Stag1 is required for mammalian embryonic development (Remeseiro et al., 2012), indicating that Stag2 cannot compensate for the loss of Stag1. To examine the physiological functions of Stag2 in the mouse, we created a Stag2 ‘floxed’ mouse line (Stag2^f/f) by homologous recombination with a template that contained two LoxP sites flanking exon 8 (Figure 1A, B) and targeted a critical exon (exon 8) of Stag2, which is located on the X chromosome, using CRISPR–Cas9 (Figure 1—figure supplement 1A). The Stag2^null embryos showed severe developmental defects and underwent necrosis by E11.5 days (Figure 1—figure supplement 1B). Thus, Stag2 is required for mouse embryonic development, consistent with a previous report (De Koninck et al., 2020). Stag1 and Stag2 have nonredundant developmental functions.

To study the functions of STAG2 in adult mice, we crossed the Stag2^f/f mice with mice bearing the Rosa26^CreErt2 genomic insertion and generated Stag2^f/y;Rosa26^CreErt2 progenies. The Stag2^f/y;Rosa26^CreErt2 adult mice were injected with tamoxifen to induce Stag2 deletion in the whole body (Figure 1—figure supplement 2A). Genotyping analysis of blood extracts showed that tamoxifen-induced efficient disruption of the Stag2 gene locus in Stag2^f/y;Rosa26^CreErt2 mice (Figure 1—figure supplement 2B, C). These Stag2-deficient adult mice did not show early onset of spontaneous tumor formation, indicating that Stag2 mutation alone in somatic cells of mice is insufficient to induce tumorigenesis. The Stag2-deficient mice also did not have other obvious adverse phenotypes (Figure 1—figure supplement 2D), except that they had slightly lower body weight (Figure 1—figure supplement 2E, F), probably due to tissue homeostasis alterations reported by others (De Koninck et al., 2020).

STAG2 mutations are found in human cohesinopathy patients with mental retardation and neuropsychiatric behaviors (Soardi et al., 2017). To study the function of STAG2 in the nervous system, we generated Stag2 conditional knockout mice (Stag2 cKO) by crossing Stag2^f/f mice with Nestin-Cre mice (Giusti et al., 2014; Figure 1C, D). The progenies were born in the Mendelian ratio, but Stag2^f/y;Nes^Cre pups presented growth retardation and premature death (Figure 1E, G). More than 50% Stag2^f/y;Nes^Cre mice died aged about 3 weeks while the rest died at about 4 months. Stag2^f/y mice did not show differences discernible from wild-type (WT) littermates. Although Stag2^f/y;Nes^Cre mice did not present microcephaly, they exhibited frequent hydrocephaly that might contribute to their premature death. The Stag2^f/y;Nes^Cre mice displayed normal drinking and feeding behaviors (Figure 1H), but showed reduced plasma IGF-1 levels compared to the control mice (Figure 1J). Stag2^f/y;Nes^Cre mice showed forepaw and hindlimb clasping (Figure 1K) and limb tremors (Video 1), which were not seen in Stag2^f/y mice. These data indicate that Stag2 deficiency in the nervous system causes growth retardation and neurological defects.

Hematoxylin and eosin (H&E) staining of brain sections of Stag2^f/y;Nes^Cre mice did not reveal overt anatomical defects (Figure 2—figure supplement 1A). As revealed by immunohistochemistry assays using neuron- or astrocyte-specific antibodies, the differentiation of neurons and astrocytes in Stag2-deleted brains was largely normal (Figure 2—figure supplement 1B–E). To understand the origins of neurological defects caused by Stag2 deletion, we analyzed the gene expression changes in Stag2^f/y;Nes^Cre mouse brains at post-natal day 21 by RNA-sequencing (RNA-seq) (Figure 2A). Compared with the control groups, 105 and 62 genes were significantly down- or upregulated by more than twofolds, respectively, in the Stag2-deficient brains. The decreased expression of top differentially expressed genes (DEGs) was confirmed by reverse transcription quantitative PCR (RT-qPCR) (Figure 2B). Among the 105 downregulated DEGs in the brains of Stag2 cKO mice, 44 were enriched in myelin (Figure 2C; Thakurela et al., 2016). The ingenuity pathway analysis (IPA) pinpoints cholesterol biosynthesis pathways as the most affected canonical pathways (Figure 2D and Supplementary file 1). We further confirmed that the cholesterol biosynthesis precursors were reduced in Stag2^f/y;Nes^Cre brains (Figure 2—figure supplement 1F).

Myelin is the membrane sheath that wraps around axons to facilitate rapid nerve conduction and maintain metabolic supply (Williamson and Lyons, 2018). Dynamic myelination in the central nervous system (CNS) is critical for proper neurodevelopment, and defective myelination is associated with autoimmune and neurodegenerative diseases (Mathys et al., 2019; Wolf et al., 2021). Cholesterol biosynthesis is essential for normal myelination (Hubler et al., 2018; Saher et al., 2005). Ensheathment of neurons and gliogenesis were among the top enriched biological pathways in downregulated DEGs (Figure 2—figure supplement 2). The innate immune response was among the top enriched pathways in the upregulated DEGs. We hypothesized that depletion of STAG2 caused myelination defects in the nervous system.

Indeed, brain sections of Stag2^f/y;Nes^Cre mice showed greatly reduced luxol fast blue (LFB) staining compared to those of Stag2^f/y and Nes^Cre heterozygous mice at about P21 (Figure 3A and Figure 3—figure supplement 1A). Immunohistochemistry using antibodies against myelin proteins, Myelin basic protein (MBP) and Proteolipid protein 1 (PLP1), confirmed that Stag2 cKO mice had significant defects in myelin fiber formation at P18-P21 (Figure 3B–F). In both cerebral cortex and cerebellum, there were fewer and sparser myelin fibers in Stag2^f/y;Nes^Cre mice, as compared to the Stag2^f/y controls. Axon myelin ensheathment was further examined using transmission electron microscopy (Figure 3G). Stag2^f/y;Nes^Cre mice at P18 had significantly fewer myelin-wrapped axons at optic nerves. Collectively, these data indicate insufficient myelination in the Stag2 cKO mice. Myelination predominantly occurs at 3 weeks after birth in the mouse. The timing of premature death of Stag2 cKO mice is thus consistent with defective myelination as a contributing factor to the lethality.

We examined Stag1 and Stag2 expression patterns in P18 WT mouse brains by in situ hybridization using isotope-labeled RNA probes (Figure 3—figure supplement 1B). Both Stag1 and Stag2 were expressed at high levels in hippocampus, medial habenula, neocortex, and cerebellum granular layer. Aside from these regions, the Stag2 transcripts were detected at relatively high levels in subventricular zone, thalamus, fiber tracts, midbrain, and hindbrain regions. Stag2 is thus ubiquitously expressed in the brain.

Oligodendrocytes (OLs) are responsible for myelination in the CNS. To examine whether the OL lineage was affected by Stag2 deletion, we performed single-cell RNA-sequencing (scRNA-seq) analysis of Stag2^f/y;Nes^Cre and Stag2^f/y forebrains at P13. As revealed by clusters in the t-SNE plot, the two genotype groups had similar cellular compositions (Figure 4A, B). All cell clusters were present in Stag2^f/y;Nes^Cre brains, again indicating generally normal neural cell differentiation. Cell-type identities were discovered with feature gene expression (Figure 4—figure supplement 1A). Based on the expression changes of Stag2 and other cohesin genes in OLs, astrocytes, and neuronal lineages (Figure 4—figure supplement 1B–D), it is clear that Stag2 ablation occurred early in the NPC stage and was maintained in all differentiated cell lineages.

The OL lineage consisted of five clusters: cycling OL progenitors (OPCcycs), OL progenitors (OPCs), newly formed OLs (NFOLs), myelin-forming OLs (mFOLs), and fully matured OLs (MOLs). Quantification of the distributions of these five cell types within the OL lineage revealed a mild reduction in the proportion of MOLs in Stag2^f/y forebrains (Figure 4C). We noticed that a higher percentage of neurons was recovered in the Stag2^f/y;Nes^Cre group. Since the bulk RNA-seq results did not show global upregulation of neuron-specific genes, we suspect that neurons in Stag2^f/y;Nes^Cre had fewer myelin-wrapped axons and were easier to be dissociated and kept alive during our library preparation for scRNA-seq. Thus, from the transcriptome analysis, we did not observe overt defects in most neural cell differentiation in the Stag2-deficient forebrain regions.

We then performed trajectory inference and pseudotime analysis of the OL lineage (Figure 4—figure supplement 2A, B). Consistent with our cell-type assignment, pseudotime variables indicated continuous differentiation from OPCs to NFOLs, mFOLs, and MOLs (Figure 4—figure supplement 2C,D). The reclustering of single cells in the OL lineage along the pseudotime path revealed that more cells were present in the terminal maturation stages in the Stag2^f/y brains (Figure 4—figure supplement 2E, F). Conversely, more cells were retained at the undifferentiated stages in the Stag2^f/y;Nes^Cre brains. Strikingly, some myelination genes, including Myelin and lymphocyte protein (Mal), were specifically repressed in Stag2^f/y;Nes^Cre MOLs, with their expression in nonneural cells unaltered (Figure 4D and Figure 4—figure supplement 3A). These observations suggest that STAG2 deficiency delays the maturation of OLs and compromises myelination-specific gene expression in mature OLs. Interestingly, compared to Stag2 and genes encoding other cohesin core subunits, Stag1 transcripts are less abundant in the OL lineage, except for cycling OPCs (Figure 4—figure supplement 3B, C). The low expression of Stag1 in mature OLs might make these cells more dependent on Stag2 for function.

To confirm the transcriptional defects in the OL lineage caused by Stag2 deletion, we isolated primary OLs at intermediate differentiation stages from Stag2^f/y;Nes^Cre and Stag2^f/y forebrains at P12-P14 with antibody-conjugated magnetic beads and conducted bulk RNA-seq analysis (Figure 4E). For both genotypes, the marker genes for NFOL and mFOLs were highly expressed in the isolated primary OLs (Figure 4—figure supplement 4A), suggesting that they mainly contained these two cell types. In Stag2-deleted OLs, 271 and 292 genes were downregulated or upregulated by more than two folds, respectively (Figure 4F and Supplementary file 2). Intriguingly, the downregulated genes were generally highly expressed in WT cells, whereas the upregulated genes had low expression levels in WT cells (Figure 4—figure supplement 4B–D). The top pathways enriched in the downregulated DEGs included the cholesterol and small molecule biosynthetic pathways and oligodendrocyte differentiation (Figure 4G and Figure 4—figure supplement 5). Cilium organization and assembly are the top enriched pathways in the upregulated DEGs (Figure 4—figure supplement 6). Among the 105 downregulated DEGs identified by RNA-seq analysis of the whole brain of Stag2-deficient mice, 42 were also differentially expressed in primary oligodendrocytes (Figure 4H). The cholesterol biosynthetic pathways were recognized as the major altered pathways (Figure 4—figure supplement 4E). Thus, defective cholesterol biosynthesis and oligodendrocyte differentiation likely underly hypomyelination and neurological defects in Stag2 cKO mice.

We performed chromatin immunoprecipitation sequencing (ChIP-seq) experiments to examine the enrichment of the active transcription mark H3K27ac in Stag2^f/y and Stag2^f/y;Nes^Cre OLs and found that Stag2 loss did not appreciably affect H3K27Ac enrichment at transcription start sites (TSSs) (Figure 5A, B). Consistent with our RNA-seq results, the upregulated genes had much lower H3K27ac enrichment near their TSS, indicating that they were less active. We then checked the genomic distribution of STAG2 by ChIP-seq. Among other genomic loci, STAG2 was enriched at TSS of stable and downregulated genes, including genes in the cholesterol biosynthesis and myelination pathways (Figure 5C, D, Figure 5—figure supplement 1, and Supplementary file 2). Among the 271 downregulated DEGs, there were 210 genes (77%) with STAG2 enrichment near the transcriptional start site (TSS ± 2 kb). Thus, STAG2 occupied the promoter regions of many downregulated DEGs in oligodendrocytes. It was enriched at the TSS of upregulated genes to a lesser extent, with only 117 of the 292 (40%) upregulated DEGs exhibiting STAG2 ChIP-seq peaks at their TSS ± 2 kb regions. Stag2 loss might have indirectly affected the expression of these less active genes.

To investigate whether chromosome conformation was altered by Stag2 deletion and whether that caused transcription dysregulation, we performed high-dimensional chromosome conformational capture (Hi-C) analysis of primary OLs isolated from Stag2^f/y and Stag2^f/y;Nes^Cre mice in biological replicates (Figure 6 and Figure 6—figure supplement 1). We observed few compartment switching events in Stag2-deleted cells (Figure 6A–C). Virtually all genomic regions in Stag2-deleted cells were kept in their original compartment categories (AA or BB) (Figure 6C). Only a very small number of genomic regions switched compartments (AB or BA). Consistent with the RNA-seq data, analysis of average gene expression changes of DEGs in these genomic regions revealed that more genes located in the transcriptionally active A compartment (AA) were repressed in Stag2-deleted cells and more genes in the transcriptionally silent B compartment (BB) were upregulated (Figure 6D and Figure 6—figure supplement 1C). Genes that switched from the A compartment to the B compartment were not more repressed compared to those that remained in the A compartment. Likewise, compared to genes that stayed in the B compartment, genes located in chromatin regions that switched from compartment B to A were not significantly activated. Acute depletion of all forms of cohesin eliminates TAD formation (Wutz et al., 2017). In contrast, deletion of Stag2 had minimal impact on TAD formation in oligodendrocytes (Figure 6E–G and Figure 6—figure supplement 1D), suggesting that STAG1-cohesin compensates for the loss of STAG2-cohesin in spatial organization of chromatin at larger than megabase scales. Therefore, our analyses did not uncover evidence for compartment switching and TAD alterations being the underlying cause for the observed gene expression changes in STAG2-deficient OLs.

While TAD boundaries are largely conserved among species and cell types, chromatin interactions within each TAD are more flexible and variable in cells undergoing differentiation, tumorigenesis, and reprogramming (Dixon et al., 2015; Dixon et al., 2012). Among the intra-TAD chromatin interactions, the enhancer–promoter loops are particularly important for transcription and are often cell-type specific. We examined whether chromatin loops in OLs were affected by Stag2 loss. Compared to Stag2^f/y OLs, Stag2^f/y;Nes^Cre OLs had significantly fewer loops across almost all genomic distances (Figure 7A, B and Figure 7—figure supplement 1). The common and genotype-specific loops are reproducible in each replicate. Loops specific to Stag2^f/y;Nes^Cre OLs, which were likely mediated by STAG1-cohesin, were longer than STAG2-dependent Stag2^f/y-specific loops. When genomic distances exceeded 0.25 Mb, the loops from Stag2^f/y;Nes^Cre cells gradually gained higher scores over loops from Stag2^f/y cells (Figure 7C). Therefore, STAG1-cohesin cannot completely compensate for STAG2-cohesin during loop formation. STAG1-cohesin-mediated loops are relatively longer than STAG2-cohesin-mediated loops, consistent with published findings in HeLa cells (Wutz et al., 2020).

We then tested whether the loop number decrease in Stag2-deficient cells could be a cause for transcriptional changes. When examining the local Hi-C maps, we noticed that loops anchored at gene promoters, including those of downregulated genes, were reduced in Stag2^f/y;Nes^Cre oligodendrocytes (Figure 7D, Figure 7—figure supplement 2A–C, and Supplementary file 2). The effects were again reproducible in each replicate (Figure 7—figure supplement 2D). Promoter-anchored loops (P-loops) can potentially be promoter–promoter links, promoter–enhancer links, and gene loops. The total number of P-loops was proportionally decreased in Stag2^f/y;Nes^Cre cells (Figure 7—figure supplement 3A). Moreover, the loops anchored at the downregulated genes were stronger than those at upregulated and stable genes (Figure 7—figure supplement 3B). We then compared P-loops associated with DEGs using pile-up analysis of local contact maps. Loop enrichment at promoters of downregulated genes was reduced in Stag2^f/y;Nes^Cre cells to a greater extent than that at promoters of upregulated and stable genes (Figure 7E). Among the 162 downregulated DEGs with reduced promoter-anchored loops in the Stag2-depleted cells, 137 genes (85%) had STAG2 peaks in their promoter regions (TSS ± 2 kb). The loops anchored at downregulated genes with STAG2 binding had significantly higher loop scores, compared to those with no STAG2 binding (Figure 7—figure supplement 3C, D). This difference was still observed in Stag2-deleted cells, suggesting that the stronger looping at these gene promoters might be maintained by STAG1-cohesin or other factors in these cells. By contrast, the loops anchored at upregulated genes with STAG2 binding had lower loop scores. These differences became insignificant in the Stag2-deleted cells. The loop scores of loops anchored at stable genes were not affected by STAG2 occupancy. Taken together, our results suggest that Stag2 loss diminishes the number of, but not the strength, short chromosome loops, including promoter-anchored loops. Highly expressed genes might be more reliant on these loops for transcription and are preferentially downregulated by Stag2 loss (Figure 4—figure supplement 4B-D).

We also performed pile-up analysis of local chromatin regions flanking TSSs (Figure 7F). Strikingly, we observed a clear stripe that extended from the TSS of downregulated gene only in the direction of transcription. The formation of promoter-anchored stripes (P-stripes) on aggregated plots is consistent with one-sided loop extrusion from the promoter to the gene body. The P-stripe was still present in Stag2^f/y;Nes^Cre cells, suggesting that STAG1 could compensate for the loss of STAG2 and mediate its formation (Figure 7F and Figure 7—figure supplement 3E).

Selective ablation of Stag2 in the nervous system in the mouse causes growth retardation, neurological defects, and premature death. STAG2 loss delays the maturation of oligodendrocytes and reduces the expression of highly active myelin and cholesterol biosynthesis genes in oligodendrocytes, resulting in hypomyelination in the CNS. Hypomyelination disorders in humans and mice are known to produce abnormal neurological behaviors similar to those seen in our Stag2 cKO mice, suggesting that hypomyelination is a major underlying cause for the phenotypes in Stag2 cKO mice. The growth retardation in these mice can be explained by insufficient secretion of growth hormones, which may be a consequence of defective neuronal signaling.

Mutations of cohesin subunits and regulators, including STAG2, cause the Cornelia de Lange syndrome (CdLS) and other similar developmental diseases, collectively termed cohesinopathy. CdLS patients exhibit short stature and developmental defects in multiple tissues and organs, including the brain. Although STAG2 mutations are implicated in human cohesinopathy, these mutations are rare and hypomorphic (Soardi et al., 2017). The cohesin loader NIPBL is the most frequently mutated cohesin regulator in cohesinopathy (Mannini et al., 2013). NIPBL deficiency is expected to affect the functions of both STAG1- and STAG2-cohesin. It is possible that the partial loss of STAG2-cohesin function leads to subtle myelination defects in patients with cohesinopathy. Indeed, lack of myelination in certain brain regions of CdLS patients has been reported (Avagliano et al., 2017; Vuilleumier et al., 2002). As myelination of the CNS mostly occurs after birth and during childhood, strategies aimed at enhancing myelination might help to alleviate certain disease phenotypes and symptoms.

STAG2 promotes oligodendrocyte maturation and the expression of myelination genes in mature oligodendrocytes. Because STAG2 does not have an established cohesin-independent function, it most likely activates the myelination-promoting transcriptional program as a core component of cohesin. Consistent with previous reports (Rao et al., 2017), loss of STAG2-cohesin in oligodendrocytes does not affect genome compartmentalization, but reduces the number of relatively short chromosome loops, including promoter-anchored loops. Promoter-anchored loops at downregulated genes are reduced to a greater extent than those at stable and upregulated genes. These findings suggest that STAG2-cohesin promotes the myelination transcriptional program by forming promoter-anchored loops.

Pile-up analysis of Hi-C maps reveals the formation of asymmetric promoter-anchored stripes in the direction of transcription at downregulated genes, indicative of active loading of cohesin at TSSs followed by one-sided loop extrusion from the promoter to the gene body. The stripes are, however, not reduced in STAG2-deficient cells. Because both forms of cohesin are capable of loop extrusion, it is possible that STAG1-cohesin can compensate for the loss of STAG2-cohesin in loop extrusion. It remains to be tested whether the intrinsic kinetics and processivity of loop extrusion mediated by the two forms of cohesin are differentially regulated by cellular factors or posttranslational modifications and whether these differences contribute to their nonredundant roles in transcription regulation.

We envision three possibilities that may account for why oligodendrocytes, but not other cell types, are more severely affected by Stag2 loss in the CNS. First, STAG2-cohesin may be more abundant than STAG1-cohesin in postmitotic OLs, making them more dependent on STAG2 for proper functions. Second, STAG1-cohesin preferentially localizes to CTCF-enriched TAD boundaries whereas STAG2-cohesin is more enriched at enhancers lacking CTCF (Kojic et al., 2018). Enhancers are critical for cell-type-specific gene transcriptional programs. To cooperate with the axonal growth during postnatal neurodevelopment, enhancer-enriched transcription factors induce timely and robust gene expression in oligodendrocytes for proper myelination (Mitew et al., 2014). The high demand for enhancer function may render the transcription of myelination genes more reliant on STAG2-cohesin. Finally, the C-terminal regions of STAG1 and STAG2 are divergent in sequence and may bind to different interacting proteins and be subjected to differential regulation. STAG2 may interact with oligodendrocyte-specific transcription factors and be preferentially recruited to myelination genes. It will be interesting to investigate the interactomes of STAG1 and STAG2 in oligodendrocytes using mass spectrometry.

The mechanisms by which STAG2-dependent chromosome looping facilitates transcription are unclear at present. We propose several models that are not mutually exclusive (Figure 8). First, by forming promoter–enhancer loops, STAG2-cohesin brings the mediator complex and other enhancer-binding factors to the spatial proximity of the general transcriptional machinery at the promoter, thereby enhancing RNA polymerase II recruitment and transcription initiation. The existence of P-stripes at STAG2-dependent genes in the Hi-C maps suggests that STAG2-mediated promoter–enhancer loops may involve enhancers located in the gene body. Second, loop extrusion by STAG2-cohesin may promote transcription elongation by regulating transcription-coupled pre-mRNA processing. For example, STAG2 has been shown to interact with RNA–DNA hybrid structures termed R-loops in vitro and in cells (Pan et al., 2020; Porter et al., 2021). R-loops formed between the nascent pre-mRNA and the DNA template impede transcription elongation and need to be suppressed (Moore and Proudfoot, 2009). When traveling with the transcription machinery on DNA, STAG2-cohesin might directly suppress R-loop formation or recruit other factors, such as the spliceosome, for cotranscriptional pre-mRNA processing and R-loop resolution. Third, STAG2-cohesin may establish promoter–terminator gene loops to recycle the RNA polymerase II that has finished one cycle of transcription back to the TSS for another round of transcription. Future experiments using high-resolution Hi-C methods in oligodendrocytes and ChIP-seq experiments with additional enhancer- and promoter-specific histone marks will allow us to better define the nature of STAG2-dependent promoter-anchored loops and stripes. It will also be interesting to examine whether Stag2 deletion causes the accumulation of R-loops in downregulated genes and the incomplete splicing of their pre-mRNAs.

We have discovered a requirement for the cohesin subunit STAG2 in the myelination of the CNS in mammals. Our findings implicate hypomyelination as a contributing factor to certain phenotypes of cohesinopathy, including growth retardation and neurological disorders. We provide evidence to suggest that STAG2 promotes the myelination transcriptional program in oligodendrocytes through the formation of promoter-anchored loops. Our study establishes oligodendrocytes as a physiologically relevant cell system for dissecting the cellular functions and regulatory mechanisms of cohesin-mediated chromosome folding and genome organization.

All animals were handled in accordance with institutional guidelines of the Institutional Animal Care and Use Committee (IACUC; AAALAC unit number 000673) of University of Texas (UT) Southwestern Medical Center under the animal protocol number (APN) 102335. The Stag2 locus was targeted by inserting one neo cassette and two loxP sites flanking exon 8 via homologous recombination in the mouse embryonic stem (ES) cells. G418-selected positive ES clones were screened for successful targeting by nested PCR tests on both 5′ and 3′ integration sites of loxP. Four confirmed ES clones were then microinjected into mouse blastocysts. The chimeras were bred to the R26FLP mouse line for the removal of the neo cassette. Stag2^f/+ mice with the 129/B6 background were crossed with Stag2^f/y or WT C57BL/6J mice and maintained on this background. For the generation of the inducible system of Stag2^f/y;Rosa26^CreErt2 mice, Stag2^f/f mice were bred to the mouse strain that contains two alleles of the conditional Cre-ER^T2 cassette (B6.129-Gt(ROSA)26Sortm1(cre/ERT2)Tyj/J, JAX stock #008463) (Ventura et al., 2007). For the generation of the nervous system-specific Stag2^f/y;Nes^Cre mice, the Stag2^f/f mice were crossed with the transgenic mice carrying one allele of Cre recombinase driven by the rat nestin promoter and enhancer (Tg(Nes-cre)1Kln, JAX stock #003771) (Giusti et al., 2014; Tronche et al., 1999).

Whole-body knockout mice were generated by CRISPR–Cas9 gene editing technology. Briefly, a pair of guide RNAs (sgRNA; sequences listed in the Key Resource Table) targeting genomic sequence flanking exon 8 of the Stag2 locus were tested for cutting efficiency in cell culture, transcribed in vitro, purified, checked for integrity, and microinjected into B6C3F1 mouse zygotes along with the Cas9 mRNA (5-methylcytidine, pseudouridine, TriLink). 20 ng/μl of Cas9 mRNA and 10 or 20 μg/μl each of sgRNA were used. The injected embryos were transferred to the surrogate mother on the same day. Mosaic F₀ founders carrying the Stag2^null allele were identified by PCR genotyping. The reduction of the STAG2 protein was confirmed by western blotting in multiple tissues. The F₀ founders were crossed with WT C57BL/6J mice to generate the Stag2^+/− F₁. The mutations in F₁ were identified by Sanger Sequencing. Two mouse lines carrying genomic deletions between the Cas9 cleavage sites were chosen for the generation of Stag2^null mouse embryos.

All mice were housed in the antigen-free barrier facility with 12 hr light/dark cycles (6 AM on and 6 PM off). Mice were fed a standard rodent chow (2016 Teklad Global 16% protein rodent diet, Harlan Laboratories).

The C-terminal fragment of human STAG2 protein was expressed and purified from Escherichia coli and used as the antigen to generate rabbit polyclonal antibodies against STAG2 at YenZym. Other antibodies were purchased from the following commercial sources: anti-SMC1 (Bethyl Laboratories, A300-055A), anti-SMC3 (Bethyl Laboratories, A300-060A), anti-RAD21 (Bethyl Laboratories, A300-080A), anti-SA1 (Bethyl Laboratories, A302-579A), anti-SA2 (Bethyl Laboratories, A302-581A), anti-α-TUBULIN (Sigma-Aldrich, DM1A), anti-MBP (Abcam, ab7349), anti-PLP1 (Abcam, ab28486), and anti-H3K27ac (Abcam, ab4729).

For immunoblotting, brain hemispheres were homogenized in a Precellys tissue homogenizer (Bertin Instruments) with the lysis buffer [20 mM Tris–HCl (pH 7.7), 137 mM NaCl, 2 mM Ethylenediamine tetraacetic acid (EDTA), 10% (vol/vol) glycerol, 1% (vol/vol) Triton X-100, 0.5 mM dithiothreitol, 1 mM Phenylmethylsulfonyl fluoride (PMSF), 1 mM Na₃VO₄, 10 mM β-glycerophosphate, 5 mM NaF, and protease inhibitors (Roche)]. Homogenized brain tissues were lysed on ice for 1 hr. The lysate was then subjected to centrifugation at 20,817 × g at 4°C for 20 min and further cleared by filtering through a 0.45 μm filter. The cleared lysate was analyzed by sodium dodecyl sulfate–polyacrylamide gel electrophoresis and transferred to membranes, which was then incubated with the appropriate primary and secondary antibodies. The blots were imaged with the Odyssey Infrared Imaging System (LI-COR).

Mouse brains were fixed in 10% neutral buffered formalin solution for 48 hr followed by paraffin embedding and coronal or sagittal sectioning at 5 μm. H&E staining and LFB staining were performed by the Molecular Pathology Core at UT Southwestern Medical Center. Investigators were blinded to the genotype. Images were acquired with the DM2000 microscope (Leica) at ×1.25 resolution.

Immunohistochemistry was performed as previously described (Choi et al., 2016). Briefly, deparaffinized sections were fixed with 4% paraformaldehyde, subjected to antigen retrieval by boiling with 10 mM sodium citrate (pH 6.0), and then incubated with the indicated antibodies at 1:100 dilution. The slides were scanned with an Axioscan.Z1 microscope (Zeiss) at ×40 resolution at the Whole Brain Microscopy Facility at UT Southwestern Medical Center. Images were processed and quantified with ImageJ. For the myelinated fiber length measurement and coherency analysis, coronal sections of the brain cortex stained with the anti-MBP antibody were processed as previously described (van Tilborg et al., 2017). The myelinated axial thinning and fiber length measurement were performed by the plugin DiameterJ. The coherency analysis of myelinated axons was performed with the plugin OrientationJ.

The immunomagnetic isolation of oligodendrocytes from Stag2^f/y and Stag2^f/y;Nes^Cre P12-P14 pups was conducted using anti-O4 microbeads (Miltenyi Biotec) according to a published protocol (Flores-Obando et al., 2018). Briefly, brain cortices were dissected, pooled, minced into 1 mm³ cubes, and incubated with the Papain dissociation solution (neurobasal medium with 1% penicillin–streptomycin, 1% L-glutamine, 2% B27 supplement, 20–30 U/ml of Papain, and 2500 U DNase I) in a 37°C, 5% CO₂ incubator for more than 20 min. The enzymatic digestion was inactivated by the addition of 1 ml of fetal bovine serum (FBS). Gentle trituration by 10 ml, 5 ml and 1-ml pipettes was applied to break up cell clumps. Cells were collected by centrifugation (200 × g, 10 min), washed first with serum-containing medium (neurobasal medium with 1% penicillin–streptomycin, 1% L-glutamine, 2% B27 supplement, and 10% FBS), and then with the magnetic cell sorting (MCS) buffer (phosphate-buffered saline, pH 7.2, with 0.5% bovine serum albumin [BSA], 0.5 mM EDTA, 5 μg/ml insulin and 1 g/l glucose). The cell pellet was resuspended in the MCS buffer and incubated with anti-O4 microbeads at 10 μl/10⁷ cells at 4°C for 15 min followed by 1× wash with the MCS buffer. The O4⁺ immature oligodendrocytes were sorted through the magnetic LS columns according to the manufacturer’s instruction. Freshly prepared oligodendrocytes were directly used or fixed for subsequent analysis.

Mice were singly housed in shoebox-sized cages with a 5-day acclimation period followed with a 4-day recording period. Recorded parameters were analyzed by the TSE system and normalized to body weight. The experiments were conducted by the core personnel under the core protocol at the Metabolic Phenotyping Core at UT Southwestern Medical Center. Investigators were blinded to the genotype.

Blood samples were collected from facial bleeding without fasting. Plasma growth hormone levels were determined with the rat/mouse growth hormone ELISA kit (EMD Milipore, EZRMGH-45K). Plasma IGF-1 concentrations were measured using the mouse/rat IGF1 Quantikine ELISA kit (R&D Systems).

Brain hemispheres were preweighed and snap-frozen for extraction and measurement by mass spectrometry. The sterol extraction and quantitative analysis were conducted at the Center of Human Nutrition at UT Southwestern Medical Center as described previously (McDonald et al., 2012).

Stag2^f/y and Stag2^f/y;Nes^Cre P18 pups were transcardially perfused with 4% paraformaldehyde, 1% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4). Tissues were dissected and fixed with 2.5% (vol/vol) glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4) for at least 2 hr. After three rinses with the 0.1 M sodium cacodylate buffer, optic nerve samples were embedded in 3% agarose and sliced into small blocks. All samples were again rinsed with the 0.1 M sodium cacodylate buffer three times and postfixed with 1% osmium tetroxide and 0.8% potassium ferricyanide in the 0.1 M sodium cacodylate buffer for 3 hr at room temperature. Blocks were rinsed with water and en bloc stained with 4% uranyl acetate in 50% ethanol for 2 hr. Samples were dehydrated with increasing concentrations of ethanol, transitioned into propylene oxide, infiltrated with Embed-812 resin, and polymerized in a 60°C oven overnight. Blocks were sectioned with a diamond knife (Diatome) on a Leica Ultracut 7 ultramicrotome (Leica Microsystems) and collected onto copper grids, poststained with 2% aqueous uranyl acetate and lead citrate. Images were acquired on a Tecnai G2 Spirit transmission electron microscope (Thermo Fisher) equipped with a LaB6 source using a voltage of 120 kV. Tissue processing, sectioning, and staining were completed by the Electron Microscopy Core at UT Southwestern Medical Center.

Total RNA was extracted from brain hemispheres or isolated oligodendrocytes with Trizol. RNA integrity was determined by the Agilent BioAnalyzer 2100. TruSeq Stranded mRNA library prep kit (Illumina) was used to generate the mRNA libraries. The libraries were analyzed by the Bioanalyzer and multiplexed and sequenced using the NextSeq 500 high output kit (400 M reads) for the brain libraries or NextSeq 500 mid output kit (130 M reads) for the isolated oligodendrocytes libraries at the Next Generation Sequencing Core at UT Southwestern Medical Center.

Raw data from the sequencer were demultiplexed and converted to fastq files using bcl2fastq (v2.17, Illumina). The fastq files were checked for quality using fastqc (v0.11.2) Andrews, 2010 and fastq_screen (v0.4.4) (Wingett, 2011). Fastq files were mapped to the mm10 mouse reference genome (from iGenomes) using STAR (Dobin et al., 2013). Read counts were then generated using featureCounts (Liao et al., 2014). TMM normalization and differential expression analysis were performed using edgeR (Robinson et al., 2010). Pathway analysis was performed with the IPA software. Genes with more than 1.5-fold change and false-discovery rate FDR <0.01 were included in the brain RNA-seq pathway analysis. Genes with more than twofold change and FDR <0.05 were used for the pathway analysis of the RNA-seq data from oligodendrocytes.

Single-stranded cDNAs were converted from 2 µg of total RNA extracted from mouse brains with the high-capacity cDNA reverse transcription kit (Applied Biosystems). Quantitative PCR was conducted to determine transcript levels using gene-specific TaqMan probes (Applied Biosystems).

Single-cell suspension was prepared from forebrains of P13 Stag2^f/y or Stag2^f/y;Nes^Cre pups using the Papain Dissociation System (Worthington Biochemical, LK003150) according to the manufacturer’s instructions. Biological duplicates were made for each genotype. Single-cell RNA-seq libraries were generated with the Chromium Single Cell 3′ GEM, Library & Gel Bead Kit v3 (10× Genomics) according to the manufacturer’s guidelines. Cell density and viability were checked by the TC-20 Cell Counter (Bio-Rad). Cells were then loaded onto Chip B in the Chromium Controller (10× Genomics). 10,000 cells were targeted for each sample. The libraries were analyzed by the Bioanalyzer (Agilent) and pair-end sequenced in two flowcells of the NextSeq 500 High Output (400 M) run. The sequencing was performed at the Next Generation Sequencing Core at UT Southwestern Medical Center.

Data demultiplexing and alignment were performed using the Cell Ranger pipeline (https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/using/ mkfastq) (10× Genomics). The raw features, barcodes, and matrixes were used as input for further analysis using the R package Seurat3 (Butler et al., 2018; Stuart et al., 2019) (https://satijalab.org/seurat/). Cells were filtered by the following criteria: nFeature_RNA (200–9500) and percent.mt <10. After filtering, a total of 5834 cells in Stag2^f/y#1, 4699 cells in Stag2^f/y#2, 9050 cells in Stag2^f/y;Nes^Cre#1, and 3073 cells in Stag2^f/y;Nes^Cre#2 were used for downstream analysis. 2000 variable features were found from each normalized dataset. All datasets were then integrated using identified anchors (dims = 1:30). Standard scaling and principal component analysis, clustering (resolution = 0.5), and tSNE reduction (dims = 1:30) were performed on the integrated dataset. Cluster biomarkers were identified, and top features were examined. Clusters were then manually assigned to distinct cell-type identities with knowledge from previous studies (Cahoy et al., 2008; Dulken et al., 2019; Marques et al., 2018; Marques et al., 2016; Marton et al., 2019; Saunders et al., 2018; Zeisel et al., 2018; Zywitza et al., 2018) (http://www.brainrnaseq.org/) (http://dropviz.org/). Clusters with the same cell-type identities were merged. Five clusters of oligodendrocyte lineage [cycling oligodendrocyte progenitors (OPCcycs), oligodendrocyte progenitors (OPCs), newly formed oligodendrocytes (NFOLs), myelin-forming oligodendrocytes (mFOLs), and fully matured oligodendrocytes (MFOLs)] were identified and selected for indicated gene expression comparison and plotting using Vlnplot or FeaturePlot functions. The trajectory analysis was performed using Monocle3 (Cao et al., 2019) in the oligodendrocyte cell population. Gene density plot over pseudotime was generated as previously described (Luecken and Theis, 2019).

Chromatin immunoprecipitation (ChIP) was performed as previously described (Liu et al., 2017). Briefly, isolated oligodendrocytes were fixed with 1% formaldehyde and fragmented with a sonicator (Branson 450). The fragmented chromatin was incubated with antibodies overnight at 4°C. Dynabeads Protein A (Thermo Fisher Scientific) was used for the immunoprecipitation. Libraries were generated by the Next Gen DNA Library Kit (Active Motif) with the Next Gen Indexing Kit (Active Motif) for STAG2 ChIP-seq or the KAPA HyperPrep Kits (KAPA Systems) for histone ChIP-seq. The libraries were analyzed by the Bioanalyzer and pool sequenced with the NextSeq 500 mid output (130 M) kit. After mapping reads to the mouse genome (mm10) by bowtie2 (v2.2.3) (Langmead and Salzberg, 2012) with the parameter ‘–sensitive’, we performed filtering by removing alignments with mapping quality less than 10 and then removing duplicate reads identified by Picard MarkDuplicates (v1.127). For STAG2 ChIP-seq, Picard MarkDuplicate was used to remove duplicates together with options to use molecular identifiers (MIDs) information in the reads. Enriched regions (peaks) were identified using MACS2 (v2.0.10) (Zhang et al., 2008), with a q-value cutoff of 0.05 for peaks. Peak regions were annotated by HOMER (Ross-Innes et al., 2012).

Hi-C was performed at the Genome Technology Center at NYU Langone Health from 3.5 to 4.0 μg of DNA isolated from cells cross-linked with 2% formaldehyde at room temperature for 10 min. Experiments were performed in duplicates following the instructions from the Arima Hi-C kit (Arima Genomics, San Diego, CA). Subsequently, Illumina-compatible sequencing libraries were prepared by using a modified version of the KAPA HyperPrep library kit (KAPA BioSystems, Willmington, MA). Quality check steps were performed to assess the fraction of proximally ligated DNA labeled with biotin, and the optimal number of PCR reactions needed to make libraries. The libraries were loaded into an Illumina flowcell (Illumina, San Diego, CA) on a NovaSeq 6000 instrument for paired-end 50 reads.

Hi-C analysis was performed using the HiC-Bench pipeline (Lazaris et al., 2017; Tsirigos et al., 2012) (https://github.com/NYU-BFX/hic-bench) and HiC-Pro v3.1.0 (Servant et al., 2015). The read pairs were aligned and filtered with the following parameters: Genome-build=mm10; –very-sensitive-local –local; mapq = 20; –min-dist 25000 –max-offset 500. The Juicer ‘pre’ tool (Durand et al., 2016) (RRID: SCR_017226, v1.11.09; https://github.com/aidenlab/juicer) was used to generate the.hic file with default parameters. Sample duplicates were combined. The compartment analysis was done using the HOMER tool (Heinz et al., 2010) (http://homer.ucsd.edu/homer/index.html) with 100 kb bins. H3K27ac ChIP-seq data were used to assign A/B compartments. Eigenvector-1 bins were considered shifted (AB and BA) when the bin sign changed and the delta value was greater than 1.5. TADs and boundaries were identified at 40 kb resolution with the HiCRatio method with the follow parameters: –min-lambda=0.0 –max-lambda=1.0 –n-lambda=6 –gamma = 0 –distance = 500 kb –fdr = 0.1. TADs were also identified using the Juicer tools (v1.22.01) arrowhead at 10 and 25 kb resolution. Aggregate TAD analysis was performed on TAD boundaries by coolpup.py (Flyamer et al., 2020) or GENOVA (van der Weide et al., 2021). The.hic files were converted to.cool format for visualization and plotting with pyGenomeTracks (Lopez-Delisle et al., 2021) at 5 kb resolution.

The loops were classified into group-specific loops and common loops by using the significance cutoffs provided by Fit-HiC (Ay et al., 2014). A q-value cutoff of 0.01 was used to identify significant loops in both groups. A loop is considered ‘group-specific’ if it is only present in one group with a q value <0.01 and not present in the other group with cutoff of q val <0.1. Loop anchors were annotated with the gene promoter information (promoter defined as ±2 kb from the TSS). The genes were classified into ‘down’ and ‘up’ regulated genes using an FDR cutoff of 0.05, logFC cutoff of ±0.58 and logCPM >0. ‘stable’ or less changed genes are defined as logFC <0.38, and logCPM >0. Random 1000 genes were chosen for analysis and plotting. The active genes (logCPM >0) were also grouped in ‘high’, ‘mid’, and ‘low’ expression groups by separating the genes in three quantiles according to the logCPM values. For the loop enrichment scores, normalized contact scores were computed using Fit-HiC at 10 kb resolution and bias corrected. Pile-up analysis was performed with coolpup.py (Flyamer et al., 2020) with the KR method to balance the weight and random shift controls for distance normalization at 5 kb or using GENOVA.