Spatial Transcriptomics Series

Course 1 · Introduction to Spatial Transcriptomics

COURSE 1 OF 7

Who are you in today's study?

Each researcher below faces a different real-world spatial transcriptomics situation. Choose the profile that best matches your role or curiosity. Your decisions will shape the platform recommendation, experimental design, and next steps you receive.

Dr. Maya Chen Wet Lab Scientist · MD Anderson Cancer Center

You are studying the breast cancer tumor microenvironment — specifically, how immune cells infiltrate and spatially organize around the tumor mass. You have samples from 6 patients and access to both a shared sequencing facility and a core instrument lab with imaging-based platforms.

Your first critical decision happens before you request any reagents.

Question 1 of 2 — What type of tissue samples do you have?

Dr. Maya Chen Fresh-frozen · 6 patient samples

Fresh-frozen tissue gives you the best RNA quality and the widest platform options. Now the question is scope: do you already know which genes and cell types matter, or are you still in discovery mode?

The tumor microenvironment is complex and incompletely characterized. Choosing a targeted panel now means you can only see what you already know to look for.

Question 2 of 2 — What is the nature of your research question?

Dr. Maya Chen FFPE archival · 6 patient samples

FFPE tissue immediately narrows your platform options. Your three viable platforms are Visium FFPE, Xenium, and CosMx. Each makes a different trade-off.

Because your samples are archival, RNA integrity numbers (RIN) will vary. Plan a pre-screen on all 6 samples before committing any slides — one bad sample early can derail downstream batch analysis.

Question 2 of 2 — What matters most in your study design?

Your Platform: 10x Visium

Fresh-frozen · Whole-transcriptome discovery · Breast cancer TME

Platform Recommendation & Justification

10x Genomics Visium

Visium is the strongest choice for whole-transcriptome discovery on fresh-frozen tissue. Each 55 µm capture spot measures 18,000+ genes simultaneously. For a tumor microenvironment study, this breadth lets you detect unexpected cell populations — myeloid subtypes, fibroblast states, vascular endothelium — that you would miss with a targeted panel. With 6 patients, you can run 2 samples per slide, keeping costs manageable. Seurat and Giotto pipelines (Course 3) have mature Visium support.

Experimental Design Checklist

  • Confirm tissue sections are 10 µm and mounted correctly on Visium slides
  • Run H&E staining first — annotate tumor, stroma, and immune zones before sequencing
  • Target 50,000–100,000 reads per spot for TME analysis
  • Include at least 2 samples per patient condition (matched tumor + adjacent normal)
  • Use the same tissue block face for spatial and scRNA-seq if co-profiling

Common Mistakes to Avoid

  • Each Visium spot contains 2–20 cells — do not treat spots as single cells without deconvolution (Course 5)
  • Do not use fresh-frozen thresholds if any sample was partially thawed — re-RIN test before processing
  • Avoid processing all patients in one batch — batch effects in spatial data are severe and hard to correct
✦ Pathway complete

You are ready for Course 2: Data Import & Quality Control

In Course 2 you will load your Visium output folders into R using Seurat::Load10X_Spatial(), assess per-spot quality metrics, and filter spots before normalization.

Your Platform: 10x Xenium

Fresh-frozen · Targeted panel · Single-cell resolution

Platform Recommendation & Justification

10x Genomics Xenium

With a defined gene panel and fresh-frozen tissue, Xenium delivers subcellular resolution — you can assign individual transcripts to single cells, not just spots. For immune cell mapping in the tumor microenvironment, this distinction matters: a CD8⁺ T cell adjacent to a tumor cell reads very differently from one three cell-diameters away. Xenium panels support up to 480 genes (5K panel option available). MERFISH is an alternative if you need broader panel flexibility.

Experimental Design Checklist

  • Design your panel to include cell-type markers AND functional state markers — not just cell identity
  • Include housekeeping genes (ACTB, GAPDH) for normalization validation
  • Run a pilot on one sample section before committing all 6 patients
  • Capture at least 3 fields of view covering tumor core, invasive margin, and adjacent stroma
  • Plan for image segmentation QC — cell boundaries can be imprecise in high-density regions

Common Mistakes to Avoid

  • A 300-gene panel cannot be retrofitted into a whole-transcriptome question — commit to your scope before panel design
  • Do not skip pilot sequencing to save money — a failed full run costs 4× more
  • Xenium requires very flat, even tissue sections — uneven mounting creates focus errors across the run
✦ Pathway complete

You are ready for Course 2: Data Import & Quality Control

In Course 2 you will import Xenium output into R, assess transcript detection rates, and flag low-quality cells before downstream clustering.

Your Platform: Visium FFPE

FFPE archival tissue · Whole-transcriptome · Probe-based capture

Platform Recommendation & Justification

10x Genomics Visium FFPE

Visium FFPE uses probe-based hybridisation chemistry designed for degraded RNA. Instead of poly-A capture, probes bind directly to target sequences — making it robust against the fragmentation common in archival tissue. You will measure 18,000+ genes per spot, though gene counts will be lower than fresh-frozen (expect 2,000–4,000 vs 3,000–6,000). The spatial resolution and analysis pipeline are identical to standard Visium, so all Course 3 and 4 materials apply directly.

Experimental Design Checklist

  • Pre-screen all 6 FFPE blocks with DV200 score — aim for DV200 ≥ 50% before committing
  • Section at 5 µm (thinner than fresh-frozen standard) for better probe penetration
  • Use the FFPE-specific Visium tissue optimisation protocol — do not substitute fresh-frozen parameters
  • Apply FFPE-calibrated QC thresholds in Course 2 — do not reuse fresh-frozen cutoffs
  • Increase sequencing depth: target 70,000–120,000 reads per spot

Common Mistakes to Avoid

  • FFPE samples yield fewer genes per spot — applying fresh-frozen thresholds removes too many spots
  • Higher mitochondrial gene percentages are expected in FFPE — calibrate, do not eliminate these spots by default
  • Blocks older than 5 years may fail DV200 screening — check before planning replicates
✦ Pathway complete

You are ready for Course 2: Data Import & Quality Control

In Course 2 you will load Visium FFPE output into R, apply FFPE-specific QC filters, and learn why DV200-stratified thresholds produce more biologically valid results than fixed cutoffs.

Your Platform: CosMx SMI

FFPE archival tissue · Targeted panel · Single-cell resolution

Platform Recommendation & Justification

NanoString CosMx SMI

CosMx (Spatial Molecular Imager) is one of the few imaging-based platforms with validated FFPE protocols at single-cell resolution. It supports 1,000-gene RNA panels plus simultaneous protein detection — giving you immune phenotyping power alongside transcriptional state mapping. For a breast cancer TME study in archival tissue, this combination is difficult to match. Xenium FFPE is an emerging alternative worth evaluating if your institution already has the instrument.

Experimental Design Checklist

  • Validate the CosMx FFPE protocol on one representative block before running all 6 patients
  • Select RNA panel to include both tumor cell markers and immune cell subtypes
  • Add protein panel targets for CD8, CD4, PD-L1, and FOXP3 to enable direct immune subtype assignment
  • Plan image acquisition to cover at least 3 tumor-stroma interface regions per patient
  • Budget for data analysis time — file sizes are large and segmentation is compute-intensive

Common Mistakes to Avoid

  • Do not rely on automated cell segmentation without manual review in dense tumor regions
  • FFPE tissue folding artifacts create false-positive signals — quality-check tile images before analysis
  • CosMx RNA + protein data require different normalisation strategies — keep them separate until the integration step
✦ Pathway complete

You are ready for Course 2: Data Import & Quality Control

In Course 2 you will load CosMx output into R using the Seurat CosMx loader, assess per-cell transcript counts, apply segmentation quality filters, and prepare the data for cell-type assignment.

Dr. James Okonkwo Bioinformatician · Broad Institute

A wet-lab collaborator just dropped a dataset on your shared drive with the message: "Here's the spatial data from the mouse brain study. Let me know what you find." No methods section. No processing notes. Before you open R, you need to understand what you're working with.

You open the folder and see: a filtered_feature_bc_matrix/ directory, a spatial/ folder with tissue images, and a *.h5 file. This tells you something important.

Question 1 of 2 — What does this file structure tell you about the platform?

Dr. James Okonkwo Confirmed: Visium · Mouse brain

Correct. You open web_summary.html and confirm: 10x Visium, mouse brain, Space Ranger 2.1. Now you need to set your QC thresholds before filtering. You check the tissue image and notice most spots appear to be in cortex and hippocampus, with some in the cerebellum.

Your collaborator's email adds: "We used fresh-frozen OCT-embedded sections."

Question 2 of 2 — Which QC threshold approach should you apply?

Correct Approach: Adaptive QC

10x Visium · Mouse brain · Fresh-frozen · Data-driven thresholds

Your QC Strategy

10x Visium · Space Ranger 2.1

You identified the platform correctly from the file structure, and you chose the evidence-based QC approach. Different brain regions have dramatically different cell densities — cerebellum granule cells pack densely and produce fewer genes per spot, while cortical areas are more spread and gene-rich. Applying a single fixed cutoff across the whole section would systematically bias against one region. Inspect distributions first, set region-aware thresholds, then filter.

Your First R Commands in Course 2

  • Load data: Load10X_Spatial(data.dir, filename = "*.h5")
  • Calculate mitochondrial percentage: PercentageFeatureSet(pattern = "^mt-") (lowercase for mouse)
  • Plot violin plots of nFeature, nCount, and percent.mt
  • Spatially visualise QC metrics on the tissue image before filtering
  • Set thresholds based on inflection points in the distributions, not fixed values

Common Bioinformatics Mistakes

  • Mouse mitochondrial genes are mt- (lowercase) — using MT- (human) returns zero and skips the filter silently
  • Do not filter on nCount and nFeature independently without checking their correlation
  • Always check web_summary.html before analysis — it flags sequencing saturation and alignment rate issues
✦ Pathway complete

You are ready for Course 2: Data Import & Quality Control

In Course 2 you will work through the complete Visium QC pipeline in R — loading, inspecting, filtering, and producing a clean SpatialExperiment object ready for normalization in Course 3.

Dr. Sofia Reyes Oncologist · Memorial Sloan Kettering

You are studying colorectal cancer immune evasion. Your hypothesis: tumors with high microsatellite instability (MSI-H) have spatially distinct immune cell arrangements near the tumor-stroma boundary compared to microsatellite-stable (MSS) tumors. You have 24 FFPE blocks from a retrospective cohort — 12 MSI-H and 12 MSS.

Your clinical research fund is $45,000 for the spatial profiling component. Every dollar matters.

Question 1 of 2 — How would you characterise your research question?

Dr. Sofia Reyes Hypothesis-driven · $45,000 budget · 24 samples

Good reasoning — your hypothesis is specific enough to support a targeted approach. Now the budget constraint comes into focus. With 24 samples and $45,000, your per-sample budget is $1,875. Review the platform costs below.

Platform cost comparison — approximate per sample (reagents + sequencing)
Platform Approx. per sample FFPE compatible Single-cell resolution
Visium FFPE $1,200–1,800 ✓ Yes ✗ Spot-level
Xenium FFPE $1,500–2,200 ✓ Yes ✓ Yes
CosMx SMI $2,500–3,500 ✓ Yes ✓ Yes
MERFISH Variable Limited ✓ Yes

Question 2 of 2 — Which platform fits your per-sample budget?

Reconsider: discovery approach with budget constraint

Your Platform: Visium FFPE

FFPE · 24 samples · $45,000 budget · Whole transcriptome + deconvolution

Platform Recommendation & Budget Justification

10x Genomics Visium FFPE

At $1,200–1,800 per sample, Visium FFPE fits your 24-sample cohort within a $43,200 ceiling — leaving a buffer for library preparation variability and one failed sample replacement. RCTD, CARD, or SPOTlight deconvolution (Course 5) can estimate the proportions of CD8⁺, CD4⁺, and Treg cells per spot using a public scRNA-seq colorectal cancer reference. This is a well-established approach with peer-reviewed validation in colorectal cancer studies.

Study Design Checklist

  • Screen all 24 blocks for DV200 ≥ 50% — plan to replace up to 3 blocks from your biobank
  • Randomise sample processing order across batches to avoid MSI-H vs MSS batch confounding
  • Use the CRC Human Cell Atlas reference for deconvolution, or request a matched scRNA-seq dataset
  • Pre-register your deconvolution method choice before analysis
  • Budget sequencing at 70,000–100,000 reads per spot for FFPE tissue quality

Clinical Study Pitfalls

  • Do not co-process MSI-H and MSS samples in separate batches — batch effects will confound the primary comparison
  • Deconvolution estimates are probabilistic — report confidence intervals, not absolute cell counts
  • Archival blocks from different clinical sites may have different fixation protocols — document and report this
✦ Pathway complete

You are ready for Course 2: Data Import & Quality Control

In Course 2 you will import your 24 Visium FFPE datasets, apply FFPE-calibrated QC thresholds, and build a batch-corrected multi-sample object ready for spatial domain analysis in Course 4.

Your Platform: Xenium FFPE

FFPE · 20 samples (budget-adjusted) · Single-cell resolution · Targeted panel

Platform Recommendation & Budget Justification

10x Genomics Xenium FFPE

Reducing from 24 to 20 samples (10 MSI-H, 10 MSS) brings your total to approximately $40,000–44,000 — within budget. Xenium's single-cell resolution directly addresses your hypothesis about immune cell spatial arrangements. You will assign individual CD8⁺ T cells, CD4⁺ T cells, and FOXP3⁺ Tregs to precise coordinates within the tumor border zone. A well-powered study with direct single-cell evidence is publishable in higher-impact journals than a larger spot-level study.

Study Design Checklist

  • Design a 200–400 gene panel including immune markers, checkpoint genes, and epithelial/stromal markers
  • Include EPCAM and CDH1 to define the tumor boundary — essential for tumor-stroma distance analysis
  • Validate the FFPE Xenium protocol on 2 pilot samples before committing the full cohort
  • Plan image acquisition to capture ≥ 3 tumor-invasion-front regions per sample
  • Pre-define your spatial analysis zone (for example, 100 µm from the tumor boundary) in your analysis plan

Clinical Study Pitfalls

  • Verify your core lab has completed ≥ 5 successful FFPE Xenium runs before committing your cohort
  • Cell segmentation in dense tumor epithelium is error-prone — budget manual review time for boundary zone images
  • A 4-sample reduction in a retrospective cohort requires power analysis justification — document this in your pre-registration
✦ Pathway complete

You are ready for Course 2: Data Import & Quality Control

In Course 2 you will load your Xenium FFPE output, apply cell-level QC filters, flag segmentation artifacts in the tumor boundary zone, and prepare a clean dataset for spatial neighbourhood analysis in Course 4.