Structure-Function Analysis of the Glioma Targeting NFL-TBS.40-63 Peptide Corresponding to the Tubulin-Binding Site on the Light Neurofilament Subunit

We previously reported that a 24 amino acid peptide (NFL-TBS.40-63) corresponding to the tubulin-binding site located on the light neurofilament subunit, selectively enters in glioblastoma cells where it disrupts their microtubule network and inhibits their proliferation. Here, we analyzed the structure-function relationships using an alanine-scanning strategy, in order to identify residues essential for these biological activities. We showed that the majority of modified peptides present a decreased or total loss to penetrate in these cells, or to alter microtubules. Correspondingly, circular dichroism measurements showed that this peptide forms either β-sheet or α-helix structures according to the solvent and that alanine substitution modified or destabilized the structure, in relation with changes in the biological activities. Moreover, substitution of serine residues by phosphoserine or aspartic acid concomitantly decreased the cell penetrating activity and the structure stability. These results indicate the importance of structure for the activities, including selectivity to glioblastoma cells of this peptide, and its regulation by phosphorylation.


Introduction
Neurofilaments (NFs) are the major component of the cytoskeleton in mature neurons. They are composed of three polypeptide subunits of light (NFL, 61 kDa in humans), medium (NFM, 90 kDa), and heavy (NFH, 110 kDa) molecular weight [1]. Each subunit contains a central a-helical rod domain of ,310 residues, flanked by a N-terminal head domain and a non-ahelical C-terminal domain. NFL subunits constitute the core of the filament on which NFM and NFH are anchored through coiledcoil interactions. The C-terminal domains of NFM and NFH form side arms that project from the filament [2]. These domains have been shown to be involved in the rate of NF transport, the radial growth of axons and the regulation of inter-filament affinity [3][4][5][6][7][8]. They also bind microtubules (MTs), either directly [9][10][11] or using Microtubule-Associated Proteins (MAPs) as linkers between NFs and MTs [12]. The association with MTs depends on the phosphorylation state of the C-terminal domain: dephosphorylation of this domain by alkaline phosphatase promotes the NF/MT interaction while phosphorylation by the tau protein kinase II dissociates NFs from MTs.
We recently showed that NFs also contain sites in their Nterminal domains able to bind un-polymerized tubulin. Peptides corresponding to these motifs (named TBS for Tubulin-Binding Sites) inhibit the in vitro polymerization of MT. Moreover, a 24amino acid (AA) peptide corresponding to the second tubulinbinding site located on the NFL subunit (NFL-TBS.  penetrates selectively in T98G human glioblastoma cells where it destroys the MT network and consequently inhibits cell viability [13,14]. Considering these properties, this peptide represents an interesting candidate for malignant glioma treatment. Here, we further analyzed the structure-function of NFL-TBS.40-63 using an alanine scan strategy to clarify the cell entering mechanism. The peptides in which each AA was sequentially replaced by alanine were tested for their ability to penetrate into T98G human glioblastoma cells and to disrupt the MT cytoskeleton. In order to detect possible changes in secondary structure the different peptides were also analyzed by circular dichroism (CD) spectroscopy. Finally, since the NFL-TBS.40-63 peptide contains two phosphorylation sites at serines 17 and 19 (which correspond to serines 55 or 57 of the NFL sequence), and considering the effects of the phosphorylation status of these sites on the axonal transport [15], we investigated the properties of modified NFL-TBS.40-63 containing phosphorylated-serine 17 or 19, or aspartic acid substitutions of serines 17 or 19 to mimic permanent phosphorylation.

Results and Discussion
We previously reported the specific uptake of NFL-TBS.40-63 peptide by T98G human glioblastoma cells and its effects on their MT network, while it has no major effect on astrocytes, neurons and other cell types like Hela cells, prostate carcinoma cells, or mouse immortalized fibroblast 3T3 cells [13,14]. As illustrated in Figure 1A the NFL-TBS.40-63 peptide typically enters in glioblastoma cells and disturbs the MT network. It also enters in SW13 cells without affecting the MT network ( Figure 1B). As SW13 cells do not contain intermediate filaments, these results, as well as those previously published [13,14], suggest that intermediate filaments (IF) are not directly involved in the activities of the NFL-TBS.40-63 peptide. Here, we further investigated the structure-activity of this peptide to understand its capacity to penetrate in glioblastoma cells and to alter their MT cytoskeleton.
Revealing the importance of each amino acid by the alanine scanning strategy In order to evaluate the role of each AA of the NFL-TBS.40-63 peptide in its incorporation into glioblastoma cells and its antimicrotubule activity, we sequentially replaced each AA by alanine, a small and neutral AA ( Table 1), and human glioblastoma T98G cells were incubated with 10 mM of peptide during 6 hours. Cells containing the biotinylated peptide were detected using Alexalabeled avidin (green fluorescence) and those displaying a normal or a destroyed MT network were detected with an anti-tubulin antibody (red fluorescence) ( Figure 1A-B). Then, we counted a minimum of 200 cells per slide containing or not the peptide, and with a normal or destroyed microtubule network ( Figure 1C). As illustrated on Figure 1A glioblastoma cells contain a rich MT network. The normal NFL-TBS.40-63 enters in these cells where it destroys the MT network. On the opposite, the peptide with a Val14Ala replacement fails to enter in cells and does not affect the MT network. The NFL-TBS.40-63 peptide also enters in SW13 cells, which do not have intermediate filaments, but does not affect their MT network ( Figure 1B).
A change in activity due to the replacement of an AA may reflect the loss of a specific contact with an interacting molecule [16], like a protein or a lipid molecule, or it can be due to structural modifications of the peptide resulting from the sequence modification [17]. It should be noted that small peptides like NFL-TBS.40-63 generally exist in solution as a mixture of different conformations or sub-states, whose population can vary with changes in the environment or limited sequence modification.
From immuno-cytochemical results and their quantification ( Figure 1A and C), it appears that most of the AA substitutions by alanine reduced the cell penetration activity of the peptide, except for Ser21Ala and Ser24Ala. Moreover, most substitutions of the AA with alanine decreased the capacity of the peptide to alter the MT network when compared to the original NFL-TBS.40-63 peptide ( Figure 1C, Table 1). However, replacement of Ser21 by alanine did not affect the uptake of the peptide but reduced its capacity to alter the MT network.
Alanine substitutions at position 1, 12, 14 and 16 abolished the penetration of the peptide into cells. To confirm the crucial role played by these AA, they were replaced by functionally equivalent AA (Tyr1Phe, Leu12Val, Val14Leu and Arg16Lys) ( Figure 2, Table 1). With the exception of Arg16Lys, which displayed a lower activity, such substitutions restored the wild type activity of the peptide, confirming the crucial role of these residues.
Internalization of the NFL-TBS.40-63 peptide by glioblastoma cells may result from either its intrinsic capacity to cross the cell membrane through endocytosis, or its capacity to interact with a cell receptor. In the first case, the peptide would belong to the class of cell penetrating peptides (CPP). These peptides have been investigated for the last ten years for their high potential in cell delivery engineering, especially with a therapeutic objective. Cell penetrating peptides are generally defined as short (less than 30 AA) cationic peptides, which are able to penetrate cell membranes [18]. In general, CPP uptake involves an initial rapid electrostatic interaction of positively charged residues with the plasma membrane before their intracellular delivery, but the exact mechanism is still unclear [19]. Notably, the positive charge itself is not sufficient for internalization. Futaki et al. showed that an octamer of arginine (Arg) 8 was successfully internalized, whereas (Arg) 16 had no translocation property [20].
The NFL-TBS.40-63 peptide displays some of the CPP characteristics including its short sequence (24 AA) and its net positive charge at physiological pH (PI = 10.3). However, contrary to the known CPPs, which are composed mostly of arginines and lysines, the NFL-TBS.40-63 peptide contains only 2 arginine residues and no lysine. In addition, we showed previously that scrambled peptides (including NFL-SCR), or a peptide with the same AA composition but in a reversed sequence (NFL-SBT), do not penetrate in cells ( Table 2). As such peptides have similar positive charge, these results indicate that the positive charge of the NFL-TBS.40-63 sequence alone is not sufficient to account for its cell-penetrating ability [13]. Moreover, the exact sequence is crucial to preserve the properties of this peptide, which may reveal a preferred and specific conformation.
The properties of the NFL-TBS.40-63 peptide are affected by their phosphorylation status Neurofilaments are known phospho-proteins, and multiple aspects of their biology, including assembly and axonal transport, are regulated by their phosphorylation status. In particular, aberrant NF phosphorylation is a pathological hallmark of several human neurodegenerative disorders [21]. Phosphorylation sites located along the N-terminal domains, especially those of NFL and NFM, are essential for their assembly, their axonal transport or their interactions with MTs [22][23][24][25][26][27][28][29][30]. In particular, modifications of serines 2, 55 and 57 of NFL (i.e. replacement of serine by aspartic acid) alter the axonal transport of NFs [15]. As they are present in the NFL-TBS.40-63 peptide sequence (serine 17 and 19), we analyzed the possible consequence of the phosphorylation of these AA on the internalization of the peptide into glioma cells and on the anti-microtubular activity. Replacement of serines 17 and 19 with aspartic acid to mimic permanent phosphorylation was also examined.
Immunocytochemical studies revealed that all the phosphorylated peptides (Ser17SerP, Ser19SerP, Ser17,19SerP), as well as peptides in which Ser was replaced with Asp (Ser17Asp, Ser19Asp, Ser17,19Asp), showed a decreased internalization, when compared to the original NFL-TBS.40-63 peptide ( Figure 3, Table 2). Moreover, the MT network integrity was not altered when glioma cells were treated with these peptides. These results indicate that the phosphorylation status of serine 17 and 19 is crucial for both the internalization and the effects of the NFL-TBS.40-63 peptide on glioma cells. Phosphorylation clearly opposes these effects.
Analysis of the folding and stability of the NFL-TBS.40-63 peptide by circular dichroism To investigate whether the alteration of the activity of the peptide following changes of its sequence or its phosphorylation level may be correlated to a change in its folding and stability, we analyzed the structure of peptides by circular dichroism (CD) measurements. The CD signal in the far UV region (below 250 nm) is related to the secondary structure of proteins [31,32]. The CD spectra are usually measured in three solvents: (1) 10 mM potassium phosphate with 200 mM KF, pH 7.4, (2) 2 mM Sodium Dodecyl Sulfate (SDS) and (3) pure water. KF/phosphate buffer is similar in ionic strength and pH to the buffer used for cell treatment with peptide. The actual cell treatment buffer cannot be used for the CD measurements because of its high UV absorption in the far UV region. SDS mimics hydrophobic environment at low concentrations (2 mM) where micelle formation should not occur [33]. The peptide structure in the presence of SDS could thus indicate a potential conformation of the peptide when in contact with the membrane or the partner protein. The CD was  also measured in pure water as a reference to evaluate the effect of buffer and SDS on the peptide conformation. All the CD measurements were performed on the peptides that were used for the biological tests. The CD spectrum of the original NFL-TBS.40-63 peptide (wt NFL-TBS.40-63) in pure water indicates rather disordered conformation (black line in Figure 4A): the signal at 200 nm was negative while the ordered conformations (both a-helix and bsheet) exhibit a positive CD signal at this UV region [31,32]. However, an additional negative feature could be seen around 220 nm, suggesting partial folding into a-helix or b-sheet, because an a-helix structure exhibits a negative band at 222 nm and bsheet at 218 nm, while disordered peptides usually exhibit no or a positive signal around 220 nm. In fact, when the CD spectrum was analyzed by using Jasco secondary estimation software with Reed's reference set as reference spectra, the secondary structure was estimated to be composed of 25% b-sheet and 75% random coil. This differs from the CD spectra of a peptide with the same AA composition but in a shuffled sequence (NFL-SCR) (green line in Figure 4A) or a reverse sequence (NFL-SBT) (data not shown): the signal above 220 nm is positive in their CD spectra. These peptides exhibited typical CD spectra corresponding to a completely disordered structure, while the wild-type NFL-TBS.40-63 peptide has some tendency to be folded in a particular conformation.
Furthermore, the CD spectrum of wild-type NFL-TBS.40-63 peptide was modified by addition of 2 mM SDS (blue line in Figure 4A), which creates a hydrophobic environment and may mimic the situation around the membrane, in a way indicating more ordered structures: the signal at 200 nm became positive and the negative band around 220 nm displayed an higher intensity. Two negative bands (one centered at 222 nm and the other at 209 nm) suggest the presence of a-helical structures. The spectral analysis with Reed's reference set predicts 20% a-helix, 40% bsheet and 40% random coil. Thus, the peptide may be folded partially in an a-helical structure when it interacts with the membrane. However, addition of any mild detergents (IGEPAL CA-630, Nonidet P40, Tween 20, Triton X-100) to the peptide in pure water did not affect the CD signal.
In KF/phosphate buffer, which mimics the cell treating buffer, the peptide exhibited another type of CD spectrum: the signal at 200 nm is positive and the negative band around 220 nm was greater in intensity compared to that in pure water (blue line in Figure 4A). Since the negative band was centered at 218 nm instead of 222 nm and there was no clear negative band at 209 nm, another typical negative band for an a-helical structure, the peptide is probably folded in an antiparallel b-sheet structure. The spectral analysis with Reed's reference set predicts 7% ahelix, 56% b-sheet and 36% random coil.
The different structural characteristics observed in KF/phosphate buffer or in SDS correspond to different properties in terms of stability. In KF/phosphate buffer, the peptide was not stable when the temperature was increased: thermal unfolding occurred with Tm of about 40uC (blue line in Figure 4B). In addition, no refolding was observed when the temperature was lowered from 70 to 10uC at a rate of 1uC per min and even after 12 hours incubation at 20uC (light blue line in Figure 4B). Corresponding to this structural feature, the peptide once heated could not penetrate into the cells. In contrast, the peptide in 2 mM SDS was stable: no large signal change in CD spectrum was observed when the temperature was increased up to 70uC (red line in Figure 4B).
The dependency on the environment of the NFL-TBS.40-63 peptide structure is characteristic of its exact sequence arrangement. No modification in the CD signal of its shuffled (NFL-SCR) or reverse sequence peptide (NFL-SBT) could be observed following addition of SDS or in KF/phosphate buffer (data not shown). Moreover, both the structural characteristics of the peptide and its dependency on the environment can be greatly modified by punctual substitutions. As described below, the substitution of only one AA with alanine at certain positions of the peptide affected the CD signal, indicating major change in the structure ( Figure 5).
In summary, the CD analysis indicates that the NFL-TBS.40-63 peptide folds into secondary structure elements and exhibits both a-helix and b-sheet structures, with a more pronounced importance of the a-helical character in a hydrophobic medium relative to a polar environment. Such data argue that for this peptide, membrane penetration would necessitate the formation of specific secondary structures. However, this does not imply that the Using a similar experimental approach to that described in Figure 1, the amino-acids at position 1, 12, 14 and 16 were replaced by amino-acids with similar properties. While substitution of these AAs by alanine abolished their properties, replacement of these AAs by functionally equivalent residues restore the capacity to penetrate in cells and to alter the MT cytoskeleton. As in Figure 1A, experiments were triplicated and a minimum of 200 cells was analyzed in each experiment. doi:10.1371/journal.pone.0049436.g002 peptide necessarily acts as a CPP, as the relation between the properties of CPP and their structural folding is not straightforward. While an a-helical structure is generally found or predicted for these peptides, b-sheet structures have also been described for some CPPs [34]. Furthermore, the a-helical structure of antennapedia can adopt a b-sheet structure in the presence of a charged lipid monolayer [35].
To further evaluate the importance of the structure/function relationships we also analyzed by CD the folding and stability of the peptide following alanine substition of each AA, or their phosphorylation state.

Alanine substitutions and phosphorylation affect the folding of NFL-TBS.40-63 peptide
The CD spectra of alanine-substituted peptides can be separated in two classes ( Figure 5). The spectra of peptides in which an AA residue at C-terminal part (21 to 24) was substituted with alanine exhibited a large negative band around 218 nm in the presence of SDS or in the KF/phosphate buffer. The CD spectra of these peptides depended upon the solvent just like the spectrum of the wild-type peptide. The peptides appeared rather disordered in pure water, in mainly anti-parallel b-sheet in KF/ phosphate buffer and in a mixture of b-sheet and a-helical structures in 2 mM SDS. There was some correlation between the CD spectrum in KF/phosphate buffer and cell penetrating capacity. Gly23Ala, which exhibited only weak cell-penetrating activity, also displayed a CD signal corresponding to less ordered structure (light blue line in Figure 5A).
The CD spectra of peptides in which a residue is substituted with alanine at positions 2 to 20 (except for position 6, which is originally alanine) exhibited much weaker CD signals (red lines in Figures 5A, B and C). However, the spectra had a welldefined shape: a positive band centered at 228 nm, a negative band at 210 nm and a positive band at 200 nm. This kind of CD spectrum is reported to correspond to another type of b-sheet structure according to Yang et al. [36]. We could not fit the spectra using Reed's reference set, but obtained a reasonable fit using Yang's reference: 6% a-helix, 49% b-sheet, 28% b-turn and 16% random coil. The spectra were not affected by addition of SDS or in KF/phosphate buffer condition. The structure thus appeared very stable. Even increasing the temperature up to 80uC did not significantly affect these CD spectra (data not shown).  Almost all peptides exhibiting such CD spectra did not show any cell-penetrating capacity except Ser9Ala and Ser10Ala. These two particular peptides penetrated in cells significantly although to a much lesser extent than the wt peptide. The peptide in which the residue at position 1 was substituted with alanine exhibited CD signals corresponding to a random coil structure in all solvents, as well as a loss of biological activities. Thus, tyrosine at position 1 is important for the folding of this peptide and its activities. As exposed in the previous section, we observed that the substitution by alanine at positions 1, 12, 14 or 16, strongly perturbed the peptide activity, but the substitution with physicochemically similar AA (Tyr1Phe, Leu12Val Val14Leu and Arg16Lys) conserved the activity except for Arg16Lys. We therefore examined the CD of these peptides. The CD spectra of peptides containing Tyr1Phe, Leu12Val or Arg16Lys substitution exhibited a CD signal similar to the wt peptide in KF/ phosphate buffer. The CD spectrum of the peptide containing Val14Leu substitution corresponded to a less ordered structure than the wt peptide, but clearly different from the peptide with Val14Ala substitution. These results indicate that the physicochemical properties of these particular AA are important for the folding of the peptide in the active form. Moreover, the fact that the peptide with Arg16Lys substitution folds in a similar way to wt peptide but exhibits lower activity indicates the importance of this residue for the interaction with other proteins or with the membrane.
CD measurements showed that phosphorylation of the peptide at positions 17 and 19 destabilized the b-sheet structure ( Figure 5D), in correlation with the decrease in cell-penetrating capacity. The CD analysis estimates that the phosphorylation at position 17 (wine line in Figure 5D) decreased the b-sheet content to 40% (from 56% content in the wt peptide) and increased the random coil content to 53% (36% in the wt peptide), while that at position 19 (magenta line in Figure 5D) decreased the b-sheet content to 34% and increased the random coil content to 60%. The phosphorylation of both sites (cyan line in Figure 5D) produced changes almost similar to those produced by single phosphorylation at position 17. Again, there is a correlation between the activity and the structure of peptide.
Together, these results reveal the importance of folding for the activity: both the peptides with scrambled or reversed sequence, as well as the peptide in which tyrosine 1 is replaced by alanine, do not cross the cell membranes and are not structured. In addition, the cell penetrating activity appears to correlate with a particular structural pattern. Moreover, all the mutated peptides, which exhibit a CD spectrum analogous to that of the wild type peptide (substitution with alanine at positions 21, 22, and 24 or functionally equivalent Tyr1Phe, Leu12Val and Arg16Lys substitutions) show a notable cell penetration activity. This was generally not the case for mutations that led to an ostensibly different CD spectrum, corresponding to different structural characteristics. These results, together with the absence of internalization of NFL-TBS.40-63 peptide composed of D-amino acids, and the saturating process of internalization in glioma cells demonstrated by FACS [13], suggest that a receptor may participate to its internalization. In that case, differences in uptake patterns of the peptide between glioma cells and normal cells could be due to differences of its interaction with a particular cell surface receptor expressed specifically in glioma. Our observations leave open the two possible options for crossing the glioblastome membrane, either directly via endocytosis or through its interaction with a cell receptor.
Another interesting feature of the NFL-TBS.40-63 peptide is its sensitivity to the environment (KF/phosphate buffer or SDS), and its remarkable stabilization by the presence of SDS. Such a characteristic is also observed for mutated forms of the NFL-TBS.40-63 peptide that still penetrate in cells. Moreover, the results suggest that Tyr1, Leu12 and Ser21 interact with tubulin as their tubulin binding activity decreases upon replacement by alanine or by residues with similar chemical properties, while the cell penetration is not compromised and the CD spectra are similar. These data suggest that a different conformation is required for anti-MT activity than for crossing the membrane. The alternative fold could be stabilized by tubulin binding.
It is noteworthy that the structure of the peptide is stabilized in the KF/phosphate buffer or in the presence of SDS. This indicates that some hydrophobic contacts in the peptide are important for the folding. The fact that the replacement of residues in the Cterminal part with alanine residue, which has distinct volume occupancy, destabilizes the structure suggests that this part is involved in the folding. On the other hand, replacement of residues in the N-terminal half by alanine promotes another folding.

Molecular modeling of NFL-TBS.40-63 peptide
Two extreme folding geometries were predicted by the PEP-FOLD program for the NFL-TBS.40-63 peptide ( Figure 6). In the first, the peptide is predominantly folded into a b-hairpin geometry extended over five consecutive amino acids (residues 1 to 5 are connected to residues 18 to 14 respectively), with a short Cterminal section presenting an a-helix character ( Figure 6A). The second geometry is characterized by a predominant a-helix folding ( Figure 6B). The two geometries present almost equivalent internal energies, with respectively values of 225.0 and 226.0 kcal.mol 21 . Each of them conserved its structural characteristics, in terms of secondary structure, over 2 ns of molecular dynamics simulation in explicit solvent. The structure with dominant a-helix character showed transient instabilities of one of its four helix turns situated at the N-ter side during 300 ps, after which the helix turn recovered stable interactions. Shifts in the bstrand pairing were also observed after 1 ns, indicating some structural variability. These results show that the two structures represent conformational substates with medium to low stability, which may coexist in solution. The mutations of residues Tyr4, Ser9, Ser11, Leu12, Ser13, Val14, Arg15, Ser17, and Ser19 into alanine were found to destabilize the b-hairpin part of the peptide and stabilize an a-helix conformation.
In conclusion, the combination of our immuno-cytochemical and structural analyses strongly suggests that the NFL-TBS.40-63 peptide presents specific folds when crossing the membrane as well as when it interacts with tubulin, but that these structures are not necessarily identical. As these conformations are affected each time that the properties to penetrate or to perturb the MT network are also affected, these results indicate a good structure/function relationships. Moreover, the peptide's ability to modify its structure in different conditions such as the cell interior versus hydrophobic environment appears a hallmark of the peptide activity. This structure/function relationship suggests importance of structure, not simple amino acid composition, for the activities of the peptide, including the selectivity to glioblastoma cells. Further studies will aim at characterizing the conformational equilibrium of the NFL-TBS.40-63 peptide in different conditions (temperature, solvent composition) and its structure when interacting with membranes or tubulin, using theoretical tools like molecular dynamics simulations and experimental structure determination.

Synthetic peptides
Biotinylated NFL-TBS.40-63 peptide corresponding to the tubulin-binding site located on NFL and biotinylated analogues (Tables 1 and 2) were synthesized at more than 95% purity by Millegen (Toulouse, France). Peptides were dissolved in distilled water at a concentration of 1 mM.

Immunocytochemistry
Cells were plated on cover slips and cultured in media containing biotinylated peptides (10 mM) for 6 hours. After being washed in Phosphate Buffered Saline (PBS), the cells were fixed for 10 min in 4% Paraformaldehyde (PFA in PBS), and washed 3 times in PBS. The cells were then incubated during 10 min in a 0.5% triton X-100 permeabilization solution, and washed 3 times in PBS. Next, they were incubated in a blocking solution (PBS 5% BSA) for 15 min, then incubated with mouse anti-ß-tubulin antibody (Sigma) 1/200 overnight. Tubulin and biotinylated peptides were localized using respectively Alexa 568 nm antimouse antibody and streptavidin Alexa 488 nm (Molecular Probes) 1/200 for 1 hour, followed by washing in PBS. The preparations were counterstained with 3 mM 496-diaminido-2phenylindole (DAPI; Sigma) for 5 min, washed twice with PBS, then cover slips were mounted with an antifading solution.
Observations were performed with an Olympus confocal microscope (BX50) using Fluoview.3.1. Software or a Leica DMI6000 inverted microscope and analyzed with Metamorph 7.1.7.0. software. We counted cells containing the peptide and cells displaying a destroyed MT network. Experiments were repeated at least three times, and a minimum of 200 cells was scored during each experiment.

Circular dichroism measurements
Circular dichroism (CD) spectra were measured in a J-810 CD spectrometer (Jasco, Japan) in step mode (bandwidth: 2 nm; interval: 0.1 nm; response time: 0.125 s). The spectra were averaged over 3 scans to increase the signal to noise ratio. To optimize the measurements, a 1 cm60.2 cm mini-quartz cell with four windows (Hellema, Germany) was used in such a way that the spectra from 260 to 200 nm were measured with a 1 cm pathlength and those between 220 and 185 nm with 0.2 cm pathlength. After normalization for pathlength (1 cm), the superimposition of two separated spectra in the 200-220 nm region was verified and the entire spectra reconstituted. The temperature of measurement was controlled with a Peltier-effect temperature controller, and was usually 20uC. The structural stability of peptides was studied by measuring the variation in CD signal at 220 nm (bandwidth: 10 nm; interval: 1 s; response time: 0.5 s; pathlength: 1 cm) when the temperature was increased by 1uC/min from 20 to 85uC. All the CD experiments were performed with 20 mM peptide under 3 buffer conditions: pure water, 2 mM SDS and 10 mM potassium phosphate pH 7.4 with 200 mM KF.
The secondary structures were estimated from the CD spectra with Jasco secondary structure estimation software. The spectra from 195 to 200 nm were decomposed to a-helix, b-sheet, b-turn and random coil elements using either Reed's [37] or Yang's reference set [38] as reference spectrum of each element as described [37,38]. We usually obtained better fits using Reed's reference.

Molecular modeling
Molecular modeling was carried out using the program PEP-FOLD (web server http://bioserv.rpbs.univ-paris-diderot.fr/PEP-FOLD/, [39]). PEP-FOLD predicts ''ab initio'' the folding characteristics of peptides comprising 9 to 25 amino-acids. It is based on structural alphabet SA letters describing the possible conformation of groups of four consecutive amino acids, which are selected and assembled via a genetic algorithm [40]. Structure reconstruction and energy evaluation rely on the coarse grain force field OPEP [41]. This program has been used successfully by several groups to predict biologically relevant peptide structures [42][43][44]. The sequence of the 24 amino acid peptide NFL-TBS.40-63 and all its alanine mutants were submitted to the PEP-FOLD server. Two protocols were used, either the default protocol or one using the PSI-PRED option which incorporates information from the homology-based PSI-PRED predictor [45]. Note that a recent update of the PEP-Fold program unifies the results of the two options, while supporting up to 35 amino acids.
Each one of the two main structures predicted for the wild type peptide (respectively represented in Figure 6A and Figure 6B) was submitted to a 2 ns, fully solvated, molecular dynamics simulation using the GROMACS software version 4.5.4 [46] and the OPLS/ AA-2001 force field [47]. The simulation boxes were filled with 5714 (for the structure in Figure 6A) or 3873 (structure in Figure 6B) SPC/E water molecules. Two chlorure ions were added to ensure electric neutrality. After energy minimization, each structure was submitted to a sequence of two 100 ps equilibration phases, respectively performed under canonical (NVT) and isothermal-isobaric (NPT) ensembles. Production was conducted in the NPT ensemble, using Particle-Mesh-Ewald treatment [48] for long range electrostatics interactions.

Statistical analysis
Data are presented as mean and Standard Error of the Mean (S.E.M.) (bars). Cell countings were analyzed by Student's t test using Prism version 3.00 (GraphPad Software, San Diego, CA). We distinguished three different levels of variation with respect to control: p,0.05; p,0.005; p,0.001 (see Figure 1).