By EVOBYTE Your partner in bioinformatics
Introduction
If you’ve ever wished you could zoom into a tumor, outline just the invasive edge, and ask “what genes are on here, and not over there?”, the GeoMx Digital Spatial Profiler (DSP) is built for that moment. Instead of covering an entire slide with barcoded capture spots, GeoMx lets you draw regions of interest (ROIs) directly on stained tissue, shine UV light precisely where you care, and turn those illuminated areas into digital counts you can analyze in R or Python. It blends classic pathology with targeted next‑generation sequencing (NGS), which is why it’s become a staple for FFPE‑heavy projects, translational cohorts, and studies where morphology should drive the questions.
How GeoMx Digital Spatial Profiling Works, Step by Step
At the bench, GeoMx starts like a standard immunofluorescence session. You stain sections with morphology markers that reveal compartments or cell types—think tumor versus stroma, neurons versus glia, or immune cell neighborhoods. On the same tissue, you hybridize probe pools: antibodies for proteins or in situ hybridization probes for RNA. Each affinity reagent carries a unique oligonucleotide barcode tethered by a UV‑cleavable linker. The instrument uses a digital micromirror device to project UV only onto your selected ROIs, where those barcodes are photocleaved and collected. Downstream, you quantify the released tags either via Illumina sequencing or by the nCounter optical barcode system; most RNA projects today use NGS readout for depth and scale.
The ROI concept is flexible by design. You can outline geometric shapes, irregular morphologies, or even multiple non‑contiguous areas. Within an ROI, GeoMx can segment subcompartments—often called areas of illumination (AOIs) or “segments”—such as tumor versus microenvironment. Although the optical mask can approach cellular size, typical recommendations favor tens to hundreds of cells per AOI to ensure robust transcript and protein detection, so you should treat GeoMx as a high‑plex, ROI‑level method rather than a single‑cell platform.
For content, GeoMx offers targeted panels and a Whole Transcriptome Atlas (WTA). WTA is still probe‑based, but it tiles essentially all protein‑coding genes for human or mouse and tends to work reliably on FFPE, a setting where poly(A) capture often struggles. This targeted‑yet‑broad strategy trades a small loss in “perfectly unbiased” discovery for sensitivity and FFPE compatibility across diverse tissues.
Finally, most labs pair GeoMx with NGS: released barcodes become libraries that run on Illumina instruments, and the DRAGEN‑accelerated GeoMx NGS Pipeline turns FASTQs into digital counts fast enough for routine cohorts. The pipeline outputs files you can load into GeoMx software or open‑source tools for QC, normalization, and downstream modeling.
Inside the Data: ROIs, AOIs, DCCs, PKCs, and Images
GeoMx data are not images you analyze pixel by pixel. Instead, you get spatially registered counts per AOI, with companion files that describe what each target is, where each AOI came from, and how the tissue looked.
The core count container is the DCC, short for Digital Count Conversion. A DCC holds probe‑level counts for a single AOI together with a header of run metadata and sequencing metrics. Panel definitions live in PKC files (Probe Kit Configuration), which map those probe IDs to target genes, designate negatives, and carry the annotation needed to collapse multiple probes into one gene. In R, the GeomxTools Bioconductor package reads DCCs and PKCs, builds a NanoStringGeoMxSet object, and gives you functions for QC, background adjustment, and gene‑level summarization.
You’ll also work with images. The instrument exports OME‑TIFFs that preserve multi‑channel immunofluorescence along with ROI overlays, enabling downstream overlays and QA of segmentation choices. Repo‑maintained tooling makes it easy to recreate the on‑instrument overlay and pair expression with morphology in a reproducible way. Many public consortia, including HuBMAP, standardize GeoMx submissions around OME‑TIFF plus DCC and annotation spreadsheets so data can travel well.
Because counts originate from cleaved barcodes, UMI handling and probe QC matter. Typical pipelines filter outlier probes, then collapse probe‑level counts to gene‑level matrices. Minimum depth, alignment, and saturation checks at the AOI level keep spurious or under‑sequenced AOIs from biasing results, and recommended cutoffs are now common across publications and package vignettes.
Here’s a minimal R workflow that reads DCCs and panel metadata, then returns a gene‑by‑AOI matrix you can take into DE, deconvolution, or pathway analysis:
# R: load GeoMx counts from DCC + PKC with GeomxTools library(GeomxTools) dcc_files <- list.files("data/dcc", pattern = "\\.dcc$", full.names = TRUE) pkc_files <- list.files("data/pkc", pattern = "\\.pkc$", full.names = TRUE) gmx <- readNanoStringGeoMxSet( dccFiles = dcc_files, pkcFiles = pkc_files, phenoDataFile = "data/annotations.xlsx", phenoDataSheet = "Sheet1", phenoDataDccColName = "Sample_ID", analyte = "RNA" ) # Collapse probe-level to gene-level counts (simplified) gmx_gene <- collapseToGenes(gmx) # see vignettes for QC steps expr <- Biobase::exprs(gmx_gene) # genes x AOIs matrix
GeoMx vs. Capture‑Based Spatial Transcriptomics
It helps to think of two mental models for spatial RNA. In the capture‑based model, you place tissue onto an array that’s pre‑patterned with a dense grid of barcoded capture features. When you permeabilize the tissue, released mRNAs bind those surface oligos; afterward, you sequence and use the spatial barcodes to put counts back where they came from. Resolution depends on the feature size and spacing—historically 55‑µm spots for Visium, and now 2‑µm squares for Visium HD—while chemistry can be poly(A) capture for fresh frozen or probe‑based ligation for FFPE. The entire capture area is assayed, so you get a continuous map rather than user‑drawn ROIs.
GeoMx flips that flow. Rather than letting the slide decide what to capture everywhere, you decide where to profile. Barcodes live on probes bound within the tissue, not on the slide, and UV light releases barcodes only in your ROIs or AOIs. Spatial linking comes from the instrument’s ROI map, not from spot barcodes baked into the substrate. In practice, that means GeoMx is morphology‑driven, excels with FFPE and multiplex protein/RNA panels on the same section or serial sections, and scales well for cohorts with many small samples, biopsies, or tissue microarrays. The tradeoff is that resolution is AOI‑level, not per‑spot pixels, and discovery is guided by where you chose to shine the light.
If you’re weighing them, frame the choice around the biological question. For atlas‑style maps where you don’t yet know what matters or you need a contiguous landscape, capture‑based platforms shine. For hypothesis‑driven work keyed to histology—say, tertiary lymphoid structures, invasive margins, micro‑metastases, or cell‑state gradients hugging a lesion—GeoMx’s ROI targeting is a strength. And because both ecosystems now offer whole‑transcriptome probe sets for FFPE, the real difference is less about gene list breadth and more about how spatial information is encoded and at what scale you analyze it. Independent comparisons echo these themes: Visium and Visium HD push spatial granularity across the slide, whereas GeoMx offers flexible ROI selection with robust FFPE performance but does not aim for single‑cell readouts.
Advantages, Limitations, and Practical Tips
GeoMx stands out for FFPE compatibility, multiplexed proteogenomics on the same material, and the ability to target exactly where a pathologist or image analysis suggests the story lives. When tissue is precious or patchy, the freedom to pool AOIs from multiple small foci beats hoping they happen to land on capture pixels. The WTA panels keep you in an “almost‑unbiased” mode while still benefiting from probe sensitivity in degraded RNA. Sequencing‑ready workflows and the DRAGEN‑accelerated pipeline mean you can move from run to count matrices in hours rather than days, which is a boon for large translational cohorts.
But there are tradeoffs worth planning around. First, signal is AOI‑averaged. If your biology hinges on cell‑to‑cell variation at micrometer scales, array capture or imaging‑based methods may be a better fit. Second, ROI choice is powerful and subjective; to reduce bias, predefine ROI strategies, store overlays, and keep metadata tidy so reviewers can trace how segmentation decisions influenced results. Third, because chemistry is probe‑based, novel isoforms and unexpected non‑coding RNAs won’t appear unless represented by probes; on the flip side, probe design also helps with specificity in FFPE. Finally, probe‑level QC and gene collapsing are essential; use recommended filters for low depth, alignment, and saturation, then collapse probes to gene counts with defensible rules before modeling. These steps are now standard in vignettes and recent papers that used GeoMx at scale.
A quick example makes the ROI mindset concrete. Imagine a colorectal cancer section where a tertiary lymphoid structure abuts tumor glands. With GeoMx, you’d stain for CD3/CD20 and cytokeratin, draw a handful of lymphoid ROIs segmented into T‑cell and B‑cell AOIs, plus companion tumor AOIs at defined distances from the edge. In one run, you’d generate a proteogenomic profile that compares immune niches to immediate tumor neighbors, preserving FFPE morphology and letting you revisit the exact overlay later. The same study on a capture array would create a full‑field map, which is great for discovering unknown niches, but you might spend more time post hoc finding the precise borders you care about.
Summary / Takeaways
GeoMx DSP is a morphology‑guided, ROI‑based spatial profiling platform that converts carefully illuminated regions into digital counts using UV‑cleavable, probe‑attached barcodes. It plays particularly well with FFPE, supports high‑plex RNA and protein on the same section or serial sections, and produces analyzable outputs—DCC counts plus PKC panel metadata and OME‑TIFF images—that snap cleanly into reproducible R workflows. If you need a continuous expression map at spot‑ or pixel‑scale across the whole slide, a capture‑based platform is often right. If, however, your question is anchored to distinct histological features and you want targeted, sensitive readouts from exactly those places, GeoMx is a strong, pragmatic choice.