Severe Inflammatory Reactions in Mice Expressing a GFI1^P2A Mutant Defective in Binding to the Histone Demethylase KDM1A (LSD1)

Growth factor independent 1 (GFI1) is a transcription factor with an N-terminal SNAG domain and six C-terminal C₂H₂-type zinc fingers that mediate its DNA binding to promoter and enhancer sites (1–3). The replacement of a proline residue at position 2 in the SNAG domain of GFI1 by an alanine (P2A) abrogates the repressor activity of GFI1 completely. Biochemical studies have shown that the SNAG domain interacts directly with the histone demethylase KDM1A (LSD1) and binds in close proximity to its active site and that the P2A mutation in GFI1 disrupts this interaction (3–5). LSD1 has been described as a cofactor in multiprotein complexes that enables the demethylation of histones H3 at lysine 4 (K4), leading to repression of transcription (6, 7). Hence, the finding that GFI1 binds to LSD1 via its SNAG domain offered an explanation for how GFI1 mediates its repressor activity, namely by recruiting LSD1 to target gene promoters or enhancers leading to a demethylation of histone H3K4 in the vicinity of GFI1 binding sites (5, 8).

This view has been recently challenged by studies showing that the enzymatic activity of LSD1 is not necessary for the function of GFI1 and that a catalytically inactive LSD1 can replace a wild-type (WT) LSD1 in myeloid leukemia cells at sites where GFI1 is bound to chromatin. In this model, LSD1 serves as a scaffold or a platform for the docking of other chromatin modifying enzymes such as histone deacetylases (HDACs) that enable the repressor activity of GFI1 (9–13). Of interest in this context is the finding that some LSD1 inhibitors (14), which are candidates for antileukemic drugs (15–19), do not block the activity of LSD1 per se but lead to an eviction of LSD1 and GFI1 or GFI1B/LSD1 complexes from genomic sites (16, 20). However, more experimental evidence is needed to refine this model and explain the differential activities of GFI1/LSD1 complexes at genomic sites.

Many studies have accumulated evidence that GFI1 is a key transcriptional repressor essential for many steps during hematopoiesis (21–25). For instance, severe defects were observed in GFI1 knockout (KO) mice in myeloid differentiation, leading to the lack of neutrophils and to the accumulation of immature myeloid cells (26–29). In addition, macrophages lacking GFI1 react with enhanced production of inflammatory cytokines upon stimulation with bacteria or bacterial membrane components such as LPS, supporting a repressive role of GFI1 in inflammation (28, 30–37). Besides its already established role in repressing gene expression, recent studies in innate lymphoid cells 2 (ILC2) have suggested that GFI1 could be a transcriptional activator, especially for IL-33Ra1 (ST2), which belongs to the IL-33R family (38). Indeed, GFI1 has been shown to bind to the IL-33Ra1 promotor, and absence of GFI1 in ILC2 cells impairs the expression of IL-33Ra1 with the consequence that ILC2 cells cannot be activated properly and produced cytokines like IL-5 (38, 39). How GFI1 positively regulates the expression of this gene is incompletely understood, and it remains unclear whether LSD1 is involved in this context either as a scaffold of platform for other cofactors, as seen in myeloid leukemia cells or, alternatively, as a demethylase for the histone marks that are associated with transcriptional silencing such as H3K9 methylation. There is indeed one example in which LSD1, in cooperation with an androgen receptor, can act as an activator of transcription by removing methyl groups from H3K9, thus opening this histone residue for acetylation and chromatin relaxation necessary to activate transcription (39). These data suggest a more complex role of GFI1 on the transcriptional regulation of target genes and suggest recruitment of partners depending on the cells and the biological context.

The generation of Gfi1 knock-in mice that carry a mutation in the N-terminal part of the GFI1 coding region leading to the replacement of a proline by an alanine at amino acid position 2 of the protein (GFI1^P2A) was first described by our group in 2006 (4). This mutation disrupts the ability of GFI1 to bind to LSD1 via its SNAG domain in vivo and renders GFI1 unable to repress target reporter genes. This finding supported the initial hypothesis that GFI1 mediates its repressor activity by recruiting LSD1 to target genes promoters and enhancers in which it acts to remove methyl groups from lysine 4 of histone H3 (13, 23, 40, 41). In contrast to Gfi1 KO mice, the mutant GFI1^P2A mutant protein is still expressed from the knock-in allele in organs such as thymus, where GFI1 is known to be present, but at a higher level because the autorepression of the Gfi1 locus is abrogated in Gfi1 P2A mice (4). This mouse model offers the opportunity to study in vivo the impact of the specific binding loss between GFI1 and LSD1 without affecting other LSD1 cofactors and understand the complex role of GFI1 in biological process.

Although the previous comparison of Gfi1 KO and Gfi1 P2A mice did not reveal major differences in phenotypes, we report in this study that when both strains are housed under the same specific pathogen–free (SPF) conditions over a longer period of time, P2A mice die prematurely most likely from a constitutive inflammatory reaction. We identify in this study that inflammatory cytokines are produced by macrophages in the lungs or CD8⁺T cells in the spleen but also locally in the thymus by a so-far-not-well-characterized myeloid and eosinophilic subpopulations. We present evidence that the presence of a GFI1^P2A mutant leads to a higher derepression of genes encoding inflammatory cytokines than absence of GFI1 and discuss the implications of our findings for the current model for the molecular function of GFI1 in a complex with LSD1.

Gfi1 KO, Gfi1-EGFP, and Gfi1 P2A mouse models used in this study have been previously described (4, 27, 28, 42). Mice have been bred on to C57BL/6 genetic background for at least 10 generations and were maintained in an SPF-plus environment at the Institut de Recherches Cliniques de Montréal (IRCM). The Institutional Review Board of the IRCM approved all animal protocols, and experimental procedures were performed in compliance with IRCM and Canadian Council of Animal Care guidelines.

Bone marrow (BM), spleen, blood, lung, and thymus from mice were analyzed by flow cytometry with a Fortessa (BD Biosciences) or Cyan (Beckman Coulter) flow cytometer and FlowJo software, and the cell sorting was realized on FACSAria III Sorter (BD Biosciences). Annexin V staining was performed using the annexin V detection kit from BD Biosciences. Intracellular staining for TNF-α was done by using the eBioscience cytokine intracellular staining kit (Invitrogen), GolgiPlug (BD Biosciences), and PMA and ionomycin (Sigma-Aldrich). Intracellular staining of Foxp3 was also done using eBioscience intracellular kit for transcription factor (Invitrogen). The Abs directly conjugated to a fluorochrome or biotin were purchased from BioLegend or BD Biosciences.

Thymic myeloid/eosinophil cells used the following: CD11c-BV421 (BioLegend), Siglec–Alexa Fluor 647 (AF647; BD Biosciences), Mac1-PerCp-Cy5.5 (BioLegend), Ly-6G–PE or AF700 (BioLegend), and CD45-FITC or -PE-Cy7 (BioLegend). MHC class II–allophycocyanin-Cy7 (BioLegend), Flt3-PE (BioLegend), CXCR3-PE (BioLegend), Ly-6C–FITC (BioLegend), F4/80-PE-Cy7 (BioLegend), and CD115-PE-Cy7 (BioLegend) were used to measure their expression on thymic myeloid cells. CCR3-PE-Cy7 (BioLegend) was used to measure its expression on thymic eosinophils. Thymic double negative used the following: CD45-PE-Cy7 (BioLegend), CD44-FITC (BioLegend), CD25-AF647 (BioLegend), and lin-BV421 (biotinylated CD3, CD4, CD8, B220, Gr1, Mac1, Ter119, NK1.1, DX5, and TCRgd [BioLegend and BD Biosciences] and streptavidin [SA]–BV421 [BioLegend]). Thymic double-positive (DP) used the following: CD4-PB (BioLegend), CD8-allophycocyanin (BioLegend), and CD45-PE-Cy7 (BioLegend). Hematopoietic stem cell (HSC)/MPP used the following: Ckit-allophycocyanin (BioLegend), Sca1-PE-Cy7 (BioLegend), CD150-PerCp-Cy5.5 (BioLegend), CD48-PE (BioLegend), and lin-FITC (biotinylated B220, CD19, CD3, CD4, CD8, Gr1, Mac1, Ter119, NK1.1, DX5, CD5, IgM, and IL-7R [BioLegend and BD Biosciences] and SA-FITC [BioLegend]). granulocyte–monocyte progenitors (GMP)/megakaryocyte-erythrocyte progenitors (MEP) used the following: Ckit-allophycocyanin (BioLegend), Sca1-PE-Cy7 (BioLegend), CD16/32-FITC (BioLegend), CD34-PE (BioLegend), and lin-BV421 (biotinylated B220, CD19, CD3, CD4, CD8, Gr1, Mac1, Ter119, NK1.1, DX5, CD5, and IgM [BioLegend and BD Biosciences] and SA-BV421 [BioLegend]). Erythroid cells used the following: CD71-FITC (BioLegend) and Ter119-AF647 or PE (BioLegend and BD Biosciences). Myeloid cells used the following: Gr1-allophycocyanin (BioLegend), Mac1-PerCp-Cy5.5 (BioLegend), Ly-6G–PE (BioLegend), Ly-6C–FITC (BioLegend), and CD11c-BV421 (BioLegend). B cells used the following: B220-allophycocyanin or PerCp-Cy5.5 (BioLegend), CD19-BV421 (BioLegend), and CD45-FITC (BioLegend). T cells used the following: CD44-FITC (BioLegend), CD62L-PerCp-Cy5.5 (CD62L-biotin [BioLegend] and SA-PerCp-Cy5.5 [BioLegend]), CD4-PB (BioLegend), and CD8-allophycocyanin (BioLegend). B/T cells in transplanted mice used the following: CD45.2-allophycocyanin-Cy7, CD45.1-FITC, B220-PerCp-Cy5.5, CD19-BV421, CD4-PE, and CD8-allophycocyanin. Myeloid cells in transplanted mice used the following: CD45.2-allophycocyanin-Cy7, CD45.1-FITC, Gr1-allophycocyanin, Mac1-PerCp-Cy5.5, Ly-6G—PE, and CD11c-BV42. Alveolar macrophages (AM) in the lungs used the following: CD11c-BV421 (BioLegend), Siglec-F–AF647 (BD Biosciences), and CD45-FITC (BioLegend).

A total of 5 × 10⁶ BM cells (CD45.2) from WT, KO, or P2A mice were transplanted with 0.5 × 10⁶ CD45.1 BM carrier cells by i.v. injection into the tail vein of lethally irradiated (9.5 Gy) CD45.1⁺ mice. Blood samples were taken 4, 8, 12, and 16 wk after transplantation to monitor transplantation efficiency by flow cytometry. At week 16, the transplanted recipient mice were sacrificed, and splenic and blood cells were analyzed by flow cytometry.

Lungs were perfused with 10 ml of PBS via the right ventricle before being cut finely with scissors and scalpel. Lung pieces were enzymatically dissociated in a buffer containing collagenase IV (1 mg/ml) and DNase I (0.2 mg/ml) in RPMI medium for 45 min at 37°C. Up and down with syringe and needle (18 gauge) was performed every 10 min to help the dissociation process. At the end, lung cell suspension was filtered and washed in PBS before staining, RNA extraction, or sorting. Mechanical thymus dissociation was performed, and then the cell suspension was filtered and washed before staining and/or sorting. We obtained total thymic cells in which more than 98% of the cells are CD45⁺ cells (mostly T cell progenitors).

For RNA isolation, between 1 × 10⁶ and 3 × 10⁶ of cells were lysed in Tri-Reagent (500 μl; Life Technologies). A total of 100 μl of chloroform was added to the Tri-Reagent–lysed cells, then vortexed and incubated for 5 min at room temperature before a 15-min centrifugation at 13,000 rpm. Then, the upper aqueous phase was kept and incubated with the equivalent volume of isopropanol and incubated for 5 min at room temperature and centrifuged for 5 min at 13,000 rpm. The pellet was washed twice with ethanol 70% before being air dried and resuspended in RNase-free water. RNA extraction on fewer than 1 × 10⁶ of cells was performed by using RNeasy Micro kit from QIAGEN. The cells were put in Buffer RLT with 10% of β-mercaptoethanol, and the RNA extraction was performed by following the company’s instructions (RNeasy Micro kit; QIAGEN). RT-PCR was performed using Superscript II (Invitrogen). RNA was incubated with oligodeoxythymine (Invitrogen), EDTA (Invitrogen), and dNTP (Invitrogen) for 10 min at 65°C and then DTT (Invitrogen), Superscript buffer 5× (Invitrogen), RNaseOUT buffer (Invitrogen), and the enzyme Superscript II (Invitrogen) were added and went to PCR cycle (step 1, 25°C for 10 min; step 2, 42°C for 60 min; step 3, 70°C for 15 min) to generate cDNA for real-time quantitative PCR. Real-time quantitative PCR was performed in triplicates on the ViiA7 real-time PCR machine (Applied Biosystems) in SYBR Green Master Mix (Applied Biosystems) containing specific primers for mice.

The primers are the following: Il1b forward, TTCCCCAGGGCATGTTAAGG, and Il1b reverse, TCTTGGCCGAGGACTAAGGA; Il6 forward, GTGGCTAAGGACCAAGACCA, and Il6 reverse, TAACGCACTAGGTTTGCCGA; Osm forward, TCATCCTGAGCATGGCACTG, and Osm reverse, CGTGAGGTTCGCCTGATTCT; Tnfa forward, TTCTATGGCCCAGACCCTCA, and Tnfa reverse, TTTGCTACGACGTGGGCTAC; Gfi1 forward, GTGCCATCAGAGGAGGTGAA, and Gfi1 reverse, TCTGATAACCTGCGGCCAAT; C/ebpb forward, ACCGGGTTTCGGGACTTGA, and C/ebpb reverse, GTTGCGTAGTCCCGTGTCCA; Irf8 forward, CGTGGAAGACGAGGTTACGCTG, and Irf8 reverse, GCTGAATGGTGTGTGTCATAGGC; Nr4a1 forward, TTGAGTTCGGCAAGCCTACC, and Nr4a1 reverse, GTGTACCCGTCCATGAAGGTG; and Gapdh forward, ACTGAGCAAGAGAGGCC, Gapdh reverse, TATGGGGGTCTGGGATGGAA.

The expression of the gene of interest was calculated relative to the Gapdh mRNA (ΔC_T).

ELISA experiments for IL-1β, IL-6, TNF-α (eBioscience and Invitrogen) and oncostatin M (OSM; R&D Systems) were performed by following manufacturer’s instructions on serum samples or culture supernatants.

May–Grünwald–Giemsa stainings were performed by following manufacturer’s instruction (Polysciences), and pictures were acquired by using DM4000b microscope (Leica) and Osteomeasure software (OsteoMetrics).

BM cells were isolated from tibia, femur, and hips and cultured in DMEM medium containing 10% FBS, 1% penicillin/SA antibiotics in the presence of 50 ng/ml murine M-CSF (PeproTech). At day 4, half of the medium was removed, and fresh medium containing M-CSF was added gently to the cells. BM-derived macrophages (BMDM) were harvested at day 7 or day 8 of the differentiation after treatment or not with LPS (1 μg/ml) to perform FACS, real-time PCR, and chromatin immunoprecipitation (ChIP)–PCR experiments.

GFI1 and LSD1 ChIPs were performed on 10 × 10⁶ fresh BM cells and total thymic cells (one WT thymus, two pooled KO thymi, and three pooled P2A thymi) obtained from mice, and GFI1 ChIP was also done on BMDM. The cells were cross-linked with 1.5 mM disuccinimidyl glutarate (Thermo Fisher Scientific) for 25 min and 1% formaldehyde for the final 10 min before quenching with 125 mM glycine for 5 min. Cells were initially lysed in cell lysis buffer (5 mM PIPES, 85 mM KCl, 0.5% Nonidet P-40, freshly added protease inhibitors mixture, and 0.8 mM PMSF) for 15 min on ice with occasional vortexing. After centrifugation at 4°C at 4000 rpm for 5 min, the pellet was resuspended in nuclei lysis buffer (50 mM Tris [pH 8], 10 mM EDTA, 0.1% SDS, freshly added protease inhibitors mixture, and 0.8 mM PMSF) and incubated on ice for 20 min with occasional vortexing. The lysed cells were sonicated using a Covaris E220 to generate 200–600-bp fragments. One volume of immunoprecipitation dilution buffer (300 mM NaCl and 2% Triton X-100) was added to sonicated nuclei lysates and was immune-precipitated with 5 μg of either anti-GFI1 (AF3540; R&D Systems) or anti-LSD1 (ab17721; Abcam) at 4°C overnight. Beads were added to harvest protein‐DNA complexes. After de–cross-linking, the eluate was digested with RNase A and proteinase K. The chromatin‐associated DNA was extracted using QIAquick PCR Purification Kit (QIAGEN). H3K4me2, H3K9ac, and H3K27ac ChIP experiments were performed on BMDM. The cells were just cross-linked with 1% formaldehyde for 10 min before quenching with 125 mM glycine for 5 min. Then, the protocol was similar to the protocol used for GFI1 and LSD1 ChIP. H3K4me2 (2 μg, ab7766), H3K9ac (2 μg, ab4441), and H3K27ac (2 μg, ab4729) Abs were purchased from Abcam. The primers for the ChIP-PCR are the following: Il1b promoter forward, TCGCAAGTGTGTCATCGTGG, and Il1b promoter reverse, TGCCTACCTTTGTTCCGCAC; Il6 promoter forward, GGCGGAGCTATTGAGACTGT, and Il6 promoter reverse, AAACCGGCAAGTGAGCAGAT; Osm promoter forward, CGTCGGGCATAAAGTGGCT, and Osm promoter reverse, GCATACTGGGTCCTGGTACT; Tnfa intragenic forward, AGGATTGAGTCAGTGTCACCC, and Tnfa intragenic reverse, CCGGAAGTGAAGTGTGGGTA; Tnfa promoter forward, CTAGTCCCTTGCTGTCCTCG, and Tnfa promoter reverse, GAATGAGAGCTTTTCCCCGC; Gfi1 promoter 2 forward, GCGACGAACAGAAGCGAAAG, and Gfi1 promoter 2 reverse, CACCTCACTTTCCTGTGCCT; Gfi1 promoter 3 forward, AATTCGGGAGCTGAAGGCAA, and Gfi1 promoter 3 reverse, AGAGGGGCACACAGTTTAGC; and Chr2 forward, TGGGCATATCCCTGGAGCTT, and Chr2 reverse, GGCCATCCCACAGTCACAAC.

Two-tailed Student t test was used to calculate p values where indicated. A p value ≤0.05 was considered as statistically significant. Survival curves were analyzed by log-rank Mantel–Cox test using GraphPad Prism (GraphPad Software, La Jolla, CA). The p values are as follows: *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

We had previously demonstrated that Gfi1 P2A knock-in mice (P2A) present a similar phenotype as Gfi1 KO animals (KO), such as abnormal accumulation of myeloid cells and reduced thymic cellularity (4). However, a more detailed analysis of P2A mice and a comparison with KO animals revealed significant differences. Although KO mice have an almost normal lifespan when housed under SPF⁺ conditions, P2A animals become sick and die within a few months (Fig. 1A). It is known that absence of GFI1 can render mice sensitive to pulmonary bacterial infections and lung granulomas causing their death, in particular when they are exposed to bacteria (28, 31). However, a histopathological analysis of the lungs showed no signs of pulmonary infection or granuloma formation in sick P2A mice, excluding bacterial infections as the cause of death (Fig. 1B).

Sick P2A mice were significantly smaller compared with WT, older KO (starting at 8 wk old), and young healthy P2A mice (Fig. 1C) and showed a higher spleen index (Fig. 1D). We also observed a tendency to have decreased frequencies of erythroid cells, in particular the fraction B of the erythroid progenitors in P2A BM, which was accompanied by an increase of CD71⁺Ter119^high erythroid progenitors (fractions A and B) in P2A spleen (Supplemental Fig. 1A, 1B). These data are suggestive of ongoing extramedullary hematopoiesis because erythroid cells are normally not produced in the spleen, which may also be associated with a general extramedullary hematopoiesis in P2A mice (Fig. 1E). Moreover, sick old P2A mice had higher numbers of blasts (large unstained cells) in the blood and decreased hematocrit and hemoglobin levels compared with Gfi1 KO, healthy P2A, and WT mice, suggesting that P2A mice suffer from anemia, whereas the number of RBC is not affected (Fig. 1F). These increased numbers of blasts and decreased levels of hematocrit and hemoglobin start in sick P2A mice more than 8 wk old, and the phenotype is maintained over the time. Other blood parameters such as WBC and platelet counts were similar between all the mice (data not shown). These data indicate that the presence of a mutant GFI1 unable to bind LSD1 affects hematopoiesis differently than a complete loss of GFI1.

We found that only P2A mice had decreased numbers of long-term HSCs (LT-HSCs) and short-term HSCs (ST-HSCs) compared with WT mice, whereas MPP2 cell numbers were similarly increased as in KO mice when compared with WT animals (Fig. 2A, 2B, Supplemental Fig. 1C). Although KO mice showed increased numbers of c-Kit⁺ progenitors, especially GMP cells, in the BM compared with WT mice (43, 44), P2A mice had decreased numbers of these cell populations compared with KO and also WT animals (Fig. 2A, 2C, Supplemental Fig. 1C). MEPs were also decreased in the BM of P2A mice compared with WT mice, whereas KO mice showed similar numbers than WT mice (Fig. 2A, 2C, Supplemental Fig. 1C). As previously shown (4, 43, 44), both KO and P2A mice lack neutrophils but accumulate an abnormal Mac1⁺Gr1⁺Ly-6G⁻Ly-6C⁺ myeloid cell population in the BM, in the spleen, and in the circulating blood, where it is more extreme in the P2A mice compared with the KO mice (Fig. 2D, Supplemental Fig. 1D, 1E). The blast cells detected by the Advia in the blood and the abnormal cells found by FACS in the blood are both increased; it is thus likely that this is the same cell population. However, this accumulation of abnormal myeloid cells observed in P2A mice did not worsen with age, excluding that this phenotype is related to the health degradation of the animals over time (Fig. 2E).

The strong accumulation of myeloid cells and the abnormal blood parameters led us to hypothesize that P2A mice may be prone to develop spontaneously myeloid leukemia. To test this hypothesis, we transplanted BM cells containing a large percentage of abnormal myeloid cells from KO, P2A, or WT mice as a control into irradiated recipients. Four weeks after the transplantation, the numbers of P2A donor cells in the blood of the recipients were even lower than those of KO donor cells, whereas WT donor cells were present at the expected frequencies of almost 90% (Fig. 2F). WT donor cells were still found at these high frequencies in the blood of transplanted mice 16 wk after the transplantation, whereas the percentages of KO and P2A transplanted donor cell kept decreasing over this period (Fig. 2F). This decrease of transplanted KO cells was similar to previously reported engraftment defects (22, 45), but the decrease of transplanted P2A-derived donor cells was even more pronounced. This difference correlated well with the lower numbers of HSC and progenitors found in P2A BM compared with KO BM (Fig. 2A, 2B). These data indicated that HSCs expressing a GFI1^P2A mutant have an even lower capacity to reconstitute hematopoiesis in transplanted hosts than HSCs from KO mice but also exclude the hypothesis that P2A mice spontaneously develop a transplantable leukemia. Moreover, P2A donor myeloid cells expanded less well in the blood of transplanted hosts than WT or KO donor myeloid cells but showed similar frequencies or a slight increase in the spleen (Fig. 2G). Hence, no evidence for the expansion of a potential malignant myeloid leukemic clone was found in P2A BM that could have caused the premature death of these animals.

Sick P2A mice showed significantly elevated levels of inflammatory cytokines such as IL-1β, IL-6, and TNF-α in their serum compared with WT, KO, and healthy P2A mice, with IL-6 and TNF-α being particularly high, even in healthy P2A mice (Fig. 3A). A progressive increase of IL-6 or TNF-α with age was not observed in P2A mice until the age of 20 wk, but P2A animals that are older (20–30 wk) and become sick clearly had higher concentrations of these cytokines in the blood, notably IL-6 (Supplemental Fig. 2A). OSM, which belongs to the IL-6 family of cytokines (46), showed a tendency to be increased in sick P2A mice compared with WT or KO animals but did not reach significance (Supplemental Fig. 2B).

Numbers and the frequencies of CD4⁺ memory effector T cells (CD62L⁻CD44⁺CD4⁺) were increased in blood and spleen of KO mice compared with WT mice and showed a tendency to be even more increased in P2A than in KO animals (Fig. 3B, 3C, Supplemental Fig. 2C), although CD4⁺ T cells numbers were decreased in both KO and P2A mice (Fig. 3B, 3C). Splenic CD8⁺ T cell numbers were similar between WT, KO, and P2A mice in the spleen but were significantly higher in the blood of P2A mice than in WT and KO animals (Fig. 3D, 3E, Supplemental Fig. 2C). The increased numbers of effector memory T cells (CD62L⁻CD44⁺CD4⁺ and CD62L⁻CD44⁺CD8⁺) in the spleen became more accentuated in P2A mice with age (Supplemental Fig. 2D). The number of B cells, in which GFI1 is expressed (21, 24), were lower in the spleen and in the blood of P2A mice compared with WT and KO mice (Supplemental Fig. 2E). We also observed a higher frequency of Foxp3⁺ splenic regulatory T cells in P2A mice compared with WT and KO animals (Fig. 3F) and increased percentages of CD8⁺TNF-α⁺ T cells in the spleen of P2A mice compared with WT controls (Fig. 3G).

Mice transplanted with P2A or KO donor cells showed lower numbers of donor-derived CD4⁺ compared with WT controls, but although this reduction was also seen for CD8⁺ T cells derived from KO mice, CD8⁺ T cells derived from P2A donors were higher and reached WT levels (Supplemental Fig. 2F). Of interest, mice transplanted with P2A BM had CD45.2⁺ CD8⁺ donor cells that contained higher percentages of TNF-α–producing cells than CD45.2⁺ CD8⁺ cells derived from WT or KO BM transplants also after stimulation (Fig. 3H). This elevated TNF-α production was also observed in the total splenic P2A CD45.2⁺ donor cell population (Fig. 3H). However, only donor CD8⁺ T cells were identified as a potential source for inflammatory cytokines, whereas neither CD4⁺ T nor myeloid P2A donor cells showed increased percentages of TNF-α⁺ cells compared with WT or KO donor cells (Supplemental Fig. 2G). These data suggest that the ongoing inflammation observed in P2A mice is a cell autonomous and may be a transplantable phenotype, which can be measured despite the very low engraftment of P2A BM cells and also reveal CD8⁺ T cells, where GFI1 is known to be expressed (42), as an additional source of inflammatory cytokines notably TNF-α in P2A mice.

In addition to the hematopoietic compartment, we also observed that P2A overexpress inflammatory cytokine genes in lung cells (Supplemental Fig. 2H), with Il1b and Osm being significantly higher in P2A mice compared with WT and KO animals. Numbers of AM, known to express GFI1 in inflammatory conditions (31), were significantly increased in numbers in P2A mice compared with KO and WT animals (Fig. 3I) and showed increased mRNA expression of inflammatory cytokine genes in P2A and KO mice compared with WT mice notably for the genes encoding IL-6, OSM, and IL-1β (Fig. 3J). This suggests that lung AM also contribute to the systemic ongoing inflammation observed in P2A mice.

We also observed overexpression of Il1b, Il6, Osm, and Tnfa mRNA in total thymic cells of P2A and KO mice compared with WT controls (Fig. 4A, Supplemental Fig. 3A). The increased mRNA levels correlated with higher percentages of TNF-α⁺ cells in the thymus of P2A mice compared with WT and KO thymi (Fig. 4B), suggesting that an inflammatory reaction also occurs in the thymus of P2A mice.

To further explore the cellular source for the increased inflammatory cytokines in the thymus of P2A mice, we analyzed different subpopulations in more detail. CD11c^low/intMac1⁺Ly-6G⁻Siglec-F⁻ myeloid cells (called thymic Mac1⁺) (Fig. 4C) remained at WT levels in absolute numbers in the P2A and KO thymus but had increased frequencies among CD45⁺ cells in the thymus owing to the low numbers of thymic DP and single-positive T cells in KO and P2A animals (Fig. 4D, 4E, Supplemental Fig. 3B, 3C). CD11c^low/intMac1⁺Ly-6G⁻Siglec-F⁻ (Mac1⁺) myeloid cells, which express GFI1 (Fig. 4F, Supplemental Fig. 3D), showed increased levels of Il1b, Il6, Tnfa, and Osm mRNA in KO and P2A mice compared with the respective population of WT animals (Fig. 4G) and contained a significantly increased percentage of TNF-α⁺ cells, as shown by flow cytometry (Fig. 4H). In addition, they expressed markers of the monocytic/macrophage lineage such as CX3CR1, Ly-6C, F4/80, CD115, and MHC class II, but not FLT3, which is specific to the dendritic cell lineage (Fig. 4I, Supplemental Fig. 3E). Compared with BM-derived monocytes, these cells also show higher expression levels of Irf8 and Nr4a1 mRNA, which were previously described to be expressed in thymic macrophages (47) and show morphologic features of monocyte (Supplemental Fig. 3F, 3G). The CD11c^low/intMac1⁺Ly-6G⁻Siglec-F⁻ myeloid cells found in KO and P2A expressed the same markers than the BM abnormal myeloid cells of KO and P2A mice (Supplemental Fig. 3H), suggesting that they stem from the abnormal monocyte pool of cells found in the BM or in the blood of KO and P2A mice.

Next, we analyzed thymic eosinophils that are defined by CD11c^low/intMac1⁺Ly-6G⁻Siglec-F⁺ (48), CCR3 (49), and their characteristic morphology (Fig. 5A, 5B, Supplemental Fig. 3G). They also express GFI1 (Fig. 5C, Supplemental Fig. 3I) and showed increased expression of Il1b and Osm mRNA in P2A and KO mice compared with WT controls, both being even significantly higher in P2A than in KO animals (Fig. 5D). Thymic eosinophil total cell counts were decreased in KO and P2A mice compared with WT mice, but relative frequencies were higher in KO and P2A mice than in WT mice (Fig. 5E, 5F) because of the severe reduction of the overall thymic cellularity over time in KO and P2A mice, especially with regard to CD4⁺CD8⁺ DP cells (Fig. 4D). These data indicate that GFI1 is not only expressed in thymic lymphoid and myeloid cells but also in thymic eosinophils and suggest that GFI1 exerts a regulatory function in all these populations that affects the expression of genes for inflammatory cytokines.

The frequency of apoptotic thymic cells is known to be significantly higher in KO mice compared with WT animals, but P2A thymus show even higher proportion of apoptotic cells than KO and WT thymus (4, 25), whereas the frequencies of apoptotic cells in the BM and in the spleen are not affected by GFI1 absence or mutation (Fig. 6A). Stress signals such as cell death can activate myeloid cells and trigger the production of inflammatory cytokines (50–52), suggesting that the abnormal myeloid population found in KO and P2A mice may have a deregulated inflammatory response to stress in the thymus. To test this hypothesis, we sorted the abnormal myeloid cells found in KO and P2A BM and monocytes from WT BM, where no cell death occurs, and thus, a signal to trigger myeloid cells to produce inflammatory cytokines should be absent (Fig. 6B) and these cells treated with LPS as a proinflammatory stimulus. Sorted myeloid cells from KO and P2A mice showed a stronger response than WT cells to LPS when compared with the baseline expression of inflammation without LPS. Indeed, the induction of Il1b, Il6, and Tnfa mRNA expression was much stronger in KO cells and in P2A cells than in WT myeloid cells (Fig. 6C, Supplemental Fig. 4A), which was in correlation with increased levels of these cytokines in the supernatants of P2A and KO cells (Fig. 6D). In this study, P2A myeloid cells even showed a tendency for higher concentrations of these cytokines compared with myeloid KO cells and, in the case of TNF-α, reaches significance. This suggests that the abnormal monocyte-like population accumulating in KO and P2A mice has the potential to produce inflammatory cytokines but only after stimulation.

Next, we generated BMDM from WT, KO, and P2A BM (Supplemental Fig. 4B) to further investigate the mechanism by which GFI1 could regulate inflammatory pathways. BMDM generated from WT, KO, and P2A mice express similarly high levels of F4/80 and Mac1 as well as CD115 (Supplemental Fig. 4B), which are characteristic markers for macrophages. After treatment with LPS, we found that KO BMDM show an increased expression of Tnfa mRNA compared with LPS-treated WT BMDM in agreement with our previously reported data (28), but only P2A BMDM produced significantly higher levels of TNF-α in supernatants (Fig. 6E). A difference of Il6 mRNA expression or IL-6 production after treatment with LPS between WT, KO, and P2A BMDM was not observed (Supplemental Fig. 4C). These data suggest that absence of GFI1 or the disruption of the GFI1/LSD1 interaction through the introduction of a P2A mutation in the Gfi1 gene leads to an increased inflammatory response of myeloid cells of different origins toward stimuli, which is, in specific instances, significantly more pronounced in cells from P2A mice compared with cells from KO animals.

To test whether the GFI1^P2A mutant protein can bind DNA, we performed a quantitative ChIP-PCR experiment using primers covering previously established binding sites in the Gfi1 promoter (53). Both the WT GFI1 and the GFI1^P2A proteins were detected at the Gfi1 promoter in total thymic cells and BM cells (Fig. 7A). At the same Gfi1 promoter site, LSD1 is also present together with GFI1 in WT thymic cells but is absent in thymic cells from P2A mice (Fig. 7B). This was expected because it is well established that the P2A mutation disrupts the interaction between GFI1 and LSD1, leading to the loss of GFI1 repressive activity (2, 4, 13). However, the same experiment also showed that in BM cells, LSD1 is not readily detected at the Gfi1 promoter, although GFI1 is present (Fig. 7B), possibly because of the lower expression level of GFI1 in BM cells compared with total thymic cells. Our finding in total thymic cells indicated that a GFI1^P2A mutant protein can still occupy target genes sites, even if it no longer interacts with LSD1.

Next, we interrogated previously reported ChIP sequencing (ChIP-seq) data obtained with MLL-ENL–transformed leukemic cells (54) and found peaks for GFI1 in the Osm promoter and in Tnfa intragenic regions (Fig. 7C), but not in the promoter regions of the IL1b and IL6 genes (53). We confirmed by ChIP-PCR that GFI1 but also GFI1^P2A can occupy sites in the Osm promoter and Tnfa intragenic region (Fig. 7D), but not within regions of the Il1b and Il6 genes (Supplemental Fig. 4D). LSD1 was found to bind to the Osm promoter and Tnfa intragenic region together with GFI1 but was absent in cells expressing the GFI1^P2A mutant (Fig. 7D, Supplemental Fig. 4D). We also found GFI1 binding on Gfi1, Osm, and Tnfα promoters and the Tnfα intragenic region in WT and P2A BMDM (Fig. 7E), which suggests that GFI1 can directly regulate Osm and Tnfa genes in different cell types and again confirmed that a GFI1^P2A mutant can still occupy genomic sites despite being unable to bind to LSD1.

Because LSD1 demethylates H3K4me2 (55), we tested the presence of this histone mark at sites occupied by GFI1 in KO and P2A BMDM by ChIP-PCR with the same primers used to detect GFI1. The enrichment of the histone mark H3K4me2 on Gfi1, Osm, and Tnfα promoters and the Tnfα intragenic region had the tendency to be higher in KO and P2A cells compared with WT BMDM (Fig. 7F) and even seemed to be more elevated on the Osm and Gfi1 promoters in P2A BMDM compared with KO BMDM, but a significance at a p value below 0.05 level was not reached (Fig. 7F). Because it is also known that GFI1 recruits HDACs to promoters of target genes probably through LSD1 (9, 12, 56), we also tested H3K27 and H3K9 acetylation. ChIP-PCR demonstrated that these marks were indeed differently enriched at the Gfi1, Osm, and Tnfα promoters and the Tnfα intragenic region between WT, KO and P2A BMDM with a significant increase of H3K27ac enrichment on the Osm promoter and the Tnfα intragenic region of P2A and/or KO cells when the binding of GFI1 is the strongest (Fig. 7F, Supplemental Fig. 4E). This suggests a derepression of GFI1 target genes, especially Tnfα and Osm, occurs in the absence of GFI1, which is more pronounced in the presence of a GFI1^P2A mutant.

We report in this study that knock-in mice expressing a mutant form of GFI1 (GFI1^P2A) lacking the ability to bind LSD1 and associated histone-modifying enzymes such as HDACs die at the age of 5–7 mo, even under SPF conditions. In contrast, under the same conditions, KO mice that lack the GFI1 protein entirely survive well and do not show signs of illness. The symptoms of P2A mice include weight loss, high spleen index, an expansion of myeloid cells with infiltration of all hematopoietic organs reminiscent of the phenotype of a myeloproliferative disease that we had previously described in mice expressing lower than WT levels of GFI1 (GFI1 knockdown) (43). Nevertheless, P2A mice die earlier than GFI1 knockdown animals and show phenotypic differences such as absence of elevated platelet numbers or decreased hematocrit and hemoglobin, suggesting a different cause of death. Because we observed high levels of IL-6 or TNF-α in the peripheral blood of P2A mice and cells from lung, thymus and spleen associated with an expansion of effector memory T cells, in the absence of any damage- or pathogen-associated molecular patterns, we propose that a constitutive, ongoing immune reaction and inflammation takes place in P2A mice that contributes significantly to their death. It is therefore conceivable that the presence of a GFI1^P2A protein deregulates the immune response and inflammatory pathways in a different way than this is the case in the complete absence of GFI1.

We have reported before (4) and confirm in this study that both KO mice P2A animals accumulate an abnormal myeloid cell population that we characterized in this study as being CD11c^low/intMac1⁺Ly-6G⁻Siglec-F⁻. A direct comparison between both strains showed that this phenotype is more extreme in P2A mice and reaches almost all hematopoietic tissues through infiltration. Myeloid cells such as monocytes are known to be high producers of inflammatory cytokines (47), and it was therefore possible that this abnormal population is responsible for the overexpression of inflammatory cytokines and subsequently for the increased serum levels of TNF-α and IL-6 observed in P2A mice. However, our experiments with these cells directly sorted from BM or spleen showed that this is not the case and that mRNA levels of the genes for Il1b, Il6, Tnfa, and Osm were unchanged or even lower in P2A or KO mice compared with myeloid cells from WT animals. However, we found that this CD11c^low/intMac1⁺Ly-6G⁻Siglec-F⁻ myeloid cell subset, when isolated from the thymus, has the potential to overexpress the genes of inflammatory cytokines in KO and P2A mice and produce higher levels of TNF-α in KO and P2A mice compared with WT mice. Even though the monocyte-like cells found in the thymus do not show significant differences between KO and P2A mice in terms of mRNA expression for inflammatory cytokine genes, we could observe that P2A total CD45⁺ thymic cells produce significantly higher levels of TNF-α than the total KO thymic cells, and a similar tendency is found in the monocyte-like population, even though the significance is not reached. This suggested that a monocyte-like population in the P2A thymus has a more deregulated inflammatory response than the respective cells from WT or KO mice, but it is also possible that other cells found in the thymus could be involved in the higher TNF-α production, such as thymic macrophages.

A number of studies, including our own, had previously reported that KO mice are particularly sensitive against bacterial wall components such as LPS or other, similar agents (28, 31–33) and do not tolerate doses that normally are not affecting WT mice. In addition, we had previously shown that BMDM from KO mice strongly upregulate mRNA expression of Tnfa and other genes involved in inflammation, which attributed a new function to GFI1 as a negative regulator of the innate immune response and inflammatory reactions in general (30, 31, 33). However, these findings were made after stimulation of cells lacking GFI1 with agonists of inflammatory pathways (33) or following a stimulation of KO animals (28) with LPS or under animal housing conditions not specifically excluding pathogens. It is therefore likely that in contrast to KO mice, P2A animals suffer from a constitutive, chronic inflammation.

The thymus is an organ where cell death occurs, and it has been proposed that ongoing cell death produces DNA damage–associated patterns that can stimulate an inflammatory response and the production of inflammatory cytokines (57, 58). Indeed, KO and P2A mice show high levels of apoptosis in the thymus compared with WT mice, with P2A mice having a stronger phenotype than KO mice. This increased level of cell death could generate DNA damage–associated patterns in KO and even more in P2A thymus, and the thymic myeloid cells could respond to these signals and acquire an inflammatory profile. In agreement with this, BM cells, which are not prone to undergo cell death, do not show spontaneous overexpression of inflammatory cytokines in P2A or KO mice. However, because LPS stimulation of myeloid cells from BM induced Il1b, Il6, Tnfa, and Osm mRNA expression, which in the case of TNF-α and IL-6 production, was even higher in P2A cells than in KO cells, it is conceivable that the abnormal myeloid population found in the KO and P2A mice has the potential to become inflammatory under stimulation. It is thus plausible that the damaged cells could represent or become a stimulatory signal in the thymus of KO and P2A mice.

However, it is also possible that thymic myeloid cells produce inflammatory cytokines not only upon an internal stimulus in the KO and P2A thymus but also by responding to the systemic inflammation occurring in the mice as we also found production of inflammatory cytokines in the lung and increased numbers of memory effector T cells in the spleens of KO and P2A mice and P2A mice showed and even higher number of CD8 T cells in the spleen that produced high levels of TNF-α. Because evidence for a deregulated immune response such as increased effector memory cells and regulatory T cells is found in many organs of P2A mice, it is likely that several different sources of inflammatory cells exist that generate the phenotype seen in these animals. Where and when the inflammatory process in P2A mice begins remains open, but because the severity is exacerbated with age, it is possible that self-amplifying mechanisms of different immune cells are the reason, but further experimentation is needed to draw a definitive conclusion.

Interestingly, eosinophils in thymus had so far not been described to express Gfi1, but our Gfi1-GFP reporter mice clearly indicated that this is the case. These cells, known to be producers of OSM in the thymus (59), could be another source of inflammatory cytokines found in P2A and KO mice. Our findings support these eosinophils as a source of both IL-1β and OSM in the thymus of P2A and KO mice, with P2A eosinophils showing even higher mRNA expression of these cytokines than KO eosinophils. This suggests that at least the regulation of the expression of the Osm gene could occur directly through GFI1 because we found direct binding to the intragenic region. It is thus likely that transcriptional derepression of Osm in eosinophils and mostly in other cells where GFI1 is expressed occurs in P2A mice, contributing to the inflammation seen in these animals at higher level than KO mice, for which the derepression could be controlled.

Our data confirm not only that the GFI1/LSD1 complex is a key regulator of inflammatory pathways, a conclusion that was already supported by previous published findings (28, 31–33), but also demonstrate that the presence of a GFI1 mutant unable to bind to LSD1 is sufficient to trigger a strong inflammatory response that appears to be constitutive and fatal in mice. Our ChIP-PCR experiments indicated that the GFI1^P2A mutant protein is still able to bind to known target site for instance to specific regions in the GFI1 promoter, which must occur in absence of LSD1. This observation is different from what was described in a recent study in which GFI1 has been shown to be evicted from chromatin after in human AML cells when its interaction with LSD1 was blocked by a small molecule inhibitor that occupies the site where the SNAG domain docks into the active site in LSD1 (16). However, our data clearly indicate that in total thymic cells, BM cells and a BMDM GFI1^P2A mutant protein remains at chromatin. This is in agreement with another study in which an LSD1 inhibitor did not affect GFI1 recruitment to target genes (17). The presence of GFI1^P2A at genomic sites illustrates well the different situation in KO and P2A cells where Gfi1 binding sites are either still occupied by a GFI1^P2A mutant or free from GFI1 in KO cells because the GFI1 protein is absent. Given that GFI1^P2A protein still occupies DNA in thymic cells, BM, and BMDM at sites normally bound by GFI1, transcription at GFI1^P2A occupied target genes will be derepressed because LSD1 and associated histone-modifying enzymes cannot be recruited to these sites. This is likely the case for the Osm promoter and a Tnfα intragenic region where GFI1 and GFI1^P2A can directly bind, but not for the promoters for Il6 and Il-1b. Thus, although GFI1/LSD1 could repress the activity of these two inflammatory genes, GFI1^P2A alone loses its repressor activity because of the absence of LSD1 but also HDACs because it has been shown LSD1 functions as a scaffold to enable binding of other histone modifiers such as HDACs (16, 20). Hence, the derepression of transcription of the Osm and Tnfα genes that would be expected in cells expressing a GFI1^P2A protein should be associated with increased methylation or acetylation of histone tails. Enrichment of H3K27 acetylation and, albeit less strong, of H3K4 dimethylation was indeed seen at the Osm promoter and the Tnfα intragenic region in P2A BMDM. This would support the idea that an impaired recruitment of LSD1 and associated HDACs by the GFI1^P2A mutant leads to increase expression of OSM and TNF-α but would not explain the increased levels of IL-1β and IL-6, which is also seen in P2A mice. However, the deregulation of these two cytokines could be mediated through OSM, which has been shown to induce IL-6 (60, 61) and is also required for IL-1β production (62).

Several factors indicate that the derepression of inflammatory genes is stronger in P2A cells than in KO cells. For instance, TNF-α serum levels, the number of TNF-α–producing AM, thymic monocytic cells or eosinophils, CD8⁺ TNF-α–producing cells, and TNF-α levels reached after LPS stimulation are all higher in P2A mice than KO mice. It is likely that the sum of all these effects cause a constitutive inflammation in P2A mice that, over time, leads to their death. This deregulation is less strong in Gfi1 KO mice, and when held under conditions minimizing external inflammatory stimuli, these animals survive well. It is therefore conceivable that a mutant GFI1^P2A protein that is still present at target genes impedes other factors that in GFI1 KO cells would be able to mitigate the transcriptional derepression of inflammatory cytokine genes. Alternatively, the GFI1^P2A protein could assume the function of a transcriptional activator by recruiting coactivators, but additional experimental data would have to be obtained to support such a new biochemical function.

We thank Dr. Julie Ross for reading and correcting the manuscript, Mathieu Lapointe for technical assistance, IRCM animal facility people for excellent animal care, Eric Massicotte and Julie Lord for cell sorting, and people from the molecular biology, histology, and microscopy facilities.

The authors have no financial conflicts of interest.