. Such cancer stem cells would represent a potent reservoir for progressive metastasis63 in response to environmental changes. Why dormant cancer cells sometimes awaken only after decades3 is still enigmatic, and it cannot be excluded that different pathways converge to initiate growth of dormant lesions. Assuming CD8+ T cells are required for maintenance of dormancy it is conceivable that attrition of tissue-resident memory CD8+ T cells may result in awakening of dormant cells. Tumor resection and consequent loss of antigen may result in diminished immunological memory; tissue-resident memory cells are described to be particularly unstable in the absence of antigen64,65. Alternatively, dormant cells themselves may change, for example losing the expression of molecules preventing immune-mediated clearance66. Finally, environmental factors in the broadest sense may awaken dormant cancer cells. Such changes in the microenvironment include fibrosis67, tissue remodeling68, obesity69, inflammation70, disturbance of vascular homeostasis71, glucocorticoids72, or cigarette smoke73. How such factors induce cycling of dormant cells and whether immune cells can interfere at this stage is unknown.

Here we discovered that CD39+PD-1+CD8+ T cells mediate dormancy of disseminated cancer cells but are by themselves unable to fully eradicate all of the cancer cells. This is clinically highly relevant, as dormant cancer cells are in all likelihood the source of future metastatic disease. Because the factors determining awakening are not known and may even not be controllable, it is important to better understand the nature of dormant cancer cells and which pathways must be mobilized for their complete elimination.