Specialized, differentiated cells often perform unique tasks that require them to maintain a stable phenotype. Multiciliated ependymal cells (ECs) are unique glial cells lining the brain ventricles, important for cerebral spinal fluid circulation. While functional ECs are needed to prevent hydrocephalus, they have also been reported to generate new neurons: whether ECs represent a stable cellular population remains unclear. Via a chemical screen we found that mature ECs are inherently plastic, with their multiciliated state needing constant maintenance by the Foxj1 transcription factor, which paradoxically is rapidly turned over by the ubiquitin-proteasome system leading to cellular de-differentiation. Mechanistic analyses revealed a novel NF-κB-independent IKK2 activity stabilizing Foxj1 in mature ECs, and we found that known IKK2 inhibitors including viruses and growth factors robustly induced Foxj1 degradation, EC de-differentiation, and hydrocephalus. Although mature ECs upon de-differentiation can divide and regenerate multiciliated ECs, we did not detect evidence supporting EC’s neurogenic potential.
The adult mammalian brain has limited intrinsic capacity to regenerate. While there are resident neuronal and glial progenitors with restricted proliferative potentials, the vast majority of cell types in the adult brain are terminally differentiated. Different subpopulations of mature neurons and glia perform specific tasks, exhibit complex/distinct cellular morphologies, no longer self-renew or divide, and are generally considered the basal differentiated states for the cells. Multiciliated ependymal cells (ECs) are specialized epithelial cells lining the mammalian brain ventricles throughout, bathed by cerebral spinal fluid (CSF) along their apical surfaces1. ECs are derived from Nestin+ radial glial progenitors during their terminal differentiation in development2,3,4,5,6,7, temporally coinciding with a transition from embryonic to postnatal neurogenesis along the lateral brain ventricles8. During radial glial differentiation to a distinctive multiciliated phenotype, ECs acquire specialized interdigitating lateral membranes4,9, with planar polarity cues controlling multicilia beating orientations on their cellular surfaces10,11. Although ECs form perhaps the largest epithelial surface in the brain, their specific functions and contributions to central nervous system (CNS) physiology are poorly understood. Genetic mutant mouse models have shown that ependymal disruptions during development can result in hydrocephalus/ventriculomegaly12,13, as well as postnatal neurogenic niche assembly defects4. Hydrocephalus remains the most prevalent form of developmental CNS malformation, though interestingly inheritable genetic mutations in human hydrocephalus are relatively rare, while trauma, intraventricular hemorrhages, and CNS infections account for most cases14.
ECs were once believed to be the source of new neurons in the postnatal mammalian brain15,16. However, in recent years, this notion has been revised to an understanding that mature ECs retain the capacity to generate newborn neurons under certain environmental stimuli, such as brain injury17 or growth factor exposure18, but otherwise remain in a non-proliferative state. Since the adult mammalian CNS has limited capacity to regenerate after injury or degeneration, the notion that mature ependyma can serve as a neurogenic reservoir requires critical examination. At the heart of this long running debate on whether multiciliated ECs retain neurogenic capacity, is the phenotypic stability of a seemingly terminally differentiated multiciliated epithelial cell in the brain. However, this important and challenging problem has not been directly addressed.
The transcription factor Foxj1 belongs to the large forkhead domain DNA-binding protein family19. It is a critical regulator of motile-ciliated cellular development, including the primitive node20, multiciliated cells lining the trachea21, and the ependyma3,4. We showed previously that foxj1 expression is induced during terminal maturation of ventricular radial glial progenitors specified to becoming ECs4. And foxj1 deletion resulted in radial glial differentiation arrest and the complete lack of multiciliated ECs3,4. Performing a chemical screen for modulators of EC differentiation, we discovered unexpectedly that the mature multiciliated ependymal phenotype is not fixed, needing constant Foxj1 protein expression to prevent cellular de-differentiation back to a glial-like morphology. Paradoxically, we found that ependymal Foxj1 protein has a short half-life, requiring non-canonical IκB kinase (IKK) activity to prevent rapid degradation via the ubiquitin proteasome system (UPS). Armed with these novel molecular insights, we demonstrated that this previous unknown signaling pathway is directly targeted by serum and viruses to induce EC transformation, ventricular breakdown, and hydrocephalus. While de-differentiated mature ECs can divide and regenerate ECs, we did not find evidence to support their neurogenic capacity. We believe these findings provide a foundational basis for EC biology/brain ventricle stability research in health and disease moving forward.
Chemical screen for ependymal differentiation modulators
To identify molecular pathways that can modulate multiciliated EC differentiation from postnatal radial glial progenitors (pRGPs), we utilized an in vitro ependymal assay that we had previously developed4. Culturing primary progenitors from the lateral ventricular surface of newborn mice, we observed that roughly 52.4 ± 6.4% of these cells turned into multiciliated ECs during in vitro differentiation (Fig. 1a, b, Supplementary Fig. 1a). Consistent with in vivo markers of multiciliated EC differentiation5,6, p73 and c-Myb were also expressed during early ependymal culture differentiation and downregulated in mature ECs (Supplementary Fig. 1b). We reasoned that by altering culturing media composition and scoring percentages of multiciliated cells, this primary culture assay presented an excellent platform to screen for novel pathways regulating production/maintenance of mature ECs. Thus we initiated a chemical screen for potential small molecules that could further enrich multiciliated EC production during in vitro differentiation.
We have identified one compound MLN4924 that can consistently increase EC numbers in vitro. MLN4924 treatment during ependymal progenitor differentiation resulted in 81.1 ± 6.5% of cells in primary pRGP culture becoming multiciliated as assessed by acetylated-tubulin + DAPI immunohistochemistry (IHC) staining (Fig. 1a, b, Supplementary Fig. 1a). We also used Foxj1 transcription factor expression as a readout for the presence of ECs4, as multiciliated ECs in culture uniformly showed Foxj1 protein by IHC staining (Supplementary Fig. 1c). Consistent with the observed increase in multiciliated cells in culture, Foxj1 + DAPI IHC staining showed similarly robust increase in co-localization percentage in MLN4924-treated vs. control conditions (Fig. 1c, d, Supplementary Fig. 1d). IHC staining revealed increased Foxj1 fluorescence intensity in many cells compared to untreated controls (Fig. 1c), suggesting that MLN treatment during ependymal differentiation increased both total numbers of Foxj1+ cells, as well as Foxj1 protein content per cell. Interestingly, in fully differentiated ependymal cultures without prior MLN4924 treatment, acute MLN4924 addition for 12 h resulted in consistently higher overall level of Foxj1 IHC staining compared to untreated controls (Fig. 1e), further supported by western blot analyses (Fig. 1f, g, Supplementary Fig. 13). In freshly isolated mature ECs from P28 brain ventricular wall tissue (wholemount preparation), acute MLN4924 treatment also resulted in notable increase in Foxj1 IHC fluorescence level vs. untreated control (Fig. 1h). Activated Caspase-3 IHC staining showed no obvious differences between control and MLN4924-treated EC cultures during differentiation, and quantification of Foxj1 IHC staining fluorescence intensity in individual ECs showed a robust population increase in Foxj1 protein levels (Fig. 1i).
Inherent Foxj1 protein instability in mature ependymal cells
MLN4924 is a neddylation inhibitor that specifically blocks cullin-RING family E3 ubiquitin ligases22, currently under human clinical trials for cancer therapy23. Since MLN4924 increased Foxj1 antibody immunofluorescence in ECs (Fig. 1e, g, h), and concurrently we did not detect obvious changes to Foxj1 mRNA levels following MLN4924 treatment (Supplementary Fig. 2a), we wondered whether MLN4924 may be blocking Foxj1 protein turnover. The baseline stability of ependymal Foxj1 protein is unknown, so to test this, we first treated differentiated primary ependymal cultures with cycloheximide (CHX) to block new protein synthesis. This resulted in rapid disappearance of Foxj1 protein at 3 h, complete by 6 h after treatment onset (Fig. 2a, b, Supplementary Fig. 13). As controls, CHX-mediated degradation of Cyclin D1 and βIII-tubulin proteins in ependymal cultures occurred at similar rates to those previously reported24,25 (Supplementary Fig. 2b). Proteasomal blocker MG132 (Fig. 2c, Supplementary Fig. 13), as well as MLN4924 (Fig. 2d, Supplementary Fig. 13), stabilized Foxj1 in the presence of CHX, implicating ubiquitin proteasome-mediated protein turnover. Immunoprecipitating (IP) Foxj1 from differentiated primary ependymal cultures and western blotting for Foxj1, we noticed increased laddering of Foxj1 in the presence of MG132 (Fig. 2e, Supplementary Fig. 13). We next performed IP experiments in ependymal cultures using (1) anti-ubiquitin antibody for IP then blotting with anti-Foxj1 antibody (Fig. 2f, Supplementary Fig. 13), and in reverse (2) IP-ing with anti-Foxj1 antibody then blotting with anti-ubiquitin antibody (Fig. 2g, Supplementary Fig. 13). They also showed consistent laddering of Foxj1 protein in the presence of MG132 after 6 h incubation (Fig. 2f, g, Supplementary Fig. 13). CHX treatment of acutely isolated ependymal wholemounts from P28 animals showed a similarly rapid disappearance of Foxj1 protein in mature ECs, which can be blocked by MLN4924 (Supplementary Fig. 2c). These results revealed that the Foxj1 transcription factor is rather unstable in the mature ependyma, constantly targeted for degradation by the UPS.
We next wanted to determine the molecular pathways regulating Foxj1 protein turnover. To uncover ubiquitin ligase(s) mediating Foxj1 degradation, we reasoned that a specific inhibitor of the relevant ligase should result in increased ependymal Foxj1 protein level. We screened known chemical blockers of ubiquitin ligases in differentiated primary ependymal culture and found that in addition to MLN4924, SMER3 was somewhat effective in increasing Foxj1 protein in ependymal cultures (Fig. 2h, i, Supplementary Fig. 13). SMER3 is a known inhibitor of yeast cullin-RING family ubiquitin ligase SCFMET3026. Based on sequence homology comparisons, the closest mammalian homologs of yeast SCFMET30 are cullin-RING ligase family members Fbxw5, Fbxw7, and Fbxw1127,28,29. IHC staining for Foxj1 protein localization in differentiated ependymal cultures during CHX treatment showed concentration inside the nucleus (Fig. 2b). To rule out the potential of a modified/cytoplasmic form of Foxj1 protein not recognized by the Foxj1 antibody, we utilized a previously generated Myc-tagged version of the Foxj1 protein4. Lentiviral-mediated Foxj1-Myc expression in primary ependymal cultures, now stained with anti-Myc instead of Foxj1 antibody, showed similar nuclear localization and turnover following CHX treatment (Supplementary Fig. 2d). Using antibodies specific to Fbxw family members, we found that Fbxw5 was localized in the nucleus of differentiated ECs in vitro and in vivo (Fig. 2j, Supplementary Fig. 2e). To assay the functions of Fbxw5 in Foxj1 protein turnover, we utilized shRNAs that can efficiently knockdown Fbxw5 expression (Supplementary Fig. 2f). Lentiviral-mediated shRNA knockdown of Fbxw5 in primary ependymal cultures specifically blunted CHX-induced Foxj1 degradation (Fig. 2k, Supplementary Fig. 13), indicating Fbxw5’s functional contribution to ubiquitin-proteasome-mediated Foxj1 turnover. Consistent with this notion, Foxj1 immunoprecipitated with Fbxw5 in primary ependymal cultures, enhanced by MG132-mediated proteasomal blockade (Fig. 2l, Supplementary Fig. 13). And Fbxw5 robustly enhanced Foxj1 ubiquitination level when co-transfected into HEK293 cells (Fig. 2m, Supplementary Fig. 13).
Mature EC phenotype continues to require foxj1 expression
These findings are somewhat surprising, namely that the key transcription factor required for terminal differentiation of multiciliated ECs is itself unstable. To test whether sustained Foxj1 protein expression is required to maintain the mature cellular state after differentiation, we set out to inducibly delete Foxj1 from mature ECs. To visualize individual multiciliated ECs, we took a tamoxifen-inducible CreER genetic approach, combined with Rosa26-tdTomato (R26R-tdT) reporter mice, to generate tdTomato-labeled ECs. The 1 kb FOXJ1 human promoter element has been used extensively over the past decade to label/target multiciliated cells in mice30,31,32,33, including brain ECs3,4,17. To exclude potential mis-labeling of non-EC types by the FOXJ1-CreERt2 transgenic driver31, we used homologous recombination to knock-in CreERt2 recombinase into the endogenous foxj1 gene locus (Fig. 3a). While this fusion into the ATG start codon for foxj1 disrupted the native gene, the resulting heterozygous foxj1CreERt2/+ mice were viable, fertile, and showed no obvious abnormalities, consistent with our previous observations in foxj1KO/+ heterozygous mice4. In foxj1CreERt2/+; R26R-tdT mice, tamoxifen induction in vivo at P14, analyzed 2 weeks later, showed specific tdTomato expression in multiciliated cells (Fig. 3b, c, Supplementary Fig. 3), confirming efficient and specific ependymal targeting by the foxj1CreERt2/+ driver.
We next crossed the foxj1CreERt2/+ driver to our previously generated foxj1Flox/+ conditional allele4 to produce foxj1CreERt2/Flox mice. R26R-tdT was also crossed into the background to enable lineage-tracing experiments. These foxj1CreERt2/Flox heterozygous mutant mice, similar to foxj1KO/Flox mice that we generated previously4, showed no obvious phenotypic abnormalities in the absence of tamoxifen injection. In foxj1CreERt2/Flox; R26R-tdT conditional mutant mice (cKO), tamoxifen induction in vivo at P14, analyzed 2 weeks later showed highly effective removal of Foxj1 protein in tdTomato+ cells (Supplementary Fig. 4a). Upon foxj1 inducible-deletion, these tdTomato+ mutant cells lost multicilia and their associated basal bodies (Fig. 3c), as well as aberrantly upregulated Glast2,4 and GFAP expression (Fig. 3d, Supplementary Fig. 4b, c). These results are in contrast to control foxj1CreERt2/+; R26R-tdT mice induced with the same tamoxifen schedule, showing tdTomato labeling in multiciliated Glast−, GFAP−, Foxj1+ ECs (Fig. 3c, d). More strikingly, while control ECs exhibited epithelial morphology, Foxj1 cKO mutant ECs extended complex basal cellular processes (Fig. 3d–f), terminating on CD31+ vascular endothelial cells (Fig. 3g), showing morphological features of subependymal glial progenitors34. Nissl staining of P28 Foxj1 cKO mutant brain sections, tamoxifen-induced at P14, showed clear ventriculomegaly/hydrocephalus (Supplementary Fig. 4d, e). Repeating these experiments via P30 tamoxifen induction, analyzed 2 weeks later, showed the same Foxj1 cKO mutant EC morphological changes, loss of multicilia and their associated basal bodies, as well as ventriculomegaly/hydrocephalus (Supplementary Fig. 5). The presence of Foxj1+ multiciliated ECs (not targeted by CreER) in tamoxifen-injected mutant mice showed that these phenotypic changes seen in Foxj1-deleted ECs was not induced by hydrocephalus (Fig. 3e, Supplementary Fig. 5a, e). These results demonstrated that continued Foxj1 protein expression is required for mature ECs to maintain their terminally differentiated multiciliated state, preventing hydrocephalus. They also showed that remarkably, the tissue stability of the ependyma is controlled by the turnover homeostasis of a single transcription factor.
Non-canonical IKK2 control of EC Foxj1 protein stability
While MLN4924 can effectively block ependymal Foxj1 protein degradation, which biological pathways may accelerate this process? We reasoned that ependymal changes during MLN4924 treatment may reveal the relevant cellular machineries. We therefore performed transcriptome profiling, comparing MLN4924-treated vs. untreated primary ependymal cultures using expression microarrays. To examine molecular signatures, ependymal cultures were harvested for RNA 36 h following start of DMSO-only vs. MLN4924-treated conditions. Western blot analysis was performed concurrently to validate Foxj1 protein stabilization by MLN4924 treatment. Unsupervised hierarchical clustering clearly segregated transcriptional profiles based on treatment groups (Fig. 4a), suggesting a common cellular consequence in response to drug treatment. MLN4924 resulted in the upregulation of 899 genes and downregulation of 804 genes (>2-fold change, P < 0.001). Multiple genes in the dataset annotated as signal transduction regulators (Supplementary Fig. 6a), and the top five pathways implicated by gene ontology analysis included IKK/NF-κB signaling (Fig. 4b, c). MLN4924 is an effective inhibitor of cullin-RING E3 ligases that are important for cellular homeostasis in a variety of processes22,35; thus it was expected that “signal transduction regulation” would emerge as a top pathway in our analyses. In fact, there is extensive overlap in this gene cohort with those indicated in the IKK/NF-κB pathway (Fig. 4b, c, Supplementary Fig. 6a). While MLN4924 has been reported to alter IKK/NF-κB signaling in B-cell lymphoma cell line36, IKK/NF-κB functions in ECs as well as Foxj1 homeostasis are unknown.
The ependyma is in direct contact with the CSF, which is commonly used for diagnosing human CNS infections. The IKK/NF-κB pathway is a complex signaling hub, critically important in immune-mediated cellular responses37, thus we wondered whether it is involved in ependymal Foxj1 protein stability. We treated differentiated primary ependymal cultures with IKK-specific blockers, and found that remarkably, IKK2 inhibitor IMD-035438 resulted in rapid degradation of Foxj1 protein (Fig. 4d, e, Supplementary Fig. 14), in a similar time course to CHX treatment (Fig. 2a, b, Supplementary Fig. 13). To determine whether IMD-0354 is inducing Foxj1 degradation through UPS function, we co-added IMD-0354 with MG132 or MLN4924, and either were effective in blocking IMD-0354-induced Foxj1 degradation (Fig. 4e–g, Supplementary Fig. 14). With MLN4924 co-incubation, IMD-0354 inhibition of IKK2 did not result in major changes to Foxj1 protein levels (Fig. 4g, Supplementary Fig. 14), suggesting that potential MLN4924 modulation of IKK2 activity is not a critical contributor to maintaining EC Foxj1 protein stability. We found similar effects of IMD-0354-induced Foxj1 degradation on freshly isolated mature ependymal wholemounts (Supplementary Fig. 6b), followed by loss of multicilia consistent with Foxj1’s transcriptional regulation of multiciliogenesis (Supplementary Fig. 6c). To rule out non-specific IMD-0354 toxicity, 3 days after IMD-0354 treatment washout, we observed re-expression of Foxj1 protein in primary ependymal cultures (Supplementary Fig. 6d). Moreover, shRNA knockdown of IKK2 in primary ependymal cultures resulted in robust decrease of Foxj1 protein levels: in multiple experiments we did not detect appreciable Foxj1 staining in GFP+ IKK2 shRNA-targeted cells (Supplementary Fig. 7a–c). These results suggested that constitutive NF-κB pathway activation is required to stabilize ependymal Foxj1 protein in the physiological state, but surprisingly, SC75741 and JSH-23, compounds that specifically block NF-κB transcription factor activity did not result in ependymal Foxj1 degradation (Supplementary Fig. 7d).
The IKK complex is best known for its important role in NF-κB pathway activation39, primarily through subunit IKK2 kinase activity40,41, although NF-κB-independent/non-canonical IKK functions have also been described42. To determine whether IKK’s role in Foxj1 stability may be through direct protein phosphorylation independent of NF-κB activation, we first treated primary ependymal cultures with IMD-0354 for 1 and 2 h and performed optimized SDS-page protocol to detect molecular weight bands for Foxj1 protein. This resulted in clear distinction of two close Foxj1 bands (Fig. 4h, Supplementary Fig. 14), difficult to resolve using low-gradient SDS-page gels. Interestingly, the upper Foxj1 protein band diminished faster at the 2 h time point following IMD-0354 treatment (Fig. 4h, Supplementary Fig. 14). To determine if Foxj1 can be phosphorylated by IKK2, we performed co-transfection experiments using Myc-tagged Foxj1 + Flag-tagged IKK2 or IKK2-K44A kinase dead mutant43. When co-transfected into HEK293 cells which has low levels of endogenous IKK activity44, wild-type IKK2 resulted in two distinct Foxj1 protein bands, which changed to a single lower band with IKK2-K44A mutant (Fig. 4i, Supplementary Fig. 14) or IMD-0354 treatment (Fig. 4j, Supplementary Fig. 14), but not after JSH-23 treatment (Fig. 4j, Supplementary Fig. 14). Protein sequence analysis revealed a potential IKK2 phosphorylation site in Foxj1 that is highly conserved between species (Fig. 4k). Mutating the consensus serine32/35 sites in Foxj1 to alanine resulted in a lack of upper molecular weight protein band when co-transfected with IKK2 (Fig. 4l, Supplementary Fig. 14), indicating phosphorylation of Foxj1 by IKK2 activity. Test assay whether Foxj1 serine32/35 phosphorylation can stabilize the protein, we converted serine32/35 to aspartic acid, mimicking the phosphorylated state. Consistently, the aspartic acid mutations resulted in a more stable Foxj1 protein as compared to wild-type Foxj1 or the alanine mutations for serine32/35 (Fig. 4m, Supplementary Fig. 14). Co-IP experiments showed interactions between IKK2 and Foxj1, while IKK2 kinase activity and Foxj1 serine32/35 did not appear to be required for protein interactions (Fig. 4n, Supplementary Fig. 14).
Viral-induced Foxj1 degradation and ventricular breakdown
Having uncovered IKK2 regulation of EC Foxj1 stability, we next wanted to understand whether there are pathological conditions where IKK2 inhibition in vivo may induce hydrocephalus. There are important classes of pathogenic viruses that can suppress NF-κB activity in host cells via IKK2 inhibition45. HSV-1 is one such virus, which can result in devastating outcomes following CNS attack even if treated with antiviral therapy during the acute phase of infection46. It has been associated with hydrocephalus and ependymal ventricular wall damage47,48, although the underlying molecular pathogenesis for these outcomes remain largely unknown. Strikingly, within 12 h following HSV-1 infection of primary ependymal cultures, ECs uniformly lost Foxj1 protein expression (Fig. 5a). Consistent with our results above showing its effectiveness in blocking UPS-mediated Foxj1 protein turnover (Fig. 2d), MLN4924 was able to prolong EC Foxj1 protein expression and enhance cellular survival in HSV-1 infected primary ependymal cultures (Fig. 5a). Unlike HSV-1, cytomegalovirus (CMV) and Epstein-Barr virus (EBV) are two example NF-κB-activating viruses that can target the CNS45,49,50. We infected differentiated primary ependymal cultures with either CMV or EBV, and found no obvious changes to Foxj1 protein levels after infections (Supplementary Fig. 8a, b).
To determine whether HSV-1 can cause Foxj1 degradation in vivo, we injected HSV-1 into the lateral ventricles of P28 FOXJ1-GFP mice. The FOXJ1-GFP transgene30 is a useful tool for visualizing the entire ependymal lining instead of individual tamoxifen-induced R26R-tdTomato+ ECs. Three days after viral injection, GFP+ ependymal lining displayed dramatic epithelial breakdown, exposing the underlying brain parenchyma (Fig. 5b). MLN4924, similar to its effects in blunting HSV-1-induced Foxj1 degradation in primary ependymal cultures (Fig. 5a), when administered in vivo shortly after HSV-1 injection resulted in prolonged Foxj1 protein expression (Supplementary Fig. 8c), consistently reduced virally infected areas (Fig. 5c), and enhanced ventricular epithelial integrity (Fig. 5d). To examine whether induction of EC Foxj1 degradation by IKK-suppressing viruses may be generalizable, we next tested vaccinia virus which can also cause human viral encephalitis51. Vaccinia infection of differentiated EC cultures also resulted in Foxj1 loss, which can be blunted by MLN4924 (Supplementary Fig. 8d).
Serum-induced Foxj1 degradation and EC transformation
Growth factor receptor signaling has also been shown to inhibit IKK2 activity52,53. During in vitro primary culture, to successfully generate high percentages of multiciliated ECs from pRGPs, we needed to change the culturing media from high (10%) to low (2%) serum. Thus we reasoned it is possible that returning mature ECs to the high serum proliferation media may induce their transformation. To test this, we injected tamoxifen in foxj1CreERt2/+; R26R-tdT mice at P7 and harvested ventricular wholemounts at P14 when multiciliated ECs are mature, to perform ex vivo organotypic live-imaging experiments. Unlike in control conditions where ECs retained a rather stable morphology while incubated in 2% serum culturing media, ECs incubated in 10% serum proliferation media showed a striking change in cellular morphology from epithelial to radial glial-like shape over the course of 60 h (Fig. 6a–d, Supplementary Movie 1, 2). This morphological change observed from mature tdTomato+ ECs was relatively uniform (Fig. 6a–d). Post imaging immunohistochemical staining showed that while tdTomato+ ECs in control conditions remained GFAP− after 60 h of live-imaging, in proliferation media they became GFAP+ during their morphological transformation. Similarly, differentiated primary EC cultures treated with high serum proliferation media resulted in EC transformation and robust Foxj1 loss (Fig. 6e). MLN4924 addition to high serum proliferation media for 12 h was effective in sustaining EC Foxj1 protein expression (Fig. 6e), showing induced Foxj1 degradation via the cullin-RING ligase/UPS-dependent mechanism.
Assessing proliferation vs. neurogenic capacity of mature ECs
The de-differentiation phenotype of mature ECs raised a question of whether they can re-enter the cell cycle. To test this, we performed live-imaging on ependymal wholemounts from P14 foxj1CreERt2/+; R26R-tdT mice tamoxifen-induced at P7. While we did not detect tdTomato+ cell divisions over 3 days of live-imaging from multiple wholemounts cultured in control media conditions (which was expected since mature ECs did not de-differentiate), when cultured in proliferation media tdTomato+ ECs clearly re-entered the cell cycle (Fig. 7a, Supplementary Movie 3). In support, EdU addition to the culture media to label cells entering the cell cycle consistently marked tdTomato+/EdU+ cells in proliferation media, but not control media culture conditions (Fig. 7b). Cell division of de-differentiated multiciliated ECs raised the question of what type of cells may be generated upon their re-differentiation. Since mature ECs in primary ependymal cultures also showed cell cycle re-entry and division upon proliferation media-induced de-differentiation (Supplementary Movies 4–6), and that this primary culture can robustly survive long term in vitro culture, we EdU pulsed the mature ependymal cultures concurrent with media change to either proliferation or control media incubation for 4 days. As expected following EC de-differentiation there was dramatic reduction of multiciliated/Foxj1+ tdTomato+ cells following 4 days of proliferation media incubation, unlike control media samples with abundant multiciliated ECs (Fig. 7c, d, Supplementary Fig. 9a). However, at the end of 4 days of proliferation media exposure, changing the culturing condition back to differentiation media and incubating for 10 additional days, we observed re-differentiation of lineage-traced tdTomato+ cells into Foxj1+ multiciliated ECs (Fig. 7c–e). EdU pulse-labeling of multiciliated tdTomato+ ECs indicated cell cycle re-entry during proliferation media-induced de-differentiation (Fig. 7e). Following re-differentiation, we did not observe tdTomato+/DCX+ neuroblasts or tdTomato+/GFAP+ astrocytes. Consistent with these findings, while Foxj1 cKO mutant ECs did not upregulate early developmental markers p73, c-Myb, Pax6, and did not show activated caspase-3 activity, they similarly became Ki67+ and incorporated EdU following in vivo pulse to label cells entering the cell cycle (Supplementary Fig. 9b–g). In multiple Foxj1 control littermates, we did not detect Ki67+ nor EdU-pulsed ECs (Supplementary Fig. 9c, f, g).
The notion that ECs can serve as a progenitor source for new neurons in the postnatal brain15,16,17,18,33,54 remains controversial, as other reports describe multiciliated cells as terminally differentiated55,56. To test whether morphologically transformed ECs without foxj1 expression can become neurogenic, we first examined the olfactory bulbs (OBs) of foxj1CreERt2/Flox; R26R-tdT mutant mice injected with tamoxifen at P14, and harvested 2 and 4 weeks later. Unlike nestin-CreERtm4; R26R-tdT mice which showed robust lineage-tracing of OB interneurons post tamoxifen-induction9,57, we did not detect tdTomato+ interneurons in the OBs from multiple Foxj1 cKO mutants or littermate controls 2 or 4 weeks after tamoxifen injection (Fig. 7f, g). Cre antibody staining on P14 foxj1CreERt2/+ ventricular wholemounts showed ubiquitous Cre expression in multiciliated ECs (Supplementary Fig. 10a), giving them the capacity for tamoxifen-induced CreER recombination. While as expected two tamoxifen injections significantly increased mature ECs targeted by the foxj1CreERt2/+ driver, we did not detect tdTomato+ cells in the corresponding OBs 2 or 4 weeks following tamoxifen inductions (Supplementary Fig. 10b, c). Furthermore, following middle cerebral artery occlusion-induced stroke17 in 7-week-old mice with P14 tamoxifen induction, we did not detect lineage-traced tdTomato+ cells migrating from targeted ECs to the injured areas in Foxj1 cKO mutants or littermate controls (Supplementary Fig. 10d).
Studies on the cellular properties for Foxj1+ ECs over the last 10–15 years have relied on the human 1 kb FOXJ1 promoter regulatory region to generate genetic and viral drivers for cellular imaging and progenitor lineage-tracing3,4,17,33,58,59,60,61 and/or non-specific viral driver such as CMV16,58. We performed the same lineage-tracing experiment via the 1 kb human promoter FOXJ1-CreERt2 transgenic driver31 crossed to R26R-tdT mice, injecting tamoxifen at P7 and examined 2 weeks later. We were surprised to find extensive integration of tdTomato+ OB interneurons (Supplementary Fig. 11a). Since we did not detect similar tdTomato labeling in the OBs of either control foxj1CreERt2/+; R26R-tdT nor Foxj1 cKO mice after P7 tamoxifen induction, to distinguish mislabeling of integrated OB interneurons or postnatal NSCs by the 1 kb human FOXJ1 driver, we performed live-imaging of neuroblast migration in acute brain slice preparations from the rostral migratory stream (RMS). While we could not detect labeled cells in the RMS from foxj1CreERt2/+; R26R-tdT mice, we observed an abundance of migrating tdTomato+ neuroblasts from the FOXJ1-CreERt2; R26R-tdT RMS (Supplementary Movie 7), indicating that in addition to ECs there were mislabeled postnatal NSCs by the 1 kb human FOXJ1 promoter. In support, the widely used FOXJ1-GFP reporter mouse line30 (utilizing the same 1 kb human regulatory element), in addition to labeling ECs also expressed GFP in Foxj1-negative cells in the postnatal lateral ventricular niche (Supplementary Fig. 11b). Unlike at P7 and older ages where CreER and Foxj1 proteins are co-localized in foxj1CreERt2/+ mice, at P0 we detected numerous CreER+ but Foxj1-negative lateral ventricular pRGPs (Supplementary Fig. 11c). Following P0 tamoxifen injection in foxj1CreERt2/+; R26R-tdT animals, at P21 we observed tdTomato+ OB interneurons, as well as tdTomato+ lateral ventricular progenitors without Foxj1 protein expression nor multicilia (Supplementary Fig. 11d). With P0 tamoxifen injection, P10 live-imaging of RMS showed migrating tdTomato+ neuroblasts (Supplementary Fig. 11d), consistent with CreER-targeting of some neural progenitors transiently upregulating foxj1 transcription at this stage (thus expressing CreER but without stabilizing Foxj1 protein required for EC differentiation). We did not observe lineage-traced neurogenic capacity at any time point following P7 or P14 tamoxifen injection in foxj1CreERt2/+; R26R-tdT mice.
Recent results suggested that CD133+ ECs can give rise to neuroblasts postnatally18,54; thus there may exist a neurogenic multiciliated ependymal population that does not express Foxj1. We examined this possibility by searching for Foxj1-negative multiciliated ECs in nestin-Cre; foxj1KO/Flox mutant mice4, which restrict foxj1 deletion to the CNS. We reasoned that multiciliated ECs remaining in these mutant mice would be the best candidates. In ventricular wholemount preparations from P10 nestin-Cre; foxj1KO/Flox mutant mice, while as expected most of the ventricular surface lacked multicilia, we were able to find occasional multiciliated cells (Supplementary Fig. 12a). IHC staining revealed that these multiciliated cells uniformly expressed Foxj1, as well as CD133 (Supplementary Fig. 12b, c). These results showed that the remaining multiciliated ECs in nestin-Cre; foxj1KO/Flox mutant mice were Foxj1+ ECs which escaped Cre-mediated recombination. We did not detect the existence of Foxj1-negative multiciliated ECs.
The development and functions of multiciliated brain ECs, generally believed to be a terminally differentiated cell type, are poorly understood. To identify pathways that modulate ependymal differentiation/maturation, we uncovered an inherent cellular plasticity in mature ECs. We found that the Foxj1 transcription factor is required to maintain the multiciliated phenotype of ECs, but this important protein has a short half-life, constantly targeted for UPS-mediated degradation. To identify key molecular drivers leading to ependymal Foxj1 degradation, transcriptome and biochemical analyses revealed a non-canonical IKK2-dependent activity regulating Foxj1 protein stability. Armed with these molecular insights, we demonstrated that this previously undescribed signaling pathway is directly targeted by known pathological inhibitors of IKK2 to induce EC Foxj1 degradation and brain ventricular surface breakdown. Since the ependyma provides important barrier functions62 and directly contacts the CSF, which is frequently used to diagnose CNS diseases, we believe these results will have important implications for hydrocephalus pathophysiology and neuro-immune interactions.
ECs form a highly specialized epithelium lining the brain ventricles, with a distinct cellular feature of multiple motile cilia emanating from their apical surfaces. The ependymal developmental program involves sequential specialization of this multiciliated phenotype, including rapid duplication of basal bodies63, followed by upregulation of the Foxj1 transcription factor to induce motile multiciliary development, as well as morphological change from a glial to epithelial cell type3,4. Supposed to last the lifetime of the animal, there was little experimental evidence to suggest that mature ECs retained the ability to divide and self-renew, which will require the disassembly of multiple basal bodies and their associated motile cilia. While these cellular specializations may, in principle, indicate that mature ECs represent a stably differentiated cell population, our findings here showed that the continuous production of a single transcription factor, Foxj1, is necessary to actively prevent phenotypic transformation of mature ECs. We demonstrated that Foxj1 is constantly turned over via UPS-mediated degradation, and its removal from mature ECs resulted in de-differentiation. These results bring to mind the “Red Queen Hypothesis”, where differentiated ECs are constantly running to stand still by making and stabilizing Foxj1, whose degradation then allows ECs to re-enter cell cycle and regenerate. It will be of interest to unravel the signaling networks controlling Foxj1 protein stability in mature ECs.
To determine which biological processes may accelerate ependymal Foxj1 degradation, we performed transcriptome and biochemical analyses, and uncovered, quite unexpectedly, that a non-canonical IKK2 activity was required to stabilize Foxj1 protein. While IKK2 phosphorylation of NF-κB inhibitor IκB is well known, other IKK phosphorylation sites have been identified in proteins such as IRS-1 and Foxo3a64,65, with amino acid sequence similarities to serines 32/35 motifs in Foxj1. This IKK2 target site on Foxj1 is highly conserved evolutionarily from frogs, rodents, to human. Given the importance of IKK pathway function in diverse pathological processes, it was reasonable to assume that there may be pathological/disease states inducing IKK2 suppression in ECs. Growth factor exposure and pathogenic viruses have been previously shown to inhibit IKK2 activity52,53, and were thus good candidates for investigation. HSV-1 is an IKK inhibiting virus that can result in devastating outcomes following CNS attack, particularly in infants46. It has been associated with ventricular wall damage and hydrocephalus, although the underlying molecular pathogenesis for these outcomes remain largely unknown. Ventricular bleeding is another major cause of hydrocephalus14. Whereas obstruction of CSF flow from blood clots is a known initiator, serum factors aberrantly introduced into CSF may also disrupt ventricular integrity. Remarkably, we found that both viral infections and serum treatment can robustly induce ependymal Foxj1 degradation and EC de-differentiation. Considering that inheritable human mutations resulting in hydrocephalus are relatively rare, environmental exposures that tilt EC Foxj1 protein towards sustained degradation provide additional avenues to explore for understanding and treating acquired cases of hydrocephalus.
Consistent with our initial observation that MLN4924 was able to stabilize Foxj1 protein by inhibiting UPS-targeted degradation, in vitro/in vivo applications of MLN4924 showed attenuation of HSV-1-induced EC Foxj1 degradation and ventricular wall disruption. MLN4924 is currently in clinical trials as an anti-cancer treatment23 due to its effectiveness in blocking cell cycle progression22. While MLN4924 has a beneficial effect on preserving ependymal Foxj1 and ventricular integrity, it will likely blunt the important adaptive immune responses where proliferation of microglia, astrocytes, and peripheral immune cells are needed to restore tissue homeostasis. SMER3, while not a specific chemical inhibitor of mammalian Fbxw cullin-RING ligases, has been shown to have selective activity against related homolog SCFMET30 in yeast26. Developing similar class of compounds to target specific mammalian Fbwx ligase family members may promote EC Foxj1 protein stability while sparing cell cycle inhibition.
Once we observed that Foxj1 loss could induce de-differentiation and cell division in mature ECs, we became hopeful that this step may be the long-sought molecular prelude to new neuron production, and a link to unite divergent views in the adult neurogenesis field. To test this, we first examined the OBs of foxj1CreERt2/Flox; R26R-tdT mutant mice following postnatal tamoxifen injection. While we did not detect tdTomato+ OB interneurons in control foxj1CreERt2/+; R26R-tdT mice after tamoxifen induction, we reasoned that any tdTomato+ OB interneurons found in the conditional mutant mice would indicate ependymal neurogenic capacity following foxj1 removal. At either 2 or 4-week time points following tamoxifen injection, we did not detect tdTomato+ OB interneurons in foxj1CreERt2/Flox; R26R-tdT mutant mice. We observed similar negative results following cortical stroke. Robust numbers of tdTomato+ ECs from the lateral ventricles of these experimental animals ruled out lack of tamoxifen-induced recombination as a potential confound.
Previous studies tackling neurogenic properties of multiciliated ECs have depended on non-specific CMV viral drivers16,58, or the human 1 kb FOXJ1 promotor regulatory element to generate reporters and Cre drivers3,17,33,58,59,60,61. While Ad-CMV-Cre also targets postnatal neural stem cells effectively66,67, in principle lineage-tracing strategies utilizing the FOXJ1 promoter should be EC-specific. Repeating the same tamoxifen-inducible experiment in FOXJ1-CreERt2; R26R-tdT mice, we found robust tdTomato labeling of OB interneurons. Live-imaging of migrating tdTomato+ neuroblasts in FOXJ1-CreERt2; R26R-tdT mice, as well as FOXJ1-GFP expression in Foxj1-negative cells, revealed non-specific labeling of neurogenic population by the human 1 kb FOXJ1 promotor. While it remains possible that de-differentiated mature ECs can become neurogenic in other contexts, our results here indicate a cellular plasticity program for cell cycle re-entry and EC renewal.
Knockout and transgenic lines
All mouse experiments were performed according to an approved protocol by the Institutional Animal Care and Use Committee at Duke University. The following lines were used: foxj1Flox/+ 4; FOXJ1-GFP30; FOXJ1-CreERt2 31; nestin-Cre68. To generate the foxj1CreERt2/+ driver, CreERt2 sequence was inserted in-frame into ATG start codon of the foxj1 gene. The knock-in targeting construct was made by standard BAC recombineering techniques, electroporated into mouse embryonic stem cells by standard methods. PCR genotyping primers: P1 (TGG CTT GCA GGT ACA GGA GG), P2 (CCT CAT TGT GGC CAA CAA ACC C); 94 °C 30 s, 60 °C 1 min, 72 °C 1 min, 32 cycles. Tamoxifen (10 mg/ml, freshly dissolved in corn oil) was injected intraperitoneally, 0.15 mg per gram of body weight to induce recombination.
The Foxj1-Myc plasmid was described previously and contains a C-terminal Myc-tag downstream of the Foxj1 ORF4. The shRNA sequence for Fbxw5 (CCATACAACTGGAGTTACA) was cloned into the plvthm plasmid. The Fbxw5 construct contains an N-terminal 3× -Myc tag and is cloned into pcDNA3 plasmid. The pRK5-HA-Ubiquitin-WT containing an N-terminal HA tag was acquired from addgene.com. pCR-IKK2(Wt)-Flag and pCR-IKK2(K44A) mutant was acquired from addgene.com. Lentiviral constructs containing validated mouse shRNAs against IKK2 or a scrambled control were purchased from Geneocopia. Site-directed mutagenesis kit (New England Biolabs) was used to convert Foxj1 serines 32 and 35 to aspartic acid or alanine residues. Inserts containing the mutations of interest were sub-cloned into PWPXLD plasmid and sequenced to ensure no off-target mutations.
Primary ependymal cultures and chemical screen
ECs were grown in 24-well dishes from isolated radial glial progenitors as previously described4. For chemical screen, we used pathway selective custom library purchased from Selleck Chemicals. For immunofluorescence staining, cell cultures were grown on 12 mm glass coverslips, washed once in PBS then fixed in either 2% or 4% paraformaldehyde solution. Fixed cells were permeabilized in PBS containing 0.1% TritonX-100 (PBST) and blocked using 10% donkey serum in PBST. Primary antibody solutions were incubated with cells overnight at 4 °C and conjugated secondary antibodies were incubated for 3 h at 4 °C. DAPI was incubated for 20 min at room temperature and coverslips were placed onto slides with aqua-poly/mount for confocal imaging. During differentiation MLN4924 was included during the switch to differentiation media at 0.25 µM and incubated for 2 days, then sequentially halved with differentiation media for 3 additional days. SMER3 was incubated in mature ependymal cultures at a final concentration of 5 µM for 12 h. MG132 was incubated with ependymal cultures at a final concentration of 25 µM in ependymal differentiation media. Cycloheximide was used at a final concentration of 20 µg/ml in ependymal differentiation media. Ependymal cultures were incubated with JSH-23 and SC75741 at a final concentration of 25 and 5 µM, respectively. For EC re-differentiation after de-differentiation, EC cultures made from foxj1CreER/+; R26R-tdT animals were: (1) fully differentiated as described above, and CreER-dependent tdTomato expression was achieved by treating cultures with 4-OH-tamoxifen (1 µg/ml, Sigma) for 24 h, followed by 7 days of culturing in differentiation media. (2) The EC cultures were then treated with proliferation media for 4 days, EdU (1 µg/ml, Sigma) was added for the last 2 days. (3) Culture was switched back to differentiation media for further 10 days prior to analyses.
Total RNA was extracted from EC cultures using RNA easy mini Kit (Qiagen) and cDNA was prepared using Superscript Vilo cDNA synthesis kit (Invitrogen). Quantitative PCR analysis was performed using SYBR Green as previously described57 using primer sets Foxj1 F: 5′-tcactctgtcggccatctac-3′ and R: 5′-ttcttgaaggccccactgag-3′; GAPDH F: 5′-catggccttccgtgttcct-3′ and R: 5′- tgatgtcatcatacttggcaggtt-3′.
Immunohistochemistry and imaging
IHC of brain coronal sections were prepared as previously described9. Brain ventricular lateral wall wholemounts were isolated as previously described4 and fixed directly in paraformaldehyde solution (4 or 2%) followed by permeabilization in 0.1% PBST solution. Donkey serum at 10% in PBST was used to block tissue sections and wholemounts prior to immunolabeling. All images were acquired on Leica TCS SP5 confocal microscope, with control and experimental samples imaged under identical settings. Primary antibodies solutions were incubated overnight with tissue and fluorescently conjugated secondary were incubated for 3 h and 4 °C. For treatment of wholemounts with MLN4924, cycloheximide, and IMD0354 drugs, tissue was incubated in 0.5 µM, 25 µg/ml, and 0.5 µM concentration respectively for the indicated lengths of time. The following antibodies and dilutions were used: Foxj1 (rabbit, 1:1000, Sigma); Acetylated-tubulin (mouse, 1:1000, Sigma); VP16 (mouse, 1:1000, Santa Cruz); anti-CMV (mouse, 1:1000, Millipore); JF186/EBV (mouse, 1:500, gift of Micah Luftig); anti-vaccinia virus (mouse, 1:1000, Abcam); Fbxw5 (goat, 1:500, Santa Cruz); RFP (rabbit, 1:1000, Rockland); Acetyl-tubulin (mouse, 1:1000, Sigma); GLAST (mouse, 1:1000, Novus); DCX (guinea pig, 1:200, Millipore); CD31 (mouse, 1:1000, Invitrogen); HA (mouse, 1:1000, Genescript); Myc (mouse, 1:1000, thermo-scientific); Cre (mouse, 1:250, Millipore); Mcidas (rabbit, 1:250, Biorbyt); p73 (mouse, 1:250, Novus Biologicals); PAX6 (mouse, 1:250, abcam); Ki67 (rabbit, 1:1000, abcam); IKK2 (mouse, 1:1000 blot, Novus Biologicals); cyclin D1 (mouse, 1:2000, BD Biosciences); BIII-tubulin (mouse, 1:1000, R&D Systems); c-Myb (mouse, 1:250, Santa Cruz); Activated-Caspase 3 (rabbit, 1:250, Cell Signaling). Live-imaging was performed as previously described57, using upright Leica SP8 confocal microscope. For organotypic cultures, lateral ventricle ependymal wholemounts were isolated as previously described4, placed on collagen-coated membrane inserts (Millipore), incubated in control media (DMEM/F12, 2% FBS, 1% glutamine, and 1% Pen/Strep) or proliferation media (DMEM/F12, 10% FBS, EGF (20 ng/ml), bFGF (20 ng/ml), 1% glutamine, and 1% Pen/Strep), and live-imaged on Zeiss Cell Observer® System equipped with incubation system set at 37 °C and 5% CO2. For 3D ependymal morphological quantification: confocal z-stack images of tdTomato+ ECs were first reconstructed in Imaris 3D view, and number of major basal extensions quantified via Imaris 3D filament tracer software module. Ependymal soma position was used as start point, and protrusions <5 µm were excluded from analyses to focus on significant cellular extensions.
Biochemistry and immunoprecipitation
Cell cultures or tissue samples were lysed and prepared for immunoblotting as described previously4,69. Lysates were resolved on Tris-Glycine SDS-PAGE gels (Bio-Rad) and transferred onto PVDF membranes (Bio-Rad). Antibodies were diluted in PBS containing 0.1% Triton-X100 with 2% BSA, followed by overnight incubation at 4 °C. Detection was accomplished through secondary antibodies conjugated to horseradish peroxidase (Cell Signaling) and treated with enhanced chemiluminescence. To separate the Foxj1 doublet lysates were allowed to migrate through premade 4–20% gradient Tris-Glycine gels (Bio-Rad) for 1.5 h at 170 volts. Immunoprecipitations were performed on ependymal cultures lysed in buffer containing 0.02 M Hepes buffer, 0.15 M NaCl, 1% Triton-X100, sodium orthovanadate (Sigma), protease inhibitor cocktail (Roche), 10 mM NaF. Supernatants were collected from lysates spun at 15 K rpm for 20 min, and primary antibodies incubated either 4 h or overnight at 4 °C. Secondary antibodies conjugated to agarose beads were incubated for 2 h at 4 °C. Beads were washed five times in lysis buffer, harvested in 5× PAGE, processed for SDS-PAGE analysis and western blotting. HEK293 cells were grown on 24-well culture dishes. Twenty-four hours after plating HEK293 cells were transfected with equal amounts of plasmids for Foxj1-Myc-WT, IKK2-WT, or IKK2 mutant. Twenty-four hours later transfection media was replaced with low serum media, then harvested 16 h later for SDS-PAGE and western blotting. To separate the Foxj1 doublet lysates were allowed to migrate through premade 4–20% gradient Tris-Glycine gels (Bio-Rad) for 1.5 h at 170 volts. The following antibodies and dilutions were used for the study: Foxj1 (rabbit, 1:2000, Sigma); Flag (mouse, 1:2000, Sigma); HA (mouse, 1:2000, Genescript); Myc (mouse, 1:2000, Thermo-Scientific); ubiquitin (mouse, 1:2000, IMGENEX); Fbxw5 (goat, 1:2000, Santa Cruz); actin (rabbit, 1:5000, Sigma).
Lentiviruses were produced in 293T17 cells through triple-transfection of transfer plasmids (pLVTHM or pWPXLD), psPAX2, and pMD2.G. Lentivirus containing media was collected, filtered through 0.45 micrometer membranes. To concentrate lentiviruses media was spun for 2 h at 25 K rpm using a swinging bucket rotor and a Beckman Coulter ultracentrifuge. For HSV-1, vero cells were grown to confluence on 10 cm tissue culture dishes in DMEM supplemented with 10% fetal bovine serum at 37 °C in 5% CO2. Cells were infected with HSV-1 KOS strain70. After 48 h media containing HSV-1 virus was collected, cleared through centrifugation, and filtered through 0.45 micrometer pore membranes and stored at −80 °C. Viral titers were determined through plaque assay of Vero cells. HSV-1-KOS was maintained at a stock of 107 plaque-forming units/ml.
In vitro and in vivo viral infections
ECs were cultured in 24-well plates and cells were infected with an MOI of 2 for HSV-1-KOS or vaccinia virus. After 4 h media was replaced and cells assayed at the time points indicated. For HSV-1 and MLN4924 co-treatment, ECs were treated with HSV-1-KOs for 4 h then treated with MLN4924. MLN4924 was added 24 h after initial treatment with vaccinia virus. CMV virus and EBV virus were incubated with ependymal cultures for 4 h and media was replaced and ECs cultured for an additional 4 days. For in vivo infections mice were anesthetized by gas phase isoflurane mixed with oxygen. HSV-1-KOS virus in 0.12 µl (2×104 pfu) was injected intraventricularly through stereotaxic parameters in 3–4-week-old mice similarly to previously described method4. Coordinates for intraventricular injections (zeroed from bergma): X = 0.9 µm, Y = 2.5 µm, Z = 2.2 µm. For MLN4924 drug treatment mice were first given MLN4924 intraventricularly at a final concentration of 0.5 µM. This was followed shortly by an injection of HSV1 within the same injection site. Mice were assayed at the time points indicated. To assay barrier breakdown DAPI+ nuclei were used to measure gaps on the ependymal surface.
In vivo injections and injury
For in vivo EdU labeling, 1 mg of EdU (in 100 µl of saline) was administered through intraperitoneal injection. Seven-week-old anesthetized animals were induced for focal ischemic brain injury through middle cerebral artery occlusion as described17.
Affymetrix Gene Chip microarray data underwent strict quality control processing using the “simpleaffy” package in the Bioconductor 2 suite from the R statistical programming environment. Log-scale Robust Multiarray Analysis from the “affy” package was used to normalize the data and eliminate systematic differences across the arrays. Differential expression of genes across the conditions was identified from a moderated test statistic employed by the “limma” package. The False Discovery Rate (FDR) method was used to correct for multiple hypothesis testing. Gene set enrichment analysis was performed to identify differentially regulated pathways and gene ontology terms for each of the comparisons performed. Heatmap of the differentially expressed genes (FDR ≤ 0.05) for each 〈Condition〉 vs. 〈Condition〉 comparison. Gene expression has been z-score normalized and the samples and genes are clustered by correlation distance with complete linkage. The heatmap shows all genes that were identified as significant in at least one of the comparisons. Heatmaps only include genes that had an FDR ≤ 0.05 and a log2FC >1 or <−1. Hierarchical clustering plot of all of the samples based on expression of all genes on the arrays was shown. The samples are clustered using a correlation distance with complete linkage. Samples are colored by their condition.
No statistical methods were used to pre-determine samples sizes, but they are similar to those reported elsewhere4,9,57. For representative images and western blots, the results were shown to be reproducible by at least three separate experiments. Blocking of experimental design was assigned by animal genotype and was not randomized. Depending on sample size, unpaired Student’s t test (≥10) or Wilcoxon two-sample test (≥5) were used for statistical comparisons between two data sets. Throughout, box-and-whisker plots were generated via http://boxplot.tyerslab.com/ to depict mean (+), median (line), low and high quartiles (boxes), and range (whiskers).
Microarray data are available from the GEO database, accession number GSE111182. All other remaining data are available within the article and Supplementary Files, or available from the authors upon request.
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank D. Tscharke (Northwestern), R. Antonia (UNC), D. Carr (Oklahoma) for helpful discussions; Duke Sequencing & Genomic Technologies Shared Resource for help with transcriptome experiment; H. Sheng and D. Warner at Duke Neuroprotection Laboratory for help with injury experiment; Duke Neurotransgenic Facility for assistance with generation of foxj1CreERt2/+ mice; B. Cullen for providing HSV-1 stock; D. Pickup for vaccinia virus stock; M. Luftig for EBV stock; S. Permar for CMV stock; D. Fromme and members of the Kuo laboratory for project assistance; V. Bennett, E. Adlaf, and D. Pickup for helpful comments on manuscript. This work was supported by Jean and George Brumley Jr. Endowment (K.A. and C.T.K.); Ruth K. Broad Foundation (P.P.-G.); the March of Dimes; US National Institutes of Health grants R01NS078192, R01NS0960960, R01MH105416 (C.T.K.).
About this article
Nature Communications (2019)