SARS-CoV-2 infects human adult donor eyes and hESC-derived ocular epithelium
Human ocular surface tissue carries SARS-CoV-2 entry machinery and supports productive replication, with the limbus most permissive in both adult donor cells and stem cell derived whole-eye cultures, where infection drives NF-kB chemokine induction and blunted interferon signaling.
- Senior authors
- Benjamin R. tenOever; Timothy A. Blenkinsop
- Correspondence
- Benjamin R. tenOever; Timothy A. Blenkinsop
Research areas & themes
Citation
Eriksen AZ, Møller R, Makovoz B, Uhl SA, tenOever BR, Blenkinsop TA. SARS-CoV-2 infects human adult donor eyes and hESC-derived ocular epithelium. Cell Stem Cell. 2021. 28(7), 1205-1220.e7.
DOI 10.1016/j.stem.2021.04.028. PMID 34022129. PMCID PMC8126605.
Correspondence is addressed to both Benjamin R. tenOever and Timothy A. Blenkinsop, who are also listed as joint on conceptualization, supervision, project administration and funding in the author contributions statement.
One-sentence contribution
Human ocular surface tissue carries SARS-CoV-2 entry machinery and supports productive replication, with the limbus most permissive in both adult donor cells and stem cell derived whole-eye cultures, where infection drives NF-kB chemokine induction and blunted interferon signaling.
Executive summary
Viral RNA had been detected in tears and conjunctiva during COVID-19, but whether ocular surface cells are themselves infected, and which ones, was unresolved. This study combines three human systems. Anterior segments from three post-mortem donors who had tested positive for SARS-CoV-2 were stained for viral antigen, and spike protein was found in the ocular surface epithelium of all three, co-localizing extensively with ACE2 and detectable in both limbal keratin 15 positive and negative cells. Cells cultured from six regions of healthy cadaver globes were then challenged directly. Spike protein appeared in cornea, limbus, sclera and retinal pigment epithelium but not in iris or choroid, while subgenomic nucleocapsid transcript was detected across all six, and ACE2 and TMPRSS2 messenger RNA were highest in the limbus. Whole-transcriptome sequencing of genetically matched cornea, limbus and sclera showed the limbus most heavily infected and dominated by an NF-kB chemokine signature, with limbal identity genes suppressed. A serine protease inhibitor nearly abolished infection. To supply a renewable model, human embryonic stem cell derived SEAM cultures were profiled at single-cell resolution, entry factors were concentrated in conjunctival and limbal clusters, and infection produced a hundredfold increase in infectious virus and a heavily infected limbal cluster. Interferon-stimulated genes were induced across the culture but were lower in the cells carrying the most viral reads.
Scientific context
Aerosol transmission was understood to be the dominant route for SARS-CoV-2, and masks leave the eyes exposed. Viral RNA had been reported in tears and ocular secretions of a small percentage of patients, coronaviral RNA had previously been found in the eye for SARS-CoV-1 and MERS-CoV, and conjunctival inoculation produced respiratory infection in rhesus monkeys and in golden hamsters. On the host side, immunohistochemistry, bulk RNA sequencing and chromatin accessibility work by other groups had shown that cornea, limbus and conjunctiva express ACE2 and TMPRSS2, though TMPRSS2 is low in ocular tissue and other serine proteases are more abundant, and published screens had widened the list of candidate entry-associated genes. What remained unknown was whether virus in tears reflects direct infection of ocular surface cells or drainage from the respiratory tract, whether some ocular tissues are more permissive than others, and what host program infection triggers there. Systemic seeding was considered unlikely by the authors on the grounds that no transfusion-associated transmission had been reported.
Central question
Can SARS-CoV-2 directly infect human ocular surface cells and replicate productively there, which ocular cell types are most permissive and why, and what transcriptional response does infection provoke in those cells?
Experimental strategy
Three complementary human systems each answer a different part of the question and cover one another's weaknesses. Post-mortem tissue from infected donors establishes that viral antigen is present in the eye in real disease, but cannot say where infection started. Cells cultured from dissected regions of healthy cadaver globes allow direct experimental challenge with a defined multiplicity and a side-by-side comparison of six ocular tissues, with cell identity verified by marker staining first, and with genetically matched cornea, limbus and sclera used for the transcriptional comparison so that donor background does not confound it. Because human eyes are available only through donation, a human embryonic stem cell derived self-formed ectodermal autonomous multizone culture provides a renewable model that contains several ocular lineages at once, and single-cell sequencing of that culture allows infected cells, uninfected cells in the same culture, and cells of a separate uninfected culture to be compared, which is what separates cell-autonomous responses from paracrine ones. Productive replication, as opposed to entry alone, is tested by plaque assay on supernatant. Entry route is probed pharmacologically with TPCK, an inhibitor of serine proteases including the TMPRSS family.
Key findings
- Ocular surface tissue from all three SARS-CoV-2 positive post-mortem donors stained positive for spike protein, predominantly in epithelium and absent from corneal stroma, in cells both positive and negative for the limbal marker keratin 15, with strong co-localization with ACE2 and some spike positive cells lacking ACE2 (Fig. 1, Fig. S1 and S2). Nucleocapsid protein was found in p63 positive cells at the limbus and co-stained with ACE2 and TMPRSS2 (Fig. S3). The authors state explicitly that these samples cannot establish whether the eye was infected first or seeded from another organ.
- Cells cultured from six regions of healthy donor globes, with identity confirmed by marker staining (Fig. 2A), showed spike protein after challenge in cornea, limbus, sclera and retinal pigment epithelium but not in iris or choroid (Fig. 2B), while subgenomic nucleocapsid transcript was detected in all six (Fig. 2E). The discrepancy between protein and subgenomic RNA detection is not resolved in the text.
- ACE2 and TMPRSS2 messenger RNA were significantly higher in limbus than in the other tissues, with ACE2 in retinal pigment epithelium the exception (Fig. 2C and 2D). Iris and choroid had lower average expression, though that difference was not statistically significant.
- In genetically matched donor material, sequencing reads mapping to the viral genome were present in cornea, limbus and sclera, and were significantly higher in limbus (Fig. 3A and 3B).
- The host response in all three tissues was dominated by NF-kB-associated genes, including CXCL1, CXCL2, CXCL3 and CXCL6, TNFAIP2, TNFAIP3 and TNFAIP6, IL6 and RELB (Fig. 3F). Chemokine induction was greater in limbus and sclera than in cornea, and SOD2, RIPOR3, TNFAIP3 and C3 rose in all three (Fig. 3F and 3G). Limbus was the only tissue with a large set of genes decreasing on infection (Fig. 3D).
- Pre-treatment with the serine protease inhibitor TPCK almost completely abolished infection of limbal cells (Fig. 3H), which the authors take as confirmation of entry and of serine protease dependence.
- Genes higher in limbus than cornea included ACE2, TMPRSS2, TMPRSS4, TMPRSS11E, SPINT1 and SPINT2, and donor limbal cells expressed TMPRSS4 about elevenfold above corneal cells. The authors propose TMPRSS4 as a candidate explanation for limbal permissivity and say directly that further studies are required to assess it.
- Genes suppressed in infected limbal cells mapped to normal limbal functions including O-glycan biosynthesis, mucin glycosylation, tight junction interactions and glycosphingolipid biosynthesis, and overlapped with gene sets downregulated by SARS-CoV-2 in ferret nasal wash, intestinal organoids, Calu-3 cells and lung (Table S2C).
- Single-cell profiling of the SEAM culture resolved eleven clusters covering surface ectoderm, ocular surface ectoderm, retinal pigment epithelium, neural lineages, dorsal optic cup and periocular mesenchyme, plus some non-ocular populations (Fig. 4 and Fig. S7). ACE2 was expressed above baseline in 11.7 percent of cells with 73.7 percent of those in ocular surface ectoderm clusters, TMPRSS2 in 266 cells with 78.2 percent in the same clusters, and 113 cells co-expressed both, concentrated in the conjunctiva and limbus population (Fig. 5).
- SEAM cultures supported productive infection, with infectious virus in the supernatant increasing a hundredfold over input and plateauing after 24 hours for at least three days (Fig. 6B).
- Integrated single-cell analysis of infected and uninfected SEAM cultures revealed a second limbal cluster present only in the infected culture, in which viral E, M, N and S transcripts were high in nearly all cells (Fig. 6E, 6F and 6H). The authors attribute the separation of this cluster to the abundance of viral and response transcripts.
- NF-kB and TNF-associated genes were elevated in the heavily infected limbal cluster above both uninfected clusters and bystander cells, which the authors interpret as a virus-driven rather than cytokine-driven signature.
- Interferon-stimulated genes rose across the infected culture including in the uninfected limbal cluster, but were lower in the cluster carrying the most viral reads, though still above the uninfected culture (Fig. 6H). The authors read the broad induction as paracrine interferon signaling and the relative deficit in infected cells as viral antagonism, and they note that the pathway as a whole apart from IRF1 and ZFP36 was less induced, suggesting antagonism at a major node.
- Of genes expressed in the SEAM limbus, 87 percent were also expressed in adult limbus, and 80 percent of corneal cluster genes were expressed in adult cornea. Genes downregulated in both systems converged on injury response, immune cell recruitment, cell-cell communication, limbal and conjunctival function, and barrier function (Fig. 7).
Mechanistic model
The study does not establish a mechanism for why the limbus is preferentially infected, and it says so. What it constrains is the following. Entry into ocular surface cells requires serine protease activity, since a broad TMPRSS-family inhibitor blocks infection, and permissivity correlates with expression of ACE2 together with several serine proteases, with TMPRSS4 standing out as elevated in limbus relative to cornea in donor cells and enriched in limbal clusters in the stem cell derived model. That correlation is offered as a candidate explanation and is not tested by loss of function. The host response has two arms whose relationship is interpreted rather than demonstrated. NF-kB-dependent chemokine induction is highest in cells carrying the most viral RNA, which the authors read as driven by something in the virus rather than by soluble mediators, while interferon-stimulated gene induction is broad across the culture but comparatively lower in those same cells, which they read as intracellular antagonism of interferon induction layered on top of paracrine signaling from neighboring cells or from debris. Two alternatives are acknowledged in the discussion, namely that interferon induction may simply not be efficiently inhibited in infected cells, and that NF-kB signaling can itself suppress interferon output. The suppression of limbal identity genes is described as possibly a consequence of the virus commandeering host transcriptional machinery, or possibly part of an injury and epithelial-to-mesenchymal response, and neither is distinguished experimentally.
Conceptual or technical advance
The work places the ocular surface on the map of tissues that can be directly and productively infected by SARS-CoV-2 in human material, and it localizes the vulnerability to a specific anatomical compartment rather than to the eye as a whole. Doing this required a model that could be infected repeatedly, since human eyes come only from donation, and the adaptation of the SEAM whole-eye differentiation protocol for virology, combined with single-cell profiling before and after infection, provides one. The single-cell design also makes a distinction that bulk work cannot, separating what a cell does because it contains virus from what it does because its neighbors do, and that distinction is what allows the interferon and NF-kB arms to be described separately. The finding that limbal identity genes are suppressed connects infection to potential loss of ocular surface function rather than only to viral output.
Relationship to the broader research program
The transcriptional framing here belongs to a line running through the tenOever laboratory's COVID-19 work, in which SARS-CoV-2 elicits a strong NF-kB-driven inflammatory program alongside a muted type I and type III interferon response, established for lung and airway systems in Blanco-Melo and colleagues and cited repeatedly in this paper for direct comparison. The same laboratory's hamster work on ocular inoculation and on nucleocapsid transcript release from dying cells is likewise cited, and the interferon antagonism framework draws on the senior author's review of antiviral defense evolution. Applying that framework to an entirely different tissue, provided and characterized by the Blenkinsop laboratory, is what the collaboration adds. Whether the ocular surface constitutes a meaningful transmission route or a viral reservoir is raised as an open question by the authors and is not answered here. Comparisons across corpus entries on tissue-specific host responses to SARS-CoV-2 would be category 3 synthesis and belong to central assembly.
Related publications
- Blanco-Melo et al. 2020, imbalanced host response to SARS-CoV-2 drives development of COVID-19. Predecessor from the tenOever laboratory, the source of the NF-kB high and interferon low signature that this paper compares against throughout.
- Hoagland et al. 2021, work from the same laboratory on SARS-CoV-2 in golden hamsters including ocular inoculation and interferon dynamics. Predecessor, cited for ocular route infection in animals and for nucleocapsid transcript release from dying cells.
- tenOever 2016, the evolution of antiviral defense systems. Review or synthesis from the same laboratory, cited for interferon as first line of defense and viral antagonism.
- Hayashi et al. 2016, self-formed ectodermal autonomous multizone cultures. Methodological foundation from another group, the differentiation protocol adopted here.
- Collin et al. 2021 and Mencucci et al. 2021, expression of SARS-CoV-2 entry factors in ocular surface tissue. Predecessors from other groups, providing the entry factor expression data this study builds on and compares with.
Limitations and boundaries
The authors list their own limitations plainly. Only three to seven genetically distinct donors were available per ocular cell type, and the three infected post-mortem donors cannot show whether the eye was the portal of entry or was seeded from an established airway infection, so the transmission question the paper motivates remains open. The SEAM cultures are stem cell derived and immature, indicated by low KRT3 and KRT12, and they contain cells at different developmental stages, which may itself contribute to the higher infectivity observed in the limbal compartment given that the limbus is a stem cell niche. Single-cell sequencing of infected cadaver donor cells was not performed, which the authors name as the experiment that would test whether a specific niche with stem cell properties is preferentially infected. Infections are at a single multiplicity of one and largely a single 24 hour time point, with growth followed for three days only in the SEAM model. The TMPRSS4 hypothesis rests on expression correlation, and the protease requirement is established with a broad inhibitor that does not identify which protease matters. The infected SEAM culture and the uninfected comparison culture were differentiated for different lengths of time, 54 days and 31 days respectively, which is a difference between the compared conditions beyond infection itself. Nothing here addresses clinical ocular disease, infectivity of tears, or whether ocular infection contributes to spread within or between people.
Audience summaries
25 words
Human ocular surface cells, especially at the limbus, carry SARS-CoV-2 entry factors and support productive infection, both in donor tissue and in stem cell derived eye cultures.
75 words
Viral RNA had been found in tears, but it was unclear whether eye cells are actually infected. Antigen was present in ocular surface tissue from three deceased COVID-19 patients, and cells cultured from donor eyes supported infection, most strongly in the limbus, the ring of tissue holding the corneal stem cell niche. Stem cell derived eye cultures reproduced this and released infectious virus. Infected cells mounted a strong inflammatory response with weak interferon signaling.
150 words
Combining post-mortem tissue from SARS-CoV-2 positive donors, primary cultures from six regions of healthy cadaver globes, and human embryonic stem cell derived whole-eye cultures, this study shows that human ocular surface cells are directly infectable and productively support SARS-CoV-2. Spike antigen was present in donor ocular surface epithelium co-localizing with ACE2. Among cultured tissues the limbus carried the highest ACE2 and TMPRSS2 expression and the most viral reads, and a broad serine protease inhibitor nearly abolished infection. Transcriptionally, infection drove an NF-kB chemokine program while suppressing genes underlying normal limbal barrier and secretory function. Stem cell derived cultures concentrated entry factors in conjunctival and limbal clusters, released a hundredfold more infectious virus than the input, and yielded a heavily infected limbal cluster in which interferon-stimulated genes were lower than in bystander cells, which the authors interpret as intracellular antagonism superimposed on paracrine interferon signaling. Whether the eye is a genuine entry portal remains untested.
Documented publication relationships
- Imbalanced Host Response to SARS-CoV-2 Drives Development of COVID-19 — predecessor.
- Leveraging the antiviral type I interferon system as a first line of defense against SARS-CoV-2 pathogenicity — predecessor.