Inducible disruption of Tet genes results in myeloid malignancy, readthrough transcription, and a heterochromatin-to-euchromatin switch

Significance

TET loss of function is strongly linked to cancer. We generated a mouse model of inducible deletion of all three Tet genes (Tet iTKO mice) to study the in vivo effects of loss of TET enzymes. Tet iTKO mice develop rapid and fatal myeloid expansion within 4 to 5 wk. The expanded myeloid cells include new cell populations characterized by upregulation of the Stefin gene cluster on mouse chromosome 16, accompanied by a heterochromatin-to-euchromatin switch encompassing the same region. Stefin/cystatin genes encode cysteine protease inhibitors that inhibit cathepsin proteases among others. We show that high stefin/cystatin gene expression correlates with poor clinical outcomes in acute myeloid leukemia patients and discuss the connection of TET deficiency with changes in genome organization.

Abstract

The three mammalian TET dioxygenases oxidize the methyl group of 5-methylcytosine in DNA, and the oxidized methylcytosines are essential intermediates in all known pathways of DNA demethylation. To define the in vivo consequences of complete TET deficiency, we inducibly deleted all three Tet genes in the mouse genome. Tet1/2/3-inducible TKO (iTKO) mice succumbed to acute myeloid leukemia (AML) by 4 to 5 wk. Single-cell RNA sequencing of Tet iTKO bone marrow cells revealed the appearance of new myeloid cell populations characterized by a striking increase in expression of all members of the stefin/cystatin gene cluster on mouse chromosome 16. In patients with AML, high stefin/cystatin gene expression correlates with poor clinical outcomes. Increased expression of the clustered stefin/cystatin genes was associated with a heterochromatin-to-euchromatin compartment switch with readthrough transcription downstream of the clustered stefin/cystatin genes as well as other highly expressed genes, but only minor changes in DNA methylation. Our data highlight roles for TET enzymes that are distinct from their established function in DNA demethylation and instead involve increased transcriptional readthrough and changes in three-dimensional genome organization.

The biochemical function of Ten-eleven translocation (TET) methylcytosine dioxygenases is to convert 5-methylcytosine to 5-hydroxymethylcytosine (5hmC) and other oxidized methylcytosines in DNA, culminating in DNA demethylation (1–3). TET enzymes are recruited to active enhancers by transcription factors, as shown for TET2 in several hematopoietic cell types (4–7), and TET2 has been shown to bind enhancer regions in mouse embryonic stem cells (8). Moreover, TET2 has been shown to associate with the SET/COMPASS complex, an H3K4 methyltransferase complex that travels with RNA polymerase II (9). In consequence, 5hmC is primarily found in highly transcribed regions in euchromatin, including the gene bodies of highly transcribed genes and the most active enhancers marked by high levels of H3K27 acetylation (H3K27Ac) (10).

TET enzymes have repeatedly been shown to function as tumor suppressors. TET2 loss-of-function mutations are frequent in diverse human hematopoietic malignancies, including myelodysplastic syndrome (MDS), myeloproliferative neoplasms, chronic myelomonocytic leukemia (CMML), diffuse large B cell lymphoma, peripheral T cell lymphoma, angioimmunoblastic T cell lymphoma, and acute myeloid leukemia (AML), among others (2, 11). In mouse models, loss of Tet1 or Tet2 skews the differentiation of hematopoietic stem cells (HSC) toward lymphoid or myeloid commitment and results in hematologic malignancies with features of B lymphoid and myeloid lineages, respectively (12–15). Tet1/2 double-knockout (DKO) mice develop B cell tumors and display a median survival of 20 mo (16), whereas mice with acute, inducible deletion of Tet2 and Tet3 (Tet2/3 DKO) develop myeloid leukemias within 3 to 7 wk (17). We extended these studies to triple deletion of TET genes for two reasons: to ask whether inducible deletion of all three TET proteins in adult mice would also give rise to cancers and whether the complete absence of TET proteins would skew hematopoietic differentiation and leukemogenesis in the B cell and/or myeloid direction.

To assess these and other outcomes, we generated Tet1^f/f Tet2^f/f Tet3^f/f (Tet triple-floxed, Tet Tfl) mice that also harbored UBC-Cre-ERT2, a fusion protein of the Cre recombinase with ERT2, an engineered estrogen receptor that is normally cytoplasmic but moves to the nucleus after tamoxifen injection. The mice also possessed a Rosa26-YFP^LSL allele in which the Enhanced Yellow Fluorescent Protein (EYFP, hereafter referred to as YFP) gene is preceded by a floxed transcriptional STOP cassette; thus, YFP expression serves as a reporter for nuclear translocation and activation of Cre-ERT2. After tamoxifen injection, 100% of injected Tet1/2/3 inducible TKO (iTKO) mice succumbed to AML within 4 to 5 wk. Loss of all TET family enzymes in hematopoietic stem/precursor cells (HSPC) led to the appearance of a new population of myeloid cells bearing neutrophil markers that displayed a striking increase in expression of the stefin/cystatin gene cluster on mouse chromosome 16, coincident with a compartment switch of this region from heterochromatin to euchromatin as judged by Hi-C analysis. Additionally, we document the occurrence of transcriptional readthrough downstream of the stefin/cystatin gene cluster and other highly expressed genes in Tet iTKO cells, consistent with a previous report that readthrough transcription causes heterochromatin-to-euchromatin transitions. Finally, we showed that in patients with AML, there was a clear association of stefin/cystatin gene expression with clinical outcome: high CSTA, CSTB, CST3, and CST7 expression was associated with increased percentages of monocytes and neutrophils in peripheral blood (PB), and genomic amplification or increased expression of CSTB was associated with decreased patient survival.

Results

Inducible Deletion of All Three Tet Genes in Mouse HSPC Results in Myeloid, Not Lymphoid, Leukemia.

To determine whether complete elimination of all TET proteins and activity would skew hematopoietic differentiation and leukemogenesis in the B cell or myeloid direction, we treated Tet Tfl Rosa26-YFP^LSL UBC-Cre-ERT2 mice with tamoxifen for 5 d. The experiments were performed in two separate locations, the United States and South Korea, respectively, with identical results. After tamoxifen injection, there were no notable abnormalities in the hematopoietic systems of control Tet Tfl or Tet Tfl Rosa26-YFP^LSL mice that lacked Cre-ERT2 expression, defined here as wild type (WT) for convenience. In contrast, between 20 and 40 d after the final tamoxifen injection, 100% of injected Tet1/2/3 iTKO mice became very sick and were sacrificed, whereas all WT mice survived (Fig. 1A). Five weeks after tamoxifen treatment, cells from bone marrow (BM) and spleen of the injected mice showed complete loss of all three Tet mRNAs (SI Appendix, Fig. S1A) as well as almost complete loss of 5hmC by anti-5hmC dot blot (18) (SI Appendix, Fig. S1B).

Fig. 1.

Acute deletion of Tet1/2/3 genes results in myeloid leukemia. (A) Left, flowchart of experiments. Mice of the indicated genotypes were injected daily with tamoxifen intraperitoneally, over five consecutive days, and monitored thereafter. Right, Kaplan–Meier curve representing percent survival of WT (n = 23) and Tet iTKO (n = 27) mice over time after tamoxifen injection. (B and C) Analysis of hematopoietic cell populations  following simultaneous deletion of Tet1, Tet2, and Tet3. Flow cytometry was performed to assess myeloid (Gr-1/Mac-1), erythroid (CD71/Ter-119), B (B220/CD19), T (CD4/CD8), and myeloid progenitor (c-Kit/Mac-1) cell populations in the BM (B, n = 9 ~ 13 per each genotype) and spleen (C, n = 11 ~ 13 per each genotype) of WT or Tet iTKO mice. Representative flow cytometry plots (Left) and summary graphs (Right) at 4 ∼ 5 wk after tamoxifen injection with means ± SEM are shown. **P < 0.005, ***P < 0.0005 (Student’s t test).

In BM, spleen, and PB, the percentages of white-blood cells were hugely increased relative to control mice, and the vast majority were Gr-1⁺ Mac-1⁺ myeloid cells (Fig. 1 B and C and SI Appendix, Fig. S1 C and D). The frequencies of c-kit⁺ Mac-1⁺ myeloid progenitors were also substantially increased (Fig. 1 B and C). In all Tet TKO mice, myeloid expansion was accompanied by progressive and massive splenomegaly, enlargement of the liver and lungs, and lymph node hypertrophy (SI Appendix, Fig. S1E); the bones were pale, and the mice developed profound anemia and lymphopenia (SI Appendix, Fig. S1F). Injection of control and Tet iTKO mice with BrdU, a thymidine analog that is incorporated into DNA during replication, followed by organ harvest 16 h later, resulted in a higher fraction of BrdU-positive CD11b⁺ (Mac-1⁺) and granulocyte–macrophage progenitors (GMP) cells in Tet iTKO compared to WT mice (SI Appendix, Fig. S1G), indicating that Tet iTKO myeloid cells proliferate to a greater extent  than WT cells. Since CD11b is one component of the Mac-1 heterodimer, we use the designations Mac-1⁺ and CD11b⁺ interchangeably in the figures and text hereafter. Analysis of cellular populations by flow cytometry showed that the frequencies of Gr-1⁺/ Mac-1⁺ myeloid cells were increased in BM and spleen (Fig. 1 B and C, Top). The BM also showed a decreased frequency of Ter-119⁺ CD71⁺ erythroid precursors (Fig. 1 B, 2nd panel) and CD19⁺ B220⁺ B cells (Fig. 1 B, 3rd panel), while the spleen showed decreased frequencies of CD19⁺ B220⁺ B cells and CD8⁺ T cells (Fig. 1C). SI Appendix, Fig. S1G, The  Mac-1⁺ population in BM and spleen (Fig. 1 B and C, Top) expressed more c-Kit, a hematopoietic progenitor cell marker (Bottom). Overall, these results indicate that Tet iTKO mice develop a primarily myeloid malignancy, thus resembling Tet2/3 iDKO mice in which Tet2/3 deficiency is induced with poly-I:C or tamoxifen (17).

In BM of Tet iTKO mice, Lin^– Sca-1⁺ c-Kit⁺ (LSK) cells, which are enriched for HSCs and multipotent progenitors (MPPs), were decreased in both frequency and number (SI Appendix, Fig. S2 A and B), as were megakaryocyte–erythrocyte progenitors (MEPs) (SI Appendix, Fig. S2 A and B), long-term and short-term HSC (LT- and ST-HSC, SI Appendix, Fig. S2 C and D) and common lymphoid progenitors (CLP) (SI Appendix, Fig. S2 E and F), whereas the frequencies and numbers of GMP, a myeloid-committed progenitor cell, were increased (SI Appendix, Fig. S2 A and B). These features are very similar to those noted previously in Tet2/3 iDKO mice (17). Subdivision of MPPs into MPP2, MPP3, and MPP4 subpopulations (19) showed decreased frequencies and numbers of MPP2 and MPP4 cells, which yield MEPs and lymphoid precursors, respectively (SI Appendix, Fig. S2 C–F), leading to a relative increase in the frequency of MPP3 cells, myeloid-biased MPPs similar to pregranulocyte/macrophage (pre-GM) progenitors, within the LSK population (19). These results are consistent with our previous findings that genes with a pre-GM gene signature are up-regulated in Tet2/3 iDKO LSK cells, compared to WT LSK, whereas those with lymphoid and premegakaryocyte/erythrocyte gene signatures are down-regulated (17). A schematic showing the skewed differentiation is shown in SI Appendix, Fig. S2G. Overall, we conclude that mice with triple TET deletion are distinct from Tet1/2 DKO mice but resemble Tet2/3 DKO mice in displaying primarily myeloid rather than lymphocyte (B or T cell) expansion.

To identify regions of altered chromatin accessibility as well as transcription factors in LSK cells potentially involved in myeloid expansion observed after TET deletion, we performed ATAC-seq on LSK cells from Tet2^f/f Tet3^f/f Mx1-Cre (Tet2/3 Dfl, Tet2/3 iDKO) mice 3 wk after poly-I:C treatment (17) (SI Appendix, Fig. S3A). The 661 regions that showed increased accessibility in Tet2/3 iDKO LSK cells compared to WT (SI Appendix, Fig. S3B) displayed enrichment of consensus binding motifs for CCAAT/enhancer binding protein (C/EBP) and PU.1, two transcription factors that are essential during the period when cells commit to the myeloid lineage from HSPC (20, 21) (SI Appendix, Fig. S3C). These regions corresponded to myeloid enhancers that displayed high levels of H3K27Ac in WT GMP, macrophages (MF), granulocytes (GN), and monocytes (Mono) (SI Appendix, Fig. S3D). In contrast, the 183 regions showing decreased accessibility in Tet2/3 iDKO LSK cells compared to WT corresponded to enhancers with high H3K27Ac in LT-HSC, ST-HSC, MPP, and CLP (SI Appendix, Fig. S3D), consistent with the strong myeloid skewing and decreased lymphoid cell populations observed in Tet2/3 iDKO mice (17).

Considering the similar phenotypes between our previously published (17), Mx1-Cre-driven Tet2/3 iDKO model and the presently described, Cre-ERT2-driven Tet1/2/3 iTKO models, we directly compared differences at the gene expression level between Tet2/3 deletion and Tet1/2/3 triple deletion (SI Appendix, Fig. S4). We lentivirally transduced BM cells from Tet2^f/f Tet3^f/f or Tet1^f/f Tet2^f/f Tet3^f/f animals with Cre recombinase, transferred them to irradiated recipients, and performed total (ribodepleted) RNA-seq on the expanded CD11b⁺ cells (SI Appendix, Fig. S4A). Consistent with the myeloid expansion that occurs in both cases, the gene expression profiles of Tet2/3 DKO and Tet1/2/3 TKO expanded myeloid cells were also very similar (SI Appendix, Fig. S4 B and C), and the gene expression changes between Tet2/3 DKO and Tet1/2/3 TKO CD11b⁺ cells largely occurred in the same direction relative to WT– Tet1/2/3 TKO and Tet2/3 DKO myeloid cells showed 6,110 and 6,745 differentially expressed genes (DEGs) respectively compared to WT, of which 5,265 genes were similarly differentially expressed in both comparisons (SI Appendix, Fig. S4C). Only 23 DEGs displayed opposite changes in gene expression in Tet1/2/3 TKO and Tet2/3 DKO myeloid cells with respect to WT, e.g., only four DEGS were significantly up-regulated in Tet2/3 DKO and simultaneously down-regulated in Tet1/2/3 TKO compared to WT myeloid cells (SI Appendix, Fig. S4C).

Tet iTKO-Related Myeloid Leukemia Arises from Early Progenitor HSPCs.

To ask what cell populations could transfer the myeloid leukemia early after TET deletion, we isolated LSK, GMP, and Mac-1⁺ cells from mice 2 d after final tamoxifen administration and transferred them to lethally irradiated CD45.1⁺ mice together with WT CD45.1⁺ BM cells (Fig. 2A). At this early time after tamoxifen treatment, the cell populations derived from BM of Tet iTKO mice before transplantation were phenotypically identical to those derived from BM of WT mice (SI Appendix, Fig. S2H). TET-deficient BM or LSK cells transferred leukemia, and the mice succumbed within 60 to 90 d, but TET-deficient GMP and Mac-1⁺ cells failed to induce leukemia and the recipient mice survived beyond 125 d after cell transfer (Fig. 2B). In mice receiving WT LSK cells, CD45.2⁺ donor cells were observed in multiple organs—BM, PB, lymph nodes (LN), and spleen (Fig. 2 C and D), whereas in mice transplanted with Tet iTKO LSK cells, immature GMP-like and more mature Mac-1⁺ cells were dominantly observed in several organs (Fig. 2 C and D). These phenotypes were nearly identical to those of the parental (donor) mice (Fig. 1 and SI Appendix, Fig. S1), indicating that early progenitor HSPCs rather than GMP-like cells are the cell of origin in the myeloid leukemias developing in Tet iTKO mice.

Fig. 2.

Only early HSPCs of Tet iTKO mice can transfer myeloid leukemia. (A) Flowchart of experiments in which whole BM cells or LSK, GMP, or Mac-1⁺ cells isolated from WT and Tet iTKO mice were transferred into irradiated recipient mice. Transfer of Tet iTKO BM or LSK cells induces myeloid malignancy within 100 d, whereas transfer of Tet iTKO GMP or Mac-1⁺ cells does not. (B) Kaplan–Meier survival plot of mice transplanted with WT or Tet iTKO cells; 2 × 10⁶ total BM cells, 1 × 10⁶ Mac-1⁺, 5,000 GMP, or 2,000 LSK cells from control (WT) and Tet iTKO mice were transferred into lethally irradiated CD45.1⁺ recipient mice together with 2 × 10⁶ helper BM cells from CD45.1⁺ mice. All recipient mice given total BM (8/8) or LSK (5/5) cells from Tet iTKO mice, but none of the mice that received GMP or Mac-1⁺ cells from Tet iTKO mice, developed myeloid leukemia. (C and D) Representative flow cytometric analysis of BM HSPCs (C) and myeloid lineage cells (D) in BM, PB, LN, or spleen of recipient mice 7 to 8 wk after adoptive transfer. (E) Flowchart of experiments in which whole BM cells, LSK, or GMP cells from WT or Tet Tfl Rosa26-YFP^LSL mice were lentivirally transduced with Cre before adoptive transfer into irradiated recipient mice. (F) Kaplan–Meier survival plot of mice transplanted with Cre- or control-transduced cells from WT or Tet Tfl Rosa26-YFP^LSL mice. 2 × 10⁶ BM cells, 5,000 GMP, or 2,000 LSK from Tet Tfl Rosa-YFP mice were isolated by FACS and infected with Cre lentivirus and then transferred into lethally irradiated CD45.1⁺ recipient mice together with 2 × 10⁶ helper BM cells from CD45.1⁺ mice. All recipient mice given total BM (11/11) or LSK (9/9) cells, but none of the mice that received GMP cells from Tet iTKO mice or BM, LSK, or GMP cells from WT control cells, developed myeloid leukemia.

Because the Cre-ERT2 system has some drawbacks (for instance, the rapid decline in erythropoiesis), we established a second model system to confirm our observations. Tet Tfl Rosa26-YFP^LSL cells were lentivirally transduced with Cre recombinase, yielding a mixed population of untransduced Tet Tfl and transduced Tet TKO progenitor cells; this is more relevant as a cancer model since only a subset of progenitor cells develop somatic Tet gene deletion. Whole BM from Tet Tfl Rosa26-YFP^LSL mice, or LSK or GMP cells purified from BM of these mice by fluorescence-activated cell sorting (FACS), was infected with a Cre-expressing lentivirus, and the transduced cells were transferred into lethally irradiated CD45.1⁺ mice together with WT CD45.1⁺ BM cells (Fig. 2E). Again, TET-deficient LSK cells, but not GMP cells, were capable of transferring leukemia, and the mice succumbed within 120 d (Fig. 2F).

Cell-Autonomous Leukemia Development in Tet iTKO Mice.

To demonstrate cell-autonomous leukemia development in Tet iTKO mice, we isolated red blood cell-depleted leukemic splenocytes from WT or Tet iTKO mice at 4 wk following tamoxifen administration and transferred them to sublethally irradiated CD45.1⁺ recipient mice, respectively (SI Appendix, Fig. S5). None of the recipient mice transplanted with Tet iTKO splenocytes survived longer than 25 d after transplantation (SI Appendix, Fig. S5A). Similar to acute TET deletion in primary mice, mice receiving Tet TKO cells showed leukocytosis and enlarged spleen and liver (SI Appendix, Fig. S5 B and C). The leukocytosis was associated with significant expansion of myeloid lineage cells and with severe anemia (SI Appendix, Fig. S5 D and E). Furthermore, mice receiving Tet iTKO cells showed expansion of Mac-1⁺ cells in the BM, spleen, and PB, whereas cells of lymphoid and erythroid lineages were reduced (SI Appendix, Fig. S5 F–H). Again, Tet deficiency led to a substantial increase in the proportion of cells expressing the progenitor marker, c-Kit (SI Appendix, Fig. S5 F and G, Bottom) in BM and spleen, and to an increase in the percentage of GMP cells within Lin⁻c-Kit⁺ (LK) cells (SI Appendix, Fig. S5I). Similar results were obtained when leukemic BM cells from WT or Tet iTKO mice were transferred into recipient mice (SI Appendix, Fig. S6). Together, these results show that inducible deletion of all three TET genes results in the rapid, cell-autonomous development of leukemia in Tet iTKO mice.

Increased Expression of the Stefin/Cystatin Gene Cluster in TET iTKO BM Cells.

We used single-cell RNA-sequencing (scRNA-seq) to determine whether cell populations in BM were altered in Tet iTKO compared to WT mice (Fig. 3 A and B and SI Appendix, Fig. S7A). We profiled a total of 40,151 single cells from total BM of two WT and two Tet iTKO mice using the 10× genomics platform (SI Appendix, Fig. S7A and Methods) and partitioned them into 24 different cell populations (Fig. 3 A and B) based on lineage markers known to be highly expressed in specific hematopoietic cell populations (22) (Fig. 3B and SI Appendix, Fig. S7 B and C). For instance, the erythroid Ery 1 population was defined based on expression of the known erythroid genes Epor, Klf1, Car1, and Car2 (carbonic anhydrase 1 and 2) and Hba-a2 (hemoglobin A) (Fig. 3B and SI Appendix, Fig. S7 B and C). In agreement with the data from flow cytometry (Fig. 1B), Tet iTKO BM showed a striking decrease in erythroid (Ery1 and Ery2), lymphoid (lymph, B cell 1, and B cell 2), and certain myeloid (cDC) populations (SI Appendix, Fig. S7C).

Fig. 3.

Appearance of new myeloid cell populations in Tet iTKO BM. (A) Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction for single-cell RNA-seq data, displaying the different identified populations (Left) and the new cell populations present in Tet iTKO BM (Right). Myel. Prog., myeloid progenitor; Neu, neutrophil; Mono, monocyte; Macro, macrophage; Mast, mast cell; Eosin, eosinophil; cDC, conventional dendritic cell; pDC, plasmacytoid dendritic cell; Lymph, lymphoid progenitor; Ery, erythroid progenitor. (B) Expression of key marker genes across the different cell populations identified in (A). The dot size encodes the percentage of cells within each cell population that show expression of the indicated gene, and the color intensity encodes the average expression level of the indicated gene across all cells within each cell population. (C) Volcano plot comparing gene expression in cell populations Neu1 to Neu3 (specific to Tet iTKO) compared to Neu4 to Neu7, using single-cell RNA-seq data. (D) Expression levels of stefin/cystatin genes Stfa1, Stfa2l1, and Cstdc4 in WT and Tet iTKO cells, across different myeloid progenitor and neutrophil cell populations. (E) Mean average plots of pairwise comparison of gene expression determined by polyA⁺ RNA-seq in Tet iTKO vs Tet Tfl CD11b⁺ cells. P-values were calculated using the Wald test (as implemented in DESeq2), and adjusted using the Benjamini–Hochberg method. DEGs (adjusted P < 0.05, fold change (log₂ scale) ≥1 or ≤−1) are highlighted in orange (Tet iTKO > WT) or blue (WT > Tet iTKO). Differentially expressed Stefin/Cystatin genes found on mouse chromosome 16 are labeled.

Notably, in Tet iTKO BM cells, we observed three new populations of myeloid cells largely absent in WT cells, which we labeled Neu1, Neu2, and Neu3, based on the fact that they expressed at least nine neutrophil markers (Fig. 3 A and B). We also identified five additional cell populations, Neu4-8, in WT cells expressing the same nine neutrophil markers. We then directly compared gene expression in these two sets of neutrophil populations, the novel cell subpopulations Neu1 to Neu3 present only in Tet iTKO BM cells and the subpopulations Neu4 to Neu7 represented predominantly in WT cells. Of nine genes strongly up-regulated (log₂ fold change > 1) in the novel Neu1 to Neu3 myeloid populations observed upon Tet triple deletion, eight were members of the stefin/cystatin gene cluster located on mouse chromosome 16 (Fig. 3C and SI Appendix, Figs. S7D and S8A), a family of cysteine protease inhibitors whose expression is often dysregulated in human cancers (23–25). The new myeloid populations Neu1 to Neu3 showed the highest levels of expression of the stefin/cystatin gene cluster on mouse chromosome 16 (SI Appendix, Fig. S7D); however, upregulation of these genes was not limited to the Neu1 to Neu3 populations since we also observed increased expression of these genes in Tet iTKO cells compared to WT cells in other myeloid populations (Fig. 3D). We confirmed this observation arising from scRNA-seq data analysis by performing bulk polyA⁺ RNA-sequencing of expanded CD11b⁺ cells from Tet iTKO and WT mice, isolated ~4 wk after tamoxifen injection, which revealed that among the 1,058 differentially up-regulated genes were all 10 members of the stefin/cystatin gene cluster on mouse chromosome 16 (Fig. 3E); similar results were observed in bulk total (ribodepleted) RNA-seq from Cre-transduced Tet1^f/f Tet2^f/f Tet3^f/f cells (SI Appendix, Fig. S4F). In the mouse genome, in addition to the chromosome 16 cluster mentioned above, there is a second stefin/cystatin gene cluster on mouse chromosome 2 (SI Appendix, Fig. S8B) and two isolated stefin/cystatin genes, Cstb and Cst6, located on chromosomes 10 and 19, respectively. Of these additional genes, we were able to detect expression of only Cst3, Cstb, and Cst7, and Cst7 showed perceptible upregulation only in the polyA⁺ RNA-seq dataset (Fig. 3 E and SI Appendix, Fig. S4F).

Stefin/cystatin genes were previously reported to be up-regulated in other mouse models of aggressive AMLs (26). To define the stage of differentiation at which the stefin/cystatin genes were up-regulated, we analyzed RNA-seq data from LSK cells isolated from WT and Tet iDKO mice (17) 2.5 wk after Tet2/3 deletion (SI Appendix, Fig. S7E). Three genes in the stefin/cystatin gene cluster—Stfa1, Cstdc5, and Cstdc6—were significantly up-regulated even in Tet iDKO LSK cells, which are enriched in HSPC, indicating that the change in expression of this gene cluster was initiated at the early precursor stage in HSPC, prior to myeloid differentiation.

Role of the Stefin/Cystatin Gene Cluster in Human AML.

Stefin/cystatin gene products are cysteine protease inhibitors whose functions have been primarily studied in the context of inhibition of the cathepsin family of cysteine proteases, which have critical roles in protein turnover within the lysosome as well as in other intracellular and extracellular locations (27). Stefins/cystatins have been categorized into two classes based on their structure (25): Type I cystatins (also known as Stefins) are unglycosylated, primarily intracellular proteins that are ~100 amino acids in length and lack disulfide bonds, whereas type II cystatins are extracellular inhibitors that are ~120 amino acids in length and have two intrachain disulfide bonds. SI Appendix, Fig. S8C shows a genome browser view and table of the human cystatin genes located in a cluster on human chromosome 20 and also lists three isolated cystatin genes, CSTA, CSTB, and CST6 located on chromosomes 3, 21, and 11, respectively. New members of the stefin/cystatin gene family are still emerging with each new genome annotation, and in many cases, the subcellular localizations and functions of the newly annotated gene products remain obscure.

To explore whether stefin/cystatin expression could play a role in human myeloid malignancies, we examined the association between stefin/cystatin mRNA expression and the clinical features observed in patients with AML (28). There was a positive correlation of CSTA mRNA expression with the percentage of circulating monocytes and neutrophils present in the PB of AML patients (Fig. 4 A and B). A similar positive correlation was observed between CSTB and CST3 expression and the percentage of circulating monocytes (Fig. 4A) and between CST7 and neutrophil percentage (Fig. 4B). In a different AML dataset (29), we observed that CSTB genomic gain/amplification was associated with shorter overall survival (Fig. 4C, median survival of 4.96 mo in patients with CSTB amplification, vs. 12 mo in patients with two copies of the CSTB locus). Additionally, when stratifying patients by CSTB expression (1st quartile vs. 4th quantile), high CSTB expression (4th quartile) was associated with shorter survival (median survival of 7.96 mo, compared to 16.08 mo in the 1st quartile) (Fig. 4D).

Fig. 4.

Association between cystatin expression and clinical features in human AML. (A) Spearman correlation between CSTA, CSTB, and CST3 gene expression with percentage of monocytes in peripheral blood (PB) in AML patients. Each dot represents a different patient. (B) Spearman correlation between CSTA and CST7 gene expression with percentage of neutrophils in PB in AML patients. Each dot represents a different patient. (C) Kaplan–Meier survival curves of human AML patients, stratifying by CSTB genomic gain/amplification. P values were calculated using the log-rank test. (D) Kaplan–Meier survival curves of human AML patients, stratifying by highest quartile vs. lowest quartile in terms of CSTB gene expression. P values were calculated using the log-rank test.

Given these data, we attempted a functional analysis of the role of stefin/cystatin genes in Tet iTKO cells. We generated Tet Tfl Rosa26-YFP^LSL Cas9-IRES-GFP Cre-ERT2 mice, treated them with tamoxifen for 5 d, isolated LK cells 2 d later, transduced them with a mixture of lentiviral vectors encoding different sgRNAs targeting individual Stefin genes in the chromosome 16 cluster as well as several positive and negative control sgRNAs, and transferred the transduced cells to lethally irradiated recipient mice. Approximately 7 wk after transfer, we extracted DNA from BM cells and amplified and sequenced the region containing the different sgRNAs (SI Appendix, Fig. S7F). By measuring sgRNA abundance in expanded Tet iTKO cells, we found that sgRNAs targeting maintenance DNA methyltransferase Dnmt1 dropped in abundance (SI Appendix, Fig. S7 G and H). However, gRNAs targeting individual Stefin genes were not significantly enriched or depleted relative to nontargeting sgRNAs (SI Appendix, Fig. S7G), most likely due to redundancy among genes in this family (30). Because of this  redundancy, it may be challenging to determine whether stefin/cystatin depletion affects oncogenic progression in mouse or human systems.

The Increased Expression of Stefin/Cystatin Genes Is Not Related to Changes in DNA Methylation.

We previously showed that Tet gene disruption was associated with a striking genome-wide alteration of DNA methylation patterns in all cell types analyzed (31). Briefly, Tet-deficient cells showed focal increases of DNA methylation in euchromatin (Hi-C A compartment), coupled with widespread losses of DNA methylation in heterochromatin (Hi-C B compartment) (31). To ask whether increased expression of the stefin/cystatin gene cluster correlated with changes in DNA CpG methylation and HiC compartments, we performed whole-genome bisulfite sequencing (WGBS) and Hi-C on WT and Tet TKO CD11b⁺ cells (SI Appendix, Fig. S4A and Methods).

From the WGBS analysis, we identified 85,635 high-confidence regions that were differentially methylated (DMRs) genome-wide. Of these, 68,927 DMRs were hypermethylated, with a median size of 1,182 bp, and 16,709 DMRs were classified as hypomethylated, with a median size of 2,400 bp, consistent with the focal increases and widespread decreases in DNA methylation that we previously reported for TET-deficient cells (31). DMRs that were hypermethylated in Tet TKO (Cre-transduced Tet Tfl Rosa26-YFP^LSL) compared to WT control (Cre-transduced Rosa26-YFP^LSL) CD11b⁺ cells were mostly intragenic and 20% of them overlapped a promoter region; in contrast, 63% of hypomethylated DMRs in Tet TKO compared to control Cd11b⁺ cells were intergenic and distal to transcription start sites (TSS) (SI Appendix, Fig. S9A). Of the 8,549 genes with hypermethylated DMRs within 1 kb of their TSS, most (6,050; 71%) showed no changes in gene expression, and a similar number showed down-regulated or up-regulated expression compared to WT (~1,250; ~15% in each case) (SI Appendix, Fig. S9B). Moreover, there were only very minor changes in DNA methylation around the stefin/cystatin gene cluster in Tet TKO cells (Fig. 5A); data for a few hypermethylated DMRs surrounding Stfa1, including one directly upstream of the Stfa gene, are shown in Fig. 5A and SI Appendix, Fig. S9C). Thus, as we have noted previously for other models of TET deficiency (17, 32), there is not a straightforward relation between gene expression and DNA methylation at gene promoters (SI Appendix, Fig. S9B).

Fig. 5.

Compartment switching from heterochromatin to euchromatin in Tet iTKO cells is strongly apparent at the stefin/cystatin gene cluster. (A) Top, Genome browser view of the stefin/cystatin gene cluster comparing WT and Tet iTKO CD11b⁺ cells. RNA expression profiles (tracks 1-4), Hi-C PC1 values (track 5-6), hypermethylated (track 7) and hypomethylated (track 8) differentially methylated regions, and DNA methylation changes (track 9) are shown. Bottom, Zoomed-in view of the Stfa1 gene, comparing gene expression and DNA methylation in control (WT) and Tet iTKO Cd11b⁺ cells. RNA expression profiles (tracks 1-4), differentially hypermethylated regions (DMR #1-3, track 5), and DNA methylation differences (%Tet TKO-%WT) (track 6) are shown. (B) Scatterplot of pairwise comparison of Hi-C PC1 values between WT (x-axis) and Tet iTKO (y-axis) CD11b⁺ cells; color scale indicates the correlation of the Hi-C interaction profile between both conditions (lower correlation values indicate that the same locus interacts with different regions in WT and Tet iTKO cells). Highlighted are the windows overlapping the stefin/cystatin gene cluster (chr16: 36,150,000 to 36,300,000). (C) Relation between RNA expression changes (log₂ fold change) determined by total (ribodepleted) RNA-seq and euchromatin/heterochromatin compartment changes in WT vs Tet iTKO CD11b⁺ cells. Only the genes that exhibited compartment switching (from B to A on the right and from A to B on the left) and significant upregulation (red and black) or downregulation (blue) are plotted. The significant gene expression change was defined as greater than twofold change with a P-value < 0.05. Red dots correspond to the Stefin/cystatin cluster genes on chromosome 16.

The Stefin/Cystatin Gene Cluster Undergoes a Compartment Switch from Heterochromatin to Euchromatin in Tet iTKO Cells.

From the Hi-C analysis (at 50kb resolution), we found that the whole region of 300 kb encompassing the stefin/cystatin gene cluster underwent a change from the Hi-C B compartment in WT (Tet Tfl CD11b⁺) cells to the Hi-C A compartment in Tet iTKO CD11b⁺ cells (Fig. 5 A and B). Triple Tet gene deletion resulted in a heterochromatin-to-euchromatin switch of this region (Fig. 5A, tracks 5 and 6), associated with a striking upregulation of the expression of all genes in the stefin/cystatin gene cluster (Fig. 5A, tracks 1-4; Fig. 5C). This compartment switch was observed only in the stefin/cystatin gene cluster and few other genomic regions, however, since at the whole-genome level, the vast majority (>98%) of 50kb windows remained in the same compartment (heterochromatin or euchromatin) following Tet gene deletion (Fig. 5B and SI Appendix, Fig. S10A); only a few (641/43,844; ~1.5%) underwent compartment switching in Tet-deficient compared to Tet Tfl (WT) CD11b⁺ cells. Of these, most windows (537/641; 83.8%) were located in heterochromatin in Tet Tfl (WT) cells and moved into euchromatin upon Tet gene deletion (SI Appendix, Fig. S10A).

As expected from our previous analysis (31), the majority of genomic regions (50kb windows) that remained in euchromatin after Tet gene deletion (A-to-A) gained DNA methylation on average (SI Appendix, Fig. S10B, purple violin plots; SI Appendix, Fig. S10C); this is expected since TET proteins and 5hmC are predominantly localized to euchromatin where they mediate DNA demethylation. In contrast, consistent with our previous findings (31), the majority of genomic regions that remained in heterochromatin after Tet gene deletion (B-to-B) showed a decrease in average DNA methylation (SI Appendix, Fig. S10B, blue violin plots; SI Appendix, Fig. S10C). At a global level, DNA methylation did not appear strikingly altered in the 50kb windows that changed genomic compartments (SI Appendix, Fig. S10B, green and orange violin plots; SI Appendix, Fig. S10C); however, at the level of individual CpGs, DNA methylation was increased at the majority of euchromatic CpGs in Tet iTKO cells, regardless of whether these CpGs were in euchromatin or heterochromatin in the parental Tet Tfl cells (see Stefin locus in Fig. 5A).

Tet iTKO Cells Display ReadThrough Transcription in Highly Expressed Genes.

In human monocyte-derived macrophages infected with influenza A, transcription into heterochromatin, particularly readthrough transcription past gene ends, was shown to be causal for compartment switching from heterochromatin to euchromatin (33). Along the same lines, by analyzing either total transcripts or nascent (unspliced) transcripts in total (ribodepleted) RNA-seq data, we observed RNA signal downstream of stefin genes throughout the stefin/cystatin cluster on chromosome 16 in Tet TKO but not in control myeloid cells, and the RNA signal extended for kilobases past the gene ends (Fig. 6 A and B). Thus, the heterochromatin-to-euchromatin transition of the stefin/cystatin gene cluster in Tet TKO cells was accompanied by increased intronic and readthrough transcription and coordinated upregulation of stefin/cystatin genes.

Fig. 6.

Tet iTKO cells display intronic and readthrough transcription downstream of highly expressed genes. (A) Genome browser tracks of total (ribodepleted) RNA-seq (tracks 1– 4) and Hi-C PC1 compartment changes (tracks 6-7) at the stefin/cystatin gene cluster in WT and Tet iTKO CD11b⁺ cells. A region containing the Stfa1 gene is highlighted and shown in (B). (B) Total RNA levels in the Stfa1 locus in WT and Tet iTKO CD11b⁺ cells, at two different scales (Left: 0 to 5 RPM (reads per million of mapped reads); Right: 0 to 0.5 RPM). (C) Genome tracks of total RNA levels in WT and Tet iTKO CD11b⁺ cells at the Gpx1 locus. RNA levels within the gene body and downstream of the gene end are indicated in the figure. These regions are used to quantify transcriptional readhrough, following a previously published approach (34). Gpx1 is not differentially expressed between WT and Tet iTKO cells and is among the top 100 most highly expressed genes in both conditions. (D) Distribution of readthrough transcription levels (log₁₀ ratio of reads downstream of gene ends vs. read within gene body) for top 100 (Left) or top 501 to 1000 (Right) most highly expressed genes, across two biological replicates from WT or Tet iTKO CD11b⁺ cells.

The readthrough transcription observed in influenza-infected human macrophages was most apparent in highly transcribed genes (33, 34). Since as noted above, genes in the stefin/cystatin cluster were among the most differentially up-regulated genes in Tet TKO myeloid cells (Fig. 3 C–E and SI Appendix, Fig. S7D), we asked whether the readthrough transcription observed in the stefin/cystatin gene cluster might be a general phenomenon occurring primarily in highly transcribed genes in Tet TKO cells. By quantifying the ratio of transcription downstream of gene ends to that within the gene body using RNA-seq data (34) (Fig. 6C), we calculated gene expression and readthrough transcription across two biological replicates from control or Tet TKO cells and quantified transcriptional readthrough in genes categorized by their expression level in each sample. The distributions of readthrough ratios were shifted toward higher values in Tet TKO cells compared to WT control CD11b⁺ cells, particularly among the most highly expressed genes (Fig. 6D and SI Appendix, Fig. S11A). While we observed a clear shift in the distribution among the top 100 most highly expressed genes, this shift disappeared as one went down the ranking by gene expression levels (SI Appendix, Fig. S11A). Readthrough transcription in Tet TKO cells occurred in other differentially up-regulated genes, such as downstream of the myeloperoxidase gene (Mpo1) (SI Appendix, Fig. S11B), but it was not limited to this category, as genes equally highly expressed in Tet TKO and WT CD11b⁺ cells—such as the histone gene cluster in chromosome 13 (SI Appendix, Fig. S11C), the glutathione peroxidase-1 (Gpx1) gene (Fig. 6C), and the metalloproteinase-8 (Mmp8) gene, in which readthrough transcription continued into the downstream gene Mmp27 (SI Appendix, Fig. S11D)—displayed higher levels of transcriptional readthrough in Tet TKO cells. Together, these observations suggest a role for TET proteins in modulating transcription termination and emphasize that increased intronic and readthrough transcription of highly expressed genes, especially of long gene clusters,  may underlie heterochromatin-to-euchromatin transitions in Tet TKO cells.

Discussion

In this study, we show that acute, inducible depletion of all three TET proteins results in rapid myeloid expansion, culminating in the development of aggressive AML in mice. Only BM cells and LSK cells could transfer the disease to irradiated recipient mice, suggesting that triple Tet1/2/3 deficiency resulted in myeloid skewing, increased proliferation, and oncogenic transformation of early HSPC. Early HSPC are also thought to be cancer-initiating cells in malignancies associated with TET2 loss-of-function mutations in humans (11) as well as in myeloid expansion arising from double Tet2/3 deficiency (17) or Dnmt3a deficiency (35) in mice. Notably, the expanded myeloid cells in Tet iTKO mice contained new myeloid populations whose predominant feature was a heterochromatin-to-euchromatin switch, associated with increased expression of a single gene cluster encoding a subset of stefin/cystatin genes.

We previously showed that inducible double deletion of Tet2 and Tet3 in mice resulted, unexpectedly, in the rapid onset of AML that was fatal within 4 to 5 wk (17). This outcome was very different from that observed in mice with germline deletion of Tet2 alone, which developed late-onset diseases resembling human MDS and CMML (11, 13, 14). The difference indicated that Tet2 and Tet3 have partially redundant activities in mouse HSPC [as later confirmed by another study (36)], prompting us to examine the physiological outcome of complete TET loss of function through acute, inducible deletion of all three Tet genes. Indeed, Tet iTKO mice developed aggressive AML with the same rapid time course as Tet2/3 iDKO mice and displayed the same skewing of hematopoietic differentiation to the myeloid lineage and subsequent myeloid transformation. Given that mice with germline disruption of both Tet1 and Tet2 exhibit predominantly B cell rather than myeloid expansion (16), it is possible that Tet3 contributes more prominently than Tet2 to maintaining the proper balance of HSPC differentiation to myeloid and lymphoid lineages in mice. This hypothesis is currently being tested.

The most striking feature of the expanded myeloid cells in Tet iTKO mice was the appearance of new myeloid populations that bore neutrophil markers and showed a consistent upregulation of the stefin/cystatin gene cluster on mouse chromosome 16. Stefins are cysteine protease inhibitors (23–25) whose physiological roles are discussed in more detail below. The mouse genome contains two stefin/cystatin gene clusters, on chromosomes 2 and 16 respectively, as well as two isolated Cst genes—Cstb on chromosome 10 and Cst6 on chromosome 19. Of these, only genes in the chromosome 16 cluster were consistently up-regulated in expanded Tet iTKO myeloid cells. Our study reveals the importance of cystatin/stefin overexpression in myeloid expansion in human AML. Expression of mRNA encoding the type I cystatins CSTA and CSTB, 98-amino-acid proteins which display 54% amino-acid sequence identity (24), correlated positively with the percentage of circulating neutrophils and monocytes in the PB of AML patients. Furthermore, CSTB DNA amplification and mRNA overexpression were associated with shorter survival in these same AML patients (28, 29). Supporting the human data, the percentage of myeloid cells in PB of germline Tet2-deficient mice correlated with several markers of the preleukemic myeloproliferative disorder observed in these mice—myeloid cell expansion, extramedullary hematopoiesis and splenomegaly (37). Moreover, another aggressive mouse AML initiated by the simultaneous expression of NUP98-HOXD13 and NUP98-PHF23 also displays a dramatic upregulation of stefin genes (26), pointing to an association between stefin/cystatin overexpression and the aggressiveness of AML. There appears to be considerable redundancy between stefin/cystatin superfamily members, however (30), making it challenging to determine in either mouse or human systems whether stefin/cystatin depletion or overexpression affects oncogenic progression or survival. Future studies will be needed to determine the effects of genetic manipulation of stefin/cystatin genes in HSPC in Tet iTKO cells and to determine the contribution of stefin/cystatin expression to human cancers.

The increased expression of a subset of chromosome 16 stefin/cystatin genes in TET-deleted cells was already apparent, albeit less pronounced, in hematopoietic progenitor (LSK) cells; this was accentuated in the expanded myeloid cells and was associated with large scale changes in genome organization that occurred after TET deletion. Specifically, increased expression of the chromosome 16 stefin/cystatin gene cluster correlated with compartment switching of the entire cluster from the transcriptionally inactive Hi-C B heterochromatic compartment to the transcriptionally permissive Hi-C A euchromatic compartment. The cause-and-effect relationship between chromatin compartment switching and transcriptional upregulation is not well understood. Loss of CCCTC-binding factor (CTCF) and cohesin did not result in a global change in the organization of chromatin compartments measured in bulk populations (38); in contrast, both replication timing (39, 40) and transcriptional elongation (33) have been reported to affect the 3-dimensional compartmentalization of the genome. The effects of replication timing appear to vary with cell type since depletion of a regulator of replication timing, RIF1, resulted in different types of compartment switching in a human embryonic stem cell line (H9) compared to the colon cancer cell line HCT116 (40). A potentially more generally applicable model invokes increased transcription in heterochromatin (33, 41). In an in vitro model of influenza A infection, transcriptional elongation, particularly readthrough transcription into heterochromatin, resulted in compartment switching from heterochromatin to euchromatin by disrupting cohesin-mediated loops (33). Noncoding transcription has been reported to affect chromatin compartmentalization (42) and topologically associating domains (43) and to regulate expression of nearby genes (44). For instance, transcription of the noncoding RNA ThymoD during T cell development results in heterochromatin-to-euchromatin compartment switches and affects local DNA methylation levels, leading to demethylation of specific CpG residues which impact CTCF binding (42).

In conclusion, our molecular analysis of the aggressive myeloid leukemias developing in Tet iTKO mice in vivo has identified several effects of profound TET deficiency that appear unrelated to the established role of TET proteins in reversing DNA demethylation. The overall pattern of gene expression changes in Tet TKO myeloid cells shows no clear relation to changes in DNA methylation; moreover, the heterochromatin-to-euchromatin compartment switch and the associated increase in expression of the stefin/cystatin gene cluster in Tet iTKO myeloid cells are more closely related to high-level expression and transcriptional readthrough than to changes in DNA methylation. A few previous studies have identified functions for TET proteins that are independent of DNA demethylation. Notably, all of these involve heterochromatin: among others, maintenance of DNA methylation in heterochromatin (31), transcriptional repression of repetitive elements through modulation of KAP1/TRIM28 binding (45), and establishment of the heterochromatic histone modifications H3K9me3 and H4K20me3 (46). Our results suggest two additional functions for TET proteins that appear unrelated to DNA demethylation: TET proteins preserve three-dimensional genome organization and euchromatin-heterochromatin compartmentalization, at least at specific genomic loci, and facilitate proper transcriptional termination, especially of highly expressed genes. The latter observation may be related to the finding that the most highly expressed genes also have the highest levels of 5hmC (10, 47). Future studies will establish whether these two functions are interrelated, whether they are conserved among different cell types, and whether they require TET catalytic activity.

Materials and Methods

Detailed materials and methods are described in SI Appendix, Materials and Methods.

Acute Deletion of Tet1, Tet2, and Tet3 Genes.

Tet1^fl/fl, Tet2^fl/fl, and Tet3^fl/fl mice have been described previously (48–50). Rosa26-stop-EYFP (Rosa26-YFP^LSL) (strain #006148) and UBC-Cre-ERT2 (strain #008085) mice were purchased from Jackson Laboratories. To delete floxed alleles using the Cre-ERT2 recombinase, tamoxifen (Sigma) was solubilized at 10 mg mL⁻¹ in corn oil (Sigma) and delivered into mice by intraperitoneal injection of 2 mg tamoxifen per mouse every day for five consecutive days. The day of the last tamoxifen injection was designated Day 0 (Fig. 1A). All animal procedures were approved by the La Jolla Institute (LJI) or Ulsan National Institute of Science and Technology (UNIST) Institutional Animal Care and Use Committee and were conducted in accordance with institutional guidelines.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Next-generation sequencing data have been deposited in the Gene Expression Omnibus (GEO) repository, accession number GSE222726 (51). Previously published RNA-seq data of Tet2/3 iDKO LSK cells are available under accession number GSE72630. All study data are included in the article and/or SI Appendix.