RESEARCH GRANTS 2019-21-all cycl : simplebooklet.com

Molecular OmicsRESEARCH ARTICLEPlease do not adjust marginsPlease do not adjust marginsReceived 00th January 20xx,Accepted 00th January 20xxDOI: 10.1039/x0xx00000xAnalyses of Pseudoexfoliation aqueous humor lipidome Vanessa Collaoa-c, Jada Morrisa-c, Muhammad Zain Chauhana-d, Leila Abdelrahmana-c, Jose Marίa Martίnez-de-la-Casae, f , Beatriz Vidal-Villegase, Barbara Burgos-Blascoe, and Sanjoy K. Bhattacharya a-c*Pseudoexfoliation syndrome (PEX) is a systemic disorder that manifests as fluffy, proteinaceous fibrillar material throughout the body. In the eye such deposits result in glaucoma (PEXG), due to impeding aqueous humor outflow. Serum lipid alterations and increased lipid peroxidation have been reported in PEX. We report first ever comprehensive lipid profiling of the aqueous humor (AH) of PEXG. Our untargeted lipidomic analysis of 23 control, 19 primary open angle glaucoma (POAG), 9 PEX, and 14 PEXG AH resulted in identification of 489 lipid species within 26 lipid classes across PEX, PEXG, POAG, and control AH. Multiple cholesterol esters (ChE), phosphatidylcholines (PC), triglycerides (TG), and ceramides (Cer) were present in higher concentrations in the PEXG AH than all other groups. The CerG2GNAc1(d34:1) was enriched in control samples and depleted both in PEX and PEXG samples. Machine learning prediction with three supervised logistic regression binary classification tasks showed 1) POAG vs control, with 86% accuracy 2) PEXG vs control, with 71% accuracy and 3) PEX vs control, with 86% accuracy, respectively. In conclusion, analysis showed that control (mean peak area 13.54 ± 6) had, on average, greater total lipid content than PEX, PEXG, and POAG AH samples. Elevations in Apolipoprotein A-I (APOA1) correlated to increased abundance of PC lipid species in the AH of patients with PEXG. The PC (18:0/18:2), PC (36:2), and PC (34:1e) are in low concentrations for PEX but highly concentrated in PEXG, despite both having similar material deposits, suggesting they are fundamentally different in composition. IntroductionPseudoexfoliation syndrome (PEX) is a generalized basement membrane disorder that manifests as white, fluffy, proteinaceous fibrillar material throughout the body. It is a systemic, age-related disorder of the extracellular matrix with ocular manifestations 1. PEX is asymmetrically bilateral, meaning each eye may begin showing signs of PEX at different time points, with unilateral being the precursor to bilateral involvement 2,3. Pseudoexfoliation material (PEXM), resembling amyloid, is commonly found deposited on anterior segment structures bathed by the aqueous humor (AH), including the iris, lens capsule, and other structures of the anterior segment. 1,3. The exact PEXM composition remains to be determined. It has been hypothesized to consist of elastic microfibrils and components from smooth muscle cells, vascular endothelial cells, and melanocytes3. These cells may contribute to the active production of PEXM. AH is produced by the ciliary body (CB) epithelium, bathes the entire anterior chamber and filters out of the anterior chamber through the trabecular meshwork (TM) and Schlemm’s canal (SC) 4. Accumulation of PEXM in the TM impedes AH outflow causing fluctuations in intraocular pressure (IOP) and eventually elevated IOP 3. Cumulative exposure to elevated IOP over a period can cause damage to the optic nerve and lead to the development of pseudoexfoliation glaucoma (PEXG), a secondary open-angle glaucomatous optic neuropathy3. PEXG is the most common cause of secondary open-angle glaucoma, affecting approximately 2% of the US population greater than 50 years old. IOP is the only modifiable factor regulated by the inflow and outflow of AH targeted when monitoring PEXG progression3. The PEX is a systemic disease. Therefore, blood serum has been studied for its lipid and lipoprotein composition among patients with PEX or PEXG 5. The monocyte count/high-density lipoprotein (HDL) ratio (MHR) and lymphocyte count/monocyte count ratio (LMR) studies found the mean MHRs to be significantly higher in patients diagnosed with PEX and PEXG than the control group. The mean LMRs were also lower among the PEX and PEXG groups than controls, but the difference was not significant 5. Several PEX patients have been noted to also suffer from cardiovascular disease, dyslipidemia or hypercholesteremia, that warranted an investigation into atherogenic indices in patients with PEX using traditional serum lipid profiles and non-traditional serum lipid ratios 5-8. Traditional serum lipid profiles included total cholesterol (TC), TG, low-density lipoprotein cholesterol (LDL-c), high-density lipoprotein cholesterol (HDL-c) and non-HDL-c while non-traditional serum lipid ratios were TC/HDL-c, TG/HDL-c, LDL-c/HDL-c and non-HDL-c/HDL-c 6. Overall, this study did not show a strong relationship between PEX and an increased risk for vascular disease 6. The lipids and C-reactive protein serum levels in patients with PEX have also been evaluated only to conclude there were no significant differences in concentration of a.Bascom Palmer Eye Institute, University of Miami, Miami, Florida, USAb.Miami Integrative Metabolomics Research Center, University of Miami, Miamic.Vision Science and Investigative Ophthalmology Graduate Program, University of Miami, Miami, Florida, USAd.Department of Ophthalmology, Jones Eye Institute, University of Arkansas for Medical Sciences Little Rock, AR e.Servicio de Oftalmología, Hospital Clínico San Carlos; IdISSC, Madrid, SpainDepartamento de Inmunología, Oftalmología y ORL, Facultad de Medicina, Universidad Complutense de Madrid; Madrid, Spain. † Electronic Supplementary Information (ESI) available. See DOI:Page 1 of 10 Molecular OmicsMolecular Omics Accepted ManuscriptPublished on 11 April 2022. Downloaded on 4/12/2022 1:20:24 AM. View Article OnlineDOI: 10.1039/D1MO00495F

Page 3

ARTICLE Mol. Omics2 | Mol. Omics., 2021, 00, 1-3 This journal is © The Royal Society of Chemistry 20xxPlease do not adjust marginsPlease do not adjust marginslipids and high-sensitivity C-reactive protein between PEX and control groups 7. However, an independent study determined the significance of atherosclerosis in the development of PEX and PEXG and found that the serum lipid profile was disturbed significantly in patients with PEX and PEXG 8.The PEXG warrants study of local fluid and tissues to complement systemic studies. Lipid metabolites, specifically their peroxidation products are hypothesized to be biomarkers for oxidative stress. Open-angle glaucomas (OAG) show signs of such oxidative stress systemically and locally 9. Lipids are soluble molecules that are major components of cell membranes. Lipids also participate in cell signaling 10,11. Repertoire of cellular lipids undergo dynamic changes because of physiological, pathological, and environmental conditions 11. The lipid peroxidation products glutathione (GSH), glutathione disulfide (GSSG), and Thiobarbituric acid reactive species (TBARS) in PEX AH samples have been measured and compared with age-matched control AH. A decreased GSH, an increased GSSG resulting in a 1.7-fold decrease GSH/GSSG was found in PEX AH samples. The observed higher levels of TBARS and GSSG indicated high oxidative stress in the PEX AH 12. Despite several genetic/genomic 13-15 proteomic 16, and metabolomics studies17,18, lipidomic analyses on PEXG samples are yet to be performed. PEXG offers excellent characterized clinical tissues. PEXG allow easier access to the anterior segment for non-invasive imaging and IOP measurements as well as visual acuity measurements. For these reasons we undertook analyses of PEX and diagnosed PEXG patients along with controls and primary OAG (POAG) patients as a distinctly different group of glaucoma. With current advances in machine-learning we also thought it would be appropriate to test whether lipids alone can serve as predictors of the PEXG group from among other choices.ResultsPatient characteristicsWe obtained clinically characterized cataract control, PEX, PEXG and POAG patients. The patient age, gender and other available details are summarized in Table 1 (see Table S1 for further details). The average age of patients was 71, 74, 78, and 76 years old for control, POAG, PEX, and PEXG, respectively. The gender distribution varied between diseased groups with a greater proportion of males in POAG group (57.9%) and a greater number of females in control, PEX, and PEXG groups (56.5%, 55.6%, and 57.1%). We also recorded a few comorbidities such as occurrence of dyslipidemia and hypercholesterolemia. About 22.2% of PEX and 50% of PEXG suffered from dyslipidemia and hypercholesterolemia compared to 30.4%, 26.3% for control and POAG, respectively.Lipid Identification and ProfilingOur untargeted lipidomic analysis of 23 non-glaucomatous control, 19 POAG, 9 PEX, and 14 PEXG AH with 13 deuterated lipid internal standards for normalization among the lipid classes resulted in the combined identification of 489 lipid species within 26 lipid classes across PEX, PEXG, POAG, and control AH. The mean total lipid content in the AH across samples showed that control AH (mean peak area 13.54 ± 56.1) had, on average, greater total lipid content than PEX (4.21 ± 10.90), PEXG (9.08 ± 25.97), and POAG (5.66 ± 15.75) samples (p =<0.001) (Fig. 1A). Data were subsequently log2 transformed and mean centered for analyses (Fig. S1A). The lipid classes with the highest concentration in the AH were found to be triglycerides and phosphatidylcholines species (Fig. S1B). Moreover, this correlated to having the most represented structural categories as glycerolipids (GL) and glycerophospholipids (GP) (Fig. S1C). It was found that the total chain length of lipids in the AH is highly variable, with 36 chain lengths having the highest representation (Fig. S2A). Total double bonds of lipids in the AH were found to be highest among lipids with zero and two double bonds (Fig.S 2B). Combined k-means clustering, and principal component analysis (PCA) revealed 4 overlapping clusters (Fig. 1B). Principle component (PC) 1 explained 35.1% of the variance and PC2 explained 15.1% of the variance across all samples. The identified lipids were subsequently analyzed via heatmap, partial least-squares discriminant (PLS-DA) and variable importance in projection (VIP) score analysis. The heatmap using a Ward clustering algorithm shows the concentration of the top 70 lipid species found across all groups and their relative concentrations (Fig. 1C). Notably, multiple cholesterol esters (ChE), phosphatidylcholines (PC), triglycerides (TG), and ceramides (Cer) were present in higher concentrations for the PEXG AH samples. Some of the lipids found in high concentrations in the PEXG samples are ChE(16:0), ChE(20:3), ChE(18:1), ChE(18:3), ChE(22:6), ChE(18:2), ChE(20:4), PC(16:0/16:0), PC(16:0/18:2), TG(18:1/18:1/20:4), and Cer(t18:0/24:0). The top 9 significantly different lipid species from two-way ANOVA are shown in Figure 1D. Notably, ChE(18:1) and Cer(t18:0/24:0) were also found to have differential abundances between dyslipidemia and non-dyslipidemia groups as well as between glaucoma groups (Fig. S3 A,B). We subsequently ran an ANOVA-simultaneous component analysis to resolve the impact of patients with systemic dyslipidemia on the AH lipidome. We found that the greatest variation occurred between control and PEXG groups (Figure S4A). Interestingly, in patients with dyslipidemia, the variation in the lipidome was noted to be the greatest in the POAG and control groups, while the PEX and PEXG groups were most similar for these patients (Figure S3B).GroupAverage Age (years)GenderDyslipidemia (%)Total Statin UsersTotal Statin Users with DyslipidemiaControl 70.810 Male /13 Female 30.4% 7 7POAG 74.211 Male / 8 Female 26.3% 5 4PEX 78.34 Male /5 Female 22.2% 3 2PEXG 75.86 Male /8 Female 50.0% 7 7Table 1. Summary of AH donorsPage 2 of 10Molecular OmicsMolecular Omics Accepted ManuscriptPublished on 11 April 2022. Downloaded on 4/12/2022 1:20:24 AM. View Article OnlineDOI: 10.1039/D1MO00495F

Page 4

Mol. Omics ARTICLEThis journal is © The Royal Society of Chemistry 20xx Mol. Omics., 2021, 00, 1-3 | 3Please do not adjust marginsPlease do not adjust marginsThe PLS-DA 3D plot showed that component 1 explained 35.1% of the variance which revealed a distinct separation of control samples from PEX, PEXG, and POAG with one sample as an outlier (Fig. 2A). Further examining the top 15 lipid species through VIP analysis, it was found that several PC and SM species were enriched in PEXG samples. The CerG2GNAc1(d34:1) had the highest VIP score at 4.0 and was enriched in control samples and notably depleted in PEX and PEXG samples, which was one similar feature between these two groups. In addition, Cer(t18:0/25:0) and SM(d36:2) were both highly enriched in PEX and PEXG compared to other groups. It is reasonable to hypothesize that the lipid profile of PEX and PEXG AH samples would be similar. However, the heatmap and VIP score plot showed contrasting concentrations between PEX and PEXG samples. Both conditions have PEXM, but their lipid profile differs from these few samples. For example, PC(18:0/18:2), PC(36:2), and PC(34:1e) are in low concentrations for PEX but highly concentrated in PEXG (Fig. 2B). a)-4 -2 0 2024Log2 (FC)-Log10 (P-value)c)b)Fig. 3 Differential Expression of Lipids Between Control and PEXG AH. (a) Volcano plot of significantly different lipids between control (left) and PEXG (right). (b) List of differentially regulated lipid species in control (elevated in left) and PEXG (elevated in right). Lipids with significantly different abundances are denoted by intensity of red circles (y-axis). (c) List of differentially regulated lipid classes in control (elevated in up) and PEXG (elevated in down). Lipids with significantly different abundances by red bars (y-axis). Integrated Proteomic and Lipidomic AnalysisIn noting the greatest variation in the lipidome between control and PEXG groups, we subsequently followed-up with an integrated proteomic and lipidomic analysis focusing on differentially regulated molecules for these two groups. A volcano plot revealed several lipid species that were enriched and depleted in the PEXG group (Fig. 3A). The most significantly enriched lipid species and class in PEXG was found to be ChE, with ChE (20:5) being the most significantly enriched lipid species and Cer (18:1/24:1) being the most depleted (Fig. 3B). These findings correlated to ChE and Cer lipid classes being of the most enriched and depleted classes in PEXG AH, respectively (Fig. 3C). We found that lipids with chain lengths of 20 were the most significantly enriched and chain lengths of 48 were the most depleted in PEXG AH compared to control (Fig. 4B). In addition, we found that lipids species with 5 double bonds were the most enriched and 1 double bond was the most depleted (Fig. 4C). Interestingly, we found that ChE lipids species that contained 5 double bonds was significantly elevated when compared to controls (Fig. 4D). Lipid-related gene set enrichment analysis from differentially regulated lipids in PEXG was conducted utilizing KEGG and GO databases. AMPK, cholesterol metabolism, and sphingolipid signaling pathways were the most significantly impacted pathways in PEXG compared to controls (Fig. 4A). a)b)c)d)Fig. 4 Lipid Related Gene Enrichment and Lipid Characteristic Analysis. (a) Lipid-related gene set enrichment analysis utilizing KEGG pathways. The circular network diagram displays the significant relationships (P < 0.05) between lipid-related pathways and genes based on their lipid classes that were significantly differentially regulated in PEXG. Color of the node is according to –log10(P-value) and their sizes represent the number of lipid-related genes in the pathway. The width of the line indicates the value of gene similarity between the pathways. (b) Differentially regulated lipid chain lengths in control (elevated in up) and PEXG (elevated in down). Lipids with significantly different abundances according to threshold are denoted by red bars (y-axis). (c) Differentially regulated lipids according to double bond character in control (elevated in up) and PEXG (elevated in down). Lipids with significantly different abundances according to threshold are denoted by red bars (y-axis). (d) Total lipid between control and PEXG. We subsequently interrogated the metabolic flow and regulatory proteins of the differentially regulated lipids in PEXG utilizing metabolic networks between small molecules within Reactome Network topologies (Figure S5A)19. We noted several biochemical reactions in sphingomyelin conversion pathways were differentially regulated. We then integrated previously published AH proteomic data from publicly available datasets of patients with PEXG and lipidomic findings to determine if proteomic and lipidomic AH analysis correlated. We found that elevations in Apolipoprotein A-I (APOA1) correlated to increase abundance of PC lipid species in the AH of patients with PEXG (Fig. 5). Fig. 5 Multi-omics network analysis of AH proteome and lipidome. The network included differentially expressed proteins and lipids in PEXG aqueous humor. Proteins are labeled in red and lipids in yellow. Increases in APOA1 (blue outer rim) were correlated to changes in phosphatidylcholine lipid species (yellow circle) in PEXG. Machine-learning predictionThe log2 transformed values of lipid quantification were placed in three supervised binary classification tasks stratified by patients: 1) POAG vs control, 2) PEXG vs control, and 3) PEX vs control vectors. The feature vectors were fed into a logistical regression statistical classifier and 10-fold cross-validation was performed. The accuracy of correctly predicting these three groups were 86%, 71% and 86% respectively (Fig. 6). Page 3 of 10 Molecular OmicsMolecular Omics Accepted ManuscriptPublished on 11 April 2022. Downloaded on 4/12/2022 1:20:24 AM. View Article OnlineDOI: 10.1039/D1MO00495F

Page 5

ARTICLE Mol. Omics4 | Mol. Omics., 2021, 00, 1-3 This journal is © The Royal Society of Chemistry 20xxPlease do not adjust marginsPlease do not adjust marginsPOAG vs control PEXG vs controlPEX vs control10-fold CrossVal: 86% Accuracy10-fold CrossVal: 71% Accuracy10-fold CrossVal: 86% Accuracya) b) c)Fig. 6 Confusion matrices for classifying the three different conditions. A 0 indicates a control sample and a 1 indicates a sample with the respective condition. The resulting accuracy is reported after 10-fold cross validation with the logistic regression classifier. (a) POAG vs Control. (b) PEXG vs Control. (c) PEX vs Control. On average, the logistic regression was highly specific but lacked sensitivity.After training the models, the feature importance of the lipids was determined for each classification task. The five most-important features for each of the three task stratifications have been presented (Fig. 7). POAG vs control PEXG vs controlPEX vs controla) b) c)Fig. 7 The Logistic Regression Feature importance. (a) POAG vs Control, (b) PEXG vs Control and, (c) PEX vs Control. The importance is based on the absolute value of the weight coefficient assigned to each lipid feature in thelogistic regression. For POAG and PEXG the most important lipid for classification was Cer(d36.0) and Cer(m36.0), respectively. Cer(m36.0) also played an important role in the PEX classifier.We rely on the coefficients associated with each feature in the logistic regression’s parameters to signify importance. The absolute value of the coefficients has been used when calculating the feature’s contributions. DiscussionPEXG is a secondary glaucoma whose etiology is understood in terms of pathological deposit. It is the proteinaceous deposit that obstructs and impedes the AH outflow from anterior eye chamber resulting in elevation of IOP, whose cumulative exposure over time in turn damages the optic nerve3. PEX is a systemic disease, often associated with cardiovascular dysfunction; therefore, systemic lipid profiles such as serum lipid profile aberrations have been associated with PEX and PEXG 5-8. The PEXG subjects have been subjected to extensive genetic 13,14 and/or genomic 15 analyses. The AH as well as TM samples of PEXG have also been subjected to proteomic 16, and metabolomics studies 17,18. However, unlike systemic analyses of serum for lipids, lipidomic analyses on PEXG AH remained to be performed, which motivated us to perform the current study. The PEXG metabolite profile showed a number of unique metabolites when compared to controls, which were mainly consisted of amino acids and other non-lipid entities due to analyses being performed on water soluble rather than organic phase metabolites therefore excluding all lipids 17. PEX proteomic studies have outlined the pathways in which genes, proteins and the environment influence the development of PEX 20 . Lipidomic studies previously published have focused on sphingolipids, phospholipids, cholesterol, and glycosphingolipids of glaucomatous tissues (i.e., POAG) and AH 21-24. In this study, we expanded our analysis to include PEX, PEXG, POAG and controls with the inclusion of additional lipids like cholesterol esters, TGs, diacylglycerols, and ceramides. Currently, there is not enough available research to provide a definitive answer to explain how the proteome, lipidome, and metabolome converge. Analysis of the PEX lipidome is the first of its kind and an essential first step towards constructing a metabolite network with proteins and lipids unique to PEX. Over 400 lipid species were identified in our analyses (Fig. 1). However, we focused on the top 70 lipids because of their prominent contrasting patterns. Although these lipids are found in the AH, they may have originated from the TM, CB, lens, or other anterior segment structures 21,25. The CB actively produces AH 26. An avenue for potential future investigation is to analyze the secretions from each part of the anterior segment structures from PEX and PEXG eyes to determine whether the lipid secretion change from any region contribute to observed differences in AH profile as was performed in a prior study on POAG using in vitro and in vivo methods 26. The limitations of PEXG are accurate knowledge of onset and severity. Very frequently PEXG is observed in the clinic at an already advance stage. In eyes with asymmetric bilateral presentation, lack of timely monitoring of the unaffected eye often results in glaucoma advancement in eyes that is presented as non-glaucomatous (PEX only) during initial visits. Our AH patient data revealed that many patients have a personal history of dyslipidemia (Table 1, Table S1). Dyslipidemia is a risk factor of the cardiovascular disease characterized by increased TG, low-density lipoprotein (LDL) levels, and decreased high-density lipoprotein (HDL) levels 27. Patients of PEXG reports interference of IOP when on some select cholesterol lowering medication such as Ezetimibe. However, systemic studies are yet to be carried out to validate these patients’ reported observations. Lipid pathway impairment have been shown to result in pigment dispersion28. In the heatmap, TGs are in high concentration for some of the PEXG AH samples and half of those samples came from patients with dyslipidemia (Fig. 1C). Several studies have explored the relationship between PEX and vascular disease involving abnormal lipid levels but result in contrasting conclusions about the significance of the relationship. Blood serum with PEX or PEXG has shown higher monocyte count/high-density lipoprotein (MHR) 5. The MHR has also been postulated as a predictive biomarker of inflammation 5. This is consistent with the finding that the serum lipid profile was disturbed significantly in patients with PEX and PEXG 8. Conversely, lack of such correlation also has been found in other studies. Since dyslipidemia is a risk factor for cardiovascular disease and many studies have differing conclusions about its relationship with PEX, this area still requires more comprehensive future research 5-8. At present cross talk of records between dyslipidemia and ophthalmologic findings are suboptimal. Better integration of detailed records of both these aspects will be very helpful to derive insights. Unlike serum AH is a local and clear fluid, generated by CB. In our study, the lipid profiles of PEX and PEXG showed increased and decreased lipid concentrations for specific classes, which corresponded to the patients with a history of dyslipidemia. There seems to be a relationship between dyslipidemias and PEX, therefore it would be beneficial to conduct a future study comparing AH lipid profiles from PEX patients to controls. A potential link between abnormal cholesterol metabolism and PEX is indicated by association of cholesterol metabolizing CYP39A1 variants with exfoliation syndrome29. An intriguing aspect is that Polymorphism Phenotyping version 2 (Polyphen-2) HumDiv software identified several damaging (for example, F175L, C31Y, D157N) and non-damaging or benign variants (H164L, N183D, E224K, M244L, M426V, M73V) enzymatic activities are within similar range, which also warrants greater understanding of the lipidome in affected individuals29.Lipid changes have been found in AH as well as TM of other glaucomas. A comparative study found most phospholipids being common between control and POAG TM but a few of them unique to either group 25. Targeted lipidomic analysis has been performed for POAG but not for PEX or PEXG24. Another independent lipidomic study revealed significant increases in the concentrations of 14 sphingomyelin (SM) and 5 cholesteryl ester (ChoE) species from open angle glaucoma (OAG) AH samples 9. The univariate analysis of the Page 4 of 10Molecular OmicsMolecular Omics Accepted ManuscriptPublished on 11 April 2022. Downloaded on 4/12/2022 1:20:24 AM. View Article OnlineDOI: 10.1039/D1MO00495F

Page 6

Mol. Omics ARTICLEThis journal is © The Royal Society of Chemistry 20xx Mol. Omics., 2021, 00, 1-3 | 5Please do not adjust marginsPlease do not adjust marginsprofile showed, 37 out of 110 lipids to undergo statistically significant concentration changes in OAG compared to controls, with most of the species belonging to chemical groups diacylglycerophosphocholines (DAPC) and 1-ether, 2-acylglycerophosphocholines (MEMAPC) 9. Members of an endogenous class (eicosanoids) of lipids known as prostaglandins originally found in the iris are the only lipids that currently works best to lower IOP 22. These small molecules are derived from arachidonic acid and are pro-inflammatory. Prostaglandins play a role in inflammation, pain modulation, allergies, bone formation, and many other biological processes. Prostaglandin analogs (latanoprost, bimatoprost, tafluprost, and travoprost) are the leading medication for glaucoma treatment that can lower IOP by increasing AH outflow via the uveoscleral pathway instead of the conventional pathway 22,30. The conventional pathway begins with the production of AH by the CB and then flows from the posterior chamber to the anterior chamber to drain through the TM and SC. The uveoscleral pathway, which accounts for 5-15% of IOP regulation, drains through the ciliary muscles and exits via the supraciliary space 30. Plasmalogen and a phospholipid have been found to serve as predictors of glaucoma severity in POAG 31. The biological role of these lipids is yet to be investigated. Our current investigation also has identified lipids whose biological role need to be understood. The metabolome wide association studies are consistent with hydrophobic metabolites or lipids to be associated with open angle glaucoma. Although dyslipidemia and other conditions were not segregated in those analyses 32. Currently, there is no clinical prediction for the bilateral development of PEX. Since in a large number of individuals the disease presentation is asymmetric, accurately predicting those with potentially rapid bilateral development will be of clinical importance for aggressive follow up and treatment. As a first step towards such investigation, we used machine-learning on our datasets to evaluate if AH lipidomic profile could accurately predict the type of glaucoma (Fig. 6, 7). In humans, the AH sampling of the unaffected eye despite presence of visible material is prohibited. Currently, there are a few animal models that capture anterior segment deposit formation but none of them capture the clinical features of human PEXG accurately 33. The limited animal models are also not freely available. However, the AH samples collected for PEX and PEXG from cataract or other surgeries are great resources together with serum lipid analyses to determine predictability of disease severity. Together with other omic studies they offer promise of insight into pathways and molecules that can alter the patient care landscape of PEXG glaucoma. The identification of molecules is also likely to provide important insight into their biological role in the function of cells in anterior eye segments. MethodsData reporting Samples for all MS-based analysis were randomized prior to data acquisition. To estimate an appropriate group size for this study, statistical analysis was performed incorporating our preliminary data employing SAS/STAT software in consultation with our Biostatistical core experts. The following assumptions were made that the AH samples are independent for each protein/lipid, that equal sample sizes exist between comparative groups and that the proportion of a given protein/lipid being present in the total of groups (control, singular disease groups) may or may not be equal to that in a given group. The proportion of proteins/lipids that could be detected in different groups are based on our preliminary data. The minimum number of samples needed per group for coverage of 100% in groups (concomitant with 83.3% lipids in a singular group) to achieve 80% power (type I error or false positive rate 0.05) and 90% power (type I error rate at 0.01) are 4 and 6, respectively. With these group sizes we expect to cover 83-100% proteins/lipids across different groups covering over 95% unique proteins/lipids. Similar size determination was made for control and different groups. The AH analyses were randomized and performed by two different mass spectrometry operators with identities unknown to them. Analyses was repeated using three different operators with identity of samples masked to them. Machine-learning was run by two different operators at two different time points with the same dataset. One of them was an independent validation by someone not involved in the study. Human donors and chemicals All AH samples collected came from human donors during cataract surgery. The study protocol was approved by the Hospital Clinico San Carlos Ethics Committee (21/165-E), Madrid, Spain. AH was collected during essential ophthalmic procedures that allowed paracentesis after obtaining informed consents from human subjects without collecting identifiers but all disease and medication history. All experiments were performed in compliance with relevant laws following institutional guidelines. A total of 66 samples were collected which included non-glaucomatous controls and those with pseudoexfoliation syndrome (PEX), pseudoexfoliation glaucoma (PEXG), and primary open-angle glaucoma (POAG). The clinicians provided samples from the Department of Ophthalmology of Hospital Clínico San Carlos in Madrid, Spain. The patient samples consisted of 19 POAG, 9 PEX, 14 PEXG, and 23 controls (see Table 1 and Table S1). Approximately 50-120µl volume of AH was collected by paracentesis and stored in -80ºC immediately upon acquisition until analysis. AH was extracted after 7-10 days and analyzed immediately. Controlled analyses was performed to ensure no significant alteration of lipid profile due to these steps34. Ultra-high performance liquid chromatography mass spectrometry (UHPLC-MS) using the Thermo Scientific Vanquish Horizon system and Q Exactive mass spectrometer were used to analyze the AH samples. LC-MS grade chemicals used for the lipid extractions and mass spec analysis included chloroform, methanol, water, acetonitrile, and isopropyl alcohol. Lipid Extraction Lipids were extracted from POAG, PEX, PEXG, and control AH samples using the Bligh and Dyer method 35. EquiSPLASHTM Lipidomix quantitative mass spec internal standard was spiked into each sample to normalize the lipids (Figure S7 and S6). For the extraction, a 2:1 chloroform: methanol solution with 100 µL of water was added to promote phase separation followed by centrifugation at 4°C at 14,000 rpm for 20 minutes. After centrifugation, separation was observed, and the extracted lipids were collected from the lower, organic chloroform phase with a syringe. Once the lipids were isolated and transferred to 2 mL glass vials, they were dried in a Speed-Vac for about 90 minutes. The upper, aqueous, water/methanol phase containing most of the proteins was collected and stored for a Bradford protein assay to measure protein concentration in the future. When the samples were dehydrated, Page 5 of 10 Molecular OmicsMolecular Omics Accepted ManuscriptPublished on 11 April 2022. Downloaded on 4/12/2022 1:20:24 AM. View Article OnlineDOI: 10.1039/D1MO00495F

Page 7

ARTICLE Mol. Omics6 | Mol. Omics., 2021, 00, 1-3 This journal is © The Royal Society of Chemistry 20xxPlease do not adjust marginsPlease do not adjust marginsthey were flushed with argon gas for storage at -80°C until ready for mass spec analysis. To prepare the samples for mass spec analysis, the dried lipids were resuspended in a 1:1 acetonitrile: isopropyl alcohol solution, sonicated for 30 minutes, and aliquoted into mass spec vials.Ultra-high performance liquid chromatography (UHPLC) – mass spectrometry The Vanquish Horizon UHPLC system (Thermo) used reversed-phase chromatographic separation with an Accucore Vanquish C18+ column: 1.5µm, 2.1x150 mm (Thermo). The column and sample tray temperature were held at 55°C and 4°C. The two solvents utilized were composed of LC-MS grade chemicals. Solvent A was a 50:50 mixture of acetonitrile:water, 0.1% formic acid (FA) with five mmol of ammonium formate. Solvent B was a mixture of 88% isopropyl alcohol, 10% acetonitrile, and 2% water with five mmol of ammonium formate. The flow rate was 260 µL min-1 with an injection volume of 10 µL. The gradient was 0-10% solvent B for 0-2 minutes, 30% solvent B over 2-3.5 minutes, 50% solvent B over 3.5-7 minutes, 60% solvent B over 7-17 minutes, 70% solvent B over 17-18 minutes, 80% solvent B over 18-20 minutes, 95% solvent B over 20-22 minutes, 100% solvent B over 22-28.1 minutes, and 10% solvent B over 28.1-30 minutes. The lipids were analyzed using a Q Exactive Mass Spectrometer (Thermo) with a heated electrospray ionization source (HESI) attached that operated in positive and negative ion modes separately. The HESI source spray voltage was 2.50 kV (negative mode) and 4.00 kV (positive mode), and the heated capillary temperature was held at 325°C for both negative and positive modes. The sheath gas flow rate was 30 (negative mode) and 35 (positive mode). The auxiliary gas was 15, and the S-lens radiofrequency was 70 for both negative and positive modes. Full scan resolution was 70,000 with a scan range of 300-900 m/z (negative mode) and 250-1200 m/z (positive mode). The automatic gain control (AGC) target was 1 x 106, and the maximum injection time (IT) was 100 ms. Data-dependent parameters used a resolution of 17,500, AGC target of 1x105, maximum IT of 50 ms, 1.0 m/z isolation window, an intensity threshold of 2x104 (negative mode) and 1.6x105 (positive mode) with a dynamic exclusion time of 8 s, and the normalized collision energy (NCE) was set to 20, 30, and 40 in negative and positive modes. The lipids were identified using MS and MS/MS ions (Fig. S6). Following established procedures EquiSPLASHTM Lipidomix quantitative mass spec internal standard (Fig. S7) spiked into each sample enabled identification of MS and MS/MS ions to identify lipids. Proteomic datasets The quantitative proteomic datasets were presented as part three studies: Kliuchnikova et al. 201636, raw data associated with this study (labeled quantification) was retrieved from Proteomexchange database with accession number: PXD002623 and Taube et al. 201937, Hardenborg et al. 200938. The last two studies had their quantitative data (label free quantitative data) deposited as supplementary files rather than into a repository or database. Raw data files were obtained from each study report and used for comparative analysis to identify significantly altered proteins via identified peptides. Statistical Analysis Lipid analysis was completed using LipidSearchTM 4.2 software (Thermo) and MetaboAnalyst 5.0. Lipid species were identified and normalized against the standard internal lipids using LipidSearch. Data were subsequently log2 transformed and mean centered across samples. Then, the exported data was uploaded to MetaboAnalyst. Machine Learning and predictionsThe mass spectrometric lipid identification data were log2 transformed and mean centered across samples. Then, the exported data was uploaded to MetaboAnalyst 4.0. The results were organized and presented via a principal component analysis (PCA) plot, hierarchical clustering heatmap, partial least squares-discriminant analysis (PLS-DA), and variable in projection score plots. We then transformed the lipid expression values into a series of feature vectors for each of the patients. We also stratified the patients based on disease to generate three different supervised binary classification tasks: predicting 1) POAG vs control, 2) PEXG vs control, and 3) PEX vs control.The lipid quantification values log2 transformed using MetaboAnalyst 4.0 stratified by patients in three supervised binary classification tasks: predicting 1) POAG vs control, 2) PEXG vs control, and 3) PEX vs control vectors were used for further analyses. We fed the feature vectors into a logistical regression statistical classifier. The classifier had a sigmoid activation function and cross-entropy loss. We also set it to 1000 max-iterations for the solver to converge on a solution. We later performed 10-fold cross-validation using the Sklearn Python library to calculate our results. Figure 6 shows our results. The results point to an accuracy of about 86% for the POAG vs control group, 71% for the PEXG vs control group, and 86% for the PEX vs control group. After training the models, we then investigated the feature importance of the lipids in determining for each classification task. We rely on the coefficients associated with each feature in the logistic regression’s parameters to signify importance. We take the absolute value of the coefficients when calculating the feature’s contributions. All these calculations were performed using the Sklearn Python library. Figure 7 presents the five most-important features for each of the three task stratifications. For our predictive analytics, we primarily worked with Sklearn version 0.22.239 and Python 3.7.11 (https://www.python.org/downloads/release/python-3711/).Data availability and AccessionsAll lipidomics data will be available at the Metabolomics Workbench, http://www.metabolomicsworkbench.org, (Project IDs: ST001936). We have used published proteomics data in PRIDE proteomeXchange (http://www.proteomexchange.org/), (accession IDs PXD002623). Any additional data may be made available upon reasonable request to the corresponding author.Page 6 of 10Molecular OmicsMolecular Omics Accepted ManuscriptPublished on 11 April 2022. Downloaded on 4/12/2022 1:20:24 AM. View Article OnlineDOI: 10.1039/D1MO00495F

Page 8

Mol. Omics ARTICLEThis journal is © The Royal Society of Chemistry 20xx Mol. Omics., 2021, 00, 1-3 | 7Please do not adjust marginsPlease do not adjust marginsFig 8 Diseased states demonstrate distinct differences in lipid composition, with varying overlap. Variation inlipids between diseased states of normal control, POAG, PEX, and PEXG. Control group demonstrate enrichedlevels of CerG2GNAc1(d34:1) though depleted in both in PEX and PEXG samples. Phosphatidylcholines (PC),triglycerides (TG), and ceramides (Cer) are present in higher concentrations in the PEXG AH than all othergroups. The PC (18:0/18:2), PC (36:2), and PC (34:1e) are in low concentrations for PEX but highlyconcentrated in PEXG.ConclusionsWe identified 489 lipid species encompassing 26 lipid classes across PEX, PEXG, POAG, and control AH. Control AH showed greater total lipid content than PEX, PEXG, and POAG samples. Several cholesterol esters (ChE), phosphatidylcholines (PC), triglycerides (TG), and ceramides (Cer) species were present in higher concentrations for the PEXG AH samples. Conversely, CerG2GNAc1(d34:1) was enriched in control samples and depleted both in PEX and PEXG samples. Interaction component analyses between PEX and PEXG divided between dyslipidemia and non- and groups showed the smallest difference for PEXG with respect to lipids. Several PC species are in low concentrations for PEX but highly concentrated in PEXG, despite both having similar material deposits, suggesting lipid composition may help distinguish them (Fig. 8). Increased abundance of PC lipid species in the AH of patients with PEXG correlates with elevations in Apolipoprotein A-I (APOA1) demonstrating convergence of lipids and proteins in this pathway. Machine learning prediction is able to distinguish all three groups and control mostly with 86% accuracy. Author Contributions S. K. B. conceived the study. J. M. carried out the lipidomics analysis. V.C. contributed to initial analyses of proteomic datasets. B.V.V. and B. B. B., J.M.C. (J. M. de-la Casa) collected samples and clinical data. M. Z. C. J.M.C. and S. K. B. co-wrote the manuscript. J. M., M. Z. C., J.M.C., S. K.B. wrote the Transparent Methods (see Supplemental Information). Conflicts of interest There are no conflicts to declare.AcknowledgementsThis work was supported by an unrestricted grant from Research to Prevent Blindness, grants from The Glaucoma Foundation, New York, NIH Grants EY14801, EY031292. Metabolomics workbench is an effort of NIH Common Fund's Metabolomics Data Repository and Coordinating Center supported by U2C DK119886. Notes and references‡All lipidomics data will be available at the Metabolomics Workbench, http://www.metabolomicsworkbench.org, (Project IDs: ST001936).1 Eman Elhawy, G. K., Cecilia Q Dong, and John Danias. Pseudoexfoliation syndrome, a systemic disorder with ocular manifestations. Human Genomics, 2012, 6, doi:10.1186/1479-7364-6-22.2 Zheng, X. et al. Anterior segment optical coherence tomography analysis of clinically unilateral pseudoexfoliation syndrome: evidence of bilateral involvement and morphologic factors related to asymmetry. Invest Ophthalmol Vis Sci, 2011, 52, 5679-5684, doi:10.1167/iovs.11-7274.3 Lee, R. K. The molecular pathophysiology of pseudoexfoliation glaucoma. Curr Opin Ophthalmol, 2008, 19, 95-101, doi:10.1097/ICU.0b013e3282f49cda.4 Carreon, T., van der Merwe, E., Fellman, R. L., Johnstone, M. & Bhattacharya, S. K. Aqueous outflow - A continuum from trabecular meshwork to episcleral veins. Prog Retin Eye Res, 2017, 57, 108-133, doi:10.1016/j.preteyeres.2016.12.004.5 Mirza, E., Oltulu, R., Katipoglu, Z., Mirza, G. D. & Ozkagnici, A. Monocyte/HDL Ratio and Lymphocyte/Monocyte Ratio in Patients with Pseudoexfoliation Syndrome. Ocul Immunol Inflamm, 2020, 28, 142-146, doi:10.1080/09273948.2018.1545913.6 Mirza, E. Atherogenic indices in pseudoexfoliation syndrome. Eye (Lond), 2019, 33, 1911-1915, doi:10.1038/s41433-019-0506-z.7 Lesiewska, H. et al. Chosen Vascular Risk Markers in Pseudoexfoliation Syndrome: An Age-Related Disorder. J Ophthalmol, 2017, 2017, 5231095, doi:10.1155/2017/5231095.8 Janicijevic, K. et al. The importance of developing atherosclerosis in pseudoexfoliation glaucoma. Vojnosanit Pregl, 2017, 74, 8-12, doi:10.2298/VSP150320058J.9 Cabrerizo, J., Urcola, J. A. & Vecino, E. Changes in the Lipidomic Profile of Aqueous Humor in Open-Angle Glaucoma. J Glaucoma, 2017, 26, 349-355, doi:10.1097/IJG.0000000000000603.10 Edwards, G. et al. Endogenous ocular lipids as potential modulators of intraocular pressure. J Cell Mol Med, 2020, 24, 3856-3900, doi:10.1111/jcmm.14975.11 Yang, K. & Han, X. Lipidomics: Techniques, Applications, and Outcomes Related to Biomedical Sciences. Trends Biochem Sci, 2016, 41, 954-969, doi:10.1016/j.tibs.2016.08.010.12 Gartaganis, S. P. et al. Glutathione and lipid peroxide changes in pseudoexfoliation syndrome. Curr Eye Res, 2005, 30, 647-651, doi:10.1080/02713680590968367.13 Krumbiegel, M. et al. Genome-wide association study with DNA pooling identifies variants at CNTNAP2 associated with pseudoexfoliation syndrome. Eur J Hum Genet, 2011, 19, 186-193, doi:10.1038/ejhg.2010.144.Page 7 of 10Molecular OmicsMolecular Omics Accepted ManuscriptPublished on 11 April 2022. Downloaded on 4/12/2022 1:20:24 AM. View Article OnlineDOI: 10.1039/D1MO00495F

Page 9

ARTICLE Mol. Omics8 | Mol. Omics., 2021, 00, 1-3 This journal is © The Royal Society of Chemistry 20xxPlease do not adjust marginsPlease do not adjust margins14 Zagajewska, K. et al. GWAS links variants in neuronal development and actin remodeling related loci with pseudoexfoliation syndrome without glaucoma. Exp Eye Res, 2018, 168, 138-148, doi:10.1016/j.exer.2017.12.006.15 Anastasopoulos, E. et al. Association of LOXL1 polymorphisms with pseudoexfoliation, glaucoma, intraocular pressure, and systemic diseases in a Greek population. The Thessaloniki eye study. Invest Ophthalmol Vis Sci, 2014, 55, 4238-4243, doi:10.1167/iovs.14-13991.16 Gonzalez-Iglesias, H. et al. Comparative proteomic study in serum of patients with primary open-angle glaucoma and pseudoexfoliation glaucoma. J Proteomics, 2014, 98, 65-78, doi:10.1016/j.jprot.2013.12.006.17 Myer, C. et al. Aqueous humor metabolite profile of pseudoexfoliation glaucoma is distinctive. Mol Omics, 2020, 16, 425-435, doi:10.1039/c9mo00192a.18 Myer, C. et al. Differentiation of soluble aqueous humor metabolites in primary open angle glaucoma and controls. Exp Eye Res, 2020, 194, 108024, doi:10.1016/j.exer.2020.108024.19 Lin, W. J. et al. LipidSig: a web-based tool for lipidomic data analysis. Nucleic Acids Res, 2021, 49, W336-W345, doi:10.1093/nar/gkab419.20 Morris, J. et al. Proteomics of pseudoexfoliation materials in the anterior eye segment. Adv Protein Chem Struct Biol, 2021, 127, 271-290, doi:10.1016/bs.apcsb.2021.03.004.21 Aribindi, K. et al. Cholesterol and glycosphingolipids of human trabecular meshwork and aqueous humor: comparative profiles from control and glaucomatous donors. Curr Eye Res, 2013, 38, 1017-1026, doi:10.3109/02713683.2013.803123.22 Edwards, G., Aribindi, K., Guerra, Y., Lee, R. K. & Bhattacharya, S. K. Phospholipid profiles of control and glaucomatous human aqueous humor. Biochimie, 2014, 101, 232-247.23 Trzeciecka, A. et al. Shotgun Sphingolipid Analysis of Human Aqueous Humor. Methods Mol Biol, 2018, 1695, 97-107, doi:10.1007/978-1-4939-7407-8_10.24 Azbukina, N. V. et al. Targeted Lipidomic Analysis of Aqueous Humor Reveals Signaling Lipid-Mediated Pathways in Primary Open-Angle Glaucoma. Biology (Basel), 2021, 10, doi:10.3390/biology10070658.25 Aribindi, K., Guerra, Y., Lee, R. K. & Bhattacharya, S. K. Comparative phospholipid profiles of control and glaucomatous human trabecular meshwork. Invest Ophthalmol Vis Sci, 2013, 54, 3037-3044, doi:10.1167/iovs.12-10517.26 Margolis, M. J., Martinez, M., Valencia, J., Lee, R. K. & Bhattacharya, S. K. Phospholipid secretions of organ cultured ciliary body. J Cell Biochem, 2018, 119, 2556-2566, doi:10.1002/jcb.26419.27 Liu, H. H. & Li, J. J. Aging and dyslipidemia: a review of potential mechanisms. Ageing Res Rev, 2015, 19, 43-52, doi:10.1016/j.arr.2014.12.001.28 Weber, M. et al. Ocular phenotypes in a mouse model of impaired glucocerebrosidase activity. Sci Rep, 2021, 11, 6079, doi:10.1038/s41598-021-85528-4.29 Genetics of Exfoliation Syndrome, P. et al. Association of Rare CYP39A1 Variants With Exfoliation Syndrome Involving the Anterior Chamber of the Eye. JAMA, 2021, 325, 753-764, doi:10.1001/jama.2021.0507.30 Doucette, L. P. & Walter, M. A. Prostaglandins in the eye: Function, expression, and roles in glaucoma. Ophthalmic Genet, 2017, 38, 108-116, doi:10.3109/13816810.2016.1164193.31 Acar, N. et al. Red blood cell plasmalogens and docosahexaenoic acid are independently reduced in primary open-angle glaucoma. Exp Eye Res, 2009, 89, 840-853, doi:10.1016/j.exer.2009.07.008.32 Burgess, L. G. et al. Metabolome-Wide Association Study of Primary Open Angle Glaucoma. Invest Ophthalmol Vis Sci, 2015, 56, 5020-5028, doi:10.1167/iovs.15-16702.33 Anderson, M. G., Meyer, K. J., Hedberg-Buenz, A. & Fingert, J. H. Update on Animal Models of Exfoliation Syndrome. J Glaucoma, 2018, 27 Suppl 1, S78-S82, doi:10.1097/IJG.0000000000000911.34 Lyu, L., Sonik, N. & Bhattacharya, S. An overview of lipidomics utilizing cadaver derived biological samples. Expert Rev Proteomics, 2021, 18, 453-461, doi:10.1080/14789450.2021.1941894.35 Bligh, E. & Dyer, W. A Rapid Method of Total Lipid Extraction and Purification. Canadian Journal of Biochemistry and Physiology, 1959, 37, 911-917, doi:10.1139/o59-099.36 Kliuchnikova, A. A. et al. Human aqueous humor proteome in cataract, glaucoma, and pseudoexfoliation syndrome. Proteomics, 2016, 16, 1938-1946, doi:10.1002/pmic.201500423.37 Botling Taube, A., Konzer, A., Alm, A. & Bergquist, J. Proteomic analysis of the aqueous humour in eyes with pseudoexfoliation syndrome. Br J Ophthalmol, 2019, 103, 1190-1194, doi:10.1136/bjophthalmol-2017-310416.38 Hardenborg, E. et al. Protein content in aqueous humor from patients with pseudoexfoliation (PEX) investigated by capillary LC MALDI-TOF/TOF MS. Proteomics Clin Appl, 2009, 3, 299-306, doi:10.1002/prca.200780077.39 Pedregosa, F. et al. Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research, 2011, 12, 2825-2830.Page 8 of 10Molecular OmicsMolecular Omics Accepted ManuscriptPublished on 11 April 2022. Downloaded on 4/12/2022 1:20:24 AM. View Article OnlineDOI: 10.1039/D1MO00495F

RESEARCH GRANTS 2019-21-all cycl

Page 1

Page 2