T. Härd, Department of Molecular Biology, Swedish University of Agricultural Sciences (SLU), Box 590, SE-751 24 Uppsala, Sweden Fax: 0046 18 536971 Tel: 0046 18 4714055 E-mail: firstname.lastname@example.org
The molecular biology underlying protein aggregation and neuronal death in Alzheimer’s disease is not yet completely understood, but small soluble nonamyloid aggregates of the amyloid β-protein (Aβ) have been shown to play a fundamental neurotoxic role. The composition and biological action of such aggregates, known as oligomers and protofibrils, are therefore areas of intense study. However, research is complicated by the multitude of different interconverting aggregates that Aβ can form in vitro and in vivo, and by the inhomogeneity and instability of in vitro preparations. Here we review recent studies in which protein engineering, and in particular disulfide engineering, has been applied to stabilize different Aβ aggregates. For example, several techniques now exist to obtain stable and neurotoxic protofibrillar forms of Aβ, and engineered Aβ dimers, or larger aggregates formed by these, have been shown to specifically induce neuronal damage in a way that mimics Alzheimer’s disease pathology. Disulfide engineering has also revealed structural properties of neurotoxic aggregates, for instance that Aβ in protofibrils and globular oligomers adopts a β-hairpin conformation that is similar to, but topologically distinct from, the conformation of Aβ in mature amyloid fibrils. Protein engineering is therefore a workable strategy to address many of the outstanding questions relating to the structure, interconversion and biological effects of oligomers and protofibrils of Aβ.
The pathogenesis of Alzheimer’s disease (AD) is characterized by extracellular amyloid plaques formed by the amyloid β-protein (Aβ; a.k.a. Aβ peptide) in the central nervous system and intracellular tangles of phosphorylated tau protein . Aβ is a 39- to 43-residue secretase cleavage product of the amyloid precursor protein (APP). The most common forms in human cerebrospinal fluid are a 40-residue fragment (Aβ40, ∼ 90%) and a more aggregation-prone and toxic 42-residue fragment (Aβ42, ∼ 10%). The accumulation of Aβ as amyloid in the brain is caused by an imbalance in the production and clearance of these peptides .
Aβ aggregates also form in the test tube at micromolar or higher concentrations at neutral pH . The aggregation ultimately ends in the formation of insoluble 10-nm-wide fibrils that are the components of amyloid plaques. Research to elucidate the mechanisms of aggregation and fibril formation resulted in the realization that soluble prefibrillar (oligomeric and protofibrillar) species are actually more toxic to neuronal cells than fibrils (reviewed in ). A large number of different soluble aggregates, most of which are toxic, may form on or off the pathway to fibril formation (for recent reviews see [5,6]). It is now generally accepted that soluble Aβ aggregates are neurotoxic and an essential contributor to AD. It has not yet been finally proven that it is the Aβ production and clearance imbalance that causes AD. However, imbalance in the production of Aβ results in an increase in the Aβ42:Aβ40 ratio, which is associated with the formation of more neurotoxic aggregates , and impaired clearance of Aβ in the central nervous system is directly related to the onset of AD .
There is also substantial evidence in vitro [9,10] and in vivo  that points to mechanisms in which oligomeric or protofibrillar Aβ species, alone, are not directly toxic, but where the aggregates induce toxicity by acting as seeds for further aggregation. In such mechanisms, which may be referred to as ‘nucleated polymerization’, it is the rate of aggregation that determines toxicity and neuronal cell death, and therefore they depend on the presence of both aggregates (nuclei) and monomeric Aβ.
It is clear that we need to learn much more about the biochemical and structural properties of different Aβ aggregates. However, a basic difficulty in this research is that it is inherently difficult to control the aggregation status of Aβ in a test tube, as samples of soluble Aβ in water solution are unstable and sometimes very rapidly form insoluble amyloid fibrils. Ways to stabilize different aggregates have been devised, including the use of mixed solvents or detergents [12,13], chemical cross-linking [14,15] and protein engineering [16–21]. The intention of this minireview was to highlight recent work on how protein engineering by the introduction of disulfides can be used to stabilize different Aβ aggregates so that their structure and function may be studied in more homogeneous samples.
Dimers, globulomers, amylospheroids and protofibrils
Of the many soluble Aβ aggregates that have been identified in vitro and in vivo , there are some which are subject to much attention as a result of their neurotoxicity and physiological relevance in the AD brain: dimers and other low-n oligomers, globulomers, amylospheroids (ASPDs) and protofibrils. The interest in low-n Aβ oligomers comes from their direct association with AD: neurotoxic and apparently dimeric Aβ can be isolated from the cortex of the AD brain, whereas dimeric Aβ from the brains of non-AD controls are less neurotoxic [22–24]. These species are observed as dimers, and sometimes as trimers or possibly tetramers, also in SDS/PAGE electrophoresis experiments, indicating that they are resistant to dissociation by low concentrations of SDS. AD brain extracts containing dimeric Aβ are powerful inhibitors of electrophysiological long-term potentiation of memory (LTP) in hippocampal neurons . More importantly, they induce tau hyperphosphorylation and neuritic damage  and thereby provide a direct link between the Aβ and tau pathologies in AD. Hence, it appears that chemical modification of Aβ is an important part of AD, but the precise chemical structure of dimeric Aβ from AD brains is not yet known. It is also not clear if it is the dimer or higher-order assemblies formed from the dimer that is the biologically active unit of Aβ in the different assays.
Incubation of very pure synthetic Aβ42 with SDS or fatty acids induces the formation of globular oligomers – globulomers – with a molecular mass of ∼ 64 kDa [12,19]. Globulomers bind specifically to neuronal dendrites and completely block LTP in hippocampal neurons . Globulomer formation is distinct (off-pathway) from the aggregation process that leads to fibril formation . Globulomers contain a mixture of parallel and antiparallel β-sheet secondary structures, as described in more detail below.
A recent addition to the physiologically relevant soluble Aβ aggregates are the ASPDs , which can be isolated from the brain of patients with AD and dementia with Lewy bodies . These have an apparent mass of 128 kDa, are 7.2 nm wide, as measured by atomic force microscopy (AFM), contain lower-molecular-weight aggregates (apparent trimers) as building blocks and induce apoptosis in neuronal cell lines.
Protofibrillar Aβ was one of the first reported nonfibrillar forms [28,29]. Typical protofibrils are ∼ 5–6 nm wide and up to 200 nm long. They are more flexible than mature fibrils, as judged by their curly appearance under transmission electron microscopy (TEM) and AFM, and they appear to be composed of approximately spherical beads. Annular (pore-like) protofibrils can also be observed  or even be prepared as almost homogeneous samples . Protofibrils are precursors to fibril formation, i.e., they interconvert into fibrils. (This does not exclude that other forms of Aβ oligomers also can act as seeds for fibril formation). Additional motivation for the focus on protofibrils comes from the link between enhanced protofibril formation and early-onset AD  and that levels of high-molecular-weight oligomers are elevated in the cerebrospinal fluid of AD patients .
In protein engineering, it is common to introduce double-cysteine mutations to make a new covalent disulfide bond that links different peptide regions in the tertiary structure of a folded protein or two subunits in the quaternary structure. A number of studies demonstrate that proper disulfide engineering can increase the thermal and chemical stability of folded proteins dramatically [33,34], improve catalytic activity or binding affinity [35,36] or be used to lock a protein into a desired conformation . In practice, the effect of properly placed disulfides in proteins is, of course, the ‘stabilization’ of the folded state. However, thermodynamically speaking, the stabilizing effect arises from a decrease of the entropy of the unfolded state, resulting in a net increase in the free energy of unfolding . Likewise, when stabilizing a particular protein aggregate, such as those described here, it is the free energy of alternative aggregate structures that is made unfavorable. Aggregate stabilization can also be kinetic rather than thermodynamic (i.e. the stabilized aggregate may be made to form quickly and persists owing to high kinetic barriers that slow down the interconversion into more stable aggregates, such as amyloid fibrils). Kinetic stabilization is presumably in action in some of the cases of protofibril stabilization by intermolecular disulfides described below. The introduction of proper disulfides in a known structure is normally straightforward, even though certain geometric restrictions must be met to avoid unfavorable strain .
Intramolecular disulfides: Aβ hairpins in protofibrils and globulomers
Stabilizing Aβ globulomers: the Aβ(L17C/L34C) hairpin
Globulomers are, as mentioned, spherical Aβ oligomers that form in SDS- or fatty acid-containing solutions. Initially, in 0.2% SDS, smaller 16 kDa aggregates, referred to as preglobulomers, form and these associate into the 64 kDa globulomers upon dilution of the SDS content. The solution NMR studies, carried out by Yu et al. , of preglobulomers formed by Aβ42 (with an N-terminal methionine) are the most detailed structural characterization reported so far of any soluble oligomeric form of wild-type Aβ. Several resonances in the NMR spectrum of preglobulomers were resolved and could be assigned. Different isotopic labeling and peptide-mixing schemes then allowed the authors to distinguish between interchain and intrachain NOE connectivities. NOE and chemical shift data were used to determine the structures of dimeric Aβ units within the preglobulomer. In these, residues 18-23 and 28-33 formed an intramolecular two-stranded antiparallel β-sheet connected by a turn involving residues 24-27. The dimer was then formed by intermolecular interactions by the two 34-39 fragments in a two-stranded parallel β-sheet.
Based on this conformation, Yu et al. introduced an L17C/L34C double mutation to stabilize the hairpin with an intramolecular disulfide. The L17C/L34C mutant forms stable and homogenous globulomers that are recognized by a conformation-selective monoclonal antibody towards wild-type globulomers . These stable globulomers can be used for further structural and functional studies.
It is very interesting to note that the intrachain NOE connectivities which are observed in NMR spectra of the preglobulomer  are consistent with the hairpin conformation of Aβ that we previously observed in complex with the ZAβ3 Affibody [40,41] (Fig. 1A). Monomeric Aβ does not adopt a unique conformation in an aqueous solution . Nevertheless, NMR experiments and molecular modeling indeed suggest that Aβ has a propensity to form ‘β-hairpin’ conformations in which the nonpolar fragments 17–21 and 31–36 are extended strands connected by a turn in the region 24–28 [42–45]. The observation of precisely such structures of Aβ in globulomers and in complex with a binding protein therefore adds considerable support to the notion that the β-hairpin is an accessible, and even pervasive, structure of nonfibrillar Aβ.
It is therefore also not inconceivable that the disulfide in the L17C/L34C double mutant results in a structure which is similar to that of the A21C/A34C double mutant (Aβcc; described below), which was engineered based on the structure of Affibody-bound Aβ . This is because the two double mutations are expected to lock the two-stranded antiparallel β-sheet of the hairpin into the same register. The distance between the two β-carbons of L17 and L34 in Aβ in the Affibody complex is, on the other hand, almost 6 Å (i.e. the L17C/L34C disulfide engineered by Yu et al.  might not be completely compatible with the hairpin conformation that is stabilized in Aβcc ).
The β-hairpin structure is topologically very similar to the conformation of Aβ in amyloid fibrils with a ‘cross-β’ structure [46–48]. However, there is a distinct difference in that hydrogen bonds in the hairpin are intramolecular, resulting in antiparallel β-strands, whereas they are intermolecular in fibrils, resulting instead in parallel β-sheets. Other studies also demonstrate the importance of a reverse turn at the center of the 23–28 fragment in aggregates of Aβ. For instance, a lactam bridge linking the side chains of D23 and K28 in Aβ40 increases the rate of amyloid fibril formation by a factor of ∼ 1000 .
Aβ(A21C/A30C) hairpin (Aβcc)
The kinetics of in vitro fibril formation involves a lag time, a nucleation (seeding) event and a runaway fibril-formation reaction. The aggregation is also enhanced by breakage of fibrils to form new seeds [50,51]. A comparison of structural features of monomeric Aβ and the conformation of Aβ in fibrils led us to suggest that the aggregation of Aβ involves oligomeric intermediates composed of β-hairpins that undergo a concerted conformational change, much like closing Venetian blinds, to form fibril seeds  (Fig. 1B). To test the mechanism, we sought ways to allow Aβ to form β-hairpins while restricting a conformational change into the cross-β conformation; if the hypothesis was correct, this should allow the enrichment of stable soluble oligomeric species. Examination of the β-hairpin structure observed in the Affibody complex revealed that the β-carbons of alanine 21 and alanine 30 are located in close proximity to each other on opposing strands and that a disulfide would be accommodated without disrupting any other conformational properties of the hairpin, such as backbone hydrogen bonding (Fig. 1C). The disulfide would, on the other hand, not be compatible with the cross-β structure observed in fibrils. Hence, we created the Aβcc [Aβ(A21C/A30C)] double mutant . Aβcc may potentially also form intermolecular disulfides. However, the production system , in which Aβ (with an N-terminal methionine) is obtained as an Aβ–Affibody complex, allows for purification of monomeric Aβcc with an intramolecular disulfide.
We find that Aβcc indeed readily forms soluble oligomeric and protofibrillar species but not amyloid fibrils, unless the restraining disulfide is broken by reduction. Briefly, Aβcc can form two types of oligomeric species that aggregate further into either protofibrils that are recognized by the mAb158 monoclonal antibody, which is selective for protofibrils of wild-type Aβ , or into amorphous aggregates that are recognized by the (polyclonal) A11 antibody . (There is not much cross-reactivity between the A11 and mAb158 antibodies.) We have named these two aggregation pathways based on secondary structure features of the originating oligomers. Oligomers with an apparent molecular mass of ∼ 100 kDa (as observed in size-exclusion chromatography) that contain antiparallel β-sheet secondary structure (as indicated by CD and infrared spectroscopy) aggregate along the ‘β-sheet pathway’ to form protofibrils that are detectable by mAb158. Smaller oligomers, without regular secondary structure, aggregate along the ‘coil pathway’ to form A11-binding species. (The resulting secondary structure within the A11-binding aggregates is not yet known.) There is also a cross-over option available to smaller (‘coil’) oligomers: they can form β-sheet oligomers and ultimately protofibrils when concentrated to millimolar peptide concentrations.
The 100-kDa β-sheet oligomers and/or protofibrils formed by these on the β-sheet aggregation pathway induce apoptosis in SH-SY5Y (a human-derived cell line that was subcloned three times from a bone marrow biopsy of a metastatic neuroblastoma site of a young woman)  neuroblastoma cell lines . The toxicity is comparable to, or larger than, that of wild-type Aβ oligomers, prepared as described previously . Aβ42cc forms toxic species more readily than Aβ40cc. This is because the longer peptide preferentially aggregates along the β-sheet pathway – oligomers of Aβ40 that have been forced to cross over into the β-sheet pathway in concentrated samples also induce apoptosis.
The fact that Aβcc aggregates via two distinct pathways is consistent with studies of wild-type Aβ aggregation [26,57,58]. The size and secondary structure content of β-sheet oligomers formed by Aβcc are reminiscent of those of globulomers. The β-sheet oligomers also share the non-A11 binding and apoptotic properties of naturally occurring ASPDs, described above [26,27].
Protofibrils are the most stable form of oligomeric Aβcc. TEM images reveal that these are morphologically very similar, if not identical, to protofibrils of wild-type Aβ. They are ∼ 6 nm in diameter and up to 60 nm in length and appear as curved (or flexible) beaded strings in which individual beads are spherical. It is not unlikely that the 6-nm beads correspond to the original β-sheet containing oligomers that formed the protofibrils, but changes in secondary structure content upon fibril formation cannot be completely ruled out. However, Fourier transform infrared (FTIR) spectroscopy of protofibril samples indicates that antiparallel β-sheet structure is present in these. Presumably this originates from the (restricted) hairpin conformation of Aβcc.
As mentioned, samples of soluble Aβ isolated from the brains of AD patients contain small oligomers, primarily dimers, which do not dissociate in SDS/PAGE [22,24]. Small amounts of SDS-resistant oligomers also form in in vitro preparations of wild-type Aβ42. It is interesting to note that Aβcc forms such SDS-resistant dimers and trimers more readily than wild-type Aβ. In particular, Aβ42cc species from the β-sheet oligomers only appear as dimers and trimers (no monomers) in SDS/PAGE . Still, further research is needed to determine if they also share other biochemical and structural properties with the AD dimers.
Intermolecular disulfides: dimeric Aβ in protofibrils, in vascular amyloid and as a consequence of APP processing
The fact that the physiologically active soluble Aβ species in the AD brain appears to be dimeric in SDS/PAGE has inspired several laboratories to investigate the properties of synthetic Aβ dimers [16,18,20,21]. These studies have, for instance, resulted in methods to stabilize protofibrils that are similar to those formed by wild-type Aβ and they have resulted in the detection of vascular and amyloid Aβ dimers in the brains of transgenic mice. (When reading this section it is very important to be aware that it is not proven that dimers themselves are the biologically active and toxic species of Aβ, but that dimer engineering also can facilitate the formation of toxic aggregates of higher order.)
O’Nuallain et al.  used a sterically conservative serine to cysteine mutation to link Aβ40 at position 26. The nonaggregated (Aβ40S26C)2 dimer has no detectable secondary structure and no effect on LTP in mouse hippocampal tissue. However, (Aβ40S26C)2 aggregates into protofibrils that are rich in β-sheet secondary structure and morphologically similar to the protofibrils formed by wild-type Aβ (i.e. they appear as flexible rods with a diameter of 6 nm). These protofibrils are potent inhibitors of LTP in mouse hippocampus. Moreover, they are quite stable and can persist for up to 1 month at 37 °C, before eventually forming amyloid fibrils.
It is very significant to the relevance for AD that (Aβ40S26C)2, presumably in protofibrillar form, also causes tau hyperphosphorylation and neuritic damage in hippocampal neurons . The potency of the synthetic dimer was in this regard found to be lower than that of dimeric Aβ species isolated from AD brains, which have the same physiological effects at a concentration more than 100-fold lower (0.5 nm). However, although the weaker effect of (Aβ40S26C)2 brings it out of the physiological concentration range, it is not very critical with regard to using this synthetic dimer as a model compound for research; rather, it is the specificity that is important. (In physical terms, the lower potency corresponds to a difference in the free energy of interaction in the order of a few kcal·mol−1, which, for instance, may be attributed to only a single missing hydrogen bond somewhere at an interaction surface.)
Nevertheless, it is still potentially very valuable to learn precisely why (Aβ40S26C)2 is less potent than the AD brain extracts. There are three possible explanations, that (a) the disulfide linkage in (Aβ40S26C)2 might not be optimal with regard to the precise structure–function relationship or, in other words, there are better ways to link monomers than by a cysteine at position 26, (b) the physiological action depends not only on the dimerization and/or the formation of protofibrils, but also requires the two extra residues present in the longer and more toxic Aβ42 peptide or (c) the naturally occurring neurotoxic Aβ dimer is chemically modified in some other way that affects function compared with synthetic Aβ– such as in the N-terminally truncated pyroglutamate AβpE3 that was recently shown to account for a major fraction of Aβ in the AD brain .
Aβ dimers linked at the N-terminus or the C-terminus
Yamaguchi et al.  studied two disulfide-linked dimers of Aβ40: one with an alanine-to-cysteine replacement at position 2 at the N-terminus; and one in which the C-terminus was extended with a cysteine preceded with a glycine linker (GGGC). Both peptides were found to rapidly aggregate into stable protofibrils that are rich in β-sheet secondary structure. These results suggest that both N- and C-terminal dimer linkages, as linkage at position 26, are compatible with protofibrillar three-dimensional structures. However, the terminally linked Aβ dimers were unfortunately not characterized with regard to their biological effects and it is therefore not clear whether they are physiologically relevant.
In both the case of (Aβ40S26C)2 and of the N- and C-terminally disulfide-linked Aβ dimers, the rate of protofibril formation is enhanced compared with that of wild-type peptides, whereas the rate of amyloid formation is decreased. The protofibrils formed by these Aβ dimers are therefore kinetically stabilized. Kinetic stabilization of protofibrillar intermediates occurs also with naturally occurring Aβ mutants such as the Arctic E22G mutant , and rapid protofibril formation is an attractive explanation of why this mutant is associated with early-onset AD [17,31].
The AβK28C dimer
Schmechel et al.  characterized a set of Aβ dimers formed by a lysine to cysteine mutation at position 28. This particular site was chosen because the corresponding mutation of human APP was previously observed to yield much higher levels of Aβ in transfected SH-SY5Y neuroblastoma cell lines. The authors compared aggregation properties and the secondary structure content and morphology of aggregates of K28C dimers of Aβ40, Aβ42 and the Aβ(17–40) and Aβ(17–42) fragments, corresponding to so-called p3 peptides produced by α- and γ-secretase cleavage. It was found that the K28C mutation, in combination with the presence of residues 41 and 42 in the Aβ42K28C dimer, significantly increased the β-sheet content of fibrillar aggregates, whereas the K28C dimer, in the context of the Aβ(17–40/42) fragments, inhibited mature fibril formation with retained β-sheet secondary structure in the aggregates. The different K28C aggregates were not characterized with regard to toxicity. However, most interestingly, a polyclonal serum containing antibodies raised against Aβ(17–42)K28C homodimers yielded a strong staining of both vascular and plaque amyloid in the brains of transgenic APP mice. Hence, the results provide a biochemical demonstration of how Aβ dimers may be the species that are initially deposited in vascular amyloid .
AβS8C and AβM35C dimers
Müller-Schiffmann et al.  used computational chemistry to engineer additional Aβ dimers with relevant biological activity. The engineering was, in this case, more sophisticated than that of other intermolecular dimers described above because the sites for cysteine replacements were chosen to be both conservative and to potentially allow for dimer formation early in the APP processing pathway. These criteria led to identification of the AβS8C, AβS26C and AβM35C mutations (AβS26C is the mutation studied earlier in the context of Aβ40, as described in the ‘Aβ40S26C dimer’ section).
The biological effects of these mutants was tested after producing the peptides in vivo from full-length APP in transfected Chinese hamster ovary (CHO) cell lines, which presumably yields Aβ fragments of different lengths (i.e. both Aβ40 and Aβ42). AβS8C produced in this way was found to be homogeneous in the dimeric state when detected by western blots of SDS-polyacrylamide gels, whereas AβM35C also formed higher-order aggregates. Notably, the aggregates remained relatively stable after disulfide reduction. This is analogous to what is observed with Aβcc  and in line with the notion that appropriate disulfide linking of monomers should promote the formation of native-like aggregates.
AβS8C samples showed a significant effect on excitatory postsynaptic currents in cultured cortical neurons. Furthermore, both AβS8C and AβM35C (but not AβS26C) induced small, but significant, reductions of neurite outgrowth in PC12 cells (a rat-derived cell line that is useful as a model system for neuronal differentiation). This effect was not observed with wild-type Aβ samples that did not contain dimers or higher-order oligomers. Hence, in particular AβS8C constitutes an additional model peptide that may be valuable in further studies of the molecular biology of Aβ.
Concluding remarks: what has been achieved?
The work described here demonstrates that stabilization of biologically active soluble aggregates of Aβ may be achieved by protein engineering. The ability to prepare samples of different oligomers and protofibrils of Aβ that are more homogeneous and stable than those of the wild-type peptide will allow us to differentiate between their biological effects, which should contribute to our understanding of the molecular biology of AD. Studies of engineered Aβ aggregates have also resulted in numerous new insights into the structure of Aβ in these aggregates, for instance that a β-hairpin conformation appears to be common and that certain dimer interactions are accessible in protofibrils. A number of issues remain to be resolved, in particular with regard to what extent engineered peptides actually exist in conformations identical to those of native Aβ. However, engineering of stable and homogeneous aggregate preparations may nevertheless facilitate detailed studies of structure–function relationships and potentially lead towards rational drug design. Stable aggregates might be also used to carefully control conformational epitopes when devising AD therapies based on active or passive immunization.