Neuronal Galectin-4 is required for axon growth and for the organization of axonal membrane L1 delivery and clustering


  • Silvia Velasco,

    1. Membrane Biology and Axonal Repair Laboratory, Hospital Nacional de Parapléjicos (SESCAM), Finca La Peraleda s/n, Toledo, Spain
    Search for more papers by this author
    • These authors contributed equally to this work.
  • Natalia Díez-Revuelta,

    1. Membrane Biology and Axonal Repair Laboratory, Hospital Nacional de Parapléjicos (SESCAM), Finca La Peraleda s/n, Toledo, Spain
    Search for more papers by this author
    • These authors contributed equally to this work.
  • Teresa Hernández-Iglesias,

    1. Centro de Investigación del Cáncer-IBMCC USAL-CSIC, Campus Miguel de Unamuno, Salamanca, Spain
    Search for more papers by this author
  • Herbert Kaltner,

    1. Institut für Physiologische Chemie, Tierärztliche Fakultät, Ludwig-Maximilians-Universität, München, Germany
    Search for more papers by this author
  • Sabine André,

    1. Institut für Physiologische Chemie, Tierärztliche Fakultät, Ludwig-Maximilians-Universität, München, Germany
    Search for more papers by this author
  • Hans-Joachim Gabius,

    1. Institut für Physiologische Chemie, Tierärztliche Fakultät, Ludwig-Maximilians-Universität, München, Germany
    Search for more papers by this author
  • José Abad-Rodríguez

    Corresponding author
    1. Membrane Biology and Axonal Repair Laboratory, Hospital Nacional de Parapléjicos (SESCAM), Finca La Peraleda s/n, Toledo, Spain
    • Address correspondence and reprint requests to José Abad-Rodríguez, Membrane Biology and Axonal Repair Laboratory, Hospital Nacional de Parapléjicos (SESCAM), Finca La Peraleda s/n, E-45071 Toledo, Spain. E-mail:

    Search for more papers by this author


Axon membrane glycoproteins are essential for neuronal differentiation, although the mechanisms underlying their polarized sorting and organization are poorly understood. We describe here that galectin-4 (Gal-4), a lectin highly expressed in gastrointestinal tissues and involved in epithelial glycoprotein transport, is expressed by hippocampal and cortical neurons where it is sorted to discrete segments of the axonal membrane in a microtubule- and sulfatide-dependent manner. Gal-4 knockdown retards axon growth, an effect that can be rescued by recombinant Gal-4 addition. This Gal-4 reduction, as inhibition of sulfatide synthesis does, lowers the presence and clustered organization of axon growth-promoting molecule NCAM L1 at the axon membrane. Furthermore, we find that Gal-4 interacts with L1 by specifically binding to LacNAc branch ends of L1 N-glycans. Impairing the maturation of these N-glycans precludes Gal-4/L1 association resulting in a failure of L1 membrane cluster organization. In all, Gal-4 sorts to axon plasma membrane segments by binding to sulfatide-containing microtubule-associated carriers and being bivalent, it organizes the transport of L1, and likely other axonal glycoproteins, by attaching them to the carriers through their LacNAc termini. This mechanism would underlie L1 functional organization on the plasma membrane, required for proper axon growth.

Abbreviations used

Dulbecco's modified Eagle's culture medium




dithiobis (succinimidyl propionate)


foetal bovine serum


glial-fibrillary acidic protein


green fluorescent protein


minimal essential medium




phosphate-buffered saline



Ordered molecular transport processes are critical for distinct aspects of neuronal development. Once a single axon is determined among the early neurites, a polarized molecular traffic guides particular cargo into this selected neurite to sustain outgrowth and to maintain polarity (Craig and Banker 1994; Bradke and Dotti 1997). Although such process is critical for neuronal function, the mechanisms underlying protein sorting to the axon membrane are not fully understood. In accordance with the fact that most membrane proteins are glycosylated, the relevance of glycan moieties in polarized protein traffic is emerging, based on polysaccharide unsurpassed capacity for coding biological information, as embodied in the concept of the ‘sugar code’ (Gabius 2009). The neural cell adhesion molecule L1 is an axonal glycoprotein that regulates axon elongation/branching by homophilic interactions (Cheng and Lemmon 2004; Cheng et al. 2005). At the plasma membrane, changes in its association to detergent-resistant membrane fractions (DRM or rafts) can switch L1 function from promotion to inhibition of axon outgrowth (Kleene et al. 2001). This glycoprotein provides a graphic example that cargo transport and delivery can well be determined by glycosylation. In fact, besides modulating the extent of homophilic recognition, N-glycan chains of L1 and other axonal glycoproteins such as NCAM are required for their efficient sorting to the axon by a, so far, unknown mechanism (McFarlane et al. 2000; Zuber and Roth 2009).

Among the mammalian carbohydrate-binding proteins (lectins), members of the galectin family are able to target branch-end epitopes of glycan chains, and to regulate diverse processes such as cell adhesion, differentiation, growth, and migration (André et al. 1999; Villalobo et al. 2006; Gabius et al. 2011; Kaltner and Gabius 2012). The three types of structural galectin organization enable different ways of cross-linking counterreceptors, e.g. to establish lattice-like clusters. For instance, homodimers of prototype or chimera-type galectins can select identical glycan partners, while tandem-repeat-type galectins, which present two different carbohydrate recognition domains, may target a broader range of glycan epitopes (Kasai and Hirabayashi 1996). In this group, Galectin-4 (Gal-4), mainly expressed in mammalian gastrointestinal tracts, shows a remarkable sensitivity to clustered presentation of N-glycans with N-acetyllactosamine (LacNAc) epitopes at branch ends, making selection of distinct glycoproteins possible (Wu et al. 2004; Ideo et al. 2005; André et al. 2008, 2012; Morelle et al. 2009). Of relevance for cell polarity, it displays a particular pattern of specificity among glycosphingolipids, i.e. toward the 3′-sulfated galactose headgroup connected to a 2′-hydroxylated long-chain fatty acid of sulfatides (Delacour et al. 2005; Ideo et al. 2007; Kopitz et al. 2012; Ledeen et al. 2012). These binding characteristics underlie its pivotal role to stabilize rafts, to organize cargo selection, and to facilitate apical delivery in polarized enterocyte-like HT-29 cells (Hansen et al. 2001; Braccia et al. 2003; Stechly et al. 2009). Gal-4 has been detected in olfactory neurons (Storan et al. 2004), oligodendrocytes (Wei et al. 2007) and more recently, also in cortical neurons (Stancic et al. 2012). Intriguingly, L1 sorting to the axon of retinal cells is impaired by reducing presentation of LacNAc-bearing N-glycans in the same way as Gal-4-dependent apical delivery of the mucin-like membrane glycoprotein MUC1 is in HT-19 cells (McFarlane et al. 2000; Morelle et al. 2009; Stechly et al. 2009). Thus, the concept is shaped for Gal-4 to have a role in transport of axonal glycoproteins.

According to all the above, we aimed to define the expression and the distribution of Gal-4 in primary neurons, and to investigate its putative role in axon development through the modulation of axonal glycoproteins.

Materials and methods

Cell cultures and conditioned media

Rat embryo hippocampal and cortical neurons were obtained from E18 embryos of the Wistar strain and cultured as previously described (Goslin et al. 1998). In brief, cell suspensions were plated onto poly-l-lysine (PLL)-treated glass coverslips in Petri dishes (Nalgene Nunc, Rochester, NY, USA) containing minimal essential medium (MEM, Sigma-Aldrich, St. Louis, MO, USA) supplemented with 10% horse serum (Invitrogen, Carlsbad, CA, USA) (MEM-HS). After cell attachment, coverslips were transferred to Petri dishes containing neurobasal medium (Invitrogen) supplemented with 2% B-27 (Invitrogen) and 0.5 mM l-glutamine (Sigma). For biochemical purposes, hippocampal neurons from E18 rat embryos were plated on PLL-coated plastic Petri dishes containing MEM-HS medium, at a density of 2 × 106 cells/dish. Primary cultures of purified astrocytes were prepared from the neocortex of 1–3-day-old Wistar rats, as described elsewhere (McCarthy and de Vellis 1980). The cells were seeded on PLL-coated plastic flasks (Nunc) and maintained in Dulbecco's modified Eagle's culture medium (DMEM; Invitrogen) supplemented with 10% foetal bovine serum (FBS; Invitrogen), 1% penicillin-streptomycin (PS). To obtain conditioned media of cultured hippocampal neurons and astrocytes, their culture media were replaced by Locke's solution (154 mM NaCl, 5.6 mM KCl, 2.3 mM CaCl2, 1 mM MgCl2, 3.6 mM NaHCO3, 5 mM glucose, 5 mM Hepes pH 7.2) and the cells were further cultured for 8 h. Cell-conditioned medium was then collected, centrifuged and precipitated as follows: For TCA precipitation, conditioned media were incubated with sodium deoxycholate (25 mg/mL) for 5 min at 4°C, then proteins were precipitated with 10% TCA for 1 h at −20°C, and centrifuged at 4000 g for 20 min at 4°C. For methanol/chloroform precipitation, conditioned media were mixed with 4 volumes of methanol/ chloroform/water (4 : 1 : 3 by volume), and thoroughly mixed. The solution was centrifuged, the aqueous upper layer was discarded, and the lower phase was mixed again with 4 volumes of methanol and centrifuged. The air-dried pellets from both protein precipitation procedures were resuspended in Laemmli buffer and analysed by western blotting.

Rat pheochromocytoma PC12 cells were cultured on PLL-coated plastic Petri dishes containing DMEM, supplemented with 7.5% FBS, 7.5% HS and 2 mM glutamine. PC12 cells were seeded at 50–60% confluence and supplemented with nerve growth factor (50 ng/mL final concentration). Cells were treated with 1 mM 1-deoxymannojirimycin (DMJ; Tocris Bioscience, Bristol, UK) when indicated. DMJ was refreshed every 24 h. All cultures were maintained at 37°C in a humidified 5% CO2 atmosphere.

All procedures involving animals complied with international guidelines on the ethical use of animals in the European Communities Council Directive dated 24 November 1986 (86/6091EEC) and with the guidelines of the Institutional Ethics Committee of the Hospital Nacional de Parapléjicos (SESCAM).

Cell extracts, cross-linking and immunoprecipitation

Neuron or astrocyte culture dishes were washed with cold phosphate-buffered saline (PBS) and lysed for 20 min at 4°C in radioimmunoprecipitation assay buffer (25 mM Tris-HCl pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS) supplemented with protease inhibitors (protease inhibitor cocktail; Sigma). Protein extracts were centrifuged for 15 min at 16 000 g at 4°C. Supernatants were considered as total extracts. Protein content was quantified by the BCA method (Bio-Rad LifeScience, Hercules, CA, USA). For cross-linking assays, PC12 cells were incubated with the reversible homobifunctional cross-linker dithiobis (succinimidyl propionate) (DSP; Pierce, Thermo Scientific, Rockford, IL, USA), following the manufacturer's instructions. Reaction was quenched with Tris buffer (pH 7.5), and cells were scraped in cold lysis buffer (PBS, 5 mM EDTA, 0.5% Triton X-100). After centrifugation, supernatants were incubated with anti-Gal-4 or anti-L1 antibodies at 4°C for 18 h, and then precipitated with protein-A- or protein-G-Sepharose respectively. The immunoprecipitated complexes were separated by SDS-PAGE and subjected to detection by western blotting for L1 or Gal-4.

Antibodies and recombinant galectins

Mouse monoclonal: acetylated α-tubulin (Sigma-Aldrich, St. Louis, MO, USA); glial-fibrillary acidic protein (GFAP; Invitrogen); neurofilament H (Sigma), Tuj1 (Sigma). Goat polyclonal: anti-NCAM-L1 (Santa Cruz Biotechnology, Santa Cruz, CA, USA). Rabbit polyclonal anti-Gal-4, and anti-Gal-1 (in-house production). These two anti-galectin antibodies were tested against other commonly expressed members of this family, i.e. galectins-1, -2, -3, -7, -8, and -9, to exclude cross-reactivity using systematic screening by ELISA and western blot, and any positive reaction was abolished by affinity chromatography-based depletion, using the respective galectin as bead-immobilized ligand (Lensch et al. 2006). F-Actin was detected with Alexa-Fluor-568-conjugated phalloidin (Molecular Probes, Eugene, OR, USA). Appropriate anti-species Alexa-Fluor 488-, 594- and 647- conjugated antibodies were used for immunocytochemistry (ICC) assays. Respective secondary horseradish peroxidase-conjugated antibodies were used for western blotting (all from Molecular Probes). Nuclei were labelled with 4′-6-Diamidino-2-phenylindole (Sigma). Human Gal-4 cDNA was cloned, then the protein was obtained by recombinant production, purified by affinity chromatography on lactosylated Sepharose 4B as crucial step, and controlled for purity by one- and two-dimensional gel electrophoresis and mass spectrometry, and for activity by solid-phase and cell binding assays (André et al. 2008, 2012; Martín-Santamaría et al. 2011).


For ICC analysis, cells were fixed with 4% paraformaldehyde for 15 min at 22°C, aldehyde groups were quenched with 50 mM ammonium chloride for 5 min. When indicated, membrane permeabilization was performed with PBS containing 0.1% Triton X-100 for another period of 5 min. Non-specific binding sites for protein were saturated with blocking solution 2% FBS, 2% bovine serum albumin (Sigma), 0.2% gelatine 80–100 blooms (Panreac, Barcelona, Spain) in PBS for 1 h at 22°C, cells were then incubated with primary antibodies for 1 h at 22°C or overnight at 4°C. After extensive washing with PBS, cells were incubated with fluorophore-coupled anti-species antibodies for 1 h at 22°C. Fluorescent phalloidin for F-actin staining and secondary antibodies were then incubated together with the cells. For simultaneous detection of membrane-associated Gal-4 and acetylated α-tubulin, a first-step ICC for Gal-4 detection was performed under non-permeabilizing conditions. Cells were then fixed again, and a second ICC for presence of acetylated α-tubulin with permeabilized cells followed to complete staining. Coverslips were then mounted on glass slides with Mowiol (Calbiochem-Merk, New Jersey, USA) using DABCO (Sigma) as anti-fade agent. Images were acquired either on a confocal microscope (Leica SP5, Leica Microsystems GmgH, Wetzlar, Germany) or/and an epifluorescence microscope (Leica DM5000B).

Antibody-induced patching

Cultured living neurons were washed with PBS and then incubated for 1 h at 12°C in CO2-free MEM medium, with anti-L1 alone or together with anti-Gal-4 primary antibodies. After brief washing, the cells were incubated under the same conditions with fluorophore-coupled anti-species antibodies to induce molecular patching. Cells were then fixed with 4% paraformaldehyde and mounted. Photomicrographs were taken on a confocal microscope (Leica SP5) and analysed with the ImageJ software (NIH Bethesda, MD, USA). For quantification of L1 cluster density and size, the respective channel was separated and thresholded to define positive signal clusters using the ImageJ software. Cluster number and area were then measured, compared with controls, and subjected to statistical analysis. Ten fields from ten different pictures were analysed in three different experiments, for each treatment modality.


Wistar adult rats (Harlan, Indianapolis, IN, USA) were anesthetized with pentobarbital (40 μg/g) and immediately perfused with saline followed by fixative solution of 4% paraformaldehyde. Brains were dissected out and soaked in fixative solution at 4°C until embedding. Paraffin-embedded tissues were processed into coronal or sagittal 10 μm-thick sections by using a microtome. Paraffin-embedded sections were de-paraffinated, rehydrated and exposed 1 min to 99°C for epitope retrieval. Sections were then blocked with PBS containing 0.3% Triton X-100 and 2% bovine serum albumin, and incubated with primary antibodies for 1 h at 22°C. After extensive washing with PBS, sections were incubated with fluorophore-coupled anti-species antibodies for 40 min at 22°C, thoroughly washed again, and mounted with 4′-6-Diamidino-2-phenylindole-containing anti-fade agent (Master Diagnostica, Granada, Spain).

Metabolic inhibition of sulfation, partial extraction of membrane cholesterol and disruption of actin/tubulin cytoskeleton

For bringing down cellular level of sulfation, neurons were incubated in medium containing 75 mM sodium chlorate (Sigma) for 48 h, and then processed for Gal-4 and F-actin ICC analyses. To measure sulfatide levels, neurons were scraped in water, ultrasonicated and centrifuged at low speed. Nuclear pellets were discarded and proteins in the supernatant were measured by BCA method. Lipids in the supernatant were extracted with chloroform and methanol (C/M/W 4 : 8 : 3 by vol.), and centrifuged to pellet precipitated proteins. Lipid extracts were dried and resuspended in C/M (2 : 1). Samples and sulfatide standards were then loaded on aluminium-backed silica TLC plates and run on C/M/W (65 : 25 : 4 by vol.). Plates were then sprayed with 7% sulphuric acid in methanol, and lipid bands were visualized by heating the plates in an oven at 150°C for 30 min. Bands were quantified by densitometry after background subtraction (NIH image). Partial depletion of membrane cholesterol was performed with 500 mM MβC (Sigma) solution as previously described (Abad-Rodríguez et al. 2004). Cholesterol was measured either in control or treated cell membrane extracts, by cholesterol oxidase colorimetric method (BioSystems S.A., Barcelona, Spain) following manufacturer instructions. Neurons at 3 div were incubated for 2 h at 37°C either with 7 μM nocodazole (Sigma) to depolymerize microtubules, or with 2.5 μM lantrunculin A (Sigma) to disrupt actin microfilaments. Cells treated with nocodazole were labelled for Gal-4 and phalloidin; cells treated with lantrunculin were labelled for Gal-4 and acetylated α-tubulin.

Gal-4 knockdown by small interference RNA (siRNA) and rescue by recombinant Gal-4

For Gal-4 silencing the following synthetic oligonucleotides (Sigma) were tested: siRNA1: GCACTACAAGGTCGTGGTA(dT)(dT); siRNA2: GAACTCCCTTCTA CGAATA(dT)(dT); siRNA3: TTCGCAACAGCTATATGA(dT)(dT). Neurons were transfected in suspension using the Nucleofector system (Amaxa, Gaithersburg, MD, USA) according to the manufacturer's instructions. As control, parallel nucleofections with Mission® siRNA Universal Negative Control-1 (Sigma) were carried out. A plasmid encoding the Green Fluorescent Protein (pMAX-GFP) was co-nucleofected as transfection reporter. 48 h post-nucleofection, cells were harvested and processed for biochemical analysis, or fixed for ICC and morphological evaluation. For rescue experiments, interfered and control cultures were supplemented with recombinant Gal-4 (25 μg/mL) from the moment of coverslip transfer to neurobasal medium (see Cell Cultures above). The efficiency of siRNA1, siRNA2 and siRNA3 was assessed by comparing the reduction of Gal-4 levels in neurons cultured for 24, 48 or 72 h. The most relevant silencing results were obtained using siRNA1 targeting the nucleotide region 330–348 of Gal-4 within the first carbohydrate recognition domain. Gal-4 expression knock down was biochemically ascertained as follows: 2 div neurons were detached in Ca2+/Mg2+ free HBSS containing 25 mM HEPES, 5 mM EDTA, 2% FBS, pH 7.0, and sorted in a FACS Aria cell sorter (BD Biosciences, Franklin Lakes, NJ, USA) to obtain 90–95% pure GFP+ populations. Equal amounts of GFP+ cells were lysed, and total extracts were analysed by WB for Gal-4. Anti-acetylated α-tubulin antibody served as loading control. For morphological analysis of Gal-4 knockdown neurons, cells were fixed and immunolabeled for Gal-4 and anti-acetylated α-tubulin, as detailed above. Random images (10 fields per coverslip) were obtained using an epifluorescence microscope (Leica DM5000B), and cell parameters were measured up to a total of 30 cells per experiment (four experiments). Only cells in clusters or those with inter-crossed axons were excluded from analysis because of the difficulty to unambiguously track the axon trajectory. The fluorescence intensity reflecting the presence of axonal Gal-4 in the siRNA experiments was quantitated using ImageJ software. GFP-positive neurons showing on average Gal-4 fluorescence intensity lower than 30% of control average were considered knocked-down and included in the analysis of neurite length. Neurite length was measured using ImageJ software, and the results were subjected to Student's t-test based statistical analysis.


Hippocampal and cortical neurons express Gal-4 in culture and in situ

Standard immunofluorescence methods including permeabilization show homogeneous Gal-4 staining at both axonal and somatodendritic compartments of hippocampal (Fig. 1a, arrowheads and arrows respectively) and cortical (Figure S1) neurons. Neuronal Gal-4 expression is confirmed by western blots (Wb) of cultured rat hippocampal neurons total lysates, probed with anti-Gal-4-specific antibodies (Fig. 1b, upper line, TL). Neuron cultures frequently contain a residual population of astrocytes (< 10%) and even though galectins are released to the medium in different cell systems (Nickel 2003), remnant astrocytes are not responsible for Gal-4 presence in neuronal cultures. In fact, no Gal-4 is detected in lysates from rat cortical astrocytes (Fig. 1b, central line, TL), whereas Gal-1 presence, used as positive control, is readily observed (Fig. 1b, bottom line, TL). This is further corroborated by the absence of Gal-4-dependent immunolabeling in GFAP-positive astrocytes (Fig. 1c, upper panels), in contrast to the strong Gal-1 control signal (Fig. 1c, bottom panels). In close agreement with our results in culture, Gal-4 is detected in situ on different populations of hippocampal (Fig. 1d, arrows) and cortical (Fig. 1e, arrows) neurons by double immunolabeling for Gal-4 and axonal markers on adult rat brain sections. A defined staining along the axons is evidenced (Fig. 1d and e, insets, arrowheads).

Figure 1.

Cortical and hippocampal neurons express Gal-4. (a) Hippocampal neurons cultured for 72 h were immunolabeled for Gal-4 (green) and stained with Texas-red-conjugated phalloidin for F-actin detection (red). ICC analysis was performed on Triton X-100-permeabilized cells. Gal-4 distribution in axons and dendrites is indicated by arrowheads and arrows respectively. (b) Wb analyses were performed on total cell lysates (TL) or conditioned media (CM) of hippocampal neurons (HN) and astrocytes (Ast), using anti-Gal-4 and anti-Gal-1 antibodies. rGal-4 and rGal-1 stand for recombinantly produced human galectins, used as controls. (c) Astrocytes stained for Gal-4 or Gal-1 and counter-stained for the astrocytic marker Glial-Fibrillary Acidic Protein (GFAP). (d, e) Paraffin-embedded sections of adult rat brain double-stained for Gal-4 (green) and an axon marker (either NFH or Tuj1) (red), andcounter-stained with 4′-6-Diamidino-2-phenylindole (blue). Gal-4 presence was detected in hippocampal (d) and cortical (e) neurons (arrows) and axon tracts (arrowheads within the selected region defined by the boxes after magnification). Scale bars 25 μm.

Gal-4 is sorted to the axon membrane at an early stage of development and localizes to discrete axon tracts of mature cultured neurons

Considering that Gal-4 is associated with the outer leaflet of polarized intestinal epithelial cell membrane (Stechly et al. 2009), we investigated if this could be the case also in polarized neurons by studying the galectin distribution at different stages of hippocampal neuron development in vitro (Dotti et al. 1988), using non-permeabilizing ICC. Gal-4 is already detected at the early stage 1 (Fig. 2a, stage 1), prior to the morphological specification of the axon. During this period, Gal-4 presence is seen homogeneously in permeabilized cells (Fig. 2a, inset 2, arrows). Similar immunocytochemical processing of non-permeabilized neurons reveal Gal-4 presence only in the proximal tract of a single neurite (Fig. 2a, inset 1, arrowhead). Consistently, when neurons start to morphologically define their axons, axon membrane-restricted localization of Gal-4 is observed under non-permeabilizing conditions (Fig. 2a, stage 2, 2+, inset 3, arrowhead), while it is detected over the whole permeabilized cell (Fig. 2a, stage 2, 2+, inset 4). This distribution remains unaltered later on, during stage 3 (Fig. 2, stage 3), though axonal membrane Gal-4 in this case starts to display a characteristic discontinuous labelling only on discrete tracts along the axon (Fig. 2a, inset 5 and Fig. 2b, arrowheads). This unusual discontinuous Gal-4 presentation on axonal membrane could reflect some undetected axon damage. To evaluate this possibility, non-permeabilized neuronal cultures were stained for Gal-4, permeabilized, and counter-stained for acetylated α-tubulin, an axonal microtubule marker. This staining shows that the structure of the axon remains intact both at and between the Gal-4-labeled tracts (Fig. 2b, arrowheads), indicating that the segmented Gal-4 labelling is not due to structural damage artefacts.

Figure 2.

Gal-4 detection in cultured neurons at successive stages of development. Hippocampal neurons were fixed at different time points to define the expression status at different stages of neuronal development in culture. (a) Non-permeabilized and Triton X-100-permeabilized cells were immunolabeled for Gal-4 (green) and F-actin (red) (upper panels). Details of Gal-4 localization in boxed areas (1–6) are presented at increased level of resolution in the insets of the bottom panels. Axonal and somatodendritic Gal-4 is indicated with arrows and arrowheads respectively. Gal-4 is sorted to the axon membrane during early stages (insets 1 and 3) and becomes distributed in axon membrane segments when more matured (inset 5). (b) Non-permeabilized stage 3 neurons stained for Gal-4 (green) and acetylated α-tubulin (red). Axonal microtubule cytoskeleton does not show any structural alterations. Scale bars 25 μm.

Gal-4 knockdown by siRNA retards axon growth

The observation of early axonal sorting of Gal-4 led us to look for possible effects of lowering galectin levels in axon development. To reduce cellular Gal-4 production, hippocampal neurons were nucleofected with specific siRNA, using a universal negative control siRNA (UNCsiRNA) as reference. For rescue experiments, siRNA-transfected cultures were supplemented with recombinant Gal-4. In all cases neurons were co-transfected with a GFP-encoding cDNA to ascertain efficiency of vector uptake. The siRNA-transfected neurons, indeed, show a very low level of Gal-4 labelling by immunofluorescence (Fig. 3a, siRNA, Gal4, arrows), while UNCsiRNA-transfected (Fig. 3a, UNCsiRNA, Gal-4, arrowheads) and non-transfected (Fig. 3a, siRNA, Gal4, arrowheads) neurons display the characteristic segmented Gal-4 axon labelling. This decreased Gal-4 expression was corroborated biochemically by Wb densitometric analysis, showing a 78 ± 22% average reduction with respect to non-transfected or UNCsiRNA-transfected neurons (Fig. 3b). Importantly, the axons of siRNA-transfected neurons are significantly shorter (85.4 ± 33.7 μm) than those from UNCsiRNA-transfected neurons (184.5 ± 41.3 μm) (Fig. 3c, black bars, *p < 0.0001 Student′s t-test), while no differences are observed for other neurites (Fig. 3c, pale-gray bars). The axon growth retardation observed in siRNA-transfected neurons was rescued by recombinant Gal-4 addition. In this case, neurons displayed axons whose average length (182.6 ± 41.2 μm) was similar to those of UNCsiRNA-transfected neurons, and significantly longer than those of siRNA-transfected neurons (Fig. 3a bottom panels, and Fig. 3c, siRNA, dark-gray bar, *p < 0.0001 Student′s t-test). These results indicate that Gal-4 synthesis is required for normal axon development, and that its absence notably retards axon elongation.

Figure 3.

Gal-4 knockdown retards axon growth. Gal-4 production in neurons was knocked down by nucleofection with siRNA1 (see methods). As control, Mission® siRNA Universal Negative Control-1 (UNCsiRNA) was used. In both cases, a plasmid encoding the Green Fluorescent Protein (pMAX-GFP) was co-nucleofected as transfection reporter. Recombinant Gal-4 (25 μg/mL; rGal-4) was added to the cultures in the case of rescue experiments (a) 72 h post-transfection, cells were fixed and immunolabeled with anti-Gal-4 antibodies (red), then cells were permeabilized and immunolabeled for α-tubulin (blue). siRNA1-transfected cells show low level of Gal-4 labelling, and they display shorter axons (arrows in upper panels) than controls (central panels, and arrowheads in upper panels). Presence of rGal-4 restores axon length to control values (bottom panels). (b) GFP-positive neurons were detached, sorted, harvested and processed for Wb analysis. siRNA1-tranfected neurons express very low levels of Gal-4 (siRNA), when compared with UNCsiRNA-transfected or control total lysates (TL). Tubulin (Tub) is shown as loading control. Densitometry quantification of Gal-4 in Wb is shown in the graph as average percentage of controls in three independent experiments. (c) Random images (10 fields per coverslip) were taken from cultures of four experiments as the one shown in (a). GFP-positive neurons with Gal-4 fluorescence intensity lower than 30% compared with controls were considered knocked-down and included in the analysis of neurite length (30 cells/experiment; four experiments). In the case of rescue experiments, only GFP expression was considered, as addition of rGal-4 masked endogenous Gal-4 decrease. Bar graph represents average length of axons and other neurites, of Gal-4 knocked down (siRNA1-Gal-4), and control (UNCsiRNA) neurons, in the presence or absence of rGal-4. Axons of Gal-4 knocked down neurons are significantly shorter than those of control cells, and the effect is rescued by rGal-4 addition. No effect is observed in other neurites (*p < 0.0001, Student′s t-test). Scale bar 25 μm.

Metabolic inhibition of sulfation impairs axonal Gal-4 membrane sorting

As Gal-4 binds with high affinity not only to sulfatide but also to cholesterol 3-sulfate (Ideo et al. 2007), a sulfated lipid present in cholesterol-rich membrane microdomains, we hypothesized it to be involved in Gal-4 sorting to the axon surface. To test this, neuronal membrane cholesterol was partially extracted from the membrane of cultured neurons by incubation with methyl-β-cyclodextrin (MβC), using conditions that perturb raft organization (Abad-Rodríguez et al. 2004). This partial cholesterol depletion (45 ± 4.7%) does not modify the segmented profile of axonal Gal-4 (Fig. 4a, arrows). This indicates that the galectin's axonal distribution is independent of total membrane cholesterol, including its sulfated form that is extracted from biological membranes by MβC at the same rate as unsubstituted cholesterol (Visconti et al. 1999). The possibility that Gal-4 binding to sulfatides could play a role in establishing the axonal distribution of Gal-4 was studied next, by impairing metabolic sulfation with sodium chlorate (Baeuerle and Huttner 1986; Garcia et al. 2007). This treatment that reduces neuron sulfatide content to 32 ± 3% of control values (Figure S3b), almost completely precludes Gal-4 localization in the axonal membrane, as shown by ICC of non-permeabilized cells (Fig. 4b, NaClO3, arrows), while untreated cultures display segmented Gal-4 labelling (Fig. 4b, control, arrows). Importantly, Gal-4 is homogeneously detected in axonal and somatodendritic compartments upon NaClO3 treatment when immunolabeled after cell membrane permeabilization, and no remarkable differences are observed when compared with untreated controls (Figure S3). This result demonstrates that Gal-4 sorting to the axon membrane is affected by inhibition of sulfation, while galectin synthesis proceeds normally.

Figure 4.

Inhibition of sulfation and microtubule depolymerisation, but not cholesterol extraction, prevents segmented Gal-4 distribution in axons. Non-permeabilized cells were labelled for presence of Gal-4 (green) and F-actin (red). (a) Neuron membrane cholesterol was partially depleted by treatment with MβC for 48 h. Pattern of Gal–4 axon labelling was not altered hereby, when compared with controls (insets, arrows). (b) Neurons incubated for 48 h with the metabolic sulfation inhibitor sodium chlorate (NaClO3, 75 mM). The typical segmented Gal-4 presence in control axons (upper insets, arrows) was precluded in axons of treated neurons (lower insets, arrows). (c) Microtubules of cultured neurons were depolymerized by incubation with Nocodazol for 2 h. The typical segmented Gal-4 profile of control axons (Control inset, arrows) is entirely abolished in axons of treated neurons (Nocodazol inset, arrowheads). Scale bars 25 μm.

Axonal Gal-4 membrane distribution requires intact microtubule cytoskeleton

In intestinal epithelial cells, sulfatide-containing post-Golgi carrier vesicles have been proposed to traffic towards their target domains along cytoskeletal structures (Weisz and Rodriguez-Boulan 2009). We hypothesized that the structural state of the cytoskeleton could influence the axonal distribution of Gal-4. In fact, segmented Gal-4 labelling on the axon membrane (Fig. 4c, arrows) is nearly undetectable after nocodazol-induced de-polymerization of neuron microtubules (Fig. 4c, arrowheads), while F-actin de-polymerization in lantrunculin-treated neurons does not affect Gal-4 axonal distribution (Figure S4). These results indicate that Gal-4 sorting to the axonal membrane requires an intact microtubule cytoskeleton, suggesting that Gal-4 could be directed towards specific locations at the axonal plasma membrane by its linkage to microtubule-associated carriers.

Gal-4 interacts with L1 and its knockdown reduces L1 cluster number and size in axon membrane

The observed retardation of axon growth by low-level Gal-4 production, and the particular Gal-4 distribution at the axon plasma membrane, suggested that this galectin could modulate where and how axon elongation-associated molecules are present at the axon surface to regulate their function. Considering that Gal-4, based on features of their N-glycans, selects distinct glycoproteins as cargo for their routing to the apical domain of intestinal cells (Morelle et al. 2009; Stechly et al. 2009), we assumed that a similar mechanism could function to deliver and organize glycoproteins at the axon plasma membrane. The N-glycosylated, axon growth-inducing glycoprotein L1 was selected as a suitable model candidate to answer this question.

To assess Gal-4/L1 interaction, both proteins were patched on the membrane of living neurons by sequential incubations with primary and secondary antibodies (Abad-Rodríguez et al. 2004; Díez-Revuelta et al. 2010). Axon tracts containing dense population of Gal-4 clusters could be observed, as well as the frequent presence of Gal-4/L1 double-stained clusters (Fig. 5a, arrowheads), supporting the concept of a Gal-4/L1 interaction. To provide further evidence and to infer an interaction by galectin binding to L1 N-glycans, PC12 cell cultures were treated or not with the α-mannosidase I inhibitor DMJ, which blocks the incorporation of hybrid- and complex-type N-glycans presenting LacNAc termini. In addition, a chemical cross-linker was applied to form conjugates between (glyco) protein complexes, cleavable by reducing agents in Laemmli buffer. Anti-L1 antibodies did co-immunoprecipitate Gal-4 (Fig. 5b, IP L1, Wb Gal-4) and importantly, DMJ treatment notably reduced the amount of immunoprecipitated Gal-4 (Fig. 5b, +DMJ IP L1, Wb Gal-4). Furthermore, inverse IP with anti-Gal-4 antibodies pulls down L1 (Fig. 5b, IP Gal-4, Wb L1) and consistently, DMJ treatment diminishes extent of L1 co-IP (Fig. 5b, +DMJ IP Gal-4, Wb IL1). The co-IP results revealed that Gal-4 and L1 form complexes, and the fact that inhibition by DMJ impairs this interaction indicates that LacNAc termini of hybrid/complex-type N-glycans in L1 are relevant as docking sites of Gal-4.

Figure 5.

Gal-4 interacts with L1, and its knockdown reduces occurrence of L1 clusters in axon membrane. (a) Living neurons in culture were incubated for 1 h at 12°C with mixed anti-L1 and anti-Gal-4 antibodies, then exposed to fluorophore-labelled anti-species antibodies to induce surface patching. Axon tracts containing a high density of Gal-4 clusters are observed, as well as Gal-4/L1 double-stained clusters (arrowheads). (b) PC12 cell cultures, incubated in the absence or presence of 1–deoxymannojirimycin (DMJ), were exposed to a reversible chemical cross-linker. Proteins in cell extracts were then immunoprecipitated with anti-L1 or anti-Gal-4 antibodies, and the immunoprecipitated material was probed, respectively, for the presence of Gal-4 or L1 by Wb. Anti-L1 antibodies pull down Gal-4 from extracts of cells not treated with the glycosylation inhibitor (IP L1) to a higher extent than when cells were treated (+DMJ IP L1). The reversed IP produces coherent results, as anti-G4 antibodies pull down L1 from control extracts (IPGal-4) to a much higher extent than in DMJ-treated cells (+DMJ IPGal-4). (c) Neurons nucleofected with either siRNA Gal-4 or control UNCsiRNA were processed to obtain L1 patching as in (a), then fixed and immunolabeled for acetylated α-tubulin (Tub). GFP was used as transfection reporter. L1 clusters in control axons are arranged in groups (UNCsiRNA, inset, arrows), while their appearance was less abundant and rather sparse in axons of interfered neurons (siRNA Gal-4, inset, arrows). (d) Quantitative analysis of L1 clusters in Gal-4 knockdown neurons. The graph on the left displays the percentage of appearance of patches in axons with respect to total cell clusters. The graph on the right displays the average area of axon clusters. Ten fields from ten different photomicrographs were analysed in three different experiments for each experimental set-up (*p < 0.05, Student's t-test). Scale bars 25 μm.

Given that L1 cluster formation in the axon membrane is required for the induction of axon elongation (Nakai 2002), the specific sorting of Gal-4 to the axon membrane, the retarded axon growth upon Gal-4 interference, and the interaction of Gal-4 with L1, prompted us to test whether L1 clustering could be modified in the absence of Gal-4. L1 clustering was measured and compared using the patching technique on living neurons transfected either with Gal-4 siRNA or UNCsiRNA. In UNCsiRNA-transfected cells, L1 clusters were relatively enriched in axons and grouped at restricted zones of the axon membrane (Fig. 5c, UNCsiRNA, arrows). In contrast, L1 clusters in siRNA-transfected neurons appeared sparse along the axon (Fig. 5c, siRNA Gal-4, arrows), they were significantly less abundant than in controls (Fig. 5d, left graph) and most importantly, their average size was significantly smaller (Fig. 5d, right graph). Taking into account that Gal-4 interference does not substantially modify the level of L1 expression (Figure S5b), these results indicate that a certain level of Gal-4 protein is required for proper axonal L1 clustering, and that this process is mediated by the interaction of Gal-4 with LacNAc termini of N-glycans presented by L1.

Inhibition of N-glycan processing impairs L1 clustering but not the segmented Gal-4 distribution on the axonal membrane

In this scenario, we hypothesized that low levels of Gal-4 or perturbed L1 N-glycosylation would impede the correct functional disposition of L1 at the axonal membrane, though, in the last case, Gal-4 localization in the axon would not be affected. In fact, DMJ-treated neurons display a segmented Gal-4 presence on the axonal membrane (Fig. 6a, DMJ, arrows) similar to control neurons (Fig. 6a, Control, arrows). In spite of this regular presence of Gal-4 in the axon, L1 abundance and grouping at restricted zones of axonal membrane (Fig. 6b, Control, arrows) are altered upon DMJ treatment, showing highly scattered disposition (Fig. 6b, DMJ, arrows) and a reduced number of L1 clusters (Fig. 6c, DMJ, left graph) which, in addition, displayed a significantly smaller average area (Fig. 6c, DMJ, right graph). Similar effects are induced by sulfation inhibition with NaClO3. L1 clusters appear less grouped than in controls (Fig. 6b, NaClO3, arrows), they are much less abundant (Fig. 6c, NaClO3, left graph), and their average size is significantly reduced (Fig. 6c, NaClO3, right graph). In contrast to DMJ treatment, in this case normal axonal Gal-4 distribution is also precluded, as shown before (Fig. 4b). Together with the fact that even in the presence of the regular Gal-4 profile the impairment of N-glycan processing prevents L1 normal spatial organization at the axon membrane, this result indicates that Gal-4 binding to both sulfatides and LacNAc termini of N-glycans in L1 is necessary for correct L1 placement, clustering and function.

Figure 6.

Reduction of number of LacNAc termini in N-glycans reduces L1 clustering but not Gal-4 axon sorting, while inhibition of sulfation impairs both processes. (a) Control and 1–deoxymannojirimycin (DMJ)-treated non-permeabilized hippocampal neurons stained for Gal-4 (green) and acetylated α-tubulin (red). Gal-4 segmented presence at the axon membrane is not altered by DMJ treatment. (b) Control, DMJ- and NaClO3-treated hippocampal neurons were processed to obtain L1 patching as in Fig. 5a, then fixed and immunolabeled for acetylated α-tubulin (Tub). L1 clusters in control axons are often arranged in groups (Control inset, arrows), while in treated neurons they are less abundant and appeared rather scattered (DMJ and NaClO3 insets, arrows). (c) Comparative analysis of L1 clusters in treated and untreated neurons. The graph on the left displays the percentage of appearance of patches in axons with respect to total cell clusters. The graph on the right displays the average area of axon clusters. Between 10 and 15 fields from 10 different photomicrographs were analysed in three different experiments for each pharmacological treatment (*p < 0.05, **p < 0.005, Student's t-test). Scale bars 25 μm.


Gal-4 is a tandem-repeat-type protein, in which the peptide linker connects two different carbohydrate recognition domains, enabling the lectin to cross-link distinct counterreceptor pairs. Binding to sulfatide and clustered N-glycans of distinct glycoproteins underlies its essential role in epithelial polarity maintenance by regulating apical membrane trafficking in enterocyte-like cells (Delacour et al. 2005; Stechly et al. 2009). Polarization is one of the main features needed for neuronal development, which requires polarized transport of biological material both, to determine the axon among undifferentiated early neurites, and to maintain its outgrowth and function as the neuron matures. In addition, given the prominent presence of sulfatides in the nervous system, and the suggested presence of Gal-4 in olfactory (Storan et al. 2004), cortical and dorsal root ganglia neurons (Wei et al. 2007; Stancic et al. 2012), we postulated that Gal-4 could play a role in neuronal differentiation.

We show that hippocampal and cortical neurons express Gal-4 in two differentially arranged spatial patterns (Fig. 1 and Figure S1), i.e. an intracellular pool, rather common for other galectins, that is homogeneously distributed over the cell and becomes detectable upon membrane permeabilization, and a previously unobserved second mode of presentation, in which Gal-4 is associated to segments along the axonal surface that can only be evidenced under non-permeabilizing conditions (Fig. 2). This particular surface staining can arise, in principle, from a putative binding of Gal-4 from exogenous sources such as residual astrocytes, typically present in neuronal cultures, given that they synthesize and secrete two other galectins (Gal-1 and Gal-3). In agreement with the negligible glial expression of Gal-4 previously reported (Stancic et al. 2012), we found that astrocytes in culture do not produce significant amounts of Gal-4, unequivocally confirming the neuronal origin of Gal-4 (Fig. 1b and c). Beyond the mere expression in neurons, Gal-4 is sorted to the axonal membrane already at early stages of neuronal development (Fig. 2), pointing to functional implications in axon formation and outgrowth. The involvement of Gal-4 in axon development was ascertained by the axon growth retardation observed upon Gal-4 knockdown with siRNA, and further supported by rescue of fairly normal growth by the addition of recombinant Gal-4 to the interfered cultures (Fig. 3). This rescue with exogenously added galectin opened the possibility that not only sorting of Gal4 to the axon, but direct binding to ligands already existing on the membrane could be involved. To set the conditions for rescue, we performed a series of preliminary experiments. First, to be sure whether exogenous rGal-4 was incorporated in cultured neurons, they were shortly incubated with FITC-conjugated rGal-4, and fixed at different times. FITC-rGal-4 was endocytosed and localized to perinuclear recycling endosomes (Rab11 positive) within 2 h (Figure S5a). After 4 h, few L1-FITC co-localizing clusters could be detected on axon membrane, suggesting that some of the exogenous Gal4 was reaching the membrane together with L1 (Figure S5a, arrows). In a second control test, neurons were pre-incubated in suspension with FITC-rGal4 to assure that all exogenous galectin was endocytosed only by the cell soma. After 48 h membrane Gal4 labeling co-localized with FITC-signal in axon segments, supporting the idea that endocytosed Gal4 traffics along the axon and reaches restricted zones of its membrane (Figure S5b, arrows). In spite of these results, we cannot completely rule out the possibility that in our rescue experiments, where rGal4 is continuously incubated for 48 h, local binding and/or local recycling at the axon membrane could play a role. Nevertheless, in consistence with our working model, we assessed that a part of the endocytosed galectin traffics along the axon and reaches restricted segments of the axon membrane.

It is well established that the transport of a plethora of essential molecules for axonal development and function depends on microtubule organization (Witte and Bradke 2008; Witte et al. 2008). Indeed, Gal-4 positioning requires intact microtubules, as indicated by the absence of axonal membrane Gal-4 upon microtubule de-polymerization by nocodazol (Fig. 4c). In contrast, perturbation of filamentous actin with lantrunculin has no detectable effects on axonal presence of Gal-4, suggesting that the actin cytoskeleton is not essentially involved in Gal-4 sorting (Figure S4). It was thus likely that Gal-4 could be transported associated to carrier vesicles along axonal microtubules towards its target sites in the axon membrane. As a matter of fact, the precedent of the enterocyte-like HT-29 cells documents that Gal-4 is capable to associate to raft-containing carrier vesicles by high-affinity binding to sulfatides that insert their up to C24-long lipophilic ceramide portion into raft domains. Following our assumption that a similar mechanism could be operating in neurons, we describe that metabolic inhibition of sulfation by sodium chlorate, an inhibitor of the formation of the sulfate group donor 3′-phosphoadenosine 5′-phosphosulfate (Baeuerle and Huttner 1986; Garcia et al. 2007), impairs the localization of Gal-4 in axonal membrane (Fig. 4b), with no influence on galectin production (Figure S3). Considering that sodium chlorate treatment reduces sufation of both, proteins and lipids, we cannot exclude definitively that its effect on neuronal Gal-4 localization could involve defects on protein sulfation. Nevertheless, Gal-4 shows low affinity for sulfated proteins as heparan- or chondroitin-sulfate, while it displays high binding affinity for sulfatides. On the other hand, participation of Gal-4 binding to other sulfated lipids, e.g. cholesterol 3-sulfate (Ideo et al. 2007), was ruled out, as cholesterol extraction with MβC (Abad-Rodríguez et al. 2004; Díez-Revuelta et al. 2010), that sequesters cholesterol 3-sulfate to the same extent as cholesterol (Visconti et al. 1999), did not affect Gal-4 axonal disposition (Fig. 4a). Altogether these data strongly suggest that binding to the 3′-sulfated galactose headgroup on sulfatides is required for axonal sorting of Gal-4 but not for its biosynthesis.

Gal-4's crucial role in the raft-based apical transport of glycoproteins in polarized epithelial cells, together with the particularly discontinuous distribution along the axonal membrane, suggested that the modulation of axon development-related molecules in the membrane could underlie the influence of Gal-4 on axon growth. L1 fulfilled several prerequisites to test this concept. L1 is sorted to the axonal membrane, where variations in its clustered arrangement affect axon elongation (Kleene et al. 2001). In addition, L1 is a glycoprotein with N-glycans that are reactive with galectin-3, known to share affinity to clustered LacNAc termini with Gal-4 (Wu et al. 2004; Díez-Revuelta et al. 2010; Krzeminski et al. 2011). Finally, L1 becomes missorted if the processing towards N-glycosylation with galectin-reactive LacNAc termini is perturbed by DMJ treatment (McFarlane et al. 2000), a similar effect as the one observed for apical glycoproteins in epithelial cells (Delacour et al. 2005; Morelle et al. 2009; Stechly et al. 2009). As shown herein, Gal-4 interacts with L1, as they become co-patched on the axons of living neurons (Fig. 5a), and they co-IP after reversible cross-linking (Fig. 5b). Functional meaning of the Gal-4/L1 interaction is reflected in the low number, high degree of dispersion, and reduced average area of L1 clusters in axons of Gal-4 knockdown neurons (Fig. 5c and d), indicating that Gal-4 is required for proper spatial organization of L1 at the membrane, in accord with the retarded axon growth observed in low-Gal-4 neurons (Fig. 3). Importantly, L1/Gal-4 co-IP was drastically reduced after cell treatment with the α-mannosidase I inhibitor DMJ (Fig. 5b), a clear indication that the hybrid/complex-type N-glycans on L1 are required for Gal-4/L1 interaction. In contrast, Gal-4/L1 interaction does not seem to be dependent on sulfatides, as co-IP experiments were performed on PC12, a cell line that do not express those sphingolipids (Townson et al. 2007). Furthermore, DMJ as well as NaClO3 treatments led to deficient L1 clustering (Fig. 6b), while only in the latter case, discontinuous distribution of Gal-4 on axon membrane remains unaffected (Fig. 6a). Thus, presentation of galectin-reactive LacNAc termini on N-glycans of L1 is the biochemical platform for its Gal-4-directed distribution on the axon membrane, this structural feature not being necessary for the transport of Gal-4 itself, which depends on sulfatide synthesis. The inhibition studies thus illustrate the relevance of the dual specificity to glycolipid/glycoprotein determinants of this tandem-repeat-type galectin.

Altogether, the polarization of Gal-4 to the axon membrane is required for proper axon formation and outgrowth. Gal-4 sorting relies on binding to raft sulfatides of microtubule-associated carriers, and is instrumental for the correct functional arrangement of the glycoprotein L1 (and likely others) at the membrane of the axon. This routing rests on cargo linking to Gal-4-bound carriers through specific binding of Gal-4 to N-glycans in the transported glycoproteins (a proposed model for this mechanism is depicted in Figure S6). Thus, we propose that Gal-4 effect on axon growth is exerted through the modulation of axon membrane glycoprotein presence and organization.


We thank María Peñas de la Iglesia (Hospital Nacional de Parapléjicos –SESCAM-) and Susana Fraile-Martín (IBMCC USAL-CSIC) for their valuable technical support. This study was funded by i) EC Marie Curie RTN (contract no. 2005-019561), ii) EU 7th Framework Program (under grant agreement no. 26060, and iii) Carlos III Health Institute and Castilla-La-Mancha Health Service (SESCAM) EMER program (EMER07/026). The authors have no conflict of interest to declare.