Ceacam1L Modulates STAT3 Signaling to Control the Proliferation of Glioblastoma-Initiating Cells

Gliomas are the most common brain tumor, possess the characteristics of glial cells, astrocytes, and oligodendrocytes, and have been classified into four grades (WHO grade 1–4) based on their pathologic features. Glioblastoma multiforme (GBM) is the most malignant glioma (WHO grade 4), and patients with GBM have a median survival of approximately 1 year. In spite of tremendous efforts to effectively treat GBM, the overall survival rates of patients with GBM have remained unchanged over the past few decades.

The discovery of GBM-initiating cells (GIC) has had a significant impact on GBM research (1). GICs have a strong self-renewal capability, express neural stem cell (NSC) markers, such as CD133 (also known as Prominin1), CD15 (also known as stage-specific embryonic antigen 1 and Lewis X), and the intermediate filament protein Nestin, and are more resistant to radio- and chemotherapies than non-GICs (2–4). GICs have also been shown to exploit the signaling pathways involved in maintaining NSCs (2–4). NSCs only exist in the subventricular zone and hippocampus (5, 6), both of which contain a special microenvironment (niche) for the maintenance of NSCs, whereas GBM arises in many areas in the brain. It currently remains unknown whether GICs generate an NSC niche anywhere or use an unknown mechanism for their maintenance in non-NSC niches.

We previously established the mouse GIC (mGIC) lines, NSCL61 and OPCL61, by overexpressing an oncogenic HRas^L61 in p53-deficient NSCs and oligodendrocyte precursor cells (OPC), respectively (7, 8). These mGICs formed transplantable GBM with hypercellularity, pleomorphism, multinuclear giant cells, mitosis, and necrosis, even when as few as 10 cells were injected into the brains of nude mice. These findings indicated that they were highly enriched in GICs. Using DNA microarray analyses, we compared the gene-expression profiles of mGICs with those of their parental cells and identified genes that were upregulated and downregulated in mGICs. By evaluating the candidate genes using human GIC (hGIC) lines and GBM tissues, we successfully selected potential GIC-specific genes (7–9).

Among the candidate genes evaluated, we focused on carcinoembryonic antigen–related cell adhesion molecule 1 (Ceacam1, also known as CD66a and BGP1). Human Ceacam1 consists of 11 splicing variants, seven of which are transmembrane proteins with either a short (Ceacam1S) or long cytoplasmic tail (Ceacam1L) whereas the others are secretion forms (10–12). Transmembrane Ceacam1 exists as a monomer, cis-/trans-homodimer, or cis-/trans-heterodimer with either Ceacam1-splicing variants or other Ceacam family members, depending on the cell environment; therefore, the intracellular domain of Ceacam1 is a monomer or a homo-/hetero-dimer (11, 12), which transmit different signals. Ceacam1 was previously shown to be involved in many different biologic functions, including angiogenesis, insulin clearance, immune modulation, and proliferation (10–12). Ceacam1L was also identified as a substrate of the insulin growth factor receptor and EGFR, both of which are frequently activated in malignant gliomas (13), whereas modified Ceacam1 acted as either an amplifier or attenuator of these receptors in a cell-dependent manner. These findings prompted us to investigate the role of Ceacam1 in GICs.

Animals were obtained from the Laboratory for Animal Resources and Genetic Engineering at the RIKEN Center for Developmental Biology (CDB) and from Charles River Japan, Inc. All mouse experiments were performed following protocols approved by the Animal Care and Use Committees of RIKEN CDB, Ehime University, and Hokkaido University. Chemicals and growth factors were purchased from Sigma and PeproTech, respectively, except where indicated.

Mouse primary neural cells (NSCs and OPCs), NSCL61, OPCL61, human NSCs (hNSC; Invitrogen), and GICs (hGICs, E3 and E6) were cultured as described previously (7–9, 14). Cells were cultured in chamber slides (Nunc) precoated with fibronectin and poly-d-lysine for immunostaining, as described previously (14).

Immunostaining of paraffin-embedded human brain tumor sections (6-μm thick) and mouse cells or brain sections was performed as described previously (7, 8). Ceacam1 was retrieved by HistoVT One according to the supplier's instructions (Nacalai Tesque). The sections were permeabilized with 0.3% TritonX-100 in PBS for penetration, treated with a blocking solution (2% skim milk, 0.3% Triton X-100, and PBS) for 1 hour, and incubated with primary antibodies for 16 hours at 4°C. Cells were fixed and immunostained as described previously (14). The following antibodies were used to detect antigens: mouse anti-Ceacam1 (1:50; R&D Systems), anti-GFAP (1:500; Chemicon, 1:400; Sigma for human cells), rat monoclonal anti-GFP (1:500; Nacalai Tesque), mouse monoclonal anti-Nestin (1:200; BD Biosciences), mouse monoclonal anti-CD15 (1:200; BD Pharmingen), rabbit polyclonal anti-STAT3 (1:100; Santa Cruz Biotechnology), rabbit polyclonal anti-phosphorylated STAT3 (1:100; Cell Signaling Technology), rabbit polyclonal anti-EGFR (1:50; Cell Signaling Technology), rabbit polyclonal anti-Caspase 3 (1:1,000, Cell Signaling Technology), and rabbit monoclonal anti-Ki67 (1:200, Thermo Scientific). Antibodies were detected with Alexa568-conjugated goat anti-rabbit IgG (1:500; Molecular Probe), Alexa488-conjugated goat anti-mouse, -rabbit, or -rat IgG (1:500; Molecular Probe) and goat anti-mouse IgG-Cy3 (1:500; Jackson ImmunoResearch). Cells were counterstained with DAPI (1 μg/mL) to visualize the nuclei.

hGICs were used according to the research guidelines of the Ehime University Graduate School of Medical Science and the Hokkaido University Institute for Genetic Medicine. The detailed characterization of hGICs has been reported in ref. 9. Poly(A)⁺ RNA was prepared using a QuickPrep mRNA Purification Kit (GE Healthcare). Control human brain total mRNA (CB) was purchased from Invitrogen. cDNA was synthesized using a Transcription First Strand cDNA Synthesis Kit (Roche).

RT-PCR was carried out as described previously (14), with the cycle parameters of 20 seconds at 94°C, 30 seconds at 57°C, and 60 seconds at 72°C for 35 cycles (GICs) or 40 cycles (GBM tissues). Cycles for gapdh were 15 seconds at 94°C, 30 seconds at 53°C, and 90 seconds at 72°C for 22 cycles. The following oligonucleotide DNA primers were synthesized: for mouse Ceacam1, the 5′ primer was 5′-ATCCTCCCAAGAGCTCTTTATC-3′, and the 3′ primer was 5′-TTTGTGCTCTGTGAGATCTCG-3′; for human Ceacam1, the 5′ primer was 5′-ACACCATGGGGCACCTCTCA-3′, and the 3′ primer was 5′-GATCGTCTTGACTGTGGTCCT-3′; for sox1, the 5′ primer was 5′-AGGGCTACATGAGCGCGTCG-3′, and the 3′ primer was 5′-CTAGATGTGCGTCAGGGGCA-3′; for aldh1a1, the 5′ primer was 5′-AGGGGCAGCCATTTCTTCTC-3′, and the 3′ primer was 5′-GGCAATGCGCATCTCATCTG-3′; for cxcr4, the 5′ primer was 5′-CTGACCTCCTCTTTGTCATCAC-3′, and the 3′ primer was 5′-GTCTTGAGGGCCTTGCGCTT-3′; for dll1, the 5′ primer was 5′-TGGTGGTCTGCGTCCGGCTG-3′, and the 3′ primer was 5′-ACCGACTGGTACTTGGTGTC-3′; for notch3, the 5′ primer was 5′-ATGGTGGAAGAGCTCATCGC-3′, and the 3′ primer was 5′-TGGCCTCCTGCTCTTCTTGG-3′; for hey1, the 5′ primer was 5′-GCGGACGAGAATGGAAACTTG-3′, and the 3′ primer was 5′-AGTCCTTCAATGATGCTCAGAT-3′; for egfr, the 5′ primer was 5′-GATGAAAGAATGCATTTGCCAAG-3′, and the 3′ primer was 5′-GGGGCTGATTGTGATAGACAGG-3′; for stat3, the 5′ primer was 5′-GTGTCAGATCACATGGGCTAA-3′, and the 3′ primer was 5′-TGCCTCCTCCTTGGGAATGT-3′; for ptpn6, the 5′ primer was 5′-GTGTCCTCAGCTTCCTGGAC-3′, and the 3′ primer was 5′-GTCTGTCCATCGCGAAATGC-3′; for ptpn11, the 5′ primer was 5′-AAAGGGGAGAGCAATGACGG-3′, and the 3′ primer was 5′-ATTCACCGTGTTTTGCAGGC-3′; for c-Src, the 5′ primer was 5′-ACATCCCCAGCAACTACGTG-3′, and the 3′ primer was 5′-AGCTTCTTCATGACCTGGGC-3′. The primers for gapdh were described previously (14).

hGICs were immunolabeled with rabbit polyclonal anti-Ceacam1 (5 μg/mL; LSBio) and mouse monoclonal anti-CD15 (5 μg/mL; BD Pharmingen), following with Alexa 488-conjugated goat anti-mouse IgG (1:400; Molecular Probe) and Cy5-conjugated goat anti-rabbit IgG (Molecular Probe; diluted 1:400). The cells were analyzed in an Aria II (Becton Dickinson) using a dual-wavelength analysis (488-nm solid-state laser and 638-nm semiconductor laser). Propidium iodide–positive (i.e., dead) cells were excluded from the analysis.

The side population (SP) was analyzed as shown previously (15). Reserpine (10 μmol/L), an inhibitor of some ABC transporters, was used to identify SP.

Complementary DNAs (cDNA) were cloned as described previously (7). Human ceacam1L cDNA was inserted into pcDNA3.1-hyg (Invitrogen), pcDNA3-2xFLAG-c and pMY-EGFP vectors to produce pcDNA3.1-hyg-hCeacam1L, pcDNA3-hCeacam1L-2xFLAG-c and pMY-EGFP-hCeacam1L. The following oligonucleotide DNA primers were synthesized: For the full-length human ceacam1L, the 5′ primer was 5′-AGCTAGCGCCACCATGGGGCACCTCTCAGCCCC-3′, and the 3′ primer was 5′-ACTCGAGTTACTGCTTTTTTACTTCTGAATA-3′; for the FLAG-tagged human ceacam1L, the 5′ primer was 5′-AGAATTCGCCACCATGGGGCACCTCTCAGCCC-3′, and the 3′ primer was 5′-ACTCGAGCTGCTTTTTTACTTCTGAATAAATTAT-3′. Mouse ceacam1L cDNA was also cloned and inserted into the pcDNA3-2xFLAG-c vector to produce pcDNA3-mCeacam1L-2xFLAG-c. The following oligonucleotide DNA primers were synthesized: the 5′ primer was 5′-AGAATTCGCCACCATGGAGCTGGCCTCAGCACA-3′, and the 3′ primer was 5′-ACTCGAGCTTCTTTTTTACTTCTGAATAAAC-3′.

To knockdown human and mouse ceacam1, short-hairpin (sh) sequences were generated using InvivoGen′s siRNA Wizard (HYPERLINK “http://www.sirnawizard.com/” http://www.sirnawizard.com/). These sh sequences were inserted into a psiRNA-h7SKhygro G1 expression vector (InvivoGen) to produce psiRNA-h7SKhygro-mceacam1sh and psiRNA-h7SKhygro-hceacam1sh. The knockdown efficiency of these vectors was analyzed by Western blotting (Supplementary Fig. S3). The sh target sequences for human and mouse ceacam1 were 5′-ACCTCGGATGGCAACCGTCAAATTGTTCAAGAGACAATTTGACGGTTGCCATCCTT-3′ and 5′-ACCTCGGGAAACACTACGGCTATAGATCAAGAGTCTATAGCCGTAGTGTTTCCCTT-3′, respectively. The control sh target (egfp) sequence was 5′-GCAAGCTGACCCTGAAGTTCA-3′.

To construct the FGC1L expression vector, we amplified a portion of the mouse GCSFR extracellular domain (AA309-626) from pBOS-I62 (a kind gift from Dr. Shigekazu Nagata of Kyoto University) and the transmembrane domain-cytoplasmic tail of Ceacam1L from pcDNA3.1-hyg-hCeacam1L as described previously (7). We then inserted these fragments into p3XFLAG-CMV-9 vector (SIGMA) to produce the p3XFLAG-GCSFR-Ceacam1L cytoplasmic tail (pFGC1L). The following oligonucleotide DNA primers were synthesized: For the mouse GCSFR extracellular domain, the 5′ primer was 5′-TGCGGCCGCGCGATGCATTCGCTCATCTCTG-3′, and the 3′ primer was 5′-AAGATCTCTCGAGTTACTGCTTTTTTACTTCTGAATA-3′; for the transmembrane domain-cytoplasmic tail of Ceacam1L, the 5′ primer was 5′-TCCATCTGACATTGTGATTGGAGTAGTGGC-3′, and the 3′ primer was 5′- ACTCGAGTTACTGCTTTTTTACTTCTGAATA-3′.

FGC1S was amplified from pFGC1L and inserted into the p3XFLAG-CMV-9 vector to produce the p3XFLAG-GCSFR-Ceacam1S cytoplasmic tail (pFGC1S). The following oligonucleotide DNA primers were synthesized: the 5′ primer was 5′-TGCGGCCGCGCGATGCATTCGCTCATCTCTG-3′, and the 3′ primer was 5′-AAGATCTCTCGAGTCATTGGAGTGGTCCTGAGCT-3′.

The nucleotide sequences of cloned cDNA were verified using the BigDye Terminator Kit version 3.1 (Applied Biosystems) and ABI sequencer model 3130xl (Applied Biosystems).

We transfected cells with the vectors using either the Nucleofector device according to the supplier's instructions (Lonza) or polyethylenimine, as previously described (7, 16).

To mark the transplanted hGICs in vivo, the cells were transfected with pMY-EGFP or pMY-EGFP-hCeacam1L vector as described previously (7, 8). The GFP-expressing hGICs (hereafter, hGICs) and ceacam1L-overexpressing hGICs were suspended in 5 μL of culture medium and injected into the brains of 5- to 8-week-old female NOD/SCID mice that had been anesthetized with 10% pentobarbital. The stereotactic coordinates of the injection site were 2 mm forward from the lambda, 2 mm lateral from the sagittal suture, and 5 mm deep.

Mouse brains were dissected, fixed in 4% paraformaldehyde overnight, transferred to 70% ethanol, processed on Tissue-Tek VIP (Sakura Finetek Japan), and embedded in paraffin. Coronal sections (6-μm thick) from the cerebral cortex were prepared on a microtome and stained with hematoxylin–eosin (H&E).

Two thousand cells were cultured in 100 μL of the culture medium in each well of a 96-well plate. To examine cell proliferation, the MTT assay was performed as follows. Ten microliters of MTT (5 mg/mL; Nacalai Tesque) was added to each well on days 0, 2 or 3, and 4 in vitro. The cells were incubated for 4 hours, the medium was replaced with 100 μL of DMSO, the cells were dissociated, and cell proliferation was quantified on a Benchmark microplate reader (Bio-Rad) with the absorption spectrum at 570 nm.

We performed a soft-agar assay to determine whether transfected cells proliferated in an anchorage-independent manner. The transfected cells were suspended in 0.3% top agar made with the optimized medium and layered onto 0.6% bottom agar made with the same medium. After the top agar had polymerized, culture medium was added and the cells were cultured for 20 days with medium changes every 3 days.

Total RNA was extracted from mouse NSCs and NSCL61 using the TRizol Plus RNA Purification System (Invitrogen). Purified RNA was then amplified and labeled with Cyanine 3 (Cy3) using the one-color Agilent Low Input Quick Amp Labeling Kit (Agilent Technologies) following the manufacturer's instructions. Labeled cRNAs were fragmented and hybridized to the Agilent mouse GE 8 × 60 K Microarray. After washing, microarrays were scanned with an Agilent DNA microarray scanner. Intensity values for each scanned feature were quantified using Agilent feature extraction software, which performed background subtractions.

Normalization was achieved using Agilent GeneSpring GX version 11.0.2. After normalization, hierarchical sample clustering of the expressed genes (DEGs) was performed with the Euclidean distance and average linkage methods (Agilent GeneSpring GX).

Accession number for microarray data: http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE70023.

Dissected mouse brains were fixed in 4% paraformaldehyde at 4°C overnight. After fixation, the brains were cryoprotected with 12% to 18% sucrose in PBS and embedded in Tissue-Tek OCT compound (Miles). Coronal sections (6-μm thick) were prepared from the cerebral cortex and stained with H&E using a standard technique or immunolabeled for GFP and either the active form of caspase-3 to detect dying cells or Ki67, a marker for proliferating cells.

Immunoprecipitation was performed as previously described (17). Cell lysates were incubated with Protein G sepharose (GE Healthcare) and the anti-FLAG (10 μm/mL) antibody for 4 hours at 4°C. The mixtures were centrifuged, and the precipitants were triple-washed and analyzed by Western blotting.

Western blotting was performed as previously described (17). The blotted membranes were probed with anti-Nestin (1:1000; BD Pharmingen, 1:200; Chemicon for human cells), anti-GFAP (1:1000; BD Pharmingen, 1:200; Chemicon for human cells), rabbit anti-STAT3 (1:1000; Santa Cruz Biotechnology), rabbit anti–phospho-STAT3 (1:1,000; Cell Signaling Technology), rabbit anti-PTPN6 (1:1,000; Cell Signaling Technology), rabbit anti-PTPN11 (1:1,000; Cell Signaling Technology), rabbit anti–phospho-PTPN11 (1:1,000; Cell Signaling Technology), rabbit anti–c-Src (1:1,000; Cell Signaling Technology), rabbit anti–phospho-c-Src (1:1,000; Cell Signaling Technology), or a mouse anti-GAPDH antibody (1:1,000; Chemicon). An ECL system (Amersham) was used for detection.

Survival data were analyzed for significance by Kaplan–Meier methods using GraphPad Prism version 4 software (P values were calculated with the log-rank test). All experiments were conducted more than three times with similar results.

We confirmed that the expression of ceacam1L was higher in NSCL61, OPCL61, and hGICs, E3 and E6 that were prepared from human GBM tissues, than in mouse parental cells (p53-deficient mNSCs and mOPCs) or normal human NSCs (hNSC; Fig. 1A; refs. 7, 9). An immunocytochemical analysis revealed that more than 70% of cultured human and mouse GICs were positive for Ceacam1 (Fig. 1B). Ceacam1L was more prominently expressed in GBM than in other malignant gliomas, anaplastic astrocytoma (AA), anaplastic oligo-astrocytoma (AOA), or anaplastic oligodendroglioma (AO; Supplementary Fig. S1A). Although ceacam1S was expressed in mouse and human GICs, the expression level of ceacam1S was significantly lower than that of ceacam1L in hGICs (Supplementary Fig. S1B and S1C). Furthermore, more than 70% of Ceacam1-positive cells were coimmunolabeled for the well-known GIC marker CD15 in the periphery of human GBM (Fig. 1C; ref. 18), whereas all Ceacam1⁺ cells were positive for EGFR, which is frequently amplified in human GBM (Fig. 1D; refs. 1, 13). Flow cytometric analysis showed that 3.2% of freshly prepared GBM cells were Ceacam1^high and more than 90% of Ceacam1^high cells (2.9% in total) were also positive for CD15 (Fig. 1E). Taken together, these results suggested that Ceacam1 was a novel GIC marker.

We analyzed the function of Ceacam1L in GICs. Using ceacam1L-specific shRNA (ceacam1sh) expression vectors (Supplementary Fig. S2), we found that the depletion of Ceacam1L inhibited the proliferation of human and mouse GICs (Fig. 2A) as well as their colony-forming ability in soft agar (Fig. 2B). In contrast, ceacam1L-overexpressing hGICs enhanced their colony-forming activity in soft agar (Fig. 2C) and killed mice more quickly than their parental cells (Fig. 2D). H&E staining revealed that ceacam1L-overexpressing hGICs formed larger tumors than parental hGICs at 40 days after injection and induced massive hemorrhages in tumors (Fig. 2E), consistent with the previous finding in which Ceacam1 acted as an angiogenesis-inducing extracellular factor (19). We further found that the tumors formed by ceacam1L-overexpressing hGICs contained more Ki67⁺ proliferating cells, but less activated caspase-3+ dying cells than control tumors (Fig. 2F and G). These results indicated that Ceacam1L regulated GIC tumorigenesis intracellularly and tumor angiogenesis extracellularly.

To identify the molecular mechanism regulated by Ceacam1L, we compared the gene expression profile of ceacam1L-overexpressing NSCs with that of NSCs, in which the endogenous expression of ceacam1L was undetectable, and found that 1,347 genes were upregulated whereas 1,286 were downregulated in ceacam1L-overexpressing NSCs (Supplementary Fig. S3A). Of these, we noted that the overexpression of ceacam1L strongly induced the expression of STAT3 target genes, including glial fibrillary acidic protein (GFAP), suppressor of cytokine signaling 3 (SOCS3), S100β, angiopoietin 1 (ANGPT1), and angiotensinogen (AGT), in NSCs (Fig. 3A). We confirmed that the overexpression of Ceacam1L induced the nuclear translocation of the phosphorylated form of STAT3 (p-STAT3) and GFAP expression in NSCs (Fig. 3B–D). These results indicated that Ceacam1L activated the STAT3 signaling pathway in NSCs.

We also compared the gene-expression profile of NSCL61 with that of ceacam1sh-overexpressing NSCL61, and found that 4,864 genes were upregulated whereas 3,984 were downregulated in ceacam1sh-overexpressing NSCL61 (Supplementary Fig. S3A). We noted a significant downregulation in the expression of stemness-related genes, including Notch factors (notch3, notch4, hey1, and dll1), aldehyde dehydrogenases (aldh-1a1, -1a3, -1a7, -1l1, -1l2, and -3b1), SRY-box transcription factors (sox-1, -2, -3, -4, and -9), ATP-Binding Cassette (ABC) transporters (abc-a1, -a3, -a7, -b8, -c3, -c5, -c10, -d1, and -g4), angpt1, egfr, chemokine receptor 4 (cxcr4), patched 1, and stat3 in ceacam1sh-overexpressing NSCL61 (Supplementary Fig. S3B). Because STAT3 is a well-known important factor for both stemness and tumorigenesis (20–23), we determined which genes were potential Ceacam1L/STAT3 targets using the Champion ChiP Transcription Factor Search Portal based on the SABiosciences' proprietary database (http://www.sabiosciences.com/chipqpcrsearch.php), and found that dll1, hey1, notch (3 and 4), sox (1, 2, and 9), abc (a1, b8, and c5), angpt1, egfr, cxcr4, patched 1, and stat3, contained the putative STAT3-binding sites (Supplementary Fig. S3B). We confirmed that the overexpression of dnSTAT3 as well as that of ceacam1sh decreased the expression of sox1, aldh1a1, cxcr4, dll1, notch3, and hey1 in hGICs (Fig. 4A and B). We verified that the overexpression of dnSTAT3 inhibited GFAP expression in E3 cells, even when cultured in differentiation medium with 10% FCS (Fig. 4C), and E3 proliferation in NSC medium (Fig. 4D).

We then examined whether Ceacam1L was indeed involved in the stemness maintenance in GICs. We found that the overexpression of Ceacam1L strongly increased the SP (9.0% ≥ 30%; Fig. 5A), whereas that of ceacam1sh abolished the population in hGICs (9.3% ≥ 1.2%; Fig. 5B). The overexpression of Ceacam1L upregulated the expression of GFAP in NSC medium and that of Nestin in 1% FCS medium, whereas the knockdown of Ceacam1 downregulated the expression of Nestin in NSC medium (Fig. 5C and D). The expression level of neuronal (Tuj1) and oligodendrocyte (GC) markers did not change by either Ceacam1L overexpression or knockdown. We verified that ceacam1L-overexpressing hGICs kept the neurosphere formation activity even when cultured sparsely, whereas the parental hGICs lost their self-renewal activity in the same condition (Supplementary Fig. S4). Furthermore, we found that ceacam1L overexpression made hGICs to be resistant to the temozolomide, which is a standard anti-GBM medicine (Supplementary Fig. S5): Although the blood concentration of temozolomide is 50 to 200 μmol/L for GBM patients, the ceacam1L-overexpressing hGICs survived even in the presence of 1 mmol/L temozolomide. Together with data that Aldh1a1, one of Ceacam1L-downstream factors, mediated temozolomide resistance (24), these findings suggested that Ceacam1L confers temozolomide resistance in GICs through the induced expression of Aldh1a1.

Ceacam1L was previously shown to activate the signaling factors, c-Src and two protein tyrosine phosphatases, nonreceptor type 6 and 11 (PTPN6 and 11, also known as SHP-1 and -2, respectively), upon phosphorylation of the Ceacam1L-cytoplasmic tail by various types of tyrosine kinase receptors (10, 11, 25). In turn, c-Src and PTPN6 then activated and inactivated STAT3, respectively, while PTPN11 modulated the Ras, NF-κB, and EGFR signaling pathways positively, and the JAK–STAT pathway negatively (26–29). We found that hGICs expressed ptpn11 and c-Src, but not ptpn6 (Fig. 6A), and that the phosphorylated forms of PTPN11 (p-PTPN11) and c-Src (p-c-Src) bound to Ceacam1L in hGICs (Fig. 6B), indicating that the Ceacam1L-dependent negative feedback signal, which blocks STAT3 activation, was abolished in hGICs.

We evaluated whether c-Src activated STAT3 in hGICs. The overexpression of a constitutive active form of c-Src (caSrc) not only increased STAT3 phosphorylation and GFAP expression, but also enhanced cell proliferation, whereas the overexpression of a dominant negative form of c-Src (dnSrc) inhibited these inductions (Fig. 6C and D). These results suggested that Ceacam1L regulated GIC proliferation through the activation of c-Src/STAT3 signaling.

Ceacam1L exists as a monomer, cis-/trans-homodimer, or cis-/trans-heterodimer with Ceacam1 splicing variants or other Ceacam family members; therefore, the intracellular domain of Ceacam1L is a monomer or homodimer, depending on the circumstances (10, 11). To determine which form of the Ceacam1L cytoplasmic tail activated c-Src/STAT3 signaling, we constructed two chimera proteins, FGC1L and FGC1S, consisting of FLAG, a part of the extracellular domain of the granulocyte colony-stimulating factor receptor, the Ceacam1-transmembrane domain, and either the cytoplasmic tail of Ceacam1L or Ceacam1S, respectively (Fig. 7A). We established FGC1L- and FGC1S-expressing hGIC and NSC lines, cultured them in the presence or absence of an anti-FLAG antibody, and then examined the phosphorylation of c-Src and STAT3 as well as the expression of GFAP. We found that p-c-Src, p-STAT3, and GFAP levels were increased in FGC1L-expressing hGICs in the absence of the antibody, whereas these inductions were not detected in FGC1L-expressing hGICs treated with the antibody or FGC1S-expressing cells (Fig. 7B). We confirmed these results using FGC1L- and FGC1S-expressing NSC lines (Fig. 7C). We demonstrated that FGC1L associated with p-c-Src and p-PTPN11 in hGICs in the absence of the antibody (Aggregation−), whereas these associations were abolished in antibody-treated FGC1L-expressing cells (Fig. 7D). We also confirmed these results using FGC1L-expressing NSC lines (Fig. 7E). Thus, these results revealed that the monomeric Ceacam1L cytoplasmic tail activated c-Src/STAT3 signaling in GICs (Fig. 7A).

Ceacam1 is known to be involved in various biologic functions, including proliferation, angiogenesis, tumorigenesis, and inhibition of both cytokine production and cytotoxic activity of immune cells, as an intracellular and intercellular factor (19, 30). We showed that Ceacam1L-overexpressing hGICs formed larger colonies in soft agar and tumors with massive hemorrhaging in the brains of immunodeficient mice, whereas the knockdown of Ceacam1 blocked GIC proliferation. In addition to previous findings in which Ceacam1 acted as a major effector of VEGF-induced angiogenesis (19, 31), we revealed that Ceacam1L regulated the expression of Angpt1, IL18, and secretogranin II, all of which play an important role in vascular development and angiogenesis in GICs (32–34). Because neovascularization is not only a common characteristic of GBM, but has also been correlated with poor outcomes, these findings indicate that Ceacam1L is an indispensable therapeutic target for GBM.

Ceacam1L is a highly glycosylated protein that carries a Lewis X (LeX) structure, also known as CD15 and SSEA1 (35, 36), which is a well-known stem cell marker expressed on many types of normal stem cells, including embryonic stem cells, bone marrow stem cells, NSCs (37–39), and GICs (18). We herein confirmed that more than 90% of CD15^high GICs were positive for Ceacam1 when GICs were immunolabeled in the presence of TritonX-100, suggesting that CD15-carrying Ceacam1 mainly existed in the cytoplasm. This may also explain why CD15-negative glioma cells self-renewed and formed tumorspheres similar to CD15-positive glioma cells (40). Because Ceacam1L was previously shown to be expressed at low levels on the surface of resting T cells and was quickly mobilized from the intracellular compartment to the cell surface following T-cell activation (41), the molecular mechanisms regulating Ceacam1L mobilization in GICs need to be elucidated in more detail.

We found that Ceacam1L positively regulated the expression of a number of stemness-related genes in GICs both directly and indirectly. Ceacam1L controlled the expression of NSC factors (dll1, notch3, hey1, and Sox1-4, 9) and astrocyte markers (GFAP and S100β) in GICs, suggesting that Ceacam1L⁺ GICs may retain the characteristics of radial glial cells or subventricular astrocytes, both of which behave as multipotential NSCs in the adult brain (42, 43). We also found that ceacam1L overexpression kept self-renewal capability in hGICs. Because Ceacam1L regulated the expression of many ABC transporters and ALDH family members in GICs, Ceacam1L appears to widely govern drug resistance in GICs. Indeed, we found that the ceacam1L-overexpressing hGICs were much more resistant to temozolomide, the standard medicine for GBM, than the parental hGICs. This is consistent with a finding that Aldh1a1, one of downstream factors of Ceacam1L, mediated temozolomide resistance in GBM, although the detail mechanism remained elusive (25). Furthermore, FGC1L-overexpressing NSCs formed larger spheres in cultures and continued to express Nestin even in differentiation medium with 1% FCS. These results suggest that Ceacam1L contributed to the maintenance of stemness in GICs. It is unlikely that STAT3, a crucial Ceacam1L signaling factor, predominated the expression of all Ceacam1L-downstream genes; therefore, the transcription factor, other than STAT3, that controls the expression of stemness-related genes needs to be identified.

When phosphorylated by EGFR, Ceacam1L was shown to activate c-Src, PTPN6, and PTPN11 and also enhance or prevent the proliferation signals of the receptor in a cell type–dependent manner (11, 44, 45). We unexpectedly found that the expression of PTPN6, which prevents the JAK-dependent STAT3 signaling pathway (46), was silenced in hGICs, suggesting that a Ceacam1L-dependent negative feedback signal to EGFR was abolished in hGICs. Taken together with the finding that PTPN11 and c-Src activated EGFR and STAT3 signals, respectively (21, 27, this manuscript), our results suggest that Ceacam1L acted as an amplifier of both EGFR and STAT3 signaling in hGICs.

We demonstrated that the monomeric Ceacam1L-cytoplasmic tail activated c-Src/STAT3 signaling, whereas its oligomerization abolished this activity. The state of Ceacam1L (monomer or oligomer) in GICs may be dependent on their circumstances, in which GIC-associating (niche) cells, such as immune cells and microglia, may express Ceacam1L-binding partners, including Ceacam1L, 1S, soluble forms of Ceacam1, and Ceacam5 (10–12). The finding that lung metastasis by Ceacam1⁺ B16F10 melanoma cells was significantly decreased in Ceacam1-knockout mice revealed that the expression of Ceacam1 on tumor-associating cells was essential for successful tumorigenesis (47). In fact, we found the association of Ceacam1⁺/EGFR⁺ cells and CD68⁺ activated microglia in human GBM tissues (Supplementary Fig. S6). It has been shown that CD68⁺ cells increased in high-grade gliomas (48), and that CD68⁺ infiltrating microglias/macrophages were involved in gliomagenesis (49). Together, these suggest that CD68⁺ microglias/macrophages play a crucial part to maintain/activate GICs in the niche. Therefore, next crucial challenge is to investigate how Ceacam1L-dependent intercellular association between GICs and their surrounding cells regulates GIC maintenance and tumor-niche formation.

No potential conflicts of interest were disclosed.

Conception and design: S. Kaneko, Y. Nakatani, T. Kondo

Development of methodology: Y. Nakatani, T. Ohnishi, T. Kondo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Kaneko, Y. Nakatani, T. Takezaki, T. Hide, D. Yamashita, N. Ohtsu, S. Terasaka, T. Kondo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Kaneko, Y. Nakatani, D. Yamashita, N. Ohtsu, T. Kondo

Writing, review, and/or revision of the manuscript: S. Kaneko, Y. Nakatani, T. Kondo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Nakatani, T. Takezaki, T. Hide, D. Yamashita, T. Ohnishi, T. Kondo

Study supervision: K. Houkin, T. Kondo

The authors thank Shigekazu Nagata for mouse GCSFR cDNA, Shizuo Akira for the mouse dominant-negative form of STAT3 (dnSTAT3) cDNA, Chitose Oneyama for the c-Src-expression vectors (caSrc and dnSrc), and Yoshihiro Tsukamoto for a temozolomide assay.

This work was supported, in part, by a research program of the Project for Development of Innovative Research on Cancer Therapeutics (P-DIRECT), Ministry of Education, Culture, Sports, Science and Technology of Japan (T. Kondo).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.