Full-length single-cell spatial transcriptomics workflow connects transcript isoforms with cell identity and tissue location instead of stopping at gene-level expression. Using the 2026 Fullscope-seq macaque brain study as a central example, this article examines spatial barcode recovery, programmed cDNA concatenation, long-read benchmarking, multi-scale differential transcript usage (DTU), RNA-ISH/RNA-FISH validation, and cross-species disease relevance. It also explains when isoform-level spatial information may add value beyond conventional short-read spatial transcriptomics.

Key takeaways

• Fullscope-seq was developed to combine single-cell spatial coordinates with full-length transcript information in large tissue sections.
• In the reported benchmark, CID recovery increased from 56.8% with Direct-ONT to 89.7% with Fullscope-seq.
• Isoform-level analysis revealed differences across cortical layers, cell subclasses, and brain regions that gene-level expression alone could miss.
• Layer-level DTU was more closely linked to cell composition, whereas regional DTU reflected both cell composition and spatial context.
• Full-length spatial profiling is most relevant when the research question depends on splice isoforms, alternative transcription start or end sites, or isoform switching—not simply where a gene is expressed.

Why Spatial Isoforms Matter

Spatial transcriptomics has changed how researchers connect molecular profiles with tissue architecture. It can show where genes are expressed, how expression domains align with anatomy, and which cell populations occupy particular regions. For many studies, those gene-level answers are sufficient.

The limitation appears when the biological question depends on which transcript version of a gene is present. A single gene can produce multiple RNA isoforms through alternative exon usage, transcription start sites, polyadenylation sites, and other RNA-processing events. Those transcript structures can vary between cell types, developmental states, and anatomical regions. Recent single-cell long-read studies have shown that isoform regulation can carry information not captured by gene-level abundance alone. [2]

This creates a three-way technical problem: researchers may want single-cell identity, spatial coordinates, and full-length transcript structure in the same experiment. Conventional short-read spatial assays are strong for spatial gene expression but often provide limited direct information about complete transcript structures. Long-read sequencing can resolve transcript architecture, but spatial barcodes must survive library preparation, sequencing, and computational recovery.

For projects centered on isoforms, the key question is therefore not simply "Which gene is here?" It becomes: Which transcript isoform is expressed, in which cell, and at which location? Researchers comparing these goals can also review our guide to short-read vs full-length transcriptomics.

Five-Step Study Workflow

The Fullscope-seq study can be understood as a five-step workflow that moves from method engineering to biological interpretation. This structure is useful because it separates technical feasibility from downstream claims.

Method and Benchmarking

First, Stereo-seq coordinate IDs (CIDs) were linked to long-read sequencing through biotin enrichment and programmed cDNA concatenation. The method was then benchmarked on macaque coronal brain sections, with attention to spatial coverage, CID recovery, and isoform detection.

Multi-Scale DTU Analysis

Next, differential transcript usage was evaluated across four spatial scales: cortical layer, subregion, cell subclass, and major brain region. A generalized linear model was used to separate the contribution of cellular composition from the contribution of spatial region.

Validation and Disease Relevance

Representative DTU genes were independently examined with RNA-ISH or RNA-FISH. Finally, human GWAS and TWAS datasets were integrated to evaluate whether macaque DTU genes overlapped neuropsychiatric disease-associated biology and whether selected transcript structures were conserved across species.

Five-step sequence: Method development → performance benchmarking → multi-scale DTU analysis → experimental validation → cross-species disease relevance.

Figure 1. Five-step full-length single-cell spatial transcriptomics workflow used to map and validate spatial isoform variation.

Building and Benchmarking Fullscope-seq

The central technical obstacle was spatial barcode retention. Stereo-seq uses coordinate IDs to associate captured molecules with known positions on a patterned array. Once cDNA is moved into a long-read workflow, failure to recover the CID means that a transcript may still be sequenced but can no longer be placed confidently back into tissue space.

Recovering Spatial CID Barcodes

In the study summary, direct long-read sequencing of cDNA with ONT lost usable CID information from roughly half of sequences because of template-switching oligonucleotide and PCR-related artifacts. Direct-ONT recovered CIDs from 56.8% of reads in the reported comparison.

Fullscope-seq addressed this problem by enriching molecules that retained the spatial barcode. Biotin-based selection was used to preferentially retain CID-containing cDNA before the long-read stage. The reported CID recovery increased to 89.7%.

This is an important distinction from simply generating longer reads. For spatial isoform analysis, a long transcript is only useful if its molecule-level transcript structure can still be connected to a valid spatial coordinate.

Programmed cDNA Concatenation

The workflow then used 15 adapter types to concatenate full-length cDNA molecules. Programmed concatenation allows multiple cDNA fragments to be linked into longer sequencing constructs while preserving the sequence information needed to separate molecules computationally afterward.

The study described compatibility with ONT, PacBio, and Cyclone long-read sequencing platforms. The main paper defines Fullscope-seq as a full-length, single-molecule, large-field-of-view spatial transcriptomics method based on programmed cDNA concatenation and designed for multiple long-read platforms. [1]

Researchers planning a related project should distinguish the experimental barcode-retention problem from the computational spatial-mapping problem. Standard Stereo-seq processing already requires mRNA spatial-position reconstruction, alignment, expression-matrix generation, and downstream clustering. The Stereo-seq CID data analysis workflow provides useful background on that gene-level pipeline.

Benchmark Metrics

The benchmark used macaque coronal brain sections up to 5 cm × 3 cm, allowing the study to test the approach across a large tissue field rather than a small region of interest. The study summary reported 6,434 novel isoforms, an average of 2.33 isoforms per gene, and multiple isoforms in more than 30% of genes.

These figures should be interpreted as study-specific performance results, not universal specifications for every tissue, platform, or sequencing design. Long-read transcript recovery depends on RNA quality, cDNA completeness, sequencing depth, transcript abundance, annotation quality, and computational filtering. Independent long-read spatial studies have likewise emphasized barcode recovery, read truncation, mapping error, and isoform quantification as critical analytical issues. [3][4]

For researchers who need large-field spatial mapping, CD Genomics also supports Stereo-seq spatial transcriptomics as part of research-use-only spatial study design.

Layer-Specific Isoform Usage

Once full-length transcripts could be assigned back to space, the study asked whether cortical layers differed in how they used transcript isoforms. In the M1-1 sample, the analysis identified 981 layer-level DTU isoforms from 576 genes.

DTU Is Not the Same as Differential Expression

Differential transcript usage (DTU) measures changes in the relative use of transcript isoforms from the same gene. A gene can therefore show little change in total expression while its dominant transcript structure changes substantially between biological contexts.

That distinction matters in the cortex, where layers differ in cellular composition, connectivity, and molecular programs. A gene-level count can combine multiple transcript forms into one value and hide a shift in exon usage, transcription start sites, or 3' ends.

Layer-Dependent Transcript Structures

The study summary highlighted several examples. CD47 showed L1-associated exon skipping. NTRK2 showed alternative 3'-end usage. DBNDD2, a gene linked in prior research to schizophrenia biology, showed a layer 6-specific transcription-start-site pattern.

A particularly informative case was DDRGK1-FS1, a newly identified isoform associated with a proximal transcription start site. It was enriched in superficial cortical layers L1–L3 but was not resolved in the short-read Stereo-seq dataset. RNA-ISH provided an independent spatial check of the pattern.

The practical lesson is not that long reads should replace short reads in every spatial study. It is that transcript-structure questions require transcript-structure evidence. When a hypothesis depends on alternative splicing, alternative TSS/TES usage, or full transcript architecture, single-cell full-length RNA sequencing can provide a complementary layer of information for appropriate study designs.

Cell-Type Isoform Heterogeneity

The analysis next moved from cortical layers to cell identity. Based on annotations covering 388,443 single cells and 23 subclasses, the study summary reported 1,105 DTU genes across cell subclasses.

Gene-Level Analysis Can Miss Isoform Differences

Approximately 85% of those DTU genes did not show corresponding differences at the overall gene-expression level. This is one of the most important results for study design because it demonstrates why isoform-level analysis can answer a different question from conventional differential expression.

Imagine two cell subclasses with similar total abundance of gene X. A gene-level analysis may classify them as similar. Yet one subclass could predominantly use isoform A while the other uses isoform B. If those isoforms differ in coding sequence, untranslated regions, or regulatory elements, the functional interpretation may also differ.

Neuronal and Glial Patterns

Astrocytes showed the largest reported number of DTU genes, with 389, followed by oligodendrocytes and L2 glutamatergic neurons. Glutamatergic neurons also showed greater isoform diversity than GABAergic neurons and non-neuronal populations in the study summary.

Functional enrichment connected many DTU events with synaptic and dendritic biology. That result is consistent with the broader observation that full-length isoform regulation in the brain varies across cell types, anatomical regions, and developmental stages. [2]

Key distinction: Gene-level differential expression asks whether total gene abundance changes. DTU asks whether the relative use of transcript isoforms changes. A gene can therefore be non-differential at the gene level but strongly differential at the isoform level.

Cell Composition vs Spatial Context

A spatial difference can arise for at least two reasons. First, two regions may contain different proportions of cell types, each with its own intrinsic isoform preferences. Second, the same cell type may use different isoforms depending on its local anatomical or molecular environment.

The study addressed this problem with a generalized linear model that decomposed isoform-use variation into cell-composition effects and spatial-region effects.

Layer Effects Favor Cell Composition

At the cortical-layer level, isoform differences were more strongly explained by cellular composition. The reported layer–subclass correlation was R = 0.561. For examples such as CD47 and NTRK2, layer-associated patterns could largely be connected to the cell subclasses enriched in those layers.

This is a useful warning for spatial studies: a region-specific molecular signal does not automatically mean that location itself caused the difference. If one region contains more astrocytes and another contains more excitatory neurons, a spatial signal may simply reflect those compositional shifts.

Regional Effects Are More Mixed

At the subregional level, the pattern changed. The study summary attributed 43% of genes to spatial-region effects alone, 13% to cell composition alone, and 44% to both factors. This indicates that anatomical context can contribute information beyond cell identity for a substantial subset of transcript-usage patterns.

RTN1 illustrates the logic. Its distinct isoform-use pattern in particular subregions remained similar across multiple cell subclasses. In that case, where the cells were located appeared to explain more of the isoform pattern than what subclass the cells belonged to.

Figure 2. Multi-scale DTU analysis separates cell-composition effects from spatial-context effects on transcript isoform usage.

For project planning, this argues for integrating cell annotation, tissue-region annotation, and isoform-level statistics rather than treating spatial coordinates as a visualization layer only.

Isoform Switching Across Brain Regions

The study expanded the analysis across seven major anatomical regions captured within the macaque brain dataset: cortex, hippocampus, cerebellum, midbrain, pons, striatum, and ventricle.

The study summary reported 1,445 region-level DTU genes. Among them, 375 genes—about 26%—showed major isoform switching.

What Is Major Isoform Switching?

Major isoform switching occurs when the transcript that is dominant for a gene in one context is no longer the dominant transcript in another. The total gene may remain expressed, but the relative ranking of its isoforms changes.

That makes isoform switching conceptually different from a simple increase or decrease in gene abundance. It asks whether tissue regions are using different transcript architectures from the same genomic locus.

Region-Specific Examples

The cerebellum showed the greatest isoform complexity in the study summary, largely associated with granule-cell biology, while the ventricular region showed the lowest diversity. IDH3B and TUBB2B displayed cerebellum-associated isoform preferences that were reproduced across two independent macaques.

DDRGK1-FS1 again provided a cross-scale example: it was enriched in superficial cortical layers but was barely detected in the cerebellum. This consistency between layer-level and region-level observations strengthens the case for treating transcript structure as a spatial phenotype rather than a purely gene-centric annotation.

Experimental Validation of DTU

Long-read analysis can generate large numbers of candidate isoforms, but novel or region-specific transcripts should not be accepted solely because a computational pipeline reports them. Long-read transcript discovery is sensitive to incomplete molecules, mapping ambiguity, splice-junction errors, and annotation artifacts. Modern curation frameworks therefore evaluate transcript structure, splice junctions, transcript ends, and other quality descriptors before novel isoforms are prioritized. [5]

RNA-ISH and RNA-FISH as Orthogonal Checks

The Fullscope-seq study used RNA-ISH or RNA-FISH to examine selected targets in independent biological samples. DDRGK1 and IDH3B were among the representative DTU genes highlighted in the study summary.

Orthogonal validation serves a different purpose from technical replication. Sequencing asks whether a transcript structure and its spatial assignment can be detected computationally. An in situ assay asks whether the candidate pattern can be observed by an independent molecular measurement in tissue.

A practical validation strategy should therefore prioritize biologically important isoforms: transcripts that define a key spatial conclusion, show large or reproducible shifts, affect relevant domains, or connect to a specific mechanistic hypothesis. It is rarely necessary—or efficient—to validate every discovered transcript.

Validation principle: For spatial isoform studies, reproducibility should be assessed across independent biological samples, while selected high-value isoforms can be checked with an orthogonal assay such as RNA-ISH or RNA-FISH.

Cross-Species Disease Relevance

The final analysis connected macaque DTU patterns with human neuropsychiatric disease genetics. The study integrated human GWAS and TWAS resources covering autism spectrum disorder, schizophrenia, bipolar disorder, neurodevelopmental disorders, epilepsy, developmental delay, and related phenotypes.

The main Nature Methods report states that spatial isoform variations were substantially enriched for neuropsychiatric disorder-associated genes and showed conservation across platforms and species. [1]

Conserved Disease-Associated Isoforms

The study summary reported 283 disease-associated DTU isoforms with conserved intron structures between macaque and human. NTRK2 and GRIN1 were among the genes associated with multiple disease datasets.

Two additional examples illustrate the translational research logic. A neuron-enriched CDC42-FS1 isoform was described as homologous to human CDC42-201 and showed disease-associated expression behavior in bipolar-disorder data. Two MYL6 isoforms corresponded to human transcripts linked to synaptic remodeling and showed differential trends in ASD-related data.

These associations should be interpreted as research evidence, not diagnostic markers. Cross-species conservation can help prioritize transcript structures for mechanistic follow-up, but disease association does not establish causality, clinical validity, or individual risk.

Full-Length vs Short-Read Spatial Data

The choice between short-read and full-length spatial approaches should follow the biological endpoint. Short-read spatial transcriptomics remains highly informative when the primary questions concern gene localization, spatial domains, cell-type composition, or pathway-level expression patterns.

Full-length or long-read spatial profiling becomes more relevant when the endpoint depends on complete transcript structure. Other recent methods have likewise shown that long-read spatial data can resolve splicing, polyadenylation, and isoform switching with spatial or near-single-cell resolution. [3][4]

Choose Short-Read When Gene-Level Mapping Is the Endpoint

If the study needs to identify spatial expression domains, map known cell populations, compare pathway activity, or characterize tissue heterogeneity at gene level, short-read spatial data may answer the question without adding a long-read layer. Researchers can explore broader spatial transcriptomics services when selecting an approach based on tissue type, spatial scale, and study endpoint.

Add Full-Length Data When Transcript Structure Is the Endpoint

Full-length profiling is more justified when the study specifically asks about alternative splicing, transcript architecture, alternative transcription start or termination sites, or region-dependent isoform switching. It can also be useful when a disease-associated locus is already known but the biologically relevant transcript remains uncertain.

Figure 3. Method-selection guide for deciding when full-length spatial transcriptomics adds value beyond short-read spatial data.

Planning a study where gene-level expression may not be enough? Discuss tissue feasibility, spatial resolution, full-length transcript requirements, and the analysis endpoint before deciding whether to combine spatial and long-read workflows.

Planning a Spatial Isoform Study

A robust spatial isoform project should start with the biological question rather than the sequencing platform. The main risk is adding technical complexity without defining which transcript-level result would change the interpretation of the study.

1. Define the Isoform Question

State whether the endpoint is gene abundance, exon usage, full transcript structure, alternative TSS/TES, isoform switching, or a combination. Predefine the tissue regions and cell populations in which the comparison matters.

If the study has only a small set of known transcript targets, a targeted spatial strategy may be more efficient than a discovery-scale full-length workflow. If the isoforms are unknown or transcript structure itself is the discovery objective, full-length sequencing has a clearer rationale.

2. Review Tissue and Spatial Feasibility

Feasibility should consider tissue preservation, section quality, RNA integrity or transcript quality, morphology, spatial coverage, biological replication, and whether the selected platform can resolve the anatomical scale of interest. Archived, degraded, FFPE, or otherwise difficult tissues require sample-specific review rather than assuming universal suitability.

3. Define Long-Read QC Before Sequencing

Useful QC metric types include:

• spatial barcode/CID recovery;
• mapped-read fraction and alignment quality;
• read-length and transcript-coverage distributions;
• full-length or structurally complete transcript fraction;
• splice-junction support;
• known versus novel isoform classifications;
• TSS/TES consistency;
• sequencing saturation or isoform discovery curves;
• replicate concordance at gene and isoform levels;
• filtering criteria for low-support or artifact-prone transcripts.

No single universal threshold applies to every tissue and platform. The acceptable range should be defined from the assay design, reference quality, sequencing technology, and intended downstream comparison. Tools such as SQANTI3 illustrate why transcript-model QC and curation are important before novel isoforms are interpreted biologically. [5]

4. Plan Validation and Replication

Use biological replicates to test whether spatial isoform patterns recur across specimens. For high-value targets, consider orthogonal validation with RNA-ISH, RNA-FISH, targeted sequencing, or another method matched to the transcript feature being tested.

5. Predefine Bioinformatics Deliverables

A spatial isoform project may include:

• raw and processed sequencing files;
• spatial coordinates and tissue-region annotations;
• transcript isoform annotation tables;
• gene- and isoform-level abundance matrices;
• DTU statistics and effect summaries;
• major isoform-switching results;
• spatial maps for prioritized transcripts;
• cell-type or subclass annotations where supported;
• QC summaries and filtering records;
• pathway or gene-set enrichment results;
• figures and tables for prioritized biological findings.

The exact deliverables should match the research question and platform. CD Genomics provides spatial transcriptomics data analysis for research projects that require spatial mapping, annotation, differential analysis, visualization, and interpretation.

Decision checklist: Before adding full-length sequencing, confirm that (1) isoform structure is a primary endpoint, (2) the tissue and spatial assay are feasible, (3) barcode recovery and transcript QC can be measured, (4) biological replication is planned, and (5) the analysis has a clear validation path.

Frequently Asked Questions

From Genes to Spatial Isoforms

Spatial transcriptomics has made "Where is this gene expressed?" a routine research question. Full-length spatial transcriptomics extends the question to which transcript isoform is expressed, in which cell type, and in which tissue location.

The Fullscope-seq macaque brain study illustrates why that extra layer can matter. Transcript usage varied across cortical layers, cell subclasses, subregions, and major brain regions. Some patterns reflected cell composition; others persisted across subclasses and were more strongly associated with anatomical context. Several disease-associated transcript structures also showed cross-species conservation. [1]

The decision to add full-length sequencing should therefore be hypothesis-driven. If the study endpoint is gene localization, short-read spatial data may be sufficient. If the endpoint is transcript structure, isoform switching, or spatial alternative splicing, a full-length layer can expose biology that gene-level counts may compress or miss.

Planning an isoform-centered spatial study? CD Genomics can help evaluate tissue feasibility, spatial platform fit, full-length RNA requirements, QC checkpoints, and bioinformatics outputs for research-use-only projects.

References

1. Liu H, Hong Y, Zhang YS, et al. Full-length single-cell spatial transcriptomics reveals spatial and cell-type-specific transcript isoforms in the primate brain. Nature Methods. 2026.
2. Joglekar A, Hu W, Zhang B, et al. Single-cell long-read sequencing-based mapping reveals specialized splicing patterns in developing and adult mouse and human brain. Nature Neuroscience. 2024.
3. Foord C, Prjibelski AD, Hu W, et al. A spatial long-read approach at near-single-cell resolution reveals developmental regulation of splicing and polyadenylation sites in distinct cortical layers and cell types. Nature Communications. 2025.
4. Fu Y, Kim H, Roy S, et al. Single cell and spatial alternative splicing analysis with Nanopore long read sequencing. Nature Communications. 2025.
5. Pardo-Palacios FJ, Arzalluz-Luque A, Kondratova L, et al. SQANTI3: curation of long-read transcriptomes for accurate identification of known and novel isoforms. Nature Methods. 2024.
6. Su J, Qu Y, Schertzer M, et al. Mapping isoforms and regulatory mechanisms from spatial transcriptomics data with SPLISOSM. Nature Biotechnology. 2026.
7. Gong C, Li S, Wang L, et al. SAW: an efficient and accurate data analysis workflow for Stereo-seq spatial transcriptomics. Gigabyte. 2024.