Lack of Involvement of CEP Adducts in TLR Activation and in Angiogenesis

Proteins that are post-translationally adducted with 2-(ω-carboxyethyl)pyrrole (CEP) have been proposed to play a pathogenic role in age-related macular degeneration, by inducing angiogenesis in a Toll Like Receptor 2 (TLR2)-dependent manner. We have investigated the involvement of CEP adducts in angiogenesis and TLR activation, to assess the therapeutic potential of inhibiting CEP adducts and TLR2 for ocular angiogenesis. As tool reagents, several CEP-adducted proteins and peptides were synthetically generated by published methodology and adduction was confirmed by NMR and LC-MS/MS analyses. Structural studies showed significant changes in secondary structure in CEP-adducted proteins but not the untreated proteins. Similar structural changes were also observed in the treated unadducted proteins, which were treated by the same adduction method except for one critical step required to form the CEP group. Thus some structural changes were unrelated to CEP groups and were artificially induced by the synthesis method. In biological studies, the CEP-adducted proteins and peptides failed to activate TLR2 in cell-based assays and in an in vivo TLR2-mediated retinal leukocyte infiltration model. Neither CEP adducts nor TLR agonists were able to induce angiogenesis in a tube formation assay. In vivo, treatment of animals with CEP-adducted protein had no effect on laser-induced choroidal neovascularization. Furthermore, in vivo inactivation of TLR2 by deficiency in Myeloid Differentiation factor 88 (Myd88) had no effect on abrasion-induced corneal neovascularization. Thus the CEP-TLR2 axis, which is implicated in other wound angiogenesis models, does not appear to play a pathological role in a corneal wound angiogenesis model. Collectively, our data do not support the mechanism of action of CEP adducts in TLR2-mediated angiogenesis proposed by others.


Introduction
Age-related macular degeneration (AMD) is a major cause of legal blindness in the elderly. The macula is a specialized area of the central retina that is enriched in photoreceptor cells and is responsible for high acuity vision. In AMD, progressive macular degeneration can impair critical daily functions such as reading, driving, and face recognition. Thus AMD can have a profound impact on quality of life. There are two forms of advanced AMD: dry and wet (neovascular) AMD [1]. AMD is thought to be a disease of the retinal pigment epithelium (RPE) cells, which provide critical support functions to adjacent photoreceptors [1]. In the early stage of disease, AMD retinas show progressive accumulation of extracellular deposits, drusen, as well as intracellular deposits, lipofuscin, at the level of the RPE. These deposits initially tend to accumulate in the macular area. Over time, RPE cells show pigmentary changes and begin to degenerate. In advanced stages, dry AMD patients exhibit substantial delineated areas of RPE atrophy, or geographic atrophy. Advanced wet AMD patients exhibit leaky blood vessels in the macula, in many cases emanating from the choriocapillaris [1].
Currently there are no treatments for dry AMD. In the Age-Related Eye Disease Study 1 (AREDS 1), dietary supplements comprised of anti-oxidants and select minerals reduced the risk of progression to advanced AMD by 25% [1]. Several therapeutic approaches are being tested in clinical trials [2] but there are no FDA-approved treatments in practice at this point. For wet AMD, anti-angiogenic treatments have been clinically proven to be efficacious [3]. However, not all patients respond to treatment and the burden of treatment is still relatively high. Thus there is a great medical need for novel treatments for AMD. The molecular details of pathogenesis in AMD are not fully established but several pathogenic mechanisms have been implicated [1]. For example, human molecular genetic data indicate the involvement of the alternative complement pathway. Another potential cause is proposed to be cumulative oxidative stress, based on preclinical studies and on the AREDS1 trial. One manifestation of oxidative stress is proposed to be the formation of CEP adducts, which are a type of advanced glycation end products [4].
Photoreceptor cells are highly enriched in docosahexaenoic acid (DHA), a labile fatty acid that is susceptible to breakdown by photo-oxidation and other forms of oxidative stress. The breakdown products include a reactive aldehyde, 4-hydroxy-7oxohept-5-enoic acid, which can condense with primary amines to form a Schiff base. In the case of proteins, 4-hydroxy-7-oxohept-5enoic acid condenses with lysinee-amines. Subsequent reactions result in a covalently attached CEP moity, yielding a stable CEP adduct [4]. In previous reports, antibodies raised against synthetic CEP reagents were used to identify, localize, and quantify CEP adducts by various immunological assays [5]. Elevated levels of CEP adducts were initially reported in proteomic studies of AMD donor eyes [5] and subsequently in AMD plasma [6,7]. Thus CEP adducts were implicated in AMD [5][6][7]. In later studies CEP adducts were reported to be pro-angiogenic, both in vitro and in vivo. These in vivo studies utilized the micropocket corneal neovascularization (CoNV) and the laser-induced choroidal neovascularization (CNV) models [8]. More recently, Toll-like receptor 2 (TLR2) was reported to mediate the CEP adductinduced angiogenesis [9]. The angiogenic activity was reported to be independent of the vascular endothelial growth factor (VEGF) pathway [8,9].
There is a medical need for novel treatments for both wet and dry AMD. CEP adducts are implicated in both forms of AMD and represent an attractive potential target for drug discovery. Thus we initiated validation studies to assess the therapeutic potential of inhibiting CEP adducts.

Synthesis of Tool Reagents
Several synthetic CEP adducts were generated according to reported procedures [10]. These tool reagents included protein (e.g. human serum albumin-CEP, or HSA-CEP), peptide (e.g. Ac-Gly-Lys-OMe-CEP, or dipeptide-CEP), and phospholipid (e.g. phosphatidyl ethanolamine-CEP, or PE-CEP) adducts, as listed in Table S1. The presence of CEP adducts and the stoichiometry of adduction was confirmed by 1 H-NMR and LC-MS/MS ( Figure  S1 and Figure S2). Invariably the presence of CEP moiety was established in the adducted samples and was never detected in the controls. CEP adduction was deemed successful by several measures. For example, 1 H-NMR analysis indicated the expected molecular signature of the CEP group in dipeptide-CEP but not the untreated dipeptide ( Figure S1B). Likewise, LC-MS/MS analysis of enzymatic hydrolyzed adducted proteins detected lysine-CEP in HSA-CEP and MSA-CEP (mouse serum albumin-CEP) but not the respective controls ( Figure S1C). For protein adducts, two controls were used: CTL1, which represents untreated protein; and CTL2, which represents treated unadducted protein. The latter control was treated exactly the same way as the corresponding CEP adduct except for one step, to prevent adduction (see Materials and Methods). No CEP moiety was detected in HSA-CTL2 or MSA-CTL2. There are a total of 59 lysines in HSA (UnitPro P02768) and 50 lysines in MSA (UnitPro P07724). We identified 14 lysine-CEP sites in HSA-CEP and 40 lysine-CEP sites in MSA-CEP ( Figure S2). These values are in fair agreement with the reported stoichiometries: 6 or 17 CEP-modified lysines for HSA-CEP [10,11], and 15 for MSA-CEP [10].

Does CEP-Adduction Affect Protein Structure?
When we analyzed HSA-CEP, HSA-CTL2, and HSA-CTL1 in structural studies we observed significant structural alterations in HSA-CEP in comparison with HSA-CTL1. However, HSA-CTL2 also incurred significant alterations, similar to but less extensive than HSA-CEP. On SDS-PAGE gels HSA-CEP and HSA-CTL2 but not HSA-CTL1 appeared to form oligomers in a ladder-like fashion ( Figure 1A). Comparison of size exclusion chromatography (SEC) profiles indicated progressive loss of the HSA monomer in the order of HSA-CTL1. HSA-CTL2. HSA-CEP ( Figure 1B). Circular dichroism (CD) also indicated a significant loss in secondary structure in HSA-CTL2 and HSA-CEP compared to HSA-CTL1 ( Figure 1C).
Collectively these data show that our synthetic CEP adducts can incur two kinds of structural changes: a) CEP-independent changes and b) CEP-dependent changes. Comparison of our HSA-CTL1 and HSA-CTL2 illustrates the CEP-independent changes in protein structure, which indicate artificial changes introduced by the procedure for generating synthetic adducts. Comparison of our HSA-CTL2 and HSA-CEP illustrates the CEP-dependent changes that occur as a result covalent CEP groups in HSA-CEP, beyond the CEP-independent changes in HSA-CTL2.

Is TLR2 Activated By CEP Adducts?
We tested several CEP adducts in a cell-based TLR2 assay, essentially as described [9]. Pam3CSK4, a known TLR2 agonist, showed dose-dependent activation of TLR2 as monitored by production of NF-kB or IL-8 in cellular assays ( Figure 2, top and middle panels). However, no TLR2 activation was detected by HSA-CEP (Figure 2), or several other CEP adducts tested: MSA-CEP, dipeptide-CEP, or PE-CEP (not shown). None of these adducts showed any cytotoxicity, as measured with CellTiter-Glo (CTG) kit ( Figure 2). Next we used THP-1 cells, which naturally express several TLRs, including TLR2. While positive controls showed specific activation of the corresponding TLR, CEP adducts failed to activate TLR2 or any other TLR that was monitored ( Figure 2, bottom panel). In vivo, CEP adducts did not induce biological effects that are mediated by TLR2. As Figure 3A shows, treatment of mice with Pam3CSK4 induced infiltration of neutrophils and macrophages in the retina. However, neither dipeptide-CEP nor MSA-CEP (not shown) induced retinal infiltration in the same assay. Representative images of this experiment are shown in Figure 3B. These results suggest that CEP adducts do not activate TLRs, including TLR2.

Are CEP Adducts or TLR2 Involved in Angiogenesis?
We used the tube formation in vitro assay to determine if CEP adducts are angiogenic, similar to a previously reported assay [9]. VEGF induced significant tube formation; however, neither CEP adducts nor Pam3CSK4 affected tube formation ( Table 1). In addition, poly (I:C) (a TLR3 agonist) and LPS (a TLR4 agonist) failed to show any effect. Representative images of the tube formation assay are shown in Figure 4.
The CEP adducts were further evaluated in the laser-induced choroidal neovascularization (CNV) model, as was reported earlier [8]. Initial laser CNV studies were performed with C57BL/6N mice, which showed no effect of MSA-CEP (File S1). In light of the rd8 mutation in the Crb1 gene reported in this strain [12], we repeated the study with C57BL/6J mice which are wildtype for Crb1 [12] and were used in the previously reported study [8]. We observed the same results with 2 experiments in each strain. Subretinal injection of VEGF significantly exacerbated CNV, while a VEGF-neutralizing antibody inhibited CNV ( Figure 5A and File S1). This is consistent with VEGF being a major proangiogenic factor in CNV. However, subretinally administered MSA-CEP, at a dose nearly identical to that used in the previous report [8], had no effect in this model ( Figure 5A and File S1). Representative images of the experiments in Figure 5A are shown in Figure 5B.
In studies with a corneal neovascularization (CoNV) mouse model, we observed that Myd88-deficiency had no significant effect on CoNV ( Figure 6) when compared with similarly treated wild-type mice. Since Myd88 deficiency abolishes TLR2 activity, this indicates that TLR2 is not required for angiogenesis in the abrasion-induced CoNV model. In contrast, CoNV is greatly dependent on VEGF-A: a) qPCR analysis showed a 30-fold increase in VEGF-A mRNA expression and b) treatment of abraded mice with VEGF-A neutralizing antibody showed significant reduction in CoNV area ( Figure S3).

Summary
Here we performed validation studies for the proposed CEP-TLR2 axis to assess the therapeutic potential for wet AMD treatment. Following a published procedure we generated synthetic CEP adducts and confirmed the presence of covalent CEP groups. Structural analyses of a CEP adduct indicated changes in tertiary structure that were not observed in the naïve protein; however, similar structural changes were observed in the treated, unadducted control. Thus the physiological relevance of the observed structural changes is uncertain. Next we attempted to reproduce some of the reported biological effects of synthetic CEP adducts. When we tested our synthetic CEP adducts in in vitro and in vivo assays, we observed neither TLR2 activation nor proangiogenic activity. We conclude that our data do not support the CEP-TLR2 hypothesis.

Structural Changes in CEP Adducts
In our hands the published protocol for generating CEP adducts worked successfully, by the criterion of the presence of covalentlylinked CEP groups in the adduct. This protocol worked robustly with all classes of reagents tested, including proteins, dipeptide, and lipid. Furthermore, the stoichiometry of adduction of each reagent was in fair agreement with the corresponding published stoichiometry.
In our attempt to understand the biological consequences of CEP adduction we initiated protein structural studies with HSA-CEP. Since the conversion of a significant number of positivelycharged lysine side chains to negatively charged CEP groups (from the carboxylate group) would greatly alter its surface electrostatic potential, we anticipated and indeed observed structural changes in HSA-CEP in comparison to untreated HSA control (HSA-CTL1). We were surprised, however, to detect similar changes in HSA-CTL2, the treated unadducted control ( Figure 1). Our interpretation is that two kinds of structural alterations can occur in synthetic CEP adducts: a) alterations that do not involve the CEP group and were observed when comparing HSA-CTL2 to HSA-CTL1; and b) alterations that occur as a result of covalentlylinked CEP groups and were observed when comparing HSA-CEP to HSA-CTL2. It is not clear which step(s) or reagents in the published adduction procedure led to the CEP-independent changes in HSA-CTL2. A candidate culprit is the organic solvent, dimethylformamide; organic solvents are known to affect the structure of some proteins. The adduction procedure entails exposure of protein to 30% dimethylformamide/PBS solution for 4 days at 37uC [10]. In searching the literature we found similar significant structural alteration in a CEP adduct published by another laboratory. A synthetic MSA-CEP adduct appeared to migrate as a continuous smear on denaturing SDS-PAGE and immunoblot, whereas the untreated MSA (the equivalent of MSA-CTL1) migrated as one predominant electrophoretic band ( Figure 1 in [13]). A treated unadducted control, the equivalent of our MSA-CTL2, was not included in the report [13]. However since we used similar procedures for generating CEP adducts, in all likelihood the published MSA-CEP incorporated both CEPindependent and -dependent changes.
Thus far, no endogenous CEP adducts have been isolated directly from any biological sources and none have been characterized in the literature. For example, the stoichiometry of CEP adduction (moles CEP per mole protein) and structural properties have not been reported for any endogenous CEP adducts. Hence at this point it would not be possible to verify that any synthetic CEP adduct is representative of endogenous ones with respect to protein structure. This caveat notwithstanding, we proceeded with biological studies to see if we could reproduce the biological effects of CEP adducts with regards to TLR2 activation and angiogenesis. In our approach we included HSA-CTL2 and MSA-CTL2 in the biological assays of the corresponding CEP adducts, so we might discern biological effects that are specific to the CEP group.

TLR2 Activation by CEP Adducts
We tested CEP-adducted protein or dipeptide in two cell-based assays: a) HEK293-TLR2 cells that specifically expressed TLR2 and, b) THP-1 cells that express multiple TLRs, including TLR2. In both assays, TLR2 activation was observed with a synthetic TLR2 agonist, Pam3CSK4, but not with synthetic CEP adducts. Specifically, we did not detect any effect of HSA-CEP in TLR2expressing HEK293 cells, as was reported (Figure S14 in [9]). Not surprisingly the controls for our CEP adducts did not have any effects, either. Our cellular assays also did not register any effect of the dipeptide-CEP, which was reported to be pro-angiogenic in several cellular and in vivo assays in a TLR2-dependent manner [9]. However, this dipeptide-CEP was not tested in the same cellbased TLR assay used for HSA-CEP [9], so a direct comparison with our cell-based data is not possible.
We also evaluated CEP adducts in an in vivo model for TLR2 activation. In this model, treatment with Pam3CSK4 induced retinal leukocyte infiltration in wild-type mice, but not in Myd882/2 nor TLR22/2 mice (not shown). However, treatment of wild-type mice with CEP adducts did not result in measurable retinal leukocyte infiltration, indicating that TLR2 was not activated by CEP adducts in vivo.

CEP-TLR2 in Angiogenesis Assays
In vitro, neither HSA-CEP nor Pam3CSK4 (TLR2 agonist) showed any pro-angiogenic effect in the tube formation assay with human umbilical vein endothelial cells (HUVECs). Likewise, agonists to other TLRs (LPS, poly (I:C)) were not pro-angiogenic, whereas VEGF was.
In vivo, synthetic MSA-CEP did not exacerbate laser-induced CNV in a mouse model as reported [8]. This was the case with two substrains of C57BL/6 mice. In the initial two laser CNV studies we used C57BL/6N mice. Subsequently, it was reported that this substrain carries the rd8 mutation in the Crb1 gene [12]. We then performed two additional laser CNV studies with the C57BL/6J substrain, which has the wild-type Crb1 gene [12]. The C57BL/6J substrain is the same one used in the previously published laser CNV study [8]. We observed similar results in all four laser CNV studies: CNV was exacerbated with exogenous VEGF, ameliorated with a VEGF-neutralizing antibody, and unaffected with MSA-CEP. Collectively, these data are consistent with a major angiogenic role for the VEGF pathway, but not for CEP adducts, in the laser-induced CNV model.
As an alternative in vivo model of ocular angiogenesis for exploring the CEP-TLR2 hypothesis, we used an abrasioninduced corneal neovascularization model (CoNV). This CoNV model is different from that used earlier with CEP adducts: in the corneal pocket CoNV model, a pellet containing synthetic HSA- CEP was implanted in the cornea [8]. Since our synthetic CEP adducts were biologically inactive in all assays so far, we aimed to use a model where endogenous -not synthetic-CEP adducts might play a role. In the abrasion-induced CoNV model, angiogenesis occurs as part of the wound healing process, induced by mechanical abrasion of the cornea. Furthermore, it has been shown that macrophages are recruited to the cornea during early stages of neovascularization [14]. This resembles the back punch model, in which wound angiogenesis entails recruitment of macrophages [9]. CEP adducts were reported to be transiently present during this time, detected by immunocytochemistry with an antibody against synthetic CEP adducts. By immunolabeling, a substantial portion of CEP adducts was present in the recruited F4/80+ macrophages [9]. Treatment with dipeptide-CEP in this model accelerated wound closure and vascularization in a TLR2-dependent manner, as shown by the comparison of TLR22/2 and TLR2+/+ mice. In the same vein, we used the abrasioninduced CoNV model and compared littermate Myd882/2 and Myd88+/+ mice. (Myd88 is required for TLR2 function.) We found no difference in CoNV area between the two groups. This indicates that, in the abrasion-induced CoNV model, TLR2 and other Myd88-dependent TLRs are not involved in angiogenesis. It is not known whether endogenous CEP adducts are present in the abrasion-induced CoNV model. We also tested topical treatment with synthetic HSA-CEP, but observed no effect on CoNV (not shown). It seems therefore the CEP-TLR2 axis proposed for other wound angiogenesis models [9] does not apply to corneal abrasion-induced wound angiogenesis model tested here.

Synthetic vs. Endogenous CEP Adducts
It has been proposed that the biological effects of CEP adducts depend solely on the presence of CEP groups and not the host carrier [9]. This was not the case in our study: neither protein CEP adducts nor dipeptide CEP adducts produced any of the published biological effects [8,9] that were tested here. Our synthetic reagents were verified for the presence of covalently-attached CEP groups. There is no obvious explanation for these discrepancies. As innate immune receptors, TLRs recognize structures and patterns so it is conceivable that changes in structure, even if artificial, could elicit TLR activation. Thus one possibility is that there are structural differences between our synthetic CEP adducts and those used in previous studies, as the reagents were prepared at different laboratories.
So far endogenous CEP adducts from any biological systems have not been isolated and therefore none have been characterized in structural studies. Furthermore, the abnormal electrophoretic patterns seen with synthetic CEP adducts reported here or by others [13] do not seem to resemble those of in vivo CEP adducts detected on immunoblots, reported either in human donor material (e.g. Figure 3 in [5]) or in the light-induced rat retinal degeneration model (e.g. Figure 1 in [15]). It is therefore uncertain which synthetic CEP adducts are representative of endogenous ones; perhaps none. The final answer awaits the isolation of endogenous CEP adducts and their characterization.
By extension, the biological effects reported with synthetic CEP adducts also need to be confirmed with endogenous CEP adducts. Our HSA-CTL2 did not show any biological activity, but neither did our HSA-CEP; hence in our study there was no concern about non-physiological biological effects, which HSA-CTL2 was intended for as a control. However, a treated unadducted control would be critically important when biological activity with a synthetic CEP adduct is observed. This is exemplified in a recent  publication where the unadducted control, ''sham-MSA'', showed biological activity in some assays. In BALB/c macrophages, sham-MSA induced the upregulation of M1 markers and of an inflammatory gene, KC, 2 to 5 fold above that of control levels ( Figures 1A and S2, respectively, in [16]). In some cases the sham-MSA effect represented 30%-40% in magnitude of the MSA-CEP effect: (IL-1b, TNFa, and KC), despite the absence of CEP groups in sham-MSA [16].
Treated unadducted controls were not used in earlier studies that reported a role for synthetic CEP adducts in TLR2 activation and in angiogenesis [8,9]. If it is possible that our CEP adducts are different from those used by others, it is also possible that our treated unadducted controls are different. The fact that our treated, unadducted controls (HSA-CTL2, MSA-CTL2) were biologically inactive does not necessarily apply to other studies in literature, as the example above illustrates. Thus the physiological relevance of synthetic CEP adducts is unclear, especially when untreated proteins were used as the only controls in the biological assays.
Polyclonal and monoclonal antibodies raised to synthetic CEP adducts have been reported to immunolabel biological samples from AMD patients in Western blots, immunohistochemical sections, and ELISAs [5][6][7]11]. The same antibodies reportedly immunolabelled biological samples in animal studies, e.g. [15,17]. However, the immunolabelled proteins were not confirmed to have any CEP moieties, by other independent assays that do not use antibodies (e.g. LC-MS/MS); i.e. the immunolabelled proteins were not confirmed to be bona fide CEP adducts. For example, several candidate CEP adducts were immunolabelled on Western blots of patient donor material and subsequently identified by LC-MS/MS analysis, however the presence of covalently-linked CEP groups was not confirmed by LC-MS/MS or other assays [5]. The fact that a synthetic CEP adduct (used a control) was immunolabelled on the same western blot is not a surprise, as CEP antibodies were raised against a synthetic CEP adduct. At this point, the antibodies against synthetic CEP adducts [5] have not yet been validated for detection of endogenous CEP adducts. This is underscored by the electrophoretic changes in synthetic CEP adducts that do not seem to resemble those of endogenous CEP adducts, as explained above. Data generated by other (nonimmunological) assays is needed to validate CEP antibodies.
Arguably the in vivo existence of CEP adducts requires confirmation, as well, since all evidence so far has been generated with these antibodies. For example, a proteomics study of AMD patient samples identified and quantified hundreds of proteins by LC-MS/MS [18]. Yet the same study reported that CEP adducts were below detection limits and ''none were reliably identified'' (supplemental information in [18]). On a promising note, an improved LC-MS/MS assay has been reported, with a sensitivity of 1 pmol or less of CEP-lysine in enzymatically-digested patient plasma samples [19]. According to ELISA data [7] the average levels of CEP adducts in AMD plasma is 37 pmol/mL, thus hopefully the improved LC-MS/MS assay can confirm the presence of CEP adducts in vivo.
In conclusion, our studies of synthetic CEP adducts did not validate the CEP-TLR2 axis in angiogenesis as proposed [8,9]. While the cause of the discrepancies is not clear, it does seem clear that the mere presence of CEP groups is not sufficient to elicit the reported biological responses. More data, ideally with endogenous CEP adducts, is needed to understand what properties of CEP adducts, if any, can lead to TLR2 activation and to angiogenesis. In light of the prevalence of AMD and the unmet medical needs, more research into the pathophysiology of CEP adducts is warranted.
As a postscript, after submission of our manuscript an independent report [20] also showed that CEP adducts alone do not induce TLR2 signalling nor related biological effects (e.g. Figures 1A and 1B), as we have reported here. Rather, the report claims that CEP adducts potentiate the effect of a synthetic TLR2 agonist, Pam3CSK4, in cultured murine bone-marrow derived macrophages [20].

Synthesis and Verification of CEP Adducts
Synthesis. All CEP adducts were synthesized as described [10]. The dipeptide, Ac-Gly-Lys-OMe, was obtained from BACHEM; HSA from AlbuminBio; MSA from AlbuminBio or Sigma; phosphatidyl ethanolamine was from Sigma. Controls for protein CEP adducts included untreated protein (CTL1) and treated unadducted protein (CTL2). The latter control was processed in the same synthesis procedure as that for CEP adducts, except 4,7-dioxoheptanoic acid 9-fluorenylmethyl ester was left out to avoid the covalent addition of CEP moiety. Protein CEP adducts, after final dialysis in PBS, were quantified by the Bradford assay and tested for endotoxins with the Endosafe-PTS kit (Charles River). For storage, samples were filtered through 0.2 mm, divided aseptically in 1-mL aliquots, and stored at 280 C.
NMR of protein hydrolysate samples. Hydrolyzed protein samples were prepared for NMR analysis by the addition of 5 mL of D2O (CIL) to 15 mL of Hydrolysate solutions. Sodium 3trimethylsilyl [2,2,3,3-d4]propionate (TMSP) as added as an internal chemical shift and quantitation reference. High-resolution 1 H-NMR spectra were acquired at 30061 K, using a standard (D-90u-acquire) pulse sequences on a Bruker-600 Avance spectrometer ( 1 H frequency of 600.26 MHz). 1 H-NMR spectra were acquired with 256 free induction decays, 65,536 complex data points, a spectral width of 7.2 kHz, and a relaxation delay of 5 s. All spectra were processed by multiplying the FID by an exponential weighting function corresponding to a line broadening of 0.3 Hz. The CEP pyrrole resonances at 1 H d 6.8 ppm, 1 H d 6.1 ppm and 1 H d 5.9 ppm were integrated relative to the aromatic resonance of phenylalanine and tyrosine using the ACD 10.0 package (Advanced Chemistry Development, Toronto, Canada).

In Vitro Assays
Cell-based TLR assays. HEK293 cells expressing TLR2 and NFKB luciferase reporter (gift from Novartis Vaccine, Siena, Italy) were seeded at 30000 per well the night before. HSA-CEP was added and incubated for either 6 hr or 24 hr. Supernatant was collected for IL-8 ELISA (R&D, cat# DY208). NFkB luciferase activity was assayed on remaining cells, using Bright-Glo (Promega, cat# E2610). HEK293 cells in a separate plate with the same treatment were used for cell viability measurement using CellTiter-Glo (Promega, cat# G7570), according to manufacturer's instruction. Thp1 (ATCC, cat# TIB-202) was primed with 0.5% DMSO for overnight, at 100,000/well, then incubate with HSA-CEP for 24 hr, with TLR ligands (Pam3CSK4, FSL1, R837 and R848 were all from Invivogen; LPS was purchased from Sigma) as controls. Supernatant was collected for IL-8 ELISA (R&D), and the remaining cells were used for cell viability measurement using CellTiter-Glo.
In vitro tube formation assay. The CellPlayer GFP Angiokit-96 by Essen BioScience (Ann Arbor, MI) was used to measure tube formation in vitro. Briefly, GFP-transfected HU-VEC were co-cultured with human fibroblasts in a specially designed medium for 11 days in a 96-well format. Cells were treated on days 1, 2, 5, 7 and 9 with VEGF, TLR agonists, or CEP-adducted or control-treated proteins. Fluorescence measurements (IncuCyte, Essen BioScience) were taken kinetically every 12 hours for the duration of the experiment and average tube length (mm/mm 2 ) was quantified on the last day of the experiment according to the manufacturer's instructions. Control wells received media alone. Reagents were obtained from the following sources: Human VEGF 165-Peprotech; Pam3CSK4-InvivoGen; LPS -Sigma-Aldrich; Poly (I:C) -InVivoGen.

In Vivo Assays
Animals. All animal experiments were approved by the Animal Care and Use Committee at the Novartis Institutes for Biomedical Research. Upon arrival at the vivarium, mice were acclimated for at least 4 days before any studies were initiated. The animals were fed standard laboratory chow and sterile water ad libitum. Genotyping was performed on genomic DNA obtained from tail snips by standard procedures. All mouse strains were genotyped for the Crb1 gene, to determine if they carried the rd8 mutation.
C57BL/6N mice were obtained from Taconic; the rd8 mutation was present in these mice. C57BL/6J mice were obtained from Jackson; the rd8 mutation was absent in these mice. Myd88-deficient mice lacking exons 2-5 were generated at Novartis Institutes for Biomedical Research. Mice were backcrossed to C57BL/6J mice for at least 10 generations; the rd8 mutation was absent in these mice. Heterozygous breeding generated littermate pups of each genotype, identified by PCR genotyping. Myd88 deficiency was also functionally confirmed by the in vivo retinal infiltration assay below: mutant vs. littermate wt mice with treated with either TLR2 agonists and with TLR4 agonists and the retinal infiltration was measured as described (not shown).

Laser-induced choroidal neovascularization
(CNV). CNV was induced by laser injury in age and sex matched on a) C57BL/6N mice and b) C57BL/6J mice. Two in vivo experiments were performed with each mouse strain. After pupil dilation with 1% cylate and 10% phenylephrine, the mice were anesthetized and the retinas were visualized with a slit lamp microscope and a cover slip.  MAB006) and a proprietary anti-mouse VEGF antibody (4G3) were reconstituted in sterile saline (Hospira) to a concentration of 0.05, 0.5, 2.5 or 3.3 mg/ml respectively. 1.4 or 1.9 mg/ml of CEP-MSA and MSA-CTL2 (control 2, mouse serum albumin treated but not adducted) or the other reagents were injected in to the subretinal space on day 0 immediately after a laser as described. After the application of laser burns and subretinal injections of test reagents, antibiotic ointment (Tobramycin or Neomycin ophthalmic ointment depending on availability) was applied to both eyes. The anti-VEGF antibody, 4G3, is a mouse anti-VEGF IgG1 antibody. It binds to mouse VEGF with an EC50 of 0.047 nM in a sandwich ELISA and neutralizes mouse VEGF binding to human VEGFR-2 with an EC50 of 0.15 nM in a binding assay (ELISA MSD).
Corneal neovascularization (CoNV). Acute CoNV was induced in 7-to 9-week old anesthetized mice by complete removal of the corneal epithelium with mechanical abrasion, as detailed [14]. At the end of the studies, mice were humanely euthanized and the area of CoNV was quantitated as described [14]. Animals were randomized prior to treatment and analysis was performed in a masked fashion. N = 5-10 mice/group.
In studies using Myd882/2 mice (Figure 6), male knockout (KO) and male wild-type littermate controls (WT) were abraded on day 0 and euthanized at the end of the study on day 21 for analysis. Myd88-deficient and littermate wild-type mice are on the C57BL6/J background and are described above.
For other CoNV studies ( Figure S3), C57BL6/N mice were used. The Crb1 gene product is expressed in the retina, but not in the cornea [22]. In the study presented in Figure S3C, animals (N = 10-12 mice/group) were injected i.p. with PBS (200 ml), control antibody (IgG1, 0.5 mg/kg) or anti-VEGF antibody (4G3, 0.5 mg/kg) on days 0, 3 and 5 post-abrasion and eyes were collected on day 6 for analysis.  Adult C57BL/6N mice (N = 5 animals/ group) were subjected to corneal abrasion on Day 0 and dissected corneas were analyzed for neovascularization area at different timepoints after abrasion. Neovascularization area progressively increased and plateaued around 2 weeks after abrasion. Statistical analysis was performed using one-way ANOVA with Dunnet's post-test, comparing each time point to Naïve. Only the statistically significant differences between groups are indicated. B) Upregulation of VEGFA Transcript. Total RNA was prepared from dissected corneas from naïve mice and or cornea-abraded mice that were euthanized on Day 1 and Day 6 post-abrasion as indicated (N = 5 to 6 animals/group). First-strand cDNA was generated using the High Capacity RNA-to-cDNA Master Mix (Applied Biosystems). Pre-amplification products were generated using the Taqman PreAmp Master Mix Kit (Applied Biosystems) and a pool of FAM-labelled Taqman assays on demand (Applied Biosystems). qPCR was performed on diluted pre-amplification products using the same Taqman assays on demand in qPCR singleplex reactions. Relative quantification (RQ) performed using DDCt method and data presented as RQ median with error bars as RQ min and RQ max. VEGFA, PECAM-1(expressed by vascular endothelial cells), and b-actin mRNA expression was normalized by expression of b-actin gene and expressed relative to naive animals. Statistical analysis was performed using one-way ANOVA with Dunnett's post-test. C) VEGF Ab inhibits CoNV. Adult C57BL/6N mice (N = 10-12 animals/group) were subjected to corneal abrasion on Day 0 and injected intraperitoneally with the reagents as indicated on Days 0, 3, and 5 post-abrasion. Reagents included PBS (vehicle), control IgG1 Ab, and anti-VEGF antibody (4G3). The antibodies were dosed at 0.5 mg/kg. On Day 6 the animals were euthanized and CoNV area was measured by fluorescence microscopy as described in Materials and Methods. Statistical analysis was performed using one-way ANOVA with Dunnett's post-test.