PERSPECTIVES

The potential of optical proteomic technologiesto individualize prognosis and guide rationaltreatment for cancer patients

Muireann T. Kelleher & Gilbert Fruhwirth & Gargi Patel & Enyinnaya Ofo &Frederic Festy & Paul R. Barber & Simon M. Ameer-Beg & Borivoj Vojnovic &Cheryl Gillett & Anthony Coolen & Gyrgy Kri & Paul A. Ellis & Tony Ng

Received: 21 May 2009 /Accepted: 28 August 2009 /Published online: 16 September 2009# The Author(s) 2009. This article is published with open access at Springerlink.com

Abstract Genomics and proteomics will improve outcomeprediction in cancer and have great potential to help in thediscovery of unknown mechanisms of metastasis, ripe fortherapeutic exploitation. Current methods of prognosisestimation rely on clinical data, anatomical staging andhistopathological features. It is hoped that translationalgenomic and proteomic research will discriminate moreaccurately than is possible at present between patients witha good prognosis and those who carry a high risk ofrecurrence. Rational treatments, targeted to the specificmolecular pathways of an individuals high-risk tumor, areat the core of tailored therapy. The aim of targeted oncologyis to select the right patient for the right drug at preciselythe right point in their cancer journey. Optical proteomics

uses advanced optical imaging technologies to quantify theactivity states of and associations between signalingproteins by measuring energy transfer between fluoro-phores attached to specific proteins. Frster resonanceenergy transfer (FRET) and fluorescence lifetime imagingmicroscopy (FLIM) assays are suitable for use in cell linemodels of cancer, fresh human tissues and formalin-fixedparaffin-embedded tissue (FFPE). In animal models,dynamic deep tissue FLIM/FRET imaging of cancer cellsin vivo is now also feasible. Analysis of protein expressionand post-translational modifications such as phosphoryla-tion and ubiquitination can be performed in cell lines andare remarkably efficiently in cancer tissue samples usingtissue microarrays (TMAs). FRET assays can be performed

M. T. Kelleher :G. Fruhwirth :G. Patel : E. Ofo :S. M. Ameer-Beg :B. Vojnovic : T. Ng (*)Richard Dimbleby Department of Cancer Research, RandallDivision & Division of Cancer Studies, Kings College London,2nd Floor, New Hunt House, Guys Medical School Campus,London SE1 1UL, UKe-mail: tony.ng@kcl.ac.uk

M. T. Kelleher : P. A. EllisDepartment Medical Oncology, Guys Hospital,London SE1 9RT, UK

F. FestyBiomaterial, Biomimetics & Biophotonics Research Group,Kings College London,London, UK

P. R. Barber : B. VojnovicGray Institute for Radiation Oncology & Biology,University of Oxford,Old Road Campus Research Building, Roosevelt Drive,Oxford OX3 7DQ, UK

C. GillettGuys & St Thomas Breast Tissue & Data Bank,Kings College London, Guys Hospital,London SE1 9RT, UK

A. CoolenDepartment of Mathematics, Kings College London,Strand Campus,London WC2R 2LS, UK

G. KriVichem Chemie Research Ltd.,Herman Ott utca 15,Budapest, Hungary

G. KriPathobiochemistry Research Group of Hungarian Academy ofScience, Semmelweis University,Budapest 1444, Bp 8. POB 260,Hungary

Targ Oncol (2009) 4:235252DOI 10.1007/s11523-009-0116-y

to quantify protein-protein interactions within FFPE tissue,far beyond the spatial resolution conventionally associatedwith light or confocal laser microscopy. Multivariate opticalparameters can be correlated with disease relapse forindividual patients. FRET-FLIM assays allow rapid screen-ing of target modifiers using high content drug screens.Specific protein-protein interactions conferring a poorprognosis identified by high content tissue screening willbe perturbed with targeted therapeutics. Future targeteddrugs will be identified using high content/throughput drugscreens that are based on multivariate proteomic assays.Response to therapy at a molecular level can be monitoredusing these assays while the patient receives treatment:utilizing re-biopsy tumor tissue samples in the neoadjuvantsetting or by examining surrogate tissues. These technolo-gies will prove to be both prognostic of risk for individualswhen applied to tumor tissue at first diagnosis andpredictive of response to specifically selected targetedanticancer drugs. Advanced optical assays have greatpotential to be translated into real-life benefit for cancerpatients.

Keywords Imaging .Molecular diagnostics . FRET. FLIM .

Personalized medicine . Breast cancer . Biomarker .

Tissue microarray . Optical proteomics

Introduction

The majority of cancer-related morbidity and deaths areas a consequence of the dissemination and growth ofsecondary metastatic tumors [1]. The clinical and molec-ular heterogeneity of cancer currently presents clinicianswith difficult problems when choosing adjuvant treatmentfor individual patients. Oncologists forecast the likelyprogression of cancer, yet wish to better predict whichpatients will respond to therapy. Both cytotoxic chemo-therapy and biologically targeted drugs are prescribed forsome patients based on existing markers. Nonetheless,tailored anticancer therapy is in its infancy. It mustcontinue to evolve and progress in order to enable rationalindividualized treatment. Oncology is slowly movingaway from empiricism, toward rational personalizedcancer treatment.

At initial diagnosis, prognostic markers estimatewhether the transition to metastatic disease is likely tooccur. Adjuvant systemic chemotherapy aims to eradicatemicro-metastases, thus reducing recurrence rates andcontributing to surgical cure [25]. However the risk ofdistant relapse cannot be eliminated. Furthermore therapyhas associated toxicity [6] and high financial cost.Clinicians need to accurately identify patients at greatestrisk of metastasis, in order to appropriately direct

chemotherapy and targeted treatments. Adjuvant Onlineis an open-access web-based tool (www.adjuvantonline.com)that predicts 10-year outcomes for solid tumors with andwithout systemic therapy [7]. From patient information andtumor characteristics the program calculates a prognosticestimate. Adjuvant online cannot categorize individualpatients. Most established estimates of prognosis do notinclude many characteristics of individual tumor biology,as the vast majority of putative prognostic factors havenot been established nor validated in large series. Weurgently need cost-effective biomarkers to identify theindividual patients at high risk of recurrence and choosethe specific therapy to which they are most likely torespond.

Adjuvant cytotoxic chemotherapy improves survival forsome patients with solid malignancy, however many ofthose treated derive no benefit at all. Improved understand-ing of cancer cell signaling has resulted in therapeuticagents against specific tumor targets. These have revolu-tionized cancer treatment, most elegantly exemplified bythe use of trastuzumab, a humanized monoclonal antibodyagainst the extracellular domain of human epidermalgrowth factor receptor type 2 (HER2/ErbB2), in patientswith invasive breast cancers that overexpress HER2.Despite the myriad of newly designed therapeutics, non-targeted cytotoxic chemotherapy is the mainstay of theadjuvant therapy, rather than a rational approach based onindividual tumor biology. This is largely because ofdifficulties in selecting patients most likely to respondto each drug. As an example, when trastuzumab is givenas a single agent for first-line treatment of ErbB2-overexpressing metastatic breast cancer, it is associatedonly with a 40% objective response rate [8]. For drugstargeting the human epidermal growth factor receptor type1 (EGFR /HER1/ErbB1), patient selection is even moredifficult. Strategies are urgently required to focus targetedtherapeutics specifically on the tumors capable of responding,thus sparing patients unnecessary toxicity, and significantlyreducing drug costs.

United Kingdom cancer survival rates are inferior tosome other European countries, partly resulting from therelatively limited use of new systemic therapies [9].Expensive targeted drugs place considerable financialburden on any health care system. The majority of patientsreceiving such drugs do not experience either restrainedtumor growth or prolonged survival. The financial burdenof targeted therapeutics will continue to rise unless rational-use strategies are developed. In addition to the financialconsiderations, it is hoped that toxicity could be spared withthe use of improved prognostics. Any patient whose tumorreceives a good prognosis will not receive unnecessary,potentially toxic treatment. Thus optical proteomics com-bined with advanced genomics has the potential to

236 Targ Oncol (2009) 4:235252

http://www.adjuvantonline.com

revolutionize treatment for the next generation of cancerpatients.

Translational science must strive to improve not only thequality and efficacy of the drugs used to combat all cancers,but also design intelligent diagnostics to accurately matchthe correct drug to the appropriate patient [10]. Efficacious,individualized therapy has the potential to revolutionizecancer care. We present here a range of approaches aimed atimproving our understanding of proteomics, exploitingoptically based technologies. Such optical techniques arehighly desirable for ultimate deployment in a clinicalsetting. Development of optical technologies to bothvisualize specific protein expressions and interactions andto quantify them is considered crucial for further progressin individualization of therapy.

Genomics and optical proteomics

The advent of gene expression profiling technologyallowed quantification of the expression of multiplegenes simultaneously in human tissue samples. The aimsof gene expression profiling in cancer are a better systemfor classifying tumors, a clarification of the origin ofthese diseases, a more accurate prognostic capacity thanwas previously available and an improved ability toselect appropriate therapy [11]. Despite advances, theidentification of genes and molecules associated withescalating incidence of metastases does not necessarilybring progress to understanding of how these eventscontribute to the process, much less teach us how tofrustrate metastasis.

Genomic information alone may thus prove insufficientas a means of identifying tumors with specific changes inthe molecular pathways that are predictive of a favorabletreatment outcome. In addition to obtaining RNA profiles,the function of the gene product, i.e., the protein, must alsobe assessed. Although the cause of a disease is frequently anaberration at the genetic level, the functional consequences aremediated via protein networks, with various components ofthe network undergoing different degrees of activation(usually as a consequence of specific post-translationalmodifications such as phosphorylation), driving oncogenesis.Following successes in cancer therapy derived from targetingthe MDM2-p53 interaction [1214], the importance ofprotein-protein interactions is increasingly recognized, bothfor understanding cell physiology and for developing noveltreatments [15].

Validated techniques involving advanced microscopywith high spatial resolution and in vivo imaging capabilitiesare required to assess and quantitatively measure post-translational modifications and protein-protein interactions,in order to expose the molecular contributors to the process

of metastasis. The use of advanced optical moleculartechniques to report on protein networks, within both cellsand tissues, is termed optical proteomics.1 The opticalapproaches herein described include the monitoring ofFrster resonance energy transfer (FRET) by fluorescencelifetime imaging microscopy (FLIM). This review describesthe utility of FRET-FLIM imaging in preclinical models ofcancer, as well as its use in cancer patient tissue.Characterizing the activation/modification states of pro-teins that are responsible for promoting cell migration, aswell as the nature and extent of the intermolecularinteractions within these protein subnetworks in individualpatient tumor samples, should enable better prediction ofwhich patients are most likely to develop metastases. Thischaracterization also has the potential to provide biologicalinformation regarding how best to interfere with metastasisand allow future therapeutic blockade of the strengths ofthe metastatic process and exploitation of the inherentweaknesses.

Recent experience with molecule-targeted therapeuticssuggests that the efficacy of such therapies would beimproved if we could selectively treat patients on the basisof aberrations in protein function/activity within specificbiochemical pathways, rather than simply the level oftarget antigen expression [16]. FRET-FLIM imaging ofprotein function and protein-protein complex formationcould potentially improve patient selection for targetedtherapy, by specifically identifying for each patientwhether the targeted pathway is active in a particulartumor sample, thus truly tailoring medicine to eachindividual.

In the context of proteomics, there are several keyfunctional events that are crucial for a better understandingof how the ~23,000 proteins encoded in the human genomeare coordinated and regulated. In this article we will focusmainly on two categories of events. First is the formation ofprotein networks, comprised of protein-protein interactionsbetween direct binding partners rather than through anintermediate bridging protein. The second is a group offurther post-translational modification events includingprotein phosphorylation, glycosylation [17], sulphatation,acetylation and ubiquitinylation [18]. These processesincrease the heterogeneity of the proteome, compared to

1 Significant advances in understanding normal cell function and indeveloping successful strategies for therapeutic intervention in diseasewill increasingly depend on our ability to study the expression, post-translational modifications and formation of protein complexes andnetworks in the cancer cells. Our definition of Optical Proteomics(please refer to http:www.opticalproteomics.org) broadly refers to theuse of optical techniques to provide all these types of informationfrom intact cells and tissues without disrupting the normal protein orcell function

Targ Oncol (2009) 4:235252 237

http:www.opticalproteomics.org

the genome. This diversity is further increased by proteomeinteractions with the genome. The proteome differs betweenand within tissues as a function of time, cell cycle andenvironment.

In the following section, some of the technologiesutilized in preclinical and patient-based clinical opticalproteomic studies of cancer are outlined.

Cancer tissue banks and tissue microarrays

Cancer tissue and data banks are of crucial importance forlarge-scale optical proteomic studies. Formalin-fixed

paraffin-embedded (FFPE) tissue is the most abundantsource of archived tumor material available and has beenprospectively collected in many institutions for decades.Thirty-year archives exist from international collaborativestudies. Original FFPE donor tissue blocks are used toproduce tissue microarrays (TMA, Fig. 1) and are rapidlyfixed in formalin prior to paraffin processing. FFPE is notsuitable for all biomarkers and so fresh frozen tissue andDNA are also archived. A well-annotated resource with arich complement of clinical data has great potential in thesearch for a cancer biomarker.

Tissue and data banks must be licensed by the HumanTissue Authority and should maintain and continue to

Fig. 1 Tissue microarray (TMA)block created using a micro-arrayer in the Guys /KCL BreastTissue Bank.a The BeecherMicroarrayer. b A wax core iscut and c removed from therecipient TMA block. d Asmaller bore tissue core isremoved from one of multipledonor case original tumorblocks. e The donor tissue coreis lowered into the space createdin the wax block. f New corein position. g Complete TMAblock and one H&E stainedTMA section on a glass slide.Each TMA block is sectionedin slices 35 mm thick andmounted on charged glass slides.De-waxing of the TMA sectionis performed for antigen retrievalprior to standard immunohisto-chemistry or fluorophore-conjugated-antibody stainingtechniques. (H:left) TMA stainedwith a Cy2 labeled antibody toezrin (mAb 2H3, middle) imagedat x10 power and (right) imagedat x40 revealing more detailedmembranous ezrin staining

238 Targ Oncol (2009) 4:235252

accrue archived cancer material. Tissue, blood and clinicaldata can only be prospectively collected on patientsfollowing informed consent. Complete and verifiedclinical-pathological data on every patient must be carefullytemplate-linked to each TMA tissue core. The fact thatclinical outcome is known for each case is pivotal to tissueand data bank studies. Putative links between a potentialbiomarker and relapse data can therefore be immediatelyascertained. A 2006 review of proteomics related to breastcancer concluded that technologies that use FFPE tumortissue will have the greatest impact, as FFPE tissue is themost abundant source of cancer tissue, allowing for a well-constructed hypothesis to be tested [19]. Any suchbiomarker would require validation prior to use on cancertissue at first diagnosis to delineate individual patient riskprofile.

TMA technology was introduced in 1998 as a tissuepreserving, high throughput (HTP) technique allowingstudy of multiple markers in large sample sets. TMAtechnology provides patient material suitable for evaluatingand validating DNA amplification using fluorescence insitu hybridisation (FISH), and protein levels using immu-nohistochemistry. TMAs provide a cost-effective methodfor examining multiple biomarkers on a large number ofpatients, both retrospectively and prospectively. In a largeseries of cases it is possible to perform multiple assays onconsecutive sections without significant depletion of refer-ence tissue resources. TMAs thus facilitate the survey oflarge numbers of tumors simultaneously, allowing therapid analysis of hundreds of markers in the same set ofspecimens [20]. The evidence that TMAs are equivalent totraditional whole tissue specimens was vital for thetechnology to gain widespread acceptance [21]. In lessthan ten years the validated technique has been proven formany different markers on various tumor types in multiplestudies [22].

Direct observation of cancer phenotype and proteindistribution within cancer tissues using optical methodshas a long history in the field of medical diagnosis. Thebiomolecular specificity possible with optical methodshas been particularly valuable in microscopy and histo-pathology. The application of optical histology methodsto the modernized version of TMAs [20], which allowexamination of high numbers of patient samples, hasbecome the most used proteomic technique in high-throughput molecular pathology research [23]. Morequantitative, optically based techniques with a high lineardynamic range of detection are required.

Immunohistochemistry has limitations, such as thesubjectivity of manual scoring which is only semi-quantitative, and exhibits nonlinearity of the stainingintensity [23]. In this review novel high-throughput assaysbased on advanced optical techniques are described that can

report on protein modifications beyond the level of proteinexpression and distribution.

Fluorescence: fluorescent protein transfection,fluorescent probes and fluorophore-conjugated proteins

A fluorophore is a molecule capable of absorbing lightenergy at specific wavelengths and re-emitting this energyat higher wavelengths. Fluorescence is the energy lossprocess through the emission of light by excited moleculesas they revert to the ground state. The fluorescence lifetime, (tau), is the average time that each fluorophore remains inan excited state [24]. Frster resonance energy transfer,FRET, is the process of energy transfer from an exciteddonor fluorophore to an acceptor fluorophore in closeproximity (Fig. 2). For FRET to occur, spectrally over-lapping fluorophores must be in close proximity. FRETefficiency (FRETeff ) depends on the distance between thetwo molecules (donor and acceptor fluorophores). TheFrster radius, R0, is the distance at which FRETeff is halfits maximum value (typically 210 nm [25]). R0 furtherdepends on the spectral characteristics of the fluorophores.This energy transfer is indirectly proportional to the sixthpower of the distance between the two fluorophores. Thisfact makes FRET a powerful indicator of molecularproximity, which in practice can only be observed ifproteins are interacting (Table 1). A far-field techniquecan thus be used to provide information at distance scalesnormally associated with near-field techniques.

If a protein is labeled with a donor fluorophore and asecond protein labeled with an acceptor fluorophore, thenFRET between donor and acceptor is interpreted as theinteraction of these proteins. Protein-protein interactionswithin a cell can be studied using microscopy methods bytagging the protein of interest with a fluorophore andintroducing DNA coding for the protein to the immortalizedcancer cell. Multiple proteins can be imaged in a single cellby transfecting each protein of interest with a differentfluorescent tag and performing sample excitation and imageacquisition at appropriate wavelengths.

Alternatively, fluorophores can be directly conjugated toantibodies against proteins involved in cancer cell migra-tion (Fig. 3). When such an antibody is applied to cancertissue the fluorescence reports on the location where eachprotein is present. Automated computer algorithms canrapidly and efficiently analyze the images of fluorescentlystained tissue, thus quantifying levels of protein expressionand subcellular localization. Several proteins can be labeledwith different fluorophores enabling simultaneous assess-ment of multiple proteins in a single tissue section,including their colocalization [2628], by automaticallyaltering the excitation wavelength of the microscope whilst

Targ Oncol (2009) 4:235252 239

capturing images in appropriate fluorophore emissionwavelength channels. Colocalization studies, while useful,can only determine the presence or absence of thefluorophore(s) on a distance scale limited by the resolutionof the microscope used: in practice this is limited to around250500 nm in most automated systems. However, thefluorescence signal contains further information about thebiophysical environment of the fluorophore and ultimatelythe tissue in which it is imaged [29]. A specific phenomenon,Frster resonance energy transfer, FRET, can be exploited toprovide information at distance scales far below theoptical resolution of the microscope. By using pairs offluorescently labeled antibodies applied to tissue specimensfrom tumor samples, nanometer proximity between thefluorophores can be determined. Combined with imaging,this is a powerful approach, as FRET yields proximityinformation well below the optical resolution limit that canbe achieved by colocalization imaging of two fluorophores[29].

Fluorescence lifetime imaging microscopy

The use of FLIM to measure FRET in live and fixed cellshas significantly improved our understanding of themolecular pathways by which extracellular and environ-mental signals are sensed by breast cancer cells, transducedthrough the cell signaling machinery, in order to triggerremodeling of the cytoskeleton, thus leading to cancer cellinvasion and metastases. These assays provide importantspatiotemporal information about the post-translationalmodifications (e.g., protein phosphorylation [3033], ubiq-uitination [34]/sumoylation [35], conformational change

associated with, e.g., GDP-GTP exchange [3638]), pro-teolytic processing [39], as well as interactions betweensignaling receptors (integrins, CD44, chemokine receptorsand receptor tyrosine kinases (RTK)), protein kinases andmany cytoskeletal remodeling proteins [4051]. FRET-FLIM techniques have reported on signaling processes inarchived pathological material [30, 52, 53]. MeasuringFRET on TMAs allows in situ quantification of post-translation modifications. Specifically, a two-antibodyFRET approach has been applied to human cancer tissuesto detect the nano-proximity between a donor fluorophore-conjugated anti-protein kinase C (PKC) or anti-epidermalgrowth factor receptor (EGFR) antibody, and an acceptorfluorophore-labelled phospho-specific antibody, providinga highly specific quantification of PKC7 or EGFRphosphorylation [52, 53] in cancer tissues.

A major obstacle to the analysis of protein function andprotein complex formation in disaggregated breast cancertissues is contamination of epithelial components by thehigh admixture of non-neoplastic stroma and inflammatorycells (fibroblasts, immune cells, blood vessels). By usingFRET-FLIM imaging approaches on FFPE tissue sections,which retain architecture, relevant portions of the tissuescan be chosen for assessment. Furthermore, preservationof protein phosphorylation and protein complexes isimproved due to the absence of a protein extractionprocedure.

Data analysis

High throughput, automatically acquired imaging of proteininteractomes in cells and tissues yields data on multiple

Cy2 is photo-excited in isolation Cy2 photo-excited in close proximity to Cy3FRET occurs: shortens remains long

donoremission

MP/SP laser excitation

donoremission

acceptoremission

FRET

Fig. 2 Jablonski representation of FRET As a fluorophore absorbslight it is excited from the ground state (S0) to a higher vibrationallevel (S1=first electronic state, S2 etc). At each energy level it existsin a number of closely spaced vibrational energy levels (horizontallines). Fluorescence results when a molecule returns to S0 from thelowest energy vibrational state of S1. The length of time spent in thehigher energy state prior to reverting to the ground state is termed tau,

, the fluorescence lifetime, and is typically in the pico- to nano-second range. The donor fluorophore (e.g., Cy2) is excited in isolation(left) if donor and acceptor (e.g., Cy3) fluorophores are in closeproximity, energy is transferred from the excited donor fluorophore tothe acceptor fluorophore, generating an excited acceptor molecule thatin turn can then emit fluorescence by reverting to its ground state

240 Targ Oncol (2009) 4:235252

protein pairs. These dense multiparametric data require ananalysis method that is rational, rapid and robust. Novelmethods of interrogating the optical data acquired inrelation to clinical outcome are necessary.

Traditional histopathological scoring systems, describingthe expression proteins in breast tumors offers informationbased on tissue morphology and architecture. Such scoring

systems can be used in isolation by individual histopathol-ogists or a consensus score can be arrived at, with multiplescorers reviewing images together. Often these systems aresubjective and difficult to analyze, with wide inter- andintra- observer variability. Automated scoring of proteindistributions performed with a computer algorithm isdesirable and indeed possible. Widefield images of fluores-

Table 1 Conditions for a successful FRET assay

Targ Oncol (2009) 4:235252 241

cent antibody-marked proteins can be used to automaticallyscore levels of protein staining in specific subcellularlocations; segmenting membrane, cytoplasm and nucleus.Iterative fitting can derive a scale and generate masks thatsegmenting regions and automatically scoring the stainingwithin each compartment. Such scoring systems arerational, rapid and robust and highly economical in termsof expert histopathologist time. They also circumvent theissues of score variability.

Automated analysis of protein colocalization is alsopossible. Correlation maps between the widefield images oftwo or more proteins can be produced and the colocalizedareas can be found by thresholding out the area of lowcorrelation [54]. Background pixels corresponding to areaswhere overlap distribution [26] is low and can be efficientlymasked out. The highly colocalized areas between the twostains are then found by calculating the Pearson distribution[27] and masking out the regions where the local Pearsoncoefficient is low. The colocalization intensity image of the

remaining highly colocalized pixels can be built and itsaverage value calculated.

Automated microscopy generates large image datasetswhich can be time consuming to process [55]. This isparticularly true for FLIM where exponential curve fittinghas to be performed to produce sets of parametric images(e.g., lifetime and interacting fraction maps). These filescan be automatically analyzed to produce a distribution oflifetime and an average lifetime. The analysis has beenmade robust to noise and to low photon counts, providingfast execution and amenability to automation and batchprocessing [56]. Processing of this type can also beperformed on parallel processors or computer clusterswhere more immediate results are required. Nevertheless,a significant obstacle to more widespread application is theconsiderably slower image acquisition time: accuratedetermination of FLIM information is considerably slowerthan more standard widefield methods. While numerousacquisition technologies are available, acquisition speed is,in general, inversely proportional to accuracy and propor-tional to the likelihood of generating artifacts. Many recentcontributions have emerged in the area of FLIM analysis[5660] and offered different technical solutions to fluores-cence lifetime imaging analysis. The technical details andindividual merits associated with each method are howeverbeyond the scope of this article.

The high dimensional optical data acquired in by FRET/FLIM TMA experiments requires novel data analysis. Forinstance, Bayesian machine learning algorithms or simpleartificial neural networks of the perceptron type [61] can beused to relate input and output data according to the imagetraits derived, termed the input data. Both clustering treesand self-organizing maps (two dimensional, discretizedtopology-conserving representations) [62] can be producedto visualize these high-dimensional data.

Preclinical utility of FRET-FLIM assays

Cell line models of human cancers

In order to improve understanding of the basic biology ofcancer, immortalized cell line models of neoplasia are oftenstudied in the laboratory. FRET-FLIM assays in cell lineshave great power in progressing beyond biochemical assaysof protein interaction and network complexity. Furthermore,preclinical data cell lines cancer studies can inform on howaberrant oncogenic molecular pathways respond to drugs atthe individual protein level.

EGFR/HER signaling plays an important role in thepathogenesis of a variety of tumor types. For instance, inbreast cancer, antibody therapy against HER2 has demon-strated efficacy in patients with advanced breast cancer and

A

B

Fig. 3 Protein colocalization does not define interaction a Proteinproximity: when two proteins are close but not interacting, thefluorophores to which they are attached are some distance apart, morethan 10 nm. FRET cannot occur between Cy2 and Cy3 (conjugated toanti-ezrin and anti-PKC, respectively). b Protein:protein interaction:The conformational changes that occur upon interaction between ezrinand PKC bring the fluorophores into close proximity. FRET canoccur. Thus FRET distinguishes protein interaction from proteinproximity

242 Targ Oncol (2009) 4:235252

may prove to prevent metastases in the adjuvant setting[63]. However, responses to HER targeted therapies do notcorrelate with receptor levels [16], and tumors eventuallyescape biological control. The basic molecular biologybehind this primary and secondary resistance is not yetclear, but optical proteomics offers a novel approach inanswering these questions.

The HER2/HER3 heterodimer has been shown tofunction as an oncogenic unit, with HER2 requiringdimerization with HER3 to drive proliferation [64]. HER3couples active HER2 to the phosphatidylinositol 3-kinase(PI3K) intracellular signaling pathway. HER3 has also beenshown to recruit PI3K in gefitinib-resistant cells, thusescaping gefitinib treatment [65]. Therefore, the quantifi-cation of FRET between HER2 and HER3, indicatingheterodimer formation should be of functional significanceand has now been measured in cancer cells by FLIM(Fig. 4a). The reported stabilizing effect of small moleculeinhibitors such as lapatinib on HER2/HER3 heterodimers[66] can now be accurately quantified in cells and tissues insitu.

In order to overcome problems with relative expressionlevels of different protein-fluorophore constructs in cells, anumber of research groups have produced sensor probes,whereby the proteins of interest are placed between thedonor and acceptor fluorophores. The Raichu-Cdc42 FRETbiosensor probe is one such example (Fig. 4b). This type ofintramolecular FRET sensor can be expressed in living cellswith negligible perturbation of endogenous protein function[67, 68], and permits monitoring of the subcellularmicroenvironment that regulates GTPase activities inliving cells. Such probes can be used to monitor proteinactivity in a living tumor using intravital techniques (pleasesee the below section on preclinical molecular pathwayevaluation in animal models using optical imaging)Furthermore, the probe plasmids can be delivered intra-tumorally by electroporation in live, anaesthetized animals,as described previously [69].

For monitoring localized small Rho GTPase activities(Cdc42, Rac1, RhoA) in cells, YFP-Raichu-CFP probeshave been reported previously [70]. The original Raichuprobes were designed for intensity measurements ofsensitized acceptor emission to detect FRET. The benefitsof using donor FLIM to detect FRET are that it is largelyindependent of fluorophore concentration, it is independentof donor-acceptor stoichiometry (as long as the acceptor isin excess) and light path length. It is therefore better suitedto studies in intact cells than intensity-based methods [32,71]. We have established Raichu-FLIM [37], whereby uponGTP binding, Rac1/Cdc42/RhoA exhibits a higher affinitytowards the corresponding RhoGTPase-interacting domain,bringing the two different fluorescent proteins of thebiosensor into close proximity and enabling FRET between

GFP and mRFP1. FRET results in a decrease in theobserved GFP fluorescence lifetime (). The Raichu-Cdc42fluorescence biosensor has been used to demonstrate spatio-temporal images of EGF-induced Cdc42 activation in cancercells (see Fig. 4c).

Resistance to many biological treatments could beunderstood better by imaging the recruited signaling path-ways in response to drug combinations. Such mechanismsare responsible for both primary resistance and diminishingresponses to targeted therapy. The mechanisms of action ofnovel inhibitors can be further elucidated at the molecularlevel by optical imaging, thus providing valuable preclin-ical information on how tumor cells overcome theirvulnerability to targeted therapy. Translating such assaysinto clinical benefit will guide the rational use of expensivedrugs. The disappointment of a tumor failing to respond totargeted therapy could be challenged and overcome byimaging the recruited resistance pathways, followed bydesigning treatment combinations specific to each patientstumor, in order to circumvent these mechanisms ofresistance.

Preclinical molecular pathway evaluation in animal modelsusing optical imaging

Animal imaging plays multiple roles in new drug develop-ment by providing, in preclinical models, importantinformation which can lead to ways of improving cancerpatient management. The objectives of the preclinicalimaging activities [72] include: [i] monitoring the responseof both primary tumors and secondary metastases duringpotential therapies and also assessment of tumor recurrenceupon cessation of treatment; [ii] analysis of the distributionof agents within the body, i.e., pharmacokinetic studies, anddose optimization; [iii] determination of the efficacy of anagent against the activity of its intended target at themolecular mechanistic level; and [iv] monitoring the effectsof agents on pathophysiological processes such as changesin the vascular volume of tumors in response to blockingproangiogenic signals.

Despite the enormous investment in genomics andscreening technologies over the past 20 years, the cost ofnew drug discovery continues to rise while approval ratesfall [73]. In the drug discovery context, the desire to exploitthe wealth of the proteome has also come face to face withthe realization that knowing a target is not the same asknowing what the target does, let alone knowing the effectsof a chemical inhibitor in diverse disease settings. Forinstance, clinical studies of EGFR inhibitors have shownresponse rates of the order of 5%15% in a variety ofcancer types [7477] and responses to ErbB-targetedtherapies do not however correlate necessarily with thereceptor levels [16]. Much needed is the means of

Targ Oncol (2009) 4:235252 243

Fig. 4 FRET/FLIM examples in cell-line models of cancer a MCF-7breast cancer cells were transfected with HER2-GFP and HER3-mRFP1. The HER2-GFP transmembrane receptor is seen within thecell membrane and intracellular compartments. HER3-mRFP1 colo-calized with HER2 at both cell membrane and several intracellular/sub-plasmalemmal vesicles. In the example shown, the FRETpopulation, or fraction of interacting molecules (F2), was determinedusing bi-exponential fitting and global analysis, as described [56].HER2-HER3 receptor dimers were found to localize to both cellprotrusions (top arrow) and internal vesicular structures includingthe sub-plasmalemmal endocytic compartment shown (white arrow)b This cartoon shows the molecular domains of the modified Raichucdc42 biosensor in its inactive and activated state. Upon activation,

cdc42 bind the PAK1 domain and consequently brings the twoattached fluorophores in close proximity thus enabling FRET. Thelevel of FRET can be measured by the fluorescence lifetime decreaseof the donor fluorophore GFP. CAAX (or CAAX-box), geneticallyengineered modification to the C-terminusof the probe to enable theprobe to monitor the activity-change at the plasma membrane [67]c A431 cells expressing the Raichu-Cdc42 biosensor were stimulatedwith EGF (100 ng/ml) for 10 min, and subsequently imaged bymultiphoton FLIM. Stimulated cells had a significantly greaterinteracting fraction (F2) compared to unstimulated cells (p

monitoring the pathway responses to a specific targetedtherapy and their translation to organ and organism levelphysiology.

Optical imaging techniques, in particular those whichcan be used to reliably measure FRET between two proteinpartners, have in recent years been used to monitor thepathway response (within the proteome) to both chemicaland genetic perturbations, as well as its timing andsubcellular localization, within live/intact cancer cells exvivo. These assays provide important spatiotemporalinformation about the post-translational modifications aswell as interactions between proteins. For determiningdirectly the efficacy of targeted cancer therapies, we havenow developed intravital FRET by FLIM assays in murinemodels of human cancers.

In order to acquire intravital imaging microscopicinformation in three dimensions, optical sectioning of asample and exclusion of out-of-focus light from fluoro-phores outside the planes of interest is crucial. The lattercan be achieved by confocal laser scanning microscopy,which uses linear single photon excitation (SPE) and aconfocal pinhole to achieve optical sectioning. However, allfluorophores within the light path are excited and are thusprone to photo-toxic effects and bleaching. Furthermore,conventional SPE light is scattered in thick samples,hampering the resolution of imaging in depth. By replacingSPE with non-linear two-photon excitation (TPE, often alsotermed MPE, multi-photon excitation) the excited volumecan be reduced drastically, thereby protecting the fluoro-phores outside the plane of interest and achieving opticalsectioning via excitation of only the section of interest [78,79]. TPE excitation is achieved by the use of lessphototoxic near-infrared light of approximately half theexcitation energy required to excite a fluorophore ascompared to SPE. Near-infrared light is also scattered to amuch lesser extent in tissue, thereby improving itspenetration depth and enabling controlled excitation offluorophores that are located even deeper within the tissue.By combining TPE with laser scanning microscopy(TPLSM), it was possible to increase the imaging depthfrom tens of micrometres up to nearly 1 millimetre [80]while protecting not only the fluorophores but also thesurrounding tissue from photo-damage. This technique wassuccessfully applied for imaging endogenous and/or intro-duced fluorophores deep inside biological tissue [8185].An additional benefit of TPE is that it also enables thelabel-free imaging of highly ordered structures like fibrillarcollagen via a process termed second harmonic generation(SHG) [86]. SHG permits the collection of informationregarding structure and concentration within biologicalmaterial as long as there are sufficient highly orderedstructures available [87]. For instance, many solid tumorshave a peripheral capsule, which contains high concen-

trations of fibrillar collagen and thereby the tumor boundarycan be easily visualized. In the context of cancer invasionand metastasis, intravital deep-tissue TPLSM has providedfurther insight into processes like the invasiveness of tumorcells and the efficacy of drugs affecting cellular motility[72, 88]. This technique was also used to investigate agentsthat modify the extracellular matrix and the diffusion ofcomponents of the interstitial fluid [89]. Furthermore, theleakiness of the (tumor) vasculature [90] and the effects ofantiangiogenic drugs [91, 92] have been studied, as well asthe efficiency of oxygen delivery [9395].

In order to establish in situ measurements of theaforementioned pathway/proteome response in thick samplesof living biological tissue, we aimed at combining TPLSMwith the advantages of measuring FRET by FLIM anddeveloping this technology for intravital FRET imaging inanimal models of cancer. With the development of thistechnique, it should be possible for the first time to gaininsight into, e.g., the efficacy of an agent against the activity ofits intended target within tumor cells, cells of the tumormicroenvironment, or metastasizing cells in vivo.

Currently, the use of intravital FRET-FLIM techniques tomonitor protein activity or protein-protein interactions is anemerging technology and the first promising results werereported very recently [96]. An example of the state-of-the-artof this novel imaging technique to probe a specific protein-protein interaction that is purported to support breast cancermetastasis is depicted in Fig. 5. Mammary adenocarcinomacells were genetically modified to express a fusion protein ofgreen fluorescent protein (GFP) and the chemokine receptorCXCR4, which is implicated in metastasis of many types ofcancer [97, 98]. Protein kinase C alpha (PKC) was reportedto be a modulator of CXCR4, recycling in human breastcancer cells by directly interacting with this receptor [46].We fused PKC to a monomeric red fluorescent protein(mRFP1) and introduced this fusion protein into themammary cells already stably expressing CXCR4-GFP. Inthis new cell line, CXCR4-GFP is the molecule beingobserved by FLIM and serves as the FRET donor. Upon itsinteraction with PKC-mRFP1 (FRET acceptor), a reductionin the fluorescence lifetime of GFP can be observed thataccounts for a direct protein-protein interaction betweenCXCR4 and PKC. Following xeno-transplantation of suchmodified cancer cells into an immuno-compromised mousemodel, we obtained solid tumors after several weeks ofgrowth. The animal was imaged under anesthesia in atemperature-controlled environment and fluorescence life-time images were acquired by intravital FLIM as described[96]. The collagen-rich tumor boundary in this animal modelwas imaged by SHG and the area where the plasmamembranes of the first tumor cells were observed wasdefined as the tumor edge. All further images were acquiredparallel to this plane and depth values were assigned relative

Targ Oncol (2009) 4:235252 245

to this optical section. Two-photon-excited GFP fluorescenceintensity images show that deep-tissue imaging wasperformed with no loss in lateral resolution while increasingimaging depth (Fig. 5). FRET between GFP and mRFP1results in shortening of the fluorescence lifetime () of GFP(red on the pseudocolor scale). The fluorescence lifetimesdetermined within the tumor core (60 m, 100 m from theedge) are constant and in agreement with the values obtainedfor a tumor in the absence of a suitable acceptor fluorophore(data not shown) suggesting no direct interaction betweenCXCR4-GFP and PKC-mRFP. Interestingly, we found thatin optical sections close to the tumor boundary (20 m fromthe edge), there is interaction between CXCR4 and PKC asrevealed by the reduced fluorescence lifetimes. We arecurrently investigating the pathophysiological significanceof the assembly of this pro-migratory receptor:kinasecomplex [46], close to the edge of tumor. Small moleculeinhibitors that can reverse this protein complex assemblymay perturb this early stage of cancer invasion. Intravital

FLIM is a technology that for the first time allows us toperform in situ protein:protein interaction measurements byFRET not only in vitro in two dimensions, but also in vivoin three dimensions. Further development of this imagingtechnology will find utility in preclinical research andfacilitate novel therapeutic and associated biomarkerdiscovery.

Clinical utility of FRET-FLIM assays

Utility of FRET-FLIM assays in archived patient material

Many studies have described protein distribution andexpression level changes which correlate with prognosisacross the spectrum of human cancers. One such studyapplied immunohistochemistry techniques to tissue micro-arrays (TMA) of breast cancer. Patient clusters were identifiedaccording to their protein expression and differences were

Mul

tipho

ton

GF

P in

tens

ity

[ns]

Flu

ores

cenc

elif

etim

e m

ap

20 m 60 m 100 m

2.05

2.65

V

Depth within tumor

Tumor edge

0 m

4 m

8 m...

Fig. 5 Combination of TPLSMwith FLIM measurementsreliably reveals fluorescencelifetime differences in vivo.Tumors were established bysubcutaneous injection ofmammary adenocarcinoma cellsstably expressing CXCR4-GFPand PKC-RFP into immuno-compromized mice. The animalswere imaged alive underanesthesia using a two-photonlaser scanning microscopecapable of detecting second-harmonic-generated (SHG) andfluorescence lifetime imaging(FLIM) data, and z-stacks wereacquired. Three optical imageplanes are shown beneath acartoon for orientation purpose.Depth values represent distancesof the corresponding opticalsection relative to the tumorboundary. Monochrome imagesrepresent two-photon intensityand time-resolved informationfrom these was used to calculatethe corresponding fluorescencelifetime maps on a pixel-by-pixelbasis. The fluorescence lifetimeimages are pseudo-colored tofacilitate recognition of areaswith increased interactionbetween CXCR4 and PKC(warm colors). The scale barcorresponds to 100 micrometres

246 Targ Oncol (2009) 4:235252

noted in terms of established prognostic factors and clinicaloutcomes [99]. This study demonstrated that tumors whichappear similar on the basis of established prognosticmethods, display wide heterogeneity when examined withrespect to protein expression.

Large-scale quantification of specific protein-proteininteractions in patient-derived tissue samples has not yetbeen documented. In our laboratory we have achieved semi-automated (with pathologist input for selection of tumorregions), spatially resolved and high-resolution FRET-FLIMassays using formalin-fixed paraffin-embedded tissue. Thehigh volume of data acquired was analyzed using novelartificial neural network techniques and correlatedwith patientclusters and clinical outcome (Fig. 6 and Kelleher et al,unpublished data).

This technology platform is evolving. Our prototypeoptical proteomic biomarker, based on a small proteinnetwork involved in cancer cell metastasis, has yieldedstatistically significant prognostic information. Furtherprospective validation is required prior to demonstratingclinical significance. Nevertheless it demonstrates how acombined optical imaging/mathematical modeling approachcan offer additional information, pertaining to proteinfunction and network formation, over simple intensity-based protein quantification.

Advances in molecular and cell biology are providingan improved understanding of cancer, from the germ lineto specific somatic changes of one patients cancer. Thepotential for individualizing prognosis is significant, andallows patients with an excellent prognosis to be sparedthe toxicity of adjuvant therapy. By assessing molecularpathways in those at high risk, improved selection oftargeted therapy will ensue. Expensive drugs will betailored to the patients whose tumor biology predictsresponse.

FRET-FLIM assays have further potential clinicalutility in the assessment of response to targeted therapy.In the neoadjuvant setting, repeat tumor biopsies couldbe analyzed to assess whether the specific aberrantpathway blocked by the targeted biological drug isindeed perturbed appropriately. In instances where tumortissue is not available, FRET-FLIM assays could also beperformed in surrogate tissue samples from patientsreceiving treatment. Clinicians could choose treatmentrationally for each patient and tailor subsequent therapiesbased on that individuals molecular response or resistancemechanisms.

Discussion

Significant advances have been made in cancer treatmentover the past 50 years, many of them empirical, more

latterly based on international randomized clinical trialsinvolving thousands of patients. Early diagnostic screen-ing program, improved surgical procedures, advancedradiotherapy techniques and combinations with adjuvantcytotoxic drugs have all contributed to survival improvementsacross the spectrum of human cancers. Modifications to all ofthese treatment modalities are continually reviewed, butfurther improvements in cancer survival are likely to bemodest.

Rational biological treatment of cancer, targeted atspecific molecular pathways has shown efficacy andpromises a changing future in oncology. The currentapproach is to combine existing breast cancer therapieswith novel agents that interfere with major signalingpathways. Nonetheless, targeted therapy is no panacea.Though it holds great promise, many patients do notrespond to targeted drugs de novo, and countless moredevelop acquired resistance. In this technologicallyadvanced post-genomic era, future advances in cancertreatment must address both the genetic profiles andproteomic signatures of individual neoplasms. Assaysaimed at the identification of tumors with explicitchanges in the molecular pathways that are specific toindividual patients, will identify the population ofpatients most likely to benefit from specific targetedtreatment. Only by discovering the oncogenic pathwaysthat propel each individuals cancer we can hope to trulytailor therapy to all patients with cancer. Dynamicpersonalized therapy, changing as the tumor evolves,will improve treatment efficacy and prolong survival formany cancer patients.

The introduction of novel targeted therapeutics hasrevolutionized cancer care and may offer the potential forcure to more patients. Better selection of patients likely torespond to these drugs will reduce costs significantly. If agoal of translational research is to prevent cancer deaths,then basic science must continue to interrogate themetastatic process. Optical proteomics aims to devise newprognostic tools by analyzing the detailed molecular make-upof individual tumors and estimating the metastatic andsurvival potential therein (Fig. 7).

In vitro diagnostic multigene index assays (IVDMIAs)have been generating excitement for more than a decade,demonstrating clinical utility in select groups of patient.Prospective clinical trials have not yet reported improvedpatient outcome on the basis of treatment decisions madeaccording to IVDMIAs, though it is hoped that this willbe the case in the near future. Combined genomics andproteomics should reduce healthcare costs by reducingthe number of cancer patients selected for adjuvanttreatment without compromising clinical outcome. Toxicitycould be spared and resources focused to combat metastaticcancer.

Targ Oncol (2009) 4:235252 247

0.00

0.05

0.10

0.15

0.20

-20- 10 0 10 20 30

A

B

Con

trol

pai

r

norm

al b

reas

t tis

sue

inva

sive

bre

ast c

ance

r

FR

ET

pai

rC

ontr

ol p

air

FR

ET

pai

r

Anti-ezrin- IgG-Cy2FRET EfficiencyMultiphotonintensity (Cy2)

Cy3

0 (%) 30

Pho

ton

Cou

nt

FRET efficiency (%)

irrelevantantibody

control-Cy3

irrelevantantibody

control-Cy3

Anti-T(P)250 PKC -Cy3

Anti-T(P)250PKC -Cy3

Normal,Anti-T(P)250 PKC -Cy3

Invasive,

Invasive,Anti-T(P)250PKC -Cy3

irrelevantantibody

control-Cy3

Normal,

irrelevantantibody

control-Cy3

(FRET donor)FRET acceptor

Fig. 6 FRET-FLIM imaging of protein complexes in patient tumormaterial. The PKC-ezrin protein complex was first detected by FRET/FLIM and was shown [45] to promote directional cancer cell motilityand hence the metastatic potential of MCF-7 cells. The direct interactionbetween endogenous ezrin and activated PKC has been demonstratedby multiphoton FLIM, in patient-derived cancer tissues that have beenlabeled with Cy2-labeled anti-ezrin-antibody (2H3) and Cy3 labeledanti-T(P)250-PKC antibody [30]. FRET efficiency (in each pixel)=1 -da/control, where da is the Cy2 lifetime of tumor section costainedwith Cy2-labelled anti-ezrin-antibody and Cy3 labeled anti-T(P)250-PKC antibody; and control is the average Cy2 lifetime (for the wholefield of view) measured in the absence of the Cy3 acceptor-labeledantibody. Another control is provided by an irrelevant Cy3-labeledantibody against a protein that is known not to interact with ezrin. Innormal breast tissue the labeled proteins are close but not interacting;

the fluorophores to which they are attached do not show FRET becausethe proteins are too far apart (cold colors in the FRET efficiency map).In contrast, in the sample of invasive breast cancer, not only are ezrinand PKC close (white arrows) but they are also interacting (warmcolors in the FRET efficiency map). The anti-T(P)250-PKC antibodyis known to interact with a non-PKC, 180 kDa nuclear antigen [30];both membrane and nuclear staining can therefore be detected in thetumors. This antibody specificity issue is overcome by donor FLIMwhich only quantifies the population of Cy2 in close proximity to Cy3acceptor, i.e., the assay requires only the donor fluorophore-labeledantibody (in this case an in-house anti-ezrin monoclonal antibody withproven specificity [45]). The resolution of the FRET efficiency mapdoes not match that of the corresponding intensity map because pixelbinning was required to increase the photon counts for lifetimeestimation

248 Targ Oncol (2009) 4:235252

The optimal use of omic technologies will requireclose collaboration between oncologists and basic scien-tists. In the analysis of high dimensional data, great caremust be taken to validate results in a separate patientpopulation. To this effect, statisticians, mathematicians,bioinformaticians and epidemiologists all have a crucialrole to play in this translation of exciting, novellaboratory research into real-life benefits for cancerpatients. There are surmountable obstacles to overcometo achieve the goal of translating advanced optical

proteomic science into real-life benefit for cancer patientsand their physicians.

Acknowledgements M. Kelleher was supported by a Guys and StThomas Charity grant (R041004). G. Fruhwirth is supported by theKings College London and University College London Comprehen-sive Cancer Imaging Centre CR-UK & EPSRC, in association withthe MRC and DoH (England) C1519/A10331. E. Ofo is a recipient ofboth a Wellcome Trust Clinical Training Fellowship (WT084982) andthe Dimbleby Cancer Care Award. Both E. Ofo (2007/8) and G. Patel(2008/9) received one-year clinical fellowship support from the

Fig. 7 Clinical utility schematic. Once an optical proteomic biomarkeris validated in cancer tissue at first diagnosis, a novel optical multi-protein biomarker should clearly differentiate individual patients ofgood prognosis, who could be spared adjuvant chemotherapy distinctfrom those patients who are of higher metastatic risk. The correlation ofthe optical image parameters with clinical outcome requires the use ofboth clustering trees and self-organizing maps (two dimensional,discretized topology-conserving representations) [62], a full description

of these methods is beyond the scope of this article. Patients with anintermediate or poor prognosis should receive tailored treatment aimedat perturbing the specific oncogenic pathways in their particular tumor,thereby increasing the likelihood of a good response to therapy. Tumoror surrogate tissue could be reassessed throughout treatment in order toensure that the pro-migratory oncogenic pathways remain blocked andto monitor the development of potential drug resistance

Targ Oncol (2009) 4:235252 249

Department of Health via the National Institute for Health Research(NIHR) comprehensive Biomedical Research Centre award to Guy's& St Thomas' NHS Foundation Trust, in partnership with King'sCollege London and Kings College Hospital NHS Foundation Trust.A. Coolen is the recipient of a Springboard Research Fellowship fromEPSRC. S. Ameer-Beg and T. Ng are supported by an endowmentfund from Dimbleby Cancer Care to Kings College London. The 2PEFLIM system was built with support from both the Medical ResearchCouncil Co-Operative Group grant (G0100152 ID 56891) and an UKResearch Councils Basic Technology Research Programme grant(GR/R87901/01). P. Barber is supported by a Cancer Research UKProgramme grant C133/A/1812, as is B. Vojnovic.

Conflict of interest statement No funds were received in support ofthis study.

Open Access This article is distributed under the terms of theCreative Commons Attribution Noncommercial License which per-mits any noncommercial use, distribution, and reproduction in anymedium, provided the original author(s) and source are credited.

References

1. Chambers AF, Groom AC, MacDonald IC (2002) Disseminationand growth of cancer cells in metastatic sites. Nat Rev Cancer2:563572

2. Early Breast Cancer Trialists' Collaborative Group (EBCTCG)(2005) Effects of chemotherapy and hormonal therapy for earlybreast cancer on recurrence and 15-year survival: an overview ofthe randomised trials. Lancet 365:16871717

3. Cole BF, Gelber RD, Gelber S et al (2001) Polychemotherapyfor early breast cancer: an overview of the randomised clinicaltrials with quality-adjusted survival analysis. Lancet 358:277286

4. Early Breast Cancer Trialists' Collaborative Group (1998) Poly-chemotherapy for early breast cancer: an overview of therandomised trials. Lancet 352:930942

5. Rubens RD (1992) Management of early breast cancer. Bmj304:13611364

6. GreenM, Hortobagyi G et al (2001) Chemotherapy for Breast Cancer.In: Hunt K, Robb G, Ueno N (eds) Breast Cancer: M.D. AndersonCancer Care Series. Springer-Verlag, New York, pp 323324

7. Ravdin PM, Siminoff LA, Davis GJ et al (2001) Computerprogram to assist in making decisions about adjuvant therapy forwomen with early breast cancer. J Clin Oncol 19:980991

8. Cardoso F, Piccart MJ, Durbecq V, Di Leo A (2002) Resistance totrastuzumab: a necessary evil or a temporary challenge? ClinBreast Cancer 3:247257 discussion 258249

9. Jonsson B, Wilking N (2007) A global comparison regardingpatient access to cancer drugs: summary. Ann Oncol 18:iii27

10. Di Leo A, Claudino W, Colangiuli D et al (2007) New strategiesto identify molecular markers predicting chemotherapy activityand toxicity in breast cancer. Ann Oncol 18(Suppl 12):xii814

11. Massague J (2007) Sorting out breast-cancer gene signatures. NEngl J Med 356:294297

12. Tovar C, Rosinski J, Filipovic Z et al (2006) Small-moleculeMDM2 antagonists reveal aberrant p53 signaling in cancer:implications for therapy. Proc Natl Acad Sci U S A 103:18881893

13. Vassilev LT (2004) Small-molecule antagonists of p53-MDM2binding: research tools and potential therapeutics. Cell Cycle3:419421

14. Vassilev LT, Vu BT, Graves B et al (2004) In vivo activation ofthe p53 pathway by small-molecule antagonists of MDM2.Science 303:844848

15. Arkin MR, Wells JA (2004) Small-molecule inhibitors of protein-protein interactions: progressing towards the dream. Nat Rev DrugDiscov 3:301317

16. Arteaga CL, Baselga J (2004) Tyrosine kinase inhibitors: whydoes the current process of clinical development not apply tothem? Cancer Cell 5:525531

17. Hollingsworth MA, Swanson BJ (2004) Mucins in cancer:protection and control of the cell surface. Nat Rev Cancer 4:4560

18. Hicke L (2001) Protein regulation by monoubiquitin. Nat RevMol Cell Biol 2:195201

19. Abramovitz M, Leyland-Jones B (2006) A systems approach toclinical oncology: Focus on breast cancer. Proteome Sci 4:5

20. Kononen J, Bubendorf L, Kallioniemi A et al (1998) Tissuemicroarrays for high-throughput molecular profiling of tumorspecimens. Nat Med 4:844847

21. Camp RL, Charette LA, Rimm DL (2000) Validation of tissuemicroarray technology in breast carcinoma. Lab Invest 80:19431949

22. Kay E, O'Grady A, Morgan JM et al (2004) Use of tissuemicroarray for interlaboratory validation of HER2 immunocyto-chemical and FISH testing. J Clin Pathol 57:11401144

23. Bertucci F, Birnbaum D, Goncalves A (2006) Proteomics of breastcancer: principles and potential clinical applications. Mol CellProteomics 5:17721786

24. Lakowicz JR (2006) Principles of Fluorescence Spectroscopy.Kluwer Academic/ Plenum, New York

25. Peter M, Ameer-Beg SM (2004) Imaging molecular interactionsby multiphoton FLIM. Biol Cell 96:231236

26. Manders EMM, Verbeek FJ, Aten JA (1993) Measurement of co-localization of objects in dual-colour confocal images. J Microsc169:375382

27. Gonzalez RC (1987) Wintz P (1987) Digital Image Processing.Reading, MA Addison-Wesley

28. Schubert W, Bonnekoh B, Pommer AJ et al (2006) Analyzingproteome topology and function by automated multidimensionalfluorescence microscopy. Nat Biotechnol 24:12701278

29. Festy F, Ameer-Beg SM, Ng T, Suhling K (2007) Imagingproteins in vivo using fluorescence lifetime microscopy. MolBiosyst 3:381391

30. Ng T, Squire A, Hansra G et al (1999) Imaging protein kinaseCalpha activation in cells. Science 283:20852089

31. Verveer PJ, Wouters FS, Reynolds AR, Bastiaens PI (2000)Quantitative imaging of lateral ErbB1 receptor signal propagationin the plasma membrane [In Process Citation]. Science 290:15671570

32. Wouters FS, Bastiaens PI (1999) Fluorescence lifetime imagingof receptor tyrosine kinase activity in cells. Curr Biol 9:11271130

33. Reynolds AR, Tischer C, Verveer PJ et al (2003) EGFR activationcoupled to inhibition of tyrosine phosphatases causes lateral signalpropagation. Nat Cell Biol 5:447453

34. Ganesan S, Ameer-Beg SM, Ng TT et al (2006) A dark yellowfluorescent protein (YFP)-based Resonance Energy-AcceptingChromoprotein (REACh) for Forster resonance energy transferwith GFP. Proc Natl Acad Sci U S A 103:40894094

35. Dadke S, Cotteret S, Yip SC et al (2007) Regulation of proteintyrosine phosphatase 1B by sumoylation. Nat Cell Biol 9:8085

36. Beutler M, Makrogianneli K, Vermeij RJ et al (2008) satFRET:estimation of Forster resonance energy transfer by acceptorsaturation. Eur Biophys J 38:6982

37. Makrogianneli K, Carlin LM, Keppler MD et al (2009) Integratingreceptor signal inputs that influence small Rho GTPase activationdynamics at the immunological synapse. Mol Cell Biol 29(11):29973006

250 Targ Oncol (2009) 4:235252

38. Carlin LM, Makrogianneli K, Fruhwirth G, Ng T (2010)Visualizing signalling in immune cells. In Nourshargh S,Marelli-Berg FM (eds): "T-Cell Trafficking, Methods in MolecularBiology." Totowa: Humana Press. In press

39. Harpur AG, Wouters FS, Bastiaens PI (2001) Imaging FRETbetween spectrally similar GFP molecules in single cells. NatBiotechnol 19:167169

40. Ng T, Shima D, Squire A et al (1999) PKCalpha regulates beta1integrin-dependent cell motility through association and control ofintegrin traffic. Embo J 18:39093923

41. Parsons M, Monypenny J, Ameer-Beg SM et al (2005) Spatiallydistinct binding of Cdc42 to PAK1 and N-WASP in breastcarcinoma cells. Mol Cell Biol 25:16801695

42. Anilkumar N, Parsons M, Monk R et al (2003) Interaction offascin and protein kinase Calpha: a novel intersection in celladhesion and motility. Embo J 22:53905402

43. Legg JW, Lewis CA, Parsons M et al (2002) A novel PKC-regulated mechanism controls CD44 ezrin association anddirectional cell motility. Nat Cell Biol 27:399407

44. Parsons M, Keppler MD, Kline A et al (2002) Site-directedperturbation of PKC-integrin interaction blocks carcinoma cellchemotaxis. Mol Cell Biol 22:58975911

45. Ng T, Parsons M, Hughes WE et al (2001) Ezrin is a downstreameffector of trafficking PKC-integrin complexes involved in thecontrol of cell motility. Embo J 20:27232741

46. Peter M, Ameer-Beg SM, Hughes MK et al (2005) Multiphoton-FLIM quantification of the EGFP-mRFP1 FRET pair forlocalization of membrane receptor-kinase interactions. Biophys J88:12241237

47. Prag S, Parsons M, Keppler MD et al (2007) Activated ezrinpromotes cell migration through recruitment of the GEF Dbl tolipid rafts and preferential downstream activation of Cdc42. MolBiol Cell 18:29352948

48. Avizienyte E, Keppler M, Sandilands E et al (2007) An active Srckinase-beta-actin association is linked to actin dynamics at theperiphery of colon cancer cells. Exp Cell Res 313:31753188

49. Carvalho RF, Beutler M, Marler KJ et al (2006) Silencing ofEphA3 through a cis interaction with ephrinA5. Nat Neurosci9:322330

50. Offterdinger M, Bastiaens PI (2008) Prolonged Egfr signalling byErbb2-mediated sequestration at the plasma membrane. Traffic 9(1):147155

51. Haj FG, Verveer PJ, Squire A et al (2002) Imaging sites ofreceptor dephosphorylation by PTP1B on the surface of theendoplasmic reticulum. Science 295:17081711

52. Keese M, Magdeburg RJ, Herzog T et al (2005) Imagingepidermal growth factor receptor phosphorylation in humancolorectal cancer cells and human tissues. J Biol Chem280:2782627831

53. Kong A, Leboucher P, Leek R et al (2006) Prognostic value of anactivation state marker for epidermal growth factor receptor intissue microarrays of head and neck cancer. Cancer Res 66:28342843

54. Bolte S, Cordelieres FP (2006) A guided tour into subcellularcolocalization analysis in light microscopy. J Microsc 224:213232

55. Esposito A, Dohm CP, Bahr M, Wouters FS (2007) Unsupervisedfluorescence lifetime imaging microscopy for high content andhigh throughput screening. Mol Cell Proteomics 6:14461454

56. Barber PR, Ameer-Beg SM, Gilbey J et al (2009) Multiphotontime-domain fluorescence lifetime imaging microscopy: practicalapplication to protein-protein interactions using global analysis. JRoyal Society Interface 6:S93S105

57. Esposito A, Gerritsen HC, Wouters FS (2005) Fluorescencelifetime heterogeneity resolution in the frequency domain bylifetime moments analysis. Biophys J 89:42864299

58. Schlachter S, Elder AD, Esposito A et al (2009) mhFLIM:resolution of heterogeneous fluorescence decays in widefieldlifetime microscopy. Opt Express 17:15571570

59. Grecco HE, Roda-Navarro P, Verveer PJ (2009) Global analysis oftime correlated single photon counting FRET-FLIM data. OptExpress 17:64936508

60. Digman MA, Caiolfa VR, Zamai M, Gratton E (2008) The phasorapproach to fluorescence lifetime imaging analysis. Biophys J 94:L1416

61. Freund Y, Schapire RE (1999) Large margin classification usingthe perceptron algorithm. Machine Learning 37:277296

62. Hamalainen T, Klapuri H, Saarinen J, Kaski K (1997) Mapping ofSOM and LVQ algorithms on a tree shape parallel computersystem. Parallel Comput 23:271289

63. Smith I, Procter M, Gelber RD et al (2007) 2-year follow-up oftrastuzumab after adjuvant chemotherapy in HER2-positive breastcancer: a randomised controlled trial. Lancet 369:2936

64. Holbro T, Beerli RR, Maurer F et al (2003) The ErbB2/ErbB3heterodimer functions as an oncogenic unit: ErbB2 requiresErbB3 to drive breast tumor cell proliferation. Proc Natl AcadSci U S A 100:89338938

65. Cao Z, Wu X, Yen L et al (2007) Neuregulin-induced ErbB3downregulation is mediated by a protein stability cascade involvingthe E3 ubiquitin ligase Nrdp1. Mol Cell Biol 27:21802188

66. Scaltriti M, Verma C, Guzman M et al (2009) Lapatinib, a HER2tyrosine kinase inhibitor, induces stabilization and accumulationof HER2 and potentiates trastuzumab-dependent cell cytotoxicity.Oncogene 28:803814

67. Nakamura T, Aoki K, Matsuda M (2005) Monitoring spatio-temporal regulation of Ras and Rho GTPase with GFP-basedFRET probes. Methods 37:146153

68. Miyawaki A (2003) Visualization of the spatial and temporaldynamics of intracellular signaling. Dev Cell 4:295305

69. Li S, Xia X, Zhang X, Suen J (2002) Regression of tumors byIFN-alpha electroporation gene therapy and analysis of theresponsible genes by cDNA array. Gene Ther 9:390397

70. Itoh RE, Kurokawa K, Ohba Y et al (2002) Activation of rac andcdc42 video imaged by fluorescent resonance energy transfer-based single-molecule probes in the membrane of living cells. MolCell Biol 22:65826591

71. Wouters FS, Verveer PJ, Bastiaens PI (2001) Imaging biochemistryinside cells. Trends Cell Biol 11:203211

72. Sahai E (2007) Illuminating the metastatic process. Nat RevCancer 7:737749

73. DiMasi JA, Hansen RW, Grabowski HG (2003) The price ofinnovation: new estimates of drug development costs. J HealthEcon 22:151185

74. Fukuoka M, Yano S, Giaccone G et al (2003) Multi-institutionalrandomized phase II trial of gefitinib for previously treatedpatients with advanced non-small-cell lung cancer (The IDEAL1 Trial) [corrected]. J Clin Oncol 21:22372246

75. Cohen EE, Rosen F, Stadler WM et al (2003) Phase II trial ofZD1839 in recurrent or metastatic squamous cell carcinoma of thehead and neck. J Clin Oncol 21:19801987

76. Saltz LB, Meropol NJ, Loehrer PJ Sr et al (2004) Phase II trial ofcetuximab in patients with refractory colorectal cancer thatexpresses the epidermal growth factor receptor. J Clin Oncol22:12011208

77. Perez-Soler R, ChachouaA,Hammond LA et al (2004) Determinantsof tumor response and survival with erlotinib in patients with non-small-cell lung cancer. J Clin Oncol 22:32383247

78. Denk W, Strickler JH, Webb WW (1990) Two-photon laserscanning fluorescence microscopy. Science 248:7376

79. Centonze VE, White JG (1998) Multiphoton excitation providesoptical sections from deeper within scattering specimens thanconfocal imaging. Biophys J 75:20152024

Targ Oncol (2009) 4:235252 251

80. Theer P, Hasan MT, Denk W (2003) Two-photon imaging to adepth of 1000 m in living brains by use of a Ti:Al2O3regenerative amplifier. Optics Letters 28:10221024

81. Helmchen F, Denk W (2002) New developments in multiphotonmicroscopy. Curr Opin Neurobiol 12:593601

82. Bousso P, Robey EA (2004) Dynamic behavior of T cells andthymocytes in lymphoid organs as revealed by two-photonmicroscopy. Immunity 21:349355

83. Rubart M (2004) Two-photon microscopy of cells and tissue. CircRes 95:11541166

84. Laiho LH, Pelet S, Hancewicz TM et al (2005) Two-photon 3-Dmapping of ex vivo human skin endogenous fluorescence speciesbased on fluorescence emission spectra. J Biomed Optics10:024016024010

85. Molitoris BA, Sandoval RM (2005) Intravital multiphotonmicroscopy of dynamic renal processes. AJP Renal Physiology288:F1084F1089

86. Freund I, Deutsch M, Sprecher A (1986) Connective tissuepolarity. Optical second-harmonic microscopy, crossed-beamsummation, and small-angle scattering in rat-tail tendon. BiophysJ 50:693712

87. Mohler W, Millard AC, Campagnola PJ (2003) Second harmonicgeneration imaging of endogenous structural proteins. Methods29:97109

88. Wyckoff JB, Pinner SE, Gschmeissner S et al (2006) ROCK- andmyosin-dependent matrix deformation enables protease-independent tumor-cell invasion in vivo. Curr Biol 16:15151523

89. Brown E, McKee T, diTomaso E et al (2003) Dynamic imaging ofcollagen and its modulation in tumors in vivo using second-harmonicgeneration. Nat Med 9:796800

90. Sorlie T, Perou CM, Tibshirani R et al (2001) Gene expressionpatterns of breast carcinomas distinguish tumor subclasses withclinical implications. Proc Natl Acad Sci U S A 98:1086910874

91. Tozer GM, Ameer-beg SM, Baker J et al (2005) Intravital imagingof tumor vascular networks using multi-photon fluorescencemicroscopy. Adv drug deliv rev 57:135152

92. Tozer GM, Prise VE, Wilson J et al (2001) Mechanisms associatedwith tumor vascular shut-down induced by combretastatin A-4phosphate: intravital microscopy and measurement of vascularpermeability. Cancer Res 61:64136422

93. Serganova I, Doubrovin M, Vider J et al (2004) Molecularimaging of temporal dynamics and spatial heterogeneity ofhypoxia-inducible factor-1 signal transduction activity in tumorsin living mice. Cancer Res 64:61016108

94. Tong RT, Boucher Y, Kozin SV et al (2004) Vascular normalizationby vascular endothelial growth factor receptor 2 blockade inducesa pressure gradient across the vasculature and improves drugpenetration in tumors. Cancer Res 64:37313736

95. Winkler F, Kozin SV, Tong RT et al (2004) Kinetics of vascularnormalization by VEGFR2 blockade governs brain tumorresponse to radiation: Role of oxygenation, angiopoietin-1, andmatrix metalloproteinases. Cancer Cell 6:553563

96. Fruhwirth GO, Matthews DR, Brock A et al (2009) Deep-tissuemultiphoton fluorescence lifetime microscopy for intravital imagingof protein-protein interactions. In Periasamy A, So PTC (eds): SPIEBIOS/Photonics West 2009. San Jose, CA, USA: SPIE:71830L71839

97. Keiichi Koizumi Shtakyis (2007) Chemokine receptors in cancermetastasis and cancer cell-derived chemokines in host immuneresponse. Cancer Science 98:16521658

98. Zlotnik A (2008) New insights on the role of CXCR4 in cancermetastasis. J Path 215:211213

99. Abd El-Rehim DM, Ball G, Pinder SE et al (2005) High-throughput protein expression analysis using tissue microarraytechnology of a large well-characterized series identifiesbiologically distinct classes of breast cancer confirming recentcDNA expression analyses. Int J Cancer 116:340350

252 Targ Oncol (2009) 4:235252

The potential of optical proteomic technologies to individualize prognosis and guide rational treatment for cancer patientsAbstractIntroductionGenomics and optical proteomicsCancer tissue banks and tissue microarraysFluorescence: fluorescent protein transfection, fluorescent probes and fluorophore-conjugated proteinsFluorescence lifetime imaging microscopyData analysisPreclinical utility of FRET-FLIM assaysCell line models of human cancersPreclinical molecular pathway evaluation in animal models using optical imaging

Clinical utility of FRET-FLIM assaysUtility of FRET-FLIM assays in archived patient material

DiscussionReferences

/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 150 /GrayImageDepth -1 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages true /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 600 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputCondition () /PDFXRegistryName (http://www.color.org?) /PDFXTrapped /False

/SyntheticBoldness 1.000000 /Description >>> setdistillerparams> setpagedevice