MeCP2 regulates Gdf11, a dosage-sensitive gene critical for neurological function

Loss- and gain-of-function of MeCP2 causes Rett syndrome (RTT) and MECP2 duplication syndrome (MDS), respectively. MeCP2 binds methyl-cytosines to finely tune gene expression in the brain, but identifying genes robustly regulated by MeCP2 has been difficult. By integrating multiple transcriptomics datasets, we revealed that MeCP2 finely regulates growth differentiation factor 11 (Gdf11). Gdf11 is down-regulated in RTT mouse models and, conversely, up-regulated in MDS mouse models. Strikingly, genetically normalizing Gdf11 dosage levels improved several behavioral deficits in a mouse model of MDS. Next, we discovered that losing one copy of Gdf11 alone was sufficient to cause multiple neurobehavioral deficits in mice, most notably hyperactivity and decreased learning and memory. This decrease in learning and memory was not due to changes in proliferation or numbers of progenitor cells in the hippocampus. Lastly, loss of one copy of Gdf11 decreased survival in mice, corroborating its putative role in aging. Our data demonstrate that Gdf11 dosage is important for brain function.


Introduction
Intellectual disability (ID) and autism spectrum disorder (ASD) affect nearly 3% of children in the United States (Zablotsky et al., 2019). These diseases can be caused by mutations in any one of hundreds of genes (Deciphering Developmental Disorders Study, 2017;Satterstrom et al., 2020), a subset of which is 'dosage-sensitive', causing neurological dysfunction by either loss-of-function mutations or an increase in expression due to copy-number gain (Rice and McLysaght, 2017). Understanding how a dosage-sensitive gene drives molecular pathogenesis can reveal the mechanisms for two diseases, the loss-of-function and increased dosage disorders (Javed et al., 2020).
Methyl-CpG binding protein 2 (MECP2) is the exemplary dosage-sensitive gene. MECP2 is an X-linked gene whose loss-of-function causes Rett syndrome (RTT; OMIM: 312750) (Amir et al., 1999) and whose duplication causes MECP2 duplication syndrome (MDS; OMIM: 300260) (Lugtenberg et al., 2009;van Esch et al., 2005). Both RTT and MDS are devastating neurological disorders characterized by intellectual disability, motor dysfunction, and seizures. Despite identification of MECP2 as the causative gene for both disorders, we still have poor understanding of MeCP2-driven pathogenesis. MeCP2 is a methyl-cytosine binding protein that regulates gene expression in the brain Gabel et al., 2015;Nan et al., 1997). Disruption of normal MeCP2 function causes subtle dysregulation of thousands of genes (Sanfeliu et al., 2019). Importantly, many of these dysregulated genes are shared between mouse models of RTT and MDS, but their expression is altered in opposite directions (Chahrour et al., 2008;Chen et al., 2015). Interrogating transcripts that are highly sensitive to MeCP2 levels can reveal disease-driving genes that could be used to develop targeted therapies for RTT and MDS . Lastly, pinpointing the genes regulated by MeCP2 can enhance our understanding of molecular mechanisms of additional genes mediating neurological phenotypes (Lavery et al., 2020;Zhou et al., 2022).
To search for genes robustly regulated by MeCP2, we integrated gene expression studies that profiled RNA collected from mouse tissues bearing various MECP2 loss-or gain-of-function alleles and discovered that growth differentiation factor 11 (Gdf11) expression is highly and positively correlated with MeCP2 protein level and function. Gdf11 is a secreted ligand of the TGFβ superfamily most closely related to myostatin/GDF8 (Morikawa et al., 2016). Gdf11 was first identified as a regulator of the anterior/posterior patterning of the skeleton (McPherron et al., 1999). Further studies have revealed that Gdf11 negatively regulates neurogenesis both during development and in the adult brain (Mayweather et al., 2021;Wu et al., 2003). Gdf11 has also been implicated as a 'rejuvenating' factor that mitigates the effects of aging and as a tumor suppressor in multiple organs (Bajikar et al., 2017;Katsimpardi et al., 2014;Liu et al., 2018;Loffredo et al., 2013;Sinha et al., 2014). Recently, mutations in GDF11 have been associated with a developmental disorder with craniofacial, vertebral, and neurological abnormalities, but the consequences of altered Gdf11 dosage on mouse behavior had not been previously assessed (Cox et al., 2019;Ravenscroft et al., 2021).
In this study, we found that Gdf11 expression is increased in mouse models of MDS and that genetically normalizing Gdf11 dosage was sufficient to improve several behavioral deficits in a mouse model of MDS. Lastly, we found that loss of one copy of Gdf11 alone is sufficient to cause broad neurological dysfunction. Our study highlights the importance of revealing key molecules downstream of dosagesensitive genes by demonstrating that Gdf11 is regulated by MeCP2 and that Gdf11 dosage modifies neurological phenotypes.

Results
MeCP2 regulates the expression of Gdf11 in the mammalian brain Normalizing MeCP2 levels through genetic or pharmacological means is sufficient to rescue nearly all behavioral and molecular deficits in a mouse model of MDS (Sztainberg et al., 2015). For example, bolus injection of an antisense oligonucleotide (ASO) targeting MECP2 acutely reduces MeCP2 protein in MDS mice and ameliorates behavioral deficits (Shao et al., 2021b). We reasoned that genes whose expression was corrected after ASO treatment and whose abundance highly correlated with MeCP2 protein levels would be candidate genes to contribute to the behavioral rescue. To this end, we calculated the correlation coefficient between the fold-change of the 100 genes rescued by ASO treatment and the fold-change of MeCP2 protein levels in the hippocampi of MECP2 duplication mice (Shao et al., 2021b;Figure 1A). We found 16 genes whose fold-changes were highly correlated (Spearman's rho >±0.75) with MeCP2 levels (Figure 1-figure supplement 1A). To further refine these candidates, we flagged genes that were predicted to be loss-of-function intolerant in humans as these genes could have biological consequences if their expression level is mildly altered (Karczewski et al., 2020;Lek et al., 2016). Of the 16 MeCP2-correlated genes, we identified four genes -growth differentiation factor 11 (Gdf11), family with sequence similarity 131 member B (Fam131b), versican (Vcan), and forkhead box protein P1 (Foxp1) with a probability of loss intolerance (pLI) greater than 0.9 (Figure 1-figure supplement 1A).
Gene expression changes in Mecp2 mutant mouse models have been inconsistent in published reports (Sanfeliu et al., 2019). To test if any of the four MeCP2-correlated, loss-intolerant transcripts were robustly sensitive to MeCP2 levels, we queried expression changes in 20 readily available  Shao et al., 2021b). Spearman correlation coefficient between Gdf11 and MeCP2 is shown in the right panel. Gray interval represents a loess fit ± standard error. (B) Log 2 fold-change of GDF11 expression in excitatory and inhibitory neurons isolated from post mortem brains from individuals with Rett syndrome and Gdf11 expression in Mecp2-knockout (Mecp2-KO), mutant (Mecp2 MT), and MECP2 overexpression (MECP2 OE) mouse model RNA-seq experiments. Abbreviations used to describe studies are as follows: CTX: cortex, CB: cerebellum, HYP: hypothalamus, DG: dentate gyrus, HIP: hippocampus, Dup.: duplication, Trip.: triplication, AAV: adeno-associated virus. Grayed color indicates p adjusted >0.1. Studies are ordered from 1 to 20 and correspond to the study number in Supplementary file 1. (C) A Gdf11knockout first allele (Gdf11 tm2b ) has a LacZ cassette with splice acceptor and polyA sequence knocked in between exons 1 and 3 of Gdf11, and exon 2 of Gdf11 is deleted. Quantitative PCR of Mecp2, Gdf11, and the LacZ reporter from the cerebellum of wild-type, Gdf11 tm2b/+ , Mecp2 -/y , Gdf11 tm2b/+ ; Mecp2 -/y , MECP2-TG1, and Gdf11 tm2b/+ ; MECP2-TG1 (n=3-4 biological replicates per genotype). n.d. denotes the given gene expression measurement was not detected. qPCR data is shown as mean ± sem. (D) CUT&RUN profiling of MeCP2 binding in MECP2 transgenic mice (MECP2-TG1), wild-type, and Mecp2-knockout hippocampi (Mecp2-KO) at the Gdf11 locus. Black bar represents a MeCP2 peak called by MACS2 software, and the gray bar represents an expanded region of ±3.5 kb of increased MeCP2 binding. Log 2 fold-change of MeCP2 occupancy within the gray region is shown relative to wild-type, and the p-value of the comparison is shown beneath the fold-change. Tracks are displayed as an aggregate of biological replicates (n=3-6 biological replicates per genotype). (E) CUT&RUN profiling of H3K27me3 in wild-type and Mecp2-knockout (Mecp2-KO) at the Gdf11 locus. Log 2 foldchange of binding within the gray region is shown relative to wild-type, and the p-value of the comparison is shown beneath the fold-change. Tracks are displayed as an aggregate of biological replicates (n=3 biological replicates per genotype). (*) p<0.05, (****) p<0.0001 from one way ANOVA followed by multiple comparisons testing.
The online version of this article includes the following figure supplement(s) for figure 1:  transcriptional profiles generated in MECP2 perturbed rat (Veeraragavan et al., 2016), mouse, and human brain samples (Supplementary file 1). Strikingly, we found that Gdf11 was significantly altered in 19 of the 20 profiles, whereas the other three genes did not consistently exhibit significant difference compared to control ( Figure 1B and Figure 1-figure supplement 1B-D). GDF11 levels were downregulated two-fold in human postmortem RTT excitatory neurons, while Gdf11 levels were downregulated in the transcriptomes of Mecp2-null male mice, Mecp2-null heterozygous female mice , and a panel of MECP2 point mutant male mice in multiple brain regions (Jiang et al., 2021;Boxer et al., 2020;Baker et al., 2013); these data suggest Gdf11 is robustly sensitive to MeCP2 function. Additionally, we also found that Gdf11 was significantly upregulated in mouse models overexpressing MECP2 ( Figure 1B). We then investigated whether Gdf11 and Mecp2 expression were correlated among individual cell types in the brain. We mined a previous single-nucleus RNA-sequencing experiment in the normal mouse brain and found that the overall Spearman's correlation between Gdf11 and Mecp2 across all cell types was 0.65 (Saunders et al., 2018). Further, Gdf11 and Mecp2 were highly correlated in neuronal and glial cell types (astrocytes, microglia, neurons, and oligodendrocytes; Spearman's rho = 0.6-0.92) (Figure 1-figure supplement  1E). Within neurons specifically, we found that Gdf11 and Mecp2 were highly correlated in both excitatory neurons (Spearman's rho = 0.65) and inhibitory neurons (Spearman's rho = 0.76) (Figure 1figure supplement 1F). Given the strong dynamic correlation of Gdf11 with MeCP2 levels upon ASO treatment and with Mecp2 expression in individual brain cell types ( Figure 1A and Figure 1-figure supplement 1E and F), these data suggest Gdf11 is also sensitive to MeCP2 levels. Taken together, these results demonstrate Gdf11 is a transcript that is robustly sensitive to both MeCP2 function and protein levels.
To further investigate if MeCP2 transcriptionally regulates the Gdf11 locus, we took advantage of a Gdf11-knockout allele that has a LacZ reporter cassette knocked into the gene (Gdf11 tm2b ; Figure 1C). We crossed this allele with both a Mecp2-knockout (Guy et al., 2001) and a MECP2 duplication (MECP2-TG1) (Collins et al., 2004) mouse model to generate a series of Mecp2 or Gdf11 singlemutants and Mecp2; Gdf11 double-mutants and performed qPCR for Mecp2, Gdf11, and LacZ in the cerebellum, a brain region where Gdf11 is highly expressed in the mouse (Lein et al., 2007;Saunders et al., 2018). The expression of Mecp2 was not changed by reduction of Gdf11, but Gdf11 was significantly decreased in Mecp2-knockout males (p<0.05) and increased in MECP2-TG1 animals (p<0.0001). Furthermore, Mecp2-knockout trended towards a further decrease of Gdf11 in the Mecp2 -/y ; Gdf11 tm2b/+ double mutants while Gdf11 was restored to wild-type expression in the MECP2-TG1; Gdf11 tm2b/+ double mutants. Last, expression of the LacZ reporter was modulated by MeCP2 levels as well, being significantly downregulated in Mecp2-knockout (p<0.05) and upregulated in MECP2-TG1 animals (p<0.05) ( Figure 1C). These results demonstrate that the Gdf11 locus is transcriptionally sensitive to MeCP2 protein levels.
While MeCP2 has a well-described role as a transcriptional repressor and its loss leads to a de-repression of gene expression, MeCP2 loss has been shown to cause a down-regulation of gene expression (Ben-Shachar et al., 2009;Boxer et al., 2020;Chahrour et al., 2008;Chen et al., 2015;Clemens et al., 2020;Lyst et al., 2013;Nan et al., 1998;Nan et al., 1997;Samaco et al., 2012). We next sought to mechanistically understand how loss of MeCP2 leads to down-regulation of Gdf11. We first investigated whether MeCP2 has enriched binding near the Gdf11 locus. We performed Cleavage Under Targets & Release Using Nuclease (CUT&RUN) (Skene and Henikoff, 2017) to profile MeCP2 binding in wild-type, Mecp2-knockout, and MECP2-TG1 mouse hippocampus (Figure 1- figure  supplement 2A and B). We next identified peaks of enriched binding using the MACS algorithm (Zhang et al., 2008), which identified a peak upstream of the Gdf11 transcriptional start site (TSS). To further define the characteristics of this region, we mined existing histone acetylation data for marks of open chromatin (H3K27ac), promoter (H3K4me3), and putative enhancers (H3K4me1) (Halder et al., 2016). The peak of MeCP2 signal coincided with a region of high H3K27ac and H3K4me1, but low H3K4me3, suggesting that MeCP2 binds a putative enhancer of Gdf11 (Figure 1-figure supplement  2D). Due to the broad binding pattern of MeCP2 Skene et al., 2010), we integrated the signal in a symmetric window straddling the peak to robustly quantify MeCP2 occupancy. We observed a significant (p<0.01) increase of MeCP2 binding within this region upstream of Gdf11 in the MECP2-TG1 hippocampus. Importantly, the MeCP2 binding signal was almost completely absent in the Mecp2-knockout hippocampus in this same region ( Figure 1D). Concurrently, we profiled the repressive histone modification H3K27me3 (Figure 1-figure supplement 2C), which has been shown to interact with and be modulated by MeCP2 . Within the window of MeCP2 binding upstream of Gdf11, we found a significant (p<0.01) increase in H3K27me3 occupancy in Mecp2knockout tissue and a trend towards a reciprocal decrease in H3K27me3 occupancy in MECP2-TG1 tissue ( Figure 1E and Figure 1-figure supplement 2E). The increase in H3K27me3 in Mecp2knockout tissue corroborates the downregulation we observe in Gdf11 expression in transcriptional profiles collected from Mecp2-deficient models ( Figure 1B). These results suggest that MeCP2 binds upstream of Gdf11, at a putative enhancer, and modifies the chromatin landscape around the Gdf11 locus to regulate Gdf11 expression.
Genetic reduction of Gdf11 ameliorates behavioral deficits of MECP2 duplication mice Since Gdf11 dysregulation is rescued upon normalization of MECP2 dosage, we sought to determine if normalization of Gdf11 alone can rescue behavioral deficits in MECP2 duplication mice. We crossed the MECP2-TG1 allele with the Gdf11 tm2b allele to normalize Gdf11 expression ( Figure 1C). We generated a cohort of wild-type, MECP2-TG1, and MECP2-TG1; Gdf11 tm2b/+ double mutant littermates and behaviorally characterized these mice with an extensive battery of assays starting at 16 weeks of age ( Next, MECP2-TG1 mice experience hypoactivity and anxiety-like behavior compared to wildtype mice when placed in an elevated plus maze apparatus, traveling less distance and entering the open arms fewer times. In contrast, we observed the double mutant animals entered the open arms and crossed the maze arms like wild-type animals, and the double mutant mice traveled significantly further (p<0.01) on the maze compared to MECP2-TG1 mice ( Figure 2B). Lastly, MECP2-TG1 mice exhibit improved learning compared to wild-type mice in a shock-tone cued fear paradigm. Specifically, mice are trained to associate a tone with a foot shock and then placed in the same environment without tone (context) or a new environment with a tone (cue). Learning in this assay is measured by the amount of time the mouse freezes, where MECP2-TG1 mice freeze more in both context and cued tests. Normalization of Gdf11 led to wild-type levels of freezing during the training routine, and normalized the cued learning compared to MECP2-TG1 mice ( Figure 2-figure supplement 1C). Contextual learning was improved in double mutant mice with no significant difference observed compared to wild-type mice ( Figure 2C). Reduction of Gdf11 did not rescue all phenotypes present in MECP2-TG1 mice, as we did not observe a rescue in motor coordination on the rotating rod, change in the time spent in dark during the light-dark assay, nor any changes in sociability. However, we did observe rescue of total activity, as measured by distance traveled, in both light-dark and social assays ( Figure 2-figure supplement 1D-G). These data suggest that normalization of aberrant Gdf11 expression improves several behavioral deficits, most notably locomotion, general activity, and hippocampal learning in MECP2-TG1 mice.

Loss of one copy of Gdf11 causes neurobehavioral deficits in mice
The previous results raised the possibility that mouse neurological function is sensitive to changes in Gdf11 dosage. Supporting this evidence, loss of one copy of GDF11 function causes a developmental disorder with neurological features in humans (Ravenscroft et al., 2021). Furthermore, GDF11 is broadly and highly expressed in both the human and mouse brains, suggesting a potentially important role in brain function (Figure 3-figure supplement 1A and B; Li et al., 2017).
To determine if loss of one copy of Gdf11 resulted in neurobehavioral deficits, we generated a cohort of Gdf11 tm2b/+ mice to characterize with a comprehensive battery of tests. We first confirmed that both Gdf11 tm2b/+ and Gdf11 tm2b/tm2b mice exhibited abnormal anterior/posterior skeletal patterning by counting the number of attached and total ribs in neonatal pups. We observed that Gdf11 tm2b/tm2b pups had an abnormal number of 10 attached at 18 total ribs (n=4), while the Gdf11 tm2b/+ pups had   McPherron et al., 1999). These skeletal abnormalities, the observed perinatal lethality of Gdf11 tm2b/tm2b pups (McPherron et al., 1999), and the 50% reduction of Gdf11 in the brains of Gdf11 tm2b/+ mice ( Figure 1C) demonstrate that the Gdf11 tm2b allele is a null allele. Interestingly, we also observed the same skeletal patterning defects in MECP2-TG1; Gdf11 tm2b/+ pups, which had an abnormal number of 8 attached and 14 total ribs (n=4), while MECP2-TG1 pups had the same number of 7 attached and 13 total ribs (n=4) as wild-type pups (Figure 3-figure supplement 1C), further suggesting that the regulatory relationship between Gdf11 and Mecp2 is brain specific. Taken together, Gdf11 tm2b heterozygotes can be used to model the effects of losing one copy of Gdf11. Given these results, we generated a cohort of wild-type and Gdf11 tm2b/+ littermates and began general health assessments at weaning (Figure 3-figure supplement 2A). We observed that these mice did not display any overt phenotypes and did not have a change in body weight in either sex through sixteen weeks of age ( Figure 3-figure supplement 2B).
We began deeper behavioral characterization of the cohort at sixteen weeks of age. In the open field assay, we observed that Gdf11 tm2b/+ mice are hyperactive as indicated by a significant increase in total distance traveled (p<0.05), a significant increase in horizontal activity counts (p<0.05), and a significant increase in vertical exploratory activity counts (p<0.01) ( Figure 3A). Gdf11 tm2b/+ mice explored the center of the arena at a similar rate and ratio as wild-type mice (Figure 3-figure supplement 2C). In the elevated plus maze, Gdf11 tm2b/+ mice also displayed hyperactivity, traveling a significantly further distance (p<0.01) and decreased anxiety-like behavior, entering the open arm a significantly higher number of times (p<0.001; Figure 3B). Gdf11 tm2b/+ mice traveled a significantly further distance (p<0.05) in the light-dark assay while spending significantly more time (p<0.05) in the dark (Figure 3figure supplement 2D). Furthermore, when placed in a three-chamber apparatus, Gdf11 tm2b/+ mice explored more both during habituation and testing phases ( Figure 3C and Figure 3-figure supplement 2E). This manifests with a significantly (p<0.05) decreased time socializing compared to control ( Figure 3C). The ratio of time spent investigating the partner mouse over the total investigation time was significantly (p<0.0001) less in Gdf11 tm2b/+ mice (Figure 3-figure supplement 2F), though importantly the Gdf11 tm2b/+ mice are still social and prefer interacting with the partner mouse over the object (p<0.001; Figure 3C). Next, we observed a significant (p<0.0001) reduction in motor coordination and learning as assessed by the rotating rod assay ( Figure 3D). Finally, we observed a trend in reduction in learning during the training phase of a shock-tone conditioning paradigm ( Figure 3-figure supplement 2G), while observing a significant reduction in contextual learning (p<0.01) and cued learning (p<0.05; Figure 3E). These data collectively suggest that neurological behaviors are sensitive to loss of one copy of Gdf11 in mice.
As Gdf11 has been implicated to naturally decrease in aging in mice (Poggioli et al., 2016), we tested if a reduction in Gdf11 dosage can alter aging by tracking the survival of a cohort of Gdf11 tm2b/+ mice. Strikingly, we found a significant (p<0.001) reduction in survival of Gdf11 tm2b/+ mice compared to wild-type littermates (n=15 wild-type and n=17 Gdf11 tm2b/+ ), exhibiting a median survival of 1.8 years ( Figure 3F). Also, a portion Gdf11 tm2b/+ mice were euthanized due to reaching ethical endpoints including poor body conditioning, self-lesioning, and development of tumors, which suggests that Gdf11 dosage plays an important role in long-term health and survival in mice.
The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1. Raw data from all behavioral assays related to Figure     (D) Motor coordination measured by the rotating rod assay. (E) Learning assessment using contextual and cued fear-conditioning. Greater freezing indicates better memory of the context. (F) Survival analysis. Central estimate of data is shown as mean ± sem. Closed circles denote male mouse data points and open circles denote female mouse data points. For behavioral measurements except fear conditioning, n=29 wild-type mice (15 male, 14 female) and n=34 Gdf11 tm2b/+ mice (16 male, 18 female). For fear conditioning assay, n=28 wild-type mice (16 male, 12 female) and n=24 Gdf11 tm2b/+ mice (9 male, 15 female). For survival analysis, n=15 wild-type mice and Loss of one copy of Gdf11 does not alter proliferation, amplifying progenitors, or neural stem cells in the adult hippocampus Learning and memory is linked to the generation of adult-born neurons in the subgranular zone (SGZ) of the mouse dentate gyrus (Kempermann and Gage, 2002). Complete loss of Gdf11 in adulthood was shown to increase proliferation in the SGZ, specifically in the number of amplifying progenitors (SOX2 positive), while impairing the overall number of newborn neurons that survive and integrate into the existing hippocampal circuitry (Mayweather et al., 2021). Additionally, neurogenesis is negatively impacted in the olfactory epithelium with loss or inhibition of GDF11 (Wu et al., 2003). To test if loss of one copy of Gdf11 altered overall proliferation in the SGZ, potentially acting as a mechanism for the changes in learning and memory observed in Gdf11 tm2b/+ mice, we labeled newborn, dividing cells by injecting Gdf11 tm2b/+ and control mice with 5-Ethyl-2'-deoxyuridine (EdU). We injected the animals with EdU (10 mg/kg) for 5 consecutive days and fixed the brains at 7 days after the first injection. We performed these injections in sixteen-week-old animals, matching the timepoint when we performed behavioral characterization. We then stained for EdU, SOX2, and GFAP to label proliferating cells (EdU + ), neural stem cells (SOX2 + GFAP + ), and amplifying progenitors (SOX2 + ) (Cope and Gould, 2019;Zhao and Praag, 2020). We did not observe any significant changes in the number or density of these cell populations in Gdf11 tm2b/+ mice using quantitative stereology (Figure 4 and Figure 4-figure supplement 1A-D). Furthermore, no gross changes in brain anatomy or volume of the dentate gyrus were observed in Gdf11 tm2b/+ mice (Figure 4-figure supplement 1E-G). These results indicate that the behavioral deficits caused by loss of one copy of Gdf11 are not due to altered proliferation in the adult mouse SGZ.

Discussion
The heterogeneity in genetic causes for IDD and ASD has precluded the development of targeted therapies for these disorders. Distilling any common molecular processes that have gone awry in multiple disorders would give us a handle to investigate further and identify drug targets. In this study, we thoroughly investigated genes regulated by the exemplar dosage-sensitive gene MECP2, the causative gene of two different neurological disorders (RTT and MDS) to discover that MeCP2 sensitively and acutely regulates Gdf11. Intriguingly, Gdf11 is inversely dysregulated in RTT and MDS and restoring normal Gdf11 dosage ameliorated several behavioral deficits in a mouse model of MDS. This potential sensitivity of Gdf11 dosage in the brain led us to further discover that reduction of Gdf11 dosage alone caused neurobehavioral deficits in mice as well. Our study strongly suggests that GDF11 signaling is dysregulated in more than one neurodevelopmental disorder (RTT, MDS, and GDF11 haploinsufficiency) and modifies MeCP2-driven abnormal behaviors.
Transcriptomics is often used to identify molecular pathways that are dysregulated in diseases. However, uncovering the core genes that are altered can be challenging due to secondary effects and neuronal dysfunction, which has been particularly problematic in dissecting the molecular pathogenesis of MeCP2 (Sanfeliu et al., 2019). In this study, we leveraged dynamic gene expression changes upon acute decrease of MECP2 in a mouse model of MDS (Shao et al., 2021b), which specifically revealed that Gdf11 is highly correlated with MeCP2 protein levels. Generalization of these results demonstrated that Gdf11 is robustly dysregulated in mouse models of MeCP2-related disorders and regulated by MeCP2 at the chromatin level ( Figure 1). These data suggest that Gdf11 is proximally downstream of MeCP2, sensitive to both MeCP2 expression and function. Ongoing efforts are looking n=17 Gdf11 tm2b/+ mice. Behavioral data were analyzed using Welsch t-test and raw measurements are provided in Figure 3-source data 1. Survival was analyzed using Mantel-Cox test. * p<0.05, ** p<0.01, *** p<0.001, and **** p<0.0001.
The online version of this article includes the following source data and figure supplement(s) for figure 3: Source data 1. Raw data from all behavioral assays related to Figure 3 and    to restore MeCP2 dosage in mouse models of RTT and MDS (Shao et al., 2021b;Sinnamon et al., 2017;Sinnett et al., 2021). Typically, these studies rely on behavioral outcomes to assess treatment efficacy. However, behavioral improvements downstream of restored MeCP2 dosage or function may take weeks and may require a large sample size to manifest (Katz et al., 2012;Shao et al., 2021b;Sztainberg et al., 2015). We propose that Gdf11 expression in the brain can be used as an endogenous reporter for MeCP2 dosage and function in preclinical studies that can leverage the sensitivity of biomolecular measurements. Adding Gdf11 expression to the repertoire of metrics to evaluate treatment outcomes may prevent preliminary and promising candidates from being scrapped if behavioral outcomes are unclear or slow to normalize.
We showed that normalization of Gdf11 dosage improved several behavioral deficits in a mouse model of MDS when normalization of Gdf11 occurred using a constitutive germline deletion. If normalization of Gdf11 dosage could be used to treat MeCP2-related disorders, we must first test if restoring Gdf11 dosage in symptomatic RTT or MDS mouse models leads to outcome improvements. These experiments will decouple if MeCP2 regulation of Gdf11 is critical during a presymptomatic window Figure 4. Loss of one copy of Gdf11 does not alter adult neurogenesis in the dentate gyrus. Quantification of markers of (A) proliferative cells (EdU) or (B-D) neural progenitor pools (SOX2 and GFAP) in the subgranular zone of the dentate gyrus in wild-type or Gdf11 tm2b/+ mice. Representative images of the dentate gyrus for the indicate stains are shown on the left. The projected total number of cells and the density of cells is shown on the plots to the right using stereology to quantify n=3-6 separate slices per animal (N=3 wild-type animals and 5 Gdf11 tm2b/+ animals) (see Methods). Each data point is the aggregated value for one animal. Closed circles denote male animals and open circles denote female animals. Data are presented as mean ± sem. Data were analyzed using Welsch's t-test, and raw measurements are provided in Figure 4-source data 1. Scale bar is 200 μm and is the same for all images.
The online version of this article includes the following source data and figure supplement(s) for figure 4: Source data 1. Raw measurements and stereology quantification of neurogenesis markers. or throughout life. Importantly, GDF11 is a secreted factor, and its protein dosage could be modulated using recombinant protein to increase dosage (for RTT or GDF11 haploinsufficiency) or using a neutralizing antibody to decrease dosage (for MDS) (Morvan et al., 2017;Ozek et al., 2018;Zhang et al., 2018). Lastly, as GDF11 is present in the blood, there is a possibility that modulation of GDF11 dosage in the circulation impacts GDF11 dosage in the brain, affording an easily accessible route to modulating GDF11 levels (Poggioli et al., 2016;Xu et al., 2020).
Our study contributes to the growing body of work that GDF11 is a dosage sensitive gene important to neural function. Loss-of-function mutations in GDF11 have recently been described to cause neurological dysfunction as part of a broader developmental disorder. These patients exhibit abnormalities including intellectual disability, speech delay, and seizures (Cox et al., 2019;Ravenscroft et al., 2021). In this study, we show that reducing Gdf11 dosage by one copy in mice also causes abnormal phenotypes, which we can now leverage to investigate mechanisms of neuronal dysfunction in GDF11 haploinsufficiency. Furthermore, a recent genomic study predicted that GDF11 is intolerant to copy number gains in humans (Collins et al., 2022). Our work suggests that increased Gdf11 expression contributes to abnormal phenotypes in mouse models of MDS. Using mouse models that overexpress GDF11 alone could be used to determine if copy number gain in GDF11 is toxic (Jones et al., 2018). These studies will be important to determine whether GDF11 can be a therapeutic target if appropriately dosed.
Our study also highlights the challenges that arise when studying the interaction of multiple dosage-sensitive genes. The decrease or gain of a dosage-sensitive gene by even as little as ±30% can cause exactly reciprocal phenotypes (Shao et al., 2021a). In our case, Gdf11 tm2b/+ mice have phenotypes that present in the opposing direction as the MECP2-TG1 mice and in the same direction of Mecp2-null male mice (Katz et al., 2012), which have reduced Gdf11 expression; thus parsing out the direct cause-and-effect relationships from a potential pseudonormalization remains difficult even with deeper molecular or neurophysiologic characterization (Lavery et al., 2020;Zhou et al., 2022). As Gdf11 expression is rescued in ASO-treated MECP2-duplication mice before behavioral rescue (Shao et al., 2021b), we speculate the normalization of Gdf11 in the MECP2-TG1; Gdf11 tm2b/+ mice plays a direct role in the improvement of neurological dysfunction of MECP2-TG1 mice. Furthermore, not all behaviors or phenotypes were rescued, like the rotarod in which MECP2-TG1 and Gdf11 tm2b/+ mice have opposing phenotypes. Additionally, skeletal abnormalities observed in Gdf11 tm2b/+ pups were not rescued by increased Mecp2 dosage in MECP2-TG1; Gdf11 tm2b/+ pups, further suggesting that pseudonormalization does not explain the overall phenotypic profile of MECP2-TG1; Gdf11 tm2b/+ mice. Lastly, as both Mecp2 and Gdf11 have developmental roles (McPherron et al., 1999;Shahbazian et al., 2002), conditional genetics will be needed to precisely parse the mechanisms by which Gdf11 and Mecp2 dosage interact to regulate neurological function.
Though GDF11 is broadly and highly expressed in the mouse and human brains, we have a limited understanding of its function. As a secreted factor, it is unclear why this gene must be expressed throughout the brain in multiple cell types (Saunders et al., 2018). This may suggest that GDF11 has a strong autocrine and paracrine role in the brain rather than an endocrine role. We must therefore catalog which neurons or brain regions secrete or sense GDF11 at the protein level and understand the functional neuronal response to GDF11 stimulation. These studies will parse out the specific relationship of GDF11 dosage and neurological phenotypes. For example, we see robust hyperactivity in Gdf11 tm2b heterozygotes, but more variable effects on anxiety-like behaviors. How do the cells or circuits driving hyperactivity in Gdf11 tm2b heterozygotes interact with the cells and circuits responsible for learning and memory? Are these behaviors influenced because specific brain regions or cell types are producing or sensing GDF11? Furthermore, we discovered that Gdf11 dosage is important to survival in mice, corroborating a role of GDF11 in aging (Katsimpardi et al., 2014;Loffredo et al., 2013;Sinha et al., 2014), but it remains unknown whether this role is due to GDF11 expression in the brain. As such, understanding the production, transduction, and transmission mechanisms of GDF11 signaling in the brain will have long standing importance to multiple fields.
In summary, our study provides a framework for identifying putative disease-modifying genes by integrating multiple -omics level datasets. Through these analyses, we linked seemingly disparate disease-causing genes, MECP2 and GDF11 in this case, by characterizing a regulatory relationship between them. Probing other disease-causing genes in this way will allow us to stitch together a network of dosage-sensitive genes, which will reveal points of regulatory and molecular convergence across a wide range of diseases. Gdf11 tm2b mice were maintained on C57BL/6J background. MECP2-TG1 and Mecp2-knockout mice were previously described (Collins et al., 2004;Guy et al., 2001), maintained on a C57BL/6J background, and available from Jackson lab (Stock Nos: 008679; RRID: IMSR_JAX:008679 and 003890; RRID: IMSR_JAX:003890). Gdf11 and MECP2 double mutant mice were generated by breeding male Gdf11 tm2b/+ mice to female MECP2-TG1 or Mecp2 +/mice. Mice were monitored daily by veterinary staff. Survival age was taken as the age in which a mouse died of natural causes or when a mouse was euthanized due to poor body condition, self-lesioning, or tumor growth according to the institutional veterinarian guidelines. All procedures to maintain and use these mice were approved by the Institutional Animal Care and Use Committee for Baylor College of Medicine and Affiliates.

RNA-seq meta-analysis Analysis of transcripts correlated with MeCP2 protein levels
To identify genes that correlate with MeCP2 protein levels, we reanalyzed the transcriptional response of humanized MECP2 duplication mice treated with an anti-MECP2 ASO over time (GSE #1512222). The differential gene expression was perform as previously described (Shao et al., 2021b), and the log 2 fold-changes of the genes significantly altered between control and MECP2 duplication mice per time point relative to control were extracted. The log 2 fold-change in MeCP2 protein levels between control-treatment and MECP2-ASO treatment for a similar cohort of treated mice was taken from Figure 4C of that paper. The Spearman correlation was calculated between protein level fold-change and gene expression level fold-change. The probability of loss intolerance (pLI) was extracted from the gnomad data base (Karczewski et al., 2020).

Extraction of gene expression changes from MeCP2 related transcriptional profiles
To extract the expression of RNA fold-changes for candidate, correlated genes across MeCP2perturbed, we extracted log 2 fold-change of each transcript (MeCP2 mutant allele / wild-type) and an associated p adjusted . A gene was considered significant if the false-discovery rate was below 0.1. Bulk RNA-sequencing data from MeCP2-perturbed models were preferred but in three unique cases: (1) Single-nuclei data from postmortem RTT tissue (Renthal et al., 2018), microarrays from two MECP2 truncation models that do not have a corresponding RNA-sequencing data set (Baker et al., 2013), and bulk nuclei RNA-seq from MECP2 AAV expressing neurons that do not have a corresponding bulk RNA-sequencing data set (Ito-Ishida et al., 2020). Data were extracted from Supplemental Tables or processed files deposited on GEO as possible. The data table from the MECP2 AAV expressing neurons was taken from the following link. An additional set of data tables were found in the source data from a previous meta-analysis . The remaining data tables were generated by obtaining raw counts from GEO and re-analyzing using DESeq2 between MeCP2 perturbed samples and wild-type samples. A summary of the studies and the sources of data tables are shown in Supplementary file 1.

Cell-type-specific correlation of Gdf11 and Mecp2 expression
To extract the cell-type-specific expression of Gdf11 and Mecp2, we obtained single-cell RNAsequencing counts and cell-type annotation files from the DropViz database (https://www.dropviz. org; Saunders et al., 2018). Using the annotation file, we divided the cells into the demarcated categories and performed Spearman correlation between Gdf11 and Mecp2. To calculate neuron subtype specific correlation of Gdf11 and Mecp2, we used Slc17a7 (Vglut1) expression to coarsely separate excitatory and inhibitory neurons. We used a threshold of ≥5 counts per cell type to denote an excitatory neuron. The Spearman correlation was calculated for both excitatory and inhibitory neurons.
To set up the gradient, 4 mL of 40% Iodixanol (3 volumes working solution +1 volumes HB buffer) was added to a 50 mL round bottomed conical tube. Then, 7.5 mL 30% Iodixanol (3 volumes working solution +2 volumes HB) was slowly overlayed on top, followed by 10 mL of the sample containing mixture prepared above. This gradient was spun at 10,000 g for 20 min at 4 °C in a hanging bucket centrifuge (Sorvall Lynx 6000) with 'decel' turned off. After centrifugation, nuclei are located at the interface between 30% and 40% iodixanol layers. Iodixanol containing supernatant above the nuclei was slowly discarded with bulb pipette. Approximately 1.5-2 mL of the interface containing nuclei were collected and placed into 2 mL microcentrifuge tube.
The number of nuclei were quantified by taking 20 µ of sample and mixing it with 2 µL 0.2 mg/mL DAPI diluted in HB buffer. After three-minute incubation at RT, the nuclei were diluted 1:10 in buffer HB and counted on a Countess II with DAPI channel to quantify.

CUT&RUN
Cleavage Under Targets & Release Using Nuclease (CUT&RUN) was performed on nuclei isolated from above following Coffin et al., 2022;Skene and Henikoff, 2017. Briefly, we performed one nuclear isolation per animal and then split the nuclei into three individual tubes for the three antibodies surveyed (MeCP2, H3K27me3 and IgG). An individual sample will refer to one antibody from a unique animal.
To activate Concanavalin A coated magnetic beads (Bangs Laboratories Inc, #BP531) for binding, we incubated 25 µL per sample with 3 x volumes binding buffer (20 mM HEPES-NaOH pH 7.5, 10 mM KCl, 1 mM CaCl 2 , 1 mM MnCl 2 ), rotated at RT for 5 min, and washed 2 x with 1 mL binding buffer. All washes are done by placing microcentrifuge tube on magnetic rack and waiting until solution is clear as beads separated from the solution. Following washes, the beads were resuspended in 50 µL binding buffer per sample.
After bead activation, 200 µL beads were added to 1.5x10 6 nuclei in the iodixanol mixture from each animal and rotated for 10 min at room temperature. Following binding of nuclei to beads, all further processing was done on ice. Bead bound nuclei from each animal was washed 2 x with 1.5 mL wash buffer (20 mM HEPES NaOH pH 7.5, 150 mM NaCl, 500 μM Spermidine [Sigma #S0266], and 0.5% Ultrapure BSA [Invitrogen #AM2618] with 1 tablet of Complete Protease Inhibitor Cocktail [Roche #11873580001] per 50 mL). After the second wash, 250 µL of nuclei bound beads (~250,000 nuclei) in wash buffer were added to individual microcentrifuge tubes corresponding to each antibody.
To initiate cleavage and release of DNA bound fragments, each sample was resuspended in 150 µL of dig-wash buffer while gently vortexing. Samples were placed on ice in 4 °C room for 10 min to equilibrate. To start digestion, while in 4 °C room, 3 µL of 100 µM CaCl 2 was added to each tube, quickly flicked, and immediately returned to ice. Following 45-min incubation on ice, 150 µL 2 x STOP 340 mM NaCl, 20 mM EDTA, 4 mM EGTA, 0.05% Digitonin, 100 µg/mL RNAse A (ThermoFisher Scientific #EN0531), 50 µg/mL Glycogen (ThermoFisher Scientific #10814010) and 1 ng E. coli spike-in / sample (Epicypher #18-1401) mixture was added to each sample. Tubes were then incubated at 37 °C for 30 min to digest RNA and release DNA. E. coli spike-in control was used for normalization of CUT&RUN signal due to differences in library amplification and/or sequencing. Supernatant was transferred to new tube and incubated with 1.5 µL 20% SDS and 5 µL 10 mg/mL Proteinase K while lightly shaking at 50 °C for 1 hr. DNA was then purified by phenol-chloroform extraction using Maxtract Tubes (129046, Qiagen) and pellet resuspended in 36.5 µL TE Buffer.

CUT&RUN next-generation sequencing library preparation
Library preparation was modified from protocols.IO (dx.doi.org/10.17504/protocols.io.bagaibse) utilizing reagents from the NEB Next II DNA Ultra Kit (New England Biolabs #E7645S) and Unique Combinatorial Dual index kit (New England Biolabs #E6442S,) with modifications outlined below. Input DNA was quantified with Qubit, and 6 ng of CUT&RUN DNA was used as input for the H3K27me3 samples and 25 µL of CUT&RUN DNA was used for both MeCP2 and IgG samples. Volume of DNA was brought up to 25 μL and 1.5 µL End Prep Enzyme Mixture and 3.5 µL Reaction buffer were added and incubated at 20 °C for 30 min and 50 °C for 60 min. After end prep, 15 µL of NEB Next Ultra Ligation Mastermix, 0.5 µL Ligation Enhancer and 1.25 µL of Adapter (1.2 pmol adapter for H3K27me3, 0.6 pmol adapter for MeCP2 and IgG) were added directly to the PCR tube, mixed by pipetting, and incubated for 15 min at 20 °C. Then, 1.5 µL of USER Enzyme is added to each tube. Finally, SPRI select beads (Beckman Coulter # B23318) were used at 1.6 x ratio to remove excess adapter and eluted in 15 µL of TE buffer.
PCR amplification was performed using 13 µL of adaptor ligated fragments, 1 µL of Unique Combinatorial Dual Index (one index per sample), 1 µL of sterile water, and 15 µL 2 x Q5 Master Mix. Fourteen cycles of PCR were performed with 10 s of denaturation at 98 °C and 10 s of annealing/extension at 65 °C. Following PCR amplification, SPRI select beads were used for two-sided size selection; 0.65 x right sided selection was performed first followed by 1.2 x left sided size selection. Sample was eluted in 15 µL TE.
For quality control, each library size distribution was determined by Agilent Tapestation HS DNA 1000 (Agilent Technologies #5067) and concentration was determined by KAPA PCR (Roche #07960140001). Libraries were pooled together at equimolar concentrations and submitted to the Baylor Genomic and RNA Profiling Core. Each library was sequenced for approximately 40 million paired end reads of 100 bp in length on a Novaseq S1 flow cells.

MeCP2 Peak Calling
Peaks were called using the MACSr package (version 1.0.0) using the following parameters: gsize = "mm", format = "BAMPE", broad = "TRUE", keepduplicates = "all", qvalue = 0.0001 (Zhang et al., 2008). Input to the peak calling function callpeak were the six replicates of wild-type MeCP2 BAM files and three replicate Mecp2-knockout MeCP2 BAM files as control. Peaks with a q-value less than 0.0001 were retained for further analysis. Peaks were associated with the closest gene using bedtools -closest and mm10 GRCm38 gene coordinates.

Visualization and Quality Control of CUT&RUN signal
Integrative Genomics Viewer (IGV) v2.11.1 was used to examine spike-in normalized bigwig tracks at individual loci (Robinson et al., 2011). We extracted the integrated density at MeCP2 peaks ±3.5 kb of each filtered peak by using the bedtools function multicov for each BAM file generated (MeCP2, H3K27me3, and IgG). We performed differential binding using the DESeq2 package (version 1.32) using the E-coli spike in control counts as normalization factor and cooksCutoff set to FALSE. DESeq2 was used to perform principal component analysis of each mark individually and then together. Quantification of MeCP2 binding was performed across all samples, pooling two batches of sample collection due to no apparent batch effects by PCA. Quantification of H3K27me3 binding was performed in two batches, with MECP2-TG1 and Mecp2-knockout samples assessed with their respective wild-type samples within the batch. ChIP-seq measurements of H3K27ac, H3K4me1, and H3K4me3 generated from CA1 hippocampal neurons were obtained from GSE74971 as bigwig files (Halder et al., 2016). The naïve, 0 hr measurement from CA1 neurons for each histone mark was evaluated. Replicate bigwig files were merged using deeptools bigwigCompare and averaged for summary figures.

RNA extraction, reverse transcription, and qPCR
Total RNA was isolated from one cerebellar hemisphere using the Qiagen miRNeasy Mini kit (Qiagen #217004). On column DNAse digestion (Qiagen #79254) was performed according to manufacturer's protocol to remove genomic DNA. Two μg of total RNA was used to synthesized cDNA using the M-MLV reverse transcriptase (Invitrogen #28025013) according to manufacturer's protocol. qRT-PCR was performed using a CFX96 Real-Time System (Bio-Rad) using PowerUp SYBR Green Master Mix (ThermoFisher #A25741), 0.4 μM forward and reverse primers, and 1:20 dilution of cDNA. The following cycling conditions were used: 95 °C for 5 min, 39 cycles of 95 °C for 11 s, 60 °C for 45 s, plate read, a final melt of 95 °C, and melt curve of 65-95°C at +0.5 °C increments. The specificity of the amplification products was verified using melt-curve analysis. The Ct values were calculated with the Bio-Rad software, and relative gene expression was calculated using the ΔΔCt method using Ppia for normalization. All reactions were performed in technical duplicate with a minimum of three biological replicates. Data are presented as mean ± SEM in main figure.
qPCR primer sequences

Elevated plus maze
The lighting in the test room was set to 750 lux, and the background noise level was set to 62 dB with a white noise generator. After habituation, the mice were placed in the center zone of a plus maze that has two open arms. The mice were placed facing one of the open arms and allowed to freely move for 10 min. The movement of the mice was tracked with the ANY-maze software system (Stoelting Inc).
The total distance and distance traveled per zone, time spent in each arm zone, distance traveled in each arm zone, and number of zone crossings were recorded and tabulated by the software.

Open-field assay
The lighting in the test room was set to 150 lux, and the background noise level was set to 62 dB with a white noise generator. After habituation, mice were placed in an open plexiglass area (40x40 x 30 cm), and their movement and behavior were tracked by laser photobeam breaks. Mice were allowed to freely move for 30 min. The total distance traveled, horizontal or vertical laser beam breaks (activity counts), entries into the center 10x10 cm, and time in the center 10x10 cm were recorded and tabulated by the AccuScan Fusion software (Omnitech Electronics Inc).

Light-dark box assay
The lighting in the test room was set to 150 lux, and the background noise level was set to 62 dB with a white noise generator. After habituation, mice were placed in an plexiglass arena with an open zone (36x20 x 26 cm) and a closed, dark zone made of black plastic (15.5 x. 20x26 cm) with a 10.5 x. 5 cm opening. Mice were allowed to freely move for 10 min. The movement of the mice was tracked by laser photobeam breaks. The total and zone-specific distance traveled, the total or zone-specific horizontal or vertical laser beam breaks (activity counts), entries into the dark zone, and time spent in either zone were recorded and tabulated by the AccuScan Fusion software (Omnitech Electronics Inc).

Three-chamber assay
The lighting in the test room was set to 150 lux, and the background noise level was set to 62 dB with a white noise generator. Age and sex matched C57BL/6J mice were used as novel partner mice, which were housed together and had no overtly aggressive behaviors observed. Two days before the test, the novel partner mice were habituated to the wire cups (3-inch diameter by 4-inch height) for 1 hr/ day for 2 days. After habituation, the test mice were placed in a clear plexiglass chamber (24.75x16.75 x 8.75 inch) with two removable partitioned that divide the chamber into three zones. Empty wire cups were placed in the left and right chambers. For the acclimation phase, the mice were placed into the center zone and the partitions removed. The mice were allowed to freely explore the apparatus for 10 min. For the socialization phase, a novel partner mouse was randomly placed into the left or right wire cup. An inanimate object as control was plaed in the wire cup of the opposing chamber. The mice were allowed to explore the apparatus again for 10 minutes. The total distance traveled was tracked by the ANY-maze software system (Stoelting Inc). For all phases, the total amount of time the mice spent rearing, sniffing, or pawing the cup was recorded manually.

Rotating rod
The lighting in the test room was at ambient levels, and the background noise level was set to 62 dB with a white noise generator. After habituation, mice were placed on an accelerating rotarod apparatus (Ugo Basile). The rod accelerated from 4 to 40 r.p.m. for 5 min. For cohorts containing the MECP2-TG1 allele (Figure 2 and Figure 2-figure supplement 1), the maximum trial length was 10 min in concordance with previous studies performing the rotarod assay in MECP2-TG1 mice and the observation that MECP2-TG1 mice perform above wild-type mice on this assay (Collins et al., 2004); for all other testing, the maximum trial length was 5 min as a standard starting assay length for an uncharacterized mouse line. Mice were tested four trials each day, with an interval of at least 30 min between trials, for four consecutive days. The time it took for each mouse to fall from the rod (i.e. latency to fall) was recorded.

Fear conditioning
Animals were habituated in an adjacent room to the testing room with a light level of 150 lux and noise level of 62 dB using a white noise generator. On the training day, mice were brought to the testing groom (ambient light, no noise) and placed into a chamber containing a grid floor that can deliver an electric shock with one mouse per chamber (Med Associates Inc). The chamber was housed within a sound-attenuating box with a digital camera, loudspeaker, and a light. The training paradigm consisted of 2 min of no noise or shock, then a tone for 30 s (5 kHz, 85 dB), ending with a foot shock for 2 s (1 mA). The tone and shock pairing was repeated after 2 min of no noise or shock. The mice were returned to their home cages after training. On the testing day, occurring 24 hr later, two tests are performed -context and cued learning. For the context test, the mice were placed in the exact same chamber with no noise or shock for 5 min. After 1-hr post-context test, the cued learning test was performed. For the cued learning test, the mice were placed in a novel environment for 6 min. The first 3 min with no noise or shock, and the last 3 min with tone only (5 kHz, 85 dB). Movement was recorded on video and freezing, as defined by absence of all movement except respiration, was quantified using the FreezeFrame software (ActiMetrics) using a bout duration of 1 s and movement threshold of 10.

Brain tissue fixation and sectioning
We used design-based stereology protocol for fixing, sectioning, and quantifying proliferation markers in the subgranular zone of the dentate gyrus (Zhao and Praag, 2020). Adult, 16-week-old mice (n=5 for Gdf11 tm2b/+ and n=3 for wild-type littermate controls) were given a subcutaneous injection of analgesia (Buprenorphine, 0.5 mg/kg) 30 min prior to anesthesia (Rodent Combo III, 1.5 mL/kg) via intraperitoneal injection. Mice were perfused with ice-cold PBS followed by 4% PFA. Brains were removed and placed in 4% PFA overnight. Brain samples were then washed with ice-cold PBS, incubated in 15% sucrose in PBS overnight, and stored in 30% sucrose in PBS with 0.01% sodium azide. Experimenter was blind to genotype throughout the sectioning, imaging, and quantification process. Serial sections of the left hemisphere along the sagittal plane at 40 μm thickness were cut on a Leica SM2010 R sliding microtome and free-floating serial sections were stored in PBS +0.01% sodium azide at 4 °C.

EdU and immunofluorescent staining
Mice were injected with 10 mg/kg of 5-ethynyl-2'-deoxyuridine (EdU; Invitrogen #E10187) by intraperitoneal injection for 5 consecutive days. Brains were perfused and sectioned as described above 2 days after the final injection. The Click-iT EdU Imaging Kit with AlexaFluor 647 dye (Invitrogen #C10340) was used to detect EdU in sections following manufactures protocol. Free-floating sections were stained with mouse anti-GFAP (Sigma #G3893, RRID:AB_477010, 1:1000 dilution) and rabbit anti-SOX2 (Abcam #ab97959, RRID: AB_2341193, 1:500 dilution) before or after EdU Click-it reaction. Sections were washed with PBS (3 X) and incubated in permeabilization buffer (0.3% Triton-X-100 in PBS) for 5 min at room temperature. Sections were placed in blocking buffer (0.3% Triton-X, 3% normal donkey serum) for 30 min at room temperature and then incubated with primary antibodies with 1% BSA in blocking buffer overnight, rocking at 4 °C. Sections were washed with PBS (3 X) then incubated with secondary antibodies at 1:1000 dilutions (donkey anti-rabbit Alexa 488; Jackson ImmunoResearch #711-545-152, RRID:AB_2313584 and donkey anti-mouse Alexa 555; Thermo Scientific #A-31570, RRID: AB_2536180) in blocking buffer with 1% BSA for 1 hr at room temperature. After final washes with PBS, sections were mounted onto slides and coverslips were placed using Prolong Diamond with DAPI mounting media (Invitrogen #P36966). Slides were allowed to dry overnight.

Imaging and stereology quantification
Stained sections were imaged on a Leica SP8X confocal microscope. Z-stack images of the entire dentate gyrus were imaged to quantify absolute numbers of EdU + , GFAP + , and SOX2 + cells per section. Images were analyzed in LAS X software to measure the area and thickness of each section (t). Image J software was used to count EdU + , SOX2 + , and GFAP + cells (Q) following a standardized quantification method (Zhao and Praag, 2020). Briefly, the section sampling interval (ssf) was 1/6 for EdU and GFAP and 1/10 for SOX2 quantification. The area sampling fraction (asf)=1, and h = 40. The following equation was used to calculate the total cell number per animal (Zhao and Praag, 2020):

Skeletal analysis
Newborn P0 pups were euthanized, and skeletal preparations were performed exactly as previously described (Rigueur and Lyons, 2014). Briefly, skin and soft tissues were removed, pups were fixed in 95% ethanol at room temperature overnight, lipids cleared with acetone at room temperature overnight, stained with 0.03% Alcian blue (Sigma #A5268) in 20% glacial acetic acid and 80% ethanol at room temperature overnight, washed 2 x with 70% ethanol and stored in 95% ethanol overnight, pre-cleared with 1% KOH for one hour at room temperature, stained with 0.005% Alizarin red (Sigma #A5533) in 1% KOH for four hours at room temperature, washed in 50% glycerol:50%-1% KOH for 3-4 days at 4 °C, and stored in 100% glycerol until imaged on stereomicroscope. Excess tissues were trimmed before counting attached and total rib segments.

Statistical analysis
Experimental analyses were performed in a blinded manner when possible. Statistical tests were performed in accordance with the experimental design. Data normality for behavioral data was assessed using the Kolmogorov-Smirnov test and the Shapiro-Wilk test for the immunofluorescence quantification. Outliers were not assessed nor removed. Single comparisons used Student's t-test, whereas multi-group comparisons used one-or two-way ANOVAs as appropriate. The specific test used in each experiment is indicated in the figure legend. In each case, *, **, ***, ****, and n.s. denote p<0.05, p<0.01, p<0.001, p<0.0001, and p>0.05, respectively.

Additional files
Supplementary files • Supplementary file 1. Summary and source of transcriptomics studies mined in this study.

• MDAR checklist
Data availability CUT&RUN data have been deposited in GEO under accession code GSE213752. This paper does not report original code. Raw data from behavioral analyses are provided in Figure 2-source data 1 and Figure 3-source data 1, and raw quantification from immunofluorescence stereology quantification are provided in Figure 4-source data 1. Any additional information required to reanalyze the data reported in this paper is available from the corresponding author upon request.