This is Part 6 of a seven-part series. See also Parts 1, 2, 3, 4, 5, 7.
7 January 2009. At the 8th annual Eibsee Meeting on Cellular Mechanisms of Alzheimer’s Disease, held October 29 to November 1 in Germany, two speakers presented advances in the development of immunotherapies for AD. Despite widely reported hiccups in the clinic, this area of drug discovery remains among the most active in Alzheimer disease translational and preclinical research, and many scientists in the field share the hope that it can eventually succeed. At this conference, Guriqbal Basi of Elan Pharmaceuticals in South San Francisco, California, reported early findings of his group’s efforts to understand, with X-ray crystallography, why one particular antibody can have a therapeutic effect while another doesn’t. Ulrike Mueller of the University of Heidelberg, Germany, introduced a promising new go at crafting a highly immunogenic, safe vaccine.
Basi set up his new data by noting first that the original observation of AD immunotherapy—immunization with Aβ reduces amyloid pathology in mice—has been widely replicated. It has also been extended to other amyloid-related pathologies of AD in various model systems. Besides plaques, markers ranging from neuritic dystrophy, gliosis, vascular amyloid burden, to synaptic loss, deficits in long-term potentiation and cognition all can improve in response to the right antibodies, Basi said. At the same time, the question of precisely what constitutes the “right antibody” has come to the fore as one of the main mysteries to solve. Outwardly similar antibodies can behave quite differently in vivo, and the underlying characteristics of that are murky at best. Basi used comparative X-ray crystallography to drill down into why one antibody has a desired effect while another does not.
The scientists started from the observation that acute (aka passive) immunization with certain antibodies differs intriguingly from chronic (aka active) immunization with Aβ constructs, in that injected antibodies can have surprisingly rapid effects in both mice and slice cultures (Dodart et al., 2002; Shankar et al., 2007). As their behavioral assay to study a series of antibodies, Elan’s colleagues at Wyeth Pharmaceuticals chose contextual fear conditioning, where a mouse first learns an aversive stimulus, then remembers it. In this test, performance decline precedes age- associated increase in Aβ load. Interestingly, performance responds differently to otherwise similar therapeutic antibodies. Consider these three monoclonal antibodies by Elan, which all bind the Aβ3-7 N-terminal epitope that is widely believed to generate effective immune responses: the 12A11 potently reverses the fear conditioning deficit, the 10D5 antibody needs 30 times the dose of the 12A11 to get the same job done, and the 12B4 can’t do it at all (see also ADDF conference report).
What’s behind this difference? The binding affinities of these three antibodies to the same linear Aβ1-10 epitope did not correlate with their ability to reverse the cognitive deficit. What did correlate, however, was their ability to recognize oligomeric species of Aβ. In immunoprecipitation experiments, the most effective antibody pulled down much more Aβ dimers and trimers than monomer and larger aggregates, whereas the other two antibodies did not, Basi reported. The amino acid sequences of the three antibodies’ light and heavy chains differ only by some 5 to 15 percent.
Collaborating with Hadar Feinberg and Bill Weis at Stanford, the Elan scientists obtained crystal structures of the 12A11, the 12B4, the 10D5 and, for good measure, the 3D6 antibody that recognizes plaques and is the basis of the current series of Bapineuzumab (see Phase 3 immunotherapy trials). The scientists expressed these antibodies as recombinant Fab fragments and co-crystallized them with Aβ 1-7. Apologies, dear reader, the pictures and movies themselves have to await formal publication, but here is a brief impressionistic sketch in words: the structures show Aβ1-7 bound to antibody in an extended conformation in a groove atop the antibody. At 1.5 to 2.95 Angstroms, the structures’ resolution is high enough to allow the scientists to define the contact points between the antibody and Aβ residues that make up the binding. When overlaying the structures of the 12A11, the 10D5, and the 12B4 antibodies, low-resolution views betray little difference—the antibodies each appear to cradle Aβ in the same configuration. But zooming in to the contacts reveals that the minor amino acid differences in the antibodies’ sequences contribute to divergent contacts between antibody and antigen in the binding pocket. Most of these differences between the antibodies occur in the complementarity-determining region (CDR) 3 loop, Basi noted. Together, these atom-to-atom differences in the contacts between Aβ and antibody may eventually explain their different in-vivo properties, Basi said.
The group used X-ray structures to draw a different comparison, as well. This one was not between cognitively active and inactive antibodies, but between two antibodies that both recognize amyloid pathology in brain tissue but do so in subtly different ways. The 12A11 tends to bind the dense core of plaques, whereas the 3D6 tends to stain more diffuse plaque in frozen sections from PDAPP mouse brains, Basi said. Their Fab/Aβ structure looked very different from each other. Where the 12A11 captures the peptide in an extended conformation, the 3D6 holds a helical one. Taken together, crystal structures of four separate N-terminal Aβ antibodies illustrate differences in Aβ conformation that may correspond to its deposited forms, as well as subtle differences in antibody-antigen binding contacts that may give clues toward understanding cognitive effects of oligomeric forms of Aβ, Basi said.
These data largely fit with two prior reports of Fab-Aβ crystal structures of antibodies raised by independent investigators (Gardberg et al., 2007; Miles et al., 2008). These investigators also saw Aβ in an extended conformation and they also saw the same antibody CDRs contacting the peptide. A main difference is that these prior antibodies are not part of a therapy development program.
Many questions remain. How the CDR loop that houses most of the different contacts might relate to oligomer properties is unknown, as is whether the oligomer-binding antibody 12A11 can prevent neuronal toxicity or loss. A crystal structure of an antibody grasping an Aβ dimer is on the to-do list, as well, Basi said.
Also at the Eibsee meeting, Ulrike Mueller, now in Heidelberg, presented news from the other side of the immunotherapy spectrum. While Basi showed exquisite, near-atomic details of how a known monoclonal antibody interacts with Aβ, Mueller presented on a polyclonal immune response to a new active vaccine that was highly potent after a single shot.
Known for her sophisticated dissection of APP’s multiple functions in knockout and knockin mice (Ring et al., 2007), Mueller recently teamed up with card-carrying immunologists to join the fray with an AD vaccine of her own. Christian Buchholz at the Paul Ehrlich Institute in Langen, Germany, has for years worked to redesign retrovirus-like particles, from which all replication ability has been removed to render them harmless. Dubbed virus-like particles (VLP), they are derived from mouse leukemia virus (termed retroparticles). VLPs serve as display platforms that present densely stacked arrays of one’s antigen of choice to B cells and T cells (Buchholz et al., 2008). One of his papers, which reported that such an experimental vaccine could break the body’s tolerance to the prion protein (Nikles et al., 2005; Buchholz et al., 2006) caught Mueller’s attention, and the scientists, with Ulrich Kalinke also of the Paul Ehrlich Institute, subsequently collaborated to create a similar vaccine against Aβ. At the conference, Mueller shared results of first mouse studies with it.
The vaccine features retroparticles studded with several thousand copies of Aβ, Mueller said. When injected intravenously into 4-month old APP23 transgenic mice once, this vaccine by 11 months of age had elicited high levels of IgG1 and IgG 2b antibodies, the isotypes thought to avoid unwanted cytotoxic inflammatory reactions. Surprisingly, this single injection induced the same concentration of antibodies in the sera of the mice as did a series of six monthly booster shots. What’s more, the antibody response arose without use of an adjuvant. (Adjuvants have at times raised safety concerns when used in elderly people with multiple co-morbidities.) Tolerance to human Aβ, which can keep antibody titers low, was not a problem with this vaccine. There were no signs of inflammation or cytotoxic T cell activation. The immune response in this study may have been effective and safe because VLP epitopes stimulate T cells directed against viral, not Aβ epitopes, Mueller said.
When the researchers analyzed the brains of the mice, they were surprised to see that the vaccine had reduced both plaque burden and the concentration of soluble pools of Aβ40 and 42. Other vaccines have been shown to reduce the former but, at least temporarily, raise the latter or leave it unchanged. “The sera from immunized mice bind to soluble Aβ on ELISA plates and to plaque Aβ on sections,” Mueller said.
Overall, Mueller said, her data on this new vaccine are consistent with those of the dendrimeric Aβ1-15 vaccine developed by Cynthia Lemere at Harvard Medical School, which is currently being studied in non-human primate models (see ARF related SfN story) and other, similar vaccines. The main differences at this early stage appear to be that Mueller’s vaccine appears to require no adjuvant and lowers soluble and fibrillar Aβ in parallel. The current study injected the retroparticle vaccine into mice before they had developed amyloid pathology, i.e., in a preventive mode. How it would perform in more advanced models that better reflect a typical human trial population is unknown at this point.
This latest candidate AD vaccine comes as part of a broader trend to develop VLP-based immunodrugs in medicine. This trend aims to deploy those particles against disease-related proteins for the treatment of chronic ailments (for reviews, see Jennings and Bachmann, 2008; Jennings and Bachmann 2008). In experimental settings, virus-like particles have been used to raise autoantibodies in models of arthritis, multiple sclerosis, and hypertension. Some are in clinical trials for HIV, Mueller said. Moreover, the approved prophylactic vaccine against cervical cancer is based on VLPs derived from human papilloma virus. The discovery of this virus’s role in cancer garnered half the Nobel Prize in Physiology or Medicine this past fall.
In Alzheimer research, virus-like particles already have left a small footprint. Four years ago, researchers led by David Morgan at the University of South Florida, showed that VLPs can help break tolerance against human Aβ in APP transgenic mice (Li et al., 2004) and Bryce Chackerian went on to show that VLP-based Aβ vaccines generate more desirable B and T cell responses than did some peptide-based vaccines such as Elan’s initial AN1792 (Chackerian et al., 2006, powID=56566.) What’s more, one VLP-based vaccine against Aβ, developed jointly by the Swiss biotechnology company Cytos and the pharma giant Novartis, was reported this past summer at the International Conference on Alzheimer’s Disease (ICAD) in Chicago to have appeared safe in initial human tests performed in Sweden. Called CAD106, this vaccine is currently recruiting for a one-year Phase 2 trial. (Notably, this trial is unusual in admitting people as young as 40 years of age—most AD clinical trials require participants to be at least 50 or 55, shutting out many patients with early-onset AD.)—Gabrielle Strobel.
This is Part 6 of a seven-part series. See also Parts 1, 2, 3, 4, 5, 7.