Development and validation of a yeast high-throughput screen for inhibitors of Aβ₄₂ oligomerization

Several aggregated forms of the amyloid β peptide (Aβ), which are generated by proteolytic processing of the amyloid precursor protein (APP), in normal brains and cerebrospinal fluid (CSF) are believed to have a crucial role in the development of Alzheimer’s disease (AD) (Hardy and Higgins, 1992; Selkoe, 1991; Younkin, 1995). Although extracellular amyloid plaques and neurofibrillary tangles formed by insoluble fibrils in brains are hallmarks of AD, recent findings suggest that smaller non-fibrillar oligomeric forms of the Aβ peptide are a more likely cause of AD. Indeed, studies in mice as well as mammalian cell culture showed that detergent-stable Aβ oligomers are potent neurotoxins (Dahlgren et al., 2002; Kayed et al., 2003; Lambert et al., 1998; Lesne et al., 2006; Walsh et al., 2002a). Recently, Aβ dimers in AD brain or CSF have been specifically identified as toxic because they (but not Aβ monomers) induce synaptic dysfunction (Klyubin et al., 2008; Walsh et al., 2002a). In addition, oligomer-specific antibodies can reduce the Aβ-induced toxicity of soluble AD brain extract (Gong et al., 2003; Lambert et al., 2001; Lee et al., 2006).

Small molecules that prevent the formation of Aβ₄₂ (a 42-residue Aβ protein) aggregates that lead to the formation of large plaques had previously been of interest (De Felice and Ferreira, 2002; Estrada and Soto, 2007; Soto et al., 1998). However, evidence for a pathological role of small soluble Aβ oligomers in early AD development led to the idea that inhibiting the formation of Aβ oligomers is a more promising strategy to prevent or treat AD (Klein et al., 2001; Walsh et al., 2002b). Although the relationship between toxic oligomers, large fibrils and plaques is unclear, at least some oligomers seem not to be precursors of large fibrils. Hence, it is possible that large fibrillar aggregates might help prevent toxic oligomers from forming (Chen et al., 2010; Cheng et al., 2007; Glabe, 2005; Harper et al., 1999; Kayed et al., 2003; Necula et al., 2007a). As a result, the ideal drug candidate might inhibit toxic oligomer formation while not inhibiting large fibril aggregation.

Cell-based assays for drug-like molecules that inhibit Aβ₄₂ aggregation are advantageous because toxic compounds are immediately discarded (Bharadwaj et al., 2010; Caine et al., 2007; Kim et al., 2006; Lee et al., 2009; Macreadie et al., 2008). Compounds that inhibit Aβ aggregation have been well studied and some of them also inhibit Aβ oligomerization (Amijee et al., 2009; Amijee and Scopes, 2009; Scherzer-Attali et al., 2010). Such compounds include modified short Aβ peptides, designed to bind to the core region of Aβ₄₂ that is involved in fibrillization, e.g. SEN304 (a methylated pentapeptide of Aβ₄₂). SEN304 has been reported to inhibit secretion of toxic sodium dodecyl sulfate (SDS)-stable oligomers in 7PA2 cells (Kokkoni et al., 2006). Other compounds that are known to inhibit Aβ₄₂ from forming toxic oligomers and that also have a therapeutic effect in AD animal models are: curcumin (Yang et al., 2005), RS-0406 (hydroxyanaline) (Nakagami et al., 2002; O’Hare et al., 2010; Walsh et al., 2005), SEN1269 (hydroxyanaline derivative; Senexis), scyllo-inositol (AZD-103) (McLaurin et al., 2000; McLaurin et al., 2006; Townsend et al., 2006), PBT1 (Clioquinol, 8-hydroxyquinolin) (Hsiao et al., 1996) and PBT2 (a copper/zinc ionophore, 8-hydroxyquinolin) (Adlard et al., 2008; Faux et al., 2010). Both scyllo-inositol (Transition Therapeutics and Elan) and PBT2 (Prana Biotechnology) are currently in clinical trials. Recent work points to compounds that bind to Aβ₄₂ as possible inhibitors of Aβ₄₂ toxicity (Alavez et al., 2011; Chen et al., 2010; Scherzer-Attali et al., 2010).

The inhibition of Aβ₄₂ oligomer formation is often assayed using pure synthetic Aβ₄₂ peptide reconstituted under conditions that favor Aβ₄₂ oligomerization over fibrillization. Prevention of oligomer formation is characterized by using Thioflavin T (ThT) and/or antibodies specific for oligomers (Chang et al., 2003; Chromy et al., 2003; Hamaguchi et al., 2009; Necula et al., 2007b; Yang et al., 2005), or by using mammalian cells that overexpress and secrete human Aβ₄₂ that forms oligomers in conditioned medium (O’Hare et al., 2010; Walsh et al., 2002a; Walsh et al., 2005).

Here, we develop a yeast in vivo assay that is specific for assessing the inhibition of Aβ₄₂ oligomerization activity. Previously, we reported a yeast Aβ oligomerization model in which the formation of SDS-stable low-n oligomers, including dimers, trimers and tetramers, of an Aβ₄₂-fusion protein is easily detected (Bagriantsev and Liebman, 2006). Briefly, an Aβ₄₂ fusion to the essential functional domain (MRF) of the translational release factor, Sup35 (Derkatch et al., 1996; Ter-Avanesyan et al., 1993), provides an Aβ₄₂ aggregation-specific probe tied to a functional Sup35 readout. The Aβ-MRF construct was overexpressed in cells lacking chromosomal SUP35. The MRF expressed in the fusion (Aβ-MRF) was shown to have reduced translational termination factor activity by a simple growth test (left panel in Fig. 1). This seems to be due to aggregation of the fusion protein into SDS-stable low-n oligomers reminiscent of the detergent-stable oligomers found in human AD brains (Gong et al., 2003; Klyubin et al., 2008; Shankar et al., 2008) (left panel in Fig. 2A). Indeed, analogous fusions made with an Aβ₄₂ aggregation-deficient mutation, F19T, F20T and I31P (Hilbich et al., 1992; Morimoto et al., 2004; Williams et al., 2004), called Aβm2-MRF, retained almost complete translation termination factor activity and formed almost no oligomers (left panel in Fig. 2A) (Bagriantsev and Liebman, 2006).

Using this system, we developed and validated a high-throughput assay to screen small chemicals for an effect on Aβ₄₂ oligomer formation. The idea for the primary high-throughput screen (HTS) was to identify small drug-like molecules that restore translation termination activity to the Aβ-MRF strains by inhibiting formation of Aβ-MRF small oligomers, seen as inhibition of growth in media lacking adenine (–Ade) (right panel in Fig. 1). Here we report two drug-like compounds that were selected by the growth assay in a pilot screen. Through biochemical analyses, these compounds were further verified to indeed inhibit Aβ-MRF oligomer formation in vivo. They also inhibited bona fide Aβ₄₂ peptide polymerization and fibril formation in vitro. Thus, the yeast HTS growth assay is a promising strategy to screen for inhibitors of Aβ₄₂ oligomerization.

When Aβ-MRF in yeast lysates was treated with 1% SDS at room temperature and resolved by SDS-PAGE, most of the Aβ-MRF was found as low-n oligomers (dimer, trimer, tetramer) and monomers (expected molecular mass 73.7 kDa, but runs as 95 kDa), although a portion of the Aβ-MRF was stuck in the well or appeared as a smear above the oligomers (left panel in Fig. 2A). To investigate whether the SDS-stable oligomers seen were present in native yeast cells or were derived from larger in vivo aggregates that were broken down by the SDS treatment, we examined lysates that were not treated with SDS before loading on the gel (right panel in Fig. 2A). These untreated samples showed similar oligomers and larger material, and this larger material did not break down into oligomers or monomers when they were excised from the gel, treated with 1% SDS at room temperature and run again on a second SDS-PAGE (Fig. 2B). However all forms of the Aβ-MRF were completely converted into monomers by boiling in 2% SDS and β-mercaptoethanol (BME; lower panel in Fig. 2A).

We also examined lysates with non-denaturing PAGE (left panel in Fig. 2C) and size exclusion chromatography (SEC; supplementary material Fig. S1), in which the Aβ-MRF was detected as a complex with 10–15 monomers and monomers and low-n oligomers were not detected. The complex was broken into monomers and low-n oligomers when the native lysate was treated with BME before loading on the gel (middle panel in Fig. 2C). One possibility is that Aβ-MRF monomers and low-n oligomers are cross-linked into, or trapped in, bigger aggregates in the lysate via disulfide bonds. These large aggregates in the lysate might be disaggregated by the reducing agent, BME, possibly returning the Aβ-MRF to the intracellular monomer and oligomer states. Indeed, there are seven cysteine residues in the fusion [five in the release factor (RF) and two in the linkers between Aβ and MRF].

Before examining the effect of compounds in our assay, we made our yeast strain more permeable to chemicals by deleting the erg6 gene (Dunstan et al., 2002). The altered sterol composition of cell membranes in the erg6 mutants causes enhanced chemical uptake (Sharma, 2006; Welihinda et al., 1994), and we verified that an erg6 deletion caused our yeast strain to become more sensitive to cycloheximide (supplementary material Fig. S2).

We tested whether known anti-aggregation compounds, SEN1269, SEN304, RS-0406 and scyllo-inositol, inhibit formation of Aβ-MRF oligomers in our assay by looking for restored Aβ-MRF translation termination activity detected as growth inhibition in media lacking adenine (–Ade) in the presence of 100 μM of the compounds. SEN1269 had no effect on growth inhibition in –Ade media at the various concentrations tested (data not shown). Because SEN304 caused toxicity in +Ade media, we cannot determine whether it restored Aβ-MRF translation termination activity in –Ade media (Fig. 3A). Curcumin and RS-0406 had no significant inhibitory effects on Aβ-MRF oligomerization (Fig. 3B). Scyllo-inositol, however, had a clear effect in our system: growth in –Ade media was inhibited to 50% of the DMSO control without causing toxicity in +Ade media (Fig. 3A). Furthermore, western blot analyses indicate that scyllo-inositol decreased oligomer formation 51.5±7% of that seen with DMSO alone, based on the ratio of oligomers to monomer (Fig. 3B). The recognition of scyllo-inositol as an anti-Aβ₄₂-oligomer compound by the yeast system validates it as a promising assay for non-toxic anti-oligomeric compounds. The identification and characteristics of AO-11, which has the most dramatic effects of all the chemicals in Fig. 3, is described below.

We used the growth assay described above to screen for small compounds that block Aβ oligomerization, by selecting for restored Aβ-MRF translation termination activity. Prior to the primary screen, HTS parameters (e.g. shaking level, time of incubation, size of liquid sample and prevention of evaporation) were optimized. The sensitivity, reproducibility and stability of the assay was assessed using a library of pre-fractionated marine bacterial extracts (a gift from William Fenical, Scripps Oceanography, La Jolla, CA) (supplementary material Fig. S3). Positive and negative controls were cells expressing, respectively, the aggregation-deficient mutant Aβm2-MRF or Aβ-MRF.

On the basis of the experimental settings and HTS protocol, a screen with 12,800 compounds was conducted using a sub-library of the DIVER and CNS sets (ChemBridge, San Diego, CA). Each assay was carried out in duplicate at a single compound concentration (22 μM). We had an initial hit rate of 0.23%: 29 out of 12,800 compounds caused decreased growth in –Ade media without a significant growth decrease in +Ade media in duplicate well plates (Fig. 4).

We focused on 17 of the 29 compounds that fell into nine structural classes and one miscellaneous group based on their size, shape and the distribution of functional groups (Table 1). These 17 compounds were purchased from the vendor at a purity of 95% and retested in triplicate at the same concentration (22 μM) and conditions as the primary screen. The quality of the test, based on Z-factor statistics (Zhang et al., 1999), confirmed the assay quality to be excellent, reflective of values of >0.8 for both assays calculated from controls. Ten of the 17 compounds inhibited more than 50% growth in –Ade media compared with growth in the presence of DMSO alone. However, six of these compounds were rejected because they also inhibited growth in +Ade media (Table 1 and Fig. 5).

Finally, to eliminate compounds that inhibit growth in –Ade media because they reduce translational read-through of stop codons directly without affecting Aβ-MRF oligomerization, each compound was also tested in cells carrying a missense mutation in SUP35 (Bradley et al., 2003), which impairs translational termination activity, causing the ade1-14 premature stop codon to be read through. Two of the remaining four compounds were eliminated because they reduced this read-through (Table 1 and Fig. 5), leaving compounds AO-11 and AO-15 (respectively shown as 11 and 15 in Table 1 and Fig. 5) to be tested further.

Effects of AO-11 and AO-15 in a dose-dependent assay of Aβ-MRF-associated translational misreading were determined by serially diluting the compounds in DMSO to varying concentrations (0–50 μM) in a 96-well plate format. They showed no or mild growth inhibition in +Ade media, but growth was significantly inhibited in –Ade media as the concentration was increased, suggesting that Aβ oligomer formation was suppressed in a dose-dependent manner (Fig. 6). Estimates for EC₅₀ (the concentration of a compound at which 50% of its maximal effect is observed in vivo) obtained using the Enzyme Kinetics module of SigmaPlot (v. 9.01 Systat) (Ratia et al., 2008) were 12.8 and 6.7 μM for AO-11 and AO-15, respectively. The stronger activity of AO-15 represented by the smaller EC₅₀ value was partially caused by cytotoxicity of the compound at higher concentrations, whereas AO-11 had no apparent cytotoxicity (Table 1 and Fig. 6).

To directly test whether the restored translational termination factor activity associated with these compounds is correlated with a decrease in the presence of SDS-stable Aβ-MRF oligomers, lysates treated and grown with AO-11 or AO-15 at various concentrations were analyzed (Fig. 7). To determine the level of SDS-stable small oligomers, lysates were treated with 1% SDS for 7 minutes at room temperature, subjected to immunoblot analysis and probed with antibodies against the RF domain of Sup35p (BE4; developed in our laboratory). In cells grown with DMSO only, there were a lot of SDS-stable Aβ-MRF oligomers and much fewer monomers. In the presence of higher concentrations of the compounds, the levels of oligomers decreased and concomitantly the level of monomeric Aβ-MRF increased. AO-11, at 100 μM, decreased oligomer formation about 50% more than the known anti-aggregation compound scyllo-inositol (Fig. 3B). The EC₅₀ values for AO-11 and AO-15 are 32.4 and 45.2 μM, respectively. The 2.5- (AO-11) or 6-fold (AO-15) higher EC₅₀ values seen here compared with the EC₅₀ values for growth inhibition in –Ade media indicate that the growth assays were highly sensitive to the gain of monomers at the expense of oligomers.

The structure of these two new hits is shown in Fig. 8. They both have properties desirable for a CNS drug (Congreve et al., 2003; Rishton, 2003): six or fewer H-bond acceptors (actually three and two); three or fewer H-bond donors (actually zero and one); molecular weight <400 (actually <300); CLogP under 5 (actually 2 and 3) and a topological polar surface area of <75 (actually 37 and 62). AO-11 was previously patented for use for drug treatment of Duchenne muscular dystrophy [CA2685540 (A1)], whereas there is no previous patent or literature on AO-15.

To further study the activity of the selected inhibitors, electron microscopy (EM) and western blotting were carried out on samples of soluble Aβ₄₂ peptide prepared in the presence and absence of AO-11 and AO-15 under defined oligomer-forming conditions (Stine et al., 2003) with slight modifications (see Methods). We tested effects of the inhibitors at 50 μM, when more than 60% of oligomer formation was inhibited in vivo. Soluble recombinant Aβ₄₂ peptide was prepared and subjected to polymerization in the presence of DMSO, AO-11 or AO-15 for 24 hours at room temperature without shaking. Aliquots of duplicate samples were taken and examined by EM (Fig. 9A) and SDS-PAGE (Fig. 9B).

In the presence of DMSO (1% v/v, the concentration in which the inhibitors were dissolved), most of the peptides formed large fibrillar networks composed of uniform long fibrils by 24 hours. A clear change in morphology with a reduction in the fibril length and the absence of large fibrillar networks throughout the EM was detected when Aβ₄₂ peptides were incubated with AO-11 or AO-15. Particularly with AO-11, which had a stronger effect than AO-15 in vivo, fibrils are difficult to detect and only amorphous precipitates with bumps or nodules were seen. The reduced numbers of fibrils observed by EM correlated with a decrease in SDS-stable large fibrils detected by anti-Aβ antibody (6E10) on western blots of SDS-PAGE performed on parallel samples. These results suggest that AO-11 and AO-15 efficiently inhibited the formation of Aβ_1–42 SDS-stable aggregates in vitro.

With accumulating reports pointing to small SDS-stable Aβ₄₂ oligomers as a major early culprit in AD pathology, inhibiting their formation has become an attractive therapeutic target. Previous generations of inhibitors of Aβ aggregation were often selected for their ability to inhibit the formation of large Aβ fibrils rather than toxic oligomers. Subsequent attempts to characterize the effects of these compounds on oligomer formation have faced challenges partly due to the myriad of possible in vitro Aβ₄₂ assemblies and difficulty in monitoring and characterizing oligomer formation (Bitan et al., 2005; Hepler et al., 2006). Furthermore, because these primary assays were conducted in vitro, many of these compounds were toxic when tested in vivo (Liu and Schubert, 2006).

Our screen is designed to specifically select against Aβ₄₂ oligomer formation. Indeed, scyllo-inositol, which is known to inhibit Aβ₄₂ from forming toxic oligomers, also inhibited oligomer formation in our assay (Fig. 3). Some small molecules that inhibit Aβ₄₂ oligomer formation might also inhibit large fibril formation. Our ultimate goal is to uncover molecules that affect oligomer formation without inhibiting fibril formation. This specificity might be crucial for success in the clinic, because fibril formation might actually be beneficial (Chen et al., 2010; Cheng et al., 2007; Necula et al., 2007a; Treusch et al., 2009).

We found that, in living yeast, Aβ-MRF aggregated into small SDS-stable low-n oligomers such as dimers, trimers and tetramers that look remarkably like the neurotoxic Aβ₄₂ oligomers secreted from CHO cells (7PA2) expressing the familial V717F Alzheimer’s-disease-causing APP (Walsh et al., 2002a). However, because yeast lysates were treated with 1% SDS before loading on acrylamide gels, the oligomers seen could have been derived from larger in vivo aggregates that were broken down by the room-temperature 1% SDS treatment. To investigate this possibility we omitted the SDS sample treatment and characterized the lysate in gels lacking detergent except for 0.1% of SDS in the running buffer. Interestingly, these untreated samples showed similar oligomers, although some larger material was also stuck in the well (Fig. 2A). The material in the well when cut out and treated with 1% SDS at room temperature did not break into oligomers (Fig. 2B) but, following boiling in 2% SDS, was converted into monomers. We next analyzed Aβ-MRF in lysates on non-denaturing gels in the absence of any detergent. Here, although they ran as 10–15mer sized oligomers, treatment of these lysates with a reducing agent released the Aβ-MRF into low-n oligomers (Fig. 2C). These results suggest that Aβ-MRF exists as low-n oligomers in yeast cells, in which it is likely to be under reduced conditions.

In this study, we used a robot to perform our yeast Aβ₄₂ oligomerization HTS campaign. Of 12,800 compounds tested, four passed this pilot proof-of-principle primary screen. We eliminated two of the four hits in a secondary screen for false positives that affected the suppression efficiency of a sup35 mutant (Bradley et al., 2003). The remaining two compounds (AO-11 and AO-15) both significantly decreased Aβ-MRF oligomer formation in vivo and affected Aβ₄₂ aggregation in vitro. Although AO-11 and AO-15 seem to affect large aggregate formation as well as oligomerization, our assay should allow the detection of compounds that specifically inhibit oligomerization and not large aggregate formation.

Assays were performed in L2333, a mutant derivative of 74-D694 (MATa ade1-14 ura3-52 leu2-3,112 trp1-289 his3-200) (Chernoff et al., 1993) with a G1256A change in SUP35 (Bradley et al., 2003), and in Aβ-MRF and Aβm2-MRF yeast also derived from 74D-694 but with a sup35::LEU2 disruption (Nakayashiki et al., 2001) and respectively containing p1364 (pRS313, CEN, Ura3, CUP1:: Aβ-MRF) or p1541 (pRS313, CEN, Ura3, CUP1:: Aβ^F19,20T/I31P-MRF) (Bagriantsev and Liebman, 2006). ERG6 of the ergosterol biosynthetic pathway was deleted (Longtine et al., 1998) in these strains to enhance permeability to small compounds.

Media, cultivation and transformation procedures were standard (Sherman, 2002). Strains were grown in complex medium (YPD), or in synthetic media lacking (–Ade), or including (+Ade) adenine. Expression of Aβ-MRF was driven by the copper-inducible CUP1 promoter with 50 μM CuSO₄.

Duplicate assays in clear flat-bottom 384-well plates (ScreenMates) in 90 μl used a Tecan Freedom EVO 200 robot (Ratia et al., 2008). Wells were filled in order with 45 μl of assay medium, 0.2 μl of library compounds (10 mM in DMSO) and 45 μl of assay medium inoculated with 1×10⁵ or 1×10⁴ cells/well for the –Ade or +Ade assays, respectively, of the assay strain grown without Aβ-MRF overexpression. Each plate contained 32 positive (Aβm2-MRF) and 32 negative (Aβ-MRF) controls with 0.2 μl of DMSO only. Plates sealed half way with parafilm were incubated at room temperature with shaking (900 rpm) until the cultures reached stationary phase (5 days for –Ade; 4 days for +Ade). OD₆₀₀ determined growth.

EC₅₀ values were estimated by fitting data to v_i=v_o/(1 + [I]/EC₅₀) with the SigmaPlot (v. 9.01 Systat) Enzyme Kinetics module, where v_i and v_o are, respectively, the reaction rates in the presence and absence of inhibitor, and [I] is the inhibitor concentration (Ratia et al., 2008).

Preparation of lysates and western blot analysis of Aβ-MRF oligomers were performed as described previously (Bagriantsev and Liebman, 2006). For SDS-PAGE, lysates of late logarithmic cells grown at 30°C in 5 ml +Ade with 50 μM CuSO₄ and DMSO or a compound were treated with 1% SDS for 7 minutes at room temperature, resolved on 7.5% Tris-Glycine gels (0% SDS, Bio-Rad) in 0.1% SDS containing Laemmli running buffer and transferred to a PVDF membrane (Bio-Rad) in Laemmli buffer with 0.1% SDS and 15% methanol.

For non-denaturing PAGE, lysates in native sample buffer (Invitrogen) with 0.5% digitonin, 10 mM PMSF, and yeast anti-protease cocktail (Sigma) 1:100 were resolved on a 3–12% Blue native Novex Bis-Tris gel (Invitrogen).

Immunodetection used monoclonal antibodies against the RF domain of Sup35p (BE4; developed by our laboratory). Signals were detected using a Western-Star chemiluminescence development kit (Applied Biosystems) and quantified using AlphaEaseFC imaging software of Alpha Imager 2200 (Alpha Innotech).

Polymerization (Stine et al., 2003) was at room temperature with mild shaking for 24 hours followed by sonication after diluting 5 mM Aβ₄₂ made from lyophilized recombinant Aβ₄₂ peptide (HFIP form, Rpeptide) dissolved in anhydrous DMSO (Sigma) into 40 μM in 10 mM Tris (pH 7.4) in the presence of AO-11, AO-15 or control DMSO. For immunoblotting, 10 μg of polymerized Aβ₄₂ was treated with 1% SDS, resolved by SDS electrophoresis in a 4–20% Tris-Glycine gel and detected by anti-Aβ antibodies (6E10, Signet Laboratories).

Samples adsorbed onto 200 hex-mesh formvar/carbon-coated copper grids and stained with 4% (w/v) uranyl acetate were viewed using a JEOL 1200EX transmission electron microscope.

Total lysate of 200 μl (0.5 mg total yeast protein in native sample buffer) was loaded onto on equilibrated analytical Superose 12 10/300 GL column (GE Healthcare) connected to a liquid chromatography system (Bio-Rad) and eluted with yeast lysis buffer (Bagriantsev and Liebman, 2006) lacking glycerol. Samples from each fraction (20 μl) were spotted onto PVDF membrane that was pre-wet with transfer buffer (25 mM Bicine, 25 mM Bis-Tris, 1 mM EDTA, pH 7.2). The membrane was washed once in TBS (20 mM Tris, 140 mM NaCl, pH 7.6) with 0.1% Tween-20, blocked and immunoblotted with anti-Sup35-RF antibodies.

This work was supported by grants to S.W.L. from the Alzheimer’s Association (IIRG-06-25468 and IIRG-10-173736) and the NIH (R21 AG02881). The authors are grateful to Dennis Selkoe (Harvard Medical School, Boston, MA) for kindly providing RS-0406 and to David Scopes (Senexis Limited, UK) for kindly providing SEN304 and SEN1269. We also thank Jack Gibbons (Electron Microscopy Facility, UIC) for performing TEM analysis. We are also grateful to Sviatoslav Bagriantsev, a former member of S.W.L.’s research group, for giving ideas and suggestions.