Introduction

Uncontrollable cell proliferation can lead to tumor growth, while premature differentiation can result in tissue degeneration. The balance between stem cell proliferation and differentiation is a crucial aspect in developmental and stem cell biology. BMP (Bone morphogenetic proteins) signaling is a key cell-signaling pathway in regulating stem cell proliferation and maintaining adult stem cell quiescence ^1–3. Moreover, BMP signaling is essential in various cell specification processes ^4–6.

The ability of a single pathway to play a diverse range of roles relies on context-specific transcriptional outputs. BMPs signal through two types of SMAD proteins: the receptor SMADs (R-SMADs) - Smad1, 5 and 8, and the co-SMAD protein SMAD4. Upon ligand binding, heterodimeric receptors like BMPRI and BMPRII phosphorylate R-SMADs, leading to the assembly and nuclear translocation of the SMAD complex with SMAD4. This complex then regulates gene expression by binding to enhancers of BMP target genes. Apart from transducing BMP signaling, SMAD4 is an essential effector in TGF-β/Activin signaling. In response to ligands such as TGF-β and Activin, SMAD4 associates with two other R-SMADs - phosphorylated SMAD 2 and 3, regulating downstream genes of the TGF-β/Activin signaling pathway. These SMADs recognize and directly bind to two main types of short DNA motifs in target enhancers via their N-terminal MH1 domain ⁷. However, since the binding is generally weak, SMAD complexes rely on various co-factors such as transcription factors, co-activators and co-repressors to activate or repress target gene expression ⁸.

In the mammalian brain, BMP and Wnt signaling pattern the brain midline where an essential structure, the ChP, emerges. Neural epithelial cells at the presumptive ChP site lose neural potential and exit cell cycle. Only a small number of cells at the boarder of the ChP primordium and cortical hem (CH) persist as slowly dividing ChP progenitors, leading to the expansion of the monolayered ChP epithelial cells ⁹. BMP signaling is essential for ChP epithelium specification. Conditional depletion of the BMP receptor BMPr1a diminishes ChP development ¹⁰, whereas ectopic BMP transforms neural progenitors into ChP cells ¹¹. Wnt signaling, peaking at CH, is also necessary for ChP epithelial cell specification ¹². Conditional depletion of ß-Catenin results in defective ChP, and overexpression of ß-Catenin expands CH at the expense of the ChP epithelium. These findings emphasize the importance of tightly controlling Wnt and BMP signaling levels for proper cell specification.

The PR domain-containing (PRDM) family protein PRDM16 determines cell fate specification in various cell types ^13–19. Previous studies, including our own, demonstrated that Prdm16 knockout (KO) mouse brains show severely reduced ChP structures ^20–22. Interestingly, it was shown that PRDM16 can interact with TGF-ß pathway SMAD proteins in vitro and impact TGF-ß signaling output in craniofacial tissues ^21,23. PRDM16 is also a downstream effector of BMP signaling during brown adipocyte specification ^24,25. A recent study reported that PRDM16 and its ortholog PRDM3 (also known as Evi1) regulate Wnt signaling during craniofacial development in zebrafish ²⁶. These findings suggest that PRDM16 may be more broadly involved in regulating BMP/TGF-ß and Wnt signaling. However, the underlying molecular mechanisms remain unclear.

Consistent with its essential roles in normal development, mutations and dysregulation of Prdm16 are linked with several human diseases, including those identified in the patients with 1p36 chromosomal aberrations and cardiomyopathy ²⁷. PRDM16 exhibits versatile functions at the molecular level, regulating chromatin accessibility and epigenetic states of its bound enhancers ^18,20. Depending on associated cofactors, PRDM16 can either repress or activate gene expression. This dual role poses a challenge when considering PRDM16 as a therapeutic target, as undesired outcome may occur. Thus, a comprehensive understanding of the regulatory roles of this protein in each specific process is crucial for effective disease treatment strategies.

In this study we have investigated the mechanisms that regulate the transition between stem cell proliferation and differentiation during ChP development. We show that Prdm16 mutant ChP cells fail to exit cell cycle, a similar phenotype to when BMP signaling is abolished ¹⁰. Using primary NSC culture, we dissected the molecular interaction of SMAD4/pSMAD1/5/8 proteins with PRDM16, and found that PRDM16 functions as a co-repressor that holds the SMAD complex at distal enhancers, repressing genes involved in cell proliferation. Finally, we validated that some of the co-regulated genes by BMP signaling and PRDM16 become derepressed in the Prdm16 mutant ChP. These findings uncover an essential function of PRDM16 in stem cell regulation and BMP and Wnt signalling.

Results

PRDM16 promotes cell cycle exiting of neural epithelial cells at the ChP primordium

To understand how Prdm16 regulates ChP epithelial specification, we first investigated the expression of this gene in the developing mouse brain. At embryonic day 10.5 (E10.5), BMP signaling specifies the presumptive ChP ¹⁰. Prdm16 mRNAs and nuclear localization of the PRDM16 protein become detectable in the presumptive ChP at this stage (Fig. 1A, supplementary Fig. 1A). As previously shown, Prdm16^cGT, a Prdm16 knockout allele (Prdm16 KO), displayed severely reduced ChP structure at E13.5 (Fig 1B) ^20–22. Expression of Ttr, a classic ChP marker gene, is reduced in both the lateral ventricle (tChPs) (Fig. 1C) and the fourth ventricle (hChP) (Supplementary Fig. 1B) in the Prdm16 KO mutant, suggesting that the function of PRDM16 is not area-restricted but generally required for ChP development.

Figure 1.

To assess the patterning of the CH and ChP regions, we analyzed the expression of Wnt2b and BMP4 using conventional RNA in situ hybridization. Wnt2b expression, which marks the CH, appeared comparable between Prdm16 KO and control brains at E11.5 and E12.5 (Supplementary Fig. 1C), indicating that CH development is not affected by the loss of Prdm16. BMP4, which labels the ChP and CH, also showed normal expression levels and spatial distribution in the mutant brain. These findings suggest that the BMP-dependent patterning of the ChP and CH domain remains intact in the absence of Prdm16.

However, the ChP epithelial layer in Prdm16 mutants remained abnormally thick, resembling the adjacent neural epithelium (Fig. 1C, Supplementary Fig. 1C). This observation led us to investigate the underlying cellular changes responsible for the ChP defects in the mutant. To assess cell proliferation, we performed a 2-hour EdU labeling at E12.5. In control embryos, ChP cells were largely EdU negative and formed a monolayer, indicating that most had exited the cell cycle. In contrast, Prdm16 mutant ChP cells marked by β-Gal remained highly proliferative (Fig. 1D). These results suggest several possibilities: ChP epithelial cells is not properly specified in the mutant, mutant ChP cells are specified but fail to exit the cell cycle to differentiate, or both processes are impaired.

To explore these possibilities, we first examined the cell type composition of Prdm16 mutant ChP cells at E12.5. We stained brain sections from control and mutant embryos with the neuronal marker Doublecortin (Dcx). In controls, Dcx-positive cells were absent from the ChP, indicating that neural epithelial cells had successfully transitioned into ChP epithelial cells and lost their neural potential (Fig. 1E). In contrast, the Prdm16 mutant ChP exhibited numerous Dcx-positive cells along the basal side of the epithelium, suggesting ongoing neurogenesis.

We next examined the expression of the NSC marker SOX2. In control brains, SOX2 is highly expressed in NSCs adjacent to the ChP epithelium but is significantly downregulated within the presumptive ChP region (Supplementary Fig. 1D). This downregulation was absent in the Prdm16 mutant. Together, these results indicate that at E12.5, Prdm16 depletion disrupts the normal transition of neural progenitors into ChP epithelial cells, maintaining cells in a proliferative, undifferentiated state.

We then asked whether PRDM16 not only regulates ChP epithelial specification but also directly restricts cell proliferation. To test this, we examined Prdm16 mutant ChP cells at a later stage. We performed double S-phase labeling by injecting EdU at E12.5 and BrdU at E13.5, followed by sample collection at E14.5. By this stage, the wild-type ChP had developed into an elongated, monolayered epithelium in which most cells were negative for both EdU and BrdU, indicating cell cycle exit. In contrast, the Prdm16 mutant ChP remained a disorganized cluster of cells, many of which were EdU-positive, BrdU-positive, or both (Fig. 1F-G). This finding suggests that although the mutant ChP can separate from the neuroepithelium at later stages, it fails to exit the cell cycle and differentiate, supporting a direct role of PRDM16 in restricting proliferation.

Given that BMP signaling is a key regulator of ChP formation and that the BMP pathway mutants such as Foxg1-cre::loxpBmpr1a also exhibit ectopic cell proliferation in the developing ChP ¹⁰, we next investigated whether PRDM16 interacts with the BMP pathway to control cell proliferation.

PRDM16 and BMP signaling collaborate to induce NSC quiescence in vitro

To investigate the molecular interplay between PRDM16 and BMP signaling in regulating cell proliferation, we turned to primary NSCs as an in vitro model. BMP signaling is known to maintain quiescence in adult NSC in vivo and can induce proliferative NSCs into quiescence in vitro ¹. Furthermore, previous studies have shown that embryonic cortical NSCs are responsive to BMP4 treatment ²⁸, making them a suitable system for probing downstream BMP signaling events. Based on this, we established a cell culture assay to evaluate the effects of BMP4 and PRDM16 on NSC proliferation and quiescence (Fig. 2A).

Figure 2.

Unexpectedly, we found that PRDM16 protein levels were undetectable in cultured NSCs despite high levels of Prdm16 mRNAs (Supplementary Fig. 2A-B). Given that the PRDM16 protein is normally restricted to the nucleus of NSCs in embryonic brain tissues (Fig. 1B, Supplementary Fig. 1A and ²⁰), this observation suggests that in vivo mechanisms may regulate PRDM16 protein stability and nuclear localization. Post-translational modification of PRDM16 has been identified as a key regulatory mechanism in brown adipocytes ²⁹, and we speculate a similar mechanism may operate in NSCs.

To bypass this limitation and examine PRDM16’s molecular function in regulating NSC proliferation and gene regulation, we generated a lenti-viral construct expressing 3xNSL_Flag_Prdm16 under a constitutive promoter. Infection of wild-type primary NSCs with this construct yielded a cell line with robust Prdm16 mRNA expression and detectable nuclear PRDM16 protein (Supplementary Fig. 2A-B), which we referred to as Prdm16_overexpressing (Prdm16_OE). For comparison, we attempted to establish three additional lines: wild-type NSCs infected with the empty vector (wt_CDH), Prdm16_KO NSCs infected with the empty vector (KO_CDH), and Prdm16_KO NSCs infected with 3xNSL_Flag_Prdm16. However, we were unable to generate the last line despite repeated attempts. This failure was likely due to low viral production (the Prdm16 coding sequence exceeds 3kb) and the increased sensitivity of KO NSCs to viral infection. Nevertheless, we proceeded with comparative analyses of Prdm16_OE cells against wt_CDH, KO_CDH and uninfected KO NSCs.

To assess cell proliferation rates, we labeled NSCs with EdU and mKi67. BMP4 treatment of Prdm16_OE NSCs led to a marked reduction in the number of mKi67- and EdU-positive populations (Fig. 2B-D). Following BMP4 washout, Prdm16_OE NSCs restored EdU and mKi67 labeling, indicating re-entry into the cell cycle. This reversible reduction confirms that BMP4 induces a quiescent, non-proliferative state in NSCs. In contrast, a larger proportion of Prdm16_KO cells failed to exit the cell cycle in response to BMP4, as shown by a less pronounced decrease in mKi67- and EdU-positive populations (Fig. 2B-D). Notably, Prdm16_KO cells exhibited similar properties regardless of whether they were infected with the control viral vector (Supplementary Fig. 2D-F), and we therefore used both lines interchangeably in our analyses.

Previous studies have shown that BMP4 can program ChP cell fate and activate ChP-specific genes in neural progenitors cultured in vitro ¹¹. To test whether PRDM16 cooperates with BMP signaling to induce ChP gene expression, we measured Ttr mRNA levels using reverse-transcription followed by quantitative PCR (RT-qPCR). In Prdm16_OE NSCs, Ttr expression was robustly induced following BMP4 treatment. In contrast, Prdm16_KO NSCs and NSCs lacking the Prdm16_OE construct failed to upregulate Ttr in response to BMP4 (Fig. 2E, Supplementary Fig. 2G).

Together, these results indicate that both BMP signaling and PRDM16 are required not only to restrict NSC proliferation but also to induce ChP gene expression. We next investigated the molecular mechanisms underlying this cooperation.

BMP signaling and PRDM16 cooperatively repress proliferation genes

To understand how BMP signaling and PRDM16 suppress cell proliferation, we aimed to determine the transcriptional targets of SMAD4 and pSMAD1/5/8 in cells with and without BMP4. We first applied Cleavage Under Targets and Tagmentation (CUT&TAG) experiments using a PRDM16 antibody and the available SMAD antibodies to profile their genomic binding sites. The PRDM16 antibody worked efficiently, but none of the SMAD antibodies produced libraries with sufficient sequencing reads. Subsequently we employed chromatin immunoprecipitation followed by deep-sequencing (ChIP-seq) for the SMAD proteins. Given that SMAD4 forms a complex with SMAD2/3 only in response to TGF-β/activin-type ligands, we included an antibody targeting SMAD3 as a control for non-BMP4-induced SMAD4 binding. Thus, our experiment settings enabled us to profile genomic binding sites for PRDM16, SMAD4 and two types of R_SMADs under conditions with endogenous BMPs and TGF-β/Activin, as well as ectopic BMP4.

In Prdm16_OE cells without BMP4, we identified several hundred to a few thousand ChIP-seq peaks for all three classes of SMAD proteins after normalizing to the input reads (FDR < 10%), indicating endogenous levels of BMP and TGF-β signaling in these cells (Fig. 3A, Supplementary Fig. 3A-C). Following BMP4 addition, pSMAD1/5/8 exhibited approximately a 750-fold increase in peak number, while the number of SMAD3 peaks increased by less than threefold (Fig. 3A, Supplementary Fig. 3A). Furthermore, in cells treated with BMP4, the pSMAD1/5/8 peaks largely overlap with the SMAD4 peaks, but much less with the SMAD3 peaks (Supplementary Fig. 3B). For example, pSMAD1/5/8 binds to two enhancers in the intronic region of Prdm16, with the binding intensity dramatically increased in response to BMP4 (Supplementary Fig. 3D). This suggests that Prdm16 itself is a transcriptional target of BMP signaling. Indeed, Prdm16 mRNA levels were elevated by BMP4 in wild-type but not Prdm16_KO cells (Supplementary Fig. 2B). By contrast, SMAD3 showed little ChIP-seq signal in the Prdm16 gene locus even in the presence of BMP4. These results confirm that the response of pSMAD1/5/8 is specific to BMP4, while SMAD3 is generally unresponsive to BMP4. The gained Smad3 sites likely result from an indirect effect of BMP signaling.

Figure 3.

To evaluate the regulatory activity of SMAD4/pSMAD1/5/8 in NSCs, we determined the state of transcription activation using H3K4me3 CUT&TAG signals at annotated gene transcription start sites (TSS) in BMP4-treated and untreated Prdm16_OE cells. As cell identity genes are known to be marked by extended breadth of H3K4me3 ³⁰, we reasoned that compared to RNA-seq, which measures all gene products, changes in the H3K4me3 signal at the promoter might bias toward cell identity genes between proliferation and quiescence states. Thus, we determined all BMP4-repressed and activated genes by measuring changes in the H3K4me3 CUT&TAG read coverage at annotated TSS in BMP4-treated Prdm16_OE cells compared to untreated, identifying 282 up-regulated and 429 down-regulated genes (Fig. 3B) (P<0.01). To pinpoint genes directly regulated by SMAD4 and pSMAD1/5/8, we intersected the dysregulated genes with those whose regulatory regions contain overlapping SMAD4 and pSMAD1/5/8 ChIP-seq peaks. This analysis led to the identification of 145 up-regulated and 184 down-regulated SMAD4 and pSMAD1/5/8 target genes.

To elucidate the function of BMP-regulated targets, we performed gene ontology (GO) analyses for the genes that changed expression in response to BMP4 in Prdm16_OE cells and were also bound by SMAD4 and pSMAD1/5/8. Remarkably, the downregulated genes (BMP4-repressed genes), but not the up-regulated genes (BMP4-activated genes), showed significantly enriched functional categories, with nearly all GO terms related to cell proliferation/cell cycle (Fig. 3C). A recent study reported that BMP2-induced genes are enriched for neuronal and astrocyte differentiation ³¹, while our analysis did not identify these categories. One possibility for the discrepancy is that we overexpress Prdm16 in cultured NSCs, which may reinforce BMP signaling activities in cell proliferation. Other possibilities could be the use of different BMP ligands (BMP4 versus BMP2), differences in the origin of NSC culture (ours was from E13 instead of E11 and E14), or different profiling methods (H3K4me3 versus RNA-seq).

To identify PRDM16-repressed and activated genes under active BMP signaling conditions, we compared changes of H3K4me3 coverage at TSSs between BMP4-treated Prdm16_KO cells and BMP4-treated Prdm16_OE cells. We found approximately twice as many up-regulated genes (240) as down-regulated genes (132) (Fig. 3D), suggesting that the cooperative activity of PRDM16 and BMP signaling mainly represses gene expression. We further overlapped the 184 genes repressed by BMP4 and the 240 genes repressed by PRDM16, identifying 31 common ones. These 31 genes displayed low H3K4me3 coverage in Prdm16_OE cells with BMP4 but higher H3K4me3 signal in both Prdm16_KO cells treated with BMP4 and untreated Prdm16_OE cells (Fig. 3E).

Next, we attempted to validate whether changes in TSS H3K4me3 intensity correspond to changes in mRNA levels, by conducting RT-qPCR for selected genes whose function is associated with cell proliferation from the gene ontology analysis: Wnt7b, Mybl2, Id3, Spc24 and the three Spc24-related genes (Spc25, Ndc80 and Nuf2) (Fig. 3F-G, Supplementary Fig. 3E-K). SPC24 forms the NDC80 kinetochore complex along with three other proteins: SPC25, NDC80 and NUF2 ³² (Illustrated in Supplementary Fig. 3H), and their function is essential for chromosome segregation and spindle checkpoint activity. Notably, Spc25, Ndc80 and Nuf2 appeared repressed by PRDM16 and BMP4 based on changes in H3K4me3 at the TSS (Supplementary Fig. 3E), despite not being identified as the top 31 candidates. mRNA levels for most of the tested genes followed a similar pattern to the H3K4me3 intensity: BMP4-treated Prdm16_OE cells showed the lowest expression, while either the loss of Prdm16 or absence of BMP4 led to upregulation. Spc24 gene products showed minimal amplification in qPCR, likely due to poor primer sequence quality or unstable mRNAs. Another exception is Id3 whose expression increased upon BMP4 treatment or Prdm16 depletion, indicating PRDM16 repressing Id3 but BMP4 inducing Id3. However, the activation of Id3 by BMP4 is significantly stronger in Prdm16_KO cells than that in Prdm16_OE cells, suggesting that BMP signaling normally activates Id3 but PRDM16 suppresses such activation. In contrast, Cdkn1a, a target gene of TGF-β pathway encoding the cell cycle inhibitor P21, exhibited consistent H3K4me3 coverage and mRNA levels across BMP4 treated and non-treated cells, as well as between Prdm16_OE and KO cells (Fig. 3F-G), confirming that neither PRDM16 nor BMP signaling influences Cdkn1a expression in NSCs.

PRDM16 assists genomic binding of SMAD4 and pSMAD1/5/8

To understand how PRDM16 interacts with BMP signaling, we integrated PRDM16 CUT&TAG data with SMAD ChIP-seq data, focusing on assessing PRDM16’s influence on the genomic binding of SMAD proteins.

The SEACR peak caller software ³³ identified 3337 and 7639 PRDM16 CUT&TAG common peaks from four replicates of BMP4-treated and untreated Prdm16_OE samples (FDR < 10%, see Methods), respectively (Fig. 4A). Upon BMP4 treatment, there were 5936 lost and 1634 gained PRDM16 CUT&TAG peaks, suggesting that PRDM16 regulates a distinct subset of genes in proliferating versus quiescent NSCs. Samples from Prdm16_KO cells lacked these sites, confirming the specificity of the PRDM16 CUT&TAG signal (Fig. 4B).

Figure 4.

From our overlapping analyses (see Methods), we found that over 50% of SMAD4 and pSMAD1/5/8 binding peaks were consistent in Prdm16_OE and Prdm16_KO cells (Supplementary Fig. 4A-B), indicating that deletion of Prdm16 does not affect general genomic binding ability of these proteins. Using Homer’s mergePeaks function with PRDM16 CUT&TAG and SMAD ChIP-seq data, we identified co-bound sites by PRDM16 and SMAD proteins in BMP4-treated Prdm16_OE cells. PRDM16 CUT&TAG peaks mainly overlap with SMAD4 and pSMAD1/5/8 peaks, but not much with SMAD3 peaks (Supplementary Fig. 4A-C). This result suggests that PRDM16 mainly collaborates with the SMAD4/pSMAD1/5/8 complex but not the SMAD3/SMAD4 complex in cells with high levels of BMP4.

Further examination of SMAD4 and pSMAD1/5/8 binding revealed significantly lower enrichment of SMAD4 and pSMAD1/5/8 at PRDM16 co-bound sites in Prdm16_KO cells compared to Prdm16_OE cells (Fig. 4C) (P=2.5E-6, and P=4.7E-3, respectively, two-tailed t-test). As a control, the SMAD4 and pSMAD1/5/8 sites without PRDM16 co-binding did not show such change (P>0.05). This result suggests that SMAD binding to the PRDM16 co-binding sites depends on PRDM16.

To validate the co-binding of SMAD4 and PRDM16, we selected the candidate loci, Wnt7b and Id1 (Fig. 4D-E) and applied sequential ChIP-qPCR. This experiment confirmed simultaneous binding of SMAD4 and PRDM16 to the same DNA molecules at these loci (Supplementary Fig. 4D), as SMAD4 was more enriched in the chromatin pulled by the PRDM16 antibody compared to the IgG control. Additionally, ChIP-qPCR confirmed reduced SMAD binding at the SMAD/PRDM16 co-bound site in Prdm16_KO cells compared to Prdm16_OE cells (Fig. 4F-G). Thus, PRDM16 enhances genomic binding of the SMAD proteins to specific genome regions.

PRDM16 facilitates SMAD4 binding to regions enriched for SMAD palindromic motifs

We then wondered whether there is sequence feature that distinguishes the regions co-bound by SMADs and PRDM16 from those only bound by SMADs. To this end, we checked SMAD- bound regions with PRDM16 binding and those without for two types of SMAD4 binding motifs ⁸. Together with pSMAD1/5/8 or pSMAD3, SMAD4 may bind to a palindromic motif, GTCTAGAC or direct repeats of GTCT like GTCTGTCTGTCT ^34–36; together with pSMAD1/5/8, SMAD4 associates with GC-rich SBEs (SMAD-binding elements) including GGCGCC-AN4-GNCV and GRCGNCNNNNNGTCT ^37–39. We calculated occurrence frequency of each of these motifs in binned SMAD4 ChIP-seq peaks from Prdm16_OE cells treated with BMP4 (bin 1 to 5 with low to high peak scores, Fig. 4H). Interestingly, the palindromic motif is the most enriched one. Its occurrence frequency increases with peak scores, but becomes lower in the SMAD4 peaks absent for PRDM16 co-binding. The frequency of the GTCT-triplet motif also increases with higher SMAD4 peak scores, while there is no reduction in regions without PRDM16 binding. The two GC-rich motifs show a distinct trend: they are not as highly enriched, their frequencies do not increase with higher SMAD4 peak scores, and the occurrence frequency in PRDM16 co-bound regions is either comparable to or even lower than the regions without PRDM16 binding. Together, this result suggests that PRDM16 may separate the SMAD4/pSMAD1/5/8 proteins into two types of genomic regions, one enriched for the palindromic motif where PRDM16 is present and the other with the GC-rich motifs.

SMAD4 and pSMAD1/5/8 switch genomic locations in the absence of PRDM16

Our careful inspection on the Wnt7b and Id1 loci surprisingly revealed that there are multiple ectopic SMAD4 and pSMAD1/5/8 peaks in the Prdm16 mutant sample (arrow-indicated peaks in Fig. 5A-B). We then globally assessed the extent of ectopic SMAD binding surrounding the SMAD/PRDM16 co-bound sites by quantifying differential SMAD binding intensity between Prdm16_KO and Prdm16_OE cells (Fig. 5C, FDR < 10%). In agreement with the metaplots (Fig. 4C), all of the SMAD/PRDM16 co-bound sites (blue dots) showed reduced ChIP-seq read coverage in Prdm16_KO cells. By contrast, the flanking regions within the 200kb range of a co-bound site (red dots) displayed a trend of increase of SMAD binding in Prdm16_KO cells. This result agrees with the aforementioned motif analysis that PRDM16 helps to localize the SMAD complex at specific genomic regions, and it also suggests that without Prdm16, the SMAD factors are partly redirected to neighboring genomic regions.

Figure 5.

AP1 is the potential co-factor that interacts with relocated SMAD proteins

We intended to find out how SMAD4 and pSMAD1/5/8 relocate to ectopic sites in the absence of Prdm16. By running de novo motif discovery (see Methods), we identified a number of significantly enriched DNA motifs from pSMAD1/5/8 and SMAD4 ChIP-seq peaks in Prdm16_OE and Prdm16_KO cells (Fig. 5D-E). Interestingly, in addition to the known SMAD motifs, the only other motif identified in both pSMAD1/5/8 and SMAD4 peaks from Prdm16_KO cells but not Prdm16_OE cells is the AP1 motif. Furthermore, the AP1 motif is more frequently found in the pSMAD1/5/8 and SMAD4 regions lacking PRDM16 binding than those with PRDM16 binding (Supplementary Fig. 4E). The interaction between the AP-1 complex with the SMAD proteins was reported previously ⁴⁰. Our result implies that AP-1 is a potential cofactor for the SMAD proteins in the absence of Prdm16. To validate this finding, we performed ChIP-qPCR with an antibody against c-FOS, one of the subunits of AP1, on Prdm16_OE and KO cells. Supporting the global analyses, c-FOS is enriched in the Smad4- bound regions at the Wnt7b and Id3 loci in the Prdm16_KO condition (Fig. 5F-G). There is little c-FOS binding to these regions in cells expressing Prdm16, suggesting that there is cooperativity of SMAD and AP-1 which facilitates each other’s binding when PRDM16 is absent. Alternatively, PRDM16 may suppress genomic accessibility of these regions for SMAD and AP-1 proteins, a function we reported for PRDM16 in cortical NSCs ²⁰.

Together, our genomic data and validation suggest an enhancer switch model (Fig. 5H): PRDM16 assists SMAD protein binding to repressive cis-regulatory elements that are enriched for GTCTAGAC palindromic motifs; loss of PRDM16 results in genomic relocation of SMAD proteins, presumably via the association with coactivators such as AP1; the consequence of SMAD relocation is de-repression of cell proliferation regulators.

SMADs and PRDM16 co-repressed genes in NSCs are suppressed in the developing ChP

Next, we investigated whether genes co-regulated by PRDM16 and BMP signaling in NSCs (Fig. 3C) are also involved in ChP development, where both factors limit NSC proliferation ( ¹⁰ and Fig. 1). Using published E12.5 mouse brain single cell RNA-seq (scRNA-seq) data ⁴¹, we analyzed expression of the 31 co-repressed genes across ChP epithelial (Ttr+), CH (BMP4+/Ttr-) and a few forebrain-specific radial glia (RG) (Sox2+/Hes5+) clusters (Fig. 6A, Supplementary Fig. 5A). Except for one gene Mrtfb that was not found in the scRNA-seq data, thirteen showed no or weak expression (Supplementary Fig. 5A), and seventeen had moderate to high expression in at least one cluster (Fig. 6A). Notably, these genes were generally expressed at lower levels in ChP clusters than in CH or RG clusters, suggesting repression in the ChP.

Figure 6.

To test for direct PRDM16 regulation, we performed CUT&TAG assays using the PRDM16 antibody on dissected dorsal midline tissues that mainly contain CH and the ChP (Fig 6B and Supplementary Fig. 5B) and identified 3238 common peaks across three replicates. Manual inspection of the 31 loci revealed PRDM16 binding near the TSS of 25 genes using the browser Embryonic Mouse Brain Epigenome Atlas ⁴² (Text in black in Fig. 6A and Supplementary Fig. 5A). Using an unbiased peak-to-promoter mapping, 24/31 co-repressed and 84/153 BMP-only-repressed genes had PRDM16 binding in E12.5 ChP. The enrichment in the co-repressed group (Supplementary Fig. 5C, Fisher’s Exact Test, p = 0.015) indicates a stronger regulatory role of PRDM16 on these genes in the developing ChP.

As an example, Wnt7b has two PRDM16 peaks (Fig 6C), one unique to the dorsal midline tissue. Given Wnt signaling’s role in ChP specification and NSC proliferation^12,43. Wnt7b likely represents a direct PRDM16 target in the developing ChP.

PRDM16 represses Wnt7b and Wnt activity in the developing ChP

We sought to determine which of the identified genes from NSC culture are regulated by PRDM16 in the developing ChP. In addition to Wnt7b, several other Wnt ligands are expressed in the CH and ChP region ⁴⁴. For example, Wnt3a is expressed in the dorsal-midline region but not in the forebrain, which explains why it was not identified from the forebrain-derived NSC datasets. Although there is no PRDM16 CUT&TAG peak within the Wnt3a locus in cultured NSCs, a PRDM16 CUT&TAG peak is present in the intronic region of Wnt3a in the dorsal midline tissue (Supplementary Fig. 5D). We thus examined the scRNA-seq data ⁴¹ and confirmed that multiple components in the Wnt and BMP pathways are present in the CH and ChP clusters (Supplementary Fig. 5E). To systematically measure expression changes of PRDM16 target genes in the Prdm16 mutant brain, we applied a multiplexed fluorescent in situ approach, Single-Cell Resolution in Situ Hybridization On Tissues (SCRINSHOT) ⁴⁵. In this method each mRNA molecule is hybridized with three gene-specific padlock probes and visualized by florescent dye-conjugated detection probes. A dot of fluorescent signal corresponds to one mRNA molecule, allowing quantitative measurement of gene expression in single cell resolution.

We designed probes for ten Wnt and five BMP pathway components, other genes that are co-repressed by BMP and PRDM16 in NSC culture, as well as cell-type specific markers (the ChP marker gene Ttr and Foxj1, the neural progenitor markers Sox2, Hes5 and Zbtb20, and the neuronal marker Ngn2), and conducted SCRINSHOT in three pairs of control and Prdm16 mutant brains.

In the control brain, the Ttr and Foxj1 probes detect the ChP epithelium but not the adjacent tissues at E11.5 and E12.5, confirming the signal specificity of SCRINSHOT. Several Wnt (Wnt2b, Wnt7b, Wnt5a, Wnt3a, Fzd10 and Axin2) and BMP components (BMP4 and BMP7) also display tissue specificity: the Wnt genes are more enriched at CH while the BMP genes are more at the ChP. The representative images from E11.5 and E12.5 samples are presented in Fig. 7A-B. Except Wnt2b, the other tested Wnt genes exhibit low expression in the ChP cells, consistent with the notion that Wnt signaling is present and required at the ChP ¹².

Figure 7.

Next, we quantified expression changes of these genes in Prdm16 mutant CH and ChP cells. As expression of Wnt2b was relatively unchanged in the mutant (Supplementary Fig. 1 and Fig. 7A), we used Wnt2b-expressing cells to define the CH cells and those expressing Ttr to define ChP cells in the wild-type brain slice. There were about 110 cells in each region of one brain slice. Because the mutant brain slice showed little Ttr signal, we only used the Wnt2b-expressing cells to define the anterior border between CH and neocortex and included 110 cells under the border for the CH cells, and the further 110 cells for the mutant ChP cells, as shown in the images with DAPI and cell outlines (Fig. 7A and Fig. 7D). We then measured the dot count of each gene in these 220 cells in each sample, and summarized the changes from three pairs of animals in the violin plots (Fig. 7C and Supplementary Fig. 6A). Expression of Ttr and Foxj1 is severely reduced, while Sox2, Hes5 and Ngn2 all become upregulated in the mutant ChP, indicating a fate transformation of ChP epithelial to neural cells. This agrees with our immunostaining and conventional in situ results (Fig. 1 and Supplementary Fig. 1). Similar to BMP4, other BMP components we tested including BMP7, Smad6, Smad7 and Nog are unchanged, suggesting that PRDM16 is not an upstream regulator of the BMP pathway.

Moreover, Wnt3a and Wnt7b are significantly upregulated in the Prdm16 mutant ChP (Fig. 7B-E and Supplementary Fig. 6A-C). In line with this, the Wnt target gene Axin2 is also upregulated, indicating aberrantly elevated Wnt signaling in the Prdm16 mutant ChP. Thus, the normal function of Prdm16 represses WNT signaling in the developing ChP.

Levels of Wnt activity correlate with cell proliferation in the developing ChP and CH

We further assessed the relationship between Wnt signaling and cell proliferation in the ChP and CH at E12.5, by correlating expression levels of mKi67 with Wnt3a, Wnt7b and Axin2 levels in wild-type and mutant samples. A significant increase of mKi67 expression in Prdm16 mutant ChP cells at E12.5 is accompanied by significantly increased Wnt gene expression (Fig. 7F and Supplementary Fig 6B-C). In contrast, little or no increase of mKi67 signal is found in the mutant CH, suggesting that the effect of PRDM16 is mainly cell-autonomous. We then performed linear regression and Pearson correlation analyses to assess correlations between ChP markers, BMP and Wnt genes. Interestingly, the level of Axin2 best correlates the level of mKi67, even better than that with other Wnt genes in both wild-type (r = 0.46; p < 2.2 E-35) and mutant (r= 0.56; p < 1.6 E-55) CH and ChP cells (Fig. 7G and Supplementary Fig. 6D). This result suggests that Wnt activity may be responsible for cell proliferation in the CH and ChP region. Supporting this finding, it was shown that ectopic Wnt signaling converts ChP cells into proliferative CH neural cells ¹². Similarly, in the 4^th ventricle, two Wnt genes, Wnt1 and Wnt3a, and mKi67 all became ectopically expressed in Prdm16 mutant ChP cells (Supplementary Fig. 7). Thus, PRDM16 also suppresses Wnt signaling in the hindbrain ChP.

Additionally, six cell-proliferation-related genes (Spc24, Spc25, Nuf2, Ndc80, Id3 and Mybl2) exhibited upregulation in Prdm16 mutant ChP (Supplementary Fig. 8). Interestingly, all four NDC80 complex genes became upregulated, pointing to its potential role in promoting NSC proliferation. Taken together, our results suggest that specification of the ChP epithelium requires a process transitioning proliferating NSCs into quiescence, and that this process is mediated by a suppressive role of PRDM16 on Wnt signaling and cell cycle regulators.

Discussion

In this study we reveal a novel molecular mechanism by which BMP signaling regulates cell proliferation and gene expression. We find that PRDM16 acts as a tethering factor to localize the SMAD4/pSMAD1/5/8 complex at specific genomic sites and facilitate the repressive role of the SMAD proteins. The genes co-repressed by PRDM16 and BMP signaling include those encoding Wnt pathway ligands and other cell proliferation regulators.

Combinatory activities of morphogens are exploited throughout animal development and tissue homeostasis. The crosstalk between BMP and WNT signaling is complex as it can be synergistic or antagonistic ⁴⁶. Surprisingly, both types of effects exist in the specification of ChP epithelium. Here BMP signaling induces NSC quiescence, as evidenced by the presence of ectopic proliferating cells at the ChP in Bmpr1a mutant mice ¹⁰, while Wnt signaling promotes proliferation since the gain-of-function condition of beta-Catenin phenocopies Bmpr1a mutant animals ¹². On the other hand, adding a low dose of Wnt activator to cell culture medium enhances programing efficiency of ChP epithelial cells induced by BMP4 ⁴⁷, and loss-of-function of beta-Catenin resulted in under-developed ChP structure ¹², suggesting that BMP and WNT signaling collaborate to promote ChP epithelial cell differentiation. Our work demonstrated that PRDM16 ensures the right balance of BMP and Wnt activity by maintaining a low level of Wnt gene expression in the developing ChP.

PRDM16 was shown to physically interact with SMAD3 ²³ and antagonize TGF-β-induced cell cycle arrest ⁴⁸. However, we did not detect stable PRDM16/SMAD4/pSMAD1/5/8 or PRDM16/SMAD4/SMAD3 complexes in co-immunoprecipitation assays (data not shown). Instead, sequential ChIP revealed that PRDM16 and SMAD4 bind to the same DNA molecule, supporting their cooperative function at co-occupied regulatory elements. Notably, SMADs when associated with PRDM16 compared to those without PRDM16, with SMADs preferentially bind palindromic motifs in the presence of PRDM16 and GC-rich SBE motifs in its absence, suggesting that PRDM16 modulates SMAD DNA-binding specificity and shifts SMAD regulatory output.

A recent study similarly reported that PRDM16 acts as a SMAD4 co-repressor during pancreatic ductal adenocarcinoma progression ⁴⁹. There, SMAD4 stably binds the PRDM16 promoter and switches its co-factor from the co-repressor PRDM16 (under low TGF-β) to the co-activator SMAD3 (under high TGF-β), depending on TGF-β levels. This mechanism differs from our model. Our data indicate that high BMP4 levels promote SMAD4 and pSMAD1/5/8 binding to target loci, including Prdm16, resulting in both gene repression and activation. Notably, PRDM16 contributes to the repression function of the SMAD complex even under elevated BMP conditions.

Although PRDM16 has been shown to promote proliferation by antagonizing anti-proliferation activity of TGF-β and SMAD3 ⁴⁸, our data demonstrate that PRDM16 collaborates with SMAD4 and pSMAD1/5/8 to repress proliferation genes, which is consistent with BMP- induced quiescence. Similar roles have been observed in other contexts, such as such as osteoblast differentiation and maturation ⁵⁰. Supporting this, genomic profiling revealed that under high BMP4 and PRDM16 conditions, pSMAD1/5/8 and SMAD4 occupy numerous loci without SMAD3, suggesting PRDM16 primarily modulates BMP target genes in these settings.

Moreover, while it was reported that PRDM16 enhances Wnt signaling by stabilizing nuclear beta-Catenin in the craniofacial tissue ²⁶, we detected the opposite outcome in the ChP and cultured NSCs, ectopic WNT gene expression in the absence of Prdm16, suggesting that PRDM16 represses Wnt genes in these contexts. Notably, PRDM16, Wnt and BMP co-exist in various developmental settings, such as craniofacial development ^21,26,51, heart formation ^27,52,53, limb patterning ^21,54,55, adult intestinal stem cells ^56–58 etc. We speculate a similar regulatory circuit is used in some, if not all, of these settings.

ChP epithelial cells derive from proliferating NSCs, without intermediate fate-commitment steps. This property endows ChP cells with higher plasticity and makes them more vulnerable to abnormal genetic and cellular change, to which high occurrence frequency of ChP tumors in fetuses and young children may be attributed ⁵⁹. As an essential brain structure, the ChP releases cerebrospinal fluid (CSF) and acts as a brain-blood barrier ^60–62. Its dysfunction has been linked with several types of human diseases including hydrocephalus, Alzheimer’s disease, multiple sclerosis ⁶². Deepening our knowledge on the development of ChP will provide new insights into potential therapeutics to prevent or treat ChP tumors and other ChP-related diseases.

Footnote: a study on the antagonism between PRDM16 and SMAD4 was published on Feb 24, 2023 (Hurwitz, 2023) after we posted our manuscript at BioRxiv and during our submission process.

Materials and methods

Animals

All animal procedures were approved by Swedish agriculture board (Jordbruks Verket) with document number Dnr 11553-2017 and 11766-2022. The Prdm16^cGT mice ²² were maintained by outcrossing with the FVB/NJ line.

In situ hybridization

To make probes for conventional RNA in situ hybridization, genomic regions covering one exon or full-length cDNA were PCR amplified to generate fragments with restriction enzyme overhangs. The sequences of all oligos were included in Supplementary table 1. The fragments were inserted to the pBluescript SK(II) vector. In vitro transcription was performed as previously described (He et al., 2021). The mouse brains at defined ages were dissected and fixed for 12 hours in 4% PFA, dehydrated in 25% sucrose, cryoprotected and embedded in O.C.T. The brain samples were then sectioned at 18 µm thickness on Leica cryostats CM3050s. RNA in situ hybridization was performed using digoxigenin-labeled riboprobes as described previously. Detailed protocols are available upon request. Images were taken using a Leica DMLB microscope.

Immunostaining

Immunostaining was performed according to standard protocols as previously used ⁶³. For EdU and BrdU labeling, EdU (5-ethynyl-2′-deoxyuridine) and BrdU (5-bromo-2’-deoxyuridine) (5-20 µg/g of body weight) were injected into the peritoneal cavity of pregnant mice at desired ages. EdU incorporation was detected with the Click-iT assay (Invitrogen) according to the manufacturer’s instructions. BrdU incorporation was measured by immunostaining using an antibody against rat-BrdU (Abcam). Imaging was taken on a Zeiss confocal microscope. ZEN (ZeissLSM800), ImageJ (NIH) and Photoshop (Adobe) were used for analysis and quantification.

NSC culture

Control and mutant embryonic cortices from E13.5 animals were dissected and dissociated into single cell suspension and digested with Accutase (Sigma). Cells were maintained in proliferation media (STEMCELL Technologies). Lentivirus expressing Flag-Prdm16 and a puromycin resistant gene in the pCDH vector was produced in 293J4 cells and used to infect a wild-type NSC line derived from E13.5 forebrain, and the selected NSCs were maintained in puromycin-containing medium for three passages before puromycin withdrawal.

RT-qPCR

Prdm16_OE, wild-type and mutant NSCs were cultured and treated with or without BMP4 (Sigma). After 48-hour treatment, total RNAs were extracted using TRIzol reagent (Invitrogen). Total RNA was further cleaned with Turbo DNase (Ambion) and used in reverse-transcription with RT master mix (ThermoFisher). To ensure the absence of genomic DNA, control qPCR was performed on the mock-reverse-transcribed RNA samples. The list of qPCR primers is included in supplementary table 1.

SCRINSHOT

Brain sections were prepared in the same way as for regular in situ hybridization (described above). The SCRINSHOT experiments were carried out according to the published method ⁴⁵. In brief, 3 padlock probes and 3 corresponding detection probes were designed for each gene of interest. The slides with cryo-sectioned brain slices were pretreated at 45°C to reduce moisture and fixed in 4% PFA in 1X PBS, followed by washing in PBS Tween-20 0.05% twice. Permeabilization of tissues were done by washing slides in 0.1M HCl for 2mins 15s, followed by washing in PBS Tween-20 0.05% twice. Then a stepwise dehydration was performed for the slides in 70%, 85% & 100% Ethanol and air. The SecureSeal hybridization chamber (GRACE BIO-LABS, 621501) was then mounted to cover each pair of control and Prdm16 mutant brain slices. Samples were then blocked in a probe-free hybridization reaction mixture of 1XAmplifase buffer (Lucigen, A1905B), 0.05M KCl, 20% Formamide deionized (Millipore S4117), 0.1uM Oligo-dT, 0.1ug/ul BSA (New England Biolabs, B9000S), 1U/ul RiboLock (Thermo, EO0384), and 0.2ug/ul tRNAs (Ambion, AM7119). Then hybridization of padlock probes was done by incubating samples with padlock probes (with the concentration of each one 0.01uM) mixed in blocking reagents used before (no oligo dT used in this step). The slide was then put into a PCR machine to denature at 55°C for 15 minutes and hybridize at 45°C for 120 minutes. Padlock probes were then ligated by using SplintR ligase (NEB M0375) at 25°C for 16 hours followed by RCA (rolling cycle amplification) at 30°C for 16 hours by using phi29 polymerase (Lucigen, 30221–2) and RCA primer1. Then a fixation step was applied to stabilize

RCA product in 4% PFA for 15 minutes followed by washing in PBS Tween-20 0.05%. Then the hybridization of the first 3 genes was done by mixing all 3 3’ fluorophore-conjugated detection probes of each gene in reaction reagent (2XSSC, 20% Formamide deionized, 0.1ug/ul BSA, and 0.5 ng/µl DAPI (Biolegend, 422801) followed by hybridization at 30°C for 1 hour. The slides were washed in 20% formamide in 2X SSC and then in 6X SSC, followed by dehydration in 70%, 85%, 100% Ethanol until the chamber was removed. Then the samples were preserved in SlowFade Gold Antifade mountant (Thermo, S36936) and kept in dark before imaging. After image acquisition, the first 3 detection probes were removed by using Uracil-DNA Glycosylase (Thermo, EN0362), and the slides were ready for the next round of detection probe hybridization. The procedure was repeated until all genes were hybridized and imaged.

Images were acquired with Zeiss Axio Observer 7 fluorescent microscope with an automated stage setting to fix imaging region for different hybridization rounds. Image analysis was done according to the published method ⁴⁵. In brief, the DAPI channel from each round was extracted to measure the shift of imaging. The images were then aligned for all gene channels in Zen by Creating Image Subset with the shifting value. After alignment, images were exported into TIFF files and the threshold analysis was carried out for individual channel one by one in CellProfiler with scripts provided by the published method (Alexandros et al. 2020). Then nuclear segmentation was done manually in Fiji ROI manager to obtain nuclear ROIs, followed by an expansion of 2 um in CellProfiler to obtain cell ROIs. Then signal dots were count in these cell ROIs for each gene in CellProfiler and Fiji. Summary of the dot counts for each gene was exported to Excel files.

CUT&TAG

CUT&TAG was performed according to the published method ⁶⁴. In brief, Prdm16_OE and Prdm16 mutant NSCs were cultured in NeuroCult™ Proliferation Kit (Stem cell technologies, 05702) for 3 days with or without BMP4 at concentration of 25ng/ml medium (Sigma, H4916). Cells on day 3 were then shortly rinsed with 1X PBS and resuspended in 1ml cold NE1 buffer on ice. Nuclei were lightly fixed with 0.1% formaldehyde to 1ml PBS at RT for 2 minutes and neutralized with 75mM Glycine. Nuclei were then washed for 3 times, resuspended in 1mL wash buffer and mixed with 90uL Concanavalin A-coated magnetic beads (Polyscience 86057). The nuclei/beads mix was then blocked with 800 uL cold Antibody buffer for 5 minutes and resuspended in 1.2 mL Antibody buffer and aliquoted into 8 tubes with 150 uL each. 2 uL Anti-PRDM16 (Generous gift from Bryan Bjork lab) or IgG (SIGMA-ALDRICH, I5006) was added into each tube for an overnight incubation at 4°C. The beads/nuclei/antibody mix was then washed with Dig-wash buffer and incubated with secondary antibody (1:100 dilution) for 1 hour. After further washes with the Dig-wash buffer. a pA-Tn5 adaptor complex in Dig-300 buffer was added to the beads for 1 hour reaction. The beads were then washed with Dig-300 buffer before the incubation with 300ul Tagmentation buffer at 37°C for 1 hour. Then 10ul 0.5M EDTA, 3uL 10% SDS and 2.5uL 20 mg/mL proteinase K (Invitrogen) were used to stop the reaction. Then the resultant DNA was purified with DNA clean & Concentrator kit (ZYMO Research D4013) and eluted in 25uL Elution buffer. To generate libraries, 21uL fragment DNA was mixed with 2uL 10uM Universal i5 primer, 2uL 10uM uniquely barcoded i7 primer and 25uL PCR master mix (NEB Phusion® High-Fidelity PCR Master Mix with HF Buffer, M0531). The PCR condition was as follows: 72°C 5min, 98°C 30s, repeat 12 times (98°C 10s, 63°C 10s), 72°C 1 min and hold 4°C. The libraries were cleaned with standard Ampure XP beads as previously described. Libraries from four biological replicates were produced for each condition.

ChIP

ChIP was performed as previously described ⁶⁵. For each ChIP reaction, 10 million Prdm16_OE, Prdm16_KO NSCs with or without 3-day BMP4 treatment were fixed, lysed, sonicated and made into chromatin extract. After precleared with gamma-bind-G beads, the chromatin extract was incubated with 2ug PRDM16, 2ug SMAD4 (Proteintech, 10231-1-AP), 2ug pSMAD1/5/8 (Millipore, AB3848-1) or 2ug SMAD3 (abcam, ab227223) in each ChIP reaction. In sequential ChIP assays, the same chromatin lysate was precleared and used, but with more antibodies in each reaction: 5 ug of IgG or PRDM16 antibody in the first round of immunoprecipitation. The elutes were then precleared again using gamma-bind G beads, divided into two equal halves and immunoprecipitated with 2 ug of IgG or the SMAD4 antibody. In ChIP-qPCR experiments, 2ug SMAD4 and 2 ug c-FOS antibodies (Invitrogen, MA5-15055) were used in each reaction. The precipitated DNA was reverse cross-linked, and then purified using the Qiagen PCR purification kit.

Computation analyses

ChIP-seq libraries and analyses

0.2% input and ChIPed DNA were made into libraries using the NEBNext Ultra™ II DNA Library Prep Kit and sequenced on the Illumina Nextseq500 platform. Three replicates of ChIP-seq samples, after the adaptor trimming by Trimmomatic, were mapped to the UCSC Mus musculus (mm10) genome assembly using Bowtie2 with the default parameters. The uniquely mapped reads (with mapping quality >= 20) were used for further analyses. The peaks were called by HOMER (v4.10) ⁶⁶. The reproducibility between replicates was estimated by Irreproducibility Discovery Rate (IDR), using the HOMER IDR pipeline (https://github.com/karmel/homer-idr). As suggested by the Encode IDR guideline, we used a relatively relaxed parameter “-F 2 -fdr 0.3 -P .1 -L 3 -LP .1” for the true/pseudo/pooled replicates by the HOMER peak calling. The final confident peaks were determined by an IDR < 5%. The peaks that were overlapped with mm10 blacklist were also removed.

CUT&TAG analyses

The CUT&TAG samples of Prdm16_OE and Prdm16 mutant NSCs in four replicates, and of Prdm16 from CH and ChP area at stage 12.5 embryos in three replicates, were mapped to mm10 genome assembly using Bowtie2 (bowtie2 --end-to-end --very-sensitive --no-mixed --no-discordant --phred33 -I 10 -X 700). The CUT&TAG coverage for these samples were generated by bedtools genomeCoverageBed and normalized to the library depths to give a read per million per base (RPM). By using the peak caller SEACR ³³, which was designed for calling peaks from sparse chromatin profiling data such as CUT&TAG [22], we performed peak calling with FDR < 10% for each replicated sample (SEACR_1.3.sh normalized_coverage 0.1 non stringent output_peaks). To identify confident peaks, we selected the common peaks, which were called from all replicates for each condition. Peak overlapping analysis was performed by Homer mergePeaks function with default parameters. Peak overlap analysis utilized HOMER’s mergePeaks function, specifically employing the default ’-d given’ parameter. This approach ensures that only peaks directly overlapping with each other are merged, maintaining the precise distances between peaks as provided in the input peak calls. This method is particularly beneficial for retaining the integrity of the original peak boundaries, avoiding any alterations to their relative positions. De novo motif discovery from SMAD4 and pSMAD1/5/8 peaks in Prdm16_OE NSCs and Prdm16 mutant NSCs were performed by MEME-ChIP software ⁶⁷ (parameters: -ccut 100 -meme-p 5 -dna -meme-mod anr -minw 5 - maxw 15 -filter-thresh 0.05).

The CUT&TAG samples of H3K4me3 BMP4-treated and non-treated Prdm16_OE and Prdm16_KO cells in 2 or 3 replicates were mapped to mm10 genome assembly using Bowtie2 (bowtie2 --end-to-end --very-sensitive --no-mixed --no-discordant --phred33 -I 10 -X 700). Differential analysis of TSS up-stream and down-stream 500 bp between Prdm16_OE BMP- treated vs non-BMP-treated and Prdm16_KO vs Prdm16_OE BMP-treated were performed by Limma R package ⁶⁸. Gene Ontology (GO) enrichment analysis of up-regulated and down-regulated regions was performed by PANTHER ⁶⁹.

scRNA-seq analyses

The scRNA-seq data of the developing mouse brain were obtained from⁴¹. The counts per cell were normalized by logNormCounts function in Bioconductor package “scater” and the normalized expression data per cell were used to generate gene expression violin plots. To test whether Prdm16 target gene sets of Prdm16 with and without BMP4 and ChP E12.5 are significantly enriched among the scRNA-seq gene mean expression in each identified cell type cluster, the gene set enrichment analysis (GESA) “fgesa” Bioconductor package was used [ref: Korotkevich G. et al. Fast gene set enrichment analysis. bioRxiv, 2021. http://biorxiv.org/content/early/2016/06/20/060012].

Data availability

The CUT&TAG and ChIP-seq data produced in this study have been deposited to Gene Expression Omnibus (GEO accession number: GSE275758).

Supplementary figures

Supplementary Figure 1.

Supplementary Figure 2.

Supplementary Figure 3.

Supplementary Figure 4.

Supplementary Figure 5.

Supplementary Figure 6.

Supplementary Figure 7.

Supplementary Figure 8.

Supplementary Figure 9.

Acknowledgements

We thank the animal experimental core facility and the imaging facility of Stockholm University and Bioinformatics and Expression analysis core facility at Karolinska Institute, Sweden, for their service and support. We thank Christos Samakovlis and his lab members for technical help on SCRINSHOT experiments. We also appreciate technical help from Adrian Martinez Martin. J.W. is funded by Australian Research Council Centre of Excellence for the Mathematical Analysis of Cellular Systems (CE230100001) and ANU Future Scheme. The project was supported by the research project grant from Swedish Research Council (Vetenskapsrådet, 2020-03543) and the research grants from Swedish Cancerfonden (CAN 2017/529 and 20 1046 PjF 01 H) to Q.D.

Additional information

Author contributions

Q.D. conceived and designed the project. L.H. performed all of the experiments. J.W performed all computational analysis. Q.D., J.W. and L.H. analysed and interpreted the data. Q.D. wrote the manuscript with input from L.H. and J.W.

Additional files

Reagent list