Transthyretin proteoforms of intraocular origin in human subretinal fluid

Abstract

Understanding the molecular composition of ocular tissues and fluids could inform new approaches to prevalent causes of blindness. Subretinal fluid accumulating between the photoreceptor outer segments and retinal pigment epithelium (RPE) is potentially a rich source of proteins and lipids normally cycling among outer retinal cells and choroid. Herein, intact post-translationally modified proteins (proteoforms) were extracted from subretinal fluids of five patients with rhegmatogenous retinal detachment (RRD), analyzed by tandem mass spectrometry, and compared to published data on these same proteins as synthesized by other organs. Single-nuclei transcriptomic data from non-diseased human retina/RPE were used to identify whether proteins in subretinal fluid were of potential ocular origin. Two human donor eyes with normal maculas were immunoprobed for transthyretin (TTR) with appropriate controls. The three most abundant proteins detected in subretinal fluid were albumin, TTR, and apolipoprotein A-I. Remarkably, TTR relative to the other proteins was more abundant than its serum counterpart, suggestive of TTR being synthesized predominantly locally. Six proteoforms of TTR were detected, with the relative amount of glutathionylated TTR being much higher in the subretinal fluid (12–43%) than values reported for serum (<5%) and cerebrospinal fluid (0.4–13%). Moreover, a putative glycosylated TTR dimer of 32,428 Da was detected as the fourth most abundant protein. The high abundance of TTR and putative TTR dimer in subretinal fluid was supported by analysis of available single-nuclei transcriptomic data, which showed strong and specific signal for TTR in RPE. Immunohistochemistry further showed strong diffuse TTR immunoreactivity in choroidal stroma that contrasted with vertically aligned signal in the outer segment zone of the subretinal space and negligible signal in RPE cell bodies. These results suggest that TTR in the retina is synthesized intraocularly, and glutathionylation is crucial for its normal function. Further studies on the composition, function, and quantities of TTR and other proteoforms in subretinal fluid could inform mechanisms, diagnostic methods, and treatment strategies for age-related macular degeneration, familial amyloidosis, and other retinal diseases involving dysregulation of physiologic lipid transfer and oxidative stress.

1. Introduction

Understanding the molecular composition of ocular tissues and fluids could inform new approaches to prevalent causes of blindness. The eye is small and has numerous biologically distinct tissue compartments and fluids of importance to good visual health. Subretinal fluid accumulating between the photoreceptor outer segments and retinal pigment epithelium (RPE) is potentially a rich source of proteins and lipids normally cycling among outer retinal cells and choroid to sustain vision and metabolism. When the neurosensory retina separates from the RPE, due to a hole or tear (called rhegmatogenous retinal detachment, RRD) (Quintyn and Brasseur, 2004), fluid accumulates between photoreceptors and the RPE, i.e., in the sub-retinal space. Subretinal fluid obtained during surgical repair of RRD represents an attractive biofluid for detailed exploration. Previous studies of subretinal fluid using targeted molecular approaches have identified apolipoproteins and carotenoids (Huerva et al., 1993, 1996; Schneeberger et al., 1997; Chan et al., 1998; Kowalczuk et al., 2018).

Subretinal fluid contains proteins (0.02–61 mg/mL)(Poulsen et al., 2020), lipids (0.1–2.4 mg/mL)(Lam et al., 1975), and metabolites (Heath et al., 1962; Akhmeteli et al., 1975; Lam et al., 1975, 1980; Gebhardt et al., 1983; Newsome and Wiggert, 1988; Huerva et al., 1993, 1996; Schneeberger et al., 1997; Chan et al., 1998; Shitama et al., 2008; Ding et al., 2018), which could originate from retina, RPE, or plasma transudate (Quintyn and Brasseur, 2004). In RRD, components of the vitreous body might also be present in the subretinal fluid. Compared to lipids and metabolites, subretinal proteins have been better characterized (Kowalczuk et al., 2018; Poulsen et al., 2020), but knowledge is still limited. In previous studies of subretinal fluid composition, a conventional bottom-up proteomics approach to characterize proteins was utilized (Kowalczuk et al., 2018; Poulsen et al., 2020). This approach involves five processing and bioinformatic steps, including 1) reduction of any disulfide bonds in proteins, 2) alkylation of the resultant reduced cysteine residues to avoid a reversible reaction, 3) enzymatic digestion of the reduced and alkylated proteins into small peptides, 4) mass spectrometry analysis of the peptides, and 5) reference to theoretically generated peptides in widely utilized databases (Gundry et al., 2009). This approach is sensitive in detecting the presence of various proteins. However, with the steps mentioned above, post-translationally modified forms (proteoforms) and their abundance, essential to understanding protein function (Mylona et al., 2016), is often lost (Schaffer et al., 2019).

Top-down proteomics, a relatively new approach to analyze proteoforms, could overcome the typical information loss in bottom-up proteomics (Whitelegge et al., 2006; Zhang and Ge, 2011). This top-down approach analyzes intact proteins by mass spectrometry (MS) and tandem mass spectrometry (MS/MS), with minimum or no treatment to the proteins. As a result, various proteoforms of the same proteins can be directly detected, identified, and quantified. This approach has been applied to the study of many biological samples, including serum and saliva (van der Burgt and Cobbaert, 2018). However, there are no reports on the application of this method to subretinal fluid.

Herein we profiled and identified proteoforms in subretinal fluid captured from RRD patients, via top-down proteomics. Transthyretin (TTR, originally named prealbumin) was found to be predominant and composed of six proteoforms, with the relative amount of glutathionylated TTR much higher than values reported for counterparts in serum/plasma or cerebrospinal fluid. Suspecting an intraocular origin of glutathionylated TTR based on these findings, we sought confirmation via analysis of publicly available gene expression data and immunolocalization of TTR in human eyes.

2. Methods and materials

2.1. Chemicals and reagents

Methanol (LC-MS grade, > 99.9%) and water (CHROMASOLV^™ LC-MS grade) were purchased from Sigma-Aldrich (St. Louis, MO, USA). Formic acid (99.0+%, Optima^™ LC/MS grade), primary antibody mouse anti-cytochrome c oxidase subunit 4 (COX4) (1:500), and isotype negative control antibody mouse IgG1k (1:1000) were purchased from Thermo Fisher Scientific (Norcross, GA, USA). Primary antibody of rabbit anti-TTR (1:200) was purchased from Agilent (Santa Clara, CA, USA). Biotin-conjugated horse anti-mouse/rabbit secondary antibody, avidin-biotin-alkaline phosphatase complex (ABC-AP), and alkaline phosphatase (AP) substrate were purchased from Vector Labs (Burlin-game, CA, USA). Periodic acid Schiff-hematoxylin staining (PASH kit) was purchased from Poly Scientific R&D Corp (Bay Shore, New York, USA).

2.2. Subretinal and vitreous fluids

The study protocol was approved by the Institutional Review Board at the University of Alabama at Birmingham and conformed with the Declaration of Helsinki. Subretinal fluid samples were collected from eyes of RRD patients at the time of retinal detachment repair (See Supplementary Video). All patients underwent standard 25-gauge pars plana vitrectomy. All detachments involved peripheral retina, and in 5 of 7 patients the detachments also involved the macula (macula-off) (Table 1 for clinical information). As a natural part of vitrectomy the posterior segment of the eye was constantly infused with balanced salt solution to maintain a safe intraocular pressure. Once the vitreous gel was removed and all traction was cleared from the retinal breaks, all balanced salt solution was aspirated from the vitreous cavity with a fluid air exchange. To prevent saline from entering the subretinal space and possibly affecting sample integrity, no subretinal fluid was removed for assay until the vitreous cavity was clear of fluid and filled with air. After fluid air exchange, subretinal fluid was removed through either the retinal defect or a retinotomy using a 25-gauge cannula with active foot pedal-controlled aspiration. The sample was then removed from the line with a syringe and stored for analysis. Retinal detachment repair then proceeded with endolaser and air gas exchange. Patient 6 also had a scleral buckle placed at the time of the vitrectomy due to the presence of inferior retinal pathology. Because patients 5 and 7 did not have macula-off retinal detachments, the technique was modified, as complete fluid air exchange prior to subretinal fluid collection could have led to macular detachment and subsequent poor patient visual outcomes. Accordingly, the subretinal fluid was collected in these cases prior to fluid air exchange. In these cases, the 25-gauge cannula was placed directly under the retina, through a retinotomy or the retinal defect, to directly aspirate fluid. In all cases (macula-on and macula-off), vitreous fluid samples were collected after core vitrectomy to compare with the subretinal fluid samples, using a different 25-gauge cannula and aspiration line. Single operation reattachment rate was 100%. The only complication was a vitreous hemorrhage with mild transient choroidal hemorrhage in patient 6. The vitreous hemorrhage was cleared with a second vitrectomy and the patient did well with a final visual acuity of 20/40.

Table 1.

Summary of patients’ clinical data.

Patient	Age	Sex	Past Medical History	Past Ocular History	Retinal detachment, macula ON or OFF
1	81	F	Hypertension	Age-related macular degeneration, Hypertensive Retinopathy	Off
2	78	F	Hypertension, thyroid disease, psoriasis	None	Off
3	73	M	Skin cancer	Glaucoma both eyes, Uveitis both eyes, Endophthalmitis (non-study eye);	Off
4	77	M	Diabetes, hypertension, coronary artery disease, stroke, sleep apnea	Hypertensive Retinopathy	Off
5	52	M	Hypertension, hypercholesterolemia	None	On
6	63	M	Bone cancer	Corneal burn, Lattice degeneration of the retina, Choroidal nevus (non-study eye)	Off
7	60	F	Arthritis, anxiety	None	On

All patients were White.

Supplementary video related to this article can be found at https://doi.org/10.1016/j.exer.2022.109163

Sample volume for subretinal fluid was between 150 μL and 1.5 mL and for vitreous, between 50 μL and 1 mL. All samples were kept frozen at −80 °C until ready for MS analysis. Samples from patients 2 and 5 (Table 1) were used for testing experimental conditions. The remaining five samples were successfully analyzed.

2.3. Sample preparation for mass spectrometry

Fig. 1 shows the typical workflow for sample preparation and proteoform identification. A protein precipitation method was applied to remove salts, lipids, and other small molecules that are dissolvable in 80% methanol and would otherwise interfere with the downstream protein analysis. Specifically, 50 μL subretinal or vitreous fluids were mixed with 200 μL methanol in a 2 mL Eppendorf tube and vortexed for about 20 s. This mixture was then centrifuged for 1 min at 13,300 rpm/17,000g (Fisher Microcentrifuge accuSpin Micro 17, Norcross, GA, USA) to spin down the protein precipitant. The supernatant that contained salts and other small molecules was aspirated with a pipette and discarded. The precipitant was cleaned by adding 50 μL water and 200 μL methanol into the tube and vortexing the tube. The tube was centrifuged again, and the supernatant discarded. A volume of 150 μL water containing 1% formic acid was then mixed with the precipitant in the tube by vortexing. The precipitant completely dissolved immediately. The solution was then mixed with 150 μL methanol as the working solution for MS analysis of intact proteins. During this rapid sample preparation procedure, potential effects of proteases were negligible. The addition of methanol for protein precipitation and later to the working solution further deactivated proteases. Therefore, protease inhibitors were not necessary for this study, which eliminated potential interferences to data acquisition.

Fig. 1.

Flowchart of the typical workflow for sample preparation and proteoform identification. *Putative proteoforms were determined from reported proteins (in subretinal fluid or other body fluids) and potential post-translational modifications. See text for more details.

2.4. Mass spectrometry

Each subretinal or vitreous fluid working solution was directly infused into a Synapt G2-Si mass spectrometer from Waters Corporation (Milford, MA, USA) at a flow rate of 2 μL/min using a Harvard Apparatus syringe pump (Holliston, MA, USA). The mass spectrometer was performed with electrospray ionization in the TOF mode (ion mobility capability disabled).

The typical parameters for MS analysis include source temperature at 100 °C, desolvation temperature at 200 °C, desolvation gas flow at 500 L/h, source capillary at 2.3 kV, and sampling cone voltage at 50 V. For one sample, a higher cone voltage of 100 V or 150 V was also performed, which may preferentially transfer ions with high mass-to-charge ratios (m/z) but can also induce in-source dissociation of all ions. Samples were analyzed by MS and tandem MS/MS to detect protein ions and collision-induced dissociation (CID) fragments of protein ions, respectively. In MS mode, the trap collision energy was set at 10 eV and transfer collision energy was set at 2 eV to increase the resolution while not dissociating the protein ions. In MS/MS mode, protein ions of interest were isolated, and the trap collision energy was increased to 20–50 eV, depending on the precursor ions, to dissociate the protein ions (precursor ions) into fragments. The mass spectrometer was calibrated with a mixture of sodium iodide and cesium iodide, as suggested by the manufacturer. MS data were typically acquired for 1 min, but a more extended time of up to 50 min was used for some MS/MS experiments to increase the signal-to-noise ratio.

The peaks in the mass spectra were converted to neutral masses by one of two deconvolution approaches. One approach was based on the different charge states of the same protein ions, using MaxEnt 1 function of MassLynx (Ferrige et al., 1991) (Waters Corporation, Milford, MA, USA) or UniDec (Marty et al., 2015) (Marty Lab at the University of Arizona, Tucson, AZ, USA). The other approach was based on the isotopic distribution of peaks at each charge state, using the TopFD function of the TopPIC program (Kou et al., 2016). Before deconvolution, spectra across the acquisition time were averaged, and peaks were picked in a way best for the subsequent deconvolution programs. Specifically, for MaxEnt 1, spectra were first processed with background subtraction, then deconvoluted using the Uniform Gaussian Damage model with minimum intensity ratios of 33% on the left and right. For UniDec, peak lists were generated with the “Copy Spectrum List” function of MassLynx, and the resultant peak list file was loaded to UniDec for deconvolution. The typical setting for deconvolution by UniDec included the following: “use background subtraction” for the initial data processing, a charge range of “1 to 70” for the multiply charged protein peaks, a mass range of “5,000 to 100,000 Da” for the protein masses, and “suppress lots of artifacts” to minimize harmonics and satellites peaks. While for TopFD, the peak lists were generated using the “Automatic Peak Detection” function of MassLynx with the setting of a resolution of 30,000 and background subtraction; the maximum mass and charge state was respectively set at 100,000 and 25, for MS spectrum deconvolution, and the precursor mass and the corresponding charge state for MS/MS spectrum deconvolution.

Unknown proteins were initially tentatively identified by matching their molecular weights with those of abundant proteins in serum/plasma or subretinal fluid in the literature, considering potential post-translational modifications (PTMs). Then, when possible, the identified proteins were further confirmed by matching their fragments in collision-induced dissociation MS/MS spectra to theoretical values using MS-Product (https://prospector.ucsf.edu/) or TopMSV (Choi et al., 2021) for proteins of known sequences and PTMs.

TTR was identified by searching peak lists of the fragments against the human protein database (SwissProt 2020.09.02). Specifically, fragment peaks of different charge states in mass spectra were first deisotoped and deconvolved to neutral masses. Neutral masses were then converted to singly charged masses, which were subsequently searched in MS-Tag and Batch-Tag for single and multiple MS/MS spectra, respectively (Chalkley et al., 2005; Baker et al., 2011). MS-Tag or Batch-Tag was initially designed for bottom-up proteomics; however, they can be adapted for top-down proteomics by setting appropriate parameters (Chen et al., 2012). These include “Full protein” with “nonspecific at N-term” as the digest, “Dehydro (C)” (i.e., loss of a hydrogen atom from the cysteine residue) as the fixed modification because cysteine (C) residues typically form disulfide bonds within proteins (two cysteine residues of a protein covalently-linked together with a disulfide bond, with each cysteine residue loses a hydrogen atom), 50 ppm as the tolerance for the fragment ions, and a broad precursor-ion window (500 Da) to account for potential unexpected modifications. When possible, putative TTR proteoforms were confirmed by matching the experimental masses of the fragment peaks to theoretical values with MS-Product (https://prospector.ucsf.edu/) or TopMSV (Choi et al., 2021).

2.5. Relative quantification of TTR proteoforms

The relative amounts of the TTR proteoforms were determined from the relative intensity of the corresponding peaks at the same charge state and, when possible, averaged across different charge states. Such an approach works effectively, as demonstrated by a report of the solution concentration ratios of recombinant human histone H4 mutants (11 kDa) corresponding closely with MS ratios (Pesavento et al., 2006).

2.6. Single-nuclei RNA-sequencing

To identify potential intra-retinal sources of TTR, single-nuclei RNA-sequencing data of 11 posterior segment samples including RPE from 4 total human donor eyes (Orozco et al., 2020) were downloaded from the GEO repository (GSE135133) as demultiplexed, unnormalized gene counts per nucleus. Raw sequencing data from this published study were not provided due to patient privacy concerns. All downstream analyses, apart from data normalization and dimensionality reduction/batch correction, were done using Scanpy (Wolf et al., 2018) (version 1.8.1), a Python-based suite of packages for single-cell RNA-sequencing analysis. Initially, the 11 samples were concatenated, corresponding to 121,779 total nuclei. For quality control filtering, nuclei were removed that contained >20,000 reads and >6,000 genes. No nuclei included in the publicly available geneXnuclei matrixes contained <500 counts or <700 genes, eliminating the need for minimum thresholds. Scrublet ((Wolock et al., 2019); version 0.2.1) was used to remove potential doublets and was applied to individual samples to account for sample-to-sample variation. After quality control filtering and doublet removal, 111,809 nuclei were recovered in total. Normalization was performed using Scran (Lun et al., 2016) (version 1.10.2). Dimensionality reduction (t-SNE), batch correction, and clustering (Louvain resolution of 0.1) were performed on the top 2,000 highly variable genes using DESC (Li et al., 2020) (version 2.0.2), an unsupervised deep-learning based algorithm for iterative dimensionality reduction, batch correction, and clustering of single-cell RNA-sequencing data. Cluster annotations were guided by the expression of well-defined retinal cell-type specific marker genes. Dendrogram analysis was performed on the top 50 principal components. Data visualizations were created using standard Scanpy functions.

2.7. Immunohistochemistry

Two eyes from two human donors were used for immunohistochemical localization of TTR in human retina and supporting tissues. Whole eyes were obtained from deceased donors (≥80 years of age, white, non-diabetic, and ≤6 h death-to-preservation) to Advancing Sight Network (Birmingham AL USA). Eyes were preserved in 4% buffered paraformaldehyde and screened to exclude AMD using ex vivo multi-modal imaging including optical coherence tomography, as described (Cao et al., 2021; Chen et al., 2022). Immunohistochemistry was performed as described previously (Cao et al., 2021). Detailed methods are included in Supplementary Materials. In brief, rectangles of full-thickness eye wall (5 × 8 mm) including fovea and optic nerve were dissected and processed for 12 μm thick cryosections. Other cryosections were incubated with primary antibodies of rabbit anti-TTR or mouse anti-COX4 for photoreceptor inner segment mitochondria, followed by incubation with horse anti-mouse/rabbit secondary antibody. An avidin-biotin-alkaline phosphatase detection system was developed for a red reaction product then counterstained and dehydrated through ethanol-xylene. Coverslips were mounted with permanent medium. Positive and negative controls were in-house human tissue array sections. Neighboring retina sections from one eye were used as sham controls for TTR by omitting the primary antibody or as isotype control for COX4 by probing with mouse IgG1k. Neighboring cryosections were stained with periodic acid Schiff hematoxylin to highlight outer retinal structure with special attention to the RPE basal lamina, basal laminar deposit, and Bruch’s membrane (not shown, Vogt et al., 2011). Glass slides were scanned using 20x and 40x objectives on a microscope with a robotic stage (Olympus VSI 120, Olympus, Japan).

3. Results

3.1. Mass spectrometry analysis of albumin, Apolipoprotein A-I (ApoA1), and TTR proteoforms in the subretinal fluid

3.1.1. Identification of albumin and ApoA1 proteoforms

Three high-intensity series of protein peaks with different charge states were observed in the MS spectra of proteins extracted from subretinal fluid (Fig. 2A). The masses of deconvolved peaks (determined from their charge-state distribution) for two of these series match six proteoforms of intact human serum albumin (Leblanc et al., 2018) and two proteoforms of ApoA1 (Niederkofler et al., 2003) (Fig. 2 B and C, Supplementary Table 1). Putative albumin and ApoA1 were confirmed by matching the corresponding MS/MS spectra with theoretical values of the protein fragments (Chen et al., 2008) (Supplementary Figs. 1 and 2).

Fig. 2.

A presentative mass spectrum of proteins extracted from the subretinal fluid. (A) MS spectrum, and deconvolved MS spectra of (B) albumin, (C) apolipoprotein A-I (ApoA1), and (D) transthyretin. Note that same proteins carried many different positive charges, as indicated with the numbers and “+”, in the form of proton adducts. Due to space limit, only 3 charges of ApoA1 protein peak are labeled. Several low-intensity in-source fragment peaks were also observed. The major proteoforms of the proteins are labeled with the corresponding post-translational modifications. Abbreviations: -S.P., signal peptide removal; C10G, variant with the 10th residue (not counting the 20 residues in the signal peptide) cysteine replaced by glycine; S-Sulfo, sulfonation; S-Cys, cysteinylation; S-Cysgly, cysteinylglycylation; S-GSH, glutathionylation; K107del, variant with the deletion of the 107th residue; -DA, truncated form with a loss of the N-terminal aspartic acid and alanine residues; Glc, glucose modification or glucosylation; Glyco, glycosylation. See text for experimental details.

In this study, both albumin and ApoA1 were reproducibly detected at high intensity for all five subretinal fluid samples. One concern with detecting these proteins in the subretinal fluid is the potential blood contamination, which could originate either from the surgical procedure or retinal capillaries damaged by the tear. Even though we collected samples very carefully with no noticeable blood, serum proteins, particularly abundant ones such as albumin (Hortin, 2006), could have diffused into the sub-retinal fluid. However, it is also plausible that albumin was synthesized locally. Indeed, albumin protein and mRNA were detected in the interphotoreceptor space and retina (Dodson et al., 2001; Carter-Dawson et al., 2010). Albumin has a significant role in transporting low solubility lipids such as retinoids (Futterman and Heller, 1972) and fatty acids in the retina (Spector, 1975).

Compared to albumin, the primary source of ApoA1 is more ambiguous. As an abundant protein in serum/plasma (Hortin, 2006), ApoA1 detected in this study could have leaked in part from serum/plasma. However, RPE has been reported to be the primary source of ApoA1 within the retina (Simó et al., 2008, 2009), which is consistent with ApoA1’s function in transporting retinal lipids, including retinoic acid (Getz et al., 1999). Interestingly, one ApoA1 proteoform corresponds to the mature ApoA1 (full sequence minus the signal peptide) with the deletion of the 107th residue, a proteoform possibly linked to systemic amyloidosis with impaired lipid binding (Frank and Marcel, 2000). In summary, although the presence of albumin and ApoA1 in subretinal fluid may be due to contact with blood, these proteins also have a role in intraocular lipid transfer, according to prior literature.

3.1.2. Identification and relative quantification of TTR proteoforms

In contrast to albumin and ApoA1 peaks, the third series of peaks was less frequently reported in the literature. A search of the peak lists of the corresponding MS/MS spectra against the human proteins database using MS-Tag found with high confidence that the unknown protein was TTR (Supplementary Table 2). We then tentatively identified six TTR proteoforms by matching the experimental masses of the isotope-resolved protein peaks to theoretical masses of previous reported TTR proteoforms for plasma/serum samples (Shimizu et al., 2006; Théberge et al., 2011; Henze et al., 2015). These proteoforms include mature TTR (without modification other than signal peptide removal), mature TTR with a variation in one amino acid, and mature TTR with a modification of cysteinylation, cysteinylglycylation, glutathionylation, or sulfonation at its cysteine residue (Fig. 2D, Supplementary Fig. 3, Supplementary Table 1). Three of these putative TTR proteoforms were further confirmed by matching their experimental fragments in the MS/MS spectra to the corresponding theoretical fragments (Fig. 3, Supplementary Figs. 4–6, Supplementary Tables 3 and 4). These peaks could also be automatically identified by searching peak lists of the MS/MS spectra against the human protein database using Batch-Tag (Chalkley et al., 2005; Baker et al., 2011). This gives even higher confidence in their identities (manuscript in preparation).

Fig. 3.

Confirmation of cysteinylated transthyretin (S-Cys-TTR) in the subretinal fluid. Peaks in the MS/MS spectrum correspond to fragments from collision-induced dissociation of isolated putative S-Cys-TTR ions (m/z 868.5, 16+). The labels of the peaks indicate a match between the experimental and theoretical m/z values for terminal (y and b), internal (yb) fragment ions, or water loss ion (M − H₂O) of S-Cys-TTR, with the cysteinylation modification kept intact on fragments (e.g., b₄₂) that contained the amino acid residue C10. Insets (A)-(D) expand the corresponding regions in the main spectrum to show (A) the isotopic distribution or (B)–(D) details of the matched peaks. The theoretical m/z values of the fragments were generated with MS-Product (https://prospector.ucsf.edu/). Matched peaks were manually verified. Almost all the high-intensity peaks from the experiment showed a good match, confirming the putative S-Cys-TTR to be correct. The trap collision energy was 20 eV. See text for experimental details.

Cysteine-residue modifications in proteins are known to be related to oxidative stress (Beatty et al., 2000; Jenner et al., 2003; Barnham et al., 2004; Wakamatsu et al., 2008; Berthoud and Beyer, 2009; Giacco and Brownlee, 2010; Kim et al., 2015; Nita and Grzybowski, 2016), a factor implicated in multiple diseases, including AMD (Datta et al., 2017). Such modifications were also reported to change the binding affinity of TTR (Henze et al., 2015). Therefore, quantifying the level of modifications is essential to understand mechanisms of diseases involving transport of retinoids or other molecules by TTR. We quantified the relative amounts of different TTR proteoforms in each sample (Supplementary Fig. 7) at the same charge state and then averaged the relative amount across different charge states. Except for proteoform peaks of low signal-to-noise ratios or overlapping with other species at certain charge states, the ratios at different charge states were consistent, and the mean values were reported (Fig. 4, Supplementary Table 5).

Fig. 4.

Comparison of the TTR proteoforms’ distribution between five samples. The error bars represent one standard deviation. Patient 1 had age-related macular degeneration. For details of patients see Table 1.

Other than the above discussed proteins with detailed identification of related proteoforms, one protein of 32,428.0 ± 1.3 Da was reproducibly detected at high confidence in all five subretinal samples using UniDec (Supplementary Table 6). Coincidentally, a protein of almost the same molecular weight (32.4 kDa)(Domoto et al., 2016), reported to have elevated in the blood of saturation divers, was detected in a 2D Western blot analysis using an anti-TTR antibody. These individuals dive long enough that body tissues are saturated with inert components of breathing gas under hyperbaric conditions. Further MS/MS analysis of the in-gel digested protein was 100% matched with TTR (Domoto et al., 2016). Interestingly, similar gel bands of about 34 kDa were previously detected as TTR dimer by Western analysis for both human serum and synthetic TTR-containing solution (Wilce et al., 2001). Considering the high abundance of TTR in the subretinal fluid, the 32, 428 Da protein peak we detected is probably a TTR dimer with some unknown PTM. The presence of this TTR dimer in the subretinal fluid is supported by reported detection of TTR in the RPE cytosol as a protein spot of 37 kDa on a 2D-gel image (West et al., 2003), which is within the typical error range of 32.4 kDa. A series of other peaks were detected with the masses close to this 32,428 Da peak, including some with a mass difference corresponding to a glycosylation moiety (Supplementary Table 7). Thus, we interpret this 32,428 Da peak and several others as TTR dimers with different glycosylation modifications.

Compared to albumin and ApoA1, TTR and the related putative TTR dimer appeared unlikely to originate from blood because the peak intensity of TTR relative to albumin in the intact protein mass spectra was much higher for subretinal fluid than for serum/plasma, where TTR peaks were negligible (Chen et al., 2008). It is consistent with literature reports that the amount of TTR relative to albumin in subretinal fluid (2.9%) (Berrod et al., 1993) was about 4 times that in serum/plasma (0.7%) (Hortin, 2006). To explore the intraocular role of transthyretin, we investigated gene expression and localized TTR in retinal sections.

3.2. TTR single-nuclei RNA-sequencing

To identify local versus systemic sources of proteins found in human subretinal fluid, we reanalyzed publicly available single-nuclei RNA-sequencing data from the posterior segments of 4 human donor eyes (Orozco et al., 2020). After preprocessing and quality control filtering, 111,809 nuclei were captured. Dimensionality reduction and clustering resulted in 20 distinct cell-types (Fig. 5A). Using well-defined marker genes, clusters were annotated as rods and cones, Müller glia, RPE, horizontal cells, retinal ganglion cells, 5 types of amacrine cells, 8 types of bipolar cells, and a population of mixed cell-types that includes, but is not limited to, astrocytes, endothelial cells, and microglia/immune cells (Fig. 5B). Among the 3 predominant proteins identified in all 5 subretinal fluid samples (albumin, ApoA1, and TTR), only TTR showed robust local expression in the posterior segment. TTR was strongly expressed by RPE and weakly expressed by a subtype of bipolar cells, annotated as bipolar3 (Fig. 5C) and considered an ON bipolar cell due to the simultaneous high expression of GMR6 (Fig. 5B). RPE-specific expression of TTR was more than 2.5-fold greater than that found in the bipolar3 cluster (Fig. 5D). Given the high expression of TTR within the RPE and the location of RPE bounding the subretinal space, these findings suggest that the RPE is a local source of the TTR pool ubiquitously identified in subretinal fluid. It is consistent with a previous report that RPE was the unique site of TTR synthesis in the rat eye (Cavallaro et al., 1990).

Fig. 5.

Single-nuclei RNA-sequencing of human posterior segment identifies transthyretin (TTR) expression in retinal pigmented epithelium (RPE). (A) Dimensionality reduction of 111,809 nuclei using t-distributed stochastic neighbor embedding (t-SNE) identifies 20 distinct cell-types within the human posterior segment. (B) Expression of cell-type specific marker genes by cluster. (C) TTR expression overlaid on t-SNE map. (D) Violin plots with embedded boxplots showing TTR expression by cluster.

3.3. TTR immunohistochemistry of retinal sections

Fig. 6 shows TTR immunoreactivity in the neurosensory retina, RPE and choroid of normal aged human retina using a colorimetric detection method. Appropriate positive and negative controls are shown in Supplementary Fig. 10. The choroidal stroma is deeply and uniformly stained (Fig. 6A and B). In contrast, TTR signal in the outer retina is highly compartmentalized. RPE cell bodies, despite high mRNA content (Fig. 5), have neglible signal, while in the subretinal space strong reaction product parallels the outer segments. This observation may not seem consistent with the gene data. However, it can be reasonably explained by TTR secretion into the interphotoreceptor matrix, which is supported by the following two reports. The first is that for cells grown in chambers with permeable supports, the predominant direction for TTR secretion was into the apical medium, equivalent to the interphotoreceptor matrix in vivo (Ong et al., 1994). The second is that the interphotoreceptor matrix contains a high concentration of TTR (Adler and Edwards, 2000).

Fig. 6.

Transthyretin (TTR) immunolocalization in aged human retina. (A)-(A1) and (B)-(B1). Neuro-sensory retina and choroid in normal maculas from donors >80 years. Sections were immunolabeled with TTR antibody and detected via alkaline phosphatase-linked red-colored product. Outer retina and choroid are magnified in (A1) and (B1). (C)-(C1). Same eye as (B)-(B1) was immunolabeled using anti-cytochrome oxidase 4 to localize mitochondria at the ellipsoid of photoreceptor inner segments for comparison to the TTR-positive outer segment zone. (D)-(D1). In the same eye as (A)-(A1), sham control lacking primary antibody shows tissue architecture. Yellow arrow-heads, Bruch’s membrane. Scale bars in (D)-(D1) apply to top and bottom rows, respectively. NFL, nerve fiber layer; GCL, Ganglion cell layer; IPL, inner plexiform layer; INL, inner nuclear layer; OPL, outer plexiform layer; ONL, outer nuclear layer; ELM, external limiting membrane; ISZ, inner segment zone (photoreceptor inner segments and surrounding interphotoreceptor matrix); OSZ, outer segment zone (photoreceptor inner segments and surrounding interphotoreceptor matrix and RPE apical processes); *, zone of compacted apical processes and outer segment tips; RPE, retinal pigment epithelium; BrM, Bruch’s membrane; ChC, choriocapillaris; Ch, Choroid.

Due to RPE pigmentation, double-labeling is challenging for colorimetric detection methods. Therefore to clarify the localization of TTR in outer retina, we stained adjacent sections with a reference antibody COX4 to delimit the mitochondria-rich ellipsoid portion of photoreceptor inner segments (Fig. 6C and C1). This comparison shows that for this vertically well-aligned and attached retinal specimen with minimal compaction (Fig. 6A1, asterisk), TTR in the outer segment zone and COX4 signals in photoreceptor inner segments do not overlap. We did not see a signal consistent with any bipolar cell labeling as suggested by Fig. 5.

Overall the difference in the patterns of immunoreactivity between choroid and outer retina supports a systemic source of TTR in the choroid and a local source of TTR (RPE) in the retina.

3.4. Mass spectrometry analysis of other proteins/proteoforms in the subretinal fluid

In addition to the proteins identified in all samples mentioned above, we detected several other proteins in the sample (Patient 1 in Table 1) with the highest protein content. These proteins include putative cystatin C, β₂ microglobulin (β₂M), apolipoprotein C-I (ApoC1), and transferrin. Isotopic resolved peaks were obtained for the first three proteins, corresponding to two proteoforms of cystatin C, one proteoform of β₂M, and two proteoforms of ApoC1 with the experimental masses (monoisotopic masses) within 5 ppm errors of the corresponding theoretical masses (Supplementary Fig. 8, Supplementary Table 1). In contrast, isotopic resolved peaks were not obtained for transferrin and its mass was determined by deconvolving its multiple charge state peaks (Supplementary Fig. 9a), which resulted in 79,557.1 ± 0.5 Da (most abundant mass) (Supplementary Fig. 9b, Supplementary Table 1). This mass is consistent with literature values (79,557 Da) (Abu Bakar et al., 2018) for the most abundant form of transferrin in serum samples of healthy subjects, which corresponds to mature transferrin with complete N-glycans (including 10 hexose, 8 N-acetylhexosamine, and 4 N-ace-tylneuraminic acid residues) (Abu Bakar et al., 2018). In addition, several other proteoform peaks were determined with masses of 15,922 Da, 33,998 Da, 34,129 Da, 34,414 Da, 34,525 Da, 54,408 Da, 80,068 Da, and 108,585 Da. Interestingly, two peaks were detected with deconvolved masses of 133,820 Da and 133,995 Da, which are 405 Da and 580 Da higher, respectively, than the theoretical mass (133,415 Da) of unmodified mature interphotoreceptor retinoid-binding protein, the major carrier protein for retinoid and fatty acids (Zeng et al., 2020). Further studies are needed to identify these proteoforms.

4. Discussion

4.1. Comparison with the conventional bottom-up approach

Many proteins in the subretinal fluid were previously identified using the conventional bottom-up proteomics approach (Kowalczuk et al., 2018; Poulsen et al., 2020). However, these studies did not report the specific proteoforms of these proteins, which are essential to elucidate the mechanisms of relevant diseases. Moreover, the relative amount determined in the bottom-up approach was inferred from the resulting peptides, which could be inaccurate (Schaffer et al., 2019). In contrast, our top-down MS-based study provides a bird’s eye view of the direct detection of various proteoforms of intact proteins in the subretinal fluid and identifies these proteoforms directly from their fragments by CID MS/MS.

The top-down MS approach also overcomes the challenges in identifying and localizing PTMs in peptides by the bottom-up proteomics. With bottom-up proteomics, the PTMs on peptides are often labile. Upon dissociation, peptide ions that lost only the corresponding modifications are often dominant, which overshadow other more informative fragments (Boersema et al., 2009). In contrast, in our study with collision-induced dissociation of intact TTR proteoforms, the PTMs at cysteine residues were kept intact in the fragment ions, which facilitate characterizing the PTMs. The stable PTMs on fragments of intact TTR proteoforms are consistent with previous reports on CID of proteins with PTMs (Wu et al., 2009; Chen et al., 2019), a phenomenon contrary to peptide ions (Boersema et al., 2009).

TTR was one of the most abundant proteins we detected in the subretinal fluid in our top-down studies. Surprisingly, TTR only ranked 21st of the top 50 most abundant proteins detected by the bottom-up approach, where quantification of proteins was based on the normalized intensity of peptides for a protein (Poulsen et al., 2020). This lower ranking could be because of quantification in the bottom-up approach complicated by PTMs (Larive et al., 1999).

We also analyzed proteins extracted from the vitreous fluid and observed similar patterns of abundant albumin peaks for subretinal fluid samples. However, the signal-to-noise ratio was too low to obtain more information. Further optimization of experimental conditions should increase detection sensitivity for vitreous fluid, making it possible to compare proteoforms between subretinal and vitreous fluids to further our understanding of retinal diseases.

One limitation of this study is that only a few proteins were detected, which is typical for top-down proteomics and different from bottom-up proteomics (Schaffer et al., 2019). Nevertheless, this is the first report applying top-down proteomics to subretinal fluid. By further optimizing sample preparation and other experimental conditions, this approach will be able to characterize many more proteins. Alternative setups, including nanospray ionization, ultra-high-resolution MS, and high-performance liquid chromatography coupled to MS, can also increase the number of detected proteins.

4.2. TTR monomers, dimers, and tetramers

TTR is an abundant protein in body fluids, including plasma (200–400 mg/L) (Hortin, 2006), cerebrospinal fluid (5–20 mg/L) (Vatassery et al., 1991; Hühmer et al., 2006), vitreous fluid (4–24 mg/L) (Van Aken et al., 2009), and subretinal fluid (370 mg/L) (Berrod et al., 1993). Best known for forming a complex with retinol-binding protein to transport retinol (Buxbaum and Reixach, 2009), TTR alone could also transport other low solubility molecules, including thyroxine (Buxbaum and Reixach, 2009), retinoic acid (Zanotti et al., 1995; Hyung et al., 2010), and potentially lutein and related retinoids (Pettersson et al., 1995). Most of these molecules are abundant in the retina.

Intact TTR was previously detected by mass spectrometry as monomers, dimers, and tetramers for plasma, tissue, and recombinant protein samples (Nettleton et al., 1998; Kingsbury et al., 2007; Théberge et al., 2011; Hall et al., 2013). A tetramer is the form in which TTR transports retinol and thyroxine (Buxbaum and Reixach, 2009). Monomers are thought to result from the dissociation of tetramers and amyloid-prone (Buxbaum and Reixach, 2009). While TTR dimer has been associated with amyloid diseases, most likely in disulfide-bonded form (Kingsbury et al., 2007). We detected TTR monomers and a putative dimer but not tetramers in this study.

In our studies, TTR was found abundant in subretinal fluid in six monomeric proteoforms. Among all these TTR proteoforms, glutathionylated TTR is of particular interest. Compared to serum (<5%) (Nedelkov et al., 2005; Shimizu et al., 2006; Théberge et al., 2011; Cheon et al., 2016) and cerebrospinal fluid (0.4–13%)(Théberge et al., 2011), glutathionylated TTR relative to the sum of all the major TTR proteoforms was much higher in subretinal fluid (12%–47%) (Fig. 4, Supplementary Table 5). It suggests that glutathionylation could be essential for normal retinal physiology. Compared to cysteinylation and cysteinylglycylation, glutathionylation introduced one more carboxyl group, which could facilitate transporting metabolites that contain two acidic groups through a proton-linked transition state. For instance, glutathionylated TTR more potently bound triiodothyronine (Henze et al., 2015), a compound that contains one carboxyl group and one phenol group, i.e., two acidic groups under the physiological condition. A significantly lower amount (12%) of glutathionylated TTR in the AMD subretinal fluid may relate to decreased amounts of free and total glutathione in the blood of AMD patients (Samiec et al., 1998), suggesting a positive correlation between local and systemic antioxidant defense mechanisms (Qin et al., 2011). Analysis of more samples is needed to definitively explore these relationships.

In addition to TTR monomers, a putative TTR dimer was detected in this study as an abundant protein peak of 32,428 Da. TTR dimers with similar masses (32–34 kDa) were frequently reported for serum samples in gel images (Wilce et al., 2001; Purkey et al., 2001; Higaki et al., 2016). The mass of this putative dimer we detected is 4,908 Da higher than the theoretical mass (27,520 Da) of the dimer found in amyloid deposited TTR, formed by two unmodified mature monomers linked with a disulfide bond (Kingsbury et al., 2007). Such a mass change was consistent with glycosylation on each of its monomer components (Sato et al., 2012). Glycosylation of TTR dimers may be critical to avoid amyloid formation by increasing their stability (Solá and Griebenow, 2009). Decreased amount of glycosylation moiety in TTR dimers was previously linked to amyloid (Higaki et al., 2016). Interestingly, glycosylation in TTR monomers was neither detected in this study nor reported for wild-type monomers in serum samples (Teixeira and Saraiva, 2013), suggesting glycosylation in dimers was most likely co-translated or post-translated after synthesis of these dimers. Indeed, it was previously reported that disulfide bond formation is a determinant for glycosylation (McGinnes and Morrison, 1997). In contrast to absence of glycosylation for wild-type TTR monomers, possibly due to a change in the local environment near the cysteine residue (C10), glycosylation was reported for TTR monomers of V30M (Teixeira and Saraiva, 2013), D18G (Sato et al., 2012), and G101S (Wakita et al., 2018), the mutated form of wild-type TTR with the 30th, 18th, and 101st amino acid residues valine, aspartic acid, and glycine replaced by methionine, glycine, and serine, respectively.

TTR tetramers were thought to be the primary form in the human body (Buxbaum and Reixach, 2009). However, they were not detected in subretinal fluid, possibly due to instability under our experimental conditions or actual absence. TTR tetramers were known to partially dissociate into monomers in the presence of organic solvent (Hall et al., 2013). Organic solvent methanol used in our study could have promoted the dissociation. Typically reported as 56 kDa, the accurate mass of TTR tetramers for recombinant protein samples was measured as 55, 602 ± 3 Da (McCammon et al., 2002), or a mass of 13,900 ± 1 Da for each monomer component, close to sodium ion adduct of cysteinylated TTR (theoretical mass of 13,901.89 Da). Sodium-ion adducts are common for mass spectrometry of biological samples. With such a disulfide-bonded modification at the only cysteine residue in TTR, an intermonomer disulfide bond will not be possible. Therefore, TTR tetramers were most likely formed by noncovalent interaction, which is also consistent with the observation that MS/MS of TTR tetramers predominantly generated monomers (Hall et al., 2013). Otherwise, TTR dimer peaks should be more intense upon dissociation of tetramers because the covalent disulfide bond is more difficult to break than noncovalent interactions. It would be interesting in future studies to characterize the heterogeneity in the structure and function of the TTR tetramers composed by the various monomeric proteoforms in the subretinal fluid.

As discussed above, we detected modified TTR monomers and putative dimers in the subretinal fluid. These modifications could be critical for stability and function in transport and antioxidation. At least six modifications, including signal peptide removal, S-Cys, S-CysGly, S-GSH, C10G, and sulfonation, were detected for monomers, possibly resulting from dissociation of tetramers. In contrast, the putative TTR dimers most likely contain glycosylation and intermonomer disulfide bonding modifications. Of the two common types of glycosylation, i.e., N-linked glycosylation and O-linked glycosylation, N-linked glycosylation was detected at N98 (or N118, including the signal peptide) of TTR in normal serum samples (Liu et al., 2005) and recombinant D18G TTR monomer (Sato et al., 2012). However, the exact composition of the N-glycosylation is still unknown. O-linked glycosylation of TTR has not been reported. The putative and relevant reported TTR dimers apparently differed significantly in modifications and configurations from those in TTR tetramers and monomers. As a result, these dimers were probably not from the dissociation of TTR tetramers but instead present in the solution independently and may have functions different from the tetramers, such as transporting other types of lipids or metabolites.

4.3. Implications for retinal diseases

Further characterization of TTR in the subretinal fluid will facilitate understanding various retinal diseases. Of particular interest is subretinal drusenoid deposit (SDD, or reticular pseudodrusen), an extracellular deposit between photoreceptors and RPE that is common in age-related macular degeneration. These deposits also appear in inherited retinopathies affecting the outer retina and Bruch’s membrane as well as systemic vitamin A deficiency (Curcio et al., 2013; Spaide et al., 2018; Chen et al., 2020). SDD is known to contain unesterified cholesterol, apolipoprotein E, and vitronectin and appears first in areas where rod photoreceptors are abundant. It is speculated that SDD is part of an outer retinal lipid cycling system with specific components reflecting cone- and rod-specific support systems, necessarily including multifunctional carrier proteins; TTR could serve such a role. Our immunohistochemical and gene expression results strongly support that RPE is the source of TTR in the retina, comporting with reports that TTR protein concentration (Getz et al., 1999) and DNA/mRNA expression (Buxbaum and Reixach, 2009) are high in RPE cell bodies. Our results are consistent with prior studies showing that TTR is secreted into interphotoreceptor matrix (Ong et al., 1994; Adler and Edwards, 2000). Considering TTR’s role in transporting low solubility retinoids, lipids, hormones, and metabolites, a change of TTR proteoform homeostasis could significantly decrease binding some of these small molecules and eventually contribute to a deposit. Studies to test these hypotheses are ongoing.

In amyloidosis, a soluble protein undergoes a modification, becomes misfolded and insoluble, and accumulates in the extracellular space, and TTR amyloidosis is one of the most prevalent systemic conditions. Familial TTR amyloidosis due to inherited mutations manifests in the eye, affecting primarily vitreous (Reynolds et al., 2017; Iakovleva et al., 2021; Minnella et al., 2021; Davila, 2022). Age-related TTR amyloidosis occurs in the heart and other organs (Ruberg and Berk, 2012; Tasaki et al., 2021), likely caused by increased oxidative stress with aging (Kingsbury et al., 2007; Zhao et al., 2013) that leads to formation of amyloid-prone disulfide-bonded TTR dimers (Kingsbury et al., 2007). Whether age-related amyloidosis also occurs to the eye is a subject for future research.

5. Conclusions

With top-down MS, we identified for the first time to our knowledge many proteoforms of abundant proteins in subretinal fluid. The relative amounts of TTR, its glutathionylated proteoform, and a putative TTR glycosylated dimer, were found much higher than their counterparts in serum and cerebrospinal fluid. Further studies on the composition, function, and quantities of these species and other proteoforms in subretinal fluid will inform mechanisms, diagnostic methods, and treatment strategies for age-related macular degeneration, familial amyloidosis, and other retinal diseases involving dysregulation of physiologic lipid transfer and oxidative stress.

Abbreviations

ABC-AP

avidin-biotin-alkaline phosphatase complex

AMD

age-related macular degeneration

AP

alkaline phosphatase

ApoA1

apolipoprotein A-I

β₂M

β₂ microglobulin

ApoC1

apolipoprotein C-I

C10G

the mutated form of a wild-type protein with the 10th residue (not counting the first 20 residues in the signal peptide of this and other transthyretin mutated forms discussed in this paper) cysteine replaced by glycine

CID

collision-induced dissociation

COX4

cytochrome c oxidase subunit 4

D18G

the mutated form of a wild-type protein with the 30th amino acid residue aspartic acid replaced by glycine

DA

truncated form of a wild-type protein with a loss of the N-terminal aspartic acid and alanine residues

G101S

the mutated form of a wild-type protein with the 101st amino acid residues glycine replaced by serine

Glc

glucose modification or glucosylation

Glyco

glycosylation

K107del

the mutated form of a wild-type protein with the deletion of the 107th residue lysine

m/z

mass-to-charge ratio

MS

mass spectrometry

MS/MS

tandem mass spectrometry

RPE

retinal pigment epithelium

RRD

rhegmatogenous retinal detachment

S-Cys

cysteinylation

S-Cysgly

cysteinylglycylation

S-GSH

glutathionylation

S-Sulfo

sulfonation

SDD

subretinal drusenoid deposits

TTR

transthyretin

V30M

the mutated form of a wild-type protein with the 30th amino acid residue valine replaced by methionine