GAPDH in CSF contributes to Aβ toxicity
To test whether GAPDH is released from cells affected by exogenous Aβ, we utilized a human neuroblastoma cell line, SH-SY5Y [31]. The cells were incubated with Aβ42 peptide for indicated times, and the profiles of cell death, as determined by lactate dehydrogenase activity in culture medium, and the extracellular GAPDH concentration, as measured using a sandwich-type ELISA, were compared (Fig. 1a). The matching profiles indicate that the release of GAPDH in the intercellular space is coupled to the toxic effect of exogenous Aβ42.
To determine whether GAPDH progressively accumulates in CSF of AD patients, we studied samples from first cohort of patients with Mild Cognitive Impairment (MCI) (n = 22), as well as with mild (n = 41), moderate (n = 49), and severe (n = 51) AD, stratified based on Mini-Mental State Examination (MMSE) scores [20] (Table S1). The level of GAPDH measured with the sandwich-type ELISA immunoassay was low in samples from MCI and ‘mild AD’ patients, and progressively increased with the AD stage, reaching 12.5 ± 1.5 ng/mL in the ‘severe AD’ group (Fig. 1b).
To determine whether the extracellular GAPDH impacted cell death, 50 μL of each CSF sample was mixed with 50 μL of growth medium and added to SH-SY5Y cell cultures (Fig. 1c). Cell death increased progressively as a function of AD stage, reaching 19.2 ± 0.9% for samples from the ‘severe AD’ group, suggesting that GAPDH contributes to CSF toxicity (Fig. 1c).
We next investigated the aggregation status of GAPDH and Aβ42 in CSF. Samples from the second cohort from MCI (n = 6), mild AD (n = 8), moderate AD (n = 7), and severe AD (n = 6) groups, were examined (Table S2). To prevent the loss of aggregates we avoided centrifugation. Instead, the sodium dodecyl sulfate (SDS)-insoluble aggregating material was collected by ultrafiltration using a 96-well dot-blotting manifold; the membrane was stained with an antibody against Aβ (Fig. 1d), and another identical 96-array dot-blot was stained with an anti-GAPDH antibody (Fig. 1e). The amount of Aβ42-GAPDH co-aggregates trapped on the filter strongly correlated with the disease stage. The difference in dot density between MCI samples and samples from patients with severe AD was 4-fold for Aβ42 (Fig. 1d) and 7-fold for GAPDH (Fig.1e).
To determine whether extracellular GAPDH forms a complex with Aβ in CSF of AD patients we used Fönster resonant energy transfer (FRET). To detect photon transfer from a fluorophore-donor, Alexa 488 conjugated with Aβ antibody, to a fluorophore-acceptor, CF-555 conjugated with GAPDH antibody, we obtained an emission spectrum in the range of 516-600 nm at an excitation wavelength 490 nm. We found a progressive increase in FRET signal at 568 nm, which directly correlated with the disease stage (Fig. 1f). Although the change in the FRET signal was noticeable in CSF from ‘Mild AD’ patients (6,1 ± 0,2 vs 4,8 ± 0,7 arb.un), it was not statistically significant. However, FRET changes for samples from ‘Moderate’ and ‘Severe’ AD patients were significantly higher than in the control (Fig. 1f). We conclude that the extracellular GAPDH forms a complex with Aβ in CSF of AD patients and the more advanced the stage of the disease the higher is the concentration and toxicity of these complexes.
Covalent bonding between Aβ42and GAPDH promotes toxic co-aggregation
Our findings that Aβ42 and GAPDH persist as SDS-insoluble aggregates in the CSF of patients with AD and previous reports linking tTG to AD and Huntington disease pathology [12,32] prompted us to examine the possibility of tTG-mediated covalent interaction between the two polypeptides.
tTG-mediated bond formation may occur between surface exposed lysines of GAPDH and glutamine 15 of Aβ42 [33]. To test this possibility, we designed a protein-protein interaction assay in which GAPDH was immobilized at the bottom of a 96-well plate, and then biotinylated Aβ42 or Aβ42 with a Q15A substitution (Aβ42Q15A) were added in the presence of tTG alone, or together with cystamine, a tTG inhibitor, followed by washing. tTG enhanced the binding of GAPDH to Aβ42, but not to Aβ42Q15A, by approximately 50%, whereas cystamine abolished this effect (Fig. 2a).
To further investigate the stable association between GAPDH and Aβ42, the mixtures containing GAPDH, Aβ42, tTG, and cystamine were prepared in combinations as shown in Fig. S1, incubated for 12 hours, followed by SDS gel separation and Western blotting (Fig. S1). The only visible band stained with anti-Aβ antibody has the same mobility as GAPDH (36 kDa) (upper panel). This band, which appeared after gel separation of the mixture containing GAPDH, Aβ42, and tTG, constitutes the SDS-insoluble complex of Aβ-GAPDH.
The N-terminal fragment (residues 1-16) appears to be the minimal part of Aβ capable of triggering amyloidosis in vivo [34]. We therefore examined the ability of Aβ16 and Aβ16Q15A, as well as Aβ42 and Aβ42Q15A, to bind GAPDH in the presence or absence of tTG using a filter-trap assay (Fig. 2b). Both Aβ16 and Aβ42 formed co-aggregates with GAPDH (Fig. 2 b,c). The level of aggregation was much higher in the presence of tTG. Q15A substitution greatly reduced the ability of Aβ42 and Aβ16 to form such aggregates (Fig. 2 b,c).
To further establish the role of Q15 in the Aβ-GAPDH complex, a competition protein-protein interaction assay was designed in which GAPDH was immobilized at the bottom of a 96-well plate, and biotinylated Aβ42 was added to the wells together with non-tagged Aβ peptides at various concentrations in the presence of tTG (Fig. 2d). Aβ42 and Aβ16 peptides, but not their Q15A derivatives, reduced the binding of biotinylated Aβ42 to GAPDH in a dose-dependent manner (Fig. 2d), thus directly supporting the role of Q15 in tTG-mediated Aβ-GAPDH complex formation.
We next employed atomic force microscopy (AFM) to examine the structure of GAPDH-Aβ complexes (Fig. 2e). Aβ42 forms only tiny fibril-like structures, whereas Aβ16 does not form any visible fibrils. However, in the presence of GAPDH, and especially in combination with tTG, the fibril-like structures become obvious; their size increases for both Aβ42 and Aβ16. Notably, the aggregates found in the GAPDH mixtures with Aβ16Q15A or Aβ42Q15A were amorphous, closely resembling those produced by GAPDH alone (Fig. 2e).
As shown in Fig.1c, CSF samples from AD patients are toxic to SH-SY5Y cells. To test whether Aβ peptides with a Q15A substitution retain their toxicity, cultures of SH-SY5Y cells were incubated with Aβ42/16 peptides, or their Q15A derivatives, along with GAPDH or GAPDH+tTG. GAPDH alone or with tTG was minimally toxic (Fig. 2f,g). The tTG enzyme itself was not toxic. The number of dead cells significantly increased in the presence of Aβ42/16 and GAPDH (19.5 ± 2.4% and 31.6 ± 0.3%, respectively) and further increased in the presence of tTG (31.5 ± 4.6% and 36.1 ± 0.4% for Aβ16 and Aβ42, respectively). Q15A derivatives demonstrated much less toxicity under the same conditions (Fig. 2g).
Finally, we examined whether a protein other than GAPDH, which can also be released from dying neurons, formed insoluble complexes with Aβ42. Lactate dehydrogenase (LDH) is an abundant enzyme that is solely cytoplasmic and serves as a common indicator of cell death if found intracellularly [35]. We used utrafiltration to compare the abundance of aggregates formed by Aβ42 in the presence of GAPDH+tTG or LDH+tTG (Fig. S2). The amount of Aβ42 aggregates formed in the presence of LDH+tTG did not differ statistically from those formed by Aβ42 alone (adjusted P value 0.2365). In contrast, Aβ42 aggregates formed with GAPDH were at least four-fold more abundant than those obtained with LDH (Fig. S2).
Extracellular GAPDH in murine hippocampus exacerbates AD pathology
As shown above, the extracellular complex of GAPDH-Aβ is cytotoxic for cultured neuronal cells. To test whether this would also be the case in mammals in vivo, we adopted two animal AD models: a chemical model, in which exogenous Aβ42 was introduced into the rat hippocampus, and a genetic model using 5XFAD mice. Animals were surgically infected with lentiviruses bearing either a plasmid overexpressing GAPDH (pLOC-GAPDH) or shRNA targeting GAPDH to generate, respectively, animals with high or low GAPDH expression in the hippocampus. Nine weeks after lentivirus administration, GAPDH levels in the hippocampi of the animals were estimated by immunoblotting. Rats infected with shGAPDH or pLOC-GAPDH displayed, respectively, a 69% decrease or 9-fold increase in GAPDH, as compared to untreated animals (Fig. 3a). Similar down- and up-regulation of GAPDH occurred in the wild-type (wt) and 5XFAD mice (Fig. 3b and Fig. S3).
MRI was used to estimate the damage to the rat hippocampus caused by Aβ42 introduced into the CA1 area. Image analysis showed that the injury zone in Aβ42-treated animals positively correlated with the GAPDH level in the hippocampus (Fig. 3c). The ratio between the lesion volume and the total hippocampus was the highest in the pLOC-GAPDH+Aβ42 group (14.98%) and minimal in the shGAPDH+Aβ42 group (2.88%), which is similar to that in sham-operated rats (2.52%) (Fig. 3c,d).
To determine whether the tissue damage was due to apoptotic death of hippocampal cells, histological sections of the hippocampus were stained with the TUNEL assay. The level of apoptosis positively correlated with the level of GAPDH (Fig. 3e,f). The stereotactic surgery procedure caused some cell death in the brain; the number of apoptotic cells in operated animals in all control groups was approximately 20%. However, injection of Aβ42 caused the accumulation of 39.6 ± 3.3% apoptotic cells while the injection of pLOC-GAPDH+Aβ42 caused the accumulation of 45.6 ± 4.9% apoptotic cells (Fig. 3e,f). Remarkably, the level of apoptosis in the brains of rats injected with shGAPDH+Aβ42 was similar to that of sham-operated animals (24.5 ± 0.3%) (Fig. 3e,f).
Analogous to the rat model, the level of apoptotic cells in the hippocampus of 5XFAD mice was elevated 3.5-fold by the injection of pLOC-GAPDH, as compared to untreated 5XFAD mice (35.1 ± 0.5 vs. 10.5 ± 0.7) (Fig. 3g,h). The number of apoptotic cells in mice with knocked-down GAPDH was close to that of sham-operated 5XFAD mice (13.2 ± 1.7% vs. 10.5 ± 0.7%) (Fig. 3g,h). Thus, the level of GAPDH in the hippocampus of AD-simulating animals directly correlates with the level of apoptosis associated with AD pathology.
We next examined the behavioral effect of endogenous GAPDH in AD model animals. Wistar male rats at P180 with high and low GAPDH were tested in a Morris water maze at P240 to investigate learning function. In both models of AD, animals with elevated levels of GAPDH in the hippocampus demonstrated significant cognitive impairment. The platform searching time for rats injected with pLOC-GAPDH+Aβ42 was, respectively, 1.6-fold and 2.4-fold longer than for the Aβ42 group and the untreated group (Fig. 4a). In the 5XFAD mouse model, animals injected with pLOC-GAPDH demonstrated a 2-fold delay in searching comparing to that in untreated animals. In contrast, animals injected with shGAPDH displayed a shorter searching time than untreated animals (Fig. 4b).
To correlate the degree of memory deficit with GAPDH cytotoxicity, we obtained CSF from each group of rats and estimated its effect on SH-SY5Y neuroblastoma cell survival. As expected, the most toxic CSF was obtained from the animals injected with pLOC-GAPDH+Aβ42, killing 22.2 ± 3.1% of the cells (Fig. 4c).
The enhanced cytotoxicity of CSF obtained from AD patients was linked to aggregation of Aβ-GAPDH (Fig. 1). To confirm this finding in the AD animal models, the amount of Aβ-GAPDH aggregates in CSF and hippocampi from all treated animals was estimated using ultrafiltration. SDS-insoluble aggregates containing both Aβ42 and GAPDH were found in the hippocampus and CSF of rats injected with Aβ42 and pLOC-GAPDH+Aβ42 (Fig. 5). Likewise, the level of Aβ42 and GAPDH aggregation in 5XFAD mice injected with pLOC-GAPDH was significantly higher than in control mice (Fig. 5).
GAPDH ligand reduces Aβ aggregation and ameliorates spatial memory in 5XFAD mice
Previously, we have showed that a cortisol derivative (Fig. 6a), hydrocortisone 21-hemisuccinate (RX-624), specifically binds GAPDH and reduces its aggregation [36]. To test whether RX-624 is capable of reducing aggregation of Aβ-GAPDH complexes as well, we mixed Aβ42, GAPDH and tTG in the presence of 0.3, 1.0 and 3.0 µM of RX-624 followed by ultrafiltration and immunoblotting. RX-624 diminished aggregation of Aβ-GAPDH complexes in a dose-dependent manner, reaching 80-fold reduction at 3.0 µM (Fig. 6b).
To test the effectiveness of RX-624 in animal models of AD, wt and 5XFAD mice were divided into three groups: no treatment, vehicle, and RX-624 (2 mg/kg/injection). The treatment started at P70 of age and lasted for next 150 days, one injection per week. Two weeks after the last injection spatial memory of experimental mice was evaluated using the Morris water maze. 5XFAD mice treated with RX-624 demonstrated significant memory recovery; the platform searching time was similar to that of wt mice (22.7 ± 4.4 vs. 26.2 ± 6.6 sec) and was greatly reduced compared to that of untreated or vehicle-treated 5XFAD mice (Fig. 6c). Analysis of swimming paths performed following the instructions given by Gehring et al. [37], also demonstrated that the learning capacity of 5XFAD mice treated with RX-624 approached that of wt mice (Fig. 6d).
Next, the hippocampi from two animals of all three groups were isolated and subjected to ultrafiltration. SDS-insoluble aggregates containing both Aβ and GAPDH were detected in control 5XFAD animals, whereas the hippocampi of mice treated with RX-624 were completely free of aggregates, demonstrating the crucial role of GAPDH in promoting the formation of extracellular amyloid complexes (Fig. 6e).