Limitation of immune tolerance–inducing thymic epithelial cell development by Spi-B–mediated negative feedback regulation

We recently identified candidate transcriptional regulators of mTEC development by microarray analysis (Ohshima et al., 2011). Spi-B was selected for further analysis because of its possible involvement in autoimmune disease (Liu et al., 2010). We first investigated whether RANKL signaling induces the expression of Spi-B in mTECs. RANKL stimulation is known to induce differentiation of mature mTECs expressing Aire, TSAs, and an mTEC marker, UEA-1 lectin ligand (Fig. 1 A) in in vitro organ culture of fetal thymic stroma (2′-deoxyguanosine [2DG]–fetal thymus organ culture [FTOC]; Rossi et al., 2007; Akiyama et al., 2008), which is prepared by depleting cells of hematopoietic origin from fetal thymus (Aichinger et al., 2012). Quantitative RT-PCR (qPCR) revealed that Spib mRNA was significantly up-regulated by RANKL stimulation in 2DG-FTOCs (Fig. 1 B). Addition of RANK-Fc blocked RANKL-dependent Spib expression (Fig. 1 B), confirming the requirement of RANKL–RANK interactions. RANKL-dependent up-regulation of Spib preceded that of Aire, the TSA Salivary protein 1 (Spt1), and UEA-1 (Figs. 1, A and C), suggesting that Spib is an early gene that responds to RANKL signaling.

We next investigated whether RANKL signaling induces Spib expression in an established mTEC line because 2DG-FTOCs contain several types of stromal cells, including TECs. This mTEC line was retrovirally transduced with RANK (RANK-TEC) because endogenous RANK expression was undetectable in the currently available mTEC lines. Stimulation with RANKL rapidly and continuously up-regulated Spib expression in the RANK-TECs (Fig. 1 D). These findings suggest that Spib expression is rapidly induced by RANKL signaling in mTECs.

Subsequently, we determined whether Spi-B is expressed in mTECs in vivo. Adult mouse thymic stroma cells (CD45⁻TER119⁻) were sorted by staining with anti-EpCAM (an epithelial cell marker) and UEA-1 and were subsequently processed for qPCR gene expression assays. The mTEC fraction (UEA-1⁺EpCAM⁺), which exhibits high RANK expression levels (Fig. 1 E), showed significantly higher levels of Spi-B expression than did the cTEC-rich fraction (UEA-1⁻EpCAM⁺) and the nonepithelial stromal cell fraction (EpCAM⁻; Fig. 1 E). These data were consistent with increased Spi-B expression after RANKL signaling in thymic RANK-expressing mTECs.

Next, we investigated the mechanisms regulating Spi-B expression that occur downstream of RANKL signaling. Previous studies showed that murine Spib is transcribed from two different promoters (Chen et al., 1998), which can be discriminated using qPCR (Bartholdy et al., 2006). RANKL stimulation selectively induced 1 of the Spib transcripts (Spib1) in 2DG-FTOC (Fig. 1 F). Moreover, Spib1 expression was higher in mTECs than in other stromal cell fractions (Fig. 1 G). Thus, increased Spib expression in mTECs (Fig. 1 E) follows expression of Spib1 but not that of Spib2. Because RANKL signaling selectively activates the promoter that drives Spib1 transcription, the genomic region containing the conserved noncoding sequence upstream of the Spib1 initiation site (Fig. 1 H) was linked to a luciferase gene for reporter assays (Spib-P1-luc). As expected, Spib-P1-luc was responsive to RANKL stimulation in the RANK-mTEC line (Fig. 1 H), suggesting that this region contains the promoter that drives Spib transcription.

Previous studies showed that the signal transducers TRAF6 and NIK, which mediate classical and nonclassical NF-κB activation pathways, respectively, are required during RANKL-mediated mTEC differentiation (Kajiura et al., 2004; Akiyama et al., 2005, 2008). Thus, we investigated whether overexpression of these two signal transducers activated Spib-P1 in cells. Interestingly, Spib-P1 was activated only in NIK-overexpressing cells (Fig. 1 I), whereas an artificial NF-κB promoter (3κB-luc) was effectively activated by NIK or TRAF6 overexpression (Fig. 1 I). Induction of NIK activates the RelB complex (Thu and Richmond, 2010; Sun, 2012). Overexpression of the latter effectively activated the Spib-P1 promoter (Fig. 1 I). Moreover, a dysfunctional point mutation in Nik (aly/aly mutation) markedly diminished RANKL-mediated Spib expression in 2DG-FTOCs (Fig. 1 J; Shinkura et al., 1999). Thus, these data suggested that Spib is a target gene of RANKL–NIK–NF-κB signaling that promotes mTEC development (Kajiura et al., 2004; Akiyama et al., 2008, 2012b). Therefore, we investigated the roles of Spi-B in mature mTEC development in Spi-B–deficient (Spib^−/−) mice.

Adult murine mTECs are classified into subsets according to the expression levels of MHC II. mTECs expressing high MHC II levels (mTEC^hi) express a wide variety of TSAs (Kyewski and Klein, 2006; Anderson and Takahama, 2012) and are considered more mature than those expressing low MHC II (mTEC^lo) levels (Gray et al., 2007; Anderson et al., 2009). Flow cytometry revealed that the numbers and frequencies of total mTECs and mTEC^hi were significantly increased in the absence of Spi-B (Fig. 2 A), indicating that Spi-B is a negative regulator of mature mTEC development. To confirm that expression of Spi-B in TECs, but not in thymocytes, is responsible for this phenotype, we transplanted 2DG-FTOCs from Spib^−/− and WT fetal thymus into the kidney capsules of athymic nude mice. In these chimeric mice, the thymus comprises Spi-B–deficient thymic stroma and thymocytes derived from nude mice. mTEC^hi numbers were significantly increased in the developed thymus of mice receiving Spib^−/− 2DG-FTOCs (Fig. 2 B), indicating that Spi-B in thymic stromal cells limits the development of mTEC^hi cells.

We next investigated the expression of other mTEC marker proteins in Spib^−/− thymuses using flow cytometry. The mTEC^hi subset includes Aire-expressing mTECs (Gray et al., 2007). Moreover, mTEC^hi cells express high levels of co-stimulatory molecules, such as CD80, CD86, PD-L1, and CD40 (Kyewski and Klein, 2006; Klein et al., 2009; Hinterberger et al., 2010). As expected, Aire expression levels in whole thymus were increased by the lack of Spi-B (Fig. 2 C). Consistently, the number and frequency of Aire-expressing mTECs were increased in Spib^−/− thymuses (Fig. 2 D), whereas Aire expression levels in Aire-expressing mTECs were slightly reduced by the absence of Spi-B (Fig. 2 E). Remarkably, CD80 expression was diminished in Spib^−/− mTEC^hi, including Aire-expressing mTECs (Fig. 2, F and G). Moreover, mTECs expressing high levels of CD86 were significantly reduced in Spib^−/− mice (Fig. 2 H). In contrast, expression of the other co-stimulatory molecules, PD-L1 and CD40, was not affected by the absence of Spi-B (Fig. 2 I), suggesting that Spi-B selectively up-regulates CD80 and CD86 among co-stimulatory molecules in mTEC^hi cells. Altogether, these data demonstrated two distinct novel functions of Spi-B during mTEC development: Spi-B limits mature mTEC development and promotes expression of co-stimulatory molecules CD80 and CD86 in mTECs.

Because Spi-B may selectively regulate gene expression in mTECs, we subsequently identified other genes that are regulated by Spi-B in mTECs. We sorted mTEC fractions (EpCAM⁺UEA-1⁺) from Spib^−/− and WT mice and investigated gene expression using qPCR. Indeed, expression of the TSAs Spt1 and Col2 was significantly reduced in Spib^−/− mTEC fractions compared with those in WT mTEC fractions (Fig. 3 A). Thus, Spi-B may facilitate the expression of some TSA members by an Aire-independent mechanism. Notably, the expression of OPG was significantly reduced in Spib^−/− mTECs (Fig. 3 A), and reduced OPG protein expression was confirmed by immunohistochemical staining of thymic sections (Fig. 3 B). OPG competitively inhibits RANKL–RANK interactions as a decoy receptor of RANKL (Leibbrandt and Penninger, 2008), thereby inhibiting mTEC development (Hikosaka et al., 2008). Because this inhibitory effect (Fig. 2) may relate to reduced OPG expression, we further characterized Spi-B–dependent OPG expression in mTECs.

As expected, Opg mRNA expression was induced by RANKL signaling in 2DG-FTOCs (Fig. 3 C). RANKL stimulation consecutively induced Spib and Opg expression (compare Fig. 3 C with Fig. 1 C). RANKL-dependent Opg expression was significantly reduced in the absence of Spi-B in 2DG-FTOCs (Fig. 3 D). These data suggested that RANKL signaling up-regulates Spib expression and promotes expression of its own inhibitor, OPG, in mTECs.

Subsequently, we investigated the mechanism regulating Opg expression via Spi-B in mTECs. Because Spi-B overexpression did not activate the putative proximal promoter of Opg in reporter assays (not depicted), Spi-B was expected to control Opg transcription without directly binding to the proximal promoter of the gene. Thus, we analyzed CpG methylation of Opg in sorted mTEC fractions. Initially, we observed two CpG-rich regions that were hypomethylated in mTECs compared with EpCAM⁻ thymic stromal cells (Fig. 3 E), suggesting that Opg is tissue-dependently and differentially methylated (Yagi et al., 2008) in these regions (referred to as tissue-dependently and differentially methylated region [T-DMR] #1 and T-DMR#2). T-DMRs were hypermethylated in Spib^−/− mTECs compared with control mTECs (Fig. 3 E). Thus, Spi-B expression in mTECs leads to either demethylation of CpG sites or maintenance of their hypomethylated states in Opg, which may enhance Opg expression.

Moreover, we genetically verified that enhancement of mTEC development in Spib^−/− mice was dependent on OPG expression. As reported earlier (Hikosaka et al., 2008), OPG-deficient (Opg^−/−) mice showed enhanced development of mTECs (Fig. 3 F) compared with WT controls (Figs. 2 A and 3 F [green lines]). Indeed, mTEC development was similar between Opg^−/− mice and Spi-B and OPG double-deficient (Spib^−/−Opg^−/−) mice (Fig. 3 F). Thus, Spi-B does not attenuate mTEC development in the absence of OPG, strongly suggesting that the difference in mTEC development between WT and Spib^−/− mice was caused by differential expression of OPG.

Aire-expressing mTECs in the perinatal period are essential and sufficient for establishment of long-lasting self-tolerance (Guerau-de-Arellano et al., 2009). Therefore, we investigated mTEC development in embryonic and neonatal Spib^−/− mice. Flow cytometric analyses of embryonic day (E) 15.5, E17.5, postnatal day (PN) 1, and PN8 thymuses revealed incremental development of mTECs (MHC II⁺UEA-1⁺) in Spib^−/− PN1 and PN8 mice, but not in E15.5 and E17.5 mice (Fig. 4 A). Immunohistochemical analysis confirmed enhanced development of mTECs in postnatal Spib^−/− thymuses (Fig. 4 B). Moreover, Aire expression levels in mature mTECs were significantly increased only in postnatal Spib^−/− thymuses (Fig. 4 C). Thus, negative regulation of mTEC development by Spi-B characteristically starts perinatally. Although an increment in mTEC number in response to a lack of OPG has been reported only in adult mice (Hikosaka et al., 2008), OPG function in embryonic and neonatal mTEC development remained unknown. We therefore investigated whether OPG attenuates mTEC development in a developmental stage–dependent manner, similar to Spi-B. Indeed, mTEC development was not influenced by OPG deficiency in E15.5 thymuses and minimally enhanced in E17.5 Opg^−/− thymuses (Fig. 4 D). In contrast, mTEC development was markedly enhanced in PN1 and PN8 Opg^−/− mice (Fig. 4 D). In embryonic and neonatal thymus, CD80 is a more expedient marker for mature mTECs than MHC II. Consistently, CD80^hi mature mTECs were increased in Opg^−/− PN1 and PN8, but not in E15.5 and E17.5 mice (Fig. 4 E). Overall, these data suggested that the negative regulation of mature mTEC development by the Spi-B–OPG system is initiated perinatally. We subsequently identified the stromal cell type that expresses Opg using qPCR. We found that Opg is expressed in CD80^hi mature mTECs in PN1 mice (Fig. 4 F). Interestingly, Opg expression levels in mature mTECs were significantly up-regulated perinatally and plateaued in neonatal thymuses (Fig. 4 G). This may explain the stage-dependent negative regulation of mTEC development by OPG and implies a qualitative change in mature mTECs during the perinatal period.

We confirmed the RANKL–RANK–Spi-B–OPG–RANKL feedback loop in the perinatal period. Given that such a negative feedback loop exists, the absence of OPG would enhance Spi-B expression in mTECs by increasing RANKL signaling. Indeed, Spi-B expression was significantly enhanced in Opg^−/− mTECs at PN1 (Fig. 4 H). Interestingly, Spi-B was expressed in both CD80^lo immature mTECs and CD80^hi mature mTECs (Fig. 4 H) and was also up-regulated in the absence of OPG in both mTEC types, suggesting that RANKL signaling rapidly induces Spi-B in immature mTECs before promoting differentiation into mature mTECs. Cumulatively, these data strongly suggested that Spi-B–mediated Opg expression attenuates the development of mTECs by a negative feedback regulation of RANKL signaling.

We subsequently determined whether this negative regulation via RANKL–Spi-B–OPG influences mTEC-mediated thymic functions. We investigated the effects of abolishing this negative feedback loop in Opg^−/− mice rather than Spib^−/− mice because Spi-B reduces expression of CD80, CD86, and some TSAs in mTECs (Figs. 3 and 4). As reported earlier (Khan et al., 2014), expression levels of TSAs and Aire were significantly increased in Opg^−/− thymuses (Fig. 5 A). Because mature mTECs promote T reg cell development as self-antigen–presenting cells (Hinterberger et al., 2010; Lei et al., 2011; Cowan et al., 2013; Malchow et al., 2013) and an increase in mTEC cellularity enhances thymic T reg cell development (Hauri-Hohl et al., 2014), we investigated whether OPG deficiency facilitates T reg cell development in the thymus. Indeed, numbers and frequency of Foxp3⁺ T reg cells were significantly increased in thymuses of neonatal and young Opg^−/− mice (Fig. 5 B). Moreover, the frequency of thymic T reg cell precursors (CD4⁺CD8⁻Foxp3⁻GITR⁺CD25⁺; Lio and Hsieh, 2008) was increased in Opg^−/− thymuses (Fig. 5 C). In vitro FTOC experiments in which mTECs and T reg cells develop by endogenous signaling suggested that the increase in T reg cells by OPG deficiency is caused in a thymus-intrinsic manner (Fig. 5 D). Fetal thymic stroma transplantation experiments indicated that OPG deficiency in thymic stromal cells caused an increment of T reg cell frequency in thymuses, whereas the increase in splenic T reg cells was not significant (Fig. 5 E). Overall, these data supported the idea that negative feedback regulation via RANKL–Spi-B–OPG attenuates mTEC-mediated thymic generation of T reg cells, although it is possible that more T reg cells are retained in the Opg^−/− thymus.

Because negative feedback regulation of mTEC development via RANKL–Spi-B–OPG is initiated perinatally (Figs. 4 and 5), establishing long-term self-tolerance by Aire-expressing mTECs (Guerau-de-Arellano et al., 2009), we investigated the role of this negative regulation in modulation of immunological tolerance in vivo. Recent studies suggested that Aire-expressing mTECs impede antitumor immunity (Träger et al., 2012; Malchow et al., 2013; Zhu et al., 2013). Therefore, we speculated that this feedback regulation may contribute to antitumor immunity by attenuating mTEC-dependent tolerance to tumor-derived antigens. We tested this hypothesis in mouse carcinogen-induced tumor and tumor transplantation models. To exclude any other effects caused by increased RANKL activity, we initially adoptively transplanted Opg^−/− 2DG-FTOCs (BALB/c) into kidney capsules of BALB/c nude mice (Leibbrandt and Penninger, 2008; Akiyama et al., 2012a). A limited dose of the carcinogenic polycyclic aromatic hydrocarbon, methylcholanthrene (MCA), was injected into chimeric nude mice that had received Opg^−/− 2DG-FTOCs (Opg^−/−/nude) and control mice that had received WT 2DG-FTOCs. Tumor incidence was significantly increased in Opg^−/−/nude mice compared with control mice (Fig. 6 A), indicating that thymic stromal OPG limits MCA-dependent tumor incidence in vivo.

After congenic mouse tumor cells were subcutaneously transferred into nude recipients, tumor growth was significantly increased in Opg^−/−/nude compared with WT/nude mice (Fig. 6 B). Tumor-infiltrating T cells were reduced in tumors of Opg^−/−/nude mice (Fig. 6 C). In addition, T reg cell numbers were significantly increased in the draining lymph nodes of these mice (Fig. 6 D). Titers of antitumor cell antibodies evoked by tumor cell transplantation (determined by immunochemistry) were decreased in the serum of tumor-transplanted Opg^−/−/nude mice compared with those in WT/nude control mice (Fig. 6 E), suggesting weaker humoral immune responses to tumors in the former. Notably, the kinds of tumor-derived proteins recognized by serum antibodies upon Western blotting were also reduced in tumor-transplanted Opg^−/−/nude mice (Fig. 6 E, arrows), implying that TCR repertoires against tumor-derived proteins were decreased by the absence of OPG in thymic stroma. Thus, OPG expression in thymic stromal cells enhances immune responses to and limits the growth of tumors. These data further indicated that negative regulation via RANKL–Spi-B–OPG fine-tunes induction of mTEC-mediated tolerance and may ensure immune responses to tumors.

Although RANKL signaling is known to trigger differentiation of mature mTECs, molecular events linking RANKL signaling to the expression of molecules regulating mTEC functions have been largely unexplored. Here, we demonstrated that Spi-B expression links RANKL–NF-κB signaling with the expression of CD80, CD86, some TSAs, and OPG, implying a key regulatory role for Spi-B in mTECs. Interestingly, Spi-B deficiency did not affect the expression of Aire, MHC II, PDL1, or CD40, which are up-regulated in mature mTECs. This suggests that other unidentified mechanisms must regulate the expression of these molecules downstream of RANKL–NF-κB signaling.

Spi-B regulates high expression of CD80 and CD86 in mature mTECs, suggesting its positive role in mTEC-mediated function. Interestingly, Spi-B, CD80, and NF-κB1 have been associated with primary biliary cirrhosis (Liu et al., 2010; Mells et al., 2011), implying that Spi-B–mediated CD80 expression is correlated to the onset of autoimmunity.

Regulation of gene expression by Spi-B in mTECs remains unclear. However, the nearest homologue of Spi-B, PU.1, reportedly induces nucleosome remodeling and H3K4 monomethylation (Heinz et al., 2010). These epigenetic events are usually associated with changes in DNA methylation (Jones and Baylin, 2007). Given that Spi-B promotes demethylation of T-DMRs in Opg, Spi-B may exert epigenetic controls over gene expression in mTECs. Genome-wide epigenetic and expression studies of Spib^−/− mTECs may verify this hypothesis in future.

The mechanisms that activate self-tolerance by mTECs require stochastic encounters between self-reactive T cells and mature mTECs or dendritic cells that receive TSAs from mTECs. This mechanism may be inherently imperfect (Bouneaud et al., 2000) because of its stochasticity. The negative feedback regulation described herein ensures inadequacy of mTEC-dependent tolerance and could mediate detrimental effects by increasing the prevalence of autoreactive T cells. However, our data suggest that this negative regulation may contribute to tumor immunity and may optimize the trade-off between prevention of autoimmunity and induction of tumor immunity. Future analyses of this regulatory mechanism will promote understanding of tumor immunosurveillance and may lead to the development of new therapeutic approaches for cancer.

Spib^−/− mice were established on a C57BL/6 background (Sasaki et al., 2012). The C57BL/6, BALB/cA, BALB/cA nude/nude, aly/+, and aly/aly mice were obtained from CLEA Japan. The Opg^−/− mice have been described previously (Mizuno et al., 1998). The Opg^−/− mice on a BALB/cA background were prepared by backcrossing with BALB/cA mice for 10 generations. Littermates or age-matched WT mice from the same colonies as the mutant mice were used as controls. All mice were maintained under specific pathogen–free conditions and were handled in accordance with the Guidelines for Animal Experiments of the Institute of Medical Science, The University of Tokyo (Tokyo, Japan). E0.5 was defined as the first morning a vaginal plug was observed. Pregnant female mice were checked every morning and evening, and discovery of new pups was marked as day 1.

Rabbit anti–mouse keratin-5 antibody was purchased from Covance. Biotinylated UEA-1 was obtained from Vector Laboratories. The APC-Cy7–conjugated rat anti–mouse CD45, PE-Cy7–conjugated TER119, and PE-conjugated anti–mouse EpCAM (G8.8) antibodies were obtained from BD or BioLegend. The PE-conjugated rat anti-CD45, PE-conjugated mouse TER-119, FITC-conjugated mouse MHC II (M5/114.15.1), and APC-conjugated Aire antibodies were obtained from eBioscience. PE-conjugated CD80, CD86, CD40, and PD-L1 antibodies, purified anti–mouse CD16/32, FITC-conjugated ant–mouse EpCAM (G8.8), APC-Cy7–conjugated CD4, PE–Cy-7–conjugated CD8, and APC-conjugated GITR antibodies were obtained from BioLegend. PE-conjugated CD8, FITC-conjugated CD4, and FITC-conjugated CD3e antibodies were obtained from BD. PE-conjugated anti-Foxp3 (FJK-16s) was purchased from eBioscience. 7-aminoactinomycin D and recombinant mouse RANKL were purchased from Wako Pure Chemical Industries. Anti–mouse OPG antibody was purchased from R&D Systems. A RANK-Fc chimera was obtained from Sigma-Aldrich. Alexa Fluor 546–conjugated streptavidin and Alexa Fluor 488–conjugated anti–rabbit IgG were obtained from Invitrogen.

Whole thymus was used to prepare a single-cell suspension by digestion in RPMI 1640 medium containing collagenase/disperse (Roche) and DNase I (Sigma-Aldrich). Dead cells were excluded from the analysis after 7-aminoactinomycin D staining. After the Fc-receptor was blocked with CD16/32 antibody, the cells were stained with the indicated antibodies. For the analysis of T reg cells, thymocytes were prepared and used for nuclear staining according to the supplier’s protocol (Foxp3 staining buffer set; eBioscience) and analyzed using a fluorescence-activated cell sorter (Epics XL [Beckmann Coulter] or Canto II [BD]). The single-cell suspension was separated using a Percoll gradient to obtain a TEC-rich fraction, which was stained with anti-CD45, anti–TER-119, anti-EpCAM, and UEA-1 antibodies and sorted using a cell sorter (Aria; BD).

Thymic lobes were isolated from E15 embryos and were cultured on Nucleopore filters (Whatman) placed in R10 medium containing RPMI 1640 (Invitrogen) supplemented with 10% FBS (Equitech-Bio), 2 mM l-glutamine (Wako Pure Chemical Industries), 100 U/ml penicillin (Banyu Pharmaceutical), 100 µg/ml streptomycin (Meiji Seika Kaisha), and 50 µM 2-mercaptoethanol (Wako Pure Chemical Industries). FTOCs proceeded for 7 d and were subsequently analyzed by flow cytometry to investigate mTEC and T reg cell development in vitro. To isolate fetal thymic stromal (2DG-FTOC) cells, thymic lobes were cultured in the presence of 1.35 mM 2DG (Sigma-Aldrich) for 4 d. In some experiments, the 2DG-FTOC cells were further cultured in R10 with 1 µg/ml recombinant RANKL. For the blocking experiments, a RANK-Fc chimera (5 µg/ml) was added to the medium. For each day indicated after ligation, 2DG-FTOC was recovered and used for the analyses. For the transplantation experiments, the 2DG-FTOC was further cultured for 1 d without 2DG to eliminate 2DG before transplantation.

2DG-FTOCs (six lobes each) were transplanted to the kidney capsule of 6-wk-old female nude mice. For tumor transfer experiments, Meth A (5 × 10⁶ cells) was injected subcutaneously into the nude mice receiving 2DG-FTOC at 6 wk after transplantation. Tumor sizes were measured on days 10, 12, and 15 after transplantation. At day 15, the recipient mice were killed, and the inguinal lymph nodes were prepared for flow cytometric analysis. For the carcinogen-induced model, 50 µg MCA was injected subcutaneously into nude mice receiving 2DG-FTOC.

For immunohistochemistry, tissues were embedded in OCT compound (Sakura) and frozen in liquid nitrogen. Cryostat sections (6-µm thick) were fixed with ice-cold acetone and incubated with primary antibody for 1 h at room temperature. The slides were subsequently incubated with secondary antibody for 40 min at room temperature. For immunostaining of Meth A cells, cells were fixed with methanol and incubated with serum (100× dilution) for 1 h at room temperature and then incubated with secondary antibody (anti–mouse IgG–Alexa Fluor 488) and 1 µg/ml propidium iodide for 40 min at room temperature. Confocal color images were obtained using an FV1000D (Olympus) at 20 or 60 (Fig. 6 E) magnification and Fluoview software (Olympus).

Total RNA was extracted using an RNeasy micro kit (QIAGEN) or TRIzol (Invitrogen) and was subjected to random-primed reverse transcription using a Primescript II first strand cDNA synthesis kit (Takara Bio Inc.). Real-time qPCR was performed using an ABI PRISM 7300 Sequence Detection System (Applied Biosystems) and SYBR Green Master Mix (Roche or TOYOBO). The primers used were 36B4: 5′-TCCAGGCTTTGGGCATCA-3′ and 5′-CTTTATCAGCTGCACATCACT-3′; Gapdh: 5′-ACCATGTAGTTGAGGTCAATGAAGG-3′ and 5′-GGTGAAGGTCGGTGTGAACG-3′; Spib: 5′-CTGCAAGCCCTTCAGTTACC-3′ and 5′-AAAGGCAGCAGTAGCAGGAT-3′; Aire: 5′-GGTTCTGTTGGACTCTGCCCTG-3′ and 5′-TGTGCCACGACGGAGGTGAG-3′; Spt1: 5′-GTGTTGCTTGGTGTTTCCAC-3′ and 5′-GTGTTGCTTGGTGTTTCCAC-3′; Rank: 5′-GCTGGCTACCACTGGAACTC-3′ and 5′-GTGCAGTTGGTCCAAGGTTT-3′; Spib1: 5′-CTCTGAACCACCATGCTTGCT-3′ and 5′-TCCTTCTGGGTACAAACAGCTTAA-3′; Spib2: 5′-AGGGCGGCCCTGACAT-3′ and 5′-TCCTTCTGGGTACAAACAGCTTAA-3′; Opg: 5′-GGGCGTTACCTGGAGATCG-3′ and 5′-GAGAAGAACCCATCTGGACATTT-3′; Ccl19: 5′-GCTAATGATGCGGAAGACTG-3′ and 5′-ACTCACATCGACTCTCTAGG-3′; Col2: 5′-AGAACAGCATCGCCTACCTG-3′ and 5′-CTTGCCCCACTTACCAGTGT-3′; and Csnb: 5′-GGCACAGGTTGTTCAGGCTT-3′ and 5′-AAGGAAGGGTGCTACTTGCTG-3′.

Cell pellets were subjected to bisulfite treatment using the EZ DNA methylation-Direct kit (Zymo Research) according to the manufacturer’s instructions. PCR was performed using treated DNA obtained from 200–300 cells. For bisulfite sequencing, PCR products were cloned into a pGEM-T easy vector (Promega), and >16 clones were sequenced using the ABI 3130 sequencer and Big-Dye terminator kit (Applied Biosystems).

Tumors were minced with razor blades and digested in RPMI 1640 (Wako Pure Chemical Industries) containing 1 mg/ml collagenase D (Roche), 0.1 mg/ml DNase I, and 10% FBS for 1 h, with shaking. The remaining undigested fragments were passed through a needle. After addition of EDTA (final concentration: 10 mM), the cell suspension was filtered through a nylon mesh. The cells were collected and resuspended in 5 ml RPMI 1640 containing 45% Percoll and 10% FBS, and the mixture was layered on top of the same volume of 67.5% Percoll. After the cells were centrifuged at 800 g for 20 min at room temperature, the layer between 45% and 67.5% Percoll was collected.

P-values were calculated using Student’s t test, Mann–Whitney U test with two-tailed distribution, and two-sample equivalent variance parameters. For statistical analysis of the tumor-free rate, the log-rank test was performed.

This work was supported by a Grant-in-Aid for Scientific Research from the Precursory Research for Embryonic Science and Technology (PRESTO) program of the Japan Science and Technology Agency (to T. Akiyama), grants from the Japanese Society for the Promotion of Science (to T. Akiyama), Takeda Science Foundation (to T. Akiyama), Princess Takamatsu Cancer Research Fund (to T. Akiyama), grants for Priority Area Research (to T. Akiyama and J.-i. Inoue), and a Grant-in-Aid for Challenging Exploratory Research (to T. Akiyama).

The authors declare no competing financial interests.