to ocean conditions [30][31][32] , and developing coral cell cultures that precipitate aragonite at comparable rates to intact corals 26,33,34 . ere has also been a general proliferation of sequenced coral genomes and transcriptomes (e.g., [35][36][37] ). Roles of several biomineralization proteins have been revealed including the ability of the highly acidic proteins called coral acid rich proteins (CARPs) or skeletal aspartic acid-rich proteins (SAARPs) to precipitate aragonite from seawater 38 , and the high enzyme activity of the coral skeletal carbonic anhydrase STPCA2 39,40 . However, the function of most of the approximately 100 known coral skeletal proteins remain to be established. Some protein roles may be suggested by their persistent interactions with the mineral once formed. For instance, proteins involved in nucleation of aragonite or in adhering amorphous calcium carbonate nanoparticles together toward their recrystallization to aragonite 13,41 , may be more tightly bound within the mineral and may resist degradation. In this context, fossil biomolecular data may help clarify roles of proteins currently of unknown function. us, important information may be forthcoming in addition to the utility of these protein sequences in organismal phylogenetic reconstruction.
To examine the persistence of coral skeletal proteins older than the Holocene epoch, we analyzed in-depth one modern and several Pleistocene Stage 5E Caribbean corals. As aragonite can recrystallize into secondary aragonite or calcite, with a potential loss or degradation of proteins in the process, we determined the samples' mineral integrity by x-ray di raction and inductively coupled plasma mass spectrometry analysis of element/ calcium abundance. We then used racemization analysis of free and hydrolyzed amino acids and protein visualization to establish that any sequenced proteins are likely endogenous rather than modern contaminants. Finally, we sequenced extracted proteins using liquid chromatography with tandem mass spectrometry. is work yielded the oldest known invertebrate protein sequences and suggests that highly acidic proteins resist degradation through intimate interactions with the mineral phase and may be useful targets for further analysis within the invertebrate fossil record.
Five modern and three fossil coral specimens were interrogated for the quality of their mineral preservation (SI Table 1). Fossil corals that were collected in 1975 from exposed Key Largo Formation deposits, with youngest ages of 125 to 138 kiloanna (ka) 42,43 , and later donated to the Natural History Museum of Los Angeles County (NHMLA) were loaned to the authors. X-ray di raction of these fossil specimens showed that Orbicella anularis-2 (herea er: Mann2) has recrystallized to Mg-calcite and calcite, Montastraea cavernosa-1 (herea er: Mcav1) is 70-85% aragonite and 15-30% calcite, while O. annularis-4 (herea er Mann4) is 93-100% aragonite and 0-7% calcite (Fig. 1a-l). A modern O. annularis specimen used for in-family comparison is 100% aragonite (Fig. 1m,n). Of the fossil corals, only Mann4 exhibited Mg/Ca, Sr/Ca, and B/Ca ratios within the range of modern and fossil primary aragonite (Fig. 1o-t). Together, our data suggest that the fossil specimen Mann4 remains mostly primary aragonite.
SDS-PAGE indicated that proteins were among the biomolecules extracted by acid hydrolysis of cleaned fossil skeleton powder for all fossil specimens (Fig. 2). However, these proteins were degraded as evidenced by a bias toward small peptides (i.e., in the low molecular weight area of the gel) in all specimens. Further, fossil corals Mcav1 (70-85% aragonite), Mann2 (fully recrystallized to calcite), and Mann4 (> 90% primary aragonite) exhibited total hydrolysable amino acid (THAA) D/L Asx (aspartic acid plus asparagine) values of 0.634, 0.611, and 0.380, respectively (Table 1). For comparison, THAA D/L Asx of a modern O. annularis skeleton was 0.212.
No coral proteins were sequenced from cleaned skeleton powder of Mann2 (fully recrystallized to calcite) or Mcav1 (75-80% calcite). In contrast, Mann4 (> 90% primary aragonite) yielded six coral proteins within our stringent criteria containing peptides that were sequenced by LC-MS/MS a er trypsin or trypsin-then-GluC digestion; all proteins had either more than one peptide detected at least once, or one peptide detected multiple times (Table 2 and SI Table 2). Five proteins were detected in the acid-insoluble fraction and one was detected in the soluble fraction, with no overlap between the two fractions. is is contrasted by the 61 proteins detected across all solubility and digestion fractions of proteins extracted from the newest growth of a modern O. annularis (SI Table 3). Proteins sequenced from fossil Mann4 skeleton include one of the highly acidic skeletal proteins known as acidic skeletal organic matrix protein (acidic SOMP) 27 or SAARP3 28 in Acropora spp. and P27 in Stylophora pistillata 26 , and also detected here in our modern O. annularis (SI Table 3). Although the Blast2GO annotation calls this protein acidic SOMP-like, we use SAARP3 here to be consistent with the most recently published terminology 28 and because the gene is clearly a sub-group within the CARPs4/5 or SAARPs1-3 gene family (SI Fig. 2). Other fossil proteins sequenced include a coadhesin which was also detected in our modern O. annularis, a lanC-like protein, a polyamine-modulated factor-binding protein, and two uncharacterized proteins.
Sequencing of six fossil coral proteins (Table 2), three of which are also found in modern coral skeleton ( 26,27 , present study), provides further con rmation that coral skeletal proteins are speci c components of the organic matrix embedded within individual aragonite crystals rather than simply cellular contamination 13,44 , as peripheral proteins in the fossil samples would be accessible to degradation processes over the past ~ 100 ka. In the case of SAARP3, strong interactions likely persist between the acidic domains of the protein and calcium atoms in the biomineral 13 . is protein is a member of the CARPs4/5 or SAARPs1-3 family (SI Fig. 2), one member of which has been shown to lead to the precipitation of calcium carbonate from unamended seawater 38 . Further, blasting indicates that the peptide detected in this protein does not have a BLAST hit in Homo sapiens, so it is not a contaminant of the extraction and sequencing. Having shown that the proteins sequenced from the fossil specimen are not likely contaminants (SI Tables 4 and5), we can focus on the remaining fossil proteins. Like di ractograms for powders milled within corallites are images g, i, k while di ractograms for powders milled between corallites are images. XRD patterns for a modern O. annularis are shown for most recent growth (m) and older growth from ~ 10 cm deep (n). Element/Ca ratios for all fossil samples are shown in p, r, t to be compared with values shown in o, q, s from [94][95][96][97][98][99][100][101][102] while such ratios for modern samples were determined from modern LANHM specimens for this study. SAARP3, coadhesin has been sequenced previously from modern coral skeleton [26][27][28] . Coadhesin may persist in a sequenceable form in fossil corals due to its relatively high abundance, as suggested in modern skeleton (SI Table 3). Coadhesin, a transmembrane protein, contains multiple extracellular thrombospondin type-1 repeats, possibly allowing this protein to serve a role in adhesion of calicoblastic cells to organic matrix on and in the skeleton 27 . Further, thrombospondins may form functional triple helices similar to collagen and peroxidasin 45 two proteins known from coral skeleton 26,27 . ey have been implicated in structuring the other framework proteins of the ECM in mammalian osteoblasts 46 and inhibition of osteoblast di erentiation 47 . is structural set-up of the coral ECM is engulfed by the growing mineral where it is preserved for, potentially, 100s ka. Two other proteins, a polyamine-modulated factor-binding protein that may play an adhesion and sca olding role in the biomineralization process, as suggested by the protein's adhesive properties in sperm 48 , and a lanC-like protein which may associate with the cell membrane 49 , were also detected in the fossil O. annularis skeleton but not in the modern one. One uncharacterized protein is similar to a protein previously sequenced from S. pistillata (P16 26 ) while the other is unique to O. annularis skeleton among all coral skeletal proteomes, although orthologs are present in genomes and transcriptomes across many scleractinian taxa (reefgenomics.org, NCBI).
In general, annotation of several of the detected fossil proteins suggests that they were likely involved in adhering coral cells to the growing skeleton and in structuring the physical environment of the calcifying space, bringing them into direct contact with the mineral. While 61 proteins were sequenced from the multiple fractions of modern O. annularis skeleton, only three of these, and six proteins total across all digestion fractions, were detected in fossil skeleton. is is likely due to two issues. As analysis by SDS-PAGE shows, fossil proteins had likely degraded into smaller peptides (Fig. 2). Further, racemization of Arg and Lys would minimize the e cacy of trypsin [50][51][52] , although the stochastic nature of racemization means that, even a er 100s ka, at least half of the enzyme target amino acids should be in a cleavable form. We attempted to overcome diminished e ectiveness of tryptic digestion on fossil samples by further digesting the post-trypsin peptides with GluC, which selectively cleaves a er Asp or Glu. Increased protein sequencing has been observed when using multiple enzymes on modern specimens 53,54 , and we detected two additional proteins above our cuto settings in post-GluC digestions of fossil O. annularis acid-insoluble skeletal protein. Coral skeleton is notoriously low in organic matter abundance with estimates of skeletal organic matter content as low as 0.01% 22 and as high as several percent 55 , so that the combined e ects of low starting material, protein degradation, and racemization of enzymatic digestion targets places severe limits on the sequenceability of fossil coral proteins that would be encountered. Skeletons of invertebrates with comparable amounts of starting organic matter content such as mollusk shells 19 and echinoderm tests and spines 20,21 at 0.3-4% and ~ 0.1% by weight, respectively, may face this issue as well.
In work of this nature, care must be taken to ensure that reported proteins are not modern contaminants 56 ; well-preserved specimens must be chosen and best practices in handling samples must be employed 57 . Our crystallographic and trace element analyses allowed us to select a fossil coral specimen that retained its primary aragonite mineralogy (Fig. 1). Further, in addition to extensively cleaning all skeleton powders, we handled fossil and modern powders separately in age speci c glove bags and months apart, with fossil O. annularis skeletons handled rst. We also examined biochemical signatures of age and persistence of intact proteins.
Amino acid racemization has been used for the past 50 years to study fossil samples ranging in age from 500 to 300,000 years old [58][59][60] . In the present study, Mann4 exhibited THAA D/L Asx lower than similarly-aged Atlantic and Caribbean Pleistocene-aged corals 61,62 . is may be due to the fact that most of the Mann4 THAA Asx was made up of polymerized amino acids (33.7% FAA) whereas FAA accounted for most of the coral THAA Asx in 61 (~ 60% FAA based on their Supplementary Figure EA 1; see our Supplementary document for details on quantitation), with the FAA pool being drawn from hydrolysis of terminal amino acids in degraded proteins for which racemization had already occurred [63][64][65] . In contrast to Mann4, the higher D/L Asx in Mann2 and Mcav1 may be due to degradation of proteins, potentially linked with recrystallization, as observed in protein extracted from the partially recrystallized Mcav1 and analyzed by SDS-PAGE (Fig. 2). is is similar to the process by which Asx racemization is greater when demineralization of bone results in degradation of the collagen triple helix 66,67 . Di erences in protein content between the fossil coral specimens is likely not due to di erences in collection or preservation as documentation for the fossil skeletons indicates that they were collected by the same person at the same time and then stored together in warehouses at the NHMLA a er their donation. All three Pleistocene corals also exhibited relative amino acid molar concentrations similar to archived modern specimens reported here as well as archived material from a modern Porites skeleton and modern and fossil Acropora skeletons for which amino acid relative quanti cation was performed as part of racemization analysis 61,63 (Table 1). Even Mann2, fully recrystallized to calcite, retained a THAA relative composition roughly similar to its modern counterparts with a persistent bias toward the acidic residue-containing amino acid groups, Asx and Glx. However, compared with modern coral skeletons, we observed decreased relative Asx and Ser and increased relative Val in Mann4 and Mcav1, decreased relative Glx in Mann2, and increased relative Ala in all three fossil specimens (Table 1, SI Table 6). e di erence in amino acid composition could be due to loss of highly acidic proteins, of which there are multiple types in coral skeleton 13,41,68 , and which would reduce the remaining Asx pool while increasing the relative abundance of the remaining non-acidic residues. We also observed a bias toward smaller peptides observed by SDS-PAGE (Fig. 2). Together with the non-modern D/L Asx and D/L Glx values, these are indicative of both very old proteins and of protein degradation and loss in the fossil coral skeletons. Further con rmation that we did indeed sequence fossil coral skeletons is provided by degradation signatures in detected peptides. In particular, the detected peptide in SAARP3 in the acid-insoluble SOM fraction is suggested by MS1 data to be deamidated in both asparagines in the peptide. is deamidation is a known degradation feature observed in other paleoproteomic analyses (e.g. 69,70 ). Further e ects of proteins being locked in aragonite crystals may include di erential production of isoaspartate (or gamma glutamic acid) during asparagine (or glutamine) deamidation 71,72 ; unfortunately, the low peptide yield was insu cient to pursue this analysis in the present study. Finally, our phylogenetic analysis shows that SAARP3 is a coral-speci c protein (SI Fig. 2). In sum, we are con dent that the six proteins sequenced from the fossil O. annularis, Mann4, are coral proteins of the same age as the skeleton.
It should be noted that extraction of coral proteins for sequencing is destructive. While owners of gi ed or loaned coral skeletons were informed of this ahead of time and approved such use of the specimens, we remain cognizant that our use of portions of the samples precludes analysis of these portions in the future. We therefore sought to minimize the amount of material used for sequencing. Some modern coral skeletal proteome sequencing has used 10-30 g of cleaned skeleton powder 27,28 , but comparable protein detection can be obtained from approximately 1 g of cleaned modern material 29 . In this study, we found that this smaller amount of skeleton allowed sequencing of some fossil proteins and reinforces that these biomolecules are potentially available for sequencing from invertebrate biominerals aged over 100 ka. More material may yield better detection, and hence more sequenced peptides, in future studies but specimens should be carefully chosen to minimize loss of irreplaceable samples. Further, enzymatic digestion success is likely minimized by amino acid racemization. While we show here that we can successfully sequence several known modern skeletal proteins in fossil coral specimens, future method re nement may consider inclusion of other, non-enzymatic, peptide cleavage methods; this could include the use of cyanogen bromide (e.g. 73 ), although signi cant accumulation of methionine oxidation in fossil specimens would minimize its e ectiveness 74 .
In summary, we show that fossil coral skeletons that retain their primary aragonite mineralogy preserve endogenous proteins that we extracted and sequenced by standard methods, while proteins in recrystallized fossil coral skeleton are too degraded for sequencing. To be retained in the skeleton for over 100 ka, these proteins must have been intimately associated with the aragonite crystal. Our work supports this and pushes back the age at which these phylogenetically informative skeleton biomolecules can be obtained from the invertebrate fossil record.
Colonies of fossil Orbicella annularis and Montastraea cavernosa skeletons were borrowed from the Natural History Museum of Los Angeles (NHMLA) Invertebrate Paleontology Department (Fig. 1). All specimens were originally collected from Pleistocene deposits in the Key Largo Formation (FL) aged 125 to 138 ka 42,75 . ese and modern corals of several species, also NHMLA collections, and the surface layer of a privately-owned O. annularis were analyzed for skeleton integrity (SI Table 1). Slabbed coral fragments were soaked in equal parts 30% hydrogen peroxide and 3% sodium hypochlorite a er Stoll 76 and then ground to 125 µm . Skeleton powder was cleaned three additional times before being dried at 40 °C. Cleaning was su cient to remove contaminant proteins, as determined from phosphate bu ered saline solutions soaked on cleaned skeleton powders and then concentrated on 3 kDa Amicon Ultra centrifugal lter units (Millipore), at the detection level of bichronoic acid assays (SI Table 4), of Stain-Free SDS-PAGE (Bio-Rad) and imaging (SI Fig. 3), and were at concentrations three orders of magnitude lower than that observed in cleaned coral skeleton powder by amino acid analysis 61 . All clean powder was only handled in age-speci c glove bags (i.e. separate bags for fossil and modern), and modern samples were never handled at the same time as fossil samples for any biochemical analysis.
Duplicate milled sub-samples of cleaned skeleton were cleaned a er Stoll 76 and dried at 60 °C before analysis on a Panalytical X'Pert Pro X-ray Powder Di ractometer (Malvern) on a zero di raction background plate. Spectra were analyzed in X'Pert Hi-Score so ware and relative amounts of aragonite and calcite were determined by the Reference Intensity Ratio method 77 using 01-072-1652 calcite and 00-041-1475 aragonite references for all samples. A er x ray di raction analysis, elemental composition of powders was measured on an Element XR HR-ICP-MS ( ermo Fisher) using Cibicides wuellerstor (CAM-wuell) as a consistency standard 78 .
Amino acids for racemization analysis, both free and total hydrolysable amino acids, were extracted, hydrolyzed, and evaporated to dryness from cleaned skeleton powders by standard methods 79 . All samples were prepared in duplicate and analyzed at the Northern Arizona University Amino Acid Geochronology Laboratory using standard methods with modi cations for microfossils [80][81][82] . Rehydrated samples were spiked with L-homo-arginine as an internal standard and then injected into an HPLC tted with a reverse-phase C18-packed column. 'Blank' samples were included. Approximately 1 g cleaned powder from each coral was decalci ed in 0.5 M glacial acetic acid. Acid insoluble matrix (AIM) pellets were rinsed twice in ice-cold 80% acetone whereas acid soluble matrix (ASM) was precipitated in ice-cold 100% acetone and then rinsed twice in ice-cold 80% acetone. Pellets were immediately submitted for protein sequencing. A subset of extracted proteins were separated by SDS-PAGE on 4-20% Mini-PROTEAN TGX Stain-Free™ precast gels (Bio-Rad) which were imaged on a Bio-Rad ChemiDoc XRS + imager following UV light exposure for 5 min.
Skeletal protein AIM and ASM samples were dissolved in 2% SDS bu er and digested using either a lter aided sample preparation (FASP; Mann4) 83 or a multi-enzyme digestion lter aided sample preparation protocol (MED-FASP; Mcav1, Mann2, Mann4, and the surface layer of a modern O. annularis) modi ed from 54 . Brie y, protein was dissolved in SDS bu er and placed in a 30 kDa Microcon Centrifugal Unit (Sigma Aldrich), SDS was displaced using an 8 M urea solution and then the sample was diluted to 2 M urea and digested with trypsin (Promega). Digested peptides were moved through the lter into a micro-centrifuge tube (low retention; Fisher). Any undigested material that remained on the lter was then digested with Glu-C (Promega) and peptides centrifuged into a second micro-centrifuge tube. Each fraction was analyzed separately on an nano-liquid-chromatography system coupled to a benchtop high-resolution orbitrap mass spectrometer (QE-Plus; ermo Fisher) and operated in positive ion mode with data-dependent acquisition. MS1 was performed at resolution of 70,000 (at 400 m/z) and MS2 at 17,500. Peak lists were extracted from raw spectra and processed using a Mascot (2.4; Matrix Science) server against Montastraea cavernosa, M. faveolata, and Platygyra carnosus protein databases downloaded from comparative.reefgenomics.org 35 , an O. faveolata protein database 84 , and the O. annularis genome predicted protein database 85 under NCBI BioProject 550266. A common contaminants database downloaded from the Max Planck Institute of Biochemistry, Martinsried, and a UniProt-Human database were included in the analysis to test for contaminants. We also separately ran the LC-MS/MS data in Mascot against UniProt-bacteria, UniProt-cyanobacteria, and UniProtfungi databases and con rmed that all detected peptides from those databases did not match peptides assigned to corals (SI Table 5). We also ran LC-MS/MS data for a preparation blank sample against the coral, common contaminants, and UniProt-Human databases to con rm that other contaminant from the preparation process were not assigned to corals (SI Table 5). For all Mascot runs, we applied carbamidomethylation of cysteine as a xed modi cation and oxidation of methionine, acetylation, and deamidation of asparagine and glutamine as variable modi cations. Enzyme speci city was set to trypsin with one missed cleavage allowed. Mass tolerances were set to 10 ppm and 20 mmu for precursor and product ions, respectively, and precursor charge was set to 2 + , 3 + , or 4 + . Initial decoy searches were performed in Mascot using a 1% false discovery rate to determine the appropriate signi cance value setting. Next, we performed Mascot error-tolerant searches with this signi cance setting. Only protein sequences above the cuto score with at least two independent signi cant peptides detected, or one peptide detected signi cantly multiple times, were retained. We blasted these sequences against the NCBI nr database in Blast2GO. Further, we BLASTed returned proteins against the NCBI Homo sapiens database and manually checked hits with high sequence similarity for identity of LC-MS/MS detected peptides; if 'coral' and human peptides were identical, we manually removed the protein sequence from our list of coral skeletal proteins. e mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE 86 partner repository with the dataset identi ers found in SI Table 7.
CARP4/SAARP1, CARP5/SAARP2, and P27/acidic SOMP/ SAARP3, previously sequenced from coral skeletons [26][27][28][29] , were blasted against the cnidarian predicted protein databases in comparative.reefgenomics.org 35 . ey were also blasted against NCBI and the top non-cnidarian hits with E-values better than e-20, two Crassostrea gigas sequences, were retained. Multiple sequence alignments were generated in T-Co ee 87,88 . Aligned protein sequences were trimmed using the TrimAl v1.3 alignment utility in Phylemon2 using the gappyout method 89,90 . e most appropriate model was chosen in ProtTest 3 91 , and then maximum likelihood trees were constructed in PhyML using the WAG + G + I substitution model with bootstrap set to 1000 and all other pre-set parameters 92 .
Relative content of each amino acid (THAA) was compared between the replicate data for each fossil coral skeleton reported on here versus average values from modern O. annularis, Fungia sp. Pocillopora damicornis, P. acuta, and Porites lobata, plus previously reported values from modern Acropora palmata 61 and Porites australiensis 63 , as reported or reproduced in Table 1, in RStudio 93 . Shapiro-Wilk normality tests showed that all modern amino acids exhibited normal distribution, so that Student's t-tests were applied (SI Table 6). e mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE 86 partner repository with the dataset identi ers found in SI Table 7.