Skip to main content
. 2024 Jan 24;12:RP84628. doi: 10.7554/eLife.84628

Figure 2. Spatial gene expression in the human locus coeruleus (LC) using spatially-resolved transcriptomics (SRT).

(A) Spots within manually annotated LC regions containing norepinephrine (NE) neurons (red) and non-LC regions (gray), which were identified based on pigmentation, cell size, and morphology from the H&E stained histology images, from donors Br2701 (top row) and Br8079 (bottom row). (B) Expression of two NE neuron-specific marker genes (TH and SLC6A2). Color scale indicates unique molecular identifier (UMI) counts per spot. Additional samples corresponding to A and B are shown in Figure 2—figure supplements 1 and 3A, B. (C) Boxplots illustrating the enrichment in expression of two NE neuron-specific marker genes (TH and SLC6A2) in manually annotated LC regions compared to non-LC regions in the n=8 Visium samples. Values show mean log-transformed normalized counts (logcounts) per spot within the regions per sample. Additional details are shown in Figure 2—figure supplement 3C. (D) Volcano plot resulting from differential expression (DE) testing between the pseudobulked manually annotated LC and non-LC regions, which identified 32 highly significant genes (red) at a false discovery rate (FDR) significance threshold of 10–3 and expression fold-change (FC) threshold of 3 (dashed blue lines). Horizontal axis is shown on log2 scale and vertical axis on log10 scale. Additional details and results for 437 statistically significant genes identified at an FDR threshold of 0.05 and an FC threshold of 2 are shown in Figure 2—figure supplement 8 and Supplementary file 2A. (E) Average expression in manually annotated LC and non-LC regions for the 32 genes from D. Color scale shows logcounts in the pseudobulked LC and non-LC regions averaged across n=8 Visium samples. Genes are ordered in descending order by FDR (Supplementary file 2A). (F–G) Cross-species comparison showing expression of human ortholog genes for LC-associated genes identified in the rodent LC (Mulvey et al., 2018; Grimm et al., 2004) using alternative experimental technologies. Boxplots show mean logcounts per spot in the manually annotated LC and non-LC regions per sample in the human data.

Figure 2.

Figure 2—figure supplement 1. Spot-plot visualizations of manually annotated Visium spots within regions identified as containing LC-NE neurons in SRT data.

Figure 2—figure supplement 1.

For each of the n=9 Visium capture areas (hereafter referred to as samples), the spots were manually annotated as being within the LC regions (red) or within the non-LC regions (gray) based on spots containing NE neurons, which were identified by pigmentation, cell size, and morphology on the H&E stained histology images.
Figure 2—figure supplement 2. H&E stained histology images and number of cells per spot for SRT data.

Figure 2—figure supplement 2.

(A) Hematoxylin and eosin (H&E) stained histology images for the n=9 Visium samples. Higher-resolution images can also be viewed through the Shiny web app (https://libd.shinyapps.io/locus-c_Visium/). (B) Estimated number of cells per spot within annotated LC regions in 6 Visium samples, based on application of cell segmentation software (VistoSeg, Tippani et al., 2023). Boxplots show medians, first and third quartiles, whiskers extending to the furthest values no more than 1.5 times the interquartile range from each quartile, and outliers.
Figure 2—figure supplement 3. Spatial expression of two NE neuron-specific marker genes in Visium samples for quality control (QC) in SRT data.

Figure 2—figure supplement 3.

(A-B) Spot-plot visualizations of NE neuron marker gene expression (TH and SLC6A2, A and B, respectively) in the n=9 Visium samples. Color scale shows UMI counts per spot. One sample (Br5459_LC_round2) did not show clear expression of the NE neuron marker genes. This sample was excluded from subsequent analyses, leaving n=8 Visium capture areas (samples) from 4 out of the 5 donors. The annotated LC regions are shown in Figure 2—figure supplement 1. (C) Enrichment of NE neuron marker gene expression (TH and SLC6A2) within manually annotated LC regions compared to non-LC regions in the n=8 Visium samples that passed QC. Boxplots show values as mean log-transformed normalized counts (logcounts) per spot within each region per sample, with samples represented by shapes.
Figure 2—figure supplement 4. Spot-level quality control (QC) data visualizations for Visium samples in SRT data.

Figure 2—figure supplement 4.

(A) QC metrics, medians per sample (from left to right: sum of UMI counts per spot, number of detected genes per spot, and proportion of mitochondrial reads per spot). Boxplots show median for each QC metric per sample, with samples represented by shapes. (B) Applying thresholds of 3 median absolute deviations (MADs) to the sum of UMI counts and number of detected genes for each sample identified a total of 287 low-quality spots (red) (1.4% out of 20,667 total spots), which were removed from subsequent analyses. We did not use the proportion of mitochondrial reads for spot-level QC filtering (see Methods for more details).
Figure 2—figure supplement 5. Dimensionality reduction embeddings before and after batch integration across Visium samples in SRT data.

Figure 2—figure supplement 5.

We applied a batch integration tool (Harmony, Korsunsky et al., 2019) to remove technical variation in the molecular measurements between the n=8 Visium samples from 4 donors. The integrated measurements were subsequently used as the input for spatially-aware clustering using BayesSpace (Zhao et al., 2021). (A) Principal component analysis (PCA) (top 2 PCs) calculated on molecular expression measurements, with spots labeled (left to right) by donor ID, round ID, and sample ID, without applying any batch integration. (B) Harmony embeddings (top 2 Harmony embedding dimensions) after applying Harmony batch integration on sample IDs, with spots labeled (left to right) by donor ID, round ID, and sample ID, demonstrating that the technical variation has been reduced.
Figure 2—figure supplement 6. Identifying LC and non-LC regions in a data-driven manner by spatially-aware unsupervised clustering in SRT data.

Figure 2—figure supplement 6.

We applied a spatially-aware unsupervised clustering algorithm (BayesSpace, Zhao et al., 2021) to investigate whether the LC and non-LC regions in each Visium sample could be annotated in a data-driven manner. (A) Using BayesSpace with k=5 clusters, we clustered spots from the n=8 Visium samples using the Harmony batch-integrated molecular measurements (clustering performed across samples). Cluster 4 (red) corresponds most closely to the manually annotated LC regions. The annotated LC regions are shown in Figure 2—figure supplement 1. (B) BayesSpace clustering performance evaluated in terms of concordance between cluster 4 (red) and the manually annotated LC region in each sample. Clustering performance was evaluated in terms of precision, recall, F1 score, and adjusted Rand index (ARI) (see Methods for definitions).
Figure 2—figure supplement 7. Comparison of spot-level and region-level manual annotations in SRT data.

Figure 2—figure supplement 7.

(A) We manually annotated individual Visium spots (black) overlapping with NE neuron cell bodies within the previously manually annotated LC regions (red), based on pigmentation, cell size, and morphology from the H&E stained histology images, in the n=8 Visium samples. (B) We observed relatively low overlap between spots with expression of the NE neuron marker gene TH (≥2 observed UMI counts per spot) and the set of annotated individual spots. The differences included both false positives (annotated spots that were not TH+) and false negatives (TH+ spots that were not annotated). Therefore, we did not use the spot-level annotations for subsequent analyses, and instead used the LC region-level annotations for all further analyses.
Figure 2—figure supplement 8. Results from differential expression (DE) analysis to identify expressed genes associated with LC regions in SRT data.

Figure 2—figure supplement 8.

We performed DE testing between the manually annotated LC and non-LC regions by pseudobulking spots, defined as aggregating UMI counts from the combined set of spots, within the annotated LC and non-LC regions in each sample. (A) Using a false discovery rate (FDR) significance threshold of 10-3 and an expression fold-change (FC) threshold of 3 (dashed blue lines), we identified 32 highly significant genes (red points). (B) Using standard significance thresholds of FDR <0.05 and expression FC >2, we identified 437 significant genes (red). Vertical axes are on reversed log10 scale, and horizontal axes are on log2 scale. Additional details are provided in Supplementary file 2A.
Figure 2—figure supplement 9. Results from applying nnSVG to identify spatially variable genes (SVGs) in SRT data.

Figure 2—figure supplement 9.

We applied nnSVG (Weber et al., 2023c), a method to identify spatially variable genes (SVGs), in the Visium SRT samples. We ran nnSVG within each contiguous tissue area containing a manually annotated LC region (13 tissue areas in the n=8 Visium samples) and calculated an overall ranking of top SVGs by averaging the ranks per gene from each tissue area. (A) The top 50 ranked SVGs from this analysis included a subset (11 out of 50) of genes that were highly ranked in samples from only one donor (Br8079, genes highlighted in maroon). We determined that this was due to the inclusion of a section of the choroid plexus adjacent to the LC for this donor. Bars show the number of times (out of 13 tissue areas) each gene was included within the top 100 SVGs. Rows are ordered by overall average ranking in descending order. (B) Spatial expression of CAPS, a choroid plexus marker gene, in the n=8 Visium samples. (C) Histology image showing the two tissue areas for sample Br8079_LC_round3. (D) In order to focus on LC-associated SVGs, we calculated an overall average ranking of SVGs that were each included within the top 100 SVGs in at least 10 out of the 13 tissue areas, which identified 32 highly-ranked, replicated LC-associated SVGs. Boxplots show the ranks in each tissue area. Rows are ordered by the overall average ranking in descending order.