Address correspondence and reprint requests to Marco Antenio Maximo Prado, Laboratório de Neurofarmacologia, Departamento de Farmacologia, ICB, Universidade Federal de Minas Gerais, Avenida. Antonio Carlos 6627, 31270–910, Belo Horizonte, Minas Gerais, Brazil. E-mail: email@example.com
The cellular prion protein (PrPc) is a glycosylphosphatidylinositol (GPI)-anchored plasma membrane protein whose conformational altered forms (PrPsc) are known to cause neurodegenerative diseases in mammals. In order to investigate the intracellular traffic of mammalian PrPc in living cells, we have generated a green fluorescent protein (GFP) tagged version of PrPc. The recombinant protein was properly anchored at the cell surface and its distribution pattern was similar to that of the endogenous PrPc, with labeling at the plasma membrane and in an intracellular perinuclear compartment. Comparison of the steady-state distribution of GFP-PrPc and two N-terminal deletion mutants (Δ32-121 and Δ32-134), that cause neurological symptoms when expressed in PrP knockout mice, was carried out. The mutant proteins accumulated in the plasma membrane at the expense of decreased labeling in the perinuclear region when compared with GFP-PrPc. In addition, GFP-PrPc, but not the two mutants, internalized from the plasma membrane in response to Cu2+ treatment and accumulated at a perinuclear region in SN56 cells. Our data suggest that GFP-PrPc can be used to follow constitutive and induced PrPc traffic in living cells.
The cellular prion protein (PrPc) is a glycosylphosphatidylinositol (GPI)-anchored plasma membrane protein whose function is still under debate. Conformational altered forms of PrPc are known to cause neurodegenerative diseases in humans and other mammals (Prusiner 1998; Weissmann 1999). The diseases can arise due to mutations of PrPc, but it has also been proposed that some protease resistant forms of the protein can be infectious (Prusiner 1998); the latter are known to be the etiological agent of transmissible spongiform encephalopathies.
There is no consensus regarding the mechanisms of PrPc internalization. Initial experiments using a chicken homologue of PrPc indicated that the protein might be accumulated in regions containing coated pits, and also in clathrin-coated vesicles (Shyng et al. 1994). However, evidence that mammalian PrPc, like other GPI anchored proteins, is present in DIGs in the brain and in the cultured cells has also been obtained (Gorodinsky and Harris 1995; Vey et al. 1996; Kaneko et al. 1997). Importantly, mutations and deletions that cause disease may influence PrPc traffic (Rogers et al. 1990; Lehmann and Harris 1997; Hegde et al. 1998). Hence, the possibility to study the traffic of PrPc in living cells should be explored.
In this work, we have generated a tool to follow PrPc traffic in living cells by tagging PrPc with the enhanced green fluorescent protein (GFP). We used this methodology to compare the localization and endocytosis of GFP-PrPc and two N-terminal deletion mutants in living cells; these mutants have been shown previously to cause ataxia and neuronal degeneration in transgenic mice (Shmerling et al. 1998). We visualized for the first time PrPc internalization and intracellular accumulation in the presence of Cu2+. Moreover, we show that the two N-terminal mutants of PrPc accumulate at the plasma membrane and fail to internalize in response to Cu2+. These data show the feasibility to use GFP-PrPc to study PrPc dynamics in living cells.
Materials and methods
The SN56 cells were a generous gift of Prof. Bruce Wainer (Department of Pathology, Emory University School of Medicine, Atlanta, GA, USA). SN56 cells were maintained in Dulbecco's modified Eagle's medium, 10% fetal bovine serum, 2 mm l-glutamine, and 1% penicillin/streptomycin in 25 cm2 culture bottles in a 5% CO2 atmosphere at 37°C. Medium was changed every 2 days. Cell differentiation was performed for 2 days with 1 mm dibutyryl-cyclic AMP in the same medium but lacking fetal bovine serum. During differentiation, medium was changed every day. For confocal analysis, cells were grown directly on coverslips and cultured similarly. The SN56 cells were derived from septum neurons (septum neurons X neurobalstoma N18TG2; Hammond et al. 1990) and present a number of neuronal features including expression of synaptic vesicle proteins (Barbosa et al. 1999) and neuronal type calcium channels (Romano-Silva et al. 2001). These features are increased by differentiation (Blusztajn et al. 1992; Romano-Silva et al. 2001).
The cDNA encoding the signal peptide (SP) of mouse PrPc (amino acids 1–22) was amplified by PCR using primers 5′-CCACCGGT ATCAGTCATCATGGCGAACCT-3′ and 5′-AAACCGGTAAGC AGAGGCCTACATCAGTC-3′. The PCR products were purified and cloned 5′of EGFP in the pEGFP-C1 vector (Clontech Laboratories, Palo Alto, CA, USA) using AgeI restriction sites (SP-GFP).
The region codifying amino acids 23–254 was amplified by PCR using primers 5′-CAGAATTCTAAAAAGCGGCCAAGC-3′ and 5′-CAGGATCCACCCACGATCAGGAAGATG-3′. Two deletion constructs, Δ32-121 and PrPΔ32-134, were generated using the same primers used for the 23–254 region described above and using as template the DNA of recently generated transgenic mice (kindly donated by Drs C. Weismann and H. Kretzschmar; Shmerling et al. 1998). The wild type and mutants PrPc PCR products were purified and cloned 3′of EGFP in the pEGFP-C1 vector (*GFP-PrPc) or in the SP-GFP vector using EcoRI and BamHI restriction sites (GFP-PrPc).
Plasmids were purified and their inserts entirely sequenced using the following primers: 5′-AGTGAACCGTCAGATCCGCT-3′, 5′-AACAGCTCCTCGCCCTTGCT-3′, 5′-AAGTCCGGACTCAG ATCTCG-3′ and 5′-TACAAATGTGGTATGGCTGA-3′.
The SN56 cells were plated on coverslips or in 50 mL culture bottles one day before transfection. Cell transfection was performed by liposome-mediated method (Lipofactamine 2000, Life Technologies, Gaithersbug, MD, USA) according to manufacturer's instructions. One microgram of plasmid and 2.5mL of Lipofectamine 2000 were used for each 5.5 × 104 cells. After 4 h of transfection, cells were maintained in serum-free medium and differentiated for 2 days.
Western blot analysis
Cell extracts were prepared by homogenizing the cell pellet in Phosphate-buffered saline (PBS; pH 7.4) containing 5 mm EGTA, 0.5% TritonX-100 and a protease inhibitor mixture (complete protease inhibitor tablets at twice the concentration suggested by the manufacturer; Boehringer Mannheim, Indianapolis, IN, USA) and then cell extracts were dissolved in SDS-sample buffer. Proteins from cell extracts (40 µg) were resolved on 10% SDS–PAGE and transferred overnight to a nitrocellulose membrane. For immunobloting, nitrocellulose membranes were blocked with PBS containing 0.1% Tween-20 and 5% non-fat milk and incubated with mouse anti-PrPc M183 (1 : 1000) or with a commercial monoclonal antibody against GFP (1 : 500; Clontech Laboratories, Palo Alto, CA, USA). Membranes were washed three times for 10 min with PBS containing 0.1% Tween20 and then incubated with anti-mouse IgG antibody peroxidase conjugate (1 : 15 000). After three washes, staining was revealed by enhanced chemiluminescence (ECL Plus; Amersham Life Science, Piscataway, NJ, USA).
Cells were washed twice in 0.1 m phosphate buffer (PB, 0.1 m; pH 7.4) and fixed with 3% p-formaldehyde in 0.1 m PB for 20 min. After fixation, cells were washed once with 0.1 m PB, twice with Tris-buffered saline (TBS; Tris 100 mm; NaCl 150 mm pH 7.4) and incubated with mouse anti-PrPc for 16 h at 4°C in TBS containing 1% normal goat serum and 0.1% of Triton X-100. Three different antibodies were used: M183 and M133 (produced against recombinant mouse PrPc in PrP null mice; Graner et al. 2000a) and 6H4 (commercial monoclonal antibody against peptide 144–152 from human PrP; Prionics, Zürich, Switzerland), all of which gave identical results. The non-commercial antibodies were specific for PrPc as they did not react with brain extract of PrPc null mice in western blot (not shown). Cells were washed three times with TBS and incubated with goat anti-mouse antibodies conjugated to FITC for 40 min at room temperature (25°C). Cells were then washed three times with TBS and mounted on microscope slides with Hydromount.
Imaging was performed with a Bio-Rad MRC 1024 laser scanning confocal system running the software Lasersharp 3.0 coupled to a Zeiss microscope (Axiovert 100) with a water immersion objective (40×, 1.2 NA). A water-cooled argon UV laser (488 nm) or a Krypton/argon laser was used to excite the preparation (through the 488 nm line), and emitted light was selected with band pass filters (522/35 for FITC).
Image analysis and processing were performed with Confocal Assistant 4.0, Adobe Photoshop, and Metamorph (Universal Imaging).
To test for GPI-anchoring of the fusion protein, GFP-PrPc, cells were treated with 1 U/mL of phospholipase C, phosphatidylinositol-specific (PiPLC; Sigma, St Louis, MO, USA) for 90 min at 4°C washed three times with HBSS (NaCl 137 mm, KCl 4 mm, MgSO4 1.2 mm, CaCl2 2 mm, glucose 10 mm, HEPES 25 mm) and maintained in HBSS with 1U of PiPLC during imaging. Control cells were treated identically, but without PiPLC.
PrPc internalization assay
Cells were perfused with MEM (Minimal Essential Medium; without phenol red) and after obtaining the first image (0 min), MEM with or without 500 µm Cu2+ was perfused and further optical sections were acquired. Similar results were obtained when the cells were perfused with HBSS instead of MEM. Image analysis were performed with Metamorph (Universal Imaging, Chester, PA, USA) to quantify the fluorescence intensities of GFP-PrPc, GFP-PrP Δ32-121 or Δ32-134 localized in the plasma membrane and the perinuclear region. The regions of interest were delimited by hand and integrated fluorescence intensities of the whole cell, the plasma membrane and perinuclear region were determinated. This process was repeated in the other optical sections acquired after treatment with copper. Delimitation of regions of interest was repeated twice for each image and the results obtained never showed more than 3 % variation. The fluorescence intensity of the perinuclear region (INRx min) and the plasma membrane (IPMx min) were expressed as a percentage (% NRx min or % PMx min) of total fluorescence of the cell (ITx min) for each given time (xmin, where x is 5, 15, or 30 min) and the variations of fluorescence were normalized (NNRx min or NPMx min) to the fluorescence of the first image (% NR0 min or % PM0 min) obtained.
In order to analyze the fluorescence distribution in cells expressing the different GFP-PrP constructs, and to calculate the probability that differences observed for these cells occurred by chance, we constructed cumulative frequency distributions of percentage of fluorescence present at the plasma membrane or at the perinuclear region and used the Kolmogorov–Smirnov statistic (DK-S; Press et al. 1986; Kushmerick et al. 1999).
The constructs generated to study PrPc traffic are shown in Fig. 1(a). We first cloned the cDNA for PrPc signal peptide (amino acids 1–23) N-terminal to GFP (SP-GFP). Then this construction was used for cloning the cDNA for amino acids (a.a.) 24–254 (wild-type) or the two N-terminal deletion mutants (Δ32-121 and Δ32-134). The advantage to use this strategy is that SP-GFP allows for cloning of any PrPc mutants downstream of GFP. Addition of the signal peptide N-terminal to GFP was necessary to avoid processing and excision of the GFP tag with the signal peptide after protein synthesis in cells. Moreover, PrPc cloning at the N-terminus of GFP should be avoided due to the GPI anchor. Therefore, our construction differs from the recent reported GFP-bovine prion construct generated by Negro et al. (2001), in which GFP is inserted between amino acids 42 and 43 of PrPc.
Immunoblot analysis of cells transfected with the GFP-PrPc construct shows (Fig. 1b) a protein of approximately 60 kDa that is recognized by the PrPc polyclonal antibody and also by a GFP monoclonal antibody (arrow). The protein was not present in non-transfected cells. The M183 antibody is specific for PrPc (Graner et al. 2000a) and fails to recognize proteins in PrPc KO animals (not shown). The molecular weight of the recombinant protein corresponded to the addition of the GFP tag (27 kDa) to PrPc (33 kDa). The commercial anti-GFP monoclonal antibody recognized non-specifically a protein of 56 kDa, which was also detected in non-transfected cells. The detection of endogenous PrPc expression in cell lines by western blots is particularly hard, probably because of the small amount of this protein in established lineages despite their neural origin (Scott et al. 1988). Low amounts of a 33–35 kDa endogenous PrPc is also recognized by the PrPc polyclonal antibody in transfected and non-transfected cells (arrowhead). Other two main bands of higher molecular weight are also detected in SN56 cells suggesting perhaps the existence of distinct glycosilated forms of PrPc in SN56 cells. Mouse non-immune antibodies did not stain cell extracts (irrelevant antibody, IR).
Inspection of GFP-PrPc expressing SN56 cells indicated that fluorescence labeling was found at the plasma membrane. Figure 2(a–f) show distinct optical sections of a SN56 cell transfected with GFP-PrPc. Note that the fluorescence in the plasma membrane can be easily identified in the soma and processes.
The maximum Z projection illustrates the labeling of GFP-PrPc in a digital reconstruction confirming that the plasma membrane was well labeled by the GFP-tagged protein (Fig. 2g). In the cytosol, GFP-PrPc localization was discrete and a perinuclear region was prominently labeled (Figs 2b–g and accompanying DIC image, H). Expression of SP-GFP showed a complete different pattern of expression, the protein was present in the cytosol and entered the nucleus (Fig. 2i and DIC). Moreover, adding PrPc to GFP without the signal peptide produced a protein that was retained intracellularly (*GFP-PrPc), as the plasma membrane was not labeled (Fig. 2k). The intracellular labeling by this construct was completely different from that of endogenous PrPc (compare Figs 2k and m).
Immunofluorescence experiments and confocal microscopy analyses show that SN56 cells express endogenous PrPc at the plasma membrane. Labeling was detected in the cell body of differentiated cells (Fig. 2m). Moreover, in the cytosol, a perinuclear structure was labeled (Fig. 2m), that was similar to labeling with the recombinant GFP-PrPc (Compare Figs 2g and m). Identical results were obtained with the monoclonal commercial antibody 6H4 (not shown).
Treatment of cells with PiPLC has been shown to release PrPc by hydrolyzing its GPI anchor (Borchelt et al. 1990; Harris et al. 1993). Thus, to test if GFP-PrPc was properly targeted in SN56 cells, we evaluated whether the recombinant protein was GPI-anchored in the cells. Treatment of transfected SN56 cells with PiPLC decreased plasma membrane labeling when compared with mock-treated cells (Figs 3a and b), suggesting that the protein was indeed GPI-anchored. In general, intracellular labeling was not significantly affected by the PiPLC treatment, although it might be expected that during the treatment and imaging of the cells some intracellular GFP-PrPc reached the plasma membrane and should then be released from the cells. This would contribute to an overall decrease in fluorescence inside the cell.
We next compared GFP-PrPc localization with that of the deletion mutants GFP-PrP Δ32–134 and Δ32–121. When expressed in PrPc null mice, these mutants have been shown to cause a number of neurological symptoms (Shmerling et al. 1998), but little is known on their subcellular distribution compared with PrPc. PrPc continuously cycles through the plasma membrane (Shyng et al. 1993), and it has been suggested for chicken PrPc that the N-terminal region is relevant for PrPc internalization (Shyng et al. 1995). As the N-terminal region (amino acids 23–146) of chicken PrPc presents 64% homology to mice PrPc (a.a. 23–134), we reasoned that if GFP-PrPc is continuously cycling, than these mutants might present alterations in their steady-state distribution.
The mutant proteins (Δ32-121, 32–134) were found at the plasma membrane (Figs 4b and c). A perinuclear region was also labeled by the mutants, but inspection of the images caused the impression that the perinuclear region was in general less labeled with the mutants than with GFP-PrPc and that some cells expressing the mutants had almost no intracellular labeling (Figs 4a–c). Likewise, we also had the impression that the plasma membrane had in general more of the mutant proteins than GFP-PrPc. In order to test this assumption we quantified the fluorescence intensity at the plasma membrane and the perinuclear region of cells expressing GFP-PrPc and the two mutants. These data were normalized for the total fluorescence in the cell and we then constructed plots of cumulative frequency vs. fluorescence distribution for GFP-PrPc and the mutants, as there is cell to cell variation of fluorescence distribution for each of the fusion proteins studied. Thus, the ordinate on Figs 4(d and e) represents the proportion of cells with plasma membrane or perinuclear fluorescence less than or equal to the corresponding value of the abscissa. The advantage of this method is that it gives the range of fluorescence for the total population of cells that we studied and allowed for statistical test of their distribution. Moreover, we do not have to arbitrarily choose bin size, as would be necessary for presenting the data in histograms.
It is clear that GFP-PrP Δ 32–121 and 32–134 were preferentially present at the plasma membrane compared with GFP-PrPc(Fig. 4d). In contrast, fewer cells presented labeling of the perinuclear region when the mutants were expressed compared with GFP-PrPc(Fig. 4e). These differences were statistically significant (p < 0.02 by DK-S). Therefore, the mutants accumulated in the plasma membrane at the expense of a decreased concentration in the perinuclear region.
The PrPc internalization from plasma membrane has been shown to be stimulated several fold in the presence of micromolar concentrations of Cu2+ (Pauly and Harris 1998; Sumudhu et al. 2001). However, direct observation of PrPc internalization in living cells has not been reported. We therefore determined whether GFP-PrPc would be internalized upon Cu2+ exposure. We used 500 µm of Cu2+ as this concentration has been shown to be maximally effective to internalize chicken PrPc (Pauly and Harris 1998). Figure 5(a) shows a representative cell (out of 20 examined) before (0 min) and after 5, 15 and 30 min exposure to Cu2+. There is a clear decrease of fluorescence in the plasma membrane in response to the divalent cation at all time points examined (Fig. 5a). This decrease was accompanied by an increase in fluorescence in the perinuclear region (Fig. 5a). Transfected cells kept under control conditions did not show this response over the 30-min period of the experiment and plasma membrane fluorescence remained relatively constant (Fig. 5c shows 1 out of 23 cells examined). The DIC images (Figs 5b and d) confirm that the cells were healthy at the end of the 30 min perfusion with Cu2+.
We have also made quantitative comparisons of the response of GFP-PrPc and of GFP-PrPc N-terminal mutants (Δ32–121 and 32–134) in response to Cu2+ exposure. Figure 6(a) shows that GFP-PrPc fluorescence in the plasma membrane decreases by 65% during 30 min perfusion. This effect was smaller but could also be elicited by using 100 µm or 250 µm of Cu2+ (not shown). Loss of fluorescence in the plasma membrane was fast and appeared to reflect internalization of the protein, as it was accompanied by an increase in perinuclear accumulation of GFP-PrPc(Fig. 6b). In contrast, GFP-PrP Δ32–121 and 32–134 showed only a smaller decrease in plasma membrane fluorescence, which was linear and much more similar to the decrease found for GFP-PrPc in the absence of Cu2+(Fig. 6a). There was also much less accumulation of GFP-PrP Δ32–121 and 32–134 at the perinuclear region and again, the behavior of the mutants in the presence of Cu2+ was similar to that of GFP-PrPc in the absence of Cu2+(Fig. 6b).
We show in the present work that PrPc traffic can be followed in living neuronal cells. Our data suggest that GFP-PrPc is GPI-anchored at the plasma membrane and is present in a perinuclear region of SN56 cells. During the preparation of this manuscript, Negro et al. (2001) reported a somewhat similar construct prepared with bovine PrPc. The localization of bovine GFP-PrPc in a number of cells was similar to the present report (Negro et al. 2001). Thus, together with the observation that mouse GFP-PrPc in SN56 cells was distributed similarly to endogenous PrPc and that the fusion protein was GPI-anchored, the data suggest that the localization of the fluorescent protein reflects the localization of PrPc. In neurons, early immunohistochemical evaluation also indicated the protein to be present at the plasma membrane and in intracellular membrane compartments (DeArmond et al. 1987; Piccardo et al. 1990).
Transgenic mice expressing the Δ32–121 and 32–134 deletions in a PrPc null background present pathological changes in the cerebellum and several neurological symptoms, including ataxia (Shmerling et al. 1998). It has been suggested that these truncated PrP may interact and block a putative ligand protein, which could be activated by a redundant signaling pathway in the absence of PrPc. Although it was reported that the mutants are found in the plasma membrane (Shmerling et al. 1998), comparison with the relative expression of PrPc has not been previously examined. The two mutant proteins presented the same characteristics in SN56 cells. They were more abundant in the plasma membrane but less concentrated in the perinuclear region than GFP-PrPc, suggesting that the observed difference is not an artifact. The excess of mutant proteins in the plasma membrane at steady-state compared with PrPc might be related to decreased constitutive endocytosis of the proteins. Indeed, as SN56 cells express wild-type PrPc in the plasma membrane, the mutants may fail to compete in the cycling process with the endogenous protein. PrPc cycles between the plasma membrane and an intracellular compartment and N-terminal deletion mutants of chicken PrPc present diminished endocytosis (Shyng et al. 1995), thus, Δ32-121 and Δ32-134 might be defective in constitutive endocytosis.
The present experiments also suggest that GFP-PrPc is functional, as shown by its ability to internalize after exposure to Cu2+. Previous experiments have shown that PrPc binds to 4 Cu2+ ions through the octapeptide repeats at its N-terminal region (Viles et al. 1999). PrPc is internalized by Cu2+, however, the destiny of internalized PrPc has not been investigated in intact cells.
Our results extend previous experiments by Pauly and Harris (1998) and more recent experiments by Sumudhu et al. 2001) by showing that PrPc internalized in response to Cu2+ is accumulated in a perinuclear region. The internalization of GFP-PrPc appears to be specific to the interaction of Cu2+ with GFP-PrPc, as the N-terminal mutants are not internalized under the same condition. However, the mutants have deletions that extend from the Cu2+ binding motif (amino acids 59–90). Thus, these mutants may not internalize because they lack the copper binding motifs, but they may also fail to interact with other proteins that might be involved in PrPc traffic (Martins et al. 1997). Interestingly, a protein receptor that mediates prion peptide neurotoxicity in vitro and perhaps internalization appears also to interact with a conserved region in PrPc (amino acids 114–129; Martins et al. 1997), which is deleted from GFP-PrP Δ 32–134 and partially deleted from Δ 32–121. It is noteworthy that transgenic animals having a deletion that stops after the copper binding motifs of PrPc (amino acids 32–93) are viable and have no neurological symptoms (Flechsig et al. 2000), whereas animals expressing Δ32–121 and 32–134 present some of the symptoms that are present in prion diseases (Shmerling et al. 1998). In order to distinguish whether there are different domains that might be involved in constitutive and induced internalization a collection of fluorescent mutants will have to be generated.
Although we do not address the mechanism that GFP-PrPc uses for internalization (clathrin-mediated or potocytosis-like mechanisms), our data suggest that the protein, once internalized, is targeted to membranous structures close to the nucleus. We did not identify the structures labeled with GFP-PrPc in detail at present, but bovine GFP-PrPc is colocalized with a Golgi marker at the light microscopy level (Negro et al. 2001). It is not clear whether all GPI-anchored proteins that cycle through the plasma membrane are targeted to the same subcellular compartments. A minimal GPI-anchored fluorescent protein GFP-GPI has been shown to cycle between the plasma membrane and the Golgi (Nichols et al. 2001). The folate receptor and DAF (decay accelaration factor), both GPI-anchored proteins, are present in the recycling endosomal compartment (Mayor et al. 1998). Thus, it is plausible that N-terminal amino acid sequences in GPI-anchored proteins may divert them to subcellular locations other than the Golgi (Nichols et al. 2001). Our data showing the GFP-PrPc accumulates in a perinuclear region in response to Cu2+ may suggest that at least part of intracellular PrPc may also be located in peri-centriolar endosomal compartments, although we cannot discard that the protein cycles through the Golgi or the trans-Golgi network.
Our studies using GFP-PrPc have led to two main conclusions. Firstly, we have shown that GPI-anchored GFP-PrPc can respond to Cu2+ and that once the protein is internalized in living cells it accumulates in a pericentriolar compartment. Finally, GFP-PrPc and PrP mutants can be successfully used to evaluate the steady-state distribution and induced endocytosis in living cells. It remains to be established whether GFP-PrPc could also be used to localize PrPsc accumulation in living cells.
We thank Prof. Bruce Wainer for the SN56 cells. The authors wish to express their gratitude to members of the Laboratório De Neurofarmacologia for their constant assistance during this work. In special we thank C. Kushmerick for help with statistical analysis and J. Barbosa for help with SN56 cells. We also thank Ms A. Pereira and E. E. P. Silva for technical assistance. We are indebt to Mick Brammer (Institute of Psychiatry, London) for language review. This work was supported by a FAPESP grant (VRM 99/07124–8) and also by Pronex, PADCT, CNPq and FAPEMIG (MAMP). KSL received a PhD fellowship from FAPESP (00/03629–7), and ACM received a CAPES PhD fellowship.