Quantification of integrin receptor agonism by fluorescence lifetime imaging

Both spatiotemporal analyses of adhesion signalling and the development of pharmacological inhibitors of integrin receptors currently suffer from the lack of an assay to measure integrin-effector binding and the response of these interactions to antagonists. Indeed, anti-integrin compounds have failed in the clinic because of secondary side effects resulting from agonistic activity. Here, we have expressed integrin-GFP and effector-mRFP pairs in living cells and quantified their association using fluorescence lifetime imaging microscopy (FLIM) to measure fluorescence resonance energy transfer (FRET). Association of talin with β1 integrin and paxillin with α4 integrin was dependent on both the ligand and receptor activation state, and was sensitive to inhibition with small molecule RGD and LDV mimetics, respectively. An adaptation of the assay revealed the agonistic activity of these small molecules, thus demonstrating that these compounds may induce secondary effects in vivo via integrin activation. This study provides insight into the dependence of the activity of small molecule anti-integrin compounds upon receptor conformation, and provides a novel quantitative assay for the validation of potential integrin antagonists.


Introduction
Integrins are cell-adhesion receptors that provide physical support for tissues and enable directed migration during development and tissue homeostasis (van der Neut et al., 1996;Wagner et al., 1996). At the cellular level, integrins spatially compartmentalise signalling events by tethering the contractile cytoskeleton to the plasma membrane and indirectly modulating multiple signalling networks. In mammals, genes encoding 18 ␣ and 8 ␤ integrins produce polypeptides that combine to form 24 heterodimeric receptors (Hynes, 2002), 12 of which contain the ␤1 subunit. Both subunits are noncovalently associated, type I transmembrane proteins with large extracellular and mostly short cytoplasmic domains. In recent years, substantial progress has been made towards defining the conformational changes that underpin integrin affinity regulation and identifying the effector proteins that initiate integrin signalling (Luo et al., 2007;Humphries, 2000). Transmembrane or cytoplasmic domain separation, triggered either by the binding of FERM domain-containing cytoplasmic proteins (such as talin and myosin X) (Zhang et al., 2004;Tadokoro et al., 2003;Garcia-Alvarez et al., 2003) or extracellular ligands, is currently thought to be the mechanism for the bidirectional transmembrane signal transduction that regulates adhesion (Kim et al., 2003).
In patients with inflammatory and neoplastic diseases, aberrant integrin function perturbs cellular trafficking and causes dysregulation of cellular differentiation (Mousa, 2002). Within the past decade, the first generation of anti-integrin drugs has been approved for human therapy (Leclerc, 2002). Some of these agents are small molecule mimetics of the acidic peptide active sites found in most integrin ligands (e.g. RGD and LDV). Although these agents are potent, competitive inhibitors of integrin-ligand binding in vitro and in vivo, it is now evident that they frequently retain the agonistic properties of their parent ligands. This activity can lead to biological side effects, such as platelet dysfunction, and can consequently impair the drug development process (Hantgan et al., 2007). As a result, there is a pressing need for a reporter assay to measure the agonistic activity of integrin-binding ligands and small molecules in situ. Here, we present such a system, which uses fluorescence resonance energy transfer (FRET) to measure direct integrin-effector binding in intact cells. This assay also has the potential to detect and quantify integrin signalling during normal biological processes, such as migration and differentiation.

Results and Discussion
Initially, a full-length human ␤1 integrin construct was Cterminally tagged with green fluorescent protein (GFP). Although other ␤-integrin-GFPs have been described (Ballestrem et al., 2001), ␤1-integrin-GFP has not been previously generated because the ubiquitous distribution and high endogenous expression of ␤1 integrin in most cells makes it difficult to express. This construct was therefore stably expressed to endogenous levels in immortalised ␤1-integrin-null mouse embryonic fibroblasts Both spatiotemporal analyses of adhesion signalling and the development of pharmacological inhibitors of integrin receptors currently suffer from the lack of an assay to measure integrin-effector binding and the response of these interactions to antagonists. Indeed, anti-integrin compounds have failed in the clinic because of secondary side effects resulting from agonistic activity. Here, we have expressed integrin-GFP and effector-mRFP pairs in living cells and quantified their association using fluorescence lifetime imaging microscopy (FLIM) to measure fluorescence resonance energy transfer (FRET). Association of talin with ␤1 integrin and paxillin with ␣4 integrin was dependent on both the ligand and receptor activation state, and was sensitive to inhibition with small molecule RGD and LDV mimetics, respectively. An adaptation of the assay revealed the agonistic activity of these small molecules, thus demonstrating that these compounds may induce secondary effects in vivo via integrin activation. This study provides insight into the dependence of the activity of small molecule anti-integrin compounds upon receptor conformation, and provides a novel quantitative assay for the validation of potential integrin antagonists.
(MEFs) and its functional activity tested by a combination of confocal microscopy (to confirm its presence in adhesion complexes) and cell attachment and spreading assays (supplementary material Fig. S1). To identify effectors that might undergo FRET with ␤1 integrin, four different candidate adhesion complex components were tested, each of which has been reported to bind directly to ␤ integrins: the talin head domain (residues 1-433) (Garcia-Alvarez et al. 2003), talin rod domain (residues 1984-2344) (Xing et al., 2001), ␣-actinin (Otey et al., 1990) and paxillin (Schaller et al., 1995). ␤1-integrin-GFP MEFs were transfected with mRFP conjugates of each protein, plated onto a fibronectin substrate and integrin-effector binding analysed using FLIM to measure FRET. This technique enables visualisation and quantification of protein-protein interactions by analysis of the donor lifetime decay kinetics (see Materials and Methods for full description of technique and analysis) (Parsons et al., 2005).
Specific interactions between ␤1 integrin and the talin rod domain and ␣-actinin were detected, but no interaction was observed for the talin head domain or paxillin (Fig. 1A). The substrate dependence of the interaction between the talin rod domain and ␤1 integrin was examined. As shown in Fig. 1B, a modest interaction was detected on both collagen and laminin substrates, but binding was most prominent on fibronectin, where FRET was localised to focal adhesion structures (Fig. 1A). No FRET was detected between the integrin and any of the acceptors in cells plated on a nonintegrin-binding poly-L-lysine substrate (PLL; Fig. 1B and data not shown).
Although not the primary focus of this study, these data add significantly to our understanding of integrin effectors. Despite some controversy in the literature, it is apparent that the talin rod and ␣-actinin are able to interact directly with ␤1 integrin. From other studies, the ability of the talin head to associate with ␤1 integrin is unequivocal (Garcia-Alvarez et al., 2003), but we failed to detect the association by FRET-FLIM. We speculate that the functional role of the head domain may be transient, and/or restricted to early adhesion complexes only.
Having established that ␤1-integrin-GFP associates with talin-rod-mRFP, the activation state dependence of the interaction was examined. To constrain Journal of Cell Science 121 (3) Fig. 1. Interaction of ␤1 integrin-GFP and talin-rod-RFP by FRET. (A) ␤1 integrin-GFP fibroblasts were transfected with plasmids encoding mRFP conjugates of the N-terminal 433 amino acids of talin (talin 433), the Cterminal rod domain of talin (talin rod), paxillin or ␣actinin and plated onto fibronectin. Images show the GFP multiphoton intensity image and (where appropriate) corresponding widefield CCD camera image of the RFP expression. Control (GFP integrin alone) image demonstrates a normal GFP lifetime in the absence of acceptor. Lifetime images mapping spatial FRET across the cells are depicted using a pseudocolour scale (blue, normal GFP lifetime; red, FRET). The bar graph represents average FRET efficiency of seven cells over three independent experiments. Error bars indicate s.e.m. (B) ␤1 integrin-GFP fibroblasts were transfected with talin-rod-mRFP. Cells were plated onto coverslips coated with poly-L-lysine (PLL), collagen I (COL), or laminin-1 (LN) and allowed to attach and spread for 2 hours. Control (untransfected) cells or co-transfected cells were then imaged by FLIM to detect FRET. Lifetime measurements were acquired and depicted as in A. Histogram analysis of the spread of relative FRET efficiency is an average of more than eight cells from three different experiments. the location of the clustered integrin for analysis, ␤1-integrin-GFP MEFs expressing talin-rod-mRFP were plated onto PLL and incubated with 4.2 m beads coated with fibronectin ligand or the monoclonal antibodies 12G10 (which detects the high affinity or primed state and stimulates ligand binding) (Mould et al., 1995;Mould et al., 2002), mAb13 (which detects non-ligand-occupied integrin and which blocks ligand binding) (Akiyama et al., 1989) or K20 (which is non-functionaltering and detects all conformational states) (Amiot et al., 1986;Mould et al., 2005). As shown in Fig. 2A, an interaction between integrin and talin rod was only observed with fibronectin-or 12G10coated beads. Confocal analysis of these cells also revealed actin recruitment to the bead structures in those cases where FRET was detected (Fig. 2B). These data suggest that controlling integrin conformation and therefore the activation of the extracellular domain of the integrin, either by native ligand or antibodies alone, can drive recruitment of both talin and actin to the integrin cytoplasmic domain.
To confirm the use of FRET-FLIM for detecting integrin-effector binding, and to generate an assay that might be used to test small molecule inhibitors of integrin function, the previously characterised association between ␣4 integrin and paxillin (Liu et al., 1999;Goldfinger et al., 2003) was selected. FLIM was used to analyse FRET between ␣4-integrin-GFP and paxillin-mRFP in both mouse B16F1 and human A375SM melanoma cells. A localised interaction between integrin and paxillin was detected in B16F1 cells plated on fibronectin or an ␣4-integrin-binding fragment of fibronectin (H/120), but not on PLL (Fig. 3A,B), confirming the results of the previous biochemical analyses. This interaction was also detected in cells plated on VCAM-1 (data not shown). FRET was significantly decreased when two small molecule, LDV ligand mimetic inhibitors of ␣4␤1 integrin [S976162 and S9916197 (Gläsner et al., 2005;Lin et al., 1999) see supplementary material Table S1 and Fig. S2 for characterisation] were added to pre-spread cells (Fig. 3B). Interestingly, the remaining interacting population demonstrated a spatial shift from the cell periphery to the central basal region (just below the nucleus). A similar reduction in ␣4 integrin and paxillin binding following treatment with these compounds was also seen in live human A375SM melanoma cells plated onto an activated endothelial cell layer (Fig. 3C). These data demonstrate that binding of ␣4-integrin to paxillin, detected by FRET, is both ligand dependent and sensitive to inhibition with small molecule antagonists.
To test whether the small molecule ␣4␤1 integrin inhibitors could also act as agonists, A375SM cells transfected with ␣4integrin-GFP and paxillin-RFP were plated onto PLL and incubated with K20-coated beads. As shown in Fig. 2, these beads cluster, but do not activate integrins. Cells were then treated with vehicle control (DMF) or small molecule ␣4␤1 integrin inhibitors and the interaction between ␣4 integrin and paxillin assessed by FLIM. Cells treated with either compound demonstrated a significant increase in ␣4-integrin-paxillin binding at the K20 bead interface (Fig. 4A). This effect was not seen with mAb13 beads, which would be expected to retain the ␤1 integrin in an integrin-GFP fibroblasts were transfected with talin-rod-mRFP. Cells were plated onto coverslips coated with poly-L-lysine (PLL), and allowed to attach and spread for 2 hours. Cells were then incubated with 4.2 m beads coated with FN or anti-␤1 antibodies (12G10, mAb13 or K20) for 30 minutes and subsequently imaged using FLIM as in Fig. 1. Bar graph represents average FRET efficiency of 16 cells per ligand. Efficiency was calculated using a mask of constant area around each bead region for local analysis of FRET. Error bars indicate s.e.m. (B) Cells were prepared as in A, and following incubation with beads, samples were fixed, permeabilised and stained with phalloidin-Alexa Fluor 568 to detect F-actin. Cells were then imaged by confocal microscopy.
inactive conformation and prevent ligand binding. Moreover, the effects of the compounds were dose dependent in a rank order range that paralleled their anti-adhesive activity (Fig. 4B, supplementary material Table S1, Fig. S2). We interpret these data to indicate that K20 immobilisation of ␤1 integrin allows the soluble compounds to act as agonists for ␣4␤1 integrin and trigger an activation response in the form of recruitment of cytoskeletal proteins. To extend these findings, an RGD ligand mimetic small molecule inhibitor of ␣5␤1 and ␣V␤3 integrins (V0519) (Peyman et al., 2000) was tested for its effect on binding of ␤1-integrin-GFP and talin-rod-mRFP. ␤1-integrin-GFP cells expressing talin-rod-mRFP were plated onto PLL-coated coverslips and incubated with K20-coated beads in the presence or absence of V0519. As shown in Fig. 4D, V0519 substantially increased the interaction of ␤1 integrin and talin at the bead interface.
In summary, we have established assays to detect integrineffector binding by FRET-FLIM. The availability of these assays will not only enable spatiotemporal studies of integrin signalling, but they could also form the basis for low-and highthroughput screening of small molecule inhibitors in the pharmaceutical industry. The application of direct imaging techniques to study small molecule compound effectors in situ may provide an excellent platform for future identification of therapeutic compounds that either possess or lack agonistic activity.
Journal of Cell Science 121 (3) Fig. 3. Ligand binding to the ␣4 integrin extracellular domain regulates its association with paxillin. (A) B16F1 mouse melanoma cells were transfected with ␣4-integrin-GFP and paxillin-mRFP. Cells were then plated onto FN-or PLLcoated coverslips and imaged by multiphoton FLIM as before. (B) The same cells were plated onto H120-coated coverslips followed by treatment with DMF vehicle control or stated small molecule inhibitors for 30 minutes and imaged by multiphoton FLIM. Histogram analysis of relative spread of FRET efficiencies is a mean of 18 cells per treatment compared with ␣4-integrin-GFP alone control. (C) Human A375-SM melanoma cells were transfected with ␣4integrin-GFP and paxillin-mRFP and plated onto a monolayer of TNF␣-treated activated HUVEC cells. Cells were allowed to adhere for 60 minutes, and were then treated with DMF vehicle control or stated small molecule inhibitors for 30 minutes. Cells were then imaged using multiphoton FLIM as before. Imaging integrin activation by FRET

Plasmid constructs
Generation of talin-mRFP constructs Full-length mouse talin-1 cDNA was a gift from David Critchley (University of Leicester, Leicester, UK). A 1080 bp fragment corresponding to amino acids 1984-2344 was amplified by PCR using the GGGGGGGAATTCGCTGTGTCTG -GTATCATTGC (forward, F) and AAAAAAGCGGCCGCATTGTTCCTCAA -AGTTC (reverse, R) primers to generate a 5Ј EcoRI and a 3Ј NotI restriction site. This region in the talin rod domain has been reported to contain an integrin-binding site (Fragment G) (Tremuth et al., 2004). The PCR product was then cloned into the pcDNA-RFP-C vector (a gift from Roger Tsien, UCSD, CA) to generate a C-terminal monomeric red fluorescent protein tag. Similarly, a 1299 bp fragment corresponding to amino acids 1-433 of talin was amplified by PCR and fused to a C-terminal monomeric RFP tag. This region in the head of the talin protein also contains an integrin-binding site (Calderwood et al., 1999). The primers used were GCAGAATTCCATGGTTGCGCTTTCGCTGAAGGCGCAG (F) and ACATTT -GC GGCCGCTTCCTTTACCGTCCTGAAGGACTGTTGA (R).

Generation of ␤1-GFP
Full-length human ␤1A integrin cDNA was the gift of Ken Yamada (NIDCR, NIH, Bethesda, MD). To mutate the stop codon before fusion with a 3Ј fluorescent tag, and generate a novel 3Ј ApaI restriction site, a fragment corresponding to residues 578-799 of the mature sequence was generated by PCR using the primers TTGCAAGTGTCGTGTGTGTG (F) and GTGGATCCCGGGCCCCCCCC CC -TCCTTTTCCC (R). The PCR product was then digested with SpeI and ApaI. The remaining 5Ј portion of ␤1 was excised from ␤1-pECE using KpnI and SpeI, and both fragments simultaneously ligated into pcDNA3 digested with KpnI and ApaI. The fidelity of the resulting construct was confirmed by sequencing (GenBank code XM_005799.1). The mutated, full-length ␤1 integrin construct was then cloned into pEGFP N1 (Clontech) using KpnI and ApaI. The initial 12 amino acid linker between the end of the mature ␤1 integrin sequence and the start codon of the GFP tag proved Fig. 4. Small molecule inhibitors induce integrin signalling. (A) A375-SM cells were transfected with ␣4-integrin-GFP and paxillin-mRFP and plated onto coverslips coated with PLL. Cells were then incubated with K20 or mAb13 antibody-coated beads for 30 minutes, followed by treatment with DMF or small molecule inhibitor as indicated. Cells were then imaged using multiphoton FLIM. Cumulative FRET efficiency data from a masked pre-set region around the bead in ten cells per treatment is shown in the histogram. (B) Histogram demonstrating FRET efficiency of ␣4-paxillin association at K20 beads over a dose-response range of compounds S976162 or S9916197. Results are means ± s.e.m. (C) ␤1-integrin-GFP fibroblasts were transfected with talin-rod-mRFP. Cells were then plated onto coverslips coated with PLL and allowed to attach and spread for 1 hour. Cells were then incubated with beads coated with K20 for 30 minutes, and treated with either the vehicle control DMF or V0519 compound, and subsequently imaged using FLIM as before. Cumulative FRET efficiency data from a masked pre-set region around the bead in 12 cells per treatment is shown in the histogram. insufficient for mammalian cell expression. Therefore an 18 amino acid linker was constructed. A top strand and a bottom strand oligonucleotide were designed and annealed together, digested with ApaI and AgeI and used to replace the 3Ј end of the ␤1-integrin-pEGFP N1 construct. The resulting linker sequence between the 3Ј end of the mature ␤1 integrin coding sequence and the start codon of GFP was GGGGARRRGQAGDPPVAT. To enable greater flexibility with the colour of the fluorescent tags, the new construct was initially cloned into pHcRed N1 (Clontech) and when a GFP tag was required, the RFP fluorophore was excised using AgeI and NotI, and the GFP sequence from pEGFP N1 cloned into the pHcRed backbone using the same restriction sites. The fidelity of the ␤1 integrin constructs was verified by sequencing. These constructs were successfully expressed transiently in MEF 7929 (␤1 -/-) cells. To maximise the level of stable incorporation of the ␤1-integrin-GFP DNA in ␤1 -/cells, an additional retroviral construct was used. The ␤1-integrin-pHcGreen coding sequence was excised from the pHc backbone using EcoRI and NotI. This coding sequence was ligated into the pFBneo vector (Stratagene) using the same restriction sites, resulting in a construct that produced ␤1-integrin-GFPexpressing retroviral particles when expressed by amphotropic AM12 packaging cells (a gift from Ian Hart, Cancer Research UK, London, UK).

Generation of ␣4-GFP
A full-length clone of human ␣4 integrin in pBluescript (from Yoshi Takada, Dana-Farber Cancer Center, Boston, MA) was used as a backbone. For the generation of ␣4-GFP, the coding region of the ␣4-pBluescript was removed and an AgeI (PinAI) site inserted at the 3Ј end using PCR. The ␣4 cDNA already had a 5Ј SalI site and these sites were used to subclone ␣4 into pECFP-N1. The CFP cassette was then replaced with GFP-N1 to give ␣4-pEGFP-N1. This was then digested with NheI and NotI positioned 5Ј and 3Ј to the entire ␣4-pEGFP and the hyg + cassette from pcDNA3.1 hyg + ligated using the same restriction sites. PCR primers used to insert a 3Ј AgeI (PinAI) site into the vector backbone were CAA CAG TAA AAG CAA TGA TGA TGG CGG CGG CGG CGG ACC GGT GGA CTT CTT TCA AAT TGA GAG AAT GG and GGA TAG ATA TTA GCT TTC TCC.

Generation of mRFP-paxillin
Full-length paxillin in pEGFPC2 was a gift from Vic Small (IMBA, Vienna, Austria). PCR mutagenesis was carried out using the primers below to insert an EcoRI site restriction site at the 3Ј end and a 5Ј HindIII site. The primers used were GGA TCC AAG CTT GCC GCC ATG GAC GAC CTC GAC GCC CTG C (F) and GGA TCC GAA TTC CTA GCA GAA GAG CTT GAG GAA GC (R). The PCR product was then sequentially digested and ligated into pcDNA3.1-mRFP-N (from Roger Tsien, UCSD, CA).
Generation of ␤1 -/-MEFs Immortalised ␤1 integrin -/murine embryonic fibroblasts were obtained by crossing Immorto mice (Charles River Laboratories, Wilmington, DE) carrying the thermolabile large T antigen H-2KtsA58 (Jat et al., 1991) with mice in which exon 1 of the ␤1 integrin gene was flanked by loxP sites (Graus-Porta et al., 2001) (a gift of Uli Muller, Friedrich Miescher Institute, Basel, Switzerland) and genotyped by tail-tip PCR. Mice that were homozygous for the wild-type or floxed ␤1 integrin allele and carried at least one Immorto allele were interbred and E13.5 embryos dissected for MEF preparation. Isolated MEFs were cultured at 33°C in the presence of IFN␥ (Sigma; 20 U/ml) for five passages until all non-immortalised cells had senesced. Adenovirus-Cre (AdCreM1; Microbix Inc.) was added at an MOI of 40 overnight to subconfluent cultures of a random subset of cells derived from ␤1 (fl/fl)/Immorto embryos. Deletion of ␤1 integrin expression was monitored by flow cytometry with mAb KMI6 (BD Pharmingen) and confirmed by PCR and western blot. Clones from the ␤1 -/-MEFs were derived by limited dilution and deletion of ␤1 integrin was reconfirmed for each clone by flow cytometry.

Retroviral transduction
AM12 packaging cells were maintained in Dulbecco's Modified Eagle's Medium (DMEM) containing 10% FCS. Before transfection, cells were seeded overnight into six-well plates and transfected with 1 g/well ␤1-GFP-pFBneo using FuGene 6 reagent (Roche) according to the manufacturer's instructions. 24 hours after transfection, the medium was changed and the transfected cells allowed to condition the fresh medium for a further 48 hours. The conditioned medium was removed, filtered through a 2 m pore filter, mixed 50:50 with MEF growth medium and placed onto 50% confluent ␤1 -/-MEFs. 24-48 hours after adding the retroviruscontaining medium, transduced cells were selected with 0.5 mg/ml neomycin for 10 days then assayed for GFP expression by flow cytometry using untransduced cells as the negative control. In addition to the parent population expressing a mixed level of ␤1-GFP, a subpopulation of high expressers was isolated by flow cytometry and nine individual clones were derived by limited dilution from these cells.

Cell culture and transient transfection
MEFs were routinely cultured in DMEM, 10% FCS (Sigma) supplemented with 20 U/ml mouse IFN␥ (Sigma) at 33°C. For all experiments, MEFs were cultured in growth medium without IFN␥ at 37°C to switch off the immortalising gene. B16F1 mouse melanoma cells and A375-SM human melanoma cells were both cultured in DMEM (Sigma) supplemented with 10% FCS, glutamine and penicillin-streptomycin at 37°C. For FLIM experiments, ␤1-GFP-MEFs, B16F1 or A375-SM melanoma cells were seeded overnight in six-well plates and transfected with 1 g/well of indicated plasmid DNA using FuGene 6 according to the manufacturer's instructions. Cells typically were left to express constructs for 24-72 hours following transfection (as assessed by microscopy) and were used between 24 and 48 hours post transfection for all FLIM studies. Before imaging, cells were plated on glass coverslips or Mattek dishes (for line experiments) coated with extracellular matrix proteins or poly-L-lysine (Sigma) as indicated.
Bead coating and incubation 4 m latex beads (Dynal) were washed in PBS and incubated with diluted proteins or antibodies as stated in figure legends at indicated concentrations. Anti-␤1 monoclonal antibodies used in the study are: K20 (neutral); (Amiot et al., 1986;Mould et al., 2005), 12G10 (activating) (Mould et al., 1995;Mould et al., 2002) or mAb13 (inactivating) (Akiyama et al., 1989) and were all generated in-house as previously described. Beads were left overnight at 4°C to allow even protein binding, and subsequently washed three times and re-suspended in PBS to form a 50% slurry final concentration. Plated cells were incubated with 5 l/ml of beads in growth media and incubated at 37°C for 30 minutes. Cells were then washed three times in medium and either fixed for FLIM analysis or imaged live in phenol-red free growth media at 37°C. For integrin antagonist studies, cells were pre-treated with concentrations and times specified in figure legend with V0519 (Peyman et al., 2000), S9916197 (Gläsner et al., 2005) or S976162 (Lin et al., 1999) and subjected to FLIM analysis.

Fluorescence lifetime measurements by time-correlated single photon counting (TCSPC) FLIM
Here we used FLIM to measure FRET between protein pairs, which allows the determination of spatial protein interactions (Parsons et al., 2005). Time-domain FLIM was performed with a multi-photon microscope system as described previously (Parsons et al., 2005;Peter et al., 2005). The system is based on a modified Bio-Rad MRC 1024MP workstation, comprising a solid-state-pumped femtosecond Ti:Sapphire (Tsunami, Spectra-Physics) laser system, a focal scan-head and an inverted microscope (Nikon TE200). Enhanced detection of the scattered component of the emitted (fluorescence) photons was afforded by the use of fast response (Hamamatsu R7401-P) non-descanned detectors, developed in-house, situated in the re-imaged objective pupil plane. Fluorescence lifetime imaging capability was provided by time-correlated single photon counting electronics (Becker & Hickl, SPC 700). A 40ϫ objective was used throughout (Nikon, CFI60 Plan Fluor NA 1.3) and data collected at 500±20 nm through a bandpass filter (Coherent Inc., 35-5040). Laser power was adjusted to give average photon counting rates of the order 10 4 -10 5 photons (0.0001 to 0.001 photon counts per excitation event) to avoid pulse pile up. Acquisition times up to 300 seconds at low excitation power were used to achieve sufficient photon statistics for fitting, while avoiding either pulse pile-up or significant photobleaching. Excitation was at 890 nm. Widefield acceptor (mRFP) images were acquired using a CCD camera (Hammamatsu) at <100 millisecond exposure times.

Analysis of data for FRET experiments
Data were analysed as previously described (Parsons et al., 2005;Prag et al., 2007). Briefly, bulk measurements of FRET efficiency (i.e. intensity-based methods) cannot distinguish between an increase in FRET efficiency (i.e. coupling efficiency) and an increase in FRET population (concentration of FRET species) since the two parameters are not resolved. Measurements of FRET based on analysis of the fluorescence lifetime of the donor can resolve this issue when analyzed using multiexponential decay models. For measurements of bulk interactions (i.e. where only single exponential decays are fit to the data), measured efficiencies will appear significantly lower because of the assumption that all donors are associated with one or more acceptors. The assumption that non-interacting and interacting fractions are present allows the determination of the efficiency of interaction. The FRET efficiency is related to the molecular separation of donor and acceptor and the fluorescence lifetime of the interacting fraction by: n fret = [R 0 6 / (R 0 6 +r 6 )] = 1n fret / d , where R 0 is the Förster radius, r the molecular separation, n fret is the lifetime of the interacting fraction and d the lifetime of the donor in the absence of acceptor. n fret and d can also be taken to be the lifetime of the interacting fraction and noninteracting fraction, respectively. All data were analysed using TRI2 software [developed by Paul Barber, Gray Cancer Institute, London, UK (Prag et al., 2007); supplementary material Fig. S3]. For the analysis of FRET at the bead interface, a mask of constant defined pixel size was placed over individual bead areas and the lifetimes analysed and calculated within that area alone (see supplementary material Fig. S3). Histogram data presented here are plotted as mean FRET efficiency from >10 cells per sample. Average FRET efficiency is shown ± s.e.m. ANOVA was used to test statistical significance between different populations of data. Lifetime images of example cells are presented using a pseudocolour scale whereby blue depicts normal GFP lifetime (no FRET) and red depicts lower GFP lifetime (areas of FRET).