從 scRNA-seq 出發，為什麼還需要 ST？
單細胞轉錄組學 (scRNA-seq) 解決了「不同細胞表達不同基因」的問題，但在解離 (dissociation) 過程中，細胞的空間關係徹底遺失。然而生物學的核心常常正是空間：腫瘤微環境怎麼分區？發育時細胞如何沿軸排列？神經元如何按腦區分層？這些問題 scRNA-seq 沒辦法回答。
Spatial Transcriptomics (ST) 一邊保留組織切片的二維/三維座標，一邊量化 RNA。換句話說，每個 RNA 分子都帶著一組 (x, y) 甚至 (x, y, z) 標籤回到分析中。這讓我們可以：
- 觀察基因表達跟解剖學結構（如皮層、髓質、毛細血管）的對應
- 回答「哪兩種細胞物理上相鄰時會啟動哪條訊號路徑」
- 在保留組織切片的前提下做 H&E / IF 影像 + 表達雙模態整合
- 研究稀有/邊界細胞（如腫瘤侵襲前緣、傷口邊界）
Single-cell RNA-seq (scRNA-seq) solved "different cells express different genes," but tissue dissociation completely destroys spatial context. Yet much of biology is fundamentally spatial: how is the tumor microenvironment zoned? How do cells line up along developmental axes? How do neurons stratify across cortical layers? These questions are out of reach for scRNA-seq.
Spatial Transcriptomics (ST) preserves the 2D / 3D coordinates of a tissue section while quantifying RNA. Every transcript carries an (x, y) — sometimes (x, y, z) — tag into the analysis. This lets us:
- Map gene expression onto anatomy (cortex, medulla, capillaries)
- Ask "which signaling pathway fires when these two cell types are physically adjacent"
- Combine H&E / IF imaging with expression as paired modalities on the same section
- Study rare or boundary cells (tumor invasion fronts, wound margins)
兩大技術路線：Spot-based vs Image-based
Spot-based / NGS-based
把組織貼在帶有空間條碼 (spatial barcode) 的捕捉陣列上，cDNA 合成完直接走 NGS。
代表：10x Visium、Visium HD、Slide-seq V2、Stereo-seq、GeoMx DSP。
優點：全轉錄組覆蓋（沒有 panel 限制）、適合做假說產生、和現有 scRNA-seq 工具相容。
缺點：多數情況一個 spot 包含多個細胞 (Visium 55 µm 約 1–10 個細胞)，需要做 deconvolution。
Tissue is placed on a capture array with positional barcodes; cDNA is then sequenced by NGS.
Examples: 10x Visium, Visium HD, Slide-seq V2, Stereo-seq, GeoMx DSP.
Pros: whole-transcriptome (no panel limit), great for hypothesis generation, compatible with the scRNA-seq toolchain.
Cons: each spot usually covers multiple cells (Visium 55 µm spot ≈ 1–10 cells), so deconvolution is required.
Image-based / FISH-based
用螢光探針或 in-situ sequencing 反覆顯微成像每個 RNA 分子，達到單細胞甚至 subcellular 解析度。
代表：10x Xenium、Vizgen MERFISH/MERSCOPE、NanoString CosMx SMI、STARmap。
優點：真正的單細胞解析度、保留 RNA 的 (x, y, z) 位置、適合假說驗證。
缺點：必須選定 panel（100–6000 基因），無法量到 panel 外的基因；分析瓶頸在細胞切割。
Fluorescent probes or in-situ sequencing image each RNA molecule cyclically — reaching single-cell or even subcellular resolution.
Examples: 10x Xenium, Vizgen MERFISH/MERSCOPE, NanoString CosMx SMI, STARmap.
Pros: true single-cell resolution, preserves (x, y, z) of every RNA, ideal for hypothesis testing.
Cons: requires a fixed panel (100–6 000 genes); off-panel genes are invisible. Bottleneck shifts to cell segmentation.
互動：spot 解析度 vs 細胞密度
拖動下方滑桿，觀察 spot 直徑如何決定一個 spot 內捕獲到的細胞數。藍色圓點代表細胞、淺紫色圓圈代表 spot 涵蓋範圍。
Move the sliders to see how spot diameter controls the number of cells captured per spot. Blue dots are cells; lavender circles are spot footprints.
畫面範圍：500 µm × 230 µm
ST 分析的標準骨幹（與 scRNA-seq 對照）
| 步驟 | scRNA-seq | Spatial Transcriptomics | ||
|---|---|---|---|---|
| QC | 細胞層級：nFeature、nCount、MT% | spot 層級 + 加入空間鄰近 outlier (SpotSweeper) 與組織覆蓋檢查 | Cell-level: nFeature, nCount, MT% | Spot-level + spatially-aware outlier (SpotSweeper) + tissue coverage check |
| Norm. | LogNormalize / SCTransform | 同上，但要注意 spot 含多細胞時 SCTransform 假設可能失準 | Same, but SCTransform assumptions may break when spots contain mixed cells | |
| HVG / Features | Variance / dispersion | SVG (空間變異基因)：Moran's I、SPARK-X、nnSVG | SVG (Spatially Variable Genes): Moran's I, SPARK-X, nnSVG | |
| Clustering | Louvain / Leiden in PCA | 空間區域 (Spatial domain)：BayesSpace, SpaGCN, GraphST, BANKSY | Spatial domains: BayesSpace, SpaGCN, GraphST, BANKSY | |
| Cell typing | Marker / SingleR / Azimuth | Spot-based: Deconvolution (cell2location, RCTD…)；Image-based: 直接 segmentation + 分群 | Marker / SingleR / Azimuth | Spot-based: deconvolution (cell2location, RCTD…); Image-based: segmentation + clustering |
| Cell communication | CellChat / CellPhoneDB | CellChat v2 (spatial) / COMMOT | ||
| Special | Trajectory, Integration | Niche / Microenvironment、3D 重建、histology integration | Trajectory, Integration | Niche / microenvironment, 3D reconstruction, histology integration |
什麼時候該選 ST？怎麼選平台？
🌳 ST 平台選擇
第一個 ST 物件
# 讀入 10x Visium 標準輸出 / Load standard 10x Visium output library(Seurat) vis <- Load10X_Spatial(data.dir = "sample_visium/") # vis 同時包含 RNA 表達矩陣 + spatial coordinates + low-res tissue image SpatialFeaturePlot(vis, features = "nCount_Spatial") SpatialFeaturePlot(vis, features = "MS4A1") # B-cell marker 直接畫回組織
import scanpy as sc import squidpy as sq # 讀入 10x Visium / Load Visium adata = sc.read_visium("sample_visium/") adata.var_names_make_unique() # 直接把 nCount 和某基因畫回組織 sq.pl.spatial_scatter(adata, color=["total_counts", "MS4A1"])
📝 自我檢測
1. 對於想要回答「腫瘤侵襲前緣的免疫細胞跟腫瘤細胞如何互動」這個問題，下列哪個技術最適合？
1. To answer "how do immune cells interact with tumor cells at the invasion front," which technology fits best?
2. 關於 Spot-based 與 Image-based 的差異，下列敘述何者正確？
2. Which statement comparing spot-based vs image-based ST is correct?
3. 在 ST 流程中，以下哪一個步驟是相對於 scRNA-seq 「新增」出來、特別重要的？
3. Which step is genuinely new and especially important in ST compared to scRNA-seq?