GoKawiilVertebrate scRNA and snRNA atlas collection, filtering and preprocessing
Cell atlases were retrieved from previous publications2,3,4,5. Low-quality cells in the human atlas were further filtered based on nCount (UMI) < 400. Low-quality cells in the other atlases were already filtered. To focus on neural cells in the brain, vertebrate datasets were filtered to retain only brain tissues at juvenile or adult stages. To help balance the number of cells for cross-species integration and to accommodate different proportions of neurons and glia, we randomly downsampled the human and lizard atlases to 105 neurons and 105 non-neurons, but retained the full brain atlases for mouse (67,937 neurons and 60,395 non-neuronal cells) and lamprey (18,166 neurons and 41,472 non-neuronal cells). Only protein-coding genes were retained for downstream analyses.
As the original atlases were generated using different pipelines, we applied a standardized preprocessing approach to ensure consistency. We performed SAM analysis on each individual atlas by directly invoking the SAMAP function from the SAMap package (which runs SAM internally)20. Specifically, UMI counts from each cell were first normalized to give the median total count per cell, then log 2 -transformed followed by applying the SAM function with the following parameters: preprocessing=“StandardScaler”, npcs=100, weight_PCs=False, k=20, n_genes=3000, weight_mode=‘rms’. The anndata objects were then converted to Seurat format for downstream clustering.
Amphioxus sample collection, scRNA and snRNA library construction and raw data processing
Amphioxus (B. floridae) were obtained from a stock maintained by J.-K. Yu originating from Tampa, Florida. The amphioxus and their offspring were maintained at Xiamen University under previously described conditions65. The brain (anterior to the first dorsal ocellus) and neural tube (posterior to the first dorsal ocellus) were dissected as previously described29. We constructed and sequenced one scRNA-seq and one snRNA-seq library for each tissue.
For the scRNA-seq experiment, the dissected brain (from ten adult individuals) and neural tube (from eight adult individuals) tissues were respectively washed three times in ice-cold calcium-free and magnesium-free artificial seawater (CMF-ASW)66 and then transferred into 500 µl enzyme mix (10% trypsin and 2 mg ml–1 collagenase in CMF-ASW) and incubated in a 37 °C incubator with a nutating shaker for approximately 10 min. During digestion, tissues were gently pipetted every 1–2 min to facilitate dissociation, and progress was monitored under an inverted microscope. Digestion was terminated by adding 1 ml of an ice-cold quenching solution (20% fetal bovine serum and 2 mg ml–1 glycine in CMF-ASW). Cells were passed through a 40 µm cell strainer and centrifuged at 270g at 4 °C for 5 min. The supernatant was removed, and 500 µl RNase-free 0.04% BSA in 3× PBS was added to resuspend the cells. Calcein-AM (BD Biosciences, 564061) was added to the cell suspension to a final concentration of 10 µM and incubated at 37 °C for 5 min. The cells were subsequently placed on ice then immediately processed. scRNA-seq library construction was carried out in accordance with a previous study29. The final libraries were sequenced on an Illumina NovaSeq 6000 platform.
For the snRNA-seq experiment, we used a Nucleus Isolation kit (SHBIO, 52009-10) to obtain single nuclei of the dissected tissues. RNase inhibitors (Sigma, 3335399001) were added to the reagents before use. The samples were cut and transferred to a 5 ml tube containing lysate, mixed and lysed for 2 min on ice, then filtered through a 40 μm cell filter (Sigma, BAH136800040). The nucleus count was estimated using a microscope (Leica) with DAPI reagent. After staining with 0.4% trypan blue (Sangon Biotech E607320-0001), the nucleus was observed under a ×40 microscope (Jiangnan Novel Optics XD-202). Subsequent experiments were performed if the nuclear envelopes were intact and there were few impurities. snRNA-seq libraries were prepared using a SeekOne DD Single Cell 3′ library preparation kit (SeekGene, K00202). In brief, an appropriate number of cell nuclei was mixed with reverse transcription reagent and then added to a sample well in a SeekOne DD chip S3. Subsequently, barcoded hydrogel beads and partitioning oil were dispensed into corresponding wells separately in the chip S3. After emulsion droplet generation, reverse transcription was performed at 42 °C for 90 min and inactivated at 85 °C for 5 min. Next, cDNA was purified from broken droplets and amplified by PCR. The amplified cDNA product was then cleaned, fragmented, end-repaired, A-tailed and ligated to a sequencing adaptor. Finally, indexed PCR was performed to amplify the DNA representing the 3′ polyA part of expressing genes, which also contained the cell barcode and the unique molecular index. The indexed sequencing libraries were cleaned up using VAHTS DNA Clean Beads (Vazyme N411-01), analysed by a Qubit (Thermo Fisher Scientific, Q33226) and a Bio-Fragment Analyzer (Bioptic, Qsep400). The libraries were then sequenced on a GeneMind SURFSeq 5000 with PE150 read length.
Raw reads from scRNA-seq were processed using the BD Rhapsody WTA analysis pipeline (v.1.12.1; https://bitbucket.org/CRSwDev/cwl/src/master/) on the Seven Bridges platform (https://sevenbridges.com/). Raw reads from snRNA-seq were processed using the SeekSoul Tools pipeline. scRNA and snRNA expression matrices for each sample were then filtered and processed using Seurat (v.5.0.0). Cells or nuclei with fewer than 300 detected genes, more than 4,000 detected genes, more than 10,000 UMI detected or more than a 10% MT expression ratio were filtered out (we used stricter parameters for neural tube processed by SeekGene due to its higher ambient RNA).
Clustering and annotation
To find good-quality and high-resolution cell clusters in the SAM preprocessed atlases, we performed hierarchical and iterative clustering for individual vertebrate cell atlases using the scrattch.hicat and scrattch.bigcat packages67,68 from the Allen Institute. Raw counts (UMI) were first normalized using the cpm function provided in the above packages, followed by log 2 transformation with a pseudo-count added to prevent log 2 [0]. Cells were initially classified into broad groups and hierarchically clustered on the basis of the expression of highly variable genes, principal component analysis and Jaccard–Louvain clustering. Clustering was performed iteratively in each group using the iter_clust function, continuing until no further subclusters satisfied predefined thresholds for the number of DEGs or minimum cluster size. As our analysis did not aim to resolve extremely fine-scale cell types, we applied more relaxed parameters than those typically used with this method. DEG thresholds were defined via the de_param settings: padj.th = 0.05, q1.th = 0.4, q2.th = NULL, q.diff.th = 0.5, de.score.th = 100, min.cells = 100, and min.genes = 6. Dimensionality reduction and clustering parameters were specified as follows: dim.method = “pca”, max.dim = 80, method = “louvain”. Minimum cluster sizes were set via split.size as 800, 500, 500 and 500 for human, mouse, lizard and lamprey datasets, respectively. As we did not aim to study cell types at very high resolution, we tuned the split.size parameters for each species to generate cluster numbers at a similar level across vertebrates. Clusters were then checked and merged at the end of the iteration to ensure that they were separable with scrattch.bigcat::merge_cl. We simply used Seurat FindClusters with the Louvain algorithm and resolution = 1 for amphioxus owing to the limited number of cells and nuclei in the datasets.
... continue reading