The Measured Disconnect
Yet the Human Protein Atlas places total retinal gene expression at roughly 13,033 protein-coding transcripts—some 65 percent of the entire human protein-coding genome.
The arithmetic alone exposes the weakness of treating one molecular layer as a reliable stand-in for the other. Thousands of transcripts may be detectable in retinal tissue, while fewer than two thousand proteins reliably surface in a single anatomical zone under a particular mass-spectrometry workflow. That gap is not simply technical noise. It reflects post-transcriptional regulation: the biological processes that filter, delay, redirect, modify, and degrade a messenger before—or after—it becomes a protein product.
For anyone building multiomic atlases from donor eye tissue, this discrepancy is not a footnote. It is the central interpretive hazard.
The Disconnect: Post-Transcriptional Regulation in Retinal Tissue
The retina is among the most transcriptionally active tissues in the human body. The Human Protein Atlas identifies 769 genes with elevated expression relative to other tissues, a pattern consistent with the extraordinary metabolic, sensory, and signaling demands of photoreceptors, the retinal pigment epithelium (RPE), and the synaptic circuits of the inner plexiform layers.
But transcriptional output is only the first act in a longer biochemical cascade. The retina's cellular architecture introduces regulatory constraints that are easy to miss when the tissue is reduced to a bulk RNA-seq matrix or a list of detected proteins.
Post-transcriptional regulation operates at several checkpoints:
- mRNA stability: sequence-specific RNA-binding proteins and microRNAs can protect transcripts, target them for degradation, or suppress their translation.
- Subcellular localization: a transcript may be transported to a particular cellular compartment, retained near the nucleus, or concentrated where local translation is possible.
- Translation efficiency: ribosomal occupancy varies between transcripts and can change with metabolic state, stress, light exposure, and circadian phase.
- Protein turnover: even efficiently translated proteins may have short half-lives, while others remain detectable long after the corresponding transcript has declined.
- Post-translational modification: phosphorylation, ubiquitination, cleavage, and other modifications can alter function, localization, stability, or detectability by mass spectrometry.
These checkpoints matter acutely in photoreceptors. The inner segment-to-outer segment axis demands continuous protein trafficking for disc membrane renewal, while the connecting cilium imposes a physical and functional bottleneck on the movement of molecular cargo. Translational activity is not distributed uniformly across the cell. Local protein production, transport, and degradation all shape the final proteomic signal.
A transcript may therefore be abundant in the nuclear RNA pool of a rod photoreceptor yet contribute little measurable protein. It may be sequestered in stress granules, subject to nonsense-mediated decay, translated inefficiently, or outcompeted for ribosomal occupancy by a more efficiently translated message. Conversely, a stable protein may remain detectable after its transcript has fallen below the limit of reliable measurement.
The retina expresses 65 percent of all human protein-coding genes, yet a single anatomical zone detects fewer than 2,000 proteins by mass spectrometry. The gap is not noise; it is regulatory architecture.
The consequence for multiomic researchers is straightforward but easy to violate: transcript abundance in a donor retinal homogenate is a census of mRNA molecules present at the moment of tissue arrest. It is not a prediction of protein abundance, and it is not necessarily a close proxy for it.
The two layers are coupled by biology, but the coupling is loose, nonlinear, and dependent on variables that a single transcriptomic assay cannot resolve. Translation efficiency, protein half-life, post-translational modification, cellular composition, and spatial compartmentalization all intervene between the transcript and the measurable protein.
This is the basis of donor retina transcriptome and proteome discordance. The disagreement is not evidence that one assay has failed. It is often evidence that the assays are observing different stages of the same biological process.
Benchmarking Post-Mortem Intervals: What Holds Up and What Does Not
The practical question for eye banks and research biologics programs is more immediate: does donor tissue still tell a truthful molecular story after the donor dies?
A benchmarking study published in Investigative Ophthalmology & Visual Sciences in 2026 compared donor retinas processed at 12-hour and 18-hour post-mortem intervals (PMI), with ocular cooling initiated within 8 hours of death. Using 16-plex TMT mass spectrometry alongside whole-transcriptome profiling, the researchers quantified 22,446 RNA transcripts and 6,109 proteins across the two PMI cohorts.
The result was notable for its conservatism. Protein abundance showed no statistically significant changes between the two intervals, with adjusted P values above 0.05, and only five transcripts reached the threshold for differential expression.
That finding complicates a common assumption: that every passing hour after death necessarily produces a proportionate loss of molecular fidelity. Within the tested 12-to-18-hour window, and under the stated cooling condition, the measured protein layer remained comparatively stable. That does not mean that donor tissue is biologically unchanged. It means that the particular molecular features measured by this workflow were not substantially different between the two tested intervals.
| Parameter | 12-Hour PMI | 18-Hour PMI |
|---|---|---|
| RNA transcripts quantified | 22,446 | 22,446 |
| Proteins identified by TMT-MS | 6,109 | 6,109 |
| Differentially expressed transcripts | — | 5 |
| Significant protein-abundance changes | — | None (adjusted P > 0.05) |
| Cooling prerequisite | Within 8 hours of death | Within 8 hours of death |
The stability window depends on prompt cooling, a protocol variable shaped by logistics, geography, transport, and the speed of the tissue procurement team. It also does not answer every question about the first hours after death. Short-lived regulatory proteins may degrade before collection. Phosphorylation states in signaling cascades may shift. Transient transcriptional bursts may disappear even when the bulk transcriptome remains largely measurable.
The 18-hour interval was the upper end of the tested window, not a lower bound on an acceptable delay. The result should not be read as evidence that tissue remains equivalent beyond 18 hours, or that any sample reaching that interval is automatically suitable for every assay. It supports a narrower conclusion: when cooling begins within 8 hours, the tested 12-hour and 18-hour cohorts showed substantial similarity in the measured transcriptomic and proteomic features.
That distinction matters when a database combines specimens collected under different procurement conditions. A donor record should not reduce PMI to a single number detached from cooling history. The same elapsed interval can describe materially different pre-analytic conditions if one eye was cooled promptly and another experienced a longer period at ambient temperature.
For ocular multiomics data integration, PMI is therefore not merely a metadata field. It is part of the biological interpretation.
Living vs. Post-Mortem: RNA Velocity and the Loss of Biological State Transitions
If the proteome proves comparatively resilient across the tested PMI range, the same cannot be said for the transcriptome's temporal dynamics.
A single-cell RNA sequencing study of 106,829 human retinal cells compared fresh therapeutic enucleations processed within 10 minutes with early post-mortem samples obtained within 6 hours. The difference was not limited to changes in static transcript abundance. It concerned the organization of transcriptional states over time.
In living tissue, RNA velocity analysis—an inference based on the relationship between spliced and unspliced mRNA—captured coherent directional streams. Cells were actively transcribing, processing, and transitioning between biological states in organized trajectories. Post-mortem tissue, by contrast, showed disorganized velocity fields. The transcripts were still present; their temporal coherence was not.
This distinction is essential. A post-mortem retina can retain a substantial amount of molecular information while losing the dynamic relationships that make that information biologically interpretable. RNA velocity does not simply ask which transcripts are detectable. It attempts to infer where a cell is moving within a state space. Once the tissue has been removed from circulation and normal physiological regulation has stopped, those inferred trajectories can become fragmented or misleading.
A donor retina is a census of molecular endpoints. A living retina is a film. The two are complementary, but they are not interchangeable.
The practical implication is that donor retina tissue, however well preserved, is a molecular snapshot taken at the point of tissue arrest. It can capture steady-state abundance with considerable fidelity, as the PMI benchmarking suggests. But it cannot fully reproduce the dynamic biological processes that distinguish living tissue from preserved tissue.
Those processes include circadian transcriptional oscillations in photoreceptors, microglial surveillance states that respond to synaptic activity, and transient stress-response signatures in RPE cells responding to oxidative load. Some of these states may leave durable molecular traces. Others are defined precisely by their short duration and their dependence on intact physiology.
For researchers building transcriptomic atlases, this is more than a general caveat. It affects what kinds of questions the atlas can answer.
A donor retinal atlas can support questions about cell identity, relative expression programs, disease-associated pathways, and stable molecular differences between regions or donor groups. It is less suited to reconstructing the exact temporal sequence of responses that occurred in a living eye before procurement. A living-tissue atlas and a post-mortem atlas can be aligned conceptually, but they should not be treated as interchangeable observations of the same state.
The distinction becomes especially important when transcriptomic data are paired with proteomic data. Protein stability may make the proteome appear comparatively robust across a defined PMI range, while the transcriptome has already lost information about directional state transitions. A naïve integration method can interpret that mismatch as a technical inconsistency. In reality, it may reflect different rates of biological decay.
Spatial Proteomic Gradients: Foveomacular, Juxta-Macular, Peripheral
Even within a single donor retina, protein distribution is not uniform. The 2018 LC-MS/MS regional profiling study identified a measurable spatial gradient:
- Foveomacular retina: 1,974 proteins detected on average
- Juxta-macular retina: 1,999 proteins detected on average
- Peripheral retina: 1,779 proteins detected on average
The juxta-macular zone marginally exceeded the foveomacular region in total protein identifications, while the peripheral region yielded fewer. But the more important result lies in the pathway-level differences between regions rather than in the raw count of detected proteins.
Metabolic and oxidative-stress pathway proteins showed pronounced spatial variation. That pattern is biologically plausible given the fovea's extraordinary mitochondrial demands, its avascular architecture, its dependence on choroidal diffusion, and the shift from cone-dominated central retina toward a more rod-dominated peripheral mosaic.
The same tissue label—retina—can therefore conceal multiple molecular environments. A foveomacular sample, a juxta-macular sample, and a peripheral quadrant do not represent interchangeable pieces of one homogeneous organ. They differ in cellular composition, vascular context, metabolic load, photoreceptor distribution, and exposure to local disease processes.
For proteomic studies relying on bulk tissue homogenates, this gradient is a confounder that is still too often treated as a sampling detail. Homogenizing the entire retina dilutes foveomacular-specific signals into a heterogeneous protein mixture, potentially masking the spatial signatures most relevant to diseases such as age-related macular degeneration.
Regional dissection is not a luxury of experimental design. It is a prerequisite for proteomic data that means what it appears to mean. At minimum, a study should distinguish whether the sample came from a foveal punch, a macular ring, a juxta-macular zone, or a peripheral quadrant. Without that information, an apparent difference in abundance may reflect geography rather than disease, donor biology, or treatment exposure.
Spatial metadata should also travel with the molecular data. A protein matrix without a clearly defined anatomical origin is difficult to compare across donors and even more difficult to integrate with transcriptomic data generated from another region. The resulting donor eye proteomic-transcriptomic divergence may be partly biological regulation and partly a mismatch in what tissue was actually measured.
The same principle applies within a dissected region. A bulk sample can combine photoreceptors, Müller glia, RPE cells, vascular cells, microglia, and other cellular components in different proportions. A change in protein abundance may therefore reflect altered cell composition rather than altered expression within a particular cell type. Single-cell and spatial methods can reduce that ambiguity, but they do not remove the need for careful sample definition.
Implications for Ophthalmic Multiomics and Therapeutic Discovery
The discordance between retinal transcriptome and proteome is not a methodological inconvenience. It is a biological reality with direct consequences for biomarker discovery, target validation, and the interpretation of disease-associated genetic variants.
Consider the path from a GWAS signal to a functional mechanism. A common variant associated with age-related macular degeneration may map to a gene with robust retinal mRNA expression. But if the corresponding protein is subject to rapid ubiquitin-mediated degradation, translated only under specific metabolic conditions, or modified in a way that changes its detection by mass spectrometry, the transcriptomic signal is real but functionally incomplete.
The variant may genuinely affect transcription while its pathological impact operates primarily at the protein level—through altered protein-protein interactions, aberrant localization, impaired trafficking, or gain-of-function aggregation. None of those mechanisms is guaranteed to appear in an RNA-seq assay.
This is why human retinal mRNA protein correlation should be treated as an empirical property of each gene, cell type, region, and biological condition—not as a universal rule. Some transcripts and proteins will track closely. Others will diverge because their translation is tightly regulated or because their protein products are unusually stable. Still others will appear mismatched because the RNA and protein measurements were taken from different anatomical compartments or at different stages of tissue processing.
A useful integration strategy should therefore preserve the distinction between at least three questions:
1. Is the transcript present?
This is the question most directly answered by RNA sequencing, subject to tissue quality, cell composition, and assay sensitivity.
2. Is the protein present and measurable?
This depends on abundance, extraction, peptide detectability, post-translational modification, ionization behavior, and the limits of the proteomic platform.
3. Is the protein functionally active in the relevant compartment?
Neither bulk transcriptomics nor bulk proteomics necessarily answers this. Localization, modification state, interaction partners, and turnover may determine function more directly than abundance alone.
The distinction prevents a common analytical shortcut: using a strong RNA signal as proof that a protein target is abundant, stable, accessible, or therapeutically actionable. In the retina, those inferences can fail at every stage of the regulatory chain.
The retina's capacity to maintain comparatively stable proteomic measurements across the tested post-mortem intervals, provided cooling is timely, remains a genuine asset for biobanking programs. It expands the logistical window for procurement within the conditions that were actually tested. But it does not restore the temporal dynamics lost at the transcriptomic layer, and it does not eliminate the spatial confounders that affect untargeted proteomic workflows.
For research biologics databases, the implication is practical: specimen metadata must be rich enough to explain apparent discordance rather than merely record that it exists. Cooling initiation, PMI, anatomical region, dissection method, tissue processing, and assay platform all influence whether two molecular layers can be compared responsibly. A database that stores expression values without this context may be searchable, but it is not yet a reliable foundation for multiomic interpretation.
What remains unresolved—and what the field has not yet addressed systematically—is the translational efficiency landscape of individual retinal cell subtypes. How much protein does a single Müller cell produce per transcript, per hour, under normoxic versus hypoxic conditions? How does that ratio change in photoreceptors during light adaptation, in RPE cells under oxidative stress, or in microglia responding to synaptic injury?
Single-cell sequencing atlases have mapped the transcriptomic terrain with extraordinary resolution. Proteomic atlases, constrained by the technical demands of single-cell mass spectrometry, have not kept pace. Bulk measurements provide useful regional context, but they cannot fully reconstruct the relationship between transcript and protein within a specific retinal cell.
Until that gap narrows, the donor retina multiomic story will remain a narrative told in two languages that do not translate cleanly into each other. The eye-bank specimen on the bench may offer up its transcripts with one hand while withholding its proteins with the other—not because the tissue is unusable, but because molecular preservation is selective, spatially uneven, and governed by biology that no single assay can summarize.