Specimens of G. zonipectis were collected during daylight SCUBA dives via the application of rotenone to a targeted shallow reef habitat (8–12 m) in Peava Lagoon, Western Province of the Solomon Islands (−8.784222 degrees S, 158.231345 degrees E). Immediately after collection, the G. zonipectis specimen (AMNH 277097) was placed in a narrow photographic tank and held flat against a thin plate glass (Figures 1A,B). Fluorescent macro images were produced in a dark room by covering the flash with interference bandpass excitation filters (Omega Optical, Inc., Brattleboro, VT; Semrock, Inc., Rochester, NY) to elicit fluorescence. Longpass (LP) and bandpass (BP) emission filters (Semrock) were attached to the front of the camera lens to block the excitation light and record emitted fluorescence. The eel was stored in a liquid nitrogen dry shipper and transported back to the American Museum of Natural History, New York, where it was immediately stored at −80°C. The specimen was then delivered to Baruch College.

Research and collection permits were obtained from the Ministry of Fisheries and Marine Resources (MFMR), and the Ministry of Environment, Climate Change, Disaster Management and Meteorology (MECDM), Honiara, Solomon Islands. Fishes were collected and handled in accordance with AMNH Institutional Animal Care and Use Committee (IACUC) and American Fisheries Society (AFS) guidelines, as established for the safe and humane care and handling of vertebrate animals. Fieldwork was carried out in collaboration with and permitted by the Solomon Islands MFMR and MECDM, and facilitated in the Solomon Islands by the Wildlife Conservation Society (WCS, New York; Munda, Western Province, Solomon Islands).

Gymnothorax zonipectis was identified using the following criteria: Overall body coloration brown. The posterior 2–3 laterosensory pores on the upper and lower jaws are enclosed in vertically oriented white bars that are continuous across the lower jaw. Distinct irregular dark brown marking with pale borders posterior to orbit. Body with oblique pattern of darkly pigmented broken/discontinuous vertical bars (in four longitudinal series) that become more pronounced caudally. Bars become more darkly pigmented/larger with bright white borders on posterior fins (Supplementary Figure 1).

G. zonipectis was dissected using a Lightools Research LT-9900 Illumatool Tunable Lighting System (Lightools Research) to ensure that the samples taken contained fluorescent tissue. Cross-sectional images of specimens were generated using a Zeiss-Axio Zoom V16 stereo fluorescent microscope affixed with a Nikon D4 camera. These samples were homogenized in 1X PBS using a BeadBug homogenizer (Benchmark Scientific) and centrifuged at 15,000 rcf for 10 min.

RNA extraction of the fluorescent tissue from G. zonipectis was performed using the RNeasy Fibrous Tissue Mini Kit from Qiagen. Fluorescent tissue in G. zonipectis was found below the skin of the eel. The location of the fluorescence is shown in a photo of the cross section of the eel in Figure 1E. A 30 mg piece of fluorescent tissue was cut using a scalpel for RNA extraction. The extracted RNA sample (28 μL of 75 ng/μL) was then sent to GENEWIZ, LLC (South Plainfield, NJ) for de novo transcriptome assembly and bioinformatic analysis including a BLAST search was used to search for protein candidates.

RNA Library preparation was performed by Genewiz, LLC. RNA was quantified using Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA, United States). RNA integrity measurement was completed using TapeStation (Agilent Technologies, Palo Alto, CA, United States).

NEBNext Ultra RNA Library Prep Kit for Illumina was used in library preparation using manufacturer’s instructions (NEB, Ipswich, MA, United States). mRNAs were enriched with Oligod(T) beads. Enriched mRNAs were fragmented for 15 min at 94°C. First and second strand cDNA were synthesized. cDNA fragments were end repaired and adenylated at 3′ends, and universal adapters were ligated to cDNA fragments, followed by index addition and library enrichment by PCR with limited cycles.

The Agilent TapeStation was used for library validation (Agilent Technologies, Palo Alto, CA, United States), and quantified by using Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, CA) as well as by quantitative PCR (KAPA Biosystems, Wilmington, MA, United States).

After clustering the library on flowcell, the flowcell was loaded on the Illumina HiSeq instrument (4,000 or equivalent) according to manufacturer’s instructions. The sample was sequenced using a 2 × 150 bp Paired End (PE) configuration. Image analysis and base calling were conducted by the HiSeq Control Software (HCS). Raw sequence data (.bcl files) generated from Illumina HiSeq was converted into fastq files and de-multiplexed using Illumina’s bcl2fastq 2.17 software. One mismatch was allowed for index sequence identification.

Data analysis was performed by GENEWIZ, United States. Sequence reads were trimmed to remove possible adapter and nucleotides with poor quality (error rate < 0.01) using Trimmomatic v.0.36 (Bolger et al., 2014). Then Trinity v2.5, de novo assembler (Grabherr et al., 2011), was used on combined samples per species. One de novo assembled transcriptome was created with a minimum contig length of 200 bp. Transrate v1.0.3 (Smith-Unna et al., 2016) was used to generate statistics for the de novo assembled transcriptome. EMBOSS tool getorf (Rice et al., 2000) was then used to find the open reading frames within the de novo assembled transcriptome. The de novo transcriptome assembly was then annotated using Diamond BLASTx against the NCBI NR database.

The phylogenetic tree is presented in Supplementary Figure 2A. The alignments are presented in Supplementary Figure 2B. Alignments were done using MAFFT v7.487 with the L-INS-i Iterative refinement method and default parameters (Figure 2A and Supplementary Figure 2B; Hofacker et al., 2002; Katoh et al., 2002, 2005; Katoh and Toh, 2008; Tabei et al., 2008; Katoh and Frith, 2012; Kuraku et al., 2013; Katoh and Standley, 2016; Yamada et al., 2016; Rozewicki et al., 2019). The best model of substitution was selected using ModelFinder (Kalyaanamoorthy et al., 2017). For generation of the phylogenetic tree (Supplementary Figure 2A), the following input data statistics were generated: 14 sequences with 146 amino-acid sites. Number of constant sites: 24 (= 16.4384% of all sites). Number of invariant (constant or ambiguous constant) sites: 24 (= 16.4384% of all sites). Number of parsimony informative sites: 91. Number of distinct site patterns: 143. The tree was constructed from 1,000 ultrafast bootstrap trees (Hoang et al., 2018).

Plasmids were ordered from Genscript United States in a pET−24b(+) vector with kanamycin resistance for expression in BL21(DE3) E. coli with an N-terminal 6x histidine tag. Following a chemical transformation, BL21(DE3) were plated on agar containing a 1:1,000 dilution of kanamycin (50 mg/mL). Single colonies were selected and grown overnight in 5 mL LB media containing 5 μL kanamycin at 37°C in a shaking incubator. The resulting culture was transferred to 100 mL LB with 100 μL kanamycin and left to grow at 37°C until reaching an OD₆₀₀ of 0.4. A 1:1,000 dilution of IPTG (0.1 M) was then added, and the culture was left to grow for another 3 h after which the culture was then centrifuged at 3,000 rcf for 30 min.

The pellets were then combined and resuspended into 5 mL of Tris-HCl (50 mM), NaCl (150 mM). Lysozyme (100 μL of 10 mg/mL) was added, and the sample was left at room temperature for 1 h. The sample was then centrifuged at 8,000 rcf for 30 min. The supernatant was purified using Nickel affinity chromatography. The elution buffer contained Tris-HCl (50 mM), NaCl (150 mM), and Imidazole (300 mM), pH 7.4. Protein A₂₈₀ was measured using a Cary60 UV-Vis spectrophotometer. The extinction coefficient was calculated using the ExPasy ProtParam tool (Artimo et al., 2012) to be 21,430 M^–1 cm^–1. Bilirubin was dissolved in NaOH and diluted to 10 μM in 1X PBS (1.19 mM phosphates, 13.7 mM sodium chloride, 270 μM potassium chloride), pH 7.3, for further use in fluorescence assays.

Swiss Model was used to construct a model of GymFP (Guex et al., 2009; Bienert et al., 2017; Bertoni et al., 2017; Waterhouse et al., 2018; Studer et al., 2020). The tool chose WT UnaG as the template (PDB code 4I3B, Kumagai et al., 2013). The model was rendered in PyMol (Schrödinger, 2015).

A Hitachi F-7,000 Fluorimeter was used to collect fluorescence spectra. Spectra were recorded of a 1:1 complex of bilirubin and purified GymFP (2.1 μM of each).

Fluorescence is visible in the tissue of Gymnothorax zonipectis along the entire organism and is localized below the skin (Figure 1). Solubilization of fluorescent tissue from G. zonipectis was completed in 1X PBS (1.19 mM phosphates, 13.7 mM sodium chloride, 270 μM potassium chloride, pH 7.3). Following centrifugation, the fluorescent protein remained soluble in the supernatant. Boiling the homogenized samples fully quenched fluorescence.

We searched the transcriptome for fatty acid-binding proteins. From there, we further narrowed the hits based on the presence of a GPP sequence at residues 58–60 (Gruber et al., 2015; Krivoshik et al., 2020). One sequence was found in G. zonipectis.

Sequences of GymFP, Chlopsid FP I, and UnaG were aligned to identify homologous residues (Figure 2A). Residues in purple are identical between the respective sequences. The GPP sequence motif is shown in green. GymFP is 61% identical to UnaG and 47% identical to Chlopsid FP I, according to the SIM Alignment tool on ExPasy (Huang and Miller, 1991). Alignments and phylogenetic analysis of fluorescent and non-fluorescent FABPs are included in Supplementary Figure 2. The tree illustrates not only the relation of the FPs to brain FABPs, but also the distinction of GymFP as it is phylogenetically distinct from the Anguilla and Kaupichthys FPs.

GymFP is 139 amino acids in length and has a calculated molecular weight of 15.6 kDa, similar to UnaG (Hayashi and Toda, 2009; Kumagai et al., 2013) and Chlopsid FP (Gruber et al., 2015; Krivoshik et al., 2020). Bilirubin-bound GymFP has an excitation maximum at 496 nm and a peak emission at 532 nm (Figure 3). The apo protein does not fluoresce. Chopsid FP I has EX/EM of 488/525 nm and GymFP has 496/532 nm, which is similar to that of UnaG (497/532 nm) (Figure 4).