The visual system of the longest-living vertebrate, the Greenland shark

The Greenland shark (Somniosus microcephalus [Bloch & Schneider 1801]¹; Fig. 1a; Fig. S1) is the longest-living vertebrate on Earth, with an estimated lifespan of up to 400 years². This deep-sea shark inhabits regions from the temperate North Atlantic to the frigid waters of the Arctic Ocean, enduring temperatures as low as −1.1 °C and depths approaching 3000 meters^3,4,5. In the waters surrounding Greenland, their eyes are frequently parasitized by copepods (Ommatokoita elongata), which are thought to obscure vision by attaching to the cornea^6,7,8,9. Despite their slow-moving nature, Greenland sharks are opportunistic feeders, consuming a range of prey, from fish and squid to seals and large carrion, such as whale carcasses^3,5,10. The unique combination of extreme longevity, persistent low temperatures, high-pressure conditions, and parasitised eyes presents unparalleled challenges for its ability to see – raising fundamental questions about the nature and function of the visual sensory system in this enigmatic species.

Vertebrate visual systems have evolved across diverse photic environments, from bright terrestrial habitats to the perpetual darkness of caves and the deep sea. In most vertebrate species, vision relies on two types of retinal photoreceptors: rods, optimized for low-light (scotopic) vision, and cones, specialized for bright-light (photopic) vision¹¹. Rods are highly sensitive, with abundant chromatophore-bound visual pigments (opsins) and an efficient phototransduction cascade, while cones provide saturation-resistant, broad-spectrum photic sensitivity¹². The relative proportion of rods and cones and the spectral tuning of their opsins are primarily shaped by ecological pressures^{13,14,15,16,17,18,19,20,21}. While rod visual pigments (based on the RH1 opsin) are typically maximally sensitive to blue-green wavelengths (460–530 nm), cone visual pigments (SWS1, SWS2, RH2, and LWS) span a broader range of peak sensitivities, from ultraviolet to red light (350–600 nm)²². Deep-sea and nocturnal species often exhibit rod-dominated retinas, sometimes to the complete exclusion of cones, as in certain nocturnal reptiles²³, deep-sea teleost fishes²⁴ and deep-sea sharks^{25,26,27,28,29}.

In the most extreme conditions, vision can be entirely lost. Cavefishes, for instance, have evolved in complete darkness, leading to retinal degeneration, loss of phototransduction gene expression, and widespread pseudogenization of vision-related genes^30,31,32. Given the Greenland shark’s exceptionally dim and potentially obstructed visual environment in combination with its extreme longevity, it has been speculated that it, too, may have lost its ability to see^5,8,33. However, behavioural observations suggest that these sharks may still rely on sight³⁴, and their optic tectum—a brain region that processes visual information—is comparable in size to that of other visually capable elasmobranchs³⁵. Furthermore, the Greenland shark has a tapetum lucidum, a specialized reflective layer behind the retina that enhances photon capture in low-light conditions³⁶.

In this study, to determine whether the Greenland shark possesses a lifelong functional visual system, we conducted a comprehensive integrative analysis incorporating genomics, transcriptomics, in situ hybridisation (RNAscope), ultramicrotomy, chromatin staining, mass spectrometry, in vitro opsin regeneration, and spectrophotometry. Our findings support the presence of a preserved and functional visual system in the adult Greenland shark, which seems well-adapted to extreme low-light conditions. Additionally, transcriptomic data suggest a role for DNA repair mechanisms in maintaining retinal integrity over centuries, potentially contributing to the longevity of vision in the longest-living vertebrate.

Structural and genomic basis of vision in the Greenland shark

Vertebrates typically possess a duplex retina containing both rods and cones. Many deep-sea fish species feature rod-dominated or even pure-rod retinas to enhance scotopic vision^13,28. Using histology, we show that the Greenland shark retina also exhibits a pure-rod structure. In addition, the retina displays several morphological features associated with dim-light adaptation, including densely packed and elongated rods, and comparatively thin inner retinal layers (Fig. 1b), resembling those of other deep-dwelling or nocturnal sharks^28,37,38. All retinal layers are intact in the Greenland shark, including the photoreceptor layer, the outer nuclear layer, the inner nuclear layer, and the ganglion cell layer (Fig. 1b, Fig. S45–S46). That is, even though the retinas inspected were from adult Greenland sharks estimated to be over a century old, there were no obvious signs of retinal degeneration.

To further assess whether the Greenland shark retina shows signs of age-related degeneration, we performed a TUNEL (Terminal deoxynucleotidyl transferase dUTP Nick End Labeling) assay on cryosections of the retina. This assay detects DNA fragmentation, a hallmark of apoptosis and other forms of cell death often associated with degenerative processes in vertebrates^39,40,41,42. As a positive control, sections were treated with DNase I to induce DNA breaks, which produced a robust TUNEL signal as expected, confirming the assay’s sensitivity and functionality. In contrast, no TUNEL-positive cells were observed in the untreated retinal sections from the Greenland shark (Fig. 2e). This absence of detectable DNA fragmentation suggests there is no ongoing DNA damage and cell death within the retina.

The absence of obvious signs of retinal degeneration is remarkable given that even in healthy aging, vertebrate retinas (including those of humans) undergo progressive photoreceptor loss and DNA damage over time⁴⁰. For example, at published rates of age-related rod loss (approximately 0.2–0.6% per year), a human living to 400 years of age would be expected to lose over 50–90% of their rod photoreceptors^43,44,45. Although it is worth noting that metabolic temperature effects may attenuate rates of cellular decline in cold-water species, such as the Greenland shark. Systematic studies of retinal aging in other long-lived species remain scarce. However, existing data suggest that elephants retain stable photoreceptor populations with age⁴⁶, and turtles and tortoises can exhibit slow or negligible senescence (though the retina itself was not examined in this group)⁴⁷. The oldest Greenland shark specimen analysed here was estimated to exceed 130 years of age – making it two- to four-fold older than the eldest individuals examined in elephants, turtles and tortoises. Thus, the Greenland shark provides a striking example of long-term preservation of retinal integrity, supported at both molecular and histological levels, and highlights the extraordinary potential for neuronal maintenance in extreme longevity.

Vertebrate photoreception relies on a cascade of biochemical reactions mediated by numerous phototransduction genes, many of which are specialised for either rods, cones, or non-visual photoreceptive cells, such as intrinsically photosensitive retinal ganglion cells (ipRGCs)^12,48. Deep-dwelling fish species typically rely more on scotopic vision and, hence, the rod-specific phototransduction genes^16,49,50. To examine the phototransduction gene repertoire of the Greenland shark, we generated a draft genome for this species and screened it for phototransduction genes. We retrieved functional copies for the full complement of genes required for rod-based phototransduction (Fig. 1c), including rh1, sag, gnat1, gucy2f, pde6a, pde6b, pde6d, pde6g, grk1, cnga1, gnb5, rcvrn and gngt1. The same set of rod-specific phototransduction genes was found in the genomes of five other shark species (Scyliorhinus canicula, S. torazame, Isurus oxyrinchus, Rhincodon typus, and Chiloscyllium plagiosum; Fig. 1c; Fig. S2), which typically inhabit the upper pelagic zone of tropical to warm temperate seas^{51,52,53,54,55}. Unlike in the other shark genomes inspected, only one gene typically associated with cone phototransduction (grk7) was retrieved from the Greenland shark genome, and recent evidence suggests grk7 may also function in rods in some elasmobranchs⁵⁶. The remaining cone-associated phototransduction genes present in the other sharks were either not found in our draft genome (arrc) or were pseudogenized (rh2, gnat2, pde6c, cnga3, and cngb3) (Fig. 1c). This suggests that Greenland sharks rely on rod-based vision, just like another deep-dwelling shark, the lanternshark Etmopterus spinax²⁵ and many deep-sea teleosts¹³.

We further found that the Greenland shark has a single functional copy of the visual opsin gene rh1, and the non-visual opsins opn3, opn5, rrh and va—a pattern similar to that of other sharks^29,57. Like several other shark species⁵⁸, the Greenland shark lacks a functional copy of the green-sensitive cone opsin gene (rh2), and this gene also shows signs of pseudogenization in the Greenland shark’s genome. Unique among sharks—but similar to what has happened in some cavefish species³⁰—is the pseudogenization of the non-visual opsin gene opn4 in the Greenland shark. In mammals, opn4 encodes melanopsin, a photopigment expressed in intrinsically photosensitive retinal ganglion cells (ipRGCs) that contribute to non-image-forming light responses, including circadian photoentrainment^59,60. However, in dim-light environments, such as the deep sea, rods and cones are primarily responsible for circadian entrainment⁶¹, potentially reducing the functional importance of melanopsin-based pathways. The loss of opn4 in the Greenland shark may thus reflect an adaptation to its deep-sea habitat. Supporting this idea, Greenland sharks do not exhibit a clear circadian rhythm in their diel vertical movement patterns^33,62.

Finally, by computing the ω ratio (i.e., dN/dS) and conducting selection tests, we found that some of the rod-specific phototransduction genes, including rh1, showed evidence of relaxed selection relative to other sharks, suggesting that they are under less pressure to be maintained in the Greenland shark genome compared to other species (Fig. 1e; Fig. S13–S43). However, no loss-of-function mutations were detected in this pathway, which would be expected if they were in the process of being pseudogenized³⁰. Furthermore, some rod genes (e.g., gnat1 and sag) showed evidence of diversifying positive selection, which may suggest adaptive changes related to the optimisation of rod-based vision in the Greenland shark.

Subcellular adaptations and circuitry of the Greenland shark retina

To investigate whether the retinal tissue of the Greenland shark actively maintains nuclear organization, which is indicative of active transcription and hence cellular metabolism, we assessed the presence of histone modifications associated with active (H3K27Ac) and repressive (H4K20me3) chromatin states⁶³. We found clear signals for both histone markers in all nuclear layers of the retina (Fig. 2a), indicating active maintenance of chromatin structure across all cell types. Notably, in rod cells, the active H3K27Ac marker was primarily localized to the centre of the nucleus, while the repressive H4K20me3 marker was predominantly found near the nuclear lamina (Figs. 2b, c). This nuclear organization pattern has previously been associated with diurnality (and therefore, being active in bright light) in mammals⁶⁴. That we now also found this pattern in an elasmobranch species that is predominantly active in dim light suggests that associations between nuclear architecture and diel activity may differ across lineages and ecological contexts. Overall, the presence of both active and repressive chromatin states suggests that the retina of the Greenland shark is indeed metabolically and transcriptionally active.

To assess the integrity of the visual circuit, we employed cell-type specific probes in fluorescent RNA in situ hybridization (RNAscope) to target specific retinal cell types, including rods (using a probe for rh1), GABAergic amacrine cells (gad1), rod bipolar cells (pkcα), Müller glia (glul), retinal ganglion cells (rbpms) and glycinergic amacrine cells (slc6a9)^65,66. Using this approach, we confirmed the presence of all key cell populations necessary for rod-based vision, confirming the integrity of the retinal circuitry in the Greenland shark (Fig. 2d).

Next, we examined the polyunsaturated fatty acid (PUFA) composition of the Greenland shark retina, with a particular focus on omega-3 docosahexaenoic acid (DHA, 22:6n-3) and very-long-chain polyunsaturated fatty acids (VLC-PUFAs) containing 24 or more carbon atoms. These lipids support rhodopsin function via membrane fluidity and pigment packing^67,68,69 and have been associated with counteracting cold-induced membrane rigidity⁷⁰. First, we noted that elovl2, a key gene in synthesis of VLC-PUFAs in zebrafish⁷¹ and humans⁶⁷, is expressed in the Greenland shark retina, specifically in the retinal pigment epithelium (RPE) cells (Fig. 2d). Liquid chromatography-mass spectrometry (LC-MS) analysis revealed that the retina of the Greenland shark contains an exceptionally high proportion of DHA (41%) compared to the bovine retina (26%) (Fig. 2e), and that VLC-PUFAs constituted 45% of total retinal lipids in the Greenland shark, compared to 35% in mammals. Moreover, the dominant VLC-PUFAs of the Greenland shark featured longer carbon chains (36 carbons) compared to the bovine ones (32 carbons; Fig. 2e) or to mouse⁷². The higher proportion of PUFAs and the longer carbon chains in the Greenland shark compared to other vertebrates might represent yet another adaptation to the deep-sea environment, as recently proposed⁷³. Taken together, our data show that the lipid composition of disc membranes in Greenland shark appears to be optimal to support rhodopsin function.

Rhodopsin spectral sensitivity and corneal light transmission in the Greenland shark

In vertebrates, the spectral sensitivity of the visual pigment formed by the rod opsin is typically tuned to the prevailing wavelengths of light in the environment²². This is also true for the Greenland shark. Spectroscopic analysis of purified, in vitro-expressed Greenland shark rhodopsin bound to 11-cis-retinal revealed a maximum absorbance wavelength (λ_max) of 458 nm (Fig. 3b), which is shorter than that of most shallow-dwelling sharks⁵⁸, and even particularly blue-shifted compared to other deep-sea species⁴⁹. Short-wavelength shifting of the rhodopsin λ_max is a typical adaptation found in deep-sea fishes⁴⁹, suggesting adaptive evolution of the Greenland shark rhodopsin for life in the deep sea. Furthermore, the extreme blue-shift observed in the Greenland shark rhodopsin may reflect adaptation to the highly transparent, blue-dominated light environments of deep, high-latitude Arctic waters⁷⁴.

Finally, we investigated whether the parasitized cornea of the Greenland shark allows penetration of light to the retina. Using a SpectralLED tunable light source coupled to an integrating sphere and a spectroradiometer (Fig. 3c), we measured light transmission through fixed Greenland shark corneas compared to fixed human donor corneas (Fig. 3d). All shark corneas analysed had parasites bound to the corneal edges (Fig. S47). We found that human corneas had an average transmittance of 95%, while the six shark corneas ranged from 70 to 100% transmittance. In the blue light range specifically (450–500 nm), human corneas had an average transmittance of 94%, while the shark corneas ranged from 66 to 100% transmittance. These findings suggest that Greenland shark corneas allow light to reach the retina, despite the presence of parasites.

Retinal gene expression in the Greenland shark

Given that we found a fully intact phototransduction gene repertoire and cell circuitry for rod-based vision in the Greenland shark, we sought to determine whether the expression levels of the phototransduction genes were comparable to those in other shark species and thus, biologically relevant. To address this, we sequenced bulk retinal transcriptomes from the Greenland shark and compared them to publicly available transcriptomic data from retinas of five representative shark species: S. canicula, S. torazame, I. oxyrinchus, R. typus, and C. plagiosum. Our analysis revealed that key rod phototransduction genes were expressed at biologically relevant levels in the retina of the Greenland shark, including rhodopsin (rh1), rhodopsin kinase (grk1), arrestin (sag), transducins (gnat1 and gngt1), and phosphodiesterases (pde6a, pde6b, and pde6g) (Fig. 4b). Expression levels of these genes were found to be comparable to those observed in adult specimens of other shark species (S. canicula, S. torazame, and C. plagiosum) and higher than those in species sampled as juveniles (I. oxyrinchus and R. typus) (Fig. 4b). In contrast, we observed either no expression or low expression of cone phototransduction genes in the Greenland shark (Fig. 4c), consistent with a visual system specialized for scotopic vision. Interestingly, individuals from our comparison group that were juveniles (e.g., I. oxyrinchus and R. typus) exhibited relatively lower levels of most phototransduction genes compared to adults from other species. While these patterns may reflect developmental differences, we acknowledge that species and life stage are not fully independent in our dataset, and this may contribute to the observed variation. Overall, our findings suggest that the Greenland shark preferentially relies on rod-based phototransduction for vision. The transcriptomic basis for scotopic vision in this species is fully intact, comparable to other adult elasmobranchs, and well-suited to its dim light environment.

Lastly, we investigated whether efficient DNA repair mechanisms may contribute to the preservation of retinal structure and maintenance of photoreceptor integrity in the retina of the Greenland shark, as has been suggested for other organisms and recently shown in the bowhead whale⁷⁵. Specifically, we focused on the ERCC1-XPF DNA repair complex which is known to have an important role in supporting retinal health and has been linked to retinal aging across a range of eukaryotes^{40,76,77,78,79,80,81}. Notably, UV sensitivity and early-onset vision loss are hallmarks of the human progeroid syndrome, Xeroderma Pigmentosum, caused by mutations in the ERCC1-XPF complex⁸². Similarly, ercc1 knockout mice exhibit accelerated retinal aging and degeneration⁴⁰, highlighting the complex’s role in protecting ocular tissues. We found that shark species with the longest lifespans, including the Greenland shark, retain the ercc1 gene, while shorter-lived sharks lacked this gene^{83,84,85,86,87,88}, and that the Greenland shark exhibits elevated expression of ercc4 (xpf) compared to other sharks (Fig. 3d). This suggests that a robust DNA repair system may help preserve retinal integrity and function over the extremely long lifespan of the Greenland shark. Future studies could help clarify the role of the ERCC1–XPF complex in retinal maintenance in sharks and may reveal additional DNA repair pathways that contribute to visual longevity in this species.

Preservation of vision over centuries

Our findings provide compelling evidence that the Greenland shark (S. microcephalus) retains functional vision, despite extreme longevity, corneal parasitisation, and an environment characterized by minimal light. The rod phototransduction pathway remains intact, and the loss or pseudogenization of most cone pathway genes strongly suggests a reliance on scotopic vision. Active transcription of rod phototransduction pathway genes, supported by RNA-seq, RNAscope and chromatin staining, indicates that the retina is functionally preserved. Furthermore, key visual adaptations revealed by histological analyses and in vitro opsin expression are well-aligned with the deep-sea ecology of this species, including elongated, densely packed rods and a short-wavelength shift in rhodopsin sensitivity. In addition, corneal transmission remains intact despite chronic parasitism. The absence of obvious retinal degeneration in exceptionally old individuals, alongside the preferential retention and elevated expression of DNA repair genes linked to retinal degeneration (ercc1, ercc4), suggests a potential mechanism underpinning their long retinal health span. Together, these findings highlight the extraordinary adaptability of vertebrate sensory systems in extreme environments and the remarkable preservation of organ function over hundreds of years.

Animal tissue collection and preservation

The Greenland sharks used in this study were caught between 2020 and 2024 using scientific long lines off the coast of the University of Copenhagen’s Arctic Station on Disko Island, Greenland (69°15’N, 53°34’W). All sampling was carried out in accordance with laws and regulations under a permit to collect Greenland sharks for scientific purposes. The work was carried out with authorisation from the Government of Greenland under permits (2020-26794, 2022-24744, 2023-6108, 2024-119) from the Ministry of Fisheries, Hunting & Agriculture and a non-exclusive licence (G24-051) for the utilization of Greenland genetic resources issued by the Ministry of Foreign Affairs, Business and Trade. Individuals were euthanized immediately after capture by direct spinal cord transection. Total body length (TL) was measured, and eyecups were dissected and either fixed whole in 4% paraformaldehyde [PFA; 4% (w/v) PFA in 0.01 M phosphate-buffered saline] or the retina was removed and fixed in RNAlater or 100% ethanol. Ages were estimated from TL based on the values provided in Nielsen et al. (2016), which were calculated using the von Bertalanffy growth model². Details of all individuals can be found in Table S1.

The bovine eyes used in this study (n = 5) were obtained from a commercial slaughterhouse. Each eye was dissected to remove the anterior segment, lens, and vitreous. The retina was then carefully separated from the underlying eyecup and stored at –80 °C until further processing.

Human donor tissue collection and preservation

The human donor eyes used in this study were obtained and managed in compliance with the Declaration of Helsinki. De-identified eyeballs from a 50-year-old female donor were obtained from the Willed Body Program at the University of California, Irvine (Irvine, CA) within 12 h postmortem, in compliance with ethical requirements of the program. The program is registered with the respective state boards to facilitate the distribution of deceased human remains for research purposes. Prior to tissue processing, serological testing was performed on donor blood to exclude samples positive for blood-borne pathogens. Whole eyes were dissected, the anterior segments were further isolated and 5 mm bio punches were taken for further analysis. All samples are preserved in the laboratory of Dorota Skowronska-Krawczyk.

Retinal histology

Retinal morphology was assessed for one PFA-fixed eye each from three adult specimens of S. microcephalus. Whole, enucleated eyes were post-fixed in 2.5% glutaraldehyde and 2% PFA in 0.1 M PBS, progressively dehydrated in increasing concentrations of ethanol, infiltrated with EMbed-812 resin and polymerized at 60˚ C for 48 h. For light micrographs, 1 μm-thick radial sections were cut on a Reichert-Jung Ultracut E ultramicrotome, deplastified, stained with epoxy tissue stain (Electron Microscopy Sciences; toluidine blue and basic fuchsin), and imaged under brightfield illumination on a Leica DM4 microscope.

Additionally, hematoxylin and eosin (H&E) staining was performed on 4 μm-thick paraffin sections from a separate eye. Sections were processed following a standard protocol⁸⁹: deparaffinized in xylene, rehydrated through a graded ethanol series to distilled water, stained with hematoxylin for 5 min, rinsed in running tap water, differentiated in 1% acid alcohol, and blued in an alkaline solution. Counterstaining with eosin was carried out for 1–2 min, followed by dehydration through graded ethanol, clearing in xylene, and mounting.

Whole-genome sequencing

A draft genome was sequenced for the Greenland shark, S. microcephalus. Briefly, DNA was extracted from frozen tissue from two adult retinas using Zymo Quick-DNA Miniprep Plus kit. DNA quality was assessed on an Agilent TapeStation and library preparation and sequencing was performed by the Department of Biosystems Science and Engineering (ETH Zurich). Individual libraries were prepared using NEBNext Ultra II FS DNA Library Prep kit (Illumina) with 150 bp insert size, pooled, and sequenced as 150 bp paired-end reads on an Illumina NovaSeq6000 with an SP flow cell (300 cycles). Quality control and adapter removal were performed with fastp (v.0.23.4)⁹⁰, filtered reads were assembled using ABySS (v.2.3.1) using a k-mer size of 80 and k-mer coverage threshold of two⁹¹ and the completeness of the assembly was assessed using BUSCO (v.5.4.5) using the vertebrata odb10 database⁹² (see Table S3 for assembly statistics).

Chondrichthyes phylogeny

A species tree was constructed for 17 species in the class Chondrichthyes (see Table S2 for accession numbers). All genome assemblies available for Chondrichthyes were downloaded from the NCBI Genome database. Completeness of each genome was assessed using BUSCO (v.5.4.5) using the vertebrata odb10 database and assemblies with completeness scores over 80% were retained (14 species). An additional three species were included in the phylogeny: S. microcephalus and another two species which were used for phototransduction gene mining and expression analyses (S. torazame, 76.0% complete; R. typus, 75.6% complete). All single-copy orthologs (obtained from BUSCO analysis) present in at least half of the species were aligned using MUSCLE (v.5.1)⁹³, alignments were trimmed using trimAl (v.1.4.1)⁹⁴, and maximum likelihood trees were made for each ortholog using IQ-TREE (v.2.0)⁹⁵. An unrooted species tree was estimated by inputting concatenated gene trees into ASTRAL (v.5.7.8)⁹⁶, and the tree was dated and rooted using the least-squared method in IQ-TREE and visualised in iTOL (v.6.9.1)⁹⁷.

Phototransduction and DNA repair gene mining

Phototransduction and DNA repair genes were mined from the newly generated draft genome assembly for S. microcephalus and from published genome assemblies for five other shark species (S. canicula, GCF_902713615.1_sScyCan1.1⁹⁸; S. torazame, GCA_003427355.1_Storazame_v1.0⁹⁹, R. typus, GCF_021869965.1_sRhiTyp1.1¹⁰⁰; I. oxyrinchus, GCA_026770705.1¹⁰¹; and C. plagiosum, GCF_004010195.1¹⁰²), which were selected because they also had publicly available RNAseq data. Genome annotations were obtained for all NCBI genomes, except for I. oxyrinchus. The I. oxyrinchus genome assembly was annotated by inputting publicly available RNAseq data for this species (see Table S2 for NCBI accession numbers), protein annotations for the four other shark species, and Vertebrata orthologs from OrthoDB (v.12.0)¹⁰³ into BRAKER (v.3)¹⁰⁴.

For S. microcephalus, phototransduction and DNA repair gene coding sequences (CDS) were mined using a combination of exon prediction and mapping guided by publicly available reference sequences from NCBI. For the opsin genes, we specifically searched for all known genes (including rh1, sws1, sws2, lws, and rh2) using multiple reference species. For the exon prediction approach, scaffolds of interest in the genome assembly were identified using TBLASTN (v.2.11.0)¹⁰⁵ and EXONERATE (v.2.4.0)¹⁰⁶ was used to predict exons on those scaffolds similar to reference sequences. If the exon prediction approach did not yield a CDS, or if the CDS was fragmented, the CDS was retrieved or completed using a mapping approach. Briefly, raw genomic reads were mapped to reference sequences using HISAT2 (v.2.2.1)¹⁰⁷ and the consensus sequence was extracted in Geneious Prime (v.2022.2.2; Biomatters Ltd).

Phototransduction and DNA repair genes were also mined from another five shark species with publicly available genome assemblies. For S. canicula, R. typus and C. plagiosum, functional genome annotations were available and thus, genes were mined by filtering the annotation (GFF) file to keep only matched gene names and AGAT (v.1.4.1)¹⁰⁸ was used to extract the CDS for the longest isoform per gene. For the two species without functional genome annotations (S. torazame and I. oxyrinchus), all genes were mined using the exon prediction approach described above and annotation files were manually edited to incorporate mined genes.

The identity of every gene mined was confirmed phylogenetically. To that end, gene trees for each clade of interest were generated by downloading reference sequences for outgroup species (Homo sapiens, Mus musculus and Danio rerio) and other chondrichthyan species from NCBI, aligning these with mined genes using MUSCLE, trimming alignments using trimAl, generating maximum likelihood phylogenies using IQ-TREE and rooting in iTOL (Fig. S2-S12).

Selection tests on phototransduction and DNA repair genes

The selective forces acting on the newly mined genes were assessed for S. microcephalus. For each gene clade, the protein sequences of the Greenland shark were aligned with those from other species in Chondrichthyes using MUSCLE (after removing stop codons and frameshifts present in pseudogene sequences). This protein alignment was then converted to a codon alignment using trimAl and used to construct a gene tree for each clade using IQ-TREE. The codon alignment and the corresponding gene tree were then used to compute maximum likelihood estimates of ω with PAML. Three branch models were used: (i) a free-ratio model, which allows a different ω value per branch; (ii) a two-ratio model assuming one ω for the Greenland shark branches and one ω for all other branches; and (iii) a null one-ratio model assuming that every branch has the same ω ratio. For all genes, the best model was always the free-ratio model (Model 1). The choice between Model 2 and Model 3 was always made using the mean of a likelihood ratio test, using the χ2 distribution with one degree of freedom. Finally, for each gene clade, RELAX¹⁰⁹ and aBSREL¹¹⁰ was implemented via the HyPhy framework (v.2.5.63)¹¹¹ to look for signs of relaxed selection or positive selection in the Greenland shark genes. Greenland shark branches were assigned as “test” branches, while all other branches were assigned as background branches, in sequential runs of RELAX and aBSREL.

Rhodopsin absorption spectrum measurements

The peak spectral sensitivity of the S. microcephalus rhodopsin (RHO) was assessed using in vitro protein regeneration and spectrophotometry (Fig. 3a). In brief, HEK293S cells were transfected to express Greenland shark RHO modified to contain the 1D4 epitope derived from the C-terminus of bovine rhodopsin (TETSQVAPA). The plasmid, pRP[Exp]-CBA-[SharkRH1], was constructed and packaged by VectorBuilder (vector ID: VB241011-1418sns). After 48 h, cells were pelleted down for 1D4-affinity chromatography. Immunoaffinity 1D4 resin was prepared by conjugating purified, anti-Rho antibody (1D4) to CNBr-activated Sepharose 4B beads (Cytiva). Pelleted cells were homogenized, and RHO pigments were reconstituted with 40 µM 11-cis-retinal for 1 h at room temperature to regenerate RHO from apo-opsin. Then, 10% DDM was added to solubilize the membrane-enriched pellet, followed by 1 h incubation at 4 °C and centrifugation (21,300 x g) for 5 min at 4 °C. The supernatant was then filtered through a 0.22 µm polyethersulfone membrane and incubated with 250 µL of 1D4-resin for 1 h at 4 °C. Rho-1D4-resin mixture was loaded onto a centrifuge column, washed in a buffer containing 50 mM HEPES (pH 7.5), 0.25 M NaCl, and 10% DDM, and eluted with C-terminal nonapeptide (synthesized by GenScript, NJ, USA) overnight at 4 °C. Absorption spectra were recorded using a Varian Cary 50 Scan UV-Vis spectrophotometer (Varian Australia Pty Ltd). The unbleached sample was used as blank, after which the sample was bleached for 10 min with a white-light, 875-Lumens bulb. The difference in absorption spectrum was then recorded.

Nuclear chromatin staining and quantification

Eyecups were fixed in 4% PFA overnight, cryoprotected by immersion in a sucrose gradient (10% and 20% sucrose for 1 h at room temperature, and 30% sucrose overnight at 4 °C), embedded in Tissue-Tek OCT (Sakura, Torrance, CA) and frozen on a conductive metal block placed on dry ice. After cryosectioning, sections were blocked in 5% bovine serum albumin (BSA), 0.3% TritonX-100 for 1.5 h at room temperature to minimize nonspecific binding. Sections were then incubated with primary antibodies (mouse anti-trimethyl histone H4 (sc-134216), 1:200 and rabbit anti-H3K27ac (ab-4729), 1:200) diluted in 5% BSA and 0.1% TritonX-100 overnight at 4 °C. Following three washes with PBS, sections were incubated with fluorescently-labeled secondary antibodies diluted in 5% BSA and 0.1% TritonX-100 for 1 h at room temperature. Following three washes with PBS, nuclei were counterstained with Hoescht 33342 (Thermo), and sections were mounted using ProLong Gold Antifade (Thermo). Immunostained sections were imaged on a Zeiss LSM900 confocal microscope with Airyscan 2 at 40X magnification. Quantification of nuclear chromatin distribution was performed using ImageJ. For each cell, two straight lines were drawn across the nucleus, and intensity profiles for the H4K20me3, H3K27ac, and Hoechst channels were measured. The intensity profiles were normalized within each channel and plotted against the normalized distance along each line.

Fluorescence in situ hybridization

In situ hybridization was performed using the RNAscope^® Multiplex Fluorescent Assay v2 (ACD Diagnostics) following modifications¹¹². Briefly, frozen histologic sections of fixed shark eyes were pretreated as per the manual using hydrogen peroxide and target retrieval reagents, including protease IV. Probes were then hybridized according to the protocol and then detected with TSA Plus^® Fluorophores fluorescein, cyanine 3, and cyanine 5. Sections were mounted with Prolong Gold Antifade (Thermo Fisher) and imaged (Keyence BZ-X700). Probes specific for mouse and human transcripts were designed by the manufacturer (see Table S5).

TUNEL assay

DNA fragmentation was detected using the Click-iT™ Plus TUNEL Assay Kit, Alexa Fluor™ 488 (Thermo Fisher Scientific, C10617), following the manufacturer’s protocol. As a positive control for DNA fragmentation, selected sections were treated with 1U DNase I for 30 min at 37 °C. Slides were counterstained with Hoechst 33342 and images were acquired using a Zeiss LSM900 confocal microscope with Airyscan 2 at 20X magnification.

Fatty acid analysis

Lipid purification and fatty acid quantification were performed following the method of Gao and colleagues¹¹³. Briefly, lipids from B. taurus (n = 5) and S. microcephalus (n = 2) retinas were extracted according to the methodology of Bligh and Dyer¹¹⁴. Specifically, the tissue was homogenized in 200 μL water, transferred to a glass vial, and 750 μL 1:2 (v/v) CHCl₃: MeOH was added and vortexed. Then, 250 μL CHCl₃ was added and vortexed. Finally, 250 μL ddH₂O was added and vortexed. The samples were centrifuged at 3000 RPM for 5 min at 4 °C. The lower phase was transferred to a new glass vial and dried under nitrogen stored at -20 °C until subsequent lipid analysis. The total fatty acids were released through acid hydrolysis and extracted with hexane, then the sample was evaporated under nitrogen and stored at -20 °C until subsequent lipid analysis. LC-MS/MS separation of VLC-PUFAs was achieved on an Acquity UPLC^® BEH C18 column (1.7 μm, 2.1 × 100 mm, Waters Corporation). The Q Exactive mass spectrometer (Thermo Fisher Scientific) was operated in a full MS scan mode (resolution 70,000 at m/z 200) in negative mode. For the compounds of interest, a scan range of m/z 250–800 was chosen. The identification of fatty acids was based on retention time and formula.

Corneal transmittance

We measured light transmission through 5 mm diameter bio-punches of six corneas from five Greenland sharks and a pair of corneas from a human donor for reference. Using a SpectralLED tunable light source coupled to an integrating sphere and a spectroradiometer (PhotoResearch-655 SpectraScan®), we measured spectral radiance of samples between wavelengths 425–600 nm as shown in the optical setup in Fig. 3c. Transmittance was calculated as a % representing the ratio between the radiance measurements of corneal samples to reference radiance (no corneas).

Retinal transcriptome sequencing

Retinal transcriptomes were sequenced for three individuals for S. microcephalus. Briefly, total RNA was extracted from one RNAlater-fixed retina from each of three adult sharks using the QuickRNA Miniprep kit (Zymo). RNA quality was assessed on an Agilent TapeStation and library preparation and sequencing was performed by the Department of Biosystems Science and Engineering (ETH Zurich). Individual libraries were prepared using TruSeq stranded total RNA ribo-zero gold kit (Illumina), pooled, and sequenced as 100 bp paired-end reads using BRAVO Sequencing on an Illumina NovaSeq6000 with an S4 flow cell (200 cycles). Quality control and adapter removal was performed with fastp (v.0.23.4).

Phototransduction and DNA repair gene expression

Phototransduction and DNA repair gene expression was quantified for the Greenland shark and five other shark species for which retinal transcriptome data were available on NCBI (see Table S2 for accession numbers). For the five other species, transcriptomes were pre-processed using fastp and mapped to annotated genomes using STAR (v.2.7.10b)¹¹⁵ with –outFilterMultimapNmax 1 –outFilterMatchNminOverLread 0.4 –outFilterScoreMinOverLread 0.4 options, and mapped singletons were filtered out for paired-end data. Read counts were performed using the HTSeq-count script from the HTSeq framework (v.2.0.2)¹¹⁶ and were used to calculate TPM values for all genes. For the genes of interest, TPM values were plotted against the proportion of reads mapped (calculated as the number of reads mapped divided by the total number of reads in the transcriptome) in R (v.4.4.0)¹¹⁷ and a linear regression was performed to generate an equation that describes the relationship between the two variables for each gene (Fig. S44).

For the Greenland shark, an annotation file was manually generated for all genes that were mined, and this was used to guide mapping against those sequences using STAR. Similar to the other species, mapped singletons were filtered out and read counts were performed using HTSeq-count. The equations generated using the other five shark species were used to extrapolate TPM values from the proportion of mapped reads for each gene expressed in the Greenland shark.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.