25 November 2003. Are you sometimes tempted to view APP and presenilin merely as raw material and machine for Aβ production, respectively? If so, think again, because that narrow spotlight is opening up to a broader understanding of how these two proteins function and might contribute to Alzheimer’s in their own right, as it were. At the annual meeting of the Society for Neuroscience held earlier this month in New Orleans, Jie Shen of Brigham and Women’s Hospital in Boston led a symposium on synaptic changes in Alzheimer’s disease, a topic that overlaps in large part with new areas of investigation for APP and PS. The event drew a packed auditorium, in part, perhaps, because the original Richard Morris of water maze fame was on hand, as well. (Developed in the 1980s for the study of hippocampal memory, this mouse swimming pool has become de rigueur for any Alzheimer’s mouse lab trying to prove their favorite intervention is meaningful in vivo.) New functions proposed for APP and PS included axonal transport, formation of the synapse, and synaptic transmission and its related signal transduction pathways. One provocative upshot of this new data—briefly debated at the symposium—was that presenilin mutations may cause memory loss and neurodegeneration not just through a toxic gain of function—the widely accepted notion—but in a more complex manner that involves also the loss of separate physiological functions of PS in learning and memory. Here’s a summary of the symposium presentations, plus another talk that tied in tau with axonal transport.
Presenilin: A Memory Molecule?
Shen first reminded the audience that synaptic loss remains the best early correlate of cognitive impairment in AD, and is accompanied by a “dying back” of dendrites and axons, and by aberrant sprouting (for a recent review, see Hashimoto et al., 2003). Of the known familial AD mutations, only 16 are in APP, and they are rare (see APP mutations directory; see also ARF related news story). By contrast, the literature contains about 143 mutations in presenilins 1 and 2 (see PS mutations directory; for the latest additions, see Tedde et al., 2003), and they are far more common. Do they cause AD through a gain of function or a loss of function? Speaking for the former are their dominant inheritance pattern and evidence that they increase Aβ42 levels. At the same time, however, PS mutations are scattered throughout the coding sequence of the two genes, and this is consistent with a partial loss of function, Shen said.
Shen took a genetic approach to understand the normal function of presenilin in the postnatal brain. Since PS 1 and 2 are important for embryonic development, knockout mice die too early for the study of these proteins' roles later in life. Shen and colleagues, therefore, generated a line of neuron-specific conditional PS1 knockout mice, which are normal anatomically, have lower Aβ production, and show a surprisingly subtle spatial memory deficit (see ARF related news story). Next, Shen’s group crossed these conditional PS1 knockouts with APPswe+717e transgenic mice made in Lennart Mucke’s lab and asked whether inactivating PS1 could prevent the amyloid pathology in these mice without side effects. (Besides exploring the normal role of PS, this experiment mimics genetically a γ-secretase inhibitor treatment for AD; see also ARF related New Orleans story.) As expected, these double mutants have neither amyloid pathology nor the related gliosis and inflammation. However, collaborative studies with Alfredo Kirkwood’s group at Johns Hopkins University showed that eliminating PS1 did not rescue the LTP impairment. It did not completely rescue the behavioral deficits of the APP transgenics, either, Shen's and Morris' groups found (see also 295.13). The double mutants temporarily performed better on an associative memory task, but by the age of six months, had fallen behind again, plus they performed poorly in spatial reference tests. “This mixed phenotype highlights the importance of comprehensive testing of AD models using multiple behavioral paradigms,” said Shen.
Up to this point, PS2 could compensate for PS1, potentially muddying the results. To shut this escape hatch, Shen’s lab crossed their conditional PS1 knockout to a PS2 knockout strain from Karen Duff’s group, creating mice that express PS1 until a month after birth and then continue life without any presenilin in their excitatory forebrain neurons. These mice, too, seem fine at two months of age, but go downhill from there, Shen told the audience. At this age, prior to any pathology, the mice show mild deficits in associative and spatial memory, as well as in short-term plasticity and LTP. The latter prompted the researchers to investigate NMDA receptor-mediated responses and found reductions there, as well. What’s more, NMDA receptor and CaMKinase II levels in synaptoneurosome preparations were reduced. Postdoc Carlos Saura presented these data in a slide talk (239.2).
By six months, the mice were severely impaired in spatial learning and memory, failing to learn conditioned fear or the Morris water maze. The mice also had shrunken white matter, and by nine and 16 months (few mice survive to this age), they suffer progressive neurodegeneration, with reductions in synaptophysin levels, dendritic number and complexity, number of cortical neurons, and cortical volume. Gliosis accompanies this process, Shen reported.
What molecular webs might underlie this profound phenotype? Shen’s suspicion fell on the CREB/CREM pathway, in part because prior work has established it as a mediator of synaptic plasticity, hippocampus-dependent learning, and in part because mice mutant in these factors show impaired LTP and neurodegeneration (see Pittenger et al., 2002; Chen et al., 2003). Indeed, at two months, these mice had downregulated a list of genes whose expression depends on CREB, including BDNF, c-fos, and others. Total levels of CREB itself or its phosphorylated version were unchanged, but levels of CREB-binding protein (CBP) were down, suggesting that presenilin is somehow necessary to maintain normal levels of CBP. Tau was hyperphosphorylated and p25 upregulated.
Shen invited the audience to ponder a working model for memory impairment in which problems with presenilin function would, as one of the earliest steps, lead to impaired synaptic plasticity through a downregulation of the CREB pathway. While at first blush these data seem to flatly contradict the amyloid hypothesis, Shen explained that the story is more complex than that. "There is truth to both sides as some PS mutations clearly increase Aβb42 generation. I think that, depending on the downstream targets, PS mutations can lead to the disease through both loss and gain of function," she concluded.
Richard Morris, of the University of Edinburgh, Scotland, noted that finding a suitable behavioral model for AD is a formidable task because the clinical diagnosis of this heterogeneous disease requires memory impairment plus at least one other cognitive disturbance, for example aphasia or deficits in decision-making. Even just the memory component features differences in working, episodic, remote, and semantic memory, Morris explained, with striking deficits in episodic memory early in the disease. How can one model that? By now, numerous mouse models
show behavioral deficits (for example, the Tg2576, App23, CRND8, and SAMP8 strains), most of which are assessed using the water maze and the radial arm maze introduced by Gary Arendash and Dave Morgan at the University of South Florida in Tampa.
In earlier studies of the PDAPP mouse, with Guiquan Chen at the University of Edinburgh and Karen Chen and others at Elan Biopharmaceuticals in south San Francisco, Morris tried to distinguish between the ability to encode, store, and retrieve memories (Chen et al., 2000). To that end, he devised a serial spatial-learning task that involves longitudinal testing of an animal and allows for the teasing apart of the different processes of memory. A key result of this work was that these animals seem to forget what they had learned faster because of a specific defect in memory retrieval. Morris noted that a decline in excitatory transmission might underly this effect.
APP: A Synapse Builder?
Hui Zheng of Baylor College of Medicine in Houston presented data implying a physiological role for APP and some of its family members in synapse formation. She proposed that impairments in this function could perhaps also contribute to the synaptic deficits in AD. Zheng pointed out that Aβ is but one of many APP processing products, all of which are affected by FAD mutations. She said it remains unclear whether the synaptic damage in vivo occurs via Aβ, other APP cleavage products, APP-related signaling pathways, or a combination of these. Zheng added that other APP family members such as APLP1 or 2 are also suspects, given that they are abundantly expressed in neurons and found in dendrites.
Zheng and colleagues have further examined their APP-knockout mouse (Zheng et al., 1995). In addition to gliosis, an LTP defect, and behavioral deficits in the water maze, this mouse also moves around less and has a weak grip. The mice have normal numbers of neurons, but their dendrites were abnormally short. Then, Zheng crossed APLP2-knockout mice, which appear normal, with the APP knockouts and found that the offspring died soon after birth, much like a similar set of mice created in Ulrike Muller’s lab (Heber et al., 2000). A growing number of double knockouts of either APP and APLP1 or 2 now exist, Zheng said, offering a chance to sort out the more subtle synaptic functions of these proteins.
The locomotor defect and weak grip of the mice inspired Zheng to study their neuromuscular junctions as a peripheral model of what might be going wrong with synaptic development in the late embryonic stage. The scientists found that the double knockouts had fewer synaptic vesicles on the presynaptic side, as well as a defect in the proper distribution and positioning of presynaptic vesicles opposite acetylcholine receptors on the postsynaptic membrane. Slightly later, these faulty synapses sprouted neurites excessively, Zheng reported. It’s not clear yet whether the vesicle defect has to do with transport or some other aspect of vesicle release and recycling, but it is clear that the neuromuscular junctions were defective in electrophysiological measures of presynaptic function. This synapse-forming role of APP/APLP2 is the first in-vivo demonstration of a physiological function of the APP family of proteins, Zheng said. Whether something similar happens in central synapses, and whether this could explain the behavioral and LTP deficits in APP-knockout mice, are open questions at this point, she added.
Starving Synapses: Axonal Transport a Common Theme?
If the APP family and the presenilins play important physiological roles at synapses, then surely trouble looms if their delivery dries up. In this sense, blockages of axonal transport could provide another mechanism of AD pathogenesis, and indeed, a number of groups have described axonal transport defects in different neurodegenerative diseases in recent years (see ARF related news story; ARF news; ARF news). At the symposium, Larry Goldstein of University of California, San Diego, set the stage by describing just how enormous a transport burden neurons face. Their “railway” system of tracks (the microtubules) and engines (the kinesin and dynein families of motor proteins) shuttle most of the neuron’s biosynthesis products down the axon and other materials (such as NGF) back up to the nucleus. Axons are narrow and easily jammed by stalled vesicles; distances are immense. Goldstein briefly recapped his lab’s prior work on huntingtin (see ARF related news story) and on Alzheimer’s, including Drosophila studies suggesting that APP serves as an anchor for kinesin (see ARF related news story), which mediates transport of vesicles containing APP, BACE, and γ-secretase to the synaptic terminals (see ARF related news story). This work led to a working model in which APP interacts with kinesin, perhaps with JIP-1b as a scaffold (see Inomata et al., 2003), to transport a cargo vesicle that contains the APP processing machinery. Transport defects could be physical, i.e., plugging of the axon by protein aggregates, and/or could occur in a positive feedback loop whereby stalling of vesicles would increase local processing of APP in the axon. Increased APP or impaired kinesin function could also set off this speculative cycle, as could endocytosis of Aβ, Goldstein said.
Goldstein then presented new experiments in mice that aimed to test the prediction that APP overexpression should poison axonal transport by titrating APP away from kinesin binding. Postdoc Gorazd (Goghy) Stokin first studied the morphology of cholinergic axons in basal forebrain of APPSwe mice made by Richard Bochelt, and found numerous axonal swellings, some as large as 10 microns (see also 445.6). In the electron microscope, these swellings resemble those seen in axons of APP-overexpressing fruit flies and contain numerous vesicles and organelles, including mitochondria. Some look as though they are degenerating. Initial data from an ongoing analysis of human AD tissue hint that it, too, contains similar axonal swellings, Goldstein added. Could these axonal swellings be precursors of the dystrophic neurites that are a hallmark of the AD brain?
Next, Stokin crossed the APP transgenic mice to strains deficient in a kinesin light chain gene to ask whether reducing the amount of kinesin would disrupt axonal transport in mice as it did in flies. When he looked for axonal swellings in cortex, he found none in wild-type mice, some in the APP transgenics, and abundant swellings in the double-transgenics. To ask whether reduced transport increases APP processing, the scientists measured amyloid pathology in the double-transgenic mice and found that reducing kinesin accelerated the age-related increase in Aβ levels and plaque deposition. In preliminary experiments, the scientists also saw the distribution of plaque deposition shift in such a way that the fraction in the entorhinal cortex (which projects through the perforant pathway to the dentate gyrus of the hippocampus and is affected early in AD) appeared to increase relative to deposition in the dentate gyrus (see ARF related news story). While these issues need further study, Goldstein argued that APP processing likely is regulated by kinesin in some way.
And, in turn, kinesin may be regulated by tau. In a separate session in New Orleans, Eva-Maria Mandelkow talked about a new relationship between tau and the movement of APP vesicles down the axon (336.6). She proposed that the kinase cascade MARKK-MARK controls this process by phosphorylating tau and causing it to come off the microtubules. Her presentation revolved around tau’s role in limiting vesicle transport through its competition with kinesin for microtubule binding. Overexpression of tau leads to a redistribution of organelles back to the cell body and thus starves the synapse, Mandelkow explained. Her lab has created conditional tau transgenic mice to study subtle changes early in the disease process, long before hyperphosphorylated tau detaches entirely from microtubuli, causing them to disintegrate. While these mice are still too young to use in studying brain aging, cortical neurons cultured from the mice yielded some clues. These neurons lose their dendrites and axons, and quickly die. In normal neurons, anterograde transport of APP-containing vesicles predominates, but in the tau-overexpressing neurons, anterograde transport slows down and retrograde transport predominates. Interestingly, activating the MARKK-MARK pathway in these cells restored anterograde transport, suggesting that the increased MARK activity observed in degenerating neurons may be a protective response by neurons trying to contain tau. Mandelkow believes that even a small increase in tau may be sufficient to disrupt the flow of traffic to the synapse (Timm et al., 2003).
Stepping back to take a broader view, Goldstein even wonders if different neurodegenerative diseases could reflect divergent outcomes of a common dysfunction such as axonal transport, much like different types of cancer are all outcomes of common problems with cell cycle control. A lot of future work is needed to test this leap, some of it being a search for kinesin SNPs that cause subtle differences in kinesin activity and could, over time, impair axonal transport. You can view abstracts mentioned in this story at the SfN/ScholarOne website.—Gabrielle Strobel.