LSD1 ablation promotes mammary tumor metastasis by attenuating NK cell-mediated anti-tumor immunity.

Introduction
Most cancer-related deaths are caused by metastasis 1, 2. Large-scale sequencing studies have identified the majority of driver mutations responsible for the transformation of normal cells into cancer cells 3. However, similar efforts to identify genetic mutations that specifically drive metastasis have been largely unsuccessful 3, 4, leading to the notion that metastasis may not be driven by distinct genetic alterations 3. Genetic mutations predicted to be drivers in metastatic lesions are typically already present in the corresponding primary tumors, although their frequencies may be higher in metastases 2, 5–7. The lack of metastasis-specific genetic drivers has therefore led to the hypothesis that epigenetic mechanisms, through which cancer cells acquire pro-metastatic transcriptional programs, may play a central role in metastatic progression 1, 2.
Epigenetic regulatory machinery also plays an essential role in safeguarding cells from aberrant immune responses, for example by repressing endogenous retroviruses (ERVs) within the genome 8–10. Epigenetic deregulation can unleash ERV expression, which in turn can trigger innate immune responses. Sustained activation of these pathways may result in chronic inflammation and the establishment of an immunosuppressive microenvironment that promotes tumor progression 8–10. However, how immune mechanisms triggered by epigenetic dysregulation contribute to metastatic progression remains largely unclear.
The epigenetic landscape of a cell is shaped by diverse epigenetic regulators. Lysine-specific demethylase 1 (LSD1, also known as KDM1A) is a key histone demethylase that can function as a transcriptional repressor by removing methyl groups from mono- and di-methylated lysine 4 on histone H3 (H3K4me1/me2) 11–13. LSD1 can also act as a transcriptional activator through demethylation of mono- and di-methylated lysine 9 on histone H3 (H3K9me1/me2) 14–16. In breast cancer, LSD1 has been reported to function either as a tumor suppressor 13, 17, 18 or as an oncogenic factor 19–23. These apparently contradictory roles may reflect the context-dependent functions of LSD1 in different mammary epithelial cell (MEC) populations. Alternatively, they may arise from distinct roles of LSD1 in shaping the tumor microenvironment, particularly the immune microenvironment.
The immune system plays a critical role in dictating the metastatic spread of breast cancer. On one hand, immune effector cells such as CD8+ T cells and NK cells can recognize and eliminate tumor cells, thereby restricting metastatic outgrowth. On the other hand, tumor cells can recruit immunosuppressive cell populations or promote the differentiation of tumor-infiltrating immune cells toward pro-tumorigenic states that facilitate metastasis 24. The recent clinical success of immune checkpoint blockade (ICB) therapy has generated strong interest in immunotherapy for the treatment of multiple cancers, including breast cancer 25. However, the response rate to ICB therapy remains relatively low 26, 27. To improve therapeutic efficacy, ongoing efforts focus on combining ICB with other treatment modalities 25, including epigenetic therapies 28, 29. This strategy is based on the observation that epigenetic drugs, such as inhibitors of DNA methyltransferases (DNMTs) or Histone deacetylases (HDACs), can enhance anti-tumor immune responses by modulating key pathways involved in tumor-immune interactions 28, 29. More recently, inhibition of LSD1 has been shown to induce a type I interferon (IFN) response, enhance tumor immunogenicity, stimulate anti-tumor T cell immunity, and improve responses to anti-PD-(L)1 immunotherapy 30.
Despite these promising findings, LSD1 disruption has also been linked to increased breast cancer metastasis in several studies 13, 18, 31. How LSD1 inhibition-based epigenetic therapy influences metastatic progression therefore remains unclear. In this study, we demonstrate that disruption of LSD1 in mammary tumor cells promotes lung metastasis, at least in part through attenuation of NK cell-mediated anti-tumor immunity.

Results

Induced loss of LSD1 in PyMT tumor cells increases metastasis
We reported previously that in luminal breast cancer, LSD1 largely functions as a tumor/metastasis suppressor, in part by controlling MEC-intrinsic transcriptional programs that restrain invasion and migration of luminal MECs 18. Our experimental approach utilized intraductal injection of a Cre-expressing adenovirus driven by the Keratin 8 promoter (Ad-K8-Cre) 32, 33 into mammary glands (MGs) of the autochthonous luminal breast cancer mouse model MMTV-PyMT (hereafter referred to as PyMT) 34. The PyMT mice also carried Lsd1 conditional knockout (KO) alleles (Lsd1L/L) together with a conditional Rosa26-LSL-YFP (R26Y) Cre-reporter 35. Intraductal injection of Ad-K8-Cre resulted in inactivation of Lsd1 while simultaneously activating YFP expression in the same PyMT tumor cells (Figure 1A). Using this approach, we observed significant increases in both the number and size of lung metastatic lesions in PyMT mice with induced loss of LSD1 compared with PyMT mice retaining wild-type (wt) LSD1 (Figure 1B).
Since intraductal injection of Ad-K8-Cre induced Lsd1 deletion in only a subset of PyMT tumor cells, this approach generated a mosaic system in the injected MMTV-PyMT;Lsd1L/L;R26Y females. In this system, YFP+ tumor cells lack LSD1 expression, whereas YFP− tumor cells retain LSD1 expression. In contrast, in similarly injected MMTV-PyMT;R26Y control females, both YFP+ and YFP− tumor cells remain positive for LSD1 expression. We therefore analyzed YFP+ cells in tumors from both experimental and control PyMT females. In primary mammary tumors, the two groups showed similar percentages of CD45−YFP+ tumor cells (Figure 1C–D and Figure S1A). In peripheral blood, however, the proportion of circulating CD45−YFP+ tumor cells in PyMT mice with induced LSD1-loss was ~3.5-fold higher than in control PyMT females with wt LSD1 (Figure 1E and Figure S1B). In lung metastatic lesions, those from injected MMTV-PyMT;Lsd1L/L;R26Y experimental females contained on average ~2.7-fold higher percentages of YFP+ cells than lesions from control females (Figure 1F). Together, these data indicate that although Lsd1-null PyMT tumor cells do not exhibit a pronounced clonal growth advantage in primary mammary tumors, they possess a significant advantage in seeding the lung and forming metastatic lesions.

LSD1-loss in PyMT tumor cells alters immune-related gene expression
The increased metastatic capacity of Lsd1-null PyMT tumor cells may arise from both cell-intrinsic and extrinsic mechanisms 18. To gain further insights into the enhanced metastatic potential of Lsd1-null PyMT tumor cells, we sorted YFP+ cells from primary tumors arising in the MGs of MMTV-PyMT;Lsd1L/L;R26Y females (i.e., YFP+
Lsd1-null PyMT tumor cells) and compared them with YFP+
Lsd1-wt PyMT tumor cells sorted from MMTV-PyMT;R26Y control females ~2–3 months after intraductal injection of Ad-K8-Cre. These cells were subjected to RNA-sequencing (RNA-seq).
Strikingly, gene set enrichment analysis (GSEA) 36 of the resulting expression profiles revealed that many of the top enriched gene sets in Lsd1-null PyMT tumor cells were associated with immune response, including pathways related to cytotoxic T cells and immune checkpoint regulation (Figure 2A and Figure S2A–C). Among immune-related genes, we noted that most major histocompatibility complex class I (MHC-I) genes (both classical and non-classical), as well as genes involved in immune checkpoint pathways (Figure 2B) and IFN signaling (Figure 2C), were upregulated in Lsd1-null PyMT tumor cells. To validate these findings, we utilized YFP+
Lsd1-null and Lsd1-wt PyMT cells sorted from MGs of PyMT mice ~2-weeks after intraductal injection of Ad-K8-Cre and maintained ex vivo as organoids. In parallel, we generated CRISPR-mediated KO of Lsd1 in PyMT tumor organoid cells. In both organoid systems, quantitative RT-PCR (qRT-PCR) confirmed upregulation of immune-related genes in Lsd1-null PyMT organoid cells (Figure 2D and Figure S2D). In addition, we treated PyMT tumor organoids with the LSD1 inhibitor GSK2879552 or HCI-2509 and measured expression levels of several key immune genes by qRT-PCR. We found that pharmacological inhibition of LSD1 in PyMT tumor cells also upregulated expression of these immune genes in them (Figure S2E).
In CRISPR-induced Lsd1-KO PyMT cells, as well as in PyMT tumor organoids treated with the LSD1 inhibitor GSK2879552 or HCI-2509, we also observed upregulation of transcripts derived from several mouse ERVs 37, 38 (Figure S2F–G). This likely reflects derepression of ERV loci following loss of LSD1-mediated transcriptional silencing 30. Together, these findings suggest that the observed changes in immune-related gene expression in Lsd1-null PyMT cells (i) are unlikely to result from contamination of immune cells (e.g., macrophages) within the sorted YFP+ population and (ii) instead largely represent an early cell-intrinsic response in PyMT MECs following LSD1-loss. One possible mechanism is activation of innate immune signaling in response to double-stranded RNA stress caused by ERV derepression, similar to findings in human breast cancer cells with LSD1 ablation 30.
Lastly, to validate the upregulation of MHC-I genes at the protein level, we performed fluorescence-activated cell sorting (FACS) analysis for the MHC-I molecules H2-D and H2-K, as well as the non-classical MHC-I molecule Qa-1 (encoded by H2-T23) (Figure S3A), in PyMT tumors with or without induced LSD1-loss. We observed higher fluorescence intensities for all three MHC-I molecules, along with significant increases in the proportions of H2-D+, H2-K+ and Qa-1+ cells among CD45−YFP+ tumor cells lacking LSD1 (Figure 2E–F).

LSD1-loss in PyMT tumor cells reprograms the immune microenvironment
The profound changes in the expression of immune-related genes in Lsd1-null PyMT tumor cells suggested that these alterations might reshape the tumor immune microenvironment, potentially contributing to the increased metastatic phenotype through MEC-extrinsic mechanisms. To test this possibility, we performed multi-color FACS analysis (Figure S3A) of primary PyMT tumors arising in the MGs, as well as peripheral blood samples from these mice. In primary tumors, we observed a significant increase in the CD45+ leukocyte population in tumors with induced LSD1-loss (Figure 3A). Among CD45+ cells, tumors with LSD1-loss exhibited a significant increase in the CD4+ T cell population, accompanied by reduced frequencies of CD8+ T cells, NK cells, and macrophages (Figure 3B and Figure S3B–C). Since PyMT tumors with LSD1-loss contained a larger overall CD45+ leukocyte compartment, we also calculated the proportions of individual immune populations among total live cells within the tumors. This analysis similarly revealed increased CD4+ T cells and decreased NK cells and macrophages in tumors with induced LSD1-loss (Figure 3B). Together, these results indicate that induced loss of LSD1 in PyMT tumor cells leads to substantial reprogramming of the tumor immune microenvironment.
The strong upregulation of MHC-I genes in Lsd1-null PyMT tumor cells (Figure 2B and D), together with the reduced NK cell population observed in primary PyMT tumors with induced LSD1-loss (Figure 3B and Figure S3B–C), raised the possibility that NK cells may contribute to the differences in lung metastatic burden between PyMT mice with or without LSD1-loss. We therefore examined NK cells in greater detail. NK cell maturation in primary tumors was assessed by FACS staining for CD11b and CD27, markers commonly used to define NK cell maturation stages [NK cell maturation is determined as: CD11blow CD27low (most immature) -> CD11blow CD27high -> CD11bhigh CD27high -> CD11bhigh CD27low (most mature) 39]. In addition, we analyzed CD107a, a degranulation marker associated with NK cell activation 40 (Figure S3B).
In PyMT tumors with induced LSD1-loss, we observed not only a reduction in the overall NK cell population but also a significant increase in the immature CD11blow CD27low NK cell subset, as well as a pronounced reduction in the mature CD107a+ NK cell subset (Figure 3C). Importantly, in peripheral blood, the frequency of the NK cells in PyMT mice with induced LSD1-loss was also lower than that in control PyMT females (Figure 3D–E and Figure S1B).
To directly assess the relative contributions of NK cells versus other immune populations (e.g., CD8+ T cells) in restricting lung metastasis of PyMT tumor cells, we transplanted equal numbers of PyMT tumor organoid cells into syngeneic WT, Cd8a−/−, and Il2rg−/− female recipient mice. Of note, Cd8a−/− and Il2rg−/− mice lack CD8+ T cells and NK cells 41, 42, respectively. We found that lungs from Il2rg−/− recipients developed significantly more metastatic lesions than those from either Cd8a−/− or WT recipients (Figure 3F). These results support a critical role for NK cells in suppressing PyMT tumor cell metastasis to the lung. Collectively, our data suggest that induced LSD1-loss in the PyMT model leads to both reduced abundance and impaired maturation of NK cells, which may contribute to the enhanced lung metastasis observed in this setting.

Induction of LSD1-loss in PyMT tumor cells using a CreER/tamoxifen system produces a similar phenotype
Intraductal injection of adenovirus can trigger immune response in the injected MG, although this response typically resolves over time 32. To exclude the possibility that the adenoviral injection procedure itself contributed to the increased lung metastasis phenotype, we generated MMTV-PyMT;K8-CreER;Lsd1L/L;R26Y females and inactivated Lsd1 in the same K8+
PyMT tumor cells through tamoxifen injection, which induces CreER activity and Cre-mediated disruption of Lsd1 together with activation of YFP expression (Figure 4A). Of note, we reported previously that this CreER/tamoxifen-based approach targets K8+ luminal MECs at a frequency comparable to that achieved with the Ad-K8-Cre intraductal injection approach 33.
Using this second genetic strategy, we observed similar increases in both the number and, to a lesser extent, the size of lung metastatic lesions in PyMT mice with induced LSD1-loss (Figure 4B). FACS analysis revealed significantly increased expression of the MHC-I molecules H2-D, H2-K, and Qa-1, in CD45−YFP+
PyMT tumor cells from MGs with induced LSD1-loss (Figure 4C). Within the tumor microenvironment of primary PyMT mammary cancers with induced LSD1-loss, we also observed an increase in the total CD45+ leukocyte population (Figure 4D). Among CD45+ leukocytes, significant decreases were detected in the CD8+ T cell, NK cell, and macrophage subsets (Figure 4E and Figure S4A–B). When normalized to the percentage of total live cells, PyMT tumors with induced LSD1-loss displayed reduced NK cell, CD8+ T cell, and macrophage populations, along with an increased CD4+ T cell population (Figure 4E). Among tumor-associated NK cells, we further observed a significant increase in the immature NK cell subset accompanied by a reduction in the mature NK cell subset (Figure 4F). Consistent with these findings, the frequency of NK cells in the peripheral blood of PyMT mice with tamoxifen-induced loss of LSD1 was also lower than that observed in control PyMT females (Figure 4G). Collectively, these results indicate that it is the induced loss of LSD1 in PyMT tumor cells that drives the observed reprogramming of the tumor immune microenvironment and the enhanced metastatic phenotype in the PyMT model.

LSD1-loss in PyMT tumor cells attenuates NK cell-mediated anti-tumor immunity
NK cells are innate lymphocytes that play a critical role in controlling metastatic dissemination (43 and Figure 3F). NK cell activity is regulated through the integration of signals from activating and inhibitory receptors 44. In addition to their role in antigen presentation to CD8+ T cells, MHC-I molecules can bind inhibitory receptors on NK cells and suppress their activation, thereby preventing the killing of healthy “self” cells 45, 46. The increased MHC-I expression observed in Lsd1-null PyMT tumor cells (Figure 2) may therefore reduce NK cell-mediated cytotoxicity against these cells. To test this possibility, we measured NK cell cytotoxicity toward Lsd1-null versus Lsd1-wt PyMT tumor (organoid) cells in a co-culture system. Consistent with this hypothesis, NK cells exhibited reduced killing of Lsd1-null tumor cells, as indicated by decreased Calcein release 47 (Figure 5A–B and Figure S5).
Human HLA-E encodes a non-classical MHC-I molecule whose functional homolog in mice is Qa-1, encoded by H2-T23
48, 49. HLA-E binds peptides derived from the conserved region of leader sequences of other MHC-I molecules 50, enabling its stable presentation on the cell surface. In the absence of suitable peptides, HLA-E is degraded and fails to reach the cell surface 51. Therefore, surface expression of HLA-E serves as an indicator of intact MHC-I expression and functional antigen processing machinery 52. Importantly, the inhibitory NK receptor CD94/NKG2A recognizes HLA-E/Qa-1, and their interaction suppresses NK cell activation against healthy “self” cells 53, 54. Consequently, stable surface expression of HLA-E/Qa-1 requires sufficient expression of both HLA-E/Qa-1 itself and other MHC-I molecules. In Lsd1-null PyMT tumor cells, both H2-T23 and many other MHC-I genes were expressed at significantly higher levels compared with Lsd1-wt cells (Figure 2), suggesting that Qa-1 might be more abundantly displayed on the surface of Lsd1-null tumor cells and thereby inhibit NK cell activity through the CD94/NKG2A receptor. Consistent with this idea, FACS analysis confirmed elevated Qa-1 surface levels in PyMT tumor cells with LSD1-loss (Figure 2E–F and Figure 4C).
To determine whether increased Qa-1 expression in Lsd1-null PyMT tumor cells was responsible for the attenuated NK cell activities observed in the co-culture (Figure 5B), we disrupted Qa-1 expression in PyMT tumor organoid cells by CRISPR/Cas9-mediated KO of its coding gene, H2-T23. As a negative control, CRISPR editing targeting a neutral region within the mouse Rosa26 locus (gR26) was used. Successful knockout of H2-T23 was confirmed by both FACS analysis (Qa-1 protein levels) and qRT-PCR (Figure 5C–D). When Lsd1-null PyMT tumor cells lacking H2-T23 were co-cultured with NK cells, loss of Qa-1 restored NK cell cytotoxic activity toward these tumor organoid cells (Figure 5E).

Absence of innate immune cells abolishes the difference in metastatic potential between Lsd1-null and Lsd1-wt PyMT tumor cells
To determine whether NK cells contribute to the differences in metastatic potential between Lsd1-null and Lsd1-wt PyMT tumor cells in vivo, we intraductally injected these tumor cells (maintained as organoids) into the MGs of either NOD-SCID;IL2Rγnull (NSG) (JAX) or SCID (CRL) recipient mice (Figure 6A). The key difference between these two recipient mouse strains is that NSG mice lack innate immune cells, including NK cells and macrophages, whereas SCID mice retain NK cells and functional macrophages 55.
Following intraductal injection, PyMT tumor organoid cells formed in situ tumors within the mammary ducts, subsequently invading the surrounding stroma in the MGs of recipient mice (Figure 6B), and progressed to lung metastases within approximately one month (Figure 6C–D). We found that these tumor organoid cells generated similar numbers of lung metastatic lesions in both types of recipients (Figure 6C–D). However, the sizes of the metastatic lesions differed markedly between the two models. In NSG recipients, which lack NK cells, both Lsd1-null and Lsd1-wt PyMT tumor cells formed lung metastatic lesions of comparable size (Figure 6C). In contrast, in SCID recipients, which retain NK cells, metastatic lesions derived from Lsd1-wt tumor cells were significantly smaller than those derived from Lsd1-null tumor cells (Figure 6D).
In addition to NK cells, SCID mice also possess functional macrophages. However, macrophages have been reported to promote, rather than suppress, metastatic progression in the PyMT model 56, 57, in contrast to the anti-metastatic role of NK cells. Moreover, our analysis of PyMT primary tumors with induced LSD1-loss revealed a marked reduction in macrophage abundance compared with tumors retaining wt LSD1 (Figure 3B and Figure 4E). Therefore, although macrophages may still contribute to metastatic progression in the PyMT model, they are less likely to account for the enhanced metastasis observed in PyMT tumor cells with LSD1-loss or for the smaller lung metastases formed by Lsd1-wt tumor cells in SCID recipients. Taken together, these in vivo data support a model in which innate immune cells, particularly NK cells, play an important role in restricting lung metastasis of PyMT tumor cells and contribute to the differential metastatic capacities of Lsd1-null versus Lsd1-wt PyMT tumor cells.

A negative correlation between LSD1 and HLE-A expression is conserved in human luminal breast cancer cells
To extend our observation from the PyMT mouse model to human breast cancer, we analyzed the expression of LSD1 and MHC-I genes in human luminal breast tumors. Across human breast cancer, LSD1 (KDM1A) expression showed the strongest negative correlation with HLA-E expression among MHC-I genes, particularly in Luminal A and HER2-enriched subtypes (Figure S6A). To further validate this relationship, we analyzed the largest publicly available breast cancer cohort, METABRIC 58, 59. Among 1,445 estrogen receptor-positive (ER+) luminal breast tumors in this cohort with microarray data, the top 165 and 143 cases with the highest and lowest expression levels of LSD1 (KDM1A) were classified as LSD1-high and LSD1-low groups, respectively. We found that tumors in the LSD1-low group exhibited significantly higher expression levels of multiple MHC-I genes, most notably the non-classical MHC-I gene HLA-E (Figure 7A). Importantly, GSEA of the expression profiling data revealed that many of the top enriched gene sets in LSD1-low tumors (relative to LSD1-high tumors) were immune-related, including pathways associated with IFN signaling (Figure 7B and Figure S6B). Lastly, in the human ER+ luminal breast cancer cell line MCF7 18, knockdown of LSD1 led to upregulation of several MHC-I genes, with HLA-E showing the most prominent increase (Figure S6C).
Together, our in vivo and in vitro data support a model in which LSD1 ablation in PyMT tumor cells increases their immunogenicity and elevates MHC-I expression, thereby reducing NK cell-mediated anti-tumor immunity (Figure 7C). The resulting attenuation of NK cell-mediated tumor surveillance may contribute to the enhanced lung metastatic growth observed in Lsd1-null PyMT tumors. Notably, the similar upregulation of MHC-I genes, particularly HLA-E, in human luminal breast cancer cells with reduced LSD1 expression (or LSD1 deficiency) suggests that a comparable mechanism may operate in human disease, potentially enabling tumor cells to evade NK cell-mediated elimination and thereby promoting metastasis.

Discussion
In this study, we demonstrate that, consistent with previous reports 30, 60, genetic ablation of LSD1 increases tumor cell immunogenicity. However, when the immune system fails to effectively eliminate these tumor cells, the resulting prolonged immune activation can lead to chronic inflammation and reprogramming of the tumor immune microenvironment. One consequence of this altered immune landscape is increased expression of MHC-I genes in tumor cells lacking LSD1, which attenuates NK cell-mediated anti-tumor immunity (Figure 7C). As NK cells play an essential role in restricting metastatic progression (Figure 3F) 43, these changes in MHC-I expression as well as NK cell abundance, maturation, and activity may contribute to the increased lung metastatic burden observed in PyMT mice following LSD1-loss.
This conclusion is supported by our in vivo transplantation experiment, in which differences in lung metastasis between Lsd1-null and Lsd1-wt PyMT tumor cells were observed in SCID recipients (which retain NK cells) but not in NSG recipients (which lack NK cells). In SCID mice, we observed a significant difference in the size of lung metastatic lesions (Figure 6D), although not in their number. The absence of differences in lesion number may reflect the relatively short duration of the experiment (~1 month), as the rapid growth of primary tumors in the injected MGs required early euthanasia of recipient mice. Of note, the involvement of NK cells in this phenotype does not exclude contributions from other immune cell populations to the increased lung metastasis associated with LSD1 ablation. Previous studies have shown that macrophages play a key role in PyMT tumor metastasis 57, and that CD4+ T cells promote lung metastasis of PyMT tumors by enhancing the pro-tumorigenic properties of macrophages 56. Although we observed an increased CD4+ T cell population in PyMT tumors with LSD1-loss, the macrophage population was profoundly reduced (Figure 3B and Figure 4E), suggesting that CD4+ T cells or macrophages are unlikely to be the primary immune cell populations responsible for the enhanced metastasis observed following LSD1 ablation.
In a previous study using the 4T1 mammary tumor syngeneic transplantation model, cotreatment with anti-PD-1 antibody and the LSD1 inhibitor HCI-2509 was reported to reduce tumor burden. However, closer examination of the data indicates that treatment with HCI-2509 alone resulted in a modest increase in lung metastasis (~30% increase in lesion size) 60. In a preliminary study, we similarly treated PyMT female mice with HCI-2509 for ~2 months (beginning at ~4 weeks of age, comparable to the timing of induced Lsd1 deletion in our genetic models) and observed significant increases in both the numbers and sizes of lung metastases. In these inhibitor-treated mice, analysis of PyMT tumors in the MGs also revealed increased numbers of Qa-1+, H2-K+ and H2-D+ tumor cells, accompanied by a reduction in the overall NK cell population and an increase in the immature NK cell subset among CD45+ leukocytes. Together, these observations suggest that pharmacological inhibition of LSD1, particularly when administered over a prolonged period, may also promote mammary tumor metastasis, at least in part through attenuation of NK cell-mediated anti-tumor immunity.
One important caveat is that, whereas our genetic approaches disrupt LSD1 specifically in PyMT tumor cells, pharmacological inhibition of LSD1 is expected to affect multiple cell types, including immune cells such as NK cells. Of note, recent work has shown that different classes of LSD1 inhibitors exert distinct effects on NK cells: catalytic inhibitors that block the demethylase activity of LSD1 appear to have minimal impact on NK cell function, whereas scaffolding inhibitors that disrupt LSD1-associated epigenetic complexes can impair NK cell cytolytic activity 61. HCI-2509 (also known as SP-2509) is a scaffolding inhibitor of LSD1, and thus the increased metastasis observed following HCI-2509 treatment could partly reflect its direct inhibitory effects on NK cell function. Nevertheless, these findings further support a role for impaired NK cell-mediated anti-tumor immunity in the enhanced metastasis associated with LSD1-loss. This impairment may arise through increased HLA-E(Qa-1)-CD94/NKG2A inhibitory signaling in tumor cells, direct inhibition of NK cell function by certain classes of LSD1 inhibitors, or a combination of both mechanisms. Future studies comparing different classes of LSD1 inhibitors in vivo will be important for disentangling these NK cell-related mechanisms in breast cancer metastasis.
Compared with previous studies 30, 60, which relied on transplantation models typically lasting only one to two months, our work employs an autochthonous breast cancer model in which tumors evolve under physiological conditions for more than two months. This system therefore captures the longer-term consequences of LSD1 disruption during tumor progression. Our findings suggest that although short-term LSD1 inhibition may enhance tumor cell immunogenicity and thereby improve the efficacy of ICB-based therapies, prolonged disruption of LSD1 may instead promote chronic inflammation and potentially increase metastatic risk. We identify reduced NK cell-mediated anti-tumor immunity as a key mechanism contributing to this increased metastatic potential, likely mediated in part through activation of the HLA-E(Qa-1)-CD94/NKG2A inhibitory pathway.
The recent development of therapeutic antibodies targeting NKG2A, which can enhance anti-tumor immunity by unleashing both NK and T cell activity 40, provides an intriguing potential strategy. We propose that combining LSD1 inhibitors, particularly catalytic inhibitors, with anti-NKG2A antibody therapy may both enhance checkpoint inhibitor-based immunotherapy and restore NK cell-mediated tumor surveillance, including NK cell-dependent suppression of metastasis.

Materials and methods

Mouse models
The Lsd1L/L (B6.129-Kdm1atm1.1Sho/J) conditional knockout mouse line was obtained from Dr. Stuart Orkin (Harvard Medical School). Rosa26-LSL-YFP (R26Y) [B6.129X1-Gt(ROSA)26Sortm1(EYFP)Cos/J, Strain #: 006148] reporter mice, MMTV-PyMT [FVB/N-Tg(MMTV-PyMT)634Mul/J, Strain #: 002374, or B6.FVB-Tg(MMTV-PyVT)634Mul/LellJ, Strain #: 022974] and K8-CreER [STOCK Tg(Krt8-cre/ERT2)17Blpn/J, Strain #: 017947] transgenic mice, Cd8a−/− [B6.129S2-Cd8atm1Mak/J, Strain #: 002665] and Il2rg−/− [B6.129S4-Il2rgtm1Wjl/J, Strain #: 003174] mice were obtained from The Jackson Laboratory (JAX) (Bar Harbor, ME). For this study, we first bred MMTV-PyMT mice (FVB/N background) with Lsd1L/L;R26Y mice (B6.129 background) and generated a heterozygous strain (MMTV-PyMT;Lsd1L/+;R26Y or Lsd1L/+;R26Y littermates). The heterozygous MMTV-PyMT;Lsd1L/+;R26Y males and Lsd1L/+;R26Y females (mixed FVB/B6.129 background) were set up as breeding pairs to generate the MMTV-PyMT;Lsd1L/L;R26Y experimental and MMTV-PyMT;R26Y control mice for Ad-K8-Cre intraductal injection. For the K8-CreER-based model, we first crossed K8-CreER mice (B6 background) to MMTV-PyMT;Lsd1L/L;R26Y mice (FVB/B6.129 mixed) to generate MMTV-PyMT;K8-CreER;Lsd1L/+;R26Y and K8-CreER;Lsd1L/+;R26Y mice, which were then intercrossed to generate MMTV-PyMT;K8-CreER;Lsd1L/L;R26Y experimental and MMTV-PyMT;K8-CreER;R26Y control mice. Immunodeficient NOD-SCID;IL2Rγnull (NSG, JAX Strain #: 005557) mice and SCID mice were obtained from JAX and Charles River Laboratories (CRL) (Wilmington, MA), respectively. To target luminal MECs in MMTV-PyMT;Lsd1L/L;R26Y or MMTV-PyMT;R26Y female mice, mice were anesthetized and Ad-K8-Cre adenovirus [diluted in injection medium (DMEM supplemented with 0.1% Bromophenol blue and 0.01M CaCl2)] were introduced into mammary ducts of their #4 inguinal glands via intraductal injection 32, 62. For MMTV-PyMT;K8-CreER-based females, to induce CreER activity, tamoxifen (2 mg per mouse, one injection) was introduced into adult mice (8 weeks of age) by intraperitoneal injection 63. The tumor growth was evaluated at every other day; lungs of experimental and control mice were dissected at the experimental end point and fixed in paraffin, followed by histological processing and hematoxylin and eosin (H&E) staining at the Dana-Farber/Harvard Cancer Center (DF/HCC) Rodent Histopathology Core. The whole lung tissue was sectioned sagittally and at least two sections per mouse were evaluated; each experiment (including 1–3 experimental and control mice) was repeated at least eight times (total mice: Ad-K8-Cre: experimental n=20, control n=18; K8-CreER: experimental n=15, control n=10) to evaluate the statistical significance between the experimental and control groups. The blinded evaluation of metastasis was conducted by a pathologist (R.T. Bronson). All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of Brigham and Women’s Hospital (2016N000363, 2020N000122).

Mammary tumor cell preparation
Mammary tumors were dissected and minced, and then incubated in digestion medium (DMEM/F12 with 2% Penicillin/Streptomycin, 0.1 mg ml−1 Gentamicin, 0.6% Nystatin, 2 mg ml−1 Collagenase A, 0.096 mg ml−1 Hyaluronidase) at 37°C with shaking for 1–1.5 hours. After digestion, the cells/tissues were treated sequentially with 0.25% trypsin/EDTA (37°C, 2 minutes), 5 mg ml−1 dispase with DNaseI (0.1mg ml−1, Sigma, St. Louis, MO) (37°C, 5 minutes), cold red blood cell (RBC) lysis buffer (on ice, 2–3 minutes). Between each treatment step, cells/tissues were washed with 1x PBS with 5% FCS. After treatment with the RBC lysis buffer, cells/tissues were filtered through a 40 μm cell strainer and washed with 1x PBS/5% FCS, to obtain single cell suspension 64.

Flow cytometry
Single cell suspensions obtained from MMTV-PyMT tumors with or without LSD1-loss were analysed by flow cytometric (FACS) analysis, evaluating the immune cell populations. The FACS analysis and cell sorting were performed with the BD FACSymphony and the BD FACSAria sorter (BD Biosciences), respectively. Antibodies targeting immune cells used in FACS analysis were purchased from eBiosciences: CD11b-PE-Cyanine7 (clone M1/70, Cat. No. 25–0112-82); CD19-PE (clone eBio1D3 (1D3), Cat. No. 12–0193-81); H2-D-BV786 (clone KH117, Cat. No. 744665), or from BD Biosciences: Qa-1(b)-BV711 (clone 6A8.6F10.1A6, Cat. No. 744389), or from BioLegend: CD45-PerCP/Cyanine5.5 (Clone 30-F11, Cat. No. 103131), Ly-6G-APC/Cy7 (Clone 1A8, Cat. No. 127623), F4/80-Brilliant Violet 605™ (Clone BM8, Cat. No. 123133), NK-1.1-Brilliant Violet 650™ (Clone PK136, Cat. No. 108735), CD3-APC (Clone 17A2, Cat. No. 100235), CD4-FITC (Clone GK1.5, Cat. No. 100405); CD8-APC (Clone 17A2, Cat. No. 100235); CD107a-Alexa Fluor® 700 (Clone 1D4B, Cat. No. 121627); H2-K-PE (Clone AF6–88.5, Cat. No. 116507); CD69 (APC-Cyanine7, clone H1.2F3, Cat. No. 104525), and NKp46 (PE, clone 29A1.4, Cat. No. 137603). Data analyses were performed using FlowJo 10.6.0 (FlowJo LLC ©).

Mammary organoid culture and intraductal injection
The organoid culture was established by seeding YFP+ cells sorted from MGs of MMTV-PyMT;Lsd1L/L;R26Y or MMTV-PyMT;R26Y female mice ~2 weeks after intraductal injection of Ad-K8-Cre and was conducted by following a protocol we reported previously 65. For transplantation, MMTV-PyMT tumor organoid cells (mixed background of C57BL/6 and FVB) with or without LSD-loss (20,000 cells) were injected into each #4 inguinal glands of NSG (JAX) or SCID (CRL) mice at ~5 weeks of age via intraductal injection (n=6) 32, 62. MMTV-PyMT tumor organoid cells (pure C57BL/6 background) (20,000 cells) were injected into each #4 inguinal glands of syngeneic Cd8a−/− (JAX, strain# 002665) (n=5) or Il2rg−/− (JAX, strain# 003174) (n=4) or wildtype (JAX, strain #: 000664) (n=5) female mice at ~8 weeks of age via intraductal injection. At the end point following tumor size evaluation, mice were euthanized and lung tissues from each mouse were fixed, embedded in paraffin; the resulting paraffin sections were processed for H&E-staining at the DF/HCC Rodent Histopathology Core. The blinded evaluation of metastasis was conducted by a pathologist (R.T. Bronson).

NK cell isolation and Calcein-AM cytotoxicity assay
The general NK cell cytotoxicity assay was completed as described 66. Specifically, splenic NK cells from naïve wt mice were isolated according to the manufacturer’s instruction (NK Cell Isolation Kit, mouse; #130–115-818; Miltenyi Biotec) and suspended in complete NK-cell buffer (phenol-red free RPMI 1640 supplemented with 10% FBS, β-mercaptoethanol, non-essential amino acids, L-glutamine, sodium pyruvate, all from Gibco). The purity of the enriched NK cells was further assessed by FACS using antibodies CD45 (PerCP/Cyanine5.5, clone 30-F11, 103131, BioLegend), CD11b (PE-Cyanine7, clone M1/70, 25–0112-82, eBiosciences), CD3 (APC, clone 17A2, 100235, BioLegend), CD69 (APC-Cyanine7, clone H1.2F3, 104525, BioLegend), and NKp46 (PE, clone 29A1.4, 137603, BioLegend). MMTV-PyMT tumor organoid cells with or without LSD-loss were labeled with 10–15 μM Calcein-AM (5016952; FISHER SCIENTIFIC LLC) for 20–30 minutes at 37°C/5% CO2, allowing Calcein-AM to enter organoid cells. The purified NK cells and the target organoid cells were mixed at various ratios (80:1, 40:1, 10:1) in at least three replicates in a round-bottom, 96-well plate, and incubated at 37°C/5% CO2 for 4 hours. The spontaneous (i.e., only target organoid cells in culture medium) and maximum release (only target organoid cells in medium plus 2% Triton X-100 buffer) groups were also included. Calcein release was quantified by harvesting 100 μL of cell-free supernatant to a new plate and measuring fluorescent emission at the appropriate wavelength (excitation filter: 485±9 nm; band-pass filter: 530±9 nm) using a multilabel counter (VICTOR3, Perkin Elmer, San Diego, CA, USA). Cytotoxicity data were expressed as percent release relative to the spontaneous (target cells alone) and maximum release (2% Triton X-100 treated cells) for a particular target with the equation: % of Calcein release = (Test release-spontaneous release)/ (Maximum release-spontaneous release) * 100%.

Immunofluorescence (IF) staining
IF staining was performed on sections from mammary tumors that were fixed in 10% formalin (Fisher Scientific, Hampton, NH) and embedded in paraffin. The IF staining was performed as described 33. Antigen retrieval (Citrate buffer pH 6.0, 20 min boil in microwave oven) was performed before blocking. Primary antibodies used included: anti-GFP (ab290, 1:500 or ab6673, 1:200, Abcam, Cambridge, UK), anti-Keratin 8 (K8) (MMS-162P, 1:200, Covance), The secondary antibodies used were goat anti-mouse IgG conjugated with AF647 (A31571, 1:250), goat anti-rabbit IgG conjugated with AF488 (A11008, 1:250) or with AF594 (A11037, 1:250). Slides were counterstained with DAPI (1 μg ml−1).

RNA isolation and bulk RNA sequencing
Single-cell suspensions prepared from mammary tumors post Ad-K8-Cre injection were obtained as described in the “Mammary tumor cell preparation” procedure. Cell preps were stained with DAPI and DAPI−YFP+ cells were sorted in a FACS Aria II (BD Biosciences). Total RNA was extracted using the RNeasy Micro Kit (QIAGEN, Cat# 74004). Libraries were prepared by the Dana-Farber Cancer Institute (DFCI) Molecular Biology Cores, using the Low Input mRNA Library (Clontech SMARTer) v4, following the vendor’s protocol. Library quality was checked using High Sensitivity D1000 reagents in an Agilent Tape Station. Libraries were pooled and sequenced in an Illumina NextSeq500 (Single-End 75bp reads per sample). A Visualization Pipeline for RNAseq (Viper) analysis tool, developed by the Center for Functional Cancer Epigenetics (CFCE) at DFCI, was used to generate standard outputs.

Quantitative RT-PCR (qRT-PCR)
Organoids were dissociated with TrypLE and pelleted by centrifugation for 5 minutes at 1,000 rpm, followed by RNA extraction with either TRIzol (Thermo Fisher Scientific, 15596026) or the Allprep DNA/RNA mini/micro kits (Qiagen) according to the manufacturer’s protocols. cDNA was generated with an iScript cDNA synthesis kit (Bio-RAD, 170–8891). Quantitative RT-PCR was performed using FastStart SYBR Green Master (Roche, 04913850001, IN).
The primer sequences are listed below for qRT-PCR 37, 38, 67:

CRISPR/Cas9 lentiviral vectors and organoid manipulation
The lentiCRISPR-v2 plasmid was from Addgene (No. 52961, a gift from Dr. Feng Zhang). Two individual guide RNAs (gRNAs) targeting H2-T23 or Kdm1a were designed based on http://crispr.mit.edu (with a high-quality score of above 90) (gH2-T23-1: CTTGTGCTTAGAGATCTGTG; gH2-T23-2: TACTACAATCAGAGTAACGA; gKdm1a-1: TGTAAGGCGCTTCCAGCTGC; gKdm1a-1: TCAACTTCGGCATCTACAAG). LentiGuide vectors constructed with gRNAs targeting the mouse Rosa26 region of mouse (gR26) was applied as the negative control 65. Cloning of gRNAs in lentiCRISPR-v2 plasmid was performed as described 68. All vectors were validated by Sanger sequencing. We produced concentrated lentiviral stocks, pseudo-typed with the VSV-G envelope protein, by transient transfection of their corresponding plasmids into 293T cells as previously described 69.
The procedure for organoid transfection was developed with modifications from the protocol described previously 70. Specifically, following one day of culture in full medium, as described above, tumor cell organoids were infected with lentivirus for 2–3 days before selection was initiated in full culture medium containing puromycin (1 μg ml−1) for õne week. Stably infected and edited organoid cells were expanded in full medium. The NK cell cytotoxic assay toward transfected tumor organoid cells was conducted as described in the section of “NK cell isolation and Calcein-AM cytotoxicity assay”.

Statistics
Student’s t-test (two-tailed) was used to calculate the P values. Data were reported as mean ± SEM. No statistical method was used to pre-determine the sample size for mice. No randomization or blinding was used in the in vivo studies.

Supplementary Material
Supplement 1This article includes six supplementary figures (Figures. S1–S6).