Targeting of Voltage-Gated Calcium Channel α2δ-1 Subunit to Lipid Rafts Is Independent from a GPI-Anchoring Motif

Voltage-gated calcium channels (Cav) exist as heteromultimers comprising a pore-forming α1 with accessory β and α2δ subunits which modify channel trafficking and function. We previously showed that α2δ-1 (and likely the other mammalian α2δ isoforms - α2δ-2, 3 and 4) is required for targeting Cavs to lipid rafts, although the mechanism remains unclear. Whilst originally understood to have a classical type I transmembrane (TM) topology, recent evidence suggests the α2δ subunit contains a glycosylphosphatidylinositol (GPI)-anchor that mediates its association with lipid rafts. To test this notion, we have used a strategy based on the expression of chimera, where the reported GPI-anchoring sequences in the gabapentinoid-sensitive α2δ-1 subunit have been substituted with those of a functionally inert Type I TM-spanning protein – PIN-G. Using imaging, electrophysiology and biochemistry, we find that lipid raft association of PIN-α2δ is unaffected by substitution of the GPI motif with the TM domain of PIN-G. Moreover, the presence of the GPI motif alone is not sufficient for raft localisation, suggesting that upstream residues are required. GPI-anchoring is susceptible to phosphatidylinositol-phospholipase C (PI-PLC) cleavage. However, whilst raft localisation of PIN-α2δ is disrupted by PI-PLC treatment, this is assay-dependent and non-specific effects of PI-PLC are observed on the distribution of the endogenous raft marker, caveolin, but not flotillin. Taken together, these data are most consistent with a model where α2δ-1 retains its type I transmembrane topology and its targeting to lipid rafts is governed by sequences upstream of the putative GPI anchor, that promote protein-protein, rather than lipid-lipid interactions.


Introduction
Voltage-gated calcium channels (Ca v s) represent the primary means by which changes in membrane potential are coupled to the influx of second messenger calcium ions [1]. As such, Ca v s play a major role in orchestrating diverse excitable cell functions, ranging from rapid events such as neurotransmitter release in nerves and excitation-contraction coupling in muscle, to longer lasting events such as synaptic plasticity. While it is well established that disruption of Ca v s is involved in diverse pathologies, including neuropathic pain [2] and cardiac arrhythmia [3], much less is known about how Ca v functionality is modulated, physiologically, at the cellular level [4].
Biochemical and reconstitution studies show that Ca v s comprise an a 1 subunit (<200 kDa) containing the voltage-sensing, gating and pore machineries [1], [5]. In high voltage-activated Ca v 1 and Ca v 2 family channels, a 1 is complexed in a 1:1 stoichiometry with a cytoplasmic auxiliary b subunit. These channels are also complexed with a second auxiliary (<125 kDa) subunit termed a 2 /d, which, like b subunits, enhances cell surface expression and modulates the biophysical properties of channel heteromers [1], [6], [7]. Since multiple genes encode each type of Ca v subunit and their transcripts undergo RNA splicing, Ca v s manifest a considerable potential for diversity not only in terms of biophysical function, but also in their modulation and cellular expression patterns [1], [7].
Irrespective of their location, emerging data has shown that Ca v s are organised into large heterogeneous macromolecular assemblies containing a plethora of signal transduction proteins with which they interact and co-operate to meet local and global functional demands [4], [8], [9], [10]. Defining the mechanisms by which such assemblies are constructed and distributed is therefore crucial to understanding and manipulating Ca v function [10], [11], [12]. In this regard, an important step forward has been the observation that Ca v proteins co-localise with components of specialised cholesterol-rich membrane signalling domains termed lipid rafts [13], [14], in both heterologous expression systems and native tissues [15][16][17][18][19][20][21]. While alterations in Ca v currents seen with cholesterol-depleting agents argue that raftassociation is physiologically significant, the precise effects appear to be subtype and/or tissue specific [16], [18][19][20][21]. Although different Ca v s may associate with rafts using alternate modalities [18], [22], there is now compelling evidence for a major involvement of the a 2 /d subunit [18], [20], [21]. Thus, a 2 /d subunits co-localise with the lipid raft marker proteins caveolin and flotillin when expressed alone [18], [20], [21] and are also necessary and sufficient for the targeting of Ca v 2.2 complexes to rafts [21].
Until recently, how the a 2 /d subunit might mediate Ca v raft targeting was unclear. Structurally, the a 2 /d subunit has been viewed as a type I transmembrane (TM) spanning protein (Fig. 1A) composed of a large exofacial a 2 head region linked via disulfide bonds to a smaller membrane associated d subunit [1], [7], [23], [24], [25]. Owing to the presence of features such as Von Willebrand factor A (VWA) and Cache domains, commonly found in integrins and other cell surface proteins, the a 2 region is thought to have a modular structure [6], [7], [26] affording interactions with extracellular matrix proteins such as thrombospondin [27]. Structure-function analysis has also shown that the a 2 region mediates those interactions with Ca v s that support current enhancement and the biophysical effects seen upon co-expression of a 2 /d subunits with a 1 /b complexes [28], [29]. In contrast, the d polypeptide, while affecting the voltage-dependence of Ca v s [28], has been viewed as primarily providing a means for attaching the a 2 polypeptide to the cell surface via its hydrophobic putative TMspanning domain located proximal to the short, intracellular, carboxy terminus [1], [7], [20] [23][24][25]. However, a recent study has challenged this structural model and offered a new mechanism for Ca v raft localisation by suggesting the a 2 /d subunit associates with the plasma membrane via a glycosylphosphatidylinositol (GPI) anchor attached to the d polypeptide [20]. In common with other GPI anchored proteins, GPI attachment is envisaged to occur through the action of an ER-resident GPI-transamidase which recognises, cleaves and modifies a motif located at the distal carboxy terminus [28], [30][31][32]. While such anchoring motifs do not have a strict consensus sequence, they contain common elements including a) an amino acid with a small side chain (notably G, C, D, A, N or S) known as the v site/residue, to which the GPI moiety is amide-linked, b) two adjacent residues (v+1..2) with small side chains (typically G, A and S), c) a spacer sequence of .6 hydrophilic residues, commencing at the v+3 position and d) a stretch of hydrophobic residues (particularly L) capable of spanning the membrane [30], [31].
Since GPI-anchored proteins are highly concentrated in lipid rafts [14], [33], [34], the revised model of a 2 /d subunit structure has then been used to rationalise Ca v raft targeting [20], [35] and the apparent weakness of the a 2 /d subunit-a 1 /b complex interaction [10], [36]. However, while seeming attractive in offering GPI attachment as a further regulatory locus [35], such a model requires that lipid-lipid interactions between a single d subunit GPI anchor and liquid-ordered (L o ) raft lipids [37] can specify the raft association of Ca v a 1 (+b), a large, multispanning, membrane protein complex, predicted to partition into liquiddisordered (L D ), bulk phase lipid [14], [38], [39]. Moreover, of the four mammalian a 2 /d subunits, only a 2 /d-3 shows a significant potential for GPI anchoring when analysed by predictive algorithms (Table 1). Recent evidence also indicates that the colocalisation of raft markers and a 2 /d-1 subunits (when expressed alone or with a 1 /b complexes) in cell surface aggregates demands an intact actin-based cytoskeleton [21]. However, while this is consistent with a role for actin in shaping the distribution and dynamics of GPI-anchored proteins [40], [41], [42], such observations are equally consistent with the hypothesis that a 2 /d-1 subunits reside in rafts, and/or higher order raft assemblies, via organising principles based upon protein-protein [21], [42], [43], [44] and/or specialised lipid-protein [14], [45], [46], [47] interactions.
To resolve the above hypotheses we have re-visited the raft localisation of the a 2 /d subunit using an established strategy [48], [49], [50] based on the expression of chimera, where the reported GPI-anchoring sequences in a 2 /d-1 have been swapped with those from a known Type I TM-spanning protein -PIN-G (Fig. 1B) [51]. Like its a 2 /d-2 and a 2 /d-3 counterparts, a 2 /d-1 has been described as a GPI-anchored protein [20]. However, unlike a 2 /d-2 and a 2 /d-3, the consequences of mutating the presumptive GPIanchoring motif on a 2 /d-1 raft localisation or Ca v currents have not been reported. Using imaging, electrophysiological and biochemical assays that we recently employed to analyse a 2 /d-1 in rafts [21], we now show that the raft localisation of a 2 /d-1 is preserved even after replacement of the reported GPI anchoring motif with the TM domain of PIN-G. Conversely, the GPI-anchoring motif is not sufficient to target PIN-G to lipid rafts. While the localisation of a PIN construct containing a 2 /d-1, and its GPI motif, to lipid rafts shows susceptibility to GPI-cleavage using phosphoinositide-specific phospholipase C (PI-PLC), this effect is assay-dependent and seems to lack specificity as it also disrupts the raft localisation of caveolin, but, interestingly, not flotillin. Our data therefore support a model where the raft localisation of a 2 /d-1 depends upon exofacial sequences upstream and independent of the putative GPIanchoring motif.

Construction and GPI-anchoring potential of a 2 /d-1/PIN-G chimera
To dissect the role of GPI anchoring in localising a 2 /d subunits to lipid rafts, a series of chimera were prepared between rat a 2 /d-1 and PIN-G, a functionally inert Type I TM protein reporter that lacks trafficking or post-translational modification motifs [51] ( Fig. 1). Initially, we made a PIN chimera 2PIN-a 2 /d -encoding the PIN 'head' region (i.e. signal peptide, Haemagglutin (HA) and Green fluorescent protein (GFP) tags, lacking TM and intracellular domains) fused to full length a 2 /d-1. Next, a chimera 2PIN-d cwas generated by fusing the PIN head to the distal carboxy terminal region of the d-1 polypeptide to yield a construct containing the entire purported GPI-anchoring motif of a 2 /d-1, plus 33 residues upstream, and all residues downstream of the v site (Wild type (WT) a 2 /d-1:Gly1060, [20]. Two additional constructs 2PIN-d c -PIN TMI and PIN-a 2 /d-PIN TMI -were then designed where the putative GPI anchoring motifs within PIN-d c and PIN-a 2 /d were disrupted by replacement of all d residues after the v residue, with those encoding the transmembrane and intracellular region ('TMI' residues 327-370) of PIN-G. Based upon the work of Davies et al., 2010 [20] both the PIN-a 2 /d and PIN-d c constructs are predicted to be GPI anchored by virtue of the presence of the purported a 2 /d-1 GPI-anchoring motif. However, this prediction is supported by only one (Pred-GPI) of the three independent algorithms [30], [52], [53] we employed, and even then with GPI-attachment at a different v residue to that predicted by Davies et al., [20] (Table 1). In contrast, all three algorithms predict that PIN-G, PIN-d c -PIN TMI and PIN-a 2 / d-PIN TMI are not GPI-anchored (Table 1), whereas GFP-GPI, which contains the GPI-anchoring motif of the folate receptor [54] is GPI-anchored.
The biophysical properties of PIN-a 2 /d are retained following substitution of the GPI-anchoring motif with the transmembrane and intracellular sequence of PIN-G In order to confirm that PIN-a 2 d was fully functional we compared its effects on the electrophysiological properties of Ca v 2.2/b 1b channels, with those of WT a 2 d-1. Preliminary experiments indicated that the presence of the GFP-tag on PINa 2 d caused a marked hyperpolarisation of the V 50 for activation and a slowing of both current activation and inactivation (Fig. S1).
These effects are consistent with previous reports on the biophysical effects of amino-terminal modifications of the a 2 /d subunit [55]. As a result, all subsequent electrophysiological experiments were conducted using constructs that lacked the GFP tag (deGFP; Fig. S1). As shown in Fig. 2, co-expression of PIN-a 2 d conferred on Ca v 2.2/b 1b currents the typical hallmarks associated with the presence of WT a 2 d-1. Thus, compared with Ca v 2.2/b 1b in the absence of a 2 d-1, the peak current density, I max , was In a 2 /d-1, residues 1-25 encode the signal peptide (SP). The d subunit is further subdivided into exofacial (d e ), putative transmembrane (d TM ) and intracellular (d i ) regions. The putative minimal GPI anchoring motif, located within a cysteine-rich region (d c ) proximal to the external face of the lipid bilayer, contains, in turn, the v residue (Gly1060) to which GPI is attached, a short spacer (dashed line) and a largely hydrophobic region. Indents between residues 1060 and 1061 indicate chimera fusion site where all downstream d sequences in constructs PIN-d c or PIN-a 2 /d were replaced by the transmembrane and intracellular carboxy terminal residues of PIN-G (constructs PIN-d c -PIN TMI and PIN-a 2 /d-PIN TMI ). The parent construct PIN-G (B) contains a signal peptide derived from the Igk chain, an exofacial haemagglutin (HA) epitope tag, green fluorescent protein (GFP) a carboxy terminal sequence (PIN TMI ) containing the transmembrane spanning domain from the platelet-derived growth factor receptor and a 17 residue intracellular inert region, whose modification with endocytic or other cytoplasmically exposed targeting motifs can be used to re-direct the reporter to specific intracellular organelles [51]. enhanced approximately 4-fold, the V 50 for activation was hyperpolarised by some 13 mV on average and the rate of current inactivation was enhanced (decreased T inact ) upon coexpression of PIN-a 2 d (see also Table S1). We next examined the functional effects of disrupting the GPI anchoring motif within a 2 d. Somewhat surprisingly, and in contrast to data for the a 2 d-2 and a 2 d-3 GPI-anchoring-deficient mutants [20], co-expression of PIN-a 2 /d-PIN TMI with Ca v 2.2/b 1b produced identical currents to those of channels containing either PIN-a 2 d or WT a 2 d-1. In the absence of any a 2 sequences there was no functional effect on Ca v 2.2/b 1b channels (Table S1; PIN-d).
Formation of a 2 /d puncta is independent of the GPIanchoring motif Upon expression in COS-7 cells and surface anti-HA immunostaining, PIN-a 2 /d exhibited a labelling pattern (Fig. 3A) characterised by the appearance of numerous small puncta, spread randomly over the cell surface, and matching that of WT a 2 /d-1 Table 1. Comparison of predicted GPI-anchoring potential for WT a 2 d-1, PIN-a 2 d chimera, mutant a 2 d-2 GAS:WKW and Thy-1. Big-PI and PredGPI predict v-site residues (bold and underlined in tetrapeptide sequences indicated), with the latter reported to afford the lowest rate of false positive predictions. Note that the v-site residues giving the highest potential for GPI-modification are indicated, irrespective of the protein's potential for GPI modification. While differences exist in the predicted v-site residues obtained between algorithms, these are generally in very close physical proximity. Of the four WT Ca v -a 2 d subunits only a 2 d-3 is predicted to be GPI-anchored by all three algorithms while WT a 2 d-1 is only predicted to be, using PredGPI. In addition, the predicted v-site for WT a 2 d-1 differs between algorithms (Big-Pi: CGGV; PredGPI: CGGV) and also to that reported [20](CGGV). doi:10.1371/journal.pone.0019802.t001    [21]. In contrast, such puncta were absent in cells expressing PINd c (Fig. 3C) Rather, PIN-d c labelling was distributed evenly over the cell surface and at the cell margins. Significantly, the two different patterns of labelling seen between PIN-a 2 /d and PIN-d c were retained in the derivative PIN-a 2 /d-PIN TMI (Fig. 3B) and PIN-d c -PIN TMI (Fig. 3D) constructs, where the GPI anchoring motifs had been disrupted.
Raft localisation requires a 2 /d sequences upstream of the GPI-anchoring motif Elsewhere, we have shown an intimate link between the formation of puncta and the co-localisation of a 2 /d with lipid raft proteins [21]. Consequently, the presence of puncta in constructs lacking the putative GPI anchoring motif (PIN-d c -PIN TMI and PIN-a 2 /d-PIN TMI ) and vice versa (PIN-a 2 /d and PIN-d c ), prompted us to examine and compare their raft localisation more directly. To this end, we exploited the ability of lipid raft components, including a 2 /d subunits [18], [20], [21], to migrate into low density fractions upon equilibrium centrifugation of cell lysates in sucrose density gradients containing ice-cold non-ionic detergents [14], notably Triton-X-100 [15], [56], [57]. Following centrifugation of lysates prepared at 48 h post-transfection, gradients were fractionated and fractions immunoblotted using anti-HA antibodies (Fig. 4). To control for gradient fidelity, each fraction was also analysed for the presence of the raft marker caveolin. Irrespective of the transfection condition, endogenous caveolin (22 kDa isoform) was detected as a single peak in fractions corresponding to the 5%-30% sucrose interface (Fig. 4A). In cells transfected with PIN-a 2 /d ( Fig. 4B blot i) approximately 20% of the anti-HA immunoreactivity was distributed at the 5-30% interface in caveolin-positive fractions, with the remainder locating to fractions of higher density centred on the 30-45% sucrose interface. In contrast, PIN-d c -which contains the putative GPI motif -was localised exclusively in the higher density non-raft fractions (Fig. 4B blot iii). Next we examined the distributions of constructs PIN-a 2 /d-PIN TMI (Fig. 4B blot ii) and PIN-d c -PIN TMI (Fig. 4B blot iv) -which lack the putative GPI-motif. In both cases raft/non-raft distributions of HA-immunoreactivity were the same as their parent constructs (PIN-a 2 /d: raft + non-raft and PIN-d c : non-raft, respectively). Thus, the raft localisation of PIN-a 2 /d appears independent of the GPI motif. Conversely, the presence of the GPI motif in PIN-d c is insufficient to support raft localisation, implying that upstream sequences are required.

The expression of PIN-a 2 /d cell surface puncta is resistant to PI-PLC treatment
Taken together, these data contradict the notion that the association of a 2 /d-1 with lipid rafts is specified by the proposed GPI-anchoring motif [20]. To examine this issue further we tested for the existence of a GPI anchor through its susceptibility to PI-PLC cleavage [20], [58], [59]. First, we followed the approach of Davies et al., (2010) [20] who used imaging to assay the effect of PI-PLC on the surface expression of a 2 /d constructs. For comparison we also examined the surface and total (surface + intracellular) distribution of GFP-GPI, a well-defined GPIanchored green fluorescent protein [54]. As shown in Fig. 5A-D, GFP-GPI was found throughout the cell where it was localised in both tubulovesicular structures and at the cell surface. Although known to reside in lipid rafts like other GPI-anchored proteins [54], [59], [60], GFP-GPI surface labelling was not present in the well-defined puncta seen with PIN-a 2 /d (e.g. Fig. 3), but rather it was distributed over the cell surface in a pattern reminiscent of a very fine, granular, meshwork (Fig. 5C,D). Following treatment with PI-PLC, all GFP-GPItransfected cells showed a qualitative decrease in surface (Cy5/ anti-GFP) labelling intensity and distribution compared with non-PI-PLC-treated cells (Fig. 5G-J). More quantitative comparisons based on determining the 'on cell' signal to noise ('off cell' background) ratio (S/B) of raw (i.e. non-background subtracted) images, showed that PI-PLC caused a reduction in GFP-GPI surface labelling intensity to 23% of control (i.e. 2PI-PLC) levels ((S/B)21 = 0.4460.066 n = 8 (2PI-PLC) vs (S/B)21 = 0.1060.0217 n = 8 (+PI-PLC); p = 0.0002) (Fig. 5K). In parallel, we examined the action of PI-PLC on the surface expression of PIN-a 2 /d. In contrast to GFP-GPI, and as noted above, PIN-a 2 /d showed a pattern of surface labelling comprised of numerous high intensity puncta, with little interstitial (inter-punctal) labelling ( Fig. 5L-O). Significantly, however, pre-treatment of cells with PI-PLC had no apparent effect on the labelling intensity ((S/B)21 = 0.7560.18 n = 6 (2PI-PLC) vs 0.9560.27 n = 8 (+PI-PLC), p = 0.56) (Fig. 5 R-V and Fig. S2). Equally important, using detailed particle analysis we found no effect on the dimensions or density of the PIN-a 2 /d puncta (Fig. S2). Neither the number of particles of given area (size distribution) (Fig. S2A), nor the particulate area fraction (a measure of changes in particle dimension) (Fig. S2B) were affected by PI-PLC treatment. Thus, we found no evidence for the effects predicted were PI-PLC treatment able to induce either 'stripping' (i.e. decreased particle size), disassembly (formation of smaller puncta) or both (Fig. S2C-F).

The raft distribution of both PIN-a 2 /d and caveolin in sucrose gradients is altered by PI-PLC treatment
As a further test for the presence of a GPI anchor in PIN-a 2 /d, we examined the effect of PI-PLC on the partitioning of PIN-a 2 /d in lipid raft fractions obtained using equilibrium centrifugation in sucrose gradients containing ice-cold Triton-X-100. As shown in Fig. 6, gradient analysis of lysates from cells expressing GFP-GPI (Fig. 6A, blot i) showed anti-GFP immunoreactivity exclusively in lipid raft fractions at the 5-30% sucrose interface. In contrast, lysates from cells pre-treated with PI-PLC (Fig. 6B, blot i) showed a marked shift in immunoreactivity which was now present in higher density non-raft fractions. Next, we examined lysates from cells transfected with PIN-a 2 /d. As before (Fig. 4), anti-HA immunoreactivity was detected in both the raft and non-raft fractions ( (Fig. 6A, blot ii). However, following pre-treatment of cells with PI-PLC all the anti-HA immunoreactivity appeared in the higher density, non-raft fractions (Fig. 6B, blot ii). While these data supported the contention that PIN-a 2 /d is GPI-anchored [20], it was also possible that PI-PLC might have a more globally disruptive effect on lipid raft integrity, particularly given the lack of effect of molecular disruption of the GPI anchoring motif. To examine such a possibility we, therefore, examined the effect of PI-PLC on the gradient distribution of both caveolin (Fig. 6A,B, blot iii) and flotillin (Fig. 6A,B, blot iv) -two endogenous raft markers with separate and independent modes of raft association [61], [62], which both co-localise in puncta containing a 2 /d [21]. As anticipated, both caveolin (Fig. 6A, blot iii) and flotillin (Fig. 6A, blot iv) were concentrated in raft fractions in the absence of PI-PLC pre-treatment. However, following PI-PLC pre-treatment, the distribution of caveolin (Fig. 6B, blot iii), but not flotillin (Fig. 6B, blot iv), shifted such that it was found primarily in the higher density non-raft fractions. Thus, PI-PLC appears to have a generally disruptive effect on the integrity of lipid rafts, whose detection depends upon whether caveolin or flotillin is used as a marker. For clarity, panels A and G depict just the surface (red channel, anti-GFP) labelling corresponding to the merged (red (surface) and green (GFP, surface + intracellular) images shown in B and H. Panels C and I correspond to high magnification views of the boxed areas shown in A and G, respectively. Note strong surface labelling and evidence of clustering of GFP-GPI, in the absence of PI-PLC and diminution of surface cluster and interstitial fluorescence after PI-PLC treatment. Since contiguity between GFP-GPI clusters precluded standard particle analysis, the effect of PI-PLC on GFP-GPI clustering was analysed further by generating contour maps (panels D and J) (level scale (0-255) shown to right) of the labelling seen in panels C and I, respectively. Line scans based on the contour maps were then constructed to show differences in fluorescence intensity in the absence (white and yellow in D and F) or presence (red and orange in D and F) of PI-PLC cell treatment. Panel K shows the effect of PI-PLC cell pre-treatment on the signal to background fluorescence for raw images (n.8) collected using identical imaging conditions. *** denotes statistically significant difference (P,0.001); Student's ttest. Panels L-T correspond to images from cells transfected with PIN-a 2 /d in the absence (L-N) and presence (R-T) of PI-PLC. Panels L and M (2PI-PLC) and R and S (+PI-PLC) show merged images for total (surface + intracellular)(green, GFP) and surface (red, anti-GFP)) for separate cells. Panels N and T correspond to high magnification views of the boxed areas shown in L and R (red, (surface) channel only). Note the presence of extensive PINa 2 /d clustering irrespective of whether or not the cells had been treated with PI-PLC. Panels O and U correspond to contour maps (above) of the labelling seen in panels N and T, respectively (level scale (0-255) shown to right). Line scans corresponding to the contour maps were then constructed to show differences in fluorescence intensity in the absence (white and yellow in P and Q) or presence (red and orange in P and Q) of PI-Treatment with PI-PLC alters the cellular distribution of caveolin but not flotillin To obtain further evidence for a generalised effect of PI-PLC on raft integrity, we examined the cellular distribution of caveolin and flotillin before and after PI-PLC treatment, using imaging assays (Fig. 7). As documented elsewhere [21], both of these raft marker proteins localise to puncta and large aggregates throughout permeabilised, non-PI-PLC-treated, COS-7 cells (Fig. 7A (caveolin), 7D (flotillin)). However, following pre-treatment of cells with PI-PLC there was a marked alteration in caveolin labelling to patterns consisting of patches of intense labelling proximal to the cell nucleus and the appearance of more diffuse labelling over the cell surface (Fig. 7B). In contrast, pre-treatment of cells with PI-PLC had no effect on the distribution of flotillin (Fig. 7E) which remained punctate throughout. These data are therefore consistent with those from the sucrose-density gradient experiments and support the notion that PI-PLC -a primary tool for defining a 2 /d-1, 2 and 3 as GPI-anchored proteins [20] -has indirect effects which may confound the assignment of proteins as possessing GPI anchors.

Discussion
In this study we have tested the notion that the Ca v a 2 /d-1 subunit is a GPI anchored protein, by substitution of the putative GPI-anchoring motif, including the downstream sequence formerly designated as TM-spanning, with a bona fide [51] TMspanning and intracellular sequence from the trafficking reporter, PIN-G. Using fundamentally different algorithms, each chimera is predicted to have little or no GPI anchoring potential, due to direct disruption of all residues adjacent and subsequent to the putative v+1 site and the extended intracellular domain. By replacing the GPI-anchoring motif with bulky lysine and hydrophobic amino acids throughout, our chimera, therefore, represent an even more extensive alteration of both the motif structure and the overall GPI anchoring potential (Table 1) than that achieved previously in Ca v a 2 /d-2 and Ca v a 2 /d-3, where just three v site residues were mutated [20]. Furthermore, by generating PIN chimera corresponding to a full length or truncated a 2 /d subunit, both containing the putative GPI-motif, it was possible to examine the independence of this motif from upstream residues.
Significantly, PIN-a 2 /d supports the key hallmarks of WT a 2 /d-1 functionality, notably a 4-fold enhancement of peak current density, a hyperpolarising shift in V 50 for activation and an enhanced rate of inactivation, when co-expressed with Ca v 2.2/b 1b subunits. Such current enhancement arises through direct actions on anterograde and retrograde trafficking of Ca v complexes [6], [63] and is highly susceptible to post-translational modification events [55]. Thus, PIN-a 2 /d is evidently able to undergo processing and trafficking events similar to WT a 2 /d-1 and like WT a 2 /d-1, can co-assemble with Ca v 2.2 a 1 subunits. Equally significant, the Ca v 2.2/b 1b current enhancement and kinetic features imparted by the GPI-anchordeficient PIN-a 2 /d-PIN TMI construct are identical to those of PINa 2 /d. This is especially remarkable given that the association of a 2 / d-1 with lipid rafts has been directly attributed to GPI-anchoring [20] and that disruption of either rafts [16], [18][19][20][21], or GPIanchoring [20], has been reported to affect Ca v current density. Our observation that PIN-d does not support enhancement of Ca v 2.2/ b 1b currents is entirely consistent with the known requirement for sequences in the a 2 subunit [7], [28], [29].
In both our biochemical and imaging assays PIN-a 2 /d exhibits the raft-association characteristics of WT a 2 /d-1 [21]. However, in these assays PIN-d c -which contains the putative GPIanchoring motif and 46 (33 d c and 13 GFP-linker) upstream residues between GFP and the predicted v site (CGG) showed no raft localisation. In contrast, GFP-GPI, which contains just 22 residues between GFP and the v site, is raft localised. Thus, raft localisation must depend upon additional determinants upstream of the d c sequence rather than merely the number of residues upstream of the v site. Although it is conceivable that determinants upstream of d c somehow promote GPI-anchor attachment, our observation that raft localisation is conserved in both PIN-a 2 /d and the anchor-deficient PIN-a 2 /d-PIN TMI , argues strongly against any involvement of the putative GPIanchor motif reported by Davies et al (2010) [20]. While we cannot rule out the possibility of cryptic (i.e. internal) GPI-anchor motifs these are very rare and are thought to resemble the classic carboxy terminal anchoring motifs in structure [64], [65]. Indeed, using predictive algorithms to assess the GPI-modification potential for sequentially truncated a 2 /d-1 constructs, we have been unable to detect any additional regions within d-1 that could serve as obvious GPI-anchoring motifs (Fig. S3).
Notwithstanding the above, our data do not exclude the possibility that GPI-anchoring plays an indirect role in a 2 /d raft localisation. Indeed, upon treatment with PI-PLC, PIN-a 2 /d was no longer associated with lipid rafts when assessed by sucrose gradient analysis. While this effect has been interpreted as arising via the release of a 2 and regions of d-1 up to the v site [20] ( Table 2), it appears to be non-specific since PI-PLC also prevented the raft-association of caveolin which, in contrast to GPI-anchored proteins, is localised to the inner membrane leaflet [61]. Significantly, depletion of caveolin has been reported to redistribute Type I TM proteins from raft to non-raft fractions [66] which may explain the data reported by Davies et al., [20] where flotillin was the primary raft marker ( Table 2). In support of this, our images showing that PI-PLC causes partial dispersal of caveolin, are highly reminiscent of those obtained from COS-7 cells treated with the cholesterol-depleting agent, methyl-bcyclodextrin (M-b-CD) [21]. However, while M-b-CD also disperses flotillin and prevents its co-localisation in lipid raft fractions, PI-PLC does not. Thus, PI-PLC treatment can disrupt raft integrity, but not completely. To our knowledge, potentially disruptive effects of PI-PLC on raft structure have not been examined, although phospholipase C activity and low concentrations of its end product -diacylglycerol, are known to destabilise model membranes including those containing raft lipids [67]. Quite why caveolin and flotillin should show differential raft partitioning after PI-PLC treatment is also unclear, but likely reflects their differing modes of membrane association. While both proteins are acylated, only caveolin has a transmembrane domain [61,62,68]. Irrespective of the mechanisms, a differential effect of PI-PLC on caveolin and flotillin raft localisation, clearly, warrants caution when using these markers alone to assess raft integrity.
Taken together, our chimera studies show that Ca v a 2 /d-1 raft localisation is independent of the putative GPI-anchoring motif and that this motif does not localise chimera to rafts. By inference, our data do not support the revised model for the topology, membrane association (i.e. GPI anchoring) or ability of a 2 /d-1 subunits to target Ca v s to lipid rafts. Rather, raft association -at least for a 2 /d-1 -appears to require sequences upstream of the v site that most likely mediate protein-protein rather than lipid-lipid interactions, a scenario more consistent with emerging views of raft biogenesis and aggregation [14], [42], [69].

Molecular biology
An a 2 /d-1 construct bearing an HA epitope tag between amino acid residues I612 and K613, was generated using a three step Table 2. Comparison of experimental approaches and conclusions in the present study and that of Davies et al., [20].

This study
Davies et al. [20] A) Substrates Not tested (reduced current density in a 2 /d-2/-3 on disruption of GPI anchoring motif) Key differences are our use of: a) both caveolin and flotillin as raft markers, b) a carefully controlled surface-labelling protocol, c) lysates from live cells treated with PI-PLC and d) the extensive use of chimera which ablate the purported GPI-anchoring motif. Asterisks denote the use of non-permeabilised cells without reference to controls. As we show elsewhere [21], fixative alone can cause significant cell permeabilisation. doi:10.1371/journal.pone.0019802.t002 strategy as described in Robinson et al. (2010) [21]. All PIN constructs were prepared through the sequential insertion, deletion or substitution [70] of specified rat a 2 /d-1 sequences into the PIN-G plasmid (Genbank: AY841887), using the QuikChange TM II kit (Agilent Technologies, UK) and mutagenic megaprimers prepared by PCR. Construct fidelity was confirmed by in-house sequencing (see Fig. 1 and Fig. S3C for chimera junctions).

Cell culture and transient transfection
Culture and transient transfection of COS-7 cells (European Cell Culture Collection, Health Protection Agency, U.K.), were carried out as described in Robinson et al. (2010) [21]. Transient transfections were performed in serum-free Dulbecco's modified Eagle's medium (DMEM) at a cell confluency of 60-70% using FuGene 6 (Roche Diagnostics, U.K.; imaging and electrophysiology) or Turbofect (Fermentas, U.K.; biochemical experiments) at a total DNA:reagent ratio of 1:3 (w/v), (total DNA: 2 mg for 6-well plates/35 mm dishes, 12 mg DNA for 10 cm plates). Transfections with Ca v 2.2, Ca v b 1b and Ca v a 2 /d-1 used a ratio of 3:1:1 by mass of subunit cDNA. For transfections omitting a 2 /d cDNA, the a 2 /d cDNA was replaced with pcDNA3.1 to maintain the equivalent mass ratio. Cells were maintained at 37uC, 5% CO 2 in complete medium for a total of 48 hours (including any replating step), after which cells were: a) fixed for microscopy (below), b) re-plated onto 22 mm square coverslips for electrophysiology, or c) lysed for biochemical experiments. For re-plating post-transfection, cells were detached using a non-enzymatic cell dissociation solution (Sigma Aldrich, UK) before re-seeding in fresh complete medium.

Western immunoblotting
At 48 h post-transfection, COS-7 cells were washed in PBS and lysed at 4uC in a radio-immunoprecipitation assay (RIPA) buffer with Complete MINI EDTA-free protease inhibitor cocktail (Roche, UK). The cell lysates were then passed through a 22-gauge syringe needle 10 times to shear genomic DNA, and centrifuged at 1000 g av . Supernatants were then incubated at 37uC for 15 min with Laemmli loading buffer containing 20 mM DTT and then heated to 95uC for 2 min. Sample proteins were resolved by SDS-PAGE on 10% Tris-HCl gels for 80 min at 160 V (Mini-Protean cell, BioRad, UK) and then transferred by electrophoresis (100 V for 2 h) onto nitrocellulose membranes (Whatman, UK). Air dried membranes were immersed overnight in blocking buffer (5% nonfat dry milk in Tris-buffered saline (TBS) with 0.1% Tween-20 (TTBS)), washed three times with TTBS and then incubated with the appropriate primary antibody in TTBS for 1 h at 20uC. The membranes were then re-washed with TTBS and incubated for 1 h at 20uC with the appropriate secondary HRP-conjugated antibody (1:1000) in TTBS. After further washing with TTBS, the membranes were treated with Western Lightning enhanced chemiluminescence reagent (Perkin Elmer, UK) and immunoreactive proteins detected by exposure to film (GE Life Sciences, UK).

Sucrose gradient fractionation
As we described recently [21], transiently transfected COS-7 cells were washed in PBS and lysed 48 h post-transfection with MBS (Mes-buffered saline: 25 mM Mes, pH 6.5, 150 mM NaCl) with 1% Triton-X-100 at 4uC. For a single experiment, 9610 cm dishes were used and 150 ml of MBS/Triton-X-100 was added to lyse the cells. Cells were scraped off the dish, passed through a 22gauge needle 10 times to shear genomic DNA and 450 ml of lysate was reserved for use as a control. The remaining 900 ml of lysate was mixed with 900 ml of 90% sucrose/MBS (w/v), placed in a 5 ml polypropylene centrifuge tube (Sorvall) and carefully overlaid with 1.5 ml of 30% sucrose/MBS, followed by 1.5 ml of 5% sucrose/MBS. Gradients were spun at 38,500 rpm (140,000 g av ) in a Sorvall Discovery 100SE ultracentrifuge using an AH-650 rotor for 16 h at 4uC. Post-centrifugation, 15 fractions were taken from top to bottom of the tube and analysed in subsequent Western immunoblotting. To concentrate proteins, fractions were incubated with 25% trichloroacetic acid (final), at 4uC for 30 min. Samples were centrifuged at 14,000 rpm (13,000 g av ) at 4uC for 20 min and the pellets washed twice with ice-cold acetone, ensuring not to disrupt the pellets. Pellets were dried at 42uC for 10 min before re-suspension in 50 ml of MBS and analysed by Western immunoblotting.

Immunocytochemistry
Cells for fluorescence microscopy were re-plated 24 hours posttransfection onto 13 mm coverslips coated with 0.01% poly-Llysine. To preclude fixation artefacts, all imaging experiments of surface expression were performed using a two-step protocol [21]. Briefly, COS-7 cells (48 h post-transfection) were cooled on ice to 4uC and after 10 min, treated with primary antibody diluted in PBS. After 1 h at 4uC, coverslips were washed 3 times with PBS and the cells fixed with 4% (w/v) paraformaldehyde for 20 min at 20uC. Cells were then treated with the appropriate (Cy5 or FITC) fluorophore-conjugated secondary antibody for 1 h at 20uC. In order to detect intracellular epitope expression, cells were permeabilised post-fixation with 0.5% saponin for 10 min at 20uC, prior to incubation with primary antibody. Nuclear staining was performed with DAPI (49,6-diamidino-2-phenylindole; 1 mg/ ml) for 2 min at 20uC, prior to mounting with Prolong Gold Antifade reagent (Invitrogen/Molecular Probes).

Fluorescence deconvolution microscopy and image analysis
Images of cells on coverslips were acquired on a Delta Vision RT (Applied Precision, Image Solutions, UK) restoration microscope using a 660 objective lens and appropriate wavelength filters. The images were collected using a Coolsnap HQ (Photometrics) camera with a Z optical spacing of 0.1 mm. Raw images were then deconvolved using Softworx software and displayed as maximum projections using NIH Image J ((W.S. Rasband, NIH Bethesda, USA; Wright Cell Imaging facility bundle: http://www.uhnres.utoronto.ca/facilities/wcif.htm).
Whole-cell patch-clamp electrophysiology As described previously [21], COS-7 cells were transiently transfected with Ca v 2.2:b 1b : a 2 /d-1:mut3-GFP-pMT2 cDNA in a 3:1:1:0.2 mass ratio and current recordings made 48 h posttransfection. Where a 2 d-1 or mut-3 GFP was omitted, empty pcDNA3.1 vector was substituted to maintain the equivalent mass of DNA. Electrophysiological recordings of barium currents were made from green fluorescent COS-7 cells, using the whole-cell configuration of the patch clamp technique and the following solutions [71].  (20-22uC). An Axopatch 200B amplifier (Molecular Devices, Palo Alto, CA, USA) was used for recordings which were filtered at 2 Hz and digitised at 2-44 kHz using a Digidata 1440A A/D converter (Molecular Devices). Standard current-voltage protocols involved 150 ms sweeps from a holding potential, V h of 280 mV to command voltages of 230 to +65 mV in 5 mV steps. Current density-voltage (I-V) relationships for each cell were fitted with a Boltzmann function: Where, V rev is the reversal potential, V 50 is the voltage for half maximal activation of current, g is the conductance, and k is the slope factor. Data acquisition and analysis was performed using pCLAMP software (version 10, Molecular Devices) and Origin (version 7.0, Microcal, Northampton, MA, USA).

Data analysis
All data are presented as the mean 6 standard error of the mean (S.E.M) for n trials. Statistical analysis was carried out by Student's t-test or ANOVA (one-way with Student-Newman-Keuls (SNK) post hoc correction), as appropriate, using 95% confidence limits (SigmaStat software, Jandel Scientific). Contour mapping was performed using Origin V.8 (OriginLab Corp., MA) on images converted from TIFF format to 2D matrices using the TIFFDump algorithm written by J.S Wadia [72]. Particle analysis was performed on thresholded images using NIH Image J. Inset: data re-plotted using log scale. To facilitate overlay of images from separate cells, the number of particles Np i , of given area (Ap i , (abscissa) in pixel 2 ) is expressed as a percentage of the total (N t where N t = S N Ai ). Note overlap in data, irrespective of pre-treatment with PI-PLC. B. Distribution of fractional coverage represented by PIN-a 2 d particles. Inset: data re-plotted using expanded scale. Here and elsewhere [21], we define fractional coverage as the % of the total particulate area (C t ) within a region of interest (ROI), (not the area of the ROI) accounted for by particles of area Ap i (i.e. Np i .Ap i /C t , where C t~P i~Ap 0 i~1 Np i :Ap i and Ap' is the area of the largest particle in the data set). Using this representation it is possible to discriminate cases where coverage of the total particle area arises from many small particles or a lesser number of larger particles. For example, in the simple situation where there are 4 particles each of size 10 pixel 2 and 1 particle of size 60 pixel 2 , then C t = 100, then for the smaller particles Np i / N t = 0.8 and the fractional coverage = 0.4, for the larger particle Np i /N t = 0.2 and fractional coverage = 0.6. In contrast, if the same total particulate area is comprised of 60 particles each of size 1 pixel 2 and 4 particles each of size 10 pixel 2 , then Np i /N t = 0.94 and the fractional coverage = 0.6, for the larger particles Np i / N t = 0.06 and fractional coverage = 0.4). Note overlap of data, irrespective of pre-treatment with PI-PLC. Particle analysis was performed with Image J, using the adaptive thresholding plug-in, with thresholded images checked visually for accuracy. All data were extracted from 3 images from separate experiments. C. and D. Computer modelling of the effects of particle re-distribution on fractional coverage Fractional coverage graphs (D) were determined for the three particle size distributions shown in C. Note marked, and well-defined effect of particle re-distribution. For simplicity, the distribution curves in C were generated using equations based on a binomial distribution with terms p 4 (black), 6p 2 q 2 (red) and q 4 (blue), (where q = 1-p), respectively. In each case, the number of particles N t was adjusted to give an identical total particulate area, C t = 1610 5 pixel 2 ((N t = pixel 2 , black, red and blue curves, respectively). E. and F. Computer modelling of the effects of a reduction in particle area. In these simulations the number of particles was held constant (Nt = 10 4 ), but the area of each decreased by 50% to mimic a 'stripping' effect such as that which might be seen with PI-PLC. Curves in E. were generated as in C., for the p 4 binomial distribution. From comparisons of the size distribution and fractional coverage determined experimentally (A, B) and predicted from simulations (C-F), there is no evidence that PI-PLC pre-treatment has any effect on PIN-a 2 d particle properties. (TIF) GPI-modification for dataset and a 2 d proteins shown in A inferred using Big-Pi predictor software (http://expasy.org/tools/). Proteins with positive or negative GPI modification potential are shown in blue and red, respectively. Asterisks denote proteins where the v site differs from that inferred. Right panel detailed sequence comparison of inferred (red lettering) and predicted (asterisks) v sites. In most cases the inferred v site is very close (,2 residues) to that found experimentally. C. Analysis of potential upstream GPI-anchoring motifs in the delta subunit of WT a 2 d-1 (or PIN-a 2 d)(blue) and PIN-a 2 d-PIN TMI (red). Here, the GPI anchoring potential was determined (using Big-Pi [29]) as a function of successive truncation (1 residue at a time) of the carboxy terminus. Note: based on the length of the GPI-anchoring motif, any v site is predicted to lie 20-30 residues upstream of the position of the indicated carboxy-terminal residue (abscissa). For simplification, the carboxy-terminal sequences have been renumbered starting at residue 922 in WT a 2 d-1 as shown in the corresponding sequences (i. and ii. right panel). For PIN-a 2 d, the reported GPI-anchoring motif is shown in blue lettering. For PINa 2 d-PIN TMI green lettering denotes residues derived from PIN-G. In i. and ii., the grey boxes denote hydrophobic regions. With the exception of sequences near the junction of the a 2 and d delta subunits, all regions have a much lower GPI-modification potential than the WT a 2 d C-terminus suggesting the likely absence of additional upstream GPI-anchoring motifs unmasked by proteolytic cleavage. Note, the low GPI-modification potential of both the non-truncated PIN-a 2 d and PIN-a 2 d-PIN TMI . (TIF)

Supporting Information
Table S1 Biophysical properties of Ca v 2.2/b 1b channels coexpressed with WT a 2 d-1, PIN-a 2 d, PIN-a 2 d-PIN TMI and PIN-d. I max is the maximum peak current density. Individual current density-voltage plots were fitted with a Boltzmann function: where V rev is the reversal potential, V 50,act is the voltage for half maximal activation of current, g is the conductance, and k is the slope factor. Statistical analysis used Students unpaired t-test. Asterisks denote statistically significant differences from (2)a 2 d-1, as follows: * = P,0.05, *** = P,0.001. n is the number of cells tested per treatment. (DOC)