Main

Therapies designed to clear pathogenic Aβ deposits from the brains of human patients hold great promise for the treatment and possible cure of Alzheimer's disease (AD)1. Previous studies in multiple mouse models of AD have demonstrated that passive immune therapies targeting Aβ can effectively reduce Aβ pathology, structural neuronal abnormalities and, in some cases, behavioral deficits2,3. However, these encouraging preclinical results have failed to predict clinical efficacy in recent phase 3 clinical trials4. A better understanding of the cellular mechanisms of the antibodies in the diseased brain may help to explain this discrepancy. Although such studies are not feasible in humans, two-photon Ca2+ imaging allows simultaneous monitoring of the activity of large numbers of neurons in mouse models in vivo5. Using this method, previous work in transgenic AD models overexpressing mutant human amyloid precursor protein (APP) revealed that Aβ pathology promotes hyperexcitation and tonic hyperactivity of a large number of neurons in hippocampal6 and neocortical7 brain regions, thereby impairing the normal function of neuronal circuits involved in cognitive information processing8. Such hyperactivity may increase the risk for epileptogenic activity in both animal models9 and human patients10, particularly in those carrying familial AD mutations11.

We explored whether treatment with Aβ-targeting antibodies can reverse these functional neuronal impairments. First, we used two-photon Ca2+ imaging in the PDAPP mouse model and found, consistent with previous observations in other models7,8, a marked increase of neuronal activity levels in the neocortex of 12–17-month-old plaque-bearing mice when compared with age-matched wild-type (WT) littermates (Supplementary Fig. 1a,b). The frequency distribution of the action potential–evoked neuronal Ca2+ transients was markedly shifted toward higher values in PDAPP mice (Supplementary Fig. 1c,d). Furthermore, the fraction of neurons that were hyperactive (>6 transients per min) was markedly larger in PDAPP than in WT mice, while the fraction of silent neurons (0 transients per min) was not different between PDAPP and WT mice (Supplementary Fig. 1e).

In an attempt to treat the neuronal dysfunction, we intraperitoneally injected plaque-bearing PDAPP mice with the monoclonal antibody 3D6 (the murine equivalent of bapineuzumab4) at a weekly dose of 10 mg per kg of body weight over a period of 3 months, according to standard protocols2. Control PDAPP mice received an irrelevant, isotype-matched antibody. We analyzed two independent age cohorts of mice that had different Aβ loads in their brains before the treatment. A younger group was immunized at 12 months of age and an older group was treated beginning at 14 months of age. As expected2, the treatment reduced brain Aβ burden (Fig. 1a and Supplementary Fig. 2a). In marked contrast with this marked decrease of Aβ pathology, we observed in the same transgenic mice an aggravation of neuronal dysfunction instead of the expected amelioration (Fig. 1b–g and Supplementary Fig. 2b,c). This aggravation manifested itself in several ways. First, we found in both age cohorts that the average frequency of Ca2+ transients was doubled in treated animals compared with controls (Fig. 1b and Supplementary Fig. 2b) and was almost fivefold higher than in WT mice (WT: 1.5 ± 0.24 transients per min, n = 9 mice, t = 3.68, d.f. = 6.25, P = 0.01). Second, the proportion of hyperactive neurons was three- to fourfold larger in treated than in control mice (Fig. 1c and Supplementary Fig. 2c). Finally, in a fraction of treated animals (47%), the increased hyperactivity was associated with an unusual neuronal synchrony (Fig. 1d–g). Such synchrony, which has not been observed in WT mice and is only rarely seen in untreated PDAPP mice, may further promote the epileptogenesis observed in many mouse models of AD9. Next, we wondered whether antibody treatment could ameliorate neuronal dysfunction at an earlier disease stage, before plaque deposition. For this purpose, we studied PDAPP mice at 5 months of age and found hyperactivity already at this young age. Notably, even a single dose of 3D6 antibody (30 mg per kg; for detailed protocol, see ref. 12), applied 24 h before imaging, was sufficient to aggravate hyperactivity (Fig. 1h–j). The treatment of WT mice with 3D6 antibodies had no significant effect (P = 0.875; Fig. 1k). These results suggested that the 'pro-excitatory' effect of the anti-Aβ antibody 3D6 was dependent on APP overexpression in the transgenic AD model and, most likely, the binding to Aβ, and that it can occur even in the absence of plaques.

Figure 1: Anti-Aβ antibody 3D6 reduces amyloid pathology but aggravates neuronal dysfunction.
figure 1

(a) Amyloid burden was markedly reduced in 3D6-treated (green) compared with isotype-treated control (black) PDAPP mice. Each dot represents an individual animal, the horizontal bar represents the mean and the error bars represent ±s.e.m. (control = 33.8 ± 5.7%, n = 10 mice; 3D6 = 14.0 ± 3.4%, n = 9 mice; two-sample t test, t = 2.88, d.f. = 17, P = 0.01). (b) Frequency of Ca2+ transients in control (gray) and 3D6-treated (green) PDAPP mice. The box represents the interquartile range (25th to 75th percentiles) and whiskers extend to 1.5× the interquartile range from each side, corresponding to a coverage of 99.3% of the data in case of a normally distributed population. Center line marks the mean. The difference between the two groups was significant (control = 3.8 ± 0.8 transients per min, n = 9 mice; 3D6 = 7.6 ± 1.6 transients per min, n = 7 mice; two-sample t test, t = 2.19, d.f. = 14, P = 0.046). (c) Frequency distributions and pie charts show higher fractions of hyperactive neurons (red) in 3D6-treated (56.2 ± 13.4%, n = 7 mice) when compared with isotype-treated (19.3 ± 6.5%, n = 9 mice) PDAPP mice (two-sample t test, t = 2.48, d.f. = 8.81, P = 0.036). (d,e) Top, layer 2/3 cortical neurons imaged in vivo and activity maps, where hue is determined by the frequency of spontaneous Ca2+ transients in control (d) and 3D6-treated (e) PDAPP mice. Bottom, abnormally high synchrony of Ca2+ transients in 3D6-treated mice compared with control mice. (f,g) Superimposed traces from the shaded areas in d (f) and e (g). Each color represents a different cell. (h–j) Cumulative distributions (h; control, n = 1,414 neurons; 3D6, n = 1,393 neurons; Kolmogorov-Smirnov test, P < 0.001), average frequencies of Ca2+ transients (i; control = 4.7 ± 0.8 transients per min, n = 13 mice; 3D6 = 7.8 ± 0.7 transients per min, n = 10 mice; two-sample t test, t = −2.79, d.f. = 21, P = 0.011) and fractions of hyperactive neurons (j; control = 35.2 ± 8.3%, n = 13 mice; 3D6 = 57.5 ± 5.5%, n = 10 mice; two-sample t test, t = −2.24, d.f. = 19.73, P = 0.037) in pre-depositing PDAPP mice after acute treatment with 3D6 or control antibody. (k) Acute treatment with 3D6 (green) had no detectable effect on cortical activity levels in WT mice (isotype-treated WT = 0.8 ± 0.3 transients per min, n = 3 mice; 3D6-treated WT = 0.9 ± 0.2 transients per min, n = 5 mice; two-sample t test, t = −0.16, d.f. = 6, P = 0.875). *P < 0.05; ns, not significant.

Full size image

To validate these results with a different monoclonal antibody against Aβ, we passively immunized another APP-overexpressing model of AD that exhibits neuronal hyperactivity (Fig. 2a,b), namely the Tg2576 mouse model, with β1 antibodies13. It has been shown that β1 can reduce Aβ plaque loads in aged AD transgenic mice treated for 5 months, but, overall, β1 appeared to be less effective at clearing plaques than 3D6 (ref. 13). For consistency with the 3D6 antibody therapy (see above), we treated Tg2576 mice for 3 months with β1 antibodies and found that this treatment had no major effect on soluble and insoluble Aβ levels (Supplementary Fig. 3), allowing us to ask whether the aggravation of neuronal impairments depends on the reduction in Aβ burden. Indeed, after 3 months of treatment, cortical activity levels, as well as the fractions of hyperactive neurons, were massively elevated in β1-treated animals when compared with mice that received isotype-matched control antibodies (Fig. 2a–c). This experiment suggests, similar to the observations made in 3D6-treated, pre-depositing PDAPP mice (Fig. 1h–j), that the worsening of neuronal dysfunction by anti-Aβ antibodies can occur independently of the effects on amyloid plaque pathology. In control experiments, performed to validate and extend the previous result with acute injections of 3D6 antibodies in WT mice (Fig. 1k), we administered β1 antibodies chronically over a period of 3 months to WT mice. Cortical activity levels were not significantly different between β1-treated and control WT mice (P = 0.763; Fig. 2d).

Figure 2: Worsening of neuronal dysfunction by anti-Aβ antibodies can occur independently of the effects on Aβ pathology.
figure 2

(a) Top, representative in vivo activity maps in WT (left) as well as isotype-treated (middle) and β1-treated (right) Tg2576 mice. Bottom, Ca2+ transients of neurons indicated above. The further aggravation of neuronal hyperactivity (middle) after β1 treatment (right) is clearly visible. (b) Frequency distributions and pie charts showing the fractions of hyperactive neurons (red) in control (top; 31.7 ± 8.7%, n = 7 mice) and β1-treated (bottom; 59.5 ± 8.9%, n = 10 mice) Tg2576 mice (two-sample t test, t = 2.15, d.f. = 15, P = 0.038). (c) Frequency of Ca2+ transients in control (gray; 4.7 ± 1.1 transients per min, n = 7 mice) and β1-treated (blue; 10.6 ± 2.2 transients per min, n = 10 mice) Tg2576 mice (two-sample t test, t = 2.43, d.f. = 12.91, P = 0.031). Data are presented as in Figure 1. (d) Chronic treatment with β1 (blue) had no detectable effect on cortical activity levels in WT mice (control = 1.5 ± 0.3 transients per min, 3D6 = 1.6 ± 0.3 transients per min, n = 4 mice each; two-sample t test, t = 0.32, d.f. = 6, P = 0.763). *P < 0.05; ns, not significant.

Full size image

Finally, we investigated the possible contribution of inflammatory processes to the Aβ antibody–mediated aggravation of neuronal hyperactivity (Supplementary Fig. 4). First, we performed a treatment with the steroidal anti-inflammatory agent dexamethasone. Because long-term dexamethasone treatment, comparable to the β1 treatment, is not feasible in APP transgenic mice14, we relied on a short-term, high-dose delivery (Online Methods), as is frequently used in humans with autoimmune and chronic inflammatory diseases15. The cortical activity levels in the β1-treated transgenic mice were not affected by dexamethasone treatment (Supplementary Fig. 4a). To test the consequences of an inflammatory reaction, we locally applied lipopolysaccharide (LPS; 1 mg ml−1) to neurons in the cortex of transgenic mice, but we did not detect any changes in activity levels (Supplementary Fig. 4b,c). Consistent with these results, the analysis of pro-inflammatory cytokines and chemokines in brain homogenates did not show detectable differences between β1-treated mice and the control group (Supplementary Table 1). Thus, under our experimental conditions, we found no evidence for a prominent role of inflammation in promoting or aggravating neuronal hyperactivity.

In conclusion, our results from independent cohorts of AD transgenic mice demonstrate that passive immunotherapy with Aβ-targeting antibodies was not only ineffective in treating neuronal dysfunction, but actually worsened it. Administration of two different monoclonal antibodies resulted in significantly increased numbers of pathologically hyperactive neurons and promoted, in some cases, abnormal synchrony of cortical activity. Our findings indicate that the previously reported beneficial effects of antibody treatment on neuronal structure, including recovery of neuritic dystrophy16 or prevention of synaptophysin loss17, are not sufficient for the repair of neuronal dysfunction. Some previous animal studies have shown an amelioration of behavioral deficits after antibody treatment3, and it is unclear, at this stage, how the discrepancy between such behavioral improvement and worsening of neuronal dysfunction can be resolved. Furthermore, it is currently not known whether the increase in neuronal hyperactivity, as measured by two-photon imaging, has any bearing on the efficacy, or lack thereof4, of antibodies to Aβ in human patients with AD. Our results provide evidence that the aggravation of neuronal hyperactivity is directly related to binding of the antibodies to Aβ rather than to some non-Aβ–related properties of the antibodies. Although inflammatory processes did not seem to have a prominent role under our experimental conditions, we cannot exclude the possibility that inflammatory effects of passive immunotherapy could contribute to increased neuronal hyperactivity in transgenic animals. The pro-excitatory effect was not observed in WT animals; thus, even if the increased hyperactivity involves inflammation, it must be Aβ dependent. Regardless of the precise mechanisms involved, however, the lack of improvement of neuronal function and the risk of aggravation of dysfunction by antibodies to Aβ was unexpected and highlights the urgent need to include functional in vivo assays into the toolbox of methods for the preclinical assessment of treatment strategies for AD. Together, our results suggest a cellular mechanism that, in combination with other factors18,19,20, may be the reason for the failure of anti-Aβ immunotherapies in repairing cognitive deficits4.

Methods

All experimental procedures were in compliance with institutional animal welfare guidelines and were approved by the state government of Bavaria, Germany.

Immunization design.

In the 3D6 experiments, we used female, heterozygous PDAPP transgenic mice. These mice overexpress human mutant (V717F) amyloid precursor protein (APP) throughout their lifespans and begin to deposit Aβ plaques in their brains by 6–9 months of age21,22. At 12 months of age, there is abundant plaque deposition throughout the cortex and between 12 and 16 months of age a further drastic increase in deposits is detectable22. In initial experiments, we compared 12–17-month-old PDAPP mice with age-matched WT littermates to assess differences in activity levels of layer 2/3 neurons in the fronto-parietal cortex, measured with in vivo two-photon Ca2+ imaging. We then performed passive immunization experiments using standard protocols2,23,24. We used two age cohorts (12 and 14 months of age) to start with a different Aβ burden. The animals were injected intraperitoneally once weekly with 3D6 monoclonal antibodies, which recognize amino acids 1–5 of the N terminus of human Aβ, or control antibodies (TY11-15, IgG2a isotype) at a dose of 10 mg per kg for a duration of 3 months. In a third group of PDAPP mice, 5 months of age, animals received a single intraperitoneal injection of 3D6 or control antibody at 30 mg per kg, 24 h before the in vivo imaging, according to a previous protocol12. In a final set of experiments, we injected 5 months old WT mice with 3D6 or control antibody at 30 mg per kg, 24 h before the in vivo imaging. In the β1 experiments, we used male, heterozygous Tg2576 mice that overexpress mutant human APP bearing the familial Swedish AD mutations at positions 670/671 (ref. 25). 12 months old Tg2576 mice were passively immunized weekly for 3 months by intraperitoneal injections of β1 mouse monoclonal IgG1 antibody26 recognizing amino acids 3–6 of human Aβ as minimal epitope or isotype-matched control antibody at a dose of 10 mg per kg (for better comparability of the results we used the same immunization parameters as in the 3D6 experiments).

In vivo two-photon Ca2+ imaging.

Animals were prepared for two-photon imaging as described previously7,27,28. Briefly, mice were anesthetized with isoflurane (1–1.5% (vol/vol) for induction and during surgery). A reduced concentration of isoflurane (0.8–1%) was used later during the imaging. We removed the skin above the skull and glued a custom-made recording chamber to the skull with cyanoacrylate glue (UHU GmbH). Then, we made a small cranial window over the right fronto-parietal region of the neocortex and perfused the recording chamber with warm (37 °C) artificial cerebrospinal fluid (125 mM NaCl, 4.5 mM KCl, 26 mM NaHCO3, 1.25 mM NaH2PO4, 2 mM CaCl2, 1 mM MgCl2, 20 mM glucose, pH 7.4 when bubbled with 95% O2 and 5% CO2). The dura was left intact and extreme care was taken not to damage the cortical tissue. Bolus loading of layer 2/3 cortical neurons with the Ca2+-indicator dye Oregon Green 488 BAPTA-1 AM (OGB-1) was performed using standard protocols27,28. A 50-μg aliquot of OGB-1 was dissolved in a 20% Pluronic-127 solution (wt/vol) in DMSO and further diluted in standard pipette solution (150 mM NaCl, 2.5 mM KCl, 10 mM HEPES, pH 7.4) to yield a final concentration of 0.5 mM. We guided a micropipette with a resistance of 3 to 5 MΩ to layer 2/3 and applied pressure (10 psi) for 2–4 min to eject the solution from the pipette. Activity of cortical neurons was monitored by imaging the fluorescence changes with a custom-build two-photon microscope based on Ti:Sapphire pulsing laser operating at a wavelength of 800 nm and resonant galvo-mirror system (8 or 12 kHz, GSI) through a 40×, 0.8 numerical aperture (Nikon) water immersion objective. Full-frame images were acquired at 30 or 40 Hz using custom-written software based on LabView (National Instruments). At each focal plane, activity was recorded for more than 5 min. We performed image analysis offline as described previously6,7,8. Briefly, regions of interest (ROIs) were drawn around individual somata, and then relative fluorescence change (ΔF/F) versus time traces were generated for each ROI. Ca2+ transients were identified as changes in ΔF/F that were three times larger than the s.d. of the noise band. Astrocytes were excluded from the analysis based on their selective staining by sulforhodamine 101 and their specific morphology.

Drug applications.

For the local application of lipopolysaccharide to cortical neurons, a glass micropipette (3–5-MΩ resistance) was filled with the endotoxin dissolved in extracellular saline (1 mg ml−1)29. Following a baseline recording period, lipopolysaccharide was pressure injected (0.15 bar) to the cells of interest over a duration of 40–60 s. Furthermore, a group of β1-treated Tg2576 mice (n = 5) received twice-daily intraperitoneal injections of dexamethasone (Sigma) at a dose of 5 mg per kg over 3 d before the in vivo imaging experiments30.

Determination of plaque load in 3D6 experiments.

Plaque burden was assessed by quantitative immunoperoxidase histochemistry on 40-μm free-floating vibratome sections using the anti-Aβ antibody 3D6, as described previously2,31.

Determination of Aβ in forebrain in β1 experiments.

Forebrains were homogenized in nine volumes of ice-cold Tris-buffered saline (pH 7.4) containing Complete protease inhibitor cocktail (Roche Diagnostics) using a Sonifier 450 (Branson). Soluble Aβ was extracted from 50 μl homogenate with 50 μl TBS or 50 μl 2% Triton X-100 (vol/vol) in TBS, followed by ultracentrifugation at 100,000g for 15 min. For the extraction of insoluble amyloid peptides, 50 μl forebrain homogenate was mixed with 117 μl of 100% formic acid, stored for 15 min, and neutralized. The supernatant after centrifugation at 20,300g was diluted and used for analysis. Aβ peptides were determined using the electro-chemiluminescence immuno assay kits based on antibody 6E10 from Meso Scale Discovery. Samples and standards were prepared according to the manufactures protocols.

Cytokine and chemokine quantification in brain.

Cytokine protein levels were measured using a multiplexed particle-based flow cytometric cytokine assay32. Cytokine kits were purchased from Bio-Rad, Millipore and R&D Systems. The procedures closely followed the manufacturer's instructions. The analysis was conducted using a conventional flow cytometer (Guava EasyCyte Plus, Millipore).

Statistical analysis.

We used the smallest number of animals that would be considered appropriate in the field (for example, see refs. 6,7,8,27,28). Statistical analysis was performed with SPSS. P < 0.05 was considered statistically significant.

A Supplementary Methods Checklist is available.