Rare variants in BMAL1 are associated with a neurodevelopmental syndrome

Significance

Children with neurodevelopmental disorders exhibit highly penetrant sleep and circadian dysfunction, but the underlying mechanisms are unclear. We asked whether a subset of individuals with neurodevelopmental disorders might have genetic variants in genes known to drive circadian rhythms. Through international collaboration, we identified ten individuals with very rare genetic variants in BMAL1, a core component of the molecular clock. These individuals exhibited overlapping signs and symptoms including developmental delay, autism spectrum disorder, and variably penetrant marfanoid features. We functionally tested the identified BMAL1 variants in cell culture and in vivo and found disrupted BMAL1 function. These findings demonstrate that neurodevelopmental dysfunction can be driven by variation in circadian clock genes in a subset of individuals.

Abstract

Through international gene-matching efforts, we identified 10 individuals with ultrarare heterozygous variants, including 5 de novo variants, in BMAL1, a core component of the molecular clock. Instead of an isolated circadian phenotype seen with disease-causing variants in other molecular clock genes, all individuals carrying BMAL1 variants surprisingly share a clinical syndrome manifest as developmental delay and autism spectrum disorder, with variably penetrant sleep disturbances, seizures, and marfanoid habitus. Variants were functionally tested in cultured cells using a Per2-promoter driven luciferase reporter and revealed both loss-of-function and gain-of-function changes in circadian rhythms. The tested BMAL1 variants disrupted PER2 mRNA cycling, but did not cause significant shifts in cellular localization or binding with CLOCK. Conserved variants were further tested in Drosophila, which confirmed variant-dependent effects on behavioral rhythms. Remarkably, flies expressing variant cycle, the ortholog of BMAL1, also demonstrated deficits in short- and long-term memory, reminiscent of the highly prevalent developmental delay observed in our cohort. We suggest that ultrarare variants in the BMAL1 core clock gene contribute to a neurodevelopmental disorder.

The molecular circadian clock is the primary mechanism allowing cells to maintain 24-h cycles. The mammalian molecular clock is composed of the transcription factors BMAL1 and CLOCK, which heterodimerize and bind to the E-box domain of the PER and CRY families of genes to promote their transcription (Fig. 1A). The PER and CRY proteins, in turn, heterodimerize and block the transcriptional activation induced by BMAL1 and CLOCK (Fig. 1A). This transcription–translation feedback loop takes approximately 24 h to complete a cycle and is the primary backbone of daily timekeeping.

Fig. 1.

Characteristics of BMAL1 variants presented in this study. (A) The molecular clock is composed of a transcription–translation feedback loop. BMAL1 and CLOCK promote the transcription of the PER and CRY families of genes, and then are inhibited by those protein products over a period of 24 h. (B) Schematic of human BMAL1 protein with major protein domains depicted in blue. Missense variants are listed on top and frameshift or predicted truncations are depicted on bottom. (C) Amino acids implicated in this study demonstrate marked conservation across animal phyla. (D) SpliceAI was used to predict changes in splice acceptors or donors for the three rare splice site variants presented in this study, as well as three nearby common splice site variants. Rare variants are likely to lead to a loss of splice acceptor and/or donors. 0.8 indicates a stringent cutoff, 0.5 indicates a moderate cutoff, and 0.2 indicates a low cutoff value.

Widespread adoption of clinical exome sequencing has led to the identification of variants in core components of the molecular clock resulting in Mendelian phenotypes. Variants in PER2, PER3, and CRY1 have been implicated in advanced sleep phase disorder (1–3), while variants in CRY1 cause delayed sleep phase disorder (4). As predicted, these variants in clock genes affect circadian rhythms including the timing of sleep, but do not have other associated phenotypic features.

Here, we identified rare variants in BMAL1 in 10 individuals with a neurodevelopmental phenotype characterized by developmental delay and autism spectrum disorder. Through functional testing using a cell-based reporter and a small animal model, we find these variants perturb BMAL1 function and result in circadian and memory deficits.

Results

Ultrarare Variants in BMAL1 Are Predicted to Perturb Function.

Through international collaboration facilitated by GeneMatcher (5) and review of published cohorts (6, 7), we identified 10 individuals with rare heterozygous variants in BMAL1, including five individuals with de novo variants. gnomAD, a large database of genetic sequencing data that captures natural variation throughout the population, reveals that BMAL1 is a highly constrained gene that is intolerant to loss-of-function changes (pLI = 1) and missense variation (Z-score = 3.64), raising the possibility that rare nonsynonymous coding variants have functional consequences.

We identified 3 heterozygous missense ultrarare variants, as well as 7 heterozygous ultrarare variants predicted to introduce a frameshift and/or premature truncation (Fig. 1B). BMAL1 has four well-characterized protein domains including, a) a basic helix–loop–helix (bHLH) domain that enables binding to DNA, b) and c) PAS1 and PAS2 domains, which mediate heterodimerization with CLOCK (8), and d) a PAC domain that facilitates folding of the PAS domains. One of the missense variants, BMAL1 NM_001297719.2:c.602 T>C; p.(Ile201Thr), occurs in the PAS1 domain (Fig. 1B) and is present twice in gnomAD, (v.4.1.0) with an allele frequency of 1.24e-6 and Combined Annotation Dependent Depletion (CADD) score of 23.5, suggesting it is likely deleterious (SI Appendix, Table S1). The other two missense variants, BMAL1 NM_001297719.2:c.202G>A; p.(Gly68Arg) and BMAL1 NM_001297719.2:c.1501G>A; p.(Glu501Lys), have no population frequency and CADD scores of 24.4 and 23.5, indicating they are also likely deleterious (SI Appendix, Table S1). The BMAL1 p.(Glu501Lys) variant is particularly striking because it leads to the substitution of negatively charged glutamic acid with a positively charged lysine. Indeed, in silico analysis using AlphaMissense(9) indicates this variant is predicted to disrupt protein structure (score of 0.9724 with scores between 0.564 to 1 likely to be pathogenic).

One frameshift variant occurs in the PAS2 domain [NM_001297719.2:c.1019_1020del; p.(Tyr340CysfsTer35)], while the PAC domain contains a premature truncation [NM_001351824.2:c.1212dupT; p.(Lys405Ter)] and a frameshift variant [NM_001297719.2:c.1437dupC; p.(Thr480HisfsTer22)]; these variants are absent from the gnomAD v4.1.0 database and have no known population frequency. All coding variants occur at highly conserved residues (Fig. 1C). The NM_001297719.2:c.1604del; p.(Gly535GlufsTer) variant occurs closest to the C-terminus of the variants presented here and has no population frequency (SI Appendix, Table S1). All of the predicted truncating variants described here are located upstream of the last two exons of BMAL1, and therefore, are likely to be subjected to nonsense-mediated decay.

The splice-site variants are distributed throughout the gene including before the bHLH domain (NM_001351824.2:c.141-2A>G), before the PAS2 domain (NM_001297719.2:c.821-1G>C), and after the PAC domain (NM_001297719.2:c.1524-3C>A). We used SpliceAI, a deep-learning tool, to predict the functional consequences of these intronic variants (10). Using a stringent cutoff of 0.8, we found the c.141-2A>G and c.821-1G>C variants cause loss of splice acceptor sites (Fig. 1D). Additionally, using a moderate cutoff of 0.5, c.141-2A>G leads to a loss of a splice donor and c.1524-3C>A leads to loss of splice acceptor and donor sites (Fig. 1D). The c.141-2A>G variant was present only once in gnomAD for a frequency of 6.2e-7, and the other 2 splice-site variants had no population frequency. In contrast, nearby variants within the same intron with higher population frequencies had no predicted effect on splicing (Fig. 1D). For example, c.141-12 T>G, c.821-17C>T, and c.1524-8dup are present 7, 66, and 29 times in gnomAD, respectively, and are not expected to perturb splicing (Fig. 1D). Therefore, the ultrarare BMAL1 intronic variants we present here are likely to affect splicing in contrast to more common variants.

Rare Variants in BMAL1 Are Associated with a Neurodevelopmental Phenotype.

For individuals harboring rare BMAL1 variants, we found the most prevalent phenotypic characteristics were neurodevelopmental features. All the individuals for whom phenotypic information was available (8/8; 100%) exhibited developmental delay (SI Appendix, Tables S2 and S3). For at least four individuals (4/8; 50%) this manifested as global developmental delay, indicating that two or more domains of development were involved. Speech and motor skills were most often affected in the mild-to-moderate range. Seizures were noted in 50% of individuals (3/6) and autism spectrum disorder was noted in 6 individuals. Attention-deficit/hyperactivity disorder was noted in two individuals, as were heart murmurs. Given that BMAL1 plays an important role in the molecular clock, we attempted to further characterize sleep or circadian phenotypes. We were able to obtain phenotypic information on sleep patterns in 7/10 individuals. Within this subcohort, 43% (3/7) exhibited sleep challenges.

Interestingly, 33% (2/6) of individuals for whom relevant history was available in this cohort exhibited a marfanoid habitus. In Individual six, this was described as a tall stature with dolichostenomelic features. An additional 33% (2/6) exhibited joint hypermobility with hypotonia. Thus, a musculoskeletal phenotype manifest as marfanoid habitus or joint hypermobility was present in 66% (4/6) of this cohort and may be a common feature of this clinical syndrome.

For the variants for which inheritance could be confirmed, 71% (5/7) were de novo (SI Appendix, Table S1). The BMAL1 c.141-2A>G variant in Individual #1 was inherited from a mother who had an unremarkable medical history other than postpartum depression. Individual #1’s medical history was remarkable for a) an elder sister with learning difficulties and a history of infantile spasms, b) a maternal cousin with autistic symptoms and motor coordination difficulties, and c) another maternal cousin with learning difficulties and a mood disorder. These other family members were not available for genetic testing. The BMAL1 p.(Lys405Ter) variant in Individual #6 was inherited from a father who had a history of learning disability and tics and was unable to read and write. Parental testing was not possible in Individuals #4 and #9 because the parents were deceased or not available and in Individual #10 because the child was adopted.

Other variants of unclear significance were identified in some of the individuals presented here but none provided a single unifying diagnosis (see details in SI Appendix, Table S4).

Rare Heterozygous BMAL1 Variants Perturb Protein Function.

To test the functional impact of the identified BMAL1 variants, we used a cell-based reporter assay. A well-established target of BMAL1 is Per2, whose BMAL1-dependent rhythmic expression can be measured in cultured cells (11). Human U2OS cells stably expressing a Per2-promoter driven luciferase reporter (pPer2-dLuc) were previously created (12). Then, using CRISPR-Cas9 gene editing, we attempted to create heterozygous mutant lines that harbor each of the 10 BMAL1 variants identified. Through multiple attempts, we were able to generate clonal lines for 9/10 variants. We were unable to introduce the BMAL1 p.(Lys405Ter) variant due to lack of specific gRNA.

Circadian rhythms can be synchronized across individual U2OS cells by transient exposure to dexamethasone (11). We measured bioluminescence of Per2-dLuc for 6 d following dexamethasone treatment, and computed circadian period, phase, and amplitude using BioDare2 (13). As compared to mock transfected wild-type cells (BMAL1^+/+), two heterozygous BMAL1 frameshift variants, p.(Tyr340CysfsTer35) and p.(Thr480HisfsTer22), led to a significant shortening of the period from 24.4 h to 23.6 h and 23.5 h, respectively (Fig. 2). The 3 heterozygous splice-site BMAL1 variants, c.141-2A>G, c.821-1G>C, and c.1524-3C>A, led to a phase shift. In addition, 6 variants led to a dampening of the magnitude of luminescence, including c.141-2A>G, c.821-1G>C, p.(Tyr340CysfsTer35), p.(Glu501Lys), and p.(Gly535GlufsTer); this is consistent with decreased Per2 transcription and a loss-of-function mechanism. Interestingly the BMAL1 p.(Ile201Thr) missense variant led to a significant enhancement of signal magnitude, consistent with a gain-of-function change. To more directly assess the strength of circadian cycling independent of changes in magnitude of expression, we normalized luminescence intensity to the nadir of the first 24 h for each variant and found the underlying amplitude of cycling was either preserved or enhanced as compared to control (SI Appendix, Fig. S1). Together, these data demonstrate that variant BMAL1 decreased Per2 expression but did not decrease the amplitude of circadian cycling. We were unable to create a heterozygous BMAL1 p.(Gly68Arg) line, but even in the homozygous state no phenotype was observed. In total, we observed that 8/9 of the rare BMAL1 variants tested disrupted protein function.

Fig. 2.

U2OS cells expressing Per2-dLuc reporter and BMAL1 variants reveal altered BMAL1 function. (A) Raw luminescence signal after dexamethasone synchronization was recorded for 6 d. The same genetic control condition BMAL1^+/+ is plotted in each trace to aid in comparison to variants. Traces indicate average values and thickness of the line depicts SE of the mean. n = 5 to 9 experiments/condition. (B) Circadian parameters calculated through BioDare2. One-way ANOVA with Benjamini, Krieger, and Yekutieli’s two-stage step-up method to control the false discovery rate for multiple comparisons made to the control genotype. *P < 0.05 **P < 0.01 ***P < 0.001. Data are presented as mean values ± SEM.

The conversion of luciferin to oxy-luciferin by luciferase is an ATP-dependent process, so if mutations in BMAL1 alter ATP levels, rhythms in luminescence might be affected. Therefore, we sought to directly measure PER2 cycling. We synchronized U2OS cells containing BMAL1 variants then quantified PER2 mRNA cycling to assess for acute effects postsynchronization. Rhythm amplitude was calculated using BioDare2 (13) and revealed changes in rhythm amplitudes in 7/9 mutant lines tested (Fig. 3). A phase shift was again identified in U2OS cells harboring the BMAL1 p.(Ile201Thr) missense variant (Fig. 3). Consistent with the decreased magnitude of luminescence from the Per2-dLuc assay, we identified decreased levels of PER2 mRNA at various timepoints in cells containing BMAL1 c.141-2A>G, c.821-1G>C, p.(Tyr340CysfsTer35), p.(Glu501Lys), and p.(Gly535GlufsTer) variants, as well as the p.(Gly68Arg)/p.(Gly68Arg), p.(Thr480HisfsTer22), and c.1524-3C>A conditions. Paralleling the increased magnitude of luminescence in the Per2-dLuc assay, the BMAL1 p.(Ile201Thr) variant was the only to significantly increase PER2 transcript levels at any timepoint. We sought to validate these results for another transcriptional target of BMAL1, NR1D1, and found similar results (SI Appendix, Fig. S3). The BMAL1 p.(Ile201Thr) variant again increased the amplitude of NR1D1 rhythms, while the p.(Tyr340CysfsTer35) variant decreased NR1D1 mRNA. Note that the frequency and duration of sampling for RNA cycling experiments, relative to the luciferase assays above, makes it difficult to detect small changes in phase or period. These data are consistent with the Per2-dLuc reporter assay and provide evidence that all 9 rare BMAL1 variants affect protein function such that the magnitude and/or cycling of PER2 is disrupted within 24-h of synchronization.

Fig. 3.

Altered PER2 expression caused by BMAL1 variants. qPCR results PER2 expression at six timepoints through the day in wild-type and variant BMAL1 U2OS cells. Daily oscillation of PER2 mRNA in control cell is in gray and replotted in each trace to allow for comparison to variant lines. Circadian parameters were calculated through BioDare2 and significant results are listed in respective graphs. n = 3 samples/timepoint/condition. One-way ANOVA with Benjamini, Krieger, and Yekutieli’s two-stage step-up method to control the false discovery rate for multiple comparisons made to the control genotype. To compare mRNA levels independent of circadian metrics, a mixed-effects model with Geisser–Greenhouse correction and Dunnett’s multiple comparisons test were used. *P < 0.05 **P < 0.01 ***P < 0.001. Data are presented as mean values ± SEM.

BMAL1 Expression and CLOCK Interaction in Mutant Cell Lines.

Effects of the rare BMAL1 variants tested above on circadian cycling prompted us to ask whether these variants affected BMAL1 protein expression and interaction with CLOCK. The heterozygous BMAL1 c.141-2A>G, p.(Tyr340CysfsTer35), and p.(Gly535GlufsTer) variants led to a ~50% reduction of protein expression (Fig. 4), consistent with the variant mRNAs undergoing nonsense-mediated decay. While the BMAL1 c.821-1G>C, p.(Thr480HisfsTer22), and p.1524-3C>A variants were also predicted to reduce BMAL1 expression, a normal amount of protein was observed. As expected, the missense variants did not change BMAL1 total protein expression. Coimmunoprecipitation studies of CLOCK and BMAL1 demonstrated that there were no changes in binding (Fig. 4). Additionally, immunohistochemistry of U2OS cells harboring these rare BMAL1 variants did not reveal mislocalization of BMAL1 or CLOCK (SI Appendix, Fig. S2). Overall, the c.141-2A>G, p.(Tyr340CysfsTer35), and p.(Gly535GlufsTer) variants decrease BMAL1 expression, and none of the variants markedly impair binding to CLOCK or cellular localization.

Fig. 4.

BMAL1 expression and interaction with CLOCK. (A) Representative immunoblots depicting expression of BMAL1 and CLOCK, as well as β-actin loading control, in wild-type, and variant BMAL1 cell lines (Left). Immunoprecipitation of CLOCK and probing for BMAL1 demonstrates preserved protein–protein interactions (Right). (B) Quantification of BMAL1 band intensity in input as normalized to β-actin loading control reveals decreased BMAL1 expression caused by BMAL1 c.141-2A>G, p.(Tyr340CysfsTer35), and p.(Gly535GlufsTer) variants. n = 3 replicates/condition. One-way ANOVA with Holm–Sidak’s multiple comparisons test. ***P < 0.001. Data are presented as mean values ± SEM.

Flies Expressing Orthologous BMAL1 Variants Exhibit Altered Behavioral Rhythms.

Given that the rare BMAL1 variants presented here perturb protein function, we sought to understand how these variants might affect behavior by modeling variants in the Drosophila ortholog cycle (cyc). BMAL1 and cyc are highly homologous with a DIOPT score of 13 (Fig. 5A). As intronic splice-site variants, PAC domain variants, and C-terminal variants in BMAL1 are not well conserved in cyc, we focused our attention on the BMAL1 p.(Ile201Thr) variant, which increased circadian amplitude consistent with a gain-of-function mechanism, and the BMAL1 p.(Tyr340CysfsTer35) variant, which decreased circadian amplitude consistent with a loss-of-function mechanism. Protein alignments demonstrated that BMAL1 p.Ile201 is orthologous to cyc p.Ile161, and BMAL1 p.Tyr340 is orthologous to cyc p.Phe311. Tyrosine (Y) and phenylalanine (F) are closely related amino acids with hydrophobic side chains differing by a single hydroxyl group. Thus, we generated cyc variants p.I161T and p.F311fs, which are equivalent to the corresponding BMAL1 variants, in a cyc transgenic construct. We then created flies harboring wild-type cyc under UAS control (UAS-cyc^WT), and mutant lines harboring either cyc p.I161T or cyc p.F311fs under UAS control (UAS-cyc^I161T and UAS-cyc^F311fs respectively). These flies were then crossed to homozygous cyc mutant flies (cyc⁰¹/cyc⁰¹) (14), which are arrhythmic.

Fig. 5.

Flies harboring orthologous cyc variants exhibit circadian alterations. (A) cycle is the fly ortholog of BMAL1 with a DRSC Integrative Ortholog Prediction Tool (DIOPT) score of 13, indicating high homology. The BMAL1 p.I201T variant in the PAS1 domain corresponds with the cycle p.I161T variant in the PAS1 domain. The BMAL1 p.Y340CfsTer35 variant in the PAS2 domain corresponds with the cycle p.F311fs variant in the PAS2 domain. (B) Representative actograms demonstrating that loss-of-function cyc⁰¹/cyc⁰¹ flies lack rhythmic behavior in total darkness, but when wild-type cyc (UAS-cyc^WT) or cyc p.I161T (UAS-cyc^I161T) is expressed, behavioral rhythmicity is improved in light–dark cycles. Yellow shading indicates presence of light and resumption of ZT9-9 light:dark cycle. (C) Quantification of behavioral rhythmicity. Left: percentage of flies demonstrating rhythmic behavior as measured by an FFT value above 0.01. Middle and Right: FFT values for flies with rhythmic behavior in total darkness (DD) and light–dark cycles (LD). Note that rhythmic flies were rarely seen in the cyc⁰¹/cyc⁰¹;tim-Gal4 and cyc⁰¹/cyc⁰¹;tim-Gal4> UAS-cyc^F311fs conditions in DD, so FFT values are not provided for the Middle panel. n = 82-96 flies/condition. The Mann–Whitney test was used when comparing two groups given lack of normal distribution. The Kruskal–Wallis test was used when comparing three or more groups with Dunn’s multiple comparisons test given lack of normal distribution. *P < 0.05. Data are presented as mean values ± SEM.

In complete darkness (DD), cyc⁰¹/cyc⁰¹ mutants expressing the tim-Gal4 driver were arrhythmic using a fast Fourier transform (FFT) value of 0.01 as a cutoff for rhythmicity (Fig. 5C). The tim-Gal4 driver targets cells expressing molecular clock genes, including the ~240 brain clock neurons. When wild-type cyc or cyc p.I161T expression is driven with the tim-Gal4 driver, although full rescue is not achieved, some flies become rhythmic in constant darkness (Fig. 5C). FFT values were marginally, though significantly, higher in rhythmic flies expressing cyc p.I161T in DD, relative to those expressing wild type cyc, consistent with a gain-of-function mechanism and cell culture results (Fig. 5C). In contrast, driving cyc p.F311fs expression in DD did not improve rhythmicity, consistent with a loss-of-function mechanism and cell culture results (Fig. 5C). Upon shifting from DD to a 12-h light:12-h dark cycle (LD), only 31% of cyc⁰¹/cyc⁰¹ mutants exhibit rhythmic behavior (Fig. 5C), consistent with previous results demonstrating impaired behavioral rhythmicity in cyc mutants even in cycling light:dark conditions (14). A greater percentage of mutant flies expressing wild-type cyc or cyc p.I161T demonstrated rhythmic activity in light:dark conditions (Fig. 5 B and C). Only 8% of mutant flies expressing cyc p.F311fs were rhythmic in light:dark cycles (Fig. 5 B and C), which was less than levels seen in cyc⁰¹/cyc⁰¹ mutant flies, and therefore consistent with a severe loss-of-function or possibly dominant negative mechanism. Overall, these in vivo experiments largely recapitulate both gain- and loss-of-function mechanisms observed in culture.

Orthologous BMAL1 Variants Impair Memory.

In our cohort, we found that, for all the individuals for whom clinical information was available, all experienced developmental delay. Therefore, we sought to understand whether these variants disrupt short- and long-term appetitive memory. We found that cyc⁰¹/cyc⁰¹ mutant flies were able to form normal short- and long-term memory (Fig. 6 A and B). Driving wild-type cyc with the tim-Gal4 driver in cyc⁰¹/cyc⁰¹ mutant flies did not significantly change memory, but driving either cyc p.I161T or cyc p.F311fs significantly suppressed short- and long-term memory as compared to controls. Thus, cyc gain-of-function through cyc p.I161T expression improves circadian rhythms, but impairs memory. On the other hand, cyc p.F311fs is deficient in its ability to drive circadian rhythms and also results in impaired memory, supporting the idea that this variant exerts a toxic dominant-negative effect.

Fig. 6.

Short- and long-term memory is impaired in flies expressing variant cycle. (A) Quantification of short-term memory in cyc⁰¹/cyc⁰¹ flies expressing wild-type cyc (UAS-cyc^WT), cyc p.I161T (UAS-cyc^I161T), or cyc p.F311fs (UAS-cyc^F311fs) using the tim-Gal4 driver. cyc p.I161T and cyc p.F311fs disrupts short-term memory. (B) Quantification of long-term memory in cyc⁰¹/cyc⁰¹ flies expressing wild-type cyc, cyc p.I161T, or cyc p.F311fs using the tim-Gal4 driver. cyc p.I161T and cyc p.F311fs impairs long-term memory. n = 7 to 8 replicates/condition. One-way ANOVA with Tukey’s multiple comparisons test. Data are presented as box and whiskers and showing all points.

Discussion

We identified 10 individuals with neurodevelopmental features, sleep dysfunction, and musculoskeletal symptoms associated with ultrarare variants in BMAL1. Modeling 9 of these variants in cell culture demonstrates disrupted BMAL1 transcriptional activity. Additional study of 2 of these variants conserved in Drosophila revealed both gain-of-function and dominant negative changes that affect behavioral rhythms and learning and memory. Thus, pathogenic variants in core clock genes can contribute to neurodevelopmental phenotypes.

We find that all 9 of the rare BMAL1 variants tested affected protein function to varying degrees. The 3 BMAL1 splice site variants spread throughout the gene advanced the circadian phase and decreased PER2 transcription as measured through the PER2-dLuc reporter assay, consistent with a loss-of-function mechanism (Fig. 2). The decrement in PER2 mRNA expression was confirmed within the first 24 h after synchronization predominantly at or after ZT12 (Fig. 3). The 3 missense variants produced the most varied effects: BMAL1 p.(Gly68Arg)/p.(Gly68Arg) and p.(Glu501Lys) produced partial loss-of-function changes, while the p.(Ile201Thr) variant increased both molecular and behavioral rhythmicity, consistent with a gain-of-function mechanism. Given that the BMAL1 p.(Ile201Thr) residue is in the PAS1 domain, which directly contacts CLOCK, it is possible that the p.(Ile201Thr) variant enhances dimerization with CLOCK to increase PER2 transcription. Alternatively, given that we did not detect enhanced dimerization, it may decrease interaction with PER proteins. While BMAL1 p.(Ile201Thr) enhanced circadian rhythmicity it, however, reduced memory consolidation (Fig. 6). The BMAL1 p.(Gly68Arg)/p.(Gly68Arg) variant produced the weakest phenotype even in the homozygous state. Finally, the 3 frameshift variants all decreased PER2 transcription consistent with loss-of-function. The BMAL1 p.(Tyr340CysfsTer35) and p.(Thr480HisfsTer22) variants, occurring in the PAS2 and PAC domains, shortened the circadian period and therefore may impair interaction with CLOCK, although such impairment is likely small as it was not detected in our co-IP experiment. Both the advancement of circadian phase associated with splice site variants and the shortening of circadian period due to frameshift shift variants can be caused by decreased PER2 protein levels, as has been previously suggested (15). As in the case of p.(Ile201Thr), the p.(Tyr340CysfsTer35) and p.(Thr480HisfsTer22) variants could also affect interaction with PER/CRY, in this case increasing interaction to promote feedback and thereby decrease PER2 RNA levels. In summary, our cell culture results reveal that rare BMAL1 variants cause both loss-of-function and gain-of-function changes.

When attempting to rescue the cyc⁰¹ mutant phenotype by driving wild-type cyc (UAS-cyc^WT) using the tim-Gal4 driver, we did not observe a full rescue (Fig. 5). This is similar to previous attempts to rescue core clock components including per, tim, and Clk using neuronal and clock drivers that have failed to provide a full rescue of locomotor rhythms (16, 17). A possible explanation is either too high or too low Gal4 expression, which then would affect molecular rhythms. Beyond molecular rhythms, cyc plays an important role in non-clock-mediated cellular morphogenesis. If cyc is not rescued to wild-type levels in clock neurons early in development, then axonal projections are abnormal and predicted to limit communication with downstream partners (18). We note too that the developmental timing and spatial expression of tim-Gal4 could be somewhat different from that of endogenous cyc. The lack of a full rescue we observe is less likely due to rhythmic tim-Gal4 expression, as the Gal4 protein is expected to be stable and not cycle.

Nevertheless, our behavioral data from flies support the pathogenicity of conserved BMAL1 variants. Consistent with our cell culture results, the cyc p.I161T variant improved rhythmicity over control levels suggesting a gain-of-function mechanism. Modeling of the p.(Tyr340CysfsTer35) variant in the fly (cyc p.F311fs) led to a reduction in the percentage of rhythmic flies in light–dark conditions and impaired memory as compared to loss-of-function mutants, possibly consistent with a toxic dominant-negative effect. Here, the loss-of-function change in cell culture is caused by haploinsufficiency after nonsense mediated decay, whereas the toxic dominant-negative effect in vivo is likely due to forced expression of a protein fragment. Both gain-of-function and dominant-negative variants impaired short- and long-term memory in the fly. In addition to their role in the molecular clock, both CYC and BMAL1 play a role in non-clock-mediated cellular and molecular mechanisms. For example, loss of cyc disrupts neuronal projections in clock neurons (18), and BMAL1 modulates the activity of other transcription factors, including ETS and N-MYC (19). Thus, it remains unknown if the memory impairment we observe results from a molecular timekeeping effect of cyc or nonrhythmic transcriptional regulation of genes involved in development and/or memory.

While common population variants in BMAL1 have been associated with a variety of human disorders, we now implicate rare heterozygous pathogenic variants in BMAL1 as a cause of a Mendelian neurodevelopmental disorder. Genome-wide association studies (GWAS) reveal that variants in BMAL1 are associated with insulin resistance (20, 21), cardiovascular health (21, 22), prostate cancer (23), lung cancer (24), chronotype (25), sociability (26), and neurodegeneration (27). These common variants may affect BMAL1 function through unclear mechanisms, and likely lead to relatively small effects on associated phenotypes. In contrast, the rare variants presented here have no, or little, population frequency (SI Appendix, Table S1), and were predicted to perturb BMAL1 function through in silico analyses (SI Appendix, Table S1 and Fig. 1C). Indeed, functional testing in cultured cells, and flies, confirms these analyses (Fig. 2).

Examination of DECIPHER (Database of Chromosomal Imbalance and Phenotype in Humans Using Ensembl Resources) (28) for copy number variants (CNVs) reveals 11 heterozygous microdeletions ranging in size from 2.80 Mb to 134.18 Mb. These CNVs are de novo and associated with small size at birth (6/11), intellectual disability (4/11), and arachnodactyly (2/11), among other dysmorphic features. This cohort includes an individual with a heterozygous, de novo 4.26 Mb microdeletion involving BMAL1 and associated with autism spectrum disorder and arachnodactyly, a marfanoid feature. While these features might be consistent with the cohort we present here, microdeletions involving BMAL1 also include adjacent genes implicated in neurodevelopmental syndromes and sensitive to haploinsufficiency including RRAS2 and SOX6.

Consistent with our culture findings, heterozygous cyc and Bmal1 mutant animals have been previously demonstrated to exhibit a spectrum of phenotypes. Heterozygous cyc^0/+ mutant flies exhibit behavioral rhythms, but the period of these rhythms is about 1-h longer than control flies (14). Heterozygous Bmal1^+/- mice, similarly, exhibit intact behavior rhythms (29) with subtle phenotypes. Autism-like behaviors, including altered vocalizations, deficits in socialization, and anxious behavior have now been identified in Bmal1^+/- mice (30). While Bmal1^+/- mice exhibit intact behavioral rhythms, mice heterozygous for a C-terminus truncated Bmal1 (Bmal1^+/GTΔC) exhibit a gradual dampening of rhythms through a presumed dominant-negative mechanism (31). The BMAL1 p.(Gly535GlufsTer)/+ variant we present here mirrors this C-terminus truncated allele most closely and also demonstrates a clear dampening of rhythms (Fig. 2). However, the phenotype could just reflect loss-of-function, given the ~50% decrease in BMAL1 expression likely due to nonsense mediated decay (Fig. 4), leading to haploinsufficiency.

We find that 3/6 individuals, for whom history was available, exhibited seizures (SI Appendix, Tables S2 and S3). Electrically induced seizures are more likely to occur with less current injection in Bmal1 knockout mice, indicating increased seizure susceptibility. The drivers of increased seizure risk, and whether this occurs through molecular clock-dependent or -independent mechanisms, remain unclear. BMAL1 has tissue-specific functions independent of molecular clock function (32), and this may perhaps provide an explanation for the musculoskeletal phenotypes, including marfanoid habitus, seen in our cohort.

Sleep difficulties are highly prevalent in children with autism spectrum disorder, affecting up to 80% (33, 34). The nature of these sleep challenges is variable, and may include difficulty with sleep onset, maintenance, consolidation, timing, and/or quantity. Monogenic etiologies for alterations in sleep timing (1–4) and quantity (35–39) are rapidly being identified, largely through genetic sequencing of families exhibiting strong sleep or circadian phenotypes. Several variants affecting sleep timing occur in well-characterized circadian clock genes, including PER2 (1), PER3 (2), CRY1 (4), and CRY2 (3). Perhaps unsurprisingly, variants in PER2 have now been identified in people with autism spectrum disorder with variable sleep phenotypes, including sleep timing (40). This begs a question related to causation: Is it possible that sleep dysfunction early in development can cause neurodevelopmental signs and symptoms related to autism spectrum disorder? Correlations have been drawn between the severity of autistic behaviors and sleep/circadian dysfunction (34). In our own cohort, given that BMAL1 is a core component of the molecular clock, we hypothesized that there would be a high likelihood of sleep or circadian dysfunction. Instead, we found that sleep-related diagnoses were rarely made in children with autism spectrum disorder, and only 3 of 7 individuals in whom the information was available reported sleep difficulty. As has been previously reported, although few children with autism spectrum disorder received sleep-related diagnoses, sleep-related disturbances may be more common (40) and possibly consistent with an underdiagnosis.

An exciting corollary of our findings is the possibility that early and aggressive treatment of sleep or circadian disruption might decrease the severity of neurodevelopmental disease. We find that disruption of a core component of the molecular clock causes memory impairment. Future studies focused on understanding whether improvement of behavioral rhythmicity can improve neurodevelopment are warranted. If correction of circadian behaviors like sleep can improve neurodevelopment, this has broad applicability to neurodevelopmental disorders beyond those caused by pathogenic variants in molecular clock genes.

Materials and Methods

Cell Culture.

U2OS cells were cultured in Dulbecco’s modified Eagle’s medium supplemented with 10% FBS, 1% L-glutamine, and 1% penicillin-streptomycin (Thermo Fisher Scientific) at 37 °C under 5% CO2. Mycoplasma testing was performed and found to be negative.

Generation of U2OS Cells Expressing Per2 Promoter-Driven Luciferase and BMAL1 Variants.

As previously reported (11, 12), the pPer2-dLuc lentiviral construct was gifted by Andrew C. Liu at the Department of Physiology and Functional Genomics, University of Florida. U2OS cell reporter lines stably expressing pPer2-dLuc were generated using lentivirus-mediated gene delivery and established transduction protocols (41).

U2OS cells stably expressing pPer2-dLuc were then used for knock-in edits. Single guide RNA (sgRNA) and single-stranded oligodeoxynucleotides (ssODNs) were designed and transfected by Synthego to generate edited lines also expressing pPer2-dLuc. The Inference of CRISPR Edits tool (Synthego) was used to analyze the efficiency of CRISP-mediated edits. All sgRNA, ssODNs, and PCR primers to confirm edits are provided in SI Appendix, Table S5.

Bioluminescence Recording/LumiCycle Assay.

Real-time bioluminescence of Per2 or Bmal1 dLuc reporter U2OS cells after synchronization with 1 µM Dexamethasone (Sigma) were monitored using a LumiCycle luminometer (Actimetrics, Wilmette, IL, United States) as previously described (42).

qPCR.

Total RNA was extracted from all cell lines at 6 different time points using TRIzol reagent (Thermo Fisher Scientific). RNA was reverse transcribed to generate cDNA using a High-Capacity cDNA Reverse Transcription kit (Thermo Fisher Scientific). qPCR was performed on a ViiA 7 Real-Time PCR System (Applied Biosystems) using SYBR Green PCR master mix (Thermo Fisher Scientific). GAPDH was used as a control. Primers (5′ to 3′) for qPCR used in the study are as follows:

Per2_Forward: GGATGCCCGCCAGAGTCCAGAT

Per2_Reverse: TGTCCACTTTCGAAGACTGGTCGC

NR1D1_Forward: CTGCCAGCAATGTCGCTTCAAG

NR1D1_Reverse: TGGCTGCTCAACTGGTTGTTGG

GAPDH_Forward: GTCTCCTCTGACTTCAACAGCG

GAPDH_Reverse: ACCACCCTGTTGCTGTAGCCAA

BMAL1 and CLOCK Immunostaining.

Cells were cultured on Nunc™ Lab-Tek™ II CC2™ Chamber Slide System (Thermo Scientific Catalog # 12-565-2), and then fixed with 4% paraformaldehyde (PFA) 30 min at RT, permeabilized by washing 0.3% Triton X-100 in PBS 3 times (5 min a time). Cells were blocked using 3% NDS in PBS (blocking buffer) for 20 min RT. Cells were incubated with primary Abs in blocking buffer at RT for 1 h. Mouse anti-BMAL1 (Santa Cruz) and rabbit anti-CLK (Abcam) were used. Cells were washed with PBST (PBS + 0.1% Tween-20) for three times (10 min a time). Cells were then incubated with the following secondary Abs for 1 h: goat anti-Mouse IgG (H + L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor™ Plus 488 (Thermo Fisher Scientific) and Goat anti-Rabbit IgG (H + L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor™ Plus 594 (Thermo Fisher Scientific). Cells were then washed with PBST 3 × 10 min, and finally once with 1X PBS. Cells were then mounted using mounting medium (VECTASHIELD Antifade Mounting Medium with DAPI, Vector Laboratories). The slide was cured overnight at room temperature in the dark and imaged the following day under a confocal microscope (Leica Microsystems).

Coimmunoprecipitation and Immunoblotting.

Cells were lysed with ice-cold lysis buffer containing 10 mM Tris-Cl pH7.5, 100 mM NaCl, 5 mM EDTA, 1% Triton X-100 and 0.05% SDS, and Protease inhibitor cocktail (Roche Diagnostics). Total protein concentration was measured using a DC Protein Assay kit (Bio-Rad). 1 µg of rabbit anti-CLK IgG (Abcam) and 50 µL of washed Protein A DynaBeads (Thermo Fisher Scientific) were incubated with 500 µg of total protein for 4 h at 4 °C, and then washed 6 times with ice-cold washing buffer (10 mM Tris-Cl pH7.5, 150 mM NaCl, 5 mM EDTA, 0.4% Triton X-100 and 0.05% SDS, Protease inhibitor cocktail). Co-IP samples and 10 µg input samples were loaded into wells of 15-well 4 to 12% Tris-Glycine precast gels (Thermo Fisher Scientific). Protein was separated through SDS-PAGE, transferred onto a nitrocellulose membrane, and incubated with primary antibodies including mouse anti-BMAL1 monoclonal antibody (Santa Cruz), rabbit anti-CLK antibody (Abcam) and mouse anti-βActin antibody (R&D Systems), and horseradish peroxidase conjugated secondary antibodies (Jackson ImmunoResearch). Following chemiluminescence, blots were exposed to X-ray film. Images were digitized using a scanner then imported into Image J for processing and quantification of blot intensity. In Image J, images were converted to grayscale and regions of interest were drawn around each band to measure intensity. Relative BMAL1 band intensity was calculated by dividing the BMAL1 band intensity by the β-actin band intensity for each individual sample.

Mutagenesis of cyc and UAS-cyc Fly Generation.

For generation of UAS-cyc^WT, UAS-cyc^I161T, and UAS-cyc^F311fs flies, a vector with a cyc cDNA sequence was obtained from DGRC (GM02625) and used as a template. To create the cyc p.I161T and cyc p.F311fs mutations, we used a mutagenesis kit (NEB; E0554S) with the primers listed below. Mutated cyc sequences were validated with Sanger sequencing in the University of Pennsylvania DNA core facility. Wild type or mutated cyc cDNA were then subcloned to pUASt-AttB vector (DGRC 1419), and transgenic flies were generated by BestGene Inc. (Chino Hills, CA) with the PhiC31 injection method.

cyc_Forward: GGGCTCGAGATGGAAGTTCAGGAGTTCTGCG

cyc_Reverse: GGGTCTAGATTATAAGAACACGGAATTCTTGGCG

I161T_Forward: TCCGAAGGACACCGGCAAGGTTAAG

I161T_Reverse: TGCAGGACGTCGAAC

F311fs_Forward: TCTATCTCGCGCCACTCCGG

F311fs_Reverse: GCACGTGCCGTATGTTCGGGTGATTG

Drosophila Memory Assays.

Appetitive conditioning was performed as previously described (43, 44).

For short-term memory assays, 4- to 7-d-old mixed-sex population of ~100 flies were starved (on 2 mL of 2% agar in vials) for 18 h and trained in a dark room with red light at 25 °C and 70% relative humidity. All experiments were conducted under vacuum pressure in a memory wheel, which can conduct 4 individual experiments at once. In brief Whatman paper, cut into 1-inch squares were soaked in either 1.5 M sucrose or milliQ water. Flies were exposed to odor A + sucrose for 2 min, followed by clean air for 30 s, and odor B + water for 2 min followed by clean air for 30 s. Following training, flies were immediately tested for short term memory. During short term memory training, flies were exposed to both odors at once for 2 min. Flies were then collected, placed onto complete media, and counted. Flies remained on complete media for 3 to 5 h following training and testing.

For long-term memory assays, the same flies trained and tested in short term memory were restarved on 2% agar (as above) for 18 h and tested for long term memory. Flies were exposed to both odor A and B for 2 min and collected depending on the choice that was made. Flies were then counted and discarded.

In both short-term and long-term memory assays, odor A = 3 mm odor cups supplemented with 80 uL of 3-octanol (1:80) and odor B = 4-methylcyclohexanol (1:200) in paraffin oil. Choice index (CI) was calculated as the number of flies selecting CS+ odor minus the number of flies selecting CS- odor divided by the total number of flies. Each CI is averaged with its reciprocal training counterpart to determine the Performance Index (PI), which minimizes nonassociative effects.

Quantification and Statistical Analyses.

For datasets involving 2 groups and not normally distributed, a Mann–Whitney test was implemented to detect statistical differences. For datasets involving three or more groups and not normally distributed, we used a Kruskal–Wallis test with Dunn’s correction for multiple comparisons. For datasets involving three or more groups and reasonably assumed to be approximately normally distributed, we used a one-way ANOVA with correction for multiple comparisons.

To identify quantify circadian period, phase, and amplitude in the Per2-dLuc assays, we used BioDare2 to implement a Fast Fourier Transform (FFT-NLLS)(13)

For the qPCR dataset, BioDare2 was again implemented to calculate circadian parameters. For identification of differences between groups (Genetics versus Timepoint), a mixed-effects model with a Geisser–Greenhouse correction was used, as sphericity was not assumed, as well as Dunnett’s multiple comparisons test.

To visually represent that data, mean and SE of the mean were used. Additional details regarding statistical testing, sample size, and P-values are provided in the Figure Legends. GraphPad Prism was used for all statistical analyses.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix.