GEO Accession viewer

NCBI > GEO > Accession Display

Not logged in | Login

GEO help: Mouse over screen elements for information.

Sample GSM5044303

Query DataSets for GSM5044303

Status

Public on Aug 11, 2022

Title

v3.6.6

Sample type

SRA

Source name

CJH01

Organism

Callithrix jacchus

Characteristics

il04_sex: F
il05_agedays: 2051
il06_tissue.1: WM
il06_tissue.2: Telencephalon
il06_tissue.3: pCC
il07_location: SS07.L
il08_condition: NaiveCtrl
il09_illumina: NovaS2
il10_chemistry: 10xv3
il11_batch: 6th
il12_lmindays: 0
il13_lmaxdays: 0
il14_dataset: this.manuscript

Treatment protocol

Growth protocol

Extracted molecule

total RNA

Extraction protocol

Tissue dissection: On the day of tissue harvest, marmosets were deeply anesthetized with 5% isoflurane until all visible signs of breathing were no longer detected. Animals were transcardially perfused with ice-cold artificial cerebrospinal fluid (aCSF) for 5 min with a pump. Brains were removed from the skull and submerged into ice-cold aCSF, and after removal of meninges within the solu-tion, were positioned in a custom-designed brain holder within 10 min post-perfusion. The brain was sectioned at 3 mm into 12–13 slabs in one step with a homemade blade-separator set in the solution. Each brain slab was transferred into a 6-well plate, submerged into RNAlater (RNAlater™ Stabilization Solution, AM7021, Invitrogen) with a homemade brain trap, and stored at 4°C overnight. The following day, brain slabs were positioned in 25 x 20 x 5 mm molds (Tis-sue-Tek® Cryomold®, 4557, Sakura Finetek) on ice to facilitate target sampling. Slabs were matched to MRI for each animal, and tissue annotation for gray (Liu et al., 2018a) and white matter (Liu et al., 2020) was informed by marmoset 3D MRI atlases V1 and V2. A cylinder of tissue 2 mm in diameter and 3 mm in height for each region (Figure 1B and S1) was collected with a tissue punch (EMS-core sampling tool, 69039-20, EMS). There were 5 white matter sam-ples from temporal and parietal lobes that did not exactly match in the two animals, however they were paired in lobes of the brain and showed no significant differences in later analysis (Figure 1B, SS05, SS06, and SS08). The cylinders were ejected into PCR tubes filled with 100 µL of RNAlater and stored at −80°C. Single-nucleus dissociation: Nuclei preparation was carried out as described (Matson et al., 2018), with minor modifi-cations. Briefly, on the day of dissociation, tissue samples were thawed on ice, removed from solution, dabbed with Kimwipes to remove residual RNAlater, and placed in a 1 mL douncer tube (Dounce Tissue Grinder, 357538, Wheaton). Each tissue was homogenized in 500 μL of lysis buffer containing 0.1% Triton-X100 in low sucrose buffer (0.32 M sucrose, 10 mM HEPES, 5 mM CaCl2, 3 mM MgAc, 0.1 mM EDTA, and 1 mM DTT in ddH2O, pH8) with loose pestle 25 times and tight pestle 10 times. The homogenate was filtered through a 40-μm mesh (Falcon® 40 µm Cell Strainer, 352340, Corning) to a 50-mL Falcon tube on ice. An additional 5 mL of low sucrose buffer was used to rinse the douncer tube and cell strainer. The filtered homogenate was further mixed with a handheld homogenizer (VWR® 200 Homogenizer) at a speed of ~1000 rpm to brake nuclei clumps for 5 sec. After homogenization, a serological pipet filled with 12 mL of high sucrose buffer (1 M sucrose, 10 mM HEPES, 3 mM MgAc, and 1 mM DTT in ddH2O, pH8) was placed underneath the lysate and disconnected from the pipettor, and the buffer was released from the serological pipette by gravity and set on ice. When most of high sucrose buffer was re-leased to form a density layer underneath the homogenate, the serological pipet was retrieved along the wall of the Falcon tube gently, without disturbing the low-high sucrose interface. The Falcon tube was capped and placed in a swing bucket to be centrifuged at 3,200 rcf for 30 min at 4°C. At the end of spin, the supernatant was decanted quickly without tabbing, and 1 mL of re-suspension buffer (0.02% BSA in 1X PBS, pH7.4) was added to the Falcon tube. Slow pipetting was employed to resuspend nuclei along the Falcon tube wall below the 5-mL mark to preserve nuclei integrity. Specifically, nuclei were rinsed off the wall in courses of 2 sec per trituration for 20 times total per tube. The Falcon tube was then capped and spun at 3,200 rcf for 10 min at 4°C. At the end of spin, the supernatant was removed by gently tabbing the tube until no visible liquid drop was left behind, and 200 μL of resuspension buffer was added to each sample to collect the nuclei. The nuclei suspension was filtered through a 35-μm mesh (Cell Strainer Snap Cap, 352235, Corning) twice and counted on a hemocytometer by trypan blue staining. During count-ing, the size and quantity of myelin and other debris were visually inspected under the scope, and the suspension was filtered 1–3 more times through the 35-μm mesh if necessary. Only round and dark-blue stained nuclei were considered of good quality and included in the final count.
10x Genomics Chromium Single Cell 3’ Library & Gel Bead Kit v2 and v3

Library strategy

RNA-Seq

Library source

transcriptomic

Library selection

cDNA

Instrument model

Illumina NovaSeq 6000

Description

nuclear RNA

Data processing

Alignment: The raw sequencing reads were aligned to a marmoset genome assembly, ASM275486v1 (GCA_002754865.1). To build a reference package suitable for analyzing both unspliced pre-mRNA and mature mRNA in the nuclei, as well as to include sequences of mito-chondrial genome, marmoset DNA sequence (FASTA) and annotation (GTF) files were acquired from the Ensembl release-95 and modified as follows. The complete mitochondrial sequence (NC_025586.1, GenBank) and its annotation (Wang et al., 2016) were manually added to the FASTA and GTF files. Next, a pre-mRNA GTF was made by replacing “transcript” with “exon” as the feature-type entry in the original GTF before making a reference package with CellRanger software (v3.0.2, 10x Genomics). This custom-built reference package was then used in CellRanger (version 3.1.0, 10x Genomics) to align sequencing reads for all samples. The option to estimate cell number automatically was used for most of the samples, unless otherwise speci-fied (see Table S1 for details). A filtered cell barcode-to-gene feature matrix was generated from the software and used for downstream analysis (Figure S2A).
Preprocessing & general quality control: The matrix was loaded to create an object in Seurat v3 (Butler et al., 2018; Stuart et al., 2019). Cells with <200 genes, >5000 genes, or >5% of counts mapped to mitochondrial genome were excluded. Genes observed in <5 cells were excluded. The filtered raw count matrix was then log normalized (ln(counts*100,000+1)) within each cell and scaled to account for differences in sequencing depth with Seurat. Next, DoubletFinder (McGinnis et al., 2019) was used to esti-mate and remove putative doublets to mitigate technical confounding artifacts in droplet-based sequencing data analysis. The top 3000 variable genes calculated by Seurat were used in linear dimension reduction (principal components analysis, PCA), and the top 30 PCA were used for clustering at low resolution (parameter = 0.4) to define crude cell types. These unsupervised clusters were used to provide a quick cluster annotation for homotypic doublet probability model-ing in DoubletFinder. Doublet rate was estimated by fitting a linear equation over a multiplet rate table provided by 10x Genomics. The rate = (0.0008*cell.number + 0.0527)/100) was used to calculate a Poisson distribution with and without homotypic doublet proportion to generate low confidence (DF.found.1) and high confidence (DF.found.2) doublet annotation. Unless otherwise specified, pN = 0.25, pK = 0.005 and automatic doublet removal based on DF.found.2 annotation were used, as the first line of screening. In parallel, SoupX (Young and Behjati, 2020) was used for ambient RNA background correction. Taken from the output of the 10x pipeline (raw_feature_bc_matrix), the ambient RNA from empty droplets that contained <10 unique mo-lecular identifiers (UMI) were profiled, and the “soup” contamination fraction was calculated for each cluster. Given that the nuclear transcriptome was profiled, genes that mapped to the mito-chondrial genome could be considered as ambient input. Therefore, the top mitochondrial genes (species with >1000 accumulated counts across profiled empty droplets) were used to estimate the global contamination fraction and adjust the raw count matrix. Next, the cell barcode that passed the DoubletFinder was used as an index to subset the SoupX-corrected matrix to gener-ate a new matrix as our downstream input (Figure S2A). For individual samples, a Seurat object was created, and the index labels (IL01_uniqueID, IL02_species, IL03_source, IL04_sex, IL05_ageDays, IL06_tissue.1, IL06_tissue.2, IL06_tissue.3, IL07_location, IL08_condition, IL09_illumina, IL10_chemistry, IL11_batch, IL12_LMinDays, IL13_LMaxDays, IL14_dataset, IL15_annotation) were added to the metadata as cell attributes. For each sample, 90% of nuclei were randomly selected, then all 42 samples were merged together for downstream comparison. The remaining 10% of the nuclei were set aside for classification assessment and validation.
Level 1 quality control & clustering: To divide nuclei into classes and facilitate artifact identification, nuclei were first classified using a set of parameters that does not highlight granular detail. In this round of clustering, only 50 PC for Harmony were computed to perform linear correction over IL01_uniqueID, as the el-bow plot from the preliminary analysis showing the standard deviation stopped visually decreas-ing after top 50 PC. The top 5 Harmony-corrected embeddings (H5) were used for Seurat to learn the UMAP and find cell classes at a low resolution (0.2). Canonical cell-type markers (PTPRC for immune cells, PDGFRA for OPC, MAG for oligodendrocytes, GFAP and SLC1A2 for astrocytes, LEPR and CEMIP for vasculature and meningeal cells, and CNTN5 and NRG1 for neurons) annotated 6 of the classes unambiguously. One cluster in the middle of the H5 UMAP had mixed expression of canonical markers, which suggested artifact. The “low quality” cell bar-codes that were found from the H50 condition (defined above) were overlaid on the H5 UMAP, which exclusively highlighted the putative artifact cluster. These nuclei were removed from fur-ther analysis, although the original UMAP embeddings were maintained for plotting purposes (Figure 1 and S3).
Level2 quality control & clustering: Nuclei that passed Level 1 QC were divided into 5 partitions (MIC, OPC, OLI, VAS/AST, NEU) based on the H5 UMAP result. The astrocytes and vasculature/meningeal cells were pooled into a single class prior to subclustering to facilitate artifact identification. Shared features in this class of cells were potentially explainable by their close association at CNS barriers (blood–CSF interface, blood–brain interface, and the CSF–brain interface). For each class, log-normalization and scaling were repeated from the divided raw count matrix, the top 3000 variable genes were used for 50 PC computation, Harmony correction over IL01_uniqueID, UMAP learn-ing, and clustering, as described above. The clustering resolution was iteratively increased from low to high (0–1.2), and clustering stability was tracked with clustree (v0.4.3) (Zappia and Oshlack, 2018). Aided by the branch visualization provided by the clustree, a tentative resolution that was relatively stable was selected, then differentially expressed gene (DEG) analysis on the clusters found with this parameter set was performed. The expression pattern of the top-expressed genes for each cluster within and across classes were checked, and artifacts clus-ters were manually imputed. Doublets tended to form small distinct clusters on UMAP that branched early in the clustering tree analysis with low splitting resolution, had mixed canonical marker-gene expression, and had similar expression patterns to cells in other partitions; thus, these doublets could be easily spotted and removed. For putative doublets within each class, additional rounds of DEG analysis were performed as necessary. Each time nuclei were re-moved, basic normalization, scaling, Harmony, and UMAP learning were repeated. To control for over-splitting, for clusters that appeared to be a single pile in the 2D UMAP space but were anno-tated into >1 cluster, additional rounds of DEG analysis were performed to see if binary markers could be found to label them. In addition, clustering was projected onto a 3D UMAP space to en-sure effects were not masked due to overcrowding in 2D. This strategy also helped to further elucidate cluster associations, aid decision-making with respect to groups of clusters that should be tested further, and spot potential gradient changes among clusters. If unique and or binary patterns could not be found in the current splitting resolution after these steps were performed, a step lower in resolution on the clustering tree was examined, and the analysis process was re-peated. The following compound naming convention to la

Submission date

Jan 26, 2021

Last update date

Aug 11, 2022

Contact name

Daniel S Reich

E-mail(s)

[email protected]

Organization name

NIH

Department

NINDS