Coagulation factor XIIIa cross-links amyloid β into dimers and oligomers and to blood proteins

In cerebral amyloid angiopathy (CAA) and Alzheimer's disease (AD), the amyloid β (Aβ) peptide deposits along the vascular lumen, leading to degeneration and dysfunction of surrounding tissues. Activated coagulation factor XIIIa (FXIIIa) covalently cross-links proteins in blood and vasculature, such as in blood clots and on the extracellular matrix. Although FXIIIa co-localizes with Aβ in CAA, the ability of FXIIIa to cross-link Aβ has not been demonstrated. Using Western blotting, kinetic assays, and microfluidic analyses, we show that FXIIIa covalently cross-links Aβ40 into dimers and oligomers (kcat/Km = 1.5 × 105 m−1s−1), as well as to fibrin, platelet proteins, and blood clots under flow in vitro. Aβ40 also increased the stiffness of platelet-rich plasma clots in the presence of FXIIIa. These results suggest that FXIIIa-mediated cross-linking may contribute to the formation of Aβ deposits in CAA and Alzheimer's disease.

Amyloid ␤ (A␤) 4 is a 4 kDa intrinsically disordered protein that accumulates along the cerebral vasculature during cerebral amyloid angiopathy (CAA). The accumulation of A␤ leads to the degeneration of surrounding cells and is associated with microhemorrhages (1). Although CAA is present in over 90% of patients with Alzheimer's disease (AD) (2), the mechanisms underlying A␤ deposition on blood vessels remains unclear.
There are many links between hemostasis and cerebrovascular pathology in AD. A␤ can activate platelets, induce microhemorrhages in the brain, and interact with several coagulation factors (3)(4)(5)(6). Aggregates of A␤ can activate coagulation factor XII (FXII) to initiate blood clotting, and can increase fibrin density and resistance to fibrinolysis (5,7). CAA deposits contain several coagulation factors, and antiplatelet therapy reduces accumulation of CAA deposits and improves cognitive function in mice (8). Currently, the biochemical mechanisms that connect intravascular CAA deposition and hemostasis are not clear.
A␤ is formed from the amyloid ␤ precursor protein (APP), which is expressed by several cells, including platelets, neurons, glial cells, and astrocytes. Platelets account for 95% of circulating APP (9). APP can be cleaved to generate A␤ peptides typically 40 or 42 residues long, A␤40 and A␤42, respectively. Platelets cleave APP and release both A␤40 and A␤42 when they are activated (10). In both blood and CAA deposits, A␤40 is more abundant than A␤42, whereas A␤42 is more abundant in senile plaques within the brain parenchyma, which is a hallmark of AD (11). Mutations within the A␤ sequence can alter the pathogenicity of the peptide; for example, patients with the Flemish or Italian mutation (A21G and E22K, respectively) have increased CAA deposits, whereas patients with the Arctic mutation (E22G) have more plaque burden (12)(13)(14)(15)(16). A␤40 and A␤42 can spontaneously aggregate into small, noncovalent oligomers and subsequently large aggregates, both of which are toxic to surrounding cells (17).
The formation of protein aggregates in CAA and AD may be regulated in part by transglutaminases (TGs). TGs are a family of enzymes that form ⑀-(␥-glutamyl) lysyl isopeptide bonds between their substrates, creating irreversible bonds. TG activity colocalizes with plaques in brains in AD (6). Tissue transglutaminase 2 (TG2) can induce A␤ oligomerization and aggregation in vitro and reduce its clearance (18). However, it is unknown if activated coagulation factor XIIIa (FXIIIa), a transglutaminase in blood plasma and on platelets, can cross-link A␤ in blood. FXIIIa is formed from coagulation factor XIII (FXIII), a protransglutaminase, when it is activated by thrombin in the presence of calcium during blood clotting (19). The primary function of FXIIIa is cross-linking fibrin to itself and to other proteins to stabilize blood clots from premature fibrinolysis. FXIIIa also increases clot stiffness and platelet adhesion to further reduce blood loss.
A␤ forms stable complexes with FXIIIa and colocalizes with FXIIIa and fibrin in CAA deposits of AD patients (6). Because A␤ is a substrate for TG2 and influences hemostasis in several ways (18), we tested if A␤ is a substrate for FXIIIa, and found that it is.

A␤40 is a substrate of FXIIIa
To test if A␤ could be covalently cross-linked by FXIIIa, A␤ was incubated with FXIIIa and changes in molecular mass were detected using Western blotting. FXIIIa cross-linked monomeric A␤40 into dimers and oligomers, and to FXIIIa itself (Fig.  1A). Cross-linked A␤ oligomers did not form when FXIIIa was inhibited by chelating calcium ions with EDTA or with T101, an irreversible inhibitor of FXIIIa transglutaminase activity. However, with T101, there was a distinct band near 85 kDa, corresponding to the molecular mass of FXIIIa attached to A␤40, which has been reported previously (6). Similar trends were observed with A␤42, although low concentrations of SDS-resistant oligomers formed without FXIIIa. When the only glutamine of A␤40 was mutated to asparagine (A␤40 Q15N), FXIIIa did not generate A␤ oligomers, indicating that the oligomerization depended on the glutamine residue of A␤40.
To determine the kinetic constants of FXIIIa-mediated A␤ cross-linking, the release of ammonia, a product of the transglutaminase reaction (Fig. 1B), was measured using a photometric assay. FXIIIa-mediated cross-linking of A␤40 had K m of 8.5 Ϯ 1.2 M and k cat of 1.3 Ϯ 0.5 s Ϫ1 , resulting in a catalytic efficiency of 1.5 Ϯ 0.5 ϫ 10 5 M Ϫ1 s Ϫ1 . A small molecule substrate of FXIIIa, glycine ethyl ester (GOE), had a similar catalytic efficiency of 2.0 Ϯ 0.7 ϫ 10 5 M Ϫ1 s Ϫ1 (Fig. 1C). The catalytic efficiency of a peptide with the A␤40 residues scrambled had 10-fold lower catalytic efficiency, indicating that the sequence of A␤40 is important for FXIIIa activity. We were not able to calculate the catalytic efficiency of A␤42 because A␤42 precipitated at the concentrations necessary to perform the assay.

FXIIIa covalently cross-linked A␤40 to fibrin
A␤ can bind to many proteins in blood, such as FXII, FXIII, and fibrinogen (4,5,7). To test if A␤40 could be covalently cross-linked to other blood proteins, plasma containing A␤40 was clotted, separated by SDS-PAGE, and immunoblotted against A␤. Within 10 min, distinct A␤ bands were visible around 50 kDa, 70 kDa, and 100 kDa, and much higher molecular mass ( Fig. 2A). The molecular mass of these bands were similar to those of the ␣ and ␥ chains of fibrin, the main substrates of FXIIIa. Bands with similar molecular mass as fibrin were visible after the ␥-␥ dimers were formed (Fig. 2, C and D). A␤ was still cross-linked when an inhibitor of TG2 (Z006) (20) was added to plasma, but not when T101 was added, indicating that FXIIIa, not TG2, cross-linked A␤ (Fig. S1). To confirm that A␤40 was cross-linked directly to fibrin, A␤40 was incubated with purified fibrinogen, FXIIIa, and thrombin. Bands of A␤ were visible around 50 kDa, 70 kDa, and 100 kDa, correlating to the molecular mass of the ␣ and ␥ chains of fibrin and crosslinked ␥-␥ dimers (Fig. 3, C and D). A␤40 was cross-linked to fibrin chains faster than to itself. A␤40 was not cross-linked to fibrin chains when FXIIIa was inhibited with T101. Lower concentrations of A␤40 (1 M) also cross-linked to both purified and plasmatic fibrin by FXIIIa (Fig. S2).

FXIIIa cross-linked A␤40 to platelet proteins under flow
Because platelets contain the FXIII-A subunit, which can be activated by high concentrations of Ca 2ϩ , A␤40 was incubated with platelets to test if platelet-derived FXIIIa could cross-link A␤40 to itself or to other proteins. When platelets were activated with ADP, collagen, or thrombin, different patterns of A␤ cross-linking were detected compared with when platelets were not activated (Fig. 3A). The A␤ bands formed with platelets had higher molecular mass than A␤ dimers and trimers, suggesting A␤ was cross-linked to platelet proteins. Both EDTA, which chelates the Ca 2ϩ required for FXIIIa activity and platelet activation, and T101 prevented A␤ oligomers from forming. In contrast, Z006 did not prevent A␤ oligomers from forming, indicating that FXIIIa, not TG2, is responsible for cross-linking A␤ in platelets.
To test whether FXIIIa cross-links A␤40 to blood clots formed under flow, plasma containing platelets and fluorescently tagged A␤40 were flowed through a microfluidic device.

Factor XIIIa covalently cross-links amyloid ␤
A␤40 accumulated on the clot, and colocalized directly on platelet aggregates and fibrin fibers. The co-localization of A␤40 with platelets, measured by the ratio of A␤40 fluorescence to platelet fluorescence, was significantly decreased when T101 was added, indicating that FXIIIa can cross-link A␤40 to blood clots under flow (Fig. 3, B and C).

A␤40 increases clot stiffness of PRP and PPP via FXIIIa
Because cross-linking increases fibrin stiffness (19), we tested the effect of A␤40 on fibrin clot stiffness using thromboelastography (TEG). When whole blood was clotted in the presence of A␤40, no significant difference in clot stiffness was observed (Fig. 4A). Because the contribution of red blood cells may have masked subtle differences of A␤ on clot stiffness, we tested if the influence of A␤40 on fibrin could be detected when red blood cells were removed (21). A␤40 increased the stiffness of clots formed in platelet-rich plasma (PRP) and platelet-poor plasma (PPP) by 27 and 39%, respectively (Fig. 4, B and C). The increase in clot stiffness depended on both cross-linking by FXIIIa and platelet-platelet interactions, since inhibitors of FXIIIa (T101) or integrin ␣II b ␤ 3 (eptifibatide) abrogated the increase of clot stiffness induced by A␤40.

A␤40 mutants are differentially cross-linked by FXIIIa
Certain point mutations of A␤ increase the probability of developing CAA (12)(13)(14)(15)(16). To test whether FXIIIa cross-links  Oligomerization occurred in some mutants even in the absence of FXIIIa, which is consistent with the high propensity of these mutants to aggregate. However, when the mutants were incubated with FXIIIa, different species of A␤ oligomers were formed in varying concentrations. For example, the Flemish mutation (A21G) was cross-linked to a greater extent than WT, whereas the Iowa/Dutch mutation (D23N/E22Q) was crosslinked to a lesser extent (Fig. 5A). The catalytic efficiencies of these mutant A␤40 sequences were lower than, or comparable to, the catalytic efficiency of WT A␤40 (Fig. 5B). The Flemish mutation was also cross-linked to fibrin to a greater extent than WT (Fig. 5, C and D).

Discussion
The results show that A␤40 is covalently cross-linked by FXIIIa, both to itself to form dimers and oligomers and also to other blood proteins in plasma, such as fibrin. FXIIIa also crosslinked A␤40 to clots under flow, and the cross-linking of A␤40 increased clot stiffnesses in PRP and PPP, but not in whole blood. Although the reaction occurs in vitro, the physiological relevance and significance of these reactions in vivo must be further investigated.
The cross-linking of A␤40 to fibrin chains was visible only after the ␥-␥ dimers were formed. This is consistent with the kinetic data, where the catalytic efficiency of FXIIIa and A␤ (k cat /K m ϭ 1.5 Ϯ 0.5 ϫ 10 5 M Ϫ1 s Ϫ1 ) was lower than that of fibrin ␥-chains (5.1 ϫ 10 7 M Ϫ1 s Ϫ1 ) (22). A␤ was cross-linked to fibrin at an A␤ concentration of 1 M, which is a concentration that may occur at sites of cerebrovascular damage (23).
The cross-linking of A␤ may potentially be influenced by A␤-albumin interactions. Albumin sequesters ϳ90% of A␤ in plasma and preferentially binds oligomeric A␤ to monomeric A␤ (24,25). FXIIIa cross-linked A␤ to fibrin both in buffer and in plasma at similar rates, suggesting that albumin does not play a significant role in influencing the rate of cross-linking in these conditions. However, how albumin affects the clearance of cross-linked oligomeric A␤ requires further examination.
A␤ and FXIIIa can form stable complexes in vitro, and FXIIIa is catalytically active in vessels with CAA (6). Although isopep-tide bonds formed by transglutaminases have been discovered in CAA, covalent cross-linking of A␤ by FXIIIa was not detected ex vivo previously (6). The discrepancy with the data here may be because of the higher, although physiological, FXIIIa concentrations and longer reaction times used, and potentially higher specific activity of FXIIIa. Although the cross-linking of A␤ to itself was visible only after 3 h, crosslinking to fibrin occurred within minutes.
Cross-linking of A␤ may have implications in at least two scenarios. First, A␤ may modify clot structure at sites of damage in the cerebral vasculature or at platelet aggregates. It remains to be determined what the effect on clotting might be, but it may contribute to fibrinolysis because non-cross-linked aggregates of A␤ increase resistance to fibrinolysis and activate the coagulation cascade through FXII (5,26). Cross-linking A␤ to fibrin could enhance clotting by localizing the platelet activating sequence (A␤25-A␤35) to fibrin (3). Cross-linking of A␤ may have a greater significance in arteries than veins, as arterial clots have fewer red blood cells, because A␤ increased the stiffness of PRP, but not whole blood clots.
Second, FXIIIa-mediated activity may contribute to CAA and AD pathology. Given that blocking the binding between A␤ and fibrin with a small molecule can improve cognitive impairment in mouse models of AD, covalent cross-linking of A␤ and fibrin may exacerbate CAA pathology (26). In patients with AD, there is a higher frequency of a FXIII allele (V34L) that undergoes faster activation, suggesting that accelerated cross-linking can aggravate AD development (27). The differences in crosslinking between the A␤ variants and mutants provides further insight to the potential significance of the interaction between FXIIIa and A␤. FXIIIa cross-linked A␤40 to a higher extent than A␤42, providing a potential explanation as to why A␤40 is the more prominent form of A␤ within CAA. AD patients with the Flemish mutation (A␤40 A21G) have increased CAA phenotype, and A␤40 with the Flemish mutation was cross-linked to a higher extent compared with WT A␤40 (17). An alternative hypothesis is that cross-linking of A␤ by FXIIIa is a physiological process that is separate from aggregation and amyloid accumulation.
In conclusion, synthetic A␤40 can be covalently cross-linked to itself, and to fibrin and platelet proteins by FXIIIa under flow. Given that A␤ and FXIIIa colocalize within CAA, these results

Factor XIIIa covalently cross-links amyloid ␤
provide motivation to test if FXIIIa contributes to the accumulation of intravascular deposits of A␤ in CAA.

Platelet preparation
This study was approved by the Research Ethics Board of the University of British Columbia (H12-01516), and written informed consent was obtained from all healthy volunteers in accordance with the Declaration of Helsinki. Platelets and PRP were isolated as described previously (28). Platelets were resuspended in Tyrode's buffer (119 mM NaCl, 5 mM KCl, 25 mM HEPES, 2 mM CaCl 2 , 2 mM MgCl 2 , 6 g/liter glucose, pH 6.5) or in plasma at a final concentration of 2 ϫ 10 9 platelets/ml or as otherwise specified.

Cross-linking of A␤
A␤ peptides (AnaSpec, Fremont, CA) were initially dissolved in dimethyl sulfoxide at 20 mg/ml and diluted with 25 mM HEPES buffer to 1 mg/ml. To test if A␤ is cross-linked by puri-

Kinetics assay
The rate of cross-linking of A␤ was determined by measuring the rate at which ammonia was produced during the TG reaction, using steady-state kinetics at 37°C. A calibration curve of ammonia concentration and absorbance at 570 nm was determined with a Tecan M200 plate reader (BioVision Inc., Milpitas, CA). FXIIIa was mixed with A␤ peptides or GOE (5-50 M) as amine donors, a glutamine-containing peptide (NQEQVS-PLTLLK, 1 mM), DTT (40 M), and CaCl 2 (3 mM). Aliquots from the transglutaminase reaction mixture were removed and quenched every 15 min with EDTA (15 mM). Kinetic parameters were calculated using graphing software (OriginPro 9.1).

Microfluidic analysis
Microfluidic devices were prepared from polydimethylsiloxane (PDMS) as described previously (29). The channels were coated with lipid vesicles containing tissue factor, phosphatidylserine (PS) and phosphatidylcholine (PC); the rest of the device was coated with inert PC vesicles. The vesicles were prepared and devices were coated as described previously (30). Citrated PRP containing fluorescent A␤40-TAMRA (100 g/ml, AnaSpec) and fluorescent ␣-CD42b-FITC antibodies (1:100; eBioscience, San Diego, CA) with or without T101 (500 M) was flowed into the device at 1 l/min along with a calcium-saline solution (40 mM CaCl 2 , 90 mM NaCl) at a rate of 0.33 l/min, which corresponds to venous shear rates (20 s Ϫ1 ). Clotting was monitored using an epifluorescence microscope (Leica DMI6000B). The clots were then washed with calcium-saline solution (40 mM CaCl 2 , 90 mM NaCl) at a rate of 5 l/min for 10 min and imaged. For statistical analysis, fluorescence intensi-ties were measured at five equally distributed locations along the length of the channel.

Statistical evaluation
Statistical analyses were performed using GraphPad Prism 7.0. All results presented in graphs are the mean Ϯ S.E. N indicates number of independent experiments, performed on separate days. A two-tailed unpaired Student's t test was used for all analyses. Significance was designated at p values Ͻ0.05.