Charged multivesicular body protein 1A/Chromatin modifying protein 1A (CHMP1A) is a member of the ESCRT-III (endosomal sorting complex required for transport-III) complex1–2, but is also suggested to localize to the nuclear matrix and regulate chromatin structure3. Here we show that loss-of-function mutations to human CHMP1A cause reduced cerebellar size (pontocerebellar hypoplasia) and reduced cerebral cortical size (microcephaly). CHMP1A mutant cells show impaired proliferation, with increased expression of INK4A, a negative regulator of stem cell proliferation, and chromatin immunoprecipitation suggests a loss of the normal INK4A repression by BMI in these cells. Morpholino-based knockdown of zebrafish chmp1a resulted in brain defects resembling those seen after bmi1a and bmi1b knockdown, which were partially rescued by INK4A orthologue knockdown, further supporting links between CHMP1A and BMI1-mediated regulation of INK4A. Our results suggest that CHMP1A serves as a critical link between cytoplasmic signals and BMI1-mediated chromatin modifications that regulate proliferation of CNS progenitor cells.
As part of ongoing studies of human disorders of neural progenitor proliferation, we identified three families characterized by underdevelopment of the cerebellum, pons, and cerebral cortex (Fig. 1a–d). In a consanguineous pedigree of Peruvian origin, three children in two branches were affected (Fig. 1e; Family 1). Two additional pedigrees from Puerto Rico showed similar pontocerebellar hypoplasia and microcephaly (Fig. 1e; Family 2 and 3). Brain MRI of affected individuals from all families show severe reduction of the cerebellar vermis and hemispheres. Strikingly, the cerebellar folds (“folia”) are relatively preserved despite the extremely small cerebellar size (Fig. 1a–d, Supplementary Videos 1, 2). All affected individuals had severe pontocerebellar hypoplasia, though affected individuals in Family 1 showed better motor and cognitive function than those in Family 2 and 3 (Supplementary Note, Clinical Information).
Genome-wide linkage analysis of Family 1 and 2 using single nucleotide polymorphism (SNP) microarrays implicated only one region on chromosome 16q as linked and homozygous in all six affected individuals (Fig. 1e, Supplementary Fig. 1), with a maximum multipoint LOD score of 3.68 (Fig. 1e). Although Families 2 and 3 are not highly informative for linkage analysis, their shared homozygosity provides additional support for this locus. Furthermore, Families 2 and 3 shared the same haplotype (Supplementary Fig. 1), suggesting a founder effect. Sequencing of 42 genes within the candidate interval on 16q24.3 revealed homozygous variants predicted to be deleterious in the CHMP1A gene only. CHMP1A consists of seven exons encoding a 196 amino acid protein (Supplementary Note, CHMP1A isoforms). Affected individuals in Family 2 and 3 had a homozygous nonsense variant in exon 3, predicted to prematurely terminate translation (c.88C>T; Q30X; Fig. 2a). Family 1 showed a homozygous variant in intron 2 of CHMP1A (c.28-13G>A; Fig. 2a) predicted to create an aberrant splice acceptor site leading to an 11 base pair insertion into the spliced mRNA product (Supplementary Fig. 2a). The two mutations were absent from dbSNP, 281 neurologically normal European control DNA samples (562 chromosomes), the 1000 Genomes Project database4, and approximately 5000 control exomes from the NHLBI Exome Sequencing Project. We sequenced CHMP1A in 64 individuals with other cerebellar anomalies without finding additional mutations, but none of these patients shared the rare and distinctive pattern of hypoplasia seen in the individuals with CHMP1A mutations.
RT-PCR analysis of CHMP1A in lymphoblastoid cells from affected individuals from Family 1 (CH3101 and CH3105) identified the predicted aberrant transcript with the 11 base pair insertion and a second aberrant transcript with a 21 base pair insertion, but no normal CHMP1A transcript (Supplementary Fig. 2b). In the parents of affected children from Family 1, and in unaffected control samples, only the normal transcript was detected, suggesting that the abnormal splice products are unstable. Western blot analysis revealed a single 24 kilodalton band in a normal control individual, but no corresponding band was detected in affected individuals from Families 1 or 2 (CH3101 and CH2401, respectively; Fig. 2c). Normalized to the loading control, levels of CHMP1A were 50% in the parent (CH3104). Hence this genetic study establishes CHMP1A null mutations as the cause of pontocerebellar hypoplasia and microcephaly in these pedigrees.
CHMP1A has been assigned two distinct putative functions, as both a chromatin modifying protein, and a charged multivesicular body protein1,3. CHMP1A was originally identified as a binding partner of the Polycomb group protein Pcl (Polycomblike)3. In the nucleus, it has been suggested to recruit the Polycomb group transcriptional repressor BMI1 to heterochromatin, and overexpressed CHMP1A has been shown to arrest cells in S-phase 3. In the cytoplasm, CHMP1A is part of the ESCRT-III complex (endosomal sorting complex required for transport)1–2. ESCRT-III complex localizes to endosomes and interacts with VPS4A and VPS4B5 to assist in the trafficking of ubiquitinated cargo proteins to the lysosome for degradation6.
We investigated potential relationships of CHMP1A to Polycomb function by analysis of cell lines from two patients harboring different CHMP1A mutations (CH3101 from Family 1, and CH2401 from Family 2), which show severely impaired doubling times compared to control cell lines, suggesting essential roles of CHMP1A in regulating cell proliferation (Fig. 2d). In order to examine BMI1 function in these cells, we performed quantitative PCR analysis of expression of the BMI1 target locus CDKN2A, which encodes alternative transcripts INK4A (also known as p16INK4a) and ARF (also known as p14ARF) in human. This revealed abnormally increased expression of INK4A, the isoform implicated in cerebellar development, but not of ARF (Fig. 2e), suggesting de-repression of INK4A. Chromatin immunoprecipitation with a BMI1 antibody in control cell lines showed an approximately eight-fold enrichment of BMI1 binding at INK4A promoter DNA, relative to a control region 7kb upstream, whereas cells from an affected individual (CH2401) showed only about half this effect (Fig. 2f). Enrichment of BMI1 at the ARF promoter was not substantial in this assay, and was similar in both control and cell lines from affected individuals, consistent with the specificity of regulation of the INK4A isoform by BMI1 (Fig. 2f). Bmi1 suppresses the Cdkn2a locus via Polycomb-mediated H2A monoubiquitination, and is required for neural stem cell self-renewal7. Our evidence suggests a role for CHMP1A in mediating BMI1-directed epigenetic silencing at the INK4A promoter, but not at the ARF promoter.
We further explored the relationship between CHMP1A and BMI1 using morpholino-based knockdown experiments in zebrafish. Knockdown of the zebrafish CHMP1A orthologue (chmp1a) resulted in reduced cerebellum and forebrain volume, similar to the effects of human CHMP1A mutations and zebrafish knockdown of BMI1 orthologues (Fig. 3a–e, Supplementary Fig. 3, 4). A second morpholino led to a similar phenotype, and both morpholinos were partially rescued by human CHMP1A mRNA, confirming the specificity (Supplementary Fig. 4). The cerebellum consists of 5 major cells types, with the principal cell, known as the Purkinje cell, deriving from the ventricular epithelium, whereas granule cells derive from a separate progenitor pool known as the rhombic lip. Granule cell precursors then migrate over the outer surface of the cerebellum and form the external germinal layer (EGL) before migrating radially past the Purkinje cells to settle in the internal granule layer (IGL)8. Within the chmp1a morphant cerebellum, the internal granule and molecular layers were severely affected (Fig. 3a, b), which is consistent with the relatively preserved folia pattern of the human cerebellum (thought to be primarily established by Purkinje cells) and severely reduced volume (determined mainly by granule cell quantity).
We then tested genetic interactions between chmp1a and the zebrafish orthologue of INK4A (cdkn2a). Knockdown of cdkn2a alone did not result in noticeable abnormalities, and double knockdown of chmp1a and cdkn2a resulted in partial rescue of the brain morphology defects seen with chmp1a knockdown (Fig. 3f, g). This was analogous to the rescue of the Bmi1 knockout mouse cerebellar phenotype seen in the Bmi1 and Cdkn2a double knockout mice9. Of note, there are also parallels in brain morphology between individuals with CHMP1A mutations and Bmi1-deficient mice, which show cerebellar hypoplasia10–11 (Fig. 3h, i). In Bmi1-null mice, the cerebellar architecture was generally preserved, but the thickness of the granular and molecular layers was markedly reduced10, and Bmi1-deficient mice show a modest reduction in cerebral volume10,12, similar to individuals with CHMP1A mutations (Supplementary Note, Clinical Information).
Subcellular localization of Chmp1a appears to vary depending on the cell type. Confocal images of NIH 3T3 cells show prominent exclusion of Chmp1a from the nucleus, where Bmi1 is seen (Fig. 4a). On the other hand, confocal images of HEK293T cells, while still showing predominantly cytoplasmic localization, show some nuclear immunoreactivity as well (Fig. 4b). Primary cultures of cerebellar granule cells also show predominant cytoplasmic localization, along with a speckled nuclear pattern (Fig. 4c). Overexpression of HA-tagged Chmp1a in cultured granule cells shows abundant nuclear Chmp1a with a punctate expression pattern, confirming the speckled nuclear localization of native Chmp1a (Fig. 4d), and consistent with earlier reports that Chmp1a can appear in the nucleus3. However, even with overexpression, Chmp1a and Bmi1 do not prominently co-localize within the nucleus, also in agreement with previous data3.
Immunohistochemical studies of mouse developing cerebellum and cerebral cortex revealed widespread expression of Chmp1a in dividing and postmitotic cells. Chmp1a immunoreactivity is seen in the nucleus and cytoplasm of EGL, Purkinje and IGL cells (Fig. 4e, f, Supplementary Fig. 5). In the nucleus of these cells, Chmp1a immunoreactivity is seen in a speckled pattern. These speckles may be seen adjacent to Bmi1 signals, but they usually do not colocalize (Fig. 4f, Supplementary Fig. 5). At later stages of cerebellar development (P4, P10 and P29), Chmp1a expression persists in Purkinje and granule cells (Supplementary Fig. 6). E13.5 cerebral cortex shows widespread Chmp1a expression in the neuroepithelial cells (Fig. 4g). In the postnatal cerebral cortex (P4, P10 and P29), Chmp1a expression in postmitotic neurons of the cortical plate gradually decreases, and becomes almost undetectable by P29 (Supplementary Fig. 6). These expression studies confirm that Bmi1 and Chmp1a are often expressed in the same cells. On the other hand, the absence of widespread subcellular co-localization of Bmi1 and Chmp1a suggests that the regulation of Bmi1 by Chmp1a is perhaps not mediated by direct physical interaction.
Our data implicate CHMP1A as an essential CNS regulator of BMI1, which in turn is a key regulator of stem cell self-renewal. Chmp1a’s dual cytoplasmic and nuclear localization, and its connection to the ESCRT-III complex, position Chmp1a as a potentially crucial link between cytoplasmic signals and the global regulation of stem cells via the Polycomb complex.