Single-Cell vs Single-Nucleus RNA-Seq in Donor Retina

Single-Cell vs. Single-Nucleus RNA-Seq in Donor Retina

The ophthalmic research community has, for years, operated under a quietly convenient assumption: that single-cell RNA sequencing — scRNA-seq — is the gold standard for transcriptomic profiling of human retinal tissue, and that anything less than intact, viable whole cells represents a methodological compromise. It is a tidy narrative. It is also increasingly difficult to defend.

When the tissue in question is a donor eye — retrieved hours post-mortem, cold-shipped through an eye-bank pipeline, or taken from a repository after prolonged storage — the idea that researchers can reliably dissociate living, functionally representative retinal neurons starts to look less like best practice and more like a laboratory ideal. Single-nucleus RNA sequencing, snRNA-seq, does not solve every problem. It does, however, change which problems are tractable.

That distinction matters for snRNA-seq vs scRNA-seq donor retina profiling. The choice determines which cell types remain visible, which transcripts are retained, and which populations disappear during processing. It also determines how much confidence researchers should place in a donor-retinal atlas assembled from tissue that was never collected under the conditions required for a perfect whole-cell preparation.

If ocular biobanking is going to support functional genomics in glaucoma, inherited retinal dystrophies, age-related degeneration, and other disorders, the question is not which method wins in the abstract. The more useful question is narrower: which method best matches the tissue, the preservation history, and the biological question?

The Mechanics of Transcriptomic Capture: Whole-Cell vs. Nuclear Isolation

The conceptual difference sounds straightforward. scRNA-seq profiles RNA from intact, dissociated single cells. Tissue is enzymatically and mechanically broken apart, individual cells are separated, and the captured material includes cytoplasmic mRNA alongside nuclear transcripts. In principle, this gives a broad view of the transcriptome as it existed inside the cell at the time of dissociation.

snRNA-seq takes a different route. The cell membrane is lysed, the nucleus is isolated, and the assay profiles the RNA retained in that compartment. That includes pre-mRNA and nascent transcripts containing intronic sequences, as well as mature nuclear RNA that has not yet been exported or degraded. Cytoplasmic transcripts are underrepresented by design.

The difference is not merely a matter of sample preparation. It changes the molecular composition of every captured unit.

Whole-cell protocols generally produce higher per-cell UMI counts and detect more genes per cell because they sample both nuclear and cytoplasmic compartments. For questions that depend heavily on mature cytoplasmic transcripts, that additional coverage can be valuable. It can also make scRNA-seq data appear richer when the comparison is reduced to the number of detected genes or UMIs.

But those metrics are meaningful only when the cells being measured are the cells the tissue actually contained.

A retina is not an easy tissue to dissociate. Photoreceptors are structurally specialized, retinal ganglion cells are vulnerable and relatively sparse, and neuronal processes extend through dense layers and synaptic networks. Enzymatic digestion and mechanical trituration can break those structures apart unevenly. Some cells survive as recognizable units; others fragment, lose their cytoplasm, or fail to pass through filtration and quality-control steps.

Higher transcript counts per captured unit do not compensate for a population that the protocol systematically failed to capture.

Nuclear preparations surrender part of the cytoplasmic transcriptome, but they also avoid the requirement that a post-mortem cell remain intact and viable long enough to be isolated. That trade-off is central to any single nucleus vs single cell RNA-seq postmortem eye comparison.

Nuclear RNA is not immune to degradation, and nuclei are not automatically clean or representative. A preparation can contain ambient RNA, damaged nuclei, doublets, and variable levels of cytoplasmic contamination. The point is not that snRNA-seq produces an untouched record of the living retina. It produces a different record — one that is often more compatible with the physical state of archived ocular tissue.

The two assays should therefore not be ranked on a single axis called quality. They measure overlapping but non-identical molecular compartments. A fresh, carefully handled specimen may justify whole-cell sequencing. A frozen donor retina may make nuclear isolation the more defensible starting point, even if the resulting libraries contain fewer UMIs per unit.

Overcoming Post-Mortem Limitations in Ocular Tissue Banking

Eye banks do not collect research tissue under the same conditions as a laboratory collecting a fresh experimental specimen. Donor eyes arrive after variable post-mortem intervals and may pass through logistical steps designed primarily for clinical tissue handling. The research portion may be dissected later, divided across projects, and frozen for future use. Some samples have well-documented recovery and storage histories; others carry less complete metadata.

That history affects what scRNA-seq can realistically recover. After death, cellular energy production stops, membrane integrity changes, and the distribution of RNA between cytoplasm and nucleus begins to shift. During processing, fragile retinal neurons may be lost before they ever enter a droplet or capture well. Freezing can preserve tissue for later analysis, but it does not restore the conditions required for viable whole-cell dissociation.

This is the central advantage of snRNA-seq in ocular tissue banking. The method can be applied to frozen, archived, and post-mortem material without first turning the tissue into a suspension of living whole cells. Researchers can homogenize the specimen, isolate nuclei, assess their quality, and proceed with transcriptomic or multiomic profiling even when the original cells are no longer recoverable as intact units.

For a biobank, that is more than a technical convenience. It changes the value of the collection.

A repository of frozen donor retinas may contain regionally defined samples, disease-associated tissue, matched clinical metadata, or specimens collected under conditions that cannot be recreated prospectively. If the molecular workflow requires fresh viable cells, much of that inventory remains inaccessible or produces a highly selective view of the tissue. Nuclear sequencing makes retrospective profiling more practical, although it does not eliminate the need for careful sample annotation and quality control.

The same logic applies to post-mortem eye comparisons across donors. snRNA-seq can reduce one major source of failure — the demand for intact whole-cell recovery — but it cannot erase differences in ischemic interval, storage temperature, dissection quality, or tissue region. Those variables still need to be recorded and modeled. A nuclear assay is more tolerant of compromised tissue; it is not indifferent to tissue history.

The molecular trade-off should be stated plainly:

The correct interpretation is not that nuclear sequencing preserves everything. It preserves enough of a different compartment to make samples usable that would otherwise be difficult to profile. For frozen or archived donor tissue, snRNA-seq is generally preferred because it fits the material rather than demanding that the material behave like a fresh surgical specimen.

That preference should remain qualified. If a fresh specimen is available and the research question depends on cytoplasmic RNA, scRNA-seq may be the more informative assay. Where both sample quality and resources permit, the strongest design may use the methods together rather than forcing one to answer every question.

Cell-Type Capture Bias: Why Glia and Neurons Respond Differently to Dissociation

Dissociating a retina into single cells is not a neutral act. Enzymatic digestion, mechanical trituration, filtration, and capture all impose selection pressures. The protocol does not simply reveal the original cell composition; it creates a sequence of opportunities for some populations to survive and others to disappear.

This is particularly consequential in the inner retina. Retinal ganglion cells, amacrine cells, and bipolar cells are biologically important but technically vulnerable. Their processes, connections, and layered organization make them difficult to release as intact, cleanly captured units. A low recovery rate can easily be mistaken for low abundance in the tissue itself.

Nuclear isolation changes that selection process. It does not preserve the entire cell, but it can retain nuclei from cells that would be damaged or lost during whole-cell dissociation. Comparative retinal datasets have therefore shown better recovery of several inner retinal neuronal populations with snRNA-seq, including ganglion, amacrine, and bipolar populations.

The glial picture requires more care than a simple claim that one method always captures glia better. Glial cells can be physically robust in dissociation workflows, and whole-cell assays may recover them efficiently. However, comparative evidence indicates that snRNA-seq can enrich fibrotic Müller glia relative to scRNA-seq. That enrichment is not a reason to dismiss the nuclear method; it is a reminder that each protocol produces its own compositional bias.

Müller glia are especially sensitive to tissue state. Their transcriptional profiles can reflect disease, injury, fibrosis, and the stress associated with tissue handling. If fibrotic Müller-glia nuclei are more readily represented in a nuclear preparation, their abundance and gene-expression signatures may become more visible than they would be in a matched whole-cell dataset. That signal may be biologically useful, but it should not be interpreted without considering preservation and processing history.

The comparison is better expressed as a set of tendencies rather than a universal ranking:

The practical risk is obvious. If a donor-retinal atlas is treated as a direct census, differences in cell-type proportions may be mistaken for differences between donors, regions, or disease states. A lower fraction of ganglion cells in scRNA-seq data does not necessarily mean that the donor retina contained fewer ganglion cells. A higher fraction of fibrotic Müller-glia nuclei in snRNA-seq does not automatically mean that fibrosis was more advanced in that donor.

Every retinal transcriptomic atlas is also an atlas of what its preparation allowed to survive.

That is why donor retinal transcriptomics dissociation bias belongs in the biological interpretation, not just in the methods section. Researchers should track the tissue region, post-mortem interval where available, freezing history, dissociation chemistry, nuclei quality, mitochondrial or ambient-RNA signals, and the annotation strategy used to compare cell types across platforms.

The danger is not that one method is biased and the other is objective. The danger is forgetting that both are biased in different directions.

Mapping the Human Retina: Resolution and Cluster Diversity in snRNA-seq Atlases

If snRNA-seq were merely a salvage protocol for degraded tissue, the argument for it would be narrower. It would be useful when scRNA-seq was impossible, but inferior whenever fresh tissue became available. Human retinal atlases have made that hierarchy harder to maintain.

Single-nucleus profiling of healthy human retinal tissue across foveal, macular, and peripheral regions has resolved more than 70 retinal cell types, including 13 bipolar-cell clusters and 39 amacrine-cell clusters. The importance of those results is not simply the size of the final cell-type count. It is the demonstration that nuclei from frozen donor tissue can retain enough information to distinguish fine cellular substructure across anatomically and functionally different regions.

Bipolar cells are not a single homogeneous population. Their transcriptional programs relate to the photoreceptors and circuits they connect, and their disease vulnerability may differ accordingly. The same is true of amacrine cells, whose inhibitory and modulatory roles span a wide range of retinal circuits. When an assay resolves these groups into multiple clusters, it gives researchers a more useful framework for studying regional vulnerability, circuit disruption, and disease-associated changes.

That resolution does not mean every cluster is a fully validated cell type. Clustering is influenced by sequencing depth, reference annotations, batch structure, donor composition, and the statistical choices used to define boundaries. A cluster may represent a stable biological population, a state within a population, or a technical distinction that requires confirmation. Nuclear atlases are powerful, but their labels still need to be tested against known markers, spatial data, morphology, and, where possible, orthogonal assays.

The regional dimension is equally important. Foveal, macular, and peripheral retina differ in cellular composition, metabolic demand, vascular environment, and susceptibility to disease. A donor-eye workflow that can profile each region from archived frozen material creates opportunities that are difficult to reproduce with a fresh-cell-only design. Researchers can ask whether a disease-associated program is widespread or region-specific, whether a neuronal subtype is depleted in one anatomical zone, and whether glial responses differ between central and peripheral retina.

This matters in disorders that do not affect the retina uniformly. Macular disease, for example, cannot be understood through a generic retinal average. Peripheral rod-cone degeneration raises a different set of questions about regional cell composition and stress responses. The ability to work with banked tissue from defined regions can therefore be more important than achieving the highest possible transcript count per captured unit.

A useful interpretation of snRNA-seq atlases rests on three distinctions:

1. Resolution is not the same as completeness. A nuclear dataset can distinguish many populations while still underrepresenting cytoplasmic transcripts and outer-segment-associated biology.

2. Cluster diversity is not automatically biological abundance. The number of clusters reflects both tissue heterogeneity and analytical resolution. It should not be converted directly into a claim about how common each population is in vivo.

3. Cross-platform comparisons require harmonization. A cell type may have different marker visibility in whole-cell and nuclear data. Shared annotations should be built from robust marker programs and, where possible, supported by spatial or histological evidence.

For human ocular tissue procurement, these distinctions have operational consequences. The metadata accompanying a donor sample should make it possible to interpret the assay later, when the immediate project has ended. Region, preservation state, processing interval, and assay type are not administrative details. They are part of the biological description of the specimen.

Complementary Approaches for Multiomic Ocular Discovery

The honest answer is that neither method alone gives researchers the full picture. scRNA-seq and snRNA-seq provide complementary views of donor eyes. Whole-cell sequencing offers stronger access to cytoplasmic mature transcripts when the tissue can support viable dissociation. Nuclear sequencing offers greater compatibility with frozen and archived samples and can improve access to vulnerable neuronal populations.

The choice should therefore follow the specimen and the question.

For archived frozen donor tissue, snRNA-seq is generally the preferred starting point. It avoids the most fragile step in the workflow: recovering intact living retinal cells from material that was never preserved for that purpose. It is not the only possible molecular assay, and it is not automatically superior for every readout, but it is often the most practical and reproducible route to retrospective profiling of banked eyes.

For fresh surgical specimens, organoid-derived tissue, or other carefully controlled material, scRNA-seq remains appropriate when intact cells can be recovered and the research question depends on cytoplasmic transcript coverage. The assay may provide deeper per-cell transcript capture and a closer view of mature mRNA programs. That advantage is real, provided the cell populations lost during dissociation are not the populations central to the study.

For disease questions centered on inner retinal neurons, snRNA-seq may offer a stronger representation of ganglion, amacrine, and bipolar populations, particularly when the tissue is post-mortem or frozen. For questions focused on cytoplasmic localization, mature transcript abundance, or cell states that are poorly represented in the nucleus, scRNA-seq may contribute information that a nuclear assay cannot fully replace.

For atlas construction and ocular multiomics, combining modalities can be more informative than selecting a single winner. A practical strategy may include:

• Using snRNA-seq for frozen donor retina to establish a broad cell-type and regional framework.
• Using scRNA-seq on suitable fresh or experimental material to deepen cytoplasmic transcript coverage.
• Comparing shared cell-type programs rather than raw gene counts when integrating the two assays.
• Treating differences in cell-type proportions as potentially technical until they are supported by histology, spatial profiling, or independent measurements.
• Adding chromatin or spatial information where possible to distinguish stable identity programs from preservation- or processing-associated states.
• Recording procurement and storage metadata at the sample level so later analyses can model post-mortem and banking effects instead of treating them as unexplained noise.

The database layer matters here as much as the sequencing platform. A research biologics database for human ocular tissue should not merely record that a sample is retinal or that it came from a donor eye. It should preserve the context needed to interpret the resulting molecular data: anatomical region, tissue state, processing timeline, storage condition, assay modality, and relevant donor information. Without that context, even a technically strong dataset becomes difficult to compare across studies.

The field also needs to be more disciplined about language. A nuclear profile is not a complete substitute for a whole-cell profile. A whole-cell profile is not a neutral census of the retina. Statements about enrichment, depletion, and cellular representation should be tied to the assay and the sample history. In particular, the direction of Müller-glia enrichment must not be reversed: comparative evidence can show greater representation of fibrotic Müller glia in snRNA-seq relative to scRNA-seq, while scRNA-seq and snRNA-seq may differ in other glial and neuronal recovery patterns.

The methodological choice is therefore asymmetric in practice, even if it is complementary in principle. Fresh tissue can support more than one route, depending on the protocol and the question. Frozen and archived donor tissue narrows the field toward methods that do not require viable whole-cell recovery. That makes snRNA-seq generally preferable for retrospective molecular profiling, not because it is universally better, but because it is better matched to the material most ocular repositories actually hold.

The real paradigm deficit is not in the methods themselves. It is in treating tissue banking realities as an inconvenience rather than as part of the experimental design. A donor retina is a finite specimen with a history. That history influences which transcripts remain, which nuclei can be recovered, and which cell types become visible after sequencing.

The field does not need to declare a winner. It needs to stop asking one assay to stand in for the whole retina. scRNA-seq and snRNA-seq are different windows onto the same tissue. For fresh material, the window may open widely onto cytoplasmic biology. For frozen or archived eyes, nuclear sequencing is often the more reliable way to see the cellular architecture that remains. The strongest ocular multiomic studies will treat that distinction not as a limitation to hide, but as information to design around.