Animals

Animal usage and manipulations were performed in accordance with the University of Miami Institutional Animal Care and Use Committee and Public Health Service animal care and use guidelines.

Chromatin immunoprecipitation (ChIP)

6-week-old male C57BL/6 mice were terminally anesthetized (ketamine/xylazine, 100/15mg, i.p.) and perfused with only cold PBS just prior to removal of the spinal cord. A total of 5 cords were aggregated for each immunoprecipitation (IP) reaction. Tissue preparation and ChIP protocol closely adhered to steps outlined in the MAGnify^™ Chromatin Immunoprecipitation System manual (Invitrogen, 49–2024) according to manufacturer’s instructions with only a few modifications (see below). Briefly, dissected spinal cords were immediately weighed, cleaned, and minced with a razor blade in ice cold PBS. Minced tissue was then collectively homogenized by passing repeatedly through 18G and 21G needles before crosslinking chromatin with formaldehyde at a final concentration of 1% at room temperature for 10 minutes. Glycine was added to a final concentration of 0.125 M to quench formaldehyde crosslinking, and cells were then pelleted and washed with ice-cold PBS. Cell pellets were resuspended in Lysis Buffer plus Protease Inhibitors at a concentration of 1x10^6 cells / 50 μl. Nuclear lysates were then sonicated using a Branson^® 150 Sonifier (power setting 4, 100% duty cycle for 10 × 30-s on 90-s off intervals), yielding chromatin fragments of 50–300 bp. Dynabeads^® Protein A/G (Invitrogen, 10010D) were coupled to either 5 μg of Polyclonal Rabbit anti-Olig2 antibody (Millipore, A9610) or Rabbit IgG (Invitrogen, 100005291) by mixing end-over-end for 4 hours at 4°C prior to introduction to lysates. Antibody-conjugated beads were then combined with 500 μl of lysate (~1x10^7 cells) plus 500 μl of Dilution Buffer in 1.5 mL LoBind Eppendorf tubes (Sigma-Aldrich, Z666548-250EA) and mixed by end-over-end rotation overnight at 4°C. Approximately 1% of the total sheared chromatin was set aside and served as Input Control. Beads containing Protein–DNA complexes were subsequently washed with IP Buffer, formaldehyde crosslinking reversed by heat treatment, and the DNA eluted using a 12 Tube Magnetic Separation Rack (NEB, S1509S). Here, we note that the Magnify^™ protocol is designed for use with 0.2 uL tubes, so in all subsequent steps involving end-over-end rotation we instead placed each 1.5 mL LoBind tube on its side and applied gentle agitation using a platform rocker. This small change to the protocol ensured that the beads remained submerged in solution throughout. For all steps involving bead collection via magnetic rack, 1.5 mL LoBind tubes were placed in the rack in such a way as to concentrate bead collection toward the bottom of the tube, again to keep them submerged in solution. Specifically, foam padding was added at the base of the rack to keep the tubes slightly elevated. Input Control samples were processed at the reverse crosslinking step in parallel with the ChIP’d samples. IgG IP’d DNA was analyzed by qPCR (see below) but was not included in NGS analysis.

ChIP-Seq library preparation

DNA libraries were prepared from total sheared chromatin (Input Control Library) and ChIP-enriched fragments (ChIP’d Library) using Illumina TruSeq ChIP Library Preparation Kit (Illumina, IP-202-1024). For this experiment, we used a modified TruSeq^® ChIP Sample Preparation Guide protocol (Illumina, PN 15023092 Rev.B; Oct. 2013), omitting only the gel purification step to maximize total yield of adapter-ligated library fragments. Input Control and ChIP’d libraries were prepared using recommended inputs of 100 pg/μl (50 μl), for a total of 5 ng per library prep. Libraries were indexed using TruSeq Adapters AR008 (Input Control Library; ACTTGA/A) and AR001 (ChIP’d Library; ATCACG/A). All QC steps (Input DNA quantification, size distributions, and library traces) were evaluated using a Fragment Analyzer Automated CE system with High Sensitivity NGS Fragment Analysis Kit 1bp-6000bp (Advanced Analytical, DNF-474). Libraries were sequenced at 2 x 50 bp read lengths using the Illumina MiniSeq Sequencing System and MiniSeq High-Output Reagent Kits (Illumina, FC-420-1001).

ChIP-Seq peak calling and data analysis

Sequence analysis was performed using BaseSpace Labs ChIP-Seq v.1.0.2 application (http://www.basespace.com) This application utilizes the algorithms BWA [33] to align the reads to the mouse genome, MACS2 (https://github.com/taoliu/MACS/; [34]) to identify enriched regions pulled down by ChIP, and HOMER, (http://homer.salk.edu/homer/ngs/peakMotifs.html; [35]) to identify motifs within and annotate genes proximal to these enriched regions. Custom input parameters were entered for reference genome (UCSC mouse genome; mm10) and fragment size (148 bp; Average ChIP fragment size as determined using BaseSpace BWA Aligner v1.1.4 and BAMStats 1.25; https://sourceforge.net/projects/bamstats/). Samples for both ChIP’d, and Input Control libraries were used as inputs for MACS2 to improve spatial resolution (specificity) of ChIP-region binding sites.

ChIP-Seq quality checking

Quality checking was performed following the guidelines of the ENCODE3 Transcription Factor ChIP-Seq processing pipeline v1 (https://docs.google.com/document/d/1lG_Rd7fnYgRpSIqrIfuVlAz2dW1VaSQThzk836Db99c/). The non-redundant fraction of the aligned reads was calculated as the fraction of distinct reads (reads that align to only one genomic loci and do not have any optical duplicates) divided by the total number of reads that aligned to exactly one location. The normalized strand cross-correlation coefficient and relative strand cross-correlation coefficient were calculated using phantompeakqualtools (https://code.google.com/p/phantompeakqualtools/). These cross-correlation coefficients are relative measures of the enrichment of the ChIP data for peak-like distributions of reads (see [36] for a more complete explanation).

ChIP-qPCR and statistical analysis

ChIP was performed using a protocol similar to the one described above for ChIP-Seq with one minor change to the fragmentation protocol- the Branson 150 Sonifier settings were changed to power setting 4, 100% duty cycle for 5 × 30-s on 90-s off intervals, yielding chromatin fragments of 100–800 bp. Primer pairs were designed using primer-BLAST (https://www.ncbi.nlm.nih.gov/tools/primer-blast/; [37]) to flank each targeted peak site. Each primer pair was subsequently cross-checked for spurious hybridization potential using UCSC In-Silico PCR web server (https://genome.ucsc.edu/cgi-bin/hgPcr?db=mm10) against both the human (hg38) and mouse (mm10) reference genomes to minimize spurious hybridization. Primers used in ChIP-qPCR are provided in Supplementary Materials. qPCR data were generated using an Applied Biosystems 7300 thermocycler and Rotor-Gene SYBR Green PCR Kit (Qiagen, 204074) according to manufacturer’s instructions. At least three independent ChIPs were generated and run in separate qPCR runs, the results are presented as means and presented in a box and whisker format using StatPlus (https://www.analystsoft.com). For between groups comparison a Student’s two-tailed t-test was used. Statistical significance was established when p < 0.05.

Genome-wide mapping of direct targets of Olig2 in the spinal cords of adult mice

Expression of Olig2 in CNS oligodendroglia cells persists well into adulthood; however, it is unclear what role Olig2 transcriptional regulation plays in neuroglia, particularly OPCs and in mature OLs within the adult spinal cord, both under physiological and pathological conditions. Because of this lack of clarity, we sought to identify the genome-wide transcriptional targets of Olig2 in the adult spinal cord in vivo by performing ChIP-Seq analysis on whole spinal cord tissue harvested from healthy adult mice.

Many factors contribute to variability among ChIP-Seq datasets, so to ensure reliability of our data, and in an effort to maximize site discovery and reproducibility, we endeavored to adhere to ENCODE working standards and protocols for conducting and reporting ChIP-Seq data for point-source factors such as a transcription factor [36]. These measures included using a validated ChIP-grade antibody [38–40] and aggregating data from two independent biological experiments, resulting in approximately 55 million uniquely mapped reads, and measurement of the suggested quality control metrics. PCR duplicate analysis revealed that almost all (99.3%) of these uniquely mapped reads were non-redundant. This indicates that the sequenced DNA still retained a high degree of library complexity even after the requisite PCR steps during library preparation and, therefore, the PCR amplification likely had minimal negative impact on the results obtained. Cross-correlation analysis of the aligned ChIP-Seq data revealed evidence of successful ChIP enrichment of the type expected for a point-source transcription factor with a normalized strand cross-correlation coefficient of 1.21 and a relative strand cross-correlation coefficient of 2.98 (see Materials and methods). Each of these cross-correlation coefficients are well above the minimum levels recommended by the ENCODE consortium. Together, these quality-checking metrics support the high quality of the ChIP-Seq data presented here.

The Olig2 ChIP-Seq identified 4,214 unique regions of genomic DNA bound by Olig2 in in vivo (S1 Table; all sequences have been submitted to GEO database: GEO Series accession number GSE103324). To evaluate the global distribution of Olig2-occupied loci, we plotted the number of Olig2 sites against their distance to the nearest transcription start site (TSS). We detected a strong enrichment for Olig2 occupancy within 1 kb of the nearest annotated TSS (Fig 1A), an indication that Olig2 shows a strong occupancy bias toward target promoters. Indeed, further analysis identified 37% of peaks localized to known (annotated) promoters (Fig 1B). Comparative genomic analysis using the VistaPoint tool (http://pipeline.lbl.gov/; [41]) revealed that the majority of Olig2-occupied peaks within intergenic regions or within introns were highly conserved (S1 Fig), which suggests that they may have some functional importance. Binding regions for the transcription factor Nkx2.2 (S1A Fig) and presumptive transcription factor Zfp536 (S1B Fig) were extremely well-conserved among vertebrates, from Homo sapiens to Danio rerio. Whereas peaks of other targets like transcription factors Nkx6.2 (S1C Fig) and Klf9 (S1D Fig) were also highly conserved from Homo sapiens to Gallus gallus but were not conserved in Danio rerio.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Identification of direct targets of Olig2 in the adult spinal cord in vivo.

(A) Heatmap and line plot representation of ChIP-Seq signal density for Olig2 ChIP and Input Control centered on predicted TSS. (B) Pie chart showing proportions of genomic landmarks corresponding to Olig2-bound targets. (C) Integrative Genomics Viewer (IGV) visualization of Olig2 occupancy at selected target sites. (D) Pie chart showing proportions of targets with protein-coding or non-protein-coding designation. (E) De novo motif analysis of Olig2-binding regions identified a putative E-box motif (CANNTG) among ~60% of all Olig2-bound target regions. This motif configuration is consistent with previously published motif analyses for Olig2-occupancy via ChIP-Seq.

https://doi.org/10.1371/journal.pone.0186091.g001

To investigate whether any of the Olig2-occupied sites overlapped with genomic regions associated with known enhancer activity by comparing our ChIP-Seq peaks with the 900 enhancers that have consistent tissue-specific enhancer activity and were validated using transgenic assays [42]. Indeed, we found that one of the targeted peaks identified in the ChIP-Seq associated with the transcription factor Nkx2.2, a known target of Olig2 transcriptional regulation during development, localized to a region with demonstrated enhancer activity in vivo [42]. These findings suggest Olig2 preferentially binds to genomic regions associated with putative, and in some cases documented promoter and enhancer activity.

In our analysis, the closest annotated gene to each Olig2-binding site was considered a presumed target. Several of the targets identified in our ChIP-Seq have been identified elsewhere as direct targets of Olig2, while many genes on our list are known to be expressed in oligodendroglia, including during myelination, but were not previously known as direct targets of Olig2 (Fig 1C). For example, we identified the transcription factors Nkx2.2, Nkx6.2, and both Olig1 and Olig2, known targets of Olig2 transcriptional regulation and necessary for both OL fate specification and differentiation, as targets of Olig2 binding in the adult spinal cord in vivo (Fig 1C). We also identified Gpr17 and Zeb2/Sip1, two genes that had been reported previously as direct binding targets of Olig2 in two separate in vitro ChIP-Seq studies [43,44], as targets of Olig2 in vivo (Fig 1C). In addition to known targets, we also identified as targets of Olig2 OL-specific genes like Cnp, Mbp, Cldn11, and Sirt2, as well as myelin-associated genes like Mobp, Mog, and Mag, which were previously shown to be up-regulated 119-, 98- and 38-fold, respectively, upon OL differentiation in vitro (Fig 1C; [45–47]). Taken together, the substantial overlap between our in vivo target list and previously published lists of Olig2 targets in vitro, as well as the numerous targets we identified with OL lineage-relevant functions, strongly supports our list of in vivo targets of Olig2 generated by ChIP-Seq.

Functional genomics of Olig2-occupied target regions

The majority of Olig2-bound sites were associated with a protein-coding gene, comprising 89%, or 3736 of the 4,214 Olig2-bound targets, while non-protein-coding RNAs made up 11% of the total target list, or 448 targets, and pseudogenes, snoRNAs, rRNAs, and snRNAs each comprised less than 1% of the total list of bound targets (Fig 1D). Of the 448 non-protein-coding targets, 137 were annotated as microRNAs (abbreviated miRNAs; Fig 1D). Notable among the list of targeted miRNAs was Mir219a-2 because it is abundant in the developing CNS, its induction coincides with OL differentiation, and it plays a critical role in facilitating the differentiation of proliferating OPCs into myelinating OLs [24–27]. Based on the results of our ChIP-Seq, we found that Olig2 occupies a region upstream of Mir219a-2 (or Mir219-2) locus and might therefore regulate Mir219a-2 in OLs in the adult spinal cord, revealing a potentially important mechanism whereby Olig2 regulates OL differentiation and myelination in the adult in vivo (Fig 1C). Unfortunately, not much known about the remaining 310 non-protein-coding targets identified in the ChIP-Seq due in part to the lack of annotation for long non-coding RNAs (lncRNAs). However, there were a few notable exceptions. For example, our data showed multiple Olig2 occupancy sites just upstream of two lncRNAs, Neat1 and Malat1/Neat2. A role for Malat1 in OL development has been suggested, however, its importance is unclear due to the lack of any clear phenotype in Malat1 knockout mice [48]. Our data suggests that Olig2 likely plays a regulatory role upstream of Malat1, Mir219a-2, and likely other non-protein-coding functional RNAs as a mechanism to control OL differentiation in the adult in vivo.

Comprehensive de novo mapping of Olig2 targets in vivo

To investigate whether certain DNA motifs were enriched in the Olig2-binding sites, we applied a motif-discovery algorithm, HOMER [35]. Analysis revealed that the motif CA(G/T)(C/A)TG(T/G)N was statistically overrepresented (logPval = -942) among Olig2-binding sites, appearing in ~60% of them (Fig 1E). Importantly, this resolved motif contains a canonical E-box (CANNTG), the postulated target consensus of Olig2 and the bHLH family of transcription factors. We compared our motif to those previously published for Olig2, and found that our motif is nearly identical to those found in previously reported Olig2-bound targets [38,39]. In [39] for example, the authors, used the same antibody against Olig2 as the one used in this study (see Materials and methods) and reported an E box-containing motif configuration (CAGCTGC) for Olig2-occupied targets that is strikingly similar to the one identified in our de novo analysis (Fig 1E). In [38], the authors used a different Olig2 antibody to identify targets of Olig2 in mouse embryonic stem cells in vitro, yet reported a motif (CCAGCTGCT) that is very similar to ours. This notable consistency across ChIP-Seq experiments suggests that there is strong motif conservation for Olig2 occupancy across cells and developmental stages.

Olig2 targets myelination-related genes in the adult spinal cord

To identify the major biological functions represented by the gene targets in our dataset, we looked for statistically overrepresented gene ontology (GO) terms related to biological processes using PANTHER (GeneOntology.org; [49]). The ChIP-Seq GO profile revealed a statistical overrepresentation of terms associated with oligodendroglia physiology (Fig 2 and S2 Table). The top GO terms included myelination, ensheathment, and cell cycle regulation, processes consistent with OL development (Fig 2A and S2 Table). Targets we identified with established roles in myelination and ensheathment included Mog, Mag, Mbp, Mobp, Mal, Cnp, Fth1, Opalin/Tmem10, Sirt2, and Cldn11, among many others [45–47]. We also identified Cdk4, Mcm4 and Mcm5, regulators of cell cycle/division, as targets of Olig2, suggesting that in the adult spinal cord, Olig2 continues to influence OPC proliferation via the cell cycle. In addition, we examined whether specific protein families or classes were overrepresented in our dataset. We identified an enrichment of targets associated with microtubule and actin cytoskeletal dynamics, suggesting a putative role for Olig2 in structural maintenance and remodeling of the myelin sheath (Fig 2B and S2 Table). We identified Sept9, a member of filament-forming septins that are abundant in myelin [50], and Cdc42, a member of the Rho family of GTPases required for OL maturation and formation of CNS myelin, as in vivo targets of Olig2, consistent with previous findings from a separate Olig2 ChIP-Seq study conducted in OLs in vitro [39,51]. Finally, we looked for significantly overrepresented pathways within our dataset and found that the Pdgfa signaling pathway was among the top overrepresented pathways associated with our list of ChIP-Seq targets (Fig 2C and S2 Table). Pdgf signaling pathway plays an important role in the normal development of OL lineage, as loss of Pdgfa in mice results in profound hypomyelination and reduced numbers of OPCs and OLs [52]. This further suggests that in the adult as in development, Olig2 continues to influence OPC proliferation. Taken together, the results of the GO analyses suggest that Olig2 continues to perform similar roles in adulthood as it does during late embryonic and postnatal development of the CNS, contributing to OL lineage progression and myelin production and ensheathment of axons in the adult spinal cord.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Functional GO terms overrepresented in the Olig2 ChIP-Seq dataset.

Shown are the highest ranking terms for (A) Biological Processes, (B) Protein Class, and (C) Pathways according to PANTHER (geneontology.org) and ranked according enrichment score (FE; red bars) and statistical significance (-log2p-val; black line) as indicated.

https://doi.org/10.1371/journal.pone.0186091.g002

ChIP-qPCR validation of ChIP-Seq targets

Finally, we sought to validate the ChIP-Seq results by chromatin immunoprecipitation followed by quantitative PCR (ChIP-qPCR). We included in these experiments an IgG “mock” IP condition, which served as an important negative control. For the validation, we selected 13 targets from the complete dataset, taking into account whether they had previously been identified as a direct target of Olig2 elsewhere, and whether they had a documented role in the physiology of the OL lineage. In addition, we utilized high, low, and mid-range Peak Score (PS) values, as determined by the ChIP-Seq analysis (see Materials and methods; S1 Table), when making our selection. Primers were designed to recognize the genomic region corresponding to the peak summit for each target gene, and we referred to these primer pairs as “ON-target” (S3 Table). Among the potential targets we tested by qPCR, all 13-showed enrichment as percentage of input control (Fig 3). Clip3, which had a relatively high peak score (PS = 88.28) also showed comparatively high enrichment over input (Fig 3A). Curiously, Nrxn2, which had the highest relative peak score in our subset list (PS = 94.15), and Rtkn, which also had a high peak score (PS = 74.38), showed only modest enrichment over input (Fig 3A). Previously published Olig2 targets Olig1 (PS = 62.52) and Nkx6.2 (PS = 53.86) also showed enrichment over input (Fig 3A). The other targets we tested: Cldn11 (PS = 31.26), Mag (PS = 6.70), Mbp (PS = 17.40), Mog (PS = 16.50), and Mpzl1 (PS = 10.9), showed enrichment over input commensurate with their peak scores (Fig 3A). As an additional negative control, we designed primers recognizing genomic regions 10–30 kb away from the peak summit for Nkx2.2, Opalin, and Zcchc24, which we called the “OFF-target” set of primers (Fig 3B and S3 Table). In these regions, no Olig2 peaks were detected. In contrast to the qPCR results using the ON-target primers, qPCR using OFF-target primers showed no appreciable enrichment as a percentage of input for any of the targets tested, similar to the results of the IgG mock ChIP using ON-target primers (Fig 3B). Taken together, the ChIP-qPCR enrichment data coupled with the high confidence peak summit resolution of the ChIP-Seq, demonstrate that the Olig2-bound target sites identified in the ChIP-Seq are targets of Olig2 in the adult spinal cord.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Validation of ChIP-Seq targets by ChIP-qPCR.

(A) Box plot depicting enrichment over total genomic input (%Input) for both Olig2 ChIP (Olig2) and mock IgG (IgG) control conditions for 13 gene targets identified by ChIP-Seq. (B) IGV genome browser tracks showing the location of primers targeting either a specific ON-target peak summit (green bar; ON) or OFF-target region located 10–30 kb from the nearest peak (red bar; OFF) for Nkx2.2, Opalin, and Zcchc24. Results for all ChIP-qPCR data were generated from three independent experiments.

https://doi.org/10.1371/journal.pone.0186091.g003