Collagen Plays an Active Role in the Aggregation of β2-Microglobulin under Physiopathological Conditions of Dialysis-related Amyloidosis*

Dialysis-related amyloidosis is characterized by the deposition of insoluble fibrils ofβ2-microglobulin (β2-m) in the musculoskeletal system. Atomic force microscopy inspection of ex vivo amyloid material reveals the presence of bundles of fibrils often associated to collagen fibrils. Aggregation experiments were undertaken in vitro with the aim of reproducing the physiopathological fibrillation process. To this purpose, atomic force microscopy, fluorescence techniques, and NMR were employed. We found that in temperature and pH conditions similar to those occurring in periarticular tissues in the presence of flogistic processes, β2-m fibrillogenesis takes place in the presence of fibrillar collagen, whereas no fibrils are obtained without collagen. Moreover, the morphology ofβ2-m fibrils obtained in vitro in the presence of collagen is extremely similar to that observed in the ex vivo sample. This result indicates that collagen plays a crucial role in β2-m amyloid deposition under physiopathological conditions and suggests an explanation for the strict specificity of dialysis-related amyloidosis for the tissues of the skeletal system. We hypothesize that positively charged regions along the collagen fiber could play a direct role inβ2-m fibrillogenesis. This hypothesis is sustained by aggregation experiments performed by replacing collagen with a poly-l-lysine-coated mica surface. As shown by NMR measurements, no similar process occurs when poly-l-lysine is dissolved in solution with β2-m. Overall, the findings are consistent with the estimates resulting from a simplified collagen model whereby electrostatic effects can lead to high local concentrations of oppositely charged species, such as β2-m, that decay on moving away from the fiber surface.

The deposition of ␤ 2 -microglobulin (␤ 2 -m) 2 into amyloid fibrils is the hallmark of dialysis-related amyloidosis (DRA), a disease arising as a complication of long-term hemodialysis. ␤ 2 -m is a 99-residue protein (molecular mass 11.7 kDa) that represents the light chain of the major histocompatibility complex class I (MHCI), an integral membrane protein involved in the immune response. As a result of normal MHCI catabolism, ␤ 2 -m is released in the serum from the cell surface and carried to the kidney for clearance. In the presence of kidney failure, the concentration of free circulating ␤ 2 -m can increase by up to 50-fold; the persistent increase in ␤ 2 -m concentration results in amyloid deposition, preferentially localized in the musculoskeletal system. The accumulation of ␤ 2 -m deposits has been shown to cause arthralgias, destructive osteoarthropathies, and carpal tunnel syndrome (1). Although a high concentration of ␤ 2 -m is a necessary condition for the onset of the disease, there is not a strict correlation between the disease severity and ␤ 2 -m levels (2), suggesting that other factors might be involved in ␤ 2 -m amyloid deposition.
The aggregation process of ␤ 2 -m has been the object of extensive investigation for many years. Several experiments have been designed to reproduce ␤ 2 -m amyloid deposition in vitro and to understand the molecular conformational changes involved in this process. Spontaneous assembly of ␤ 2 -m amyloid fibrils was observed in vitro under acidic conditions hardly compatible with the physiologic ones; in this case the morphology of in vitro fibrils was influenced by pH and ionic strength (3,4). It has been reported that, at neutral pH, ␤ 2 -m can form fibrils in the presence of high concentrations of free copper ions (5). Furthermore, at neutral pH, ␤ 2 -m can elongate preformed fibrils when the solution contains trifluoroethanol (6) or sodium dodecyl sulfate (7) or when the protein populates an intermediate state of the folding pathway (8). The ability of ␤ 2 -m to elongate preformed fibrils at neutral pH is also enhanced by truncation at the N terminus (9) as well as by cleavage of the peptide bond at the carbonyl side of Lys-58 (10).
Although the information resulting from these in vitro experiments can contribute to insight into the aggregation process of ␤ 2 -m, it cannot be directly related to in vivo fibrillogenesis. A further step toward the understanding of ␤ 2 -m fibrillation in vivo is required, by designing aggregation conditions as close as possible to the physiopathological ones. One of the most surprising properties of DRA is its strict specificity for tissues of the skeletal system, despite the fact that ␤ 2 -m is ubiquitously released throughout the body from every cell expressing the major histocompatibility complex class I. In all patients, in fact, bones and ligaments are primarily involved. Therefore, it might be speculated that the molecular environment that ␤ 2 -m encounters in the skeletal system might favor the local protein aggregation and formation of amyloid fibrils.
Two macromolecules highly represented in these tissues, glycosaminoglycans and collagen, have been investigated as a putative target of ␤ 2 -m deposition. Ohashi et al. (11) have found that ␤ 2 -m binds glyco-saminoglycans with a K d of ϳ1 ϫ 10 Ϫ5 . Our group has shown that ␤ 2 -m binds collagen, giving an adduct with similar affinity, whereas with the truncated ␤ 2 -m form lacking six amino acids at the N terminus the collagen adduct exhibits a 10-fold higher affinity (12).
In this work we have explored the fibrillogenesis of ␤ 2 -m under physiopathological conditions by atomic force microscopy, which is one of the most sensitive and specific techniques currently available to study amyloid aggregate morphology (13)(14)(15). We started from the inspection of ex vivo material and then settled on a procedure for monitoring by AFM the process of ␤ 2 -m aggregation in vitro under conditions as close as possible to the physiopathological ones. Thioflavin T assay was employed to check fibril formation in solution, while NMR spectroscopy was used to monitor possible changes of the protein structure in solution. Our experiments revealed a strict association between ␤ 2 -m fibrils and collagen fibrils and suggest that collagen might act as a catalyzer for fibril formation as a consequence of its positive surface charge. To test this hypothesis, we performed aggregation experiments in the presence of artificial positively charged surfaces. Finally, a simplified model of the collagen fiber was developed to analyze direct and/or orientational electrostatic effects that could lead to an accumulation of ␤ 2 -m at the charged interface.

EXPERIMENTAL PROCEDURES
Extraction and Purification of ex Vivo ␤ 2 -m Amyloid Material-Amyloid fibrils were extracted from amyloid deposits isolated from the femoral head of a patient admitted for hip replacement surgery. The material was homogenized in 2 ml of 10 mM Tris/EDTA, pH 8.0, containing 1.5 M phenylmethylsulfonyl fluoride/100 mg of tissue and centrifuged at 50,000 ϫ g in an ultracentrifuge (Beckman L8 -704; Beckman Instruments) for 30 min and the supernatant removed. After this step was repeated nine times, the absorbance measurement at 280 nm was Ͻ0.05. The pellet was then homogenized in water in the presence of 1.5 M phenylmethylsulfonyl fluoride following the procedure of Gejyo et al. (16), and six aqueous fractions were obtained. The yield in fibrils was monitored by microscopic analysis of the extracted material stained with Congo Red.
Expression and Purification of Recombinant ␤ 2 -m-Expression and purification of recombinant ␤ 2 -m and ⌬N6␤ 2 m were carried out as previously reported (9). The concentration of the protein sample was determined spectrophotometrically at 280 nm by using an extinction coefficient (A 1 cm%) of 16.17 for ␤ 2 m and 17.22 for ⌬N6␤ 2 m.
Preparation of Type I Fibrillar Collagen-Type I collagen was purified from calf skin as previously described (12). The purity of collagen was checked by SDS-PAGE, and the concentration of the collagen solutions was determined by the hydroxyproline assay according to Huszar et al. (17). Fibrillar collagen was prepared by solubilizing purified collagen in 5 mM acetic acid. The solution was diluted 1:1 with 2ϫ phosphatebuffered saline and incubated at 37°C for 30 min.
Thioflavin T Assay-Quantification of amyloid fibril formation was performed with the method described by LeVine (18). Thioflavin T (ThT) concentration was 5 mM, and the buffer used was 50 mM glycine NaOH, pH 8.5. Measurements were made using a LS50 PerkinElmer spectrofluorometer with excitation at 455 nm; emission was collected at 485 nm. The slits were set at 5 mm.
Congo Red Birefringence-Congo Red birefringence of ␤ 2 -m fibrils grown in the presence of collagen was assayed as described previously (19).
NMR Spectroscopy-1 H NMR spectra were obtained at 500.13 MHz with a Bruker Avance spectrometer on 0.1-0.63-mM solutions of pure ␤ 2 -m or mixed with poly-L-lysine (hydrobromide, viscosimetric MW av 55,200; Sigma). The average molecular mass of 50 kDa was adopted to calculate poly-L-lysine amounts to be added to the ␤ 2 -m solution. Samples were dissolved in H 2 O/D 2 O 90/10 or 95/5 with 50 -70 mM phosphate buffer, 50 -100 mM NaCl and pH in the range 6.5-6.7 or in 150 mM deuterated ammonium acetate at pH 6.4. Deuterium oxide (D 99.9%) was from Aldrich; d 4 -acetic-acid (D 99.5%) was from Cambridge Isotope Laboratories. Samples were filtered using 20-nm pore Whatman filters. The studies were carried out at 37 and 40°C. For the pure protein samples two-dimensional TOCSY (20) spectra were acquired as previously reported (21). Experiments with poly-L-lysine were carried out on filtered samples of protein at pH 6.6 (0.63 mM ␤ 2 -m, 70 mM phosphate buffer, 100 mM NaCl) and pH 6.4 (0.50 mM ␤ 2 -m, 150 mM ammonium acetate-d 3 ). The protein samples were incubated at those conditions for 24 h at 37°C before addition of calculated amounts of poly-L-lysine directly into the NMR tube at final concentrations ϳ0.04 mM (␤ 2 -m/poly-L-lysine 14 -16/1). The interaction between ␤ 2 -m and poly-L-lysine was monitored at 37°C over 7-9 days by alternating acquisitions of one-and two-dimensional diffusionordered spectroscopy NMR experiments (22). To remove the effects of convective motions, a compensated pulse sequence was used (23) including solvent suppression by excitation sculpting (24). The length of the diffusion time was between 60 and 90 ms, and the duration of the encoding/decoding gradient was 3-4 ms. Using the same two-dimensional acquisition parameter as described for TOCSY, 131 increments (128 scans/increment) were collected in the indirect dimension, with a gradient strength varying from 2 to 95% of the maximum (68 G/cm).
Data processing and analysis were performed using Felix software (Accelrys Inc., San Diego, CA), and spectra were referenced on the L23 C d H 3 resonance peak (21). Diffusion-ordered spectroscopy spectra were analyzed by means of the routines of GIFA 4.4 (25), i.e., singleexponential fitting and inverse Laplace transformation with the maximum entropy. Fitting was performed on the most intense peaks from ␤ 2 -m and poly-L-lysine, and final results were estimated from averaging over the values affected by small errors (Ͻ2%).
Mass Spectrometry-Liquid chromatography-electrospray ionization/ mass spectrometry analysis was carried out with a Q-STAR spectrometer (Applied Biosystems/PE SCIEX), operating in positive electrospray ionization mode at atmospheric pressure and coupled to an Agilent 1100 series micro-LC pump equipped with a Phenomenex Jupiter 5 C4 300 Å (150 ϫ 0.50 mm) column for standard reverse phase chromatographic elution. Typically, the pellet deposited in the NMR tube was isolated by centrifugation at 14,000 ϫ g and dissolved in 5 M guanidinium chloride (BDH) to get 0.1-0.3 mM concentration of denatured ␤ 2 -m. Aliquots of treated pellet and of NMR tube surfactant were diluted 20-fold by 0.1% trifluoroacetic acid aqueous solution before running liquid chromatography-mass spectrometry analysis. The analysis was systematically repeated by direct infusion of trifluoroacetic acidfree aqueous solution for the optimal detection of the highly charged poly-L-lysine polypeptide.
Atomic Force Microscopy-AFM images were acquired in tapping mode using a Multimode Scanning Probe microscope (Digital Instruments-Veeco, Santa Barbara, CA) equipped with an "E" scanning head (maximum scan size 10 m) and driven by a Nanoscope IV controller. For larger scan sizes, a Dimension 3000 microscope (Digital Instruments-Veeco), equipped with a "G" scanning head (maximum scan size 100 m) and driven by a Nanoscope IIIa controller, was employed. Single-beam uncoated silicon cantilevers (type OMCL-AC, Olympus, and RTESP, Veeco) were used for air imaging. For imaging in liquid, we used V-shaped gold-coated Si 3 N 4 cantilevers (200 m length, nominal spring constant 0.06 N/m) with pyramidal tips having nominal curva-ture radii of ϳ40 nm. The drive frequency was ϳ300 kHz in air and 6 kHz in liquid; the scan rate was between 0.3 and 0.8 Hz. Vertical displacements were calibrated measuring the depth of grating notches (180 nm) and the half unit cell steps (1 nm) obtained by treating freshly cleaved mica with hydrofluoric acid. The horizontal displacements of the piezoelectric tubes were calibrated using a 3-m pitch diffraction grating.
Aggregation of ␤ 2 -m-Lyophilized ␤ 2 -m was dissolved in ammonium acetate 50 mM, pH 7.4, at the concentration of 2 mg/ml and centrifuged at 16,500 ϫ g for 1 h to remove large aggregates. The supernatant was collected and filtered with a 20-nm pore filter; protein concentration was then determined using a Jasco V-530 spectrophotometer. Aggregation experiments were performed at protein concentrations in the 0.2-0.4 mg/ml range, after acidification of the protein solution to pH 6.4 by using an HCl solution at pH 2. The final ammonium acetate concentration was in the 12-30 mM range. For AFM inspection, 20-l aliquots of the sample kept in an incubator at the chosen temperature were extracted at different times during the aggregation reaction, deposited onto freshly cleaved mica, and dried under mild vacuum.
For the experiments in the presence of collagen, fibrillar collagen was sonicated for 10 min in a bath sonicator (ACAD, Genoa, Italy), washed with the ammonium acetate buffer, and, while keeping it under hydration on a microscope slide, cut into several pieces that were washed again and added to the protein solution. Aggregation was performed under the same conditions described above, except for the protein concentration, which was increased to 0.5 or 0.6 mg/ml. For AFM inspection, a piece of collagen with its surrounding protein solution was  extracted from the sample, deposited onto freshly cleaved mica, and dried under mild vacuum.
Poly-L-lysine-coated mica substrates were prepared by incubating a drop (typically 10 l) of 1 mg/ml of poly-L-lysine solution onto freshly cleaved mica for 2 min; the excess polymer was then washed out with the buffer used for the protein. Aliquots of the protein sample extracted at different aggregation times were deposited onto poly-L-lysine-coated mica, incubated for 10 min, washed with buffer, and then imaged under liquid.
Zeta Potential Measurements-Zeta potential was measured at 25°C with a Zetasizer Nano ZS (Malvern Instruments) at a protein concentration of 0.3 mg/ml in ammonium acetate 25 mM, pH 6.4. Each measurement was performed on a freshly extracted aliquot of the protein sample incubated at 40°C and pH 6.4.

RESULTS
Analysis of ex Vivo ␤ 2 -m Amyloid Material-Ex vivo fibrillar material was extracted from the amyloid deposits surgically obtained from a patient who after 20 years of chronic hemodialysis suffered from femoral fracture. The AFM analysis of this material revealed the presence of bundles of fibrils of typical length between 0.7 and 3 m and height between 1.5 and 3 nm. As the sample was dried under mild vacuum to facilitate its adhesion to the mica substrate, this procedure resulted in a reduction in the size of the imaged objects due to dehydration and/or deformation. Similar results have been obtained by other groups using tapping mode AFM in air to image amyloid fibrils (26). Previous control measurements yielded a correction factor of ϳ2.5 for monomeric globular proteins; in the case of amyloid fibrils, however, this factor may be larger (ϳ5) (14). Taking into account this correction, the measured sizes are compatible with those generally acknowledged as peculiar to amyloid fibrils (27).
Ex vivo ␤ 2 -m fibrils are closely packed laterally to generate a kind of planar sheet or plaque (Fig. 1). To obtain a reliable estimate of fibril height, the statistical analysis of the fibril height in cross-section was performed considering only measurements obtained on those fibrils that were sufficiently separated from the rest, thus avoiding artifacts  possibly induced by fibril close packing. A fibril height of 2.4 Ϯ 0.4 nm was obtained. Adjacent fibrils are often partially interconnected or supercoiled (Fig. 1a). A very interesting feature of the ex vivo material is that ␤ 2 -m fibrils are often found in association to collagen fibrils, as shown in Fig. 1b and in the surface reconstructions reported in Fig. 2. Collagen fibrils are much thicker than amyloid fibrils and exhibit the characteristic banding pattern with a periodicity of 67 nm. The height of the collagen fibril shown in Fig. 2 is 50 nm, while its width is ϳ300 nm. The latter value is much larger than expected as a result of broadening effects due to the AFM tip size and indicates that the collagen fibril is flattened on the mica surface. The measured sizes, however, are consistent with the well known diameter of collagen fibrils (28).
Designing Conditions for ␤ 2 -m Fibrillation in Vitro-To model the conditions occurring in vivo in periarticular tissues of patients subject to long-term hemodialysis, we chose to operate at pH 6.4, a value reported to be plausible in the synovial fluids in the presence of flogistic processes (29). Experiments were carried out at temperatures of 37°C, mimicking the physiologic state, and 40°C, which could represent the extreme temperature occurring in the presence of fever.
When ␤ 2 -m is dissolved at pH 6.4, isolated tangles of fibrils can already occasionally be observed at room temperature (not shown). This scanty formation of fibrils could be due to the presence of small seeds that are not removed by filtration with the 20-nm pore filter. To avoid the presence of this material, all the experiments discussed below were performed by first dissolving the protein at pH 7.4, which results in a more stable sample with negligible fibril formation, and then lowering the pH to 6.4 just at the start of the aggregation experiment. Fig. 3a shows typical patterns observed for ␤ 2 -m kept at 40°C and pH 6.4. A relatively low number of globular aggregates and rare filamentous structures are found. The height of the latter is between 0.6 and 1.0 nm; taking into account the dehydration effects discussed above, it is consistent with protofilament rather than fibril size. The number of aggregates of both kind observed over a 30 ϫ 30-m scan area at different aggregation times is shown in the histogram in Fig. 3b. These results show that at the concentrations employed in our experiments (30 -50 M), ␤ 2 -m does not exhibit massive aggregation in vitro at 40°C and pH 6.4, although there is a significant increase in the number of aggregates within 1 h from the start of the aggregation process. However, no increase in ThT fluorescence was observed, indicating that the amount of cross-␤-structured protein is too low to provide a significant fluorescence signal. After 6 days of aggregation, neither fibrils nor filaments were observed and only a few globular structures (5 Ϯ 1 over a 30 ϫ 30-m area) were found. The results obtained for ␤ 2 -m at 40°C and pH 6.4 are in agreement with those obtained under different ionic strength conditions and slightly lower temperature (37°C) by McParland et al. (30).
The same experiment was repeated with the addition of 30% of the truncated form of the protein ⌬N6␤ 2 -m, i.e., the variant devoid of the first six residues at the N terminus. In this case the sample morphology after 30 min at 40°C was similar to that observed with wild-type ␤ 2 -m, whereas after 3 h large globular aggregates 70 -300 nm high were found to coexist with straight, flat aggregates 2-3 nm high and a few hundred nm wide, resembling two-dimensional crystals (Fig. 3c). Interestingly, filament loops (indicated by the arrows in Fig. 4) often detach from the body of the flat aggregates. The filament size is compatible with that measured for ␤ 2 -m alone. The histogram in Fig. 3d shows the counts of filamentous and other structures. Compared with the experiment with wild-type ␤ 2 -m alone, a much larger number of nonfilamentous aggregates is found, whereas the number of filaments does not increase.
The aggregate morphologies observed in vitro at 40°C and pH 6.4 for ␤ 2 -m and ⌬N6␤ 2 -m-enriched ␤ 2 -m are completely different from those observed in the ex vivo amyloid material, as non-fibrillar morphologies prevail. In addition, no increase in ThT fluorescence is detected during aggregation, indicating that no amyloid formation occurs. From these experiments we have therefore concluded that in these conditions ␤ 2 -m was unable to make amyloid fibrils despite the formation of various types of other aggregates.
A new set of experiments was then carried out under the same conditions described above but in the presence of fibrillar collagen of type I. After 4 days of aggregation we observed the formation of bundles of fibrils whose morphology was extremely similar to that reported in Figs. 1 and 2 for the ex vivo sample. Also in this case fibrils were not isolated but grew with a close lateral packing. Fibril formation was more extensive as the aggregation time increased (Fig. 5, a and b). When the protein was prefiltered with a 0.22-instead of 0.02-m pore filter the aggregation kinetics was accelerated, resulting in fibril formation already after 2 days of aggregation (Figs. 5, c and d, and 6a). Because these experiments were carried out at 40°C, which is close to denaturation conditions for collagen, control experiments in the absence of ␤ 2 -m were also performed to rule out possible temperature-induced modifications and/or rearrangements of collagen that could give rise to fibrillar structures. No fibrillar structures such as those shown in Figs. 5 and 6a were observed in the absence of ␤ 2 -m. In addition, the experiment of ␤ 2 -m aggregation in the presence of collagen was repeated at 37°C; at this temperature collagen is stable (31), and the results obtained were very similar to those obtained at 40°C (not shown). The diffuse fibrillar aggregation found for ␤ 2 -m in the presence of collagen suggests that the latter may act as a catalyzer for fibril formation. Indeed, fibrillation is strictly localized and associated to the presence of the collagen surface; no fibrils were observed by AFM in aliquots of solution extracted from the sample bulk liquid and deposited on mica, while they were present only in close proximity to the collagen network. For the same reason, when performing aggregation experiments of ␤ 2 -m in the presence of collagen, it was not possible to detect any increase in ThT fluorescence, as this assay is based on analysis of the bulk solution. The ThT assay in solution is quite unreliable in this particular case for the presence of insoluble fibrillar collagen, which makes quite problematic the specimen sampling. However, the presence of fibrils surrounding the collagen is confirmed by microscopic analysis of specimens stained by Congo Red showing the typical green birefringence of amyloid fibrils (Fig. 6, b and c).
Because the collagen fiber surface exposes patches of clusters of positively charged residues, we have hypothesized that electrostatic effects could be involved in fibril formation. Therefore, to test the fibrillogenic effect of surface charge arrays, we exposed ␤ 2 -m previously incubated at 40°C and pH 6.4 to polylysine-coated mica and imaged the sample by AFM in liquid. Deposition of ␤ 2 -m on the positively charged substrate resulted in widespread fibril formation after 25 h of incubation at 40°C and pH 6.4 (Fig. 7, a and b), but fibrils were already detectable after 4.5 h of incubation (Fig. 8a). Temperature appeared to have a remarkable effect on fibrillogenesis in these conditions. Actually, deposition at pH 6.4 without incubation at 40°C did not generate any fibrils (Fig. 8b). Zeta potential measurements performed as a function of incubation time on ␤ 2 -m at 40°C and pH 6.4 show that between 2 and 24 h of incubation in these conditions a decrease of the zeta potential occurs, from the initial value of 10 Ϯ 5 to 22 Ϯ 5 mV. Therefore, both this change in surface charge (which becomes more negative) and the presence of a positively charged surface (which helps dipole crowding and alignment) are necessary for fibril formation. As a control experiment, we tested a negatively charged mica substrate and found that it does not promote ␤2-m fibril formation.
NMR Spectra-1 H NMR spectra of isolated ␤ 2 -m solution, typically at pH 6.5-6.7 at either 37 or 40°C, never exhibited any pattern suggesting partial or total protein unfolding. This conclusion was always confirmed by spectroscopic monitoring for at least 1 week after sample preparation when the solution was filtered with 20-nm pore filters prior to being submitted to NMR observation. NMR monitoring of ␤ 2 -m solutions in the presence of collagen did not appear viable because of phase heterogeneity. Therefore, NMR measurements were carried out only on ␤ 2 -m/poly-L-lysine mixtures that were checked as reproducibly homogeneous when solutions were prepared. ␤ 2 -m solutions in the presence of poly-L-lysine did not show any detectable changes of the protein NMR chemical shift and/or line width for prolonged observation intervals (typically 7-9 days) under either investigated condition (pH 6.6 in phosphate buffer and pH 6.4 in ammonium acetate). These conclusions could be safely inferred from inspection of those regions of the spectrum that were not populated by poly-L-lysine resonances. Diffusion ordered spectroscopy confirmed that no detectable interactions between ␤ 2 -m and poly-L-lysine take place under the selected experimental conditions. The calculated self-diffusion coefficient for the protein was (1.48 Ϯ 0.01) ϫ 10 Ϫ10 m 2 s Ϫ1 , which is not significantly different from the value obtained with isolated ␤ 2 -m under the same conditions, i.e., (1.53 Ϯ 0.01) ϫ 10 Ϫ10 m 2 s Ϫ1 . The self-diffusion coefficient obtained for poly-L-lysine was (0.96 Ϯ 0.01) ϫ 10 Ϫ10 m 2 s Ϫ1 , a value that reflects the molecular mass distribution (40 -60 kDa) of the homopolypeptide product and is in agreement with expectation for a globular protein of mass ϳ60 -65 kDa. Thus an interaction with poly-L-lysine should affect ␤ 2 -m diffusion more significantly than the very small effect that emerged from our experimental fitting; although aspecific, transient interactions cannot be ruled out. The NMR spectrum quality was not altered by the limited amount of precipitation that occurred after some 24 -48 h from poly-L-lysine addition. The overall resonance amplitude was only marginally affected by this precipitation in phosphate buffer (some 3% decrease), whereas in ammonium acetate an overall amplitude decrease by 10% was observed. The precipitate, collected and redissolved in guanidinium chloride, was analyzed by electrospray mass spectrometry and shown to contain only ␤ 2 -m with a contamination from the methionine-oxidized ␤ 2 -m (at most 30%) but no detectable trace of poly-L-lysine. The same precipitate submitted to AFM analysis proved to be only amorphous material, mainly globular disordered aggregates of varying dimensions (not shown).
Electrostatic Effects in ␤ 2 -m Collagen Interaction-Modeling has been performed for a simplified model of collagen fiber in order to estimate the possible electrostatic influence on ␤ 2 -m collagen interaction. Lack of knowledge of the details of collagen fiber assembly prevents any attempt to reach precise quantitative predictions. However, analysis of a simplified model of the collagen fiber shows that direct and/or orientational electrostatic effects could be substantial and could lead to very high local concentrations of oppositely charged species (such as ␤ 2 -m) that decay on moving away from the surface. The details of the model are given in the supplemental materials section.
Assuming values of 500, 333, 166 mV/nm for the electric field at the modeled collagen surface, the numerical solution of the Poisson-Boltzmann equation (32,33) gives a set of corresponding potential values. The detailed shape of the fiber could locally magnify or reduce the potential value.
Although quantitative features cannot be described by our model, an important conclusion is that the local concentration (c) of the charged species (say with charge number z) at the surface of the cylinder may be significantly different from the bulk (or free) concentration c f . Based on the Boltzmann equation, the local concentration is given as shown in Equation 1. c ϭ c f e Ϫz (Eq. 1) If the charged species has charge number z Ͼ Ͼ1 or it has a large dipole moment, electrostatic effects may be substantial. Although these considerations should be regarded with much caution because they are not easily transferable to macromolecules, some qualitative assessment of the electrostatic effects on ␤2-m collagen interaction can be attempted.
As far as ␤ 2 -m is concerned, the net charge of the intact protein at neutral pH is Ϫ2.4 and it is Ϫ4.3 for the ⌬N6 mutant (12). The dipole moment is 9.1 ϫ 10 Ϫ28 C m (2.7 ϫ 10 2 Debye), which is moderate for proteins of this size, and it is decreased to 3.4 ϫ 10 Ϫ28 C m (1.0 ϫ 10 2 Debye) for the ⌬N6 mutant. The net effect of deleting the first six residues would be therefore an increase of local concentrations with a possible decrease in orientation effects, although the concept of dipole applies poorly to macromolecules.
The increase in local concentration of protein around the collagen fiber, or other charged surfaces, would thus aid fibrillogenesis, consistent with the experimental observations. ␤ 2 -m is found at a concentra-  (34,35). According to the proposed model, its concentration at the collagen cylinder surface, at neutral pH, could be raised by a factor in the range 10 3 -10 5 . This range should increase by 20 -50-fold in hemodialysed patients, although the net ␤ 2 -m charge reduction, due to the pH decrease (12) associated with local inflammation, should downsize the effect. However, the expected values of surface concentration would still remain remarkably high (in the order of mM). These effects are pictorially shown in Fig. 9. On the other hand, the molecular dipole orientation favored by the electric field at the surface of collagen should be quite significant. If the overall dipole geometry of ␤ 2 -m were idealized as a point dipole located at the center of the molecule, in practice all molecules would be favorably oriented at the collagen surface. This exclusive single-level population should prevail also at one Debye length (ϳ8 Å), even considering the dipole as a charge separation spread over the whole molecular length (corresponding to two 35 Å-spaced, opposite charges of 1.6 elementary units). The dipole energy should become comparable with thermal energy kT, and therefore negligible, at 35 Å away from collagen surface.

DISCUSSION
It is well known that along the collagen fibers there are regions presenting an excess of positive charge (36), and it has been shown that the exposure of other amyloidogenic peptides (37) and proteins (38) to charged surfaces can highly enhance the rate of fibril formation. In particular, it has been shown that the prototypic kIV immunoglobulin light chain SMA, which has a folding motif similar to ␤ 2 -m, displays a very high propensity to make fibrils when it is exposed to the charged surface of mica (38). Fibrillogenesis on charged mica is highly productive at pH 5 but occurs also at pH 7.4, though less efficiently. On the contrary, fibrillogenesis is totally abrogated at pH 7.4 and significantly reduced at pH 5 when the light chain is exposed to hydrophobic surfacemodified mica.
It might be speculated that the mechanism by which collagen facilitates ␤ 2 -m fibrillogenesis is related to its positively charged patches. To test the effect of positive immobilized charges on ␤ 2 -m fibrillogenesis, we deposited ␤ 2 -m previously incubated at 40°C and pH 6.4 on a poly-L-lysine-coated mica. As described under "Results", under those conditions we could ascertain a temperature-dependent formation of fibrils. The two conditions, pH 6.4 and a temperature of 40°C, do not appear to have an effect on the three-dimensional structure of ␤ 2 -m in solution when considered separately. Two-dimensional NMR spectra recorded under very similar conditions (0.1 mM, pH 6.5-6.6, 40°C) exhibit the same pattern as observed at neutral pH. These results indicate that either the onset of partial unfolding transitions of the protein is absent or partially unfolded intermediates are native like and thus hard to distinguish by NMR. A possible perturbation of the solvation properties from NMR-silent intermediates could affect protein-protein interactions, but no specifically consistent evidence was ever inferred from extensive and repeated NMR observations on ␤ 2 -m solutions kept for weeks at 40°C.
It is worth noting that poly-L-lysine in solution has no effect on ␤ 2 -m fibrillogenesis at pH 6.4 and 37°C, an observation confirmed by corresponding NMR results. The small extent of deposit obtained in the NMR tube, in fact, did not exhibit any specific morphologic features but those of a globular precipitate.
The necessity of increasing the temperature to 37-40°C to obtain fibrils may suggest that temperature and pH could cooperatively trigger  an initiating conformational transition that, however, appears to be productive only in the presence of an immobilized array of positive charges. This interpretation is consistent with the results of our assessment of the role that the electric field at the collagen surface may play in concentrating locally favorably charged counterparts such as ␤ 2 -m molecules. The ensuing concentration gradient and plausible dipole orientation effects could overcome the unfavorable statistics of productive conformational transitions leading to fibril-competent nuclei (21). This effect could be considered analogous to that recently reported by Ohhashi et al. (39) concerning the induction of ␤ 2 -m fibril formation by ultrasonication. The ultrasound energy should not only increase the frequency of molecular collisions but also destroy the unproductive low energy interactions and select the high energy ones, i.e., those that are more likely to be fibril competent. Based on our results, it can be therefore concluded that an immobilized charged surface is required for fibrillogenesis.
The experimental evidence presented in this study shows that fibrillar aggregation of ␤ 2 -m under conditions close to the physiological ones is possible and that collagen could play a fundamental role in local fibrillogenesis, possibly representing an immobilized charged surface where the protein is correctly oriented for priming an ordered polymerization that could generate fibril seeds. This observation might have a fundamental implication in future therapeutic strategies for DRA, which affects approximately one million people worldwide. In fact, the demonstration that the collagen/␤ 2 -m interaction has an effect on fibrillogenesis suggests a new target for pharmaceutical strategies.