A Direct Interaction between Transforming Growth Factor (TGF)-βs and Amyloid-β Protein Affects Fibrillogenesis in a TGF-βReceptor-independent Manner*

Transforming growth factor-β (TGF-β) receptor-mediated signaling has been proposed to mediate both the beneficial and deleterious roles for this cytokine in amyloid-β protein (Aβ) function. In order to assess receptor dependence of these events, we used PC12 cell cultures, which are devoid of TGF-β receptors. Surprisingly, TGF-β potentiated the neurotoxic effects of the 40-residue Aβ peptide, Aβ-(1–40), in this model suggesting that there may be a direct, receptor-independent interaction between TGF-β and Aβ-(1–40). Surface plasmon resonance confirmed that TGF-β binds with high affinity directly to Aβ-(1–40) and electron microscopy revealed that TGF-β enhances Aβ-(1–40) oligomerization. Immunohistochemical examination of mouse brain revealed that hippocampal CA1 and dentate gyrus, two regions classically associated with Aβ-mediated pathology, lack TGF-β Type I receptor immunoreactivity, thus indicating that TGF-β receptor-mediated signaling would not be favored in these regions. Our observations not only provide for a unique, receptor-independent mechanism of action for TGF-β, but also help to reconcile the literature interpreting the role of TGF-β in Aβ function. These data support a critical etiological role for this mechanism in neuropathological amyloidoses.

The 39 -43-mer amyloid-␤ (A␤) 1 peptide is derived from the membrane-bound amyloid precursor protein (APP) as an aberrant cleavage product (1). Transgenic APP mouse models exhibit age-related extracellular amyloid deposits (plaques) and neurodegeneration as well as cerebral amyloid angiopathy (CAA) comparable to that found in human Alzheimer's disease (AD) brain (2,3). These same models respond to both active and passive immunization against A␤ as evidenced by the reduction in levels of A␤, the prevention and/or clearance of amyloid plaques, and the improvement in cognitive behavior (4). However, an effective preparation free of significant side effects in humans is still awaited. Indeed, clinical trials involving A␤ vaccination have been discontinued following the development of inflammation in patients brains (5). Cerebral microhemmorhaging has also been observed in similarly immunized mice (6). Although it has been suggested that antibodies capable of recognizing other A␤ epitopes or conformations should be screened (6), perhaps a closer examination of modulators of A␤ fibrillogenesis may reveal a target better suited for immunotherapy.
Among the modulators proposed to date, which include apolipoprotein E, cholesterol, and ␣2-macroglobulin, we were particularly interested in the cytokine transforming growth factor-␤ (TGF-␤). The TGF-␤1 isoform was recently implicated as a co-factor for amyloid deposition with the observation that cerebrovascular amyloid deposits, which are strikingly similar to those seen in patients with AD and CAA, and the ensuing neuropathological development are accelerated in bigenic mice overexpressing both TGF-␤1 and human APP (hAPP), relative to hAPP transgenic mice controls (7). These authors suggested that cerebrovascular amyloid deposition might reflect TGF-␤ receptor-mediated induction of extracellular matrix deposition. A␤ binding proteins within these extracellular matrix components could enhance the formation and/or stability of A␤ fibrils (8,9). A link between TGF-␤ and A␤ deposition was already being considered following the localization of TGF-␤ immunoreactivity to senile plaques and neurofibrillary tangle-bearing neurons in AD patient brain (7,10). TGF-␤s had also been shown to enhance the formation of amyloid deposits in rats when co-injected intracerebroventricularly with A␤-(1-40) (11) and to increase the number of A␤ plaque-like deposits in hippocampal slice culture in a subfield-selective manner (12). Finally, APP production and A␤-(1-40)/A␤-(1-42) processing were promoted by TGF-␤1 in transgenic mice (13).
In contrast to its role as a risk factor, TGF-␤1 facilitates A␤ clearance and plaque burden reduction following activation of parenchymal glial cells in TGF-␤1/hAPP bigenic mouse brain (14). TGF-␤s have also been shown to protect neuronal cell cultures against A␤-mediated insult (15)(16)(17) as a direct consequence of TGF-␤ Type II receptor activation (17).
These apparently opposing actions of TGF-␤ in the brain illustrate the multifunctionality of this cytokine and the dependence of its effects on the specific cellular context in which it is expressed. Additional investigation is needed to characterize the cellular and molecular processes that underlie the ef-fects of TGF-␤ during A␤-mediated pathology. To date, the actions of TGF-␤ have centered on receptor-mediated events. In this report, we demonstrate a direct interaction between TGF-␤ and A␤ that promotes A␤ fibrillogenesis and neurotoxicity. Our data support a role for a unique, receptor-independent mode of action for TGF-␤ and define a new molecular point of intervention for inhibiting A␤ fibrillogenesis.
Circular Dichroism Spectroscopy (CD)-A␤-  was prepared by dissolution into 50 mM phosphate buffer (pH 7.0) to yield a final concentration of 1 mg/ml. Samples were incubated at ambient temperature (22°C) without stirring and CD measurements (at 22°C) were performed periodically using a 0.1 cm pathlength quartz cell (Hellma, Forest Hills, NY) and an Aviv Model 62A DS spectropolarimeter (Aviv Associates, Lakewood, NJ). The scan rate was 1 nm/sec at a bandwidth of 1 nm. Three independent sets of experiments, each comprised of triplicate scans performed from 250 -198 nm, were done. The buffer spectrum was subtracted from the scans and the resulting functions were smoothed. Data could not be acquired at wavelengths lower than 198 nm due to saturation of the photomultiplier. However, this range is sufficient to accurately evaluate the secondary structure state of the samples (18). Protein concentrations were determined a posteriori by quantitative amino acid analysis, thus enabling accurate calculation of molar ellipticities (⌰).
Morphological Characterization of A␤-(1-40) Fibrils-A␤-  was incubated at a concentration of 1 mg/ml (ϳ235 M) in 50 mM phosphate buffer (pH 7.0) either alone or in combination with 5 g/ml (ϳ200 nM) of either TGF-␤1 or TGF-␤2 for 48 h at room temperature without agitation. A small amount of specimen was placed onto 200-mesh Formvar-coated grids, blotted, and then air-dried. The specimen was negatively stained with 1% (w/v) potassium phosphotungstic acid (pH 7.0) prior to examination with a JEM-2000FX electron microscope (JEOL, Ltd., Tokyo, Japan) using an accelerating voltage of 80 kV.
Immunolocalization of TGF-␤ Receptors in Mouse Hippocampus-Mice (C3H/C57Bl; Charles River Laboratories) were terminated by cervical dislocation, and the brains were quickly removed, frozen in liquid nitrogen and kept at Ϫ80°C until use. Animal maintenance and manipulation was performed according to the recommendations of our ethical committees (Biotechnology Research Institute-NRC). Affinitypurified polyclonal antibodies (Santa Cruz Biotechnology, Inc., Santa Cruz, CA) were used for localization of TGF-␤ RI (V-22, 2 g/ml), RII (C-16, 1 g/ml), and RIII (C-20, 2 g/ml) receptors. Antibody specificity was confirmed using biosensor and Western blot analyses. Serial sagittal sections (8-m thick) were incubated with the primary antibody in phosphate-buffered saline/1% bovine serum albumin in a moist chamber overnight at 4°C. The specificity of the immunoreactions was tested by exclusion of the primary antibody as well as by competition, i.e. preblocking the primary antibody with the immunizing peptide (2-5 M). Following several rinses in phosphate-buffered saline, the sections were incubated for 1 h at room temperature with horseradish peroxidase-conjugated secondary antibody. Antigen expression was visualized (0.3% diaminobenzidine, 0.025% H 2 O 2 , 2-10 min) by light microscopy. Cell soma were counterstained with methyl green. This series of experiments was repeated 4 -6 times using tissue obtained from both male and female mice ranging in age from 7 to 10 months.
Statistics-Data were analyzed by one-or two-way analysis of variance using p Ͻ 0.05 as the criterion for significance. Post-hoc analysis relied on Dunnet's Multiple Comparisons Test.

A␤-(1-40)-mediated Neurotoxicity Is Potentiated by TGF-␤ Isoforms in a Receptor-independent
Manner-To examine the receptor-dependence of TGF-␤ on A␤-(1-40)-mediated neurotoxicity, we chose to use cells that do not express TGF-␤ receptors (i.e. PC12 cells) and comparing these to cells that express all three TGF-␤ receptor types (i.e. SH-SY5Y) (Ref. 20 and confirmed in our laboratory). Using PC12 cells eliminates the possibility of any event being obscured by TGF-␤ RII receptormediated neuroprotection (10,17). We first confirmed the conformation state of freshly dissolved A␤-(1-40) using circular dichroism. A␤ underwent a well-defined random coil 3 ␤-sheet transition (Fig. 2), as expected of a species undergoing a conformational transition associated with fibril formation (18,21). We used neurite retraction and mitochondrial conversion of the tetrazolium redox dye, MTT, as early indicators of cytotoxicity. TGF-␤1 and TGF-␤2 alone had no observable effect on either neurite retraction or MTT dye conversion in PC12 cells (F (2,29) ϭ 0.3632, p ϭ 0.6985, Fig. 3A), as expected of cells that do not express TGF-␤ receptors. Also as expected, A␤-(1-40) treatment caused neurite retraction (Fig. 3A) and inhibition of MTT dye conversion (reduced by 25% relative to untreated PC12 cells, t 0.05,23 ϭ 3.426, p ϭ 0.0023, Fig. 3B). Both TGF-␤1 and TGF-␤2 potentiated the cytotoxic effect of A␤-(1-40) as evidenced by further neurite retraction (Fig. 3A) and a further reduction of MTT dye conversion (to 50% of control, F (2,19) Fig. 3B), however, this effect was not exacerbated by TGF-␤s which is expected of a cell model expressing all three TGF-␤ receptor types. In fact, a marginal reversal of the effect of A␤-(1-40) was observed during treatment with TGF-␤s. These combined data demonstrate that TGF-␤ receptor-mediated events supersede the receptor-independent events. The reverse peptide, A␤-(40 -1), did not exert any effect on MTT reduction either alone or in combination with TGF-␤s (data not shown). The extremely low ratio of TGF-␤s to A␤-(1-40) used during these experiments, i.e. a 1 to 50,000 molar ratio, supports the idea that TGF-␤ may have a seeding effect and emphasizes the potential physiological relevance of these observations.
TGF-␤ Isoforms Bind Directly to A␤-  with Low Nanomolar Affinities-We used a Surface Plasmon Resonance-based biosensor to test for a direct physical interaction between TGF-␤ isoforms and A␤ peptides. The biosensor would detect mass accumulation resulting from binding of the individual TGF-␤ isoforms to the covalently immobilized A␤ peptides as a change in the refractive index of the surface matrix and would generate a curve recorded in arbitrary resonance units (RUs). Both TGF-␤1 and TGF-␤2 were observed to bind significantly to freshly dissolved A␤- , with the binding of TGF-␤2 being greater than that of TGF-␤1 (Fig. 4, A and B). The extent of binding of the TGF-␤ isoforms to A␤-(12-28) was significantly lower than that observed with A␤-(1-40), while no detectable binding was observed on surfaces to which A␤- (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) was immobilized. Subsequent biosensor experiments were focused on the physiologically relevant A␤-(1-40) peptide given the overall lack of interaction of the pharmacological fragments A␤- (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) and A␤-(12-28) with TGF-␤ isoforms in the present biosensor study. The specificity of the interaction between injected TGF-␤2 and the immobilized A␤-(1-40) was confirmed by co-injection of the TGF-␤ ligand specific antibody 3C7 (Celtrix Pharmaceuticals, Inc.) (Fig. 5), thus excluding the possibility that the binding to A␤-(1-40) that we observed might be due to a protein contaminant in the commercial TGF-␤ preparation.
The binding of TGF-␤ isoforms to freshly dissolved A␤-(1-40) was characterized in more detail using the biosensor by varying the concentration of TGF-␤1 and TGF-␤2. The curves (Fig.  6, A and D) clearly indicated that both TGF-␤1 and TGF-␤2 bound to A␤-(1-40) in a concentration-dependent manner. Fitting of these binding data using nonlinear least squares analysis and numerical integration of the differential rate equations (22) demonstrated that the binding of TGF-␤1 to A␤-(1-40) could not be described well by a simple binding model (AϩB 3 AB), as judged by the variance in the residuals Neurite retraction (arrows) is readily apparent following treatment with A␤-(1-40) (bottom) and is even more evident following co-treatment with the individual TGF-␤ isoforms. B, the viability of similarly treated PC12 cells was quantitated using MTT dye conversion. Absorbance values (n ϭ 6 -8 experiments done in triplicate) are expressed as mean Ϯ S.E. percent control. A␤-(1-40) reduced MTT dye conversion to ϳ75% of control levels. TGF-␤s alone had no effect on TGF-␤ receptor-null PC12 cell viability, but they were able to potentiate the toxic effect of A␤-(1-40) (top). In contrast, TGF-␤s did not promote the A␤-(1-40)-mediated toxicity in TGF-␤ receptor-expressing SH-SY5Y cell culture (bottom). The reverse peptide, A␤-(40 -1), did not affect MTT reduction, either alone or in combination with TGF-␤s (as these data do not add anything to the figure, they are not shown). **, p Ͻ 0.01 compared with control and *, p Ͻ 0.05 compared with A␤-(1-40) alone. between the calculated and experimental data (Fig. 6B). However, the data was well represented by a rearrangement model (AϩB 3 AB 3 AB*) (Fig. 5, A and C) with an apparent K D of 60.5 Ϯ 5.2 nM. The fitting of the rearrangement model suggests that the initial TGF-␤1-A␤-(1-40) complex undergoes a kinetically detectable rearrangement, perhaps a structural transition. This observation may point to a direct effect of TGF-␤1 on A␤-(1-40) fibril formation since A␤-(1-40) is known to undergo a conformational transition from a predominantly random coil 3 ␤-sheet-rich form during fibrillogenesis. In the case of TGF-␤2 (Fig. 6B), when the curves derived from all of the TGF-␤2 concentrations were taken into account, neither the simple model (Fig. 6E) nor rearrangement model (data not shown) represented the data well. However, the lower concentration curves for TGF-␤2 could be fit using a simple model, resulting in an apparent K D of 96.1 Ϯ 17.9 nM (Fig. 6, D and F). When the 100 and 150 nM TGF-␤2 curves were predicted using the constants derived from the fitting of the lower concentration curves, the experimental curves were found to have significantly higher plateau values than the predicted curves ( Fig.  6D), illustrating the complexity of the TGF-␤2-A␤-(1-40) interaction. The greater than predicted RU values at higher TGF-␤2 concentrations may result from a change in density of the biosensor matrix due to the formation of aggregates or fibrils on the surface. The interaction between TGF-␤s and A␤-(1-40) is not generalized to growth factors. Indeed, this was confirmed by the absence of binding between A␤-(1-40) and 150 nM nerve growth factor (NGF; data not shown), which, along with TGF-␤s, is a member of the cystine-knot-containing superfamily of growth factors (23). Subsequent binding of TGF-␤2 to the same surface confirmed the presence of covalently immobilized A␤-(1-40) (data not shown).

Mouse Hippocampal Field CA1 and Dentate Gyrus Lack TGF-␤ RI Receptors-
The hippocampus is a structure particularly vulnerable during amyloid pathology. The increases in amyloid plaque burden that occur following co-treatment of hippocampal slices with A␤ and TGF-␤s (12) and in TGF-␤1/ hAPP bigenic mice (7,14) were found to be subfield-specific, indicating the context-dependent nature of TGF-␤ action. Our observations indicate that TGF-␤ enhances A␤-(1-40)-mediated neurotoxicity in a receptor-independent manner. We therefore examined the expression of all three TGF-␤ receptors within the hippocampal formation to determine if the vulnerability of particular subfields correlates with the pattern of TGF-␤ receptor expression. Although no TGF-␤ RI receptor expression was detected in field CA1 and dentate gyrus (Fig.  8A), expression was detectable, albeit weak, in the stratum pyramidale of fields CA2-CA3, while much stronger staining was found in the stratum lucidum of fields CA2-CA3 (Fig. 8C) through to the hilus of the dentate gyrus. TGF-␤ RII receptors were expressed throughout the hippocampal formation particularly in the stratum pyramidale of fields CA1-CA3 and the strata moleculare and granulosum of the dentate gyrus, with sparse staining also observed in the neuropil (Fig. 8, B and D). We were unable to detect TGF-␤ RIII receptor expression in the hippocampus. However, examination of the caudate putamen and inner granular layer of the cerebellum revealed diffuse TGF-␤ RIII receptor staining throughout these regions confirming the ability of this antibody to immunoreact with the TGF-␤ RIII receptor (data not shown). Experiments were performed in parallel to verify the specificity of the antibodies for the various TGF-␤ receptors. Competition by preadsorbtion of the antibody with the relevant immunizing peptide diminished immunostaining in all cases (data not shown).

DISCUSSION
The role of TGF-␤s in A␤ function has been the subject of much speculation given that it apparently exerts both beneficial and deleterious effects in neuronal cells through the same high-affinity cell surface receptor system (7,10,14,17). We now show that the A␤-(1-40) peptide decreases 125 I-TGF-␤1 binding to all three types of TGF-␤ receptors. These results agree with those of Huang et al. (25) who showed that A␤-(1-40) decreases TGF-␤ receptor binding but has little effect on TGF-␤ signaling (confirmed in our laboratory, data not shown), and those of Bodmer et al. (26) who also observed that APP does not diminish TGF-␤ signaling. In explanation of these results, Huang et al. (25) hypothesized that A␤-(1-40) acts as a classical receptor antagonist, i.e. that it interacts directly with a TGF-␤ receptor. We propose that these binding displacement data are more readily interpreted as resulting from ligand sequestration due to a direct interaction of ligand with A␤-(1-40) rather than from A␤-(1-40) interacting with a receptor. Ligand sequestration by A␤-(1-40) would likely reduce binding to all three receptor types equally, as observed. In contrast, if A␤-(1-40) was acting as an antagonist at a given receptor type, then it would likely interfere selectively with binding to that receptor type.
In support of our hypothesis, we provide evidence that TGF-␤ ligands interact directly with A␤-  and, at seeding concentrations, enhance the formation of structures having a gross morphology characteristic of protofibrils (Figs. 6 and 7). A tightly coiled or "beaded" substructure reminiscent of the flexible, "beaded" protofibrillar intermediates characterized by Walsh et al. (21), was particularly evident when A␤-  was coincubated with the TGF-␤2 isoform, although a looser helical structure was also observed when A␤-(1-40) was coincubated with the TGF-␤1 isoform (Fig. 7). The pathological relevance of this direct physical interaction was suggested by the ability of TGF-␤s to potentiate A␤-(1-40)-mediated neurotoxicity in TGF-␤-receptor-null PC12 cell culture, but not in SH-SY5Y cell cultures which do express TGF-␤ receptors (Fig. 3B). This receptor-independent effect was marginally greater with TGF-␤2 than with TGF-␤1 (Fig. 3). Ren and colleagues (17) had previously used another TGF-␤-receptor null cell culture, e.g. NT2/D1, to examine the role of TGF-␤ in A␤-function. They did show conclusively that NT2/D1 cells express TGF-␤ type II receptors upon differentiation, which subserves the neuroprotective effect of TGF-␤. However, TGF-␤s did not exacerbate A␤-induced cytotoxicity in undifferentiated NT2/D1 cells. This is easily explained by the fact that these authors made use of the A␤-(25-35) fragment of A␤, which does not physically interact with TGF-␤s (present study) and, thus, does not allow for our proposed TGF-␤ receptor-independent events. The importance of TGF-␤ receptor-independent events is underscored by our demonstration of the low "seeding" ratios of TGF-␤ to A␤-(1-40) required to promote A␤-(1-40) protofibril formation and neurotoxicity. The accumulation and stabilization of protofibrillar populations are believed to be obligate factors in A␤ fibrillogenesis (27). In fact, recent observations suggest that the plaque deposit might simply represent the clinical end point in the A␤ cascade, with most of the neurotoxicity being mediated by smaller oligomeric/protofibrillar conformations (28 -31). Several proteins have been shown to codeposit in amyloid plaques and a subset of these appear to modulate A␤ fibrillogenesis (reviewed in Ref. 32). Yet, relatively high ratios of modulator to A␤ are often required for enhancement of fibrillogenesis, and contradictory results have been obtained for the effects of several of these proteins on A␤ fibrillogenesis, putting into question the relevance of these results to the roles of these modulators in vivo (see Ref. 33).
Optimal TGF-␤ receptor-mediated signaling hinges on recruitment of ligand by the TGF-␤ RIII receptor, subsequent binding to the RII receptor, and transphosphorylation of the RI receptor, the latter event being an absolute requirement for signaling (19,34). Our demonstration of a lack of detectable RI receptor expression in the hippocampal CA1 pyramidal layer and dentate gyrus (Fig. 8) suggests that signaling from either TGF-␤1 or -␤2 would not be favored in these two subfields, both of which are vulnerable during amyloid insult (30). We have also previously shown that binding of the TGF-␤1 isoform to RII receptors can occur independently of RIII receptors, while binding of the TGF-␤2 isoform to RII receptors appears to require RIII receptors (35). Our present demonstration of a lack of detectable TGF-␤ RIII receptor immunoreactivity in the hippocampus therefore suggests that TGF-␤2 binding to receptors would not be favored in this structure, facilitating the interaction of this isoform of TGF-␤ with A␤- . In support of the relevance of the TGF-␤2-A␤ interaction to fibrillogenesis and neurotoxicity in vivo, it has been observed that the distribution of fibrillar A␤ deposits in brains from two APP transgenic mouse models (36 -38) as well as the regional vulnerability of AD brains (39) parallels the regional distribution of TGF-␤2 immunoreactivity in mouse (40) and rat (41) brain, i.e. frontal and entorhinal cortices and hippocampus. The fact that degenerative changes of the CA1 pyramidal layer induced by co-treatment of hippocampal slices with A␤ and TGF-␤s were limited to the TGF-␤2 isoform (12) further emphasizes the importance of this isoform in neurodegeneration. Taken together, these receptor localization data support the hypothesis that the direct, receptor-independent effect of TGF-␤ is an important component of its amyloidogenicity. This molecular mechanism does not mitigate, but rather complements, the regionally selective nature of TGF-␤'s receptor-dependent effect, for example, on extracellular matrix deposition by cerebrovascular cells or on A␤ clearance by parenchymal glial cells. Our present cell culture data suggest that TGF-␤ receptormediated events would supersede the receptor-independent events. One may also speculate as to the possibility of isoformspecific pathologies (e.g. TGF-␤1 and cerebrovascular amyloid angiopathy & TGF-␤2 and AD-like neurodegeneration) based on regional and/or cell-type specific TGF-␤ isoform and receptor expression (7,10,12,14,40,41).
The identification of a receptor-independent interaction between TGF-␤s and A␤ defines a molecular mechanism that provides for a potent amyloidogenic action of TGF-␤. We suggest that a specific targeting of the A␤-TGF-␤ interaction would diminish A␤ protofibril formation, which was recently suggested to be more relevant to neurotoxicity than late stage plaque formation.