A new Purkinje cell antibody (anti-Ca) associated with subacute cerebellar ataxia: immunological characterization
© Jarius et al. 2010
Received: 2 February 2010
Accepted: 12 March 2010
Published: 12 March 2010
Skip to main content
© Jarius et al. 2010
Received: 2 February 2010
Accepted: 12 March 2010
Published: 12 March 2010
We report on a newly discovered serum and cerebrospinal fluid (CSF) reactivity to Purkinje cells (PCs) associated with subacute inflammatory cerebellar ataxia. The patient, a previously healthy 33-year-old lady, presented with severe limb and gait ataxia, dysarthria, and diplopia two weeks after she had recovered from a common cold. Immunohistochemical studies on mouse, rat, and monkey brain sections revealed binding of a high-titer (up to 1:10,000) IgG antibody to the cerebellar molecular layer, Purkinje cell (PC) layer, and white matter. The antibody is highly specific for PCs and binds to the cytoplasm as well as to the inner side of the membrane of PC somata, dendrites and axons. It is produced by B cell clones within the CNS, belongs to the IgG1 subclass, and activates complement in vitro. Western blotting of primate cerebellum extract revealed binding of CSF and serum IgG to an 80-97 kDa protein. Extensive control studies were performed to rule out a broad panel of previously described paraneoplastic and non-paraneoplastic antibodies known to be associated with cerebellar ataxia. Screening of >9000 human full length proteins by means of a protein array and additional confirmatory experiments revealed Rho GTPase activating protein 26 (ARHGAP26, GRAF, oligophrenin-1-like protein) as the target antigen. Preadsorption of the patient's serum with human ARHGAP26 but not preadsorption with other proteins resulted in complete loss of PC staining. Our findings suggest a role of autoimmunity against ARHGAP26 in the pathogenesis of subacute inflammatory cerebellar ataxia, and extend the panel of diagnostic markers for this devastating disease.
Autoimmune cerebellar ataxia (ACA) is an etiologically and pathologically heterogeneous syndrome. Besides multiple sclerosis (MS), paraneoplastic neurological disorders (PND) are the most common cause of ACA[1, 2]. Many cases of paraneoplastic ACA are associated with serum or CSF antibodies to neuronal and/or glial antigens such as anti-Hu, anti-Yo, anti-CV2/CRMP5[5, 6], anti-Tr, anti-Zic4, anti-protein kinase C gamma (PKCγ), anti-mGluR1[10, 11], anti-PCA2, anti-ANNA3, or antibodies to voltage gated calcium channels (VGCC). In patients with non-paraneoplastic ACA, antibodies to glutamate decarboxylase[15, 16], tissue transglutaminase, glutamate receptor δ2 (GluRδ2)[18, 19], and Homer-3 have been described.
Here we report a newly discovered autoantibody to Purkinje cells in a patient with subacute cerebellar ataxia but no tumor. This antibody binds specifically to the inner membrane and cytoplasm of PC somata, dendrites and axons. It is produced intrathecally, belongs to the IgG1 subclass and activates complement in vitro. Probing of a protein microarray with the patient's serum and additional confirmatory experiments identified the Rho GTPase activating protein 26 (ARHGAP26) as the target antigen.
IHC was performed on cryosections of adult mouse cerebellum, cerebrum, and brain stem tissue (Euroimmun, Luebeck, Germany), on cryosections of adult rhesus monkey cerebellum, brain stem, cerebrum, hippocampus, hypothalamus, peripheral nerve, and intestine tissue (Euroimmun), and on cryosections of adult rat cerebellum and brain stem (Zyagen, San Diego, CA). Mouse and monkey tissue was provided unfixed as snap frozen sections (4-6 μm) and was fixed with 10% formalin in phosphate buffered saline (PBS) for 4 min before testing. For production of rat sections, animals had been perfused with 4% paraformaldehyde (PFA), the brains removed, immersed to the same fixative up to 12 hours, cryoprotected with sucrose, frozen in OCT, and finally sectioned at a thickness of 4-10 μm. For some experiments, 0.125% triton-X or 1% 3- [(3-cholamidopropyl)dimethylammonio]-1-propanesulfonate (CHAPS) in PBS were applied to the sections for 4 min. Sections were then washed in PBS, blocked with 10% goat serum or with 10% donkey serum, with respect to the secondary antibodies used, and, after three washes in chilled PBS, incubated with patient serum for 1 hour or with various commercial antibodies for three hours at room temperature (RT) or overnight at 4°C. Binding of human IgG, IgA and IgM to CNS tissue was detected by use of polyclonal goat anti-human IgG antibodies conjugated to fluorescein isothiocyanate (FITC) (Euroimmun), Alexa Fluor® (AF) 488 (Invitrogen, Karlsruhe, Germany) or AF568 (Invitrogen), polyclonal donkey anti-human IgG antibodies labeled with Rhodamin Red-X (Dianova, Hamburg, Germany), and polyclonal goat anti-human IgM and anti-human IgA antibodies conjugated to FITC (Euroimmun), respectively. Binding of the following commercial antibodies was detected using goat anti-rabbit IgG AF568 (1:200-1:300; Invitrogen), goat anti-mouse IgG AF568 (1:100-1:400; Invitrogen), donkey anti-chicken IgG Rhodamin-Red × (1:100-1:200; Dianova), or donkey anti-goat IgG AF488 (1:200-1:400) as secondary antibodies depending on the primary antibodies employed and on further secondary antibodies used in double labeling experiments: goat anti-Homer3 (1:50; Santa Cruz, Heidelberg, Germany); rabbit anti-protein kinase C gamma (PKCγ) (1:50; Santa Cruz); mouse anti-metabotropic glutamate receptor 1α (mGluR1α) (1:200-1:300; BD Pharmingen, Heidelberg, Germany); rabbit anti-glutamate receptor delta 2 (GluRδ2) (1:25-1:50; Santa Cruz); mouse anti-glutamate receptor 3 (GluR3, clone 3B3) (1:25-1:500; Millipore, Schwalbach, Germany); rabbit anti-inositol-triphosphate receptor type I (IP3RI) (1:200; Dianova); chicken anti-glial fibrillary acidic protein (GFAP) (1:1000; Encor Biotechnology, Gainesville, FL); rabbit anti-aquaporin4 (AQP4) (1:200; Sigma Aldrich, Taufkirchen, Germany); mouse anti-calbindin-D (1:50; Swant, Bellinzona, Switzerland); goat anti-parvalbumin (1:50; Swant); and anti-Rho GTPase-activating protein 26 (ARHGAP26) (1:75; Santa Cruz). For selected experiments, the patient's CSF was incubated with 3 μg of the following recombinant human proteins for 3 h at RT prior to testing: ARHGAP26 (Abnova, Taipei, Taiwan); IP3RI (Santa Cruz); and AQP4 (Abcam); the sera were then centrifuged at 11,2000 rpm for 10 min and the supernatants incubated with cerebellum sections as described above. Sections were then mounted using glycerol standard immunofluorescence mounting medium containing 4',6-diamidino-2-phenylindole (DAPI) (1:1000) (Euroimmun) or ProLong Gold antifade reagent (Invitrogen). Slides were analyzed on a Nikon 90i upright fluorescence microscope and a Nikon A1 confocal microscope (Nikon Imaging Center, University of Heidelberg, Heidelberg, Germany).
For evaluation of IgG subclasses, serum and CSF samples were tested by IHC on mouse cerebellum sections as described above, with the following modifications applied: unconjugated sheep anti-human IgG antibodies specific for IgG subclasses 1 to 4 (Binding site, Germany) were substituted for the FITC-labeled goat anti-human IgG antibody, and AF568 labeled donkey anti-sheep IgG (Invitrogen; absorbed against human IgG) was used to detect the subclass specific antibodies.
For evaluation of complement activation, mouse cerebellum sections were incubated with heat inactivated serum samples (60 min at 56°C) from our patient or controls 1:5 dilution for 60 minutes at 37°C. After three washes in chilled PBS, pooled fresh frozen serum from three healthy donors was applied as a source of complement at 1:5 dilution for 45 minutes at 37°C, followed by fixation with 10% formalin for 15 min on ice and incubation with CHAPS for 1 min. Sections were then blocked with 10% heat inactivated goat serum for 60 min and subsequently incubated with a polyclonal rabbit anti-human C3c/C3b antibody at a dilution of 1:250 for 45 minutes at 4°C (Dako, Hamburg, Germany), or a rabbit isotype control (ready for use; Zymed Laboratories, San Francisco, California) on ice. Binding of the C3c/C3b specific antibody was visualized using a goat anti-rabbit IgG AF568 antibody (1:300; 45 min; Invitrogen). Sections were washed in chilled PBS between incubations and prior to mounting.
Serum and CSF samples were tested for the presence of antibodies to cerebellar antigens by use of a commercially available western blot assay (Euroimmun, Germany). Briefly, ready-made nitrocellulose membranes containing rhesus monkey full cerebellum extract were blocked with a ready-made blocking buffer (Euroimmun) for 15 min, incubated overnight at 4°C with diluted serum (1:25) or CSF (1:4) from the patient or healthy controls, washed three times in diluted blocking buffer, and finally visualized using an alkaline phosphatase labeled anti-human IgG antibody as conjugate and BCIP (5-bromo-4-chloro-3'-indolyphosphate p-toluidine salt) and NBT (nitro-blue tetrazolium chloride) as chromogenes. The reaction of BCIP/NBT to 5-5'-dibromo-4,4'-dichloro-indigo and NBT-formazan was stopped after 10 min by application of ice cold ddH2O. Additional membrane stripes originating from the same blot were incubated with goat anti-Homer3 (1:100), rabbit anti-PKCγ (1:75), mouse anti-mGluR1α (1:200), rabbit anti-GluRδ2 (1:75), rabbit anti-IP3RI (1:500), rabbit anti-coilin (Santa Cruz) (1:100), and rabbit anti-ARHGAP26 (Santa Cruz), respectively. Binding of these antibodies to their respective antigens was visualized using the Odyssee® Infrared Imaging System after application of goat anti-rabbit IgG labeled with Infrared dye (IRdye) 800 (Rockland Immunochemicals, Gilbertsville, PA) or goat anti-mouse AF688 (Invitrogen) as secondary antibodies (1:5000). A batch-specific evaluation matrix provided by the manufacturer was used to identify molecular weight (MW) ranges for proteins detected by the patient's serum and CSF or either of the commercial antibodies used. In addition, bands detected with well established paraneoplastic antibodies (Ri, 80kDa; Yo, 62 and 34 kDa; Ri, 55 kDa; Hu, 38kDa) were used to further define MW ranges.
Oligoclonal IgG bands were assessed by standard methods employing isoelectric focusing of serum and CSF at equal concentrations and subsequent detection of IgG by immunoblotting. Quantitative expressions of intrathecal antibody synthesis were based on calculation of the CSF/serum ratios of Purkinje cell specific IgG antibodies and total IgG (QIgG [spec] = IgGspec [CSF]/IgGspec [serum], and QIgG [total] = IgGtotal [CSF]/IgGtotal [serum]) . Antibody titers were determined semi-quantitatively by indirect immunofluorescence on mouse cerebellum sections as described above. Total IgG and total albumin concentrations in serum and CSF were determined nephelometrically (BN ProSpec, Dade Behring, Germany). The intrathecal synthesis of anti-Purkinje cell antibodies was detected by calculation of the corresponding antibody index (AI): AI = QIgG [spec]/QIgG [total], if QIgG [total] < Qlim, and AI = QIgG [spec]/Qlim, if QIgG [total] > Qlim. The upper reference range of QIgG [total], Qlim, was calculated according to Reiber's formula to correct for possible underestimation of intrathecal specific synthesis due to possible blood-CSF barrier disturbance. AI values >4 were considered to be indicative of intrathecal anti-Purkinje cell IgG production.
Commercially available immunoblots were used to test for antibodies to Hu, Yo, Ri, amphiphysin, glutamate decarboxylase (GAD), CV2/CRMP5, Ma2/PNMA2, GM1, GM2, GM3, GD1a, GD1b, GT1b, GQ1b, and AMA-M2 according to the manufacturer's instructions (Euroimmun). ANAs were assessed by immunocytochemistry on HEp2 cells and further characterized by commercial immunoblots for the detection of antibodies to nRNP/Sm, Sm, and SS-A, Ro-52, SS-B, Scl-70, PM-Scl, Jo-1, PCNA, double stranded DNA (dsDNA), centromer protein B (CEPB), nucleosomes, histones, and ribosomale P proteins (Euroimmun), and by double-staining of tissue sections with a rabbit polyclonal antibody to p80-coilin (Santa Cruz).
A commercially available human protein microarray (Protoarray v5.0; Invitrogen) spotted with >9000 human full length-proteins purified from a baculovirus-based expression system was probed with the patient's serum according to the manufacturer's instructions. Briefly, the array was incubated with blocking buffer containing 50 nM HEPES, pH 7.5, 200 nM NaCl, 0.08% triton X-100, 25% glycerol, 20 nm reduced glutathione, 1 nM DTT, and a commercial synthetic block (Invitrogen) for 1 h at 4°C. After washing, the array was incubated with the patient's serum at a 1:1500 dilution in washing buffer containing PBS-0.1%Tween and synthetic block. After more washing steps, a goat anti-human IgG detection antibody labeled with AF647 (final concentration, 1 μg/μl) (Invitrogen) was applied for 90 min at 4°C. A Genepix™ 4000 B microarray scanner and corresponding application software (Genepix, Sunnyvale, CA) was used to scan the slide, and Protoarray v5.0 Prospector software (Invitrogen) to analyze the raw data.
Protran BA79 nitrocellulose membranes (0.1 μm) (Whatman) were spotted with increasing dilutions (1:2, 1:4, 1:8, 1:16, 1:32) of a 0.14 μg/μl solution of human full length ARHGAP26 (10 μl/spot) in 0.1% bovine serum albumin (BSA). After drying, membranes were blocked with 5% bovine serum albumin (BSA) in Tris-buffered saline (TBS) for 1 h at RT, washed three times in TBS with 0.05% Tween (TBS-T), and finally incubated with a 1:20 or 1:200 dilution of the patient's serum in 0.1% BSA/TBS-T for 1 h at RT. A donkey anti-human IgG antibody labeled with IRdye 700DX (Rockland) was used to detect bound IgG. Stripes were finally washed in TBS and analysed using an Odyssey™ fluorescence scanner (Licor, Lincoln, NE) and Odyssey™ 2.0.40 application software (Licor). As controls, serum samples from three healthy donors were tested in the same run.
Mixed cerebellar cultures were prepared as published previously[23, 24]. Briefly, newborn NMRI mice were killed by decapitation. Cerebella were dissected in PBS, and meninges were removed. Tissues were treated with trypsin (1% in PBS; Worthington, Freehold, NJ) for 3 min at room temperature (RT). After replacing trypsin with DNase (0.05% in BME; Worthington), cerebella were triturated successively with three fire-polished Pasteur pipettes of decreasing bore sizes. Cells were centrifuged and resuspended in PBS with DNase, and the cell slurry was passed through a 40 μM nylon mesh filter. Cells were resuspended in serum containing medium, incubated in uncoated Petri dishes for 30 min at 37°C to remove fibroblast contamination, and finally plated on coverslips coated with 500 μg/ml poly-D-lysine. Plated cells were allowed to attach overnight, and then the medium was changed to complete serum-free medium composed of BME, BSA (10 mg/ml; A-8806, Sigma), glutamax (10 mM), glucose (32 mM), 1 nM triiodothyronine (Sigma-Aldrich), penicillin-streptomycin (29 U/ml each, Invitrogen) and Sigma I-1884 supplement. Thereafter, medium was replaced every 3-4 d during culture period of 14 days.
Purkinje cell cultures were fixed with 4% paraformaldehyde and 0.05% triton-X at RT for 30 min and immunostained with the patient's serum (1:1000) or a mouse monoclonal antibody against calbindin D28k (1:500; Sigma-Aldrich) followed by incubation with an AF488- or AF555-labeled goat anti-mouse or goat anti-human IgG antibody (1:1000; Invitrogen). Cells were then photographed with an AxioCam digital camera (Zeiss, Germany) on a Zeiss Axio Imager microscope (63× objective).
Laboratory findings, treatment, and treatment response over the course of disease
Months from onset
Total protein mg/dl
IgG anti-Ca titers (IHC)
IgG total mg/dl
IgM total mg/dl
Albumin total mg/dl
3 × 1000 mg methylprednisolone (MP) intravenously followed by oral therapy over 3 weeks at an initial dose of 60 mg MP, resulting in marked neurological improvement. After tapering corticosteroids to 12.5 mg MP per day the patient experienced worsening of symptoms together with an exaggerated startle response (month 2).
5 × 500 mg methylprednisolone (MP) intravenously (month 3) and 5 × 500 mg methylprednisolone (MP) intravenously + intravenous immunoglobulins (month 4), followed by only slight and transient neurological improvement. Oral therapy with MP at an initial dose of 60 mg MP.
Plasma exchange at month 6 (after progressive neurological deterioration), followed by clinical stabilization and moderate improvement.
We identified a new serum and CSF autoantibody to Purkinje cells (PCs) in a patient with subacute cerebellar ataxia. Further experiments revealed ARHGAP26 as its target antigen. Our findings expand the panel of diagnostic serum markers of autoimmune ataxia and suggest a role of ARHGAP26 autoimmunity in the pathogenesis of this condition.
Autoantibody-associated subacute cerebellar ataxia is frequently of paraneoplastic nature[1, 2]. Despite broad diagnostics, including repeat FDG-PET-CT scans, no tumor has been found in our patient 17 months after onset. However, paraneoplastic antibodies and the associated syndromes can precede tumor diagnosis by several years. In a large study on patients with anti-Yo antibodies, the most common paraneoplastic serum reactivity associated with autoimmune cerebellar ataxia, the neurologic syndrome preceded the diagnosis of cancer by up to 15 months and in many led to that diagnosis. We can therefore not yet completely exclude a paraneoplastic origin of the syndrome.
Ataxia developed two weeks after a common cold in our patient. The close temporal relationship could indicate a causal link between the two events. Molecular mimicry has indeed been discussed to play a role in the pathogenesis of other autoimmune disorders of the nervous system such as the Guillain Barré syndrome. Alternatively, infections can operate as a disease trigger in autoimmune disease. Viral infections are well known to prompt disease activity in MS, and antecedent signs of viral infections were reported in 15-35% of neuromyelitis optica cases[29, 30]. However, given the high prevalence of viral infections, the two events might well be causally unrelated.
The antibody bound almost exclusively to PCs. In good agreement with this finding, MRI and PET-CT did not display any sites of inflammation outside the cerebellum. However, it is of notice that our patient developed an increased startle response and a brisk head retraction reflex two months after onset, suggestive of symptomatic hyperekplexia, a condition mainly found in patients with brain stem disease. Moreover, severe depression, restlessness, and anxiety occurred in our patient, which could indicate an involvement of the limbic system. We can therefore not completely rule out that other areas of the CNS were affected as well. On the other hand, rare cases of secondary hyperekplexia due to cerebellar pathology have indeed been reported in the literature, and a combination of steroid-induced and reactive depression sufficiently explains the patient's psychiatric symptoms.
There is some evidence for a direct role of the antibody in the pathogenesis of the condition. First, the antibody is highly specific for PCs. PCs are GABAergic neurons located in the cerebellar cortex and constitute the only output of motor coordination from the cerebellar cortex. Secondly, the antibody is produced intrathecally and at high titers. We found an extraordinarily high antibody index (AI) at onset (Table 1), strongly indicating that it is produced by clonally expanded B cells within the CNS. Thirdly, it belongs to the IgG1 subclass and is capable of initiating complement C3b deposition in vitro, suggesting that it may act on PCs via complement-dependent mechanisms. Antibody-induced complement-mediated cytotoxicity is a well established feature in other autoantibody-associated disorders with presumed humoral pathogenesis such as neuromyelitis optica[33, 34], though other direct effects such as antibody-dependent cell-mediated cytotoxicity or induction of apoptosis might have played a role as well. Finally, plasma exchange (PEx) and immunoadsorption (IA) was followed by clinical stabilization and moderate improvement. Unfortunately, no CSF or serum samples obtained shortly after PEx/IA were available for analysis. Testing of a paired CSF and serum sample taken after treatment with intravenous methylprednisolone and IVIG showed a decline of the Purkinje cell antibody index (AI) from 68 at month 1 to 12 at month 6, and normalization of the cell count and the blood brain barrier function. Interestingly, however, the decline of the AI value was mainly due to a decrease of the CSF titer from 1:2000 to 1:200, while the serum titer of the antibody did not change significantly.
However, passive transfer experiments, which alone could prove a direct pathogenic effect of the antibody, have not yet been performed. We can therefore not fully exclude that the antibody represents only an epiphenomenon, similar to the situation in many paraneoplastic neurological diseases, while the actual tissue damage is caused by other components of the immune system such as T cells. Indeed, the antibody targets a protein that resides at the inner side of the plasma membrane and in the cytoplasm. This is important since some authors believe that intracellular antigens might not be accessible to antibodies in vivo. Most neurological autoantibodies of proven pathogenic impact, such as antibodies to AQP4 in neuromyelitis optica [34–36], acetylcholine receptor in myasthenia gravis, VGCC in Lambert Eaton syndrome, and mGluR1 in paraneoplastic cerebellar degeneration in fact target transmembrane proteins. Moreover, passive transfer of antibodies to nuclear antigens such as anti-Yo [38–40] have not produced clinical disease in animal studies. Instead, T cell-mediated immune mechanisms directed against the target antigen of the accompanying antibody have been proposed to play a role in those disorders [41–44].
However, there is still little direct evidence for a major role of T cells. Moreover, there are conditions in which antibodies against intracellular antigens were indeed shown to be of pathogenic impact, as proven by passive transfer. Sommer et al. found dose-dependent stiffness with spasms resembling human stiff-person syndrome in rats after injection of human serum containing high titers of antibodies to amphiphysin, a protein associated with the cytoplasmic surface of synaptic vesicles. Interestingly, both amphiphysin and ARHGAP26 are involved in endocytosis. One of the main roles of amphiphysin is to recruit dynamin to sites of clathrin-mediated endocytosis in GABAergic neurons. Dynamin, a protein with GTPase activity, is important also in the clathrin-independent endocytic pathway, which is regulated by ARHGAP26, and ARHGAP26 is a strong interactor of dynamin. Both proteins, amphiphysin and ARHGAP26, contain a BAR domain and a SH3 domain, though cross-reactivity of the patient's antibody with amphiphysin was excluded in our study by three independent methods (protein microarray; commercial line blot; and IHC). The patient's IgG indeed precipitated ARHGAP26 and co-precipitated dynamin from a mouse cerebellar extract as demonstrated by mass spectroscopy (data not shown), which is in perfect accordance with a recent study reporting co-precipitation of dynamin from rat brain cytosol by a non-human, ARHGAP26-specific antibody. Besides anti-amphiphysin, a direct pathogenic effect of antibodies on intracellular proteins has also been shown for recoverin, which is located inside of retinal cells. Anti-recoverin antibodies were demonstrated to enter retinal cells actively, probably by endocytosis, and uptake of anti-recoverin (but not of control IgG) induced caspase-dependent apoptosis. Endocytotic uptake of antibodies has also been shown for a subset of anti-DNA antibodies [49–51]. Finally, a recent study demonstrated that PCs incorporate immunoglobulins of both the IgG and the IgM classes in vitro, independent of the immunoglobulin's reactivity with Purkinje cell surface antigens, giving raise to speculation as to whether paraneoplastic or other autoantibodies reactive with cytoplasmic or nuclear Purkinje cells antigens might be taken up intracellularly and could potentially produce cell injury and death.
ARHGAP26 has not previously been described as an autoimmune target in human or animal disease. However, mutations of ARHGAP26 are a rare cause of juvenile myelomonocytic leukemia, one of the most common pediatric myelodysplastic syndromes.
Further research is now needed to evaluate the exact role of anti-ARHGAP26 antibodies in the pathogenesis of cerebellar ataxia, involving the development of a passive transfer animal model. Moreover, diagnostic tests are to be developed to assess the frequency of anti-ARHGAP26 antibodies in patients with subacute ataxia.
We describe a new autoantibody to Purkinje cell somata, dendrites and axons associated with subacute cerebellitis. The antibody targets ARHGAP26 and is produced intrathecally. Our findings indicate a role of autoimmunity to ARHGAP26 in the pathogenesis of subacute inflammatory cerebellar ataxia and expand the panel of diagnostic markers for this devastating condition.
The work of SJ was supported by a Fellowship from the European Committee for Treatment and Research in Multiple Sclerosis (ECTRIMS) and by research grants from Bayer Vital GmbH and Merck Serono (to BW). We are very thankful to Mrs Brigitte Fritz and Mrs Anna Eschlbeck for outstanding technical assistance.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.