Sa

MI

a

ARRAA

KSAPLKM

1

cdart

cbm1

soap

psva

0d

Biosensors and Bioelectronics 24 (2009) 3517–3523

Contents lists available at ScienceDirect

Biosensors and Bioelectronics

journa l homepage: www.e lsev ier .com/ locate /b ios

ensing based on assessment of non-monotonous effect determined by targetnalyte: Case study on pore-forming compounds

ihaela Gheorghiu, Andreea Olaru, Aurelia Tar, Cristina Polonschii, Eugen Gheorghiu ∗

nternational Centre of Biodynamics, 1B Intrarea Portocalelor 060101 Bucharest 6, Romania

r t i c l e i n f o

rticle history:eceived 9 March 2009eceived in revised form 24 April 2009ccepted 7 May 2009vailable online 14 May 2009

a b s t r a c t

A new and exciting biosensing avenue based on assessment of the non-monotonous, concentrationdependent effect of pore formation is discussed. A novel kinetic model is advanced to relate surfaceplasmon resonance (SPR) data with actual concentrations of interacting partners. Lipid modified L1 sen-sor chip provide the accessible platform for SPR exploration of peptide–membrane interaction, withPOPC and melittin as model systems. We show that quantitative assessment of the interaction between

eywords:PRntimicrobial peptide detectionore formationipid platforminetic modelelittin

an antimicrobial peptide and lipid modified sensors is capable to provide both sensing avenues anddetailed mechanistic insights into effects of pore-forming compounds. The proposed model combinedwith appropriate design of the experimental protocol adds a new depth to the classic SPR investigation ofpeptide–lipid interaction offering a quantitative platform for detection, improved understanding of themanifold facets of the interaction and for supporting the controlled design of novel antimicrobial com-pounds. This biosensing approach can be applied to an entire set of pore-forming compounds including

d exo

antimicrobial peptides an

. Introduction

Assessment of the complete action mechanism of pore-formingompounds on their interaction with (artificial) membranes is fun-amental in understanding membrane processes with possiblepplications in disease diagnosis, toxicology and pharmaceuticalesearch (Kass, 2005; Panchal et al., 2002) and represents an impor-ant issue, yet less investigated, in biosensing.

Pore formation and subsequent membrane destabilization is aommon feature for the interaction process between a large num-er of compounds such as peptides, toxins and viruses with lipidembranes (Anderluh et al., 2003; Chah and Zare, 2008; Shai,

999).Nevertheless, the process has not been largely exploited for

ensing purposes; in a recent report (Wilkop et al., 2008) the devel-pment of an electrochemical sensing platform for detecting tracemounts of bacterial toxins via the perforation process has beenroposed.

Antimicrobial peptides belong to an important class of com-

ounds with pore-forming capabilities. They offer an attractiveolution to the problem of increasing resistance of bacteria to con-entional antibiotics, based on direct interaction with membranesnd subsequent lysis of the pathogen cell membrane. Despite the

∗ Corresponding author. Tel.: +40 213104354; fax: +40 213104361.E-mail address: egheorghiu@biodyn.ro (E. Gheorghiu).

956-5663/$ – see front matter © 2009 Elsevier B.V. All rights reserved.oi:10.1016/j.bios.2009.05.007

-toxins.© 2009 Elsevier B.V. All rights reserved.

already documented efficient antimicrobial activity against a widerange of pathogens and viruses (Asthana et al., 2004; Dempsey,1990), the potential cytotoxic activity against mammalian cells(Wessman et al., 2008), limits the direct use of these peptides astherapeutics.

Concerted efforts to modify the native antimicrobial peptides ordesign new peptides to achieve better specificity against microbialinfections while limiting host organism cytotoxicity, add everydaynew peptides to the ∼800 natural antimicrobial peptides identifiedin eukaryotes (http://www.bbcm.units.it, Antimicrobial SequencesDatabase).

Elucidation of the complete interaction mechanism representsa key step in peptide design and in detection of new pore-formingcompounds, requiring effective appraisal of these compounds andaccess to lipid platforms for quantitative assessment of the interac-tion kinetics.

Melittin, the main component of bee venom (Habermann andJentsch, 1967), is often employed as model (pore-forming) com-pound in interaction studies with natural and artificial (biomimetic)membranes (Lundquist et al., 2008). It induces membrane disrup-tion and lysis upon spontaneous binding to biological and modelmembranes (Frey and Tamm, 1991). The mode of action is depen-

dent on the net charge of substrate: it forms transmembrane poresin zwitterionic lipid bilayer via barrel-stave mechanism (Vogel andJahnig, 1986) and acts as a detergent in negatively charged mem-branes (Ladokhin and White, 2001). Similar with all pore-formingcompounds, its effect occurs via complex, multiphase process

3 nd Bio

a(

cag

vpeir(daccsd

naefh

dafiea(tvtapfw

2

2

pfdaafwtHb

2

(adimw

518 M. Gheorghiu et al. / Biosensors a

nd exhibits concentration dependency with distinct thresholdsMozsolits et al., 2003, 2001).

Upon binding on the lipid membrane and reaching a thresholdoncentration, the insertion phase occurs leading to pore formationnd subsequent phospholipid matrix destabilization and disinte-ration (Shai, 1999).

Each step of this model of interaction has been experimentallyalidated using different lipid membrane matrices: liposomes, sup-orted lipid bilayers, micelles, phospholipid multilayers (Lundquistt al., 2008; Popplewell et al., 2005) and various methods ofnvestigation: infrared spectroscopy (Frey and Tamm, 1991), fluo-escence (Matsuzaki et al., 1997), transmission electron microscopyWessman et al., 2008), X-ray diffraction (Lee et al., 2008), circularichroism (Zhu et al., 2007) and SPR (Mozsolits et al., 2001; Papond Shai, 2003). As a corollary, this interaction depends on the lipidomposition and charge, on the hydration level and on the peptideoncentration, orientation on membrane surface and protonationtate (Berneche et al., 1998; Lin and Baumgaertner, 2000). Yet, theynamic assessment of the whole process has never been reported.

Using the advantages offered by surface plasmon reso-ance (SPR) technique, i.e. label free, real time monitoring ofnalyte–ligand interaction, we aim to reveal, for the first time, thentire process of interaction between melittin, as a model of a pore-orming compound, and a model (POPC) membrane as well as toighlight related biosensing capability.

Dynamic, quantitative assessment of the concentration depen-ent of non-monotonous processes associated with the targetnalyte (i.e. pore-forming compound) interaction with a lipid modi-ed SPR platform is proposed as a novel biosensing approach. To thisnd, comprehensive SPR measurements on melittin binding to anrtificial lipid membrane (POPC) using lipid modified L1 sensor chipBiacore) are quantitatively analyzed. A numerical routine linkinghe SPR data with the actual concentrations of interacting partnersia a novel kinetic model is developed, substantiating the quanti-ative parameters governing the interaction. We envisage that thispproach is able to support accurate detection and related analysislatform, which could be further extended to membranes with dif-

erent lipid compositions and other pore-forming compounds, asell.

. Materials and methods

.1. Materials

1-Palmitoil-2-oleyl-sn-glycero-3-phosphocholine (POPC) wasurchased from Avanti Lipids (Alabaster, USA) while melittin

rom honey bee venom (89.4% purity), (3-[(3-Cholamidopropyl)imethylammonio]-1-propanesulfonate (CHAPS), bovine serumlbumin (BSA), 4-(2-Hydroxyethyl)piperazine-1-ethanesulfoniccid (HEPES), sodium chloride, sodium hydroxide, were purchasedrom Sigma–Aldrich (München, Germany). All chemical reagents

ere of analytical grade and were used without further purifica-ion. Ultrapure water (MilliQ) and chloroform were used as solvents.EPES (10 mM Hepes, 150 mM NaCl, pH 7.4) was used as runninguffer, 0.22 �m filtered and degassed prior to use.

.2. Methods

Commercial L1 sensor chips and Biacore 3000 analytical system

Biacore AB, Uppsala, Sweden), at 25 ± 0.1 ◦C operating temper-ture, were used in all biosensor experiments. The L1 surface,isplaying lipophilic anchors in a carboxymethyl dextrane matrix,

s optimal for lipid vesicles capture and formation of stable lipidatrices. The sensor surface was prepared for lipid immobilizationith two injections of CHAPS.

electronics 24 (2009) 3517–3523

2.2.1. Lipid vesicles preparationLipid vesicles were prepared by dissolving POPC (1.5 mM) in

chloroform followed by drying under vacuum using a rotary-evaporator for 3 h. After hydration with running buffer andsonication (5 cycles, 30 min/cycle), the lipid suspension wasextruded using the Mini-extruder (Avanti Lipids, Alabaster, USA)by 22–25 passages through a 1 �m-polycarbonate membranes. Theextruded stock solution was stored at +4 ◦C prior to use; before eachexperiment the samples of different POPC vesicles concentrationswere freshly prepared.

2.2.2. SPR measurementsThe experimental conditions for each step in the complete SPR

experiment: sensor preparation, lipid immobilization, removal ofloosely bound structures, melittin injection and sensor regenera-tion are as follows.

2.2.2.1. Sensor chip preparation. The normalized sensor chip L1 wastreated with two short (1 min) injections of CHAPS at a flow rate of20 �L/min (Biacore Handbook, 2003a,b).

2.2.2.2. Formation of lipid matrix. Samples of POPC vesicles (80 �L)with different lipid concentrations (0.01 mM, 0.05 mM, 0.15 mMand 0.5 mM) were applied to the sensor surface at a low flow rateof 2 �L/min. Three injections of NaOH (50 mM), each of 5 min, wereapplied at flow rate of 20 �L/min to remove loosely bound vesi-cles structures and to stabilize the baseline. The quality of the lipidmatrix is indicated by baseline stability for more than two hours.To check the lipid coverage of the L1 chip, a pulse injection of BSA(0.5 mg/mL in HEPES, 10 �L/min) was performed (Papo and Shai,2003).

The lipid matrix immobilized on the L1 chip was used as a modelcell membrane for further investigation of the pore-forming pep-tide binding.

2.2.2.3. Peptide binding. Peptide stock solution (1 mM) was pre-pared by dissolving melittin in MilliQ water and stored in lightprotected vials at +4 ◦C. Melittin spiked samples, with concentra-tions between 0.35 and 3.62 �M, were freshly prepared beforeexperiments by dilution in running buffer. Since preliminary testswith 7–30 min injection times showed incomplete interactionsteps, a long injection, 65 min, was used to reveal the completemechanism of peptide–lipid interaction. 5 �L/min flux was usedto avoid mass transport limitations.

2.2.2.4. Sensor regeneration. Injections of CHAPS, 20 mM, wereused for sensor regeneration, preparing the surface for anotheranalysis cycle.

2.2.3. Quantitative analysis of the SPR dataClassical SPR analysis relates the SPR sensorgram (i.e. time vari-

ation of the reflectance dip position, or SPR angle) to the quantity ofinterest, assuming one effective layer (characterized by an effectivethickness deff, and dielectric constant εeff) on top of the SPR chip.

In contrast, our approach relates the evolution of interactingcompounds (based on a kinetic model) to the “evolving” layers onthe chip. Via repeated application of the Fresnel equation (Reitz etal., 1993), i.e. the construction of a Transfer Matrix (Born and Wolf,1980) we relate the SPR angle shift to the surface concentration ofcompounds in the multilayer system associated with the experi-mental platform. The effective thickness di and dielectric constant

εi of each layer in the system shape the reflectivity spectrum andinfluence the SPR angle and are important parameters in the con-struction of the transfer matrix.

We have designed a home made fitting routine with the follow-ing land marks:

nd Bio

2

3

S

h

rtS

sm

2pdibt

et(idhi

m

m

N

m

TeacsiaK

tc

dependent on the concentration of the lipids in the vesicles solu-tion and on the “quality” of the L1 surface. Typical experimentaldata show immobilization levels of 5800–8800 (RU) correspond-ing to 0.15–0.5 mM injected lipid concentrations. Literature data

Fig. 1. Complete SPR sensorgram of an experimental cycle comprises distinct steps:(A) sensor preparation, (B) lipid immobilization, (C) removal of loosely bound lipid,

M. Gheorghiu et al. / Biosensors a

1. Numerical integration of the set of four coupled differential equa-tions for the concentrations of each compound in the multilayersystem.

. Relate volume concentrations to corresponding volume fractionsin the distinct layers and compute the effective permittivity εeff1,εeff2, εeff3, εeffh of the layers with thickness dm, d2, and dh respec-tively.

. Derivation of the SPR angle based on the application of the trans-fer matrix on the multilayer system.

Details of the related equations and procedures are provided asupplementary material.

To check for consistency we used WinSpal 3.01 (freeware,ttp://www.mpip-mainz.mpg.de/knoll/soft/).

As the measured SPR signal expressed in relative units is directlyelated to the SPR angle shift derived from the application of theransfer matrix, the fitting routine provides the dependency of thePR signal on the parameters of the model.

Throughout the paper, the values are mean of triplicate mea-urements. Data are shown as mean value ± standard error of theean.

.2.3.1. The kinetic model. The dynamics of the non-monotonousrocess of interaction, as revealed by the SPR signal is highly depen-ent on the balance between the kinetics of melittin association and

nsertion. The rates at which these steps take place are modulatedy the surface concentration of melittin, which is determined byhe concentration of injected melittin.

To describe the entire process, we propose a set of kineticquations which provide time evolution of both lipid and melit-in concentrations as function of several threshold concentrationsm0, mi and mL) and constants: Ka1 for association, Ka2 for melittinnsertion, Kd0 for melittin dissociation and Kdl0 for lipid membraneestabilization.The following four coupled differential equationsave been designed, corresponding to distinct steps in the complex

nteraction process of melittin–lipid interaction:

′[t] = Ka1Nm(kNl[t] − m[t]) − Kd0 · kNl[t]/Nl0 · m[t] · −m′ins[t];

(1)

′ins[t] = Ka2

(m[t] − m0

Nl[t]Nl0

)·(

miNl[t]Nl0

− mins[t])

·

H[

m[t] − m0Nl[t]Nl0

]· H

[mi

Nl[t]Nl0

− mins[t]]

(2)

′l[t]= − Kdl0Nl[t]

(m[t]−mL

Nl[t]Nl0

)· H[Nl[t]] · H

[m[t]−mL

Nl[t]Nl0

]

(3)

′L1[t] = Ka3Nm(RR−Nl[t]/Nl0) · (NCO−mL1[t]) · H[RR − Nl[t]/Nl0] ·

H[NCO − mL1[t]] (4)

hus, melittin association to the lipid modified chip (Eq. (1)) is influ-nced by: the association constant Ka1 (the type of lipid membranend electrostatic interactions affect the corresponding value), theoncentration of injected melittin, Nm and the available “bindingites” (kNl[t]) associated with the immobilized lipid. The dynam-cs of the concentration of melittin in the association layer (dm) is

ffected by the dissociation induced by flow conditions, reflected byd0, and by the kinetics of melittin insertion in the lipid matrix m′

ins.

The insertion process (Eq. (2)) is governed by the “quality” ofhe lipid matrix, reflected by Ka2 and on the concentration of asso-iated melittin: with increasing peptide binding, upon reaching a

electronics 24 (2009) 3517–3523 3519

threshold, melittin undergoes reorientation and inserts into themembrane. A distinct threshold limits the progression of insertionwhen critical concentration mi is attained.

Lipid destabilization and dissociation, Eq. (3), is triggered uponassociated melittin reaching a threshold concentration mL and itskinetics depends on Kdl0 and the actual concentration of lipid.

Upon lipid dissociation, as the injection of melittin continues,the direct association of the peptide to the L1 chip occurs. This pro-cess, Eq. (4), depends on the concentration of injected melittin, thepercentage of uncovered sensor, RR and the L1 binding sites formelittin, NC0.

Heaviside function H[x] ={

1, x ≥ 00, x < 0

is considered for thresh-

old conditions.

3. Results and discussion

SPR technique is recognized as a valuable tool for real time mon-itoring of the bimolecular interactions (Hoa et al., 2007) enablingquantitative assessment of target compounds (Feltis et al., 2008;Kumbhat et al., 2007). SPR is also suitable for a rapid characteri-zation of binding processes allowing for qualitative analysis of theinteraction process (Vornholt et al., 2007).

Many studies were recently conducted on the developmentof SPR biosensors for elucidation of different membrane inter-actions including protein–protein interactions (Yang et al., 2007)antibody–antigen interactions (Ayela et al., 2007) and pore-forminginteractions of toxins or peptides (Chah and Zare, 2008). Never-theless, for the latter case, there is no complete, quantitative SPRassessment of the whole interaction process.

3.1. Complete SPR sensorgram

The typical sensorgram presented in Fig. 1 reveals the distinctstages of sensor preparation and peptide–lipid interaction. Theamplitude of the SPR signal upon lipid immobilization (step B) is

(D) melittin injection and sensor regeneration (E). The specific steps in melitin–lipidinteraction are presented in the inset: 1—association, 2—insertion, 3—membranedestabilization and lipid dissolution. The arrows indicate the positions correspond-ing to: Tmax, when the maximum SPR signal is reached and the time points when thethreshold concentrations for initiation of insertion, m0 and lipid dissociation, mL,are attained.

3 nd Bioelectronics 24 (2009) 3517–3523

a(2eaf2

osb

aclmrdrr

ctddta

3t

eactmep

cttnaradod

itcmdlrottmEce

520 M. Gheorghiu et al. / Biosensors a

ssociate around 5000 RU for lipid bilayers and above 8000 RUBiacore Handbook, 2003b; Anderluh et al., 2005; Nakajima et al.,001) for tightly packed lipid matrices. Lower immobilization lev-ls (3000 RU) obtained for 0.01 mM have been excluded from ournalysis, as this immobilization level has been reported beforeor monolayer formation (Mozsolits et al., 2001; Popplewell et al.,007).

The SPR signal of the lipid modified L1 chip (baseline at the endf step C) is considered the “ground level” for melittin injectiontep: in the following, the sensorgrams are normalized to 0 at theeginning of peptide injection.

Upon melittin injection, the SPR data (inset Fig. 1) show a char-cteristic evolution with distinct domains corresponding to theomplex peptide–lipid interaction: (1) melittin attachment to theipid matrix, (2) melittin insertion and (3) pore formation, lipid

atrix destabilization and dissociation. This last step is clearlyevealed in the sensorgram as a peak value followed by a steepescent of the SPR data. A similar SPR sensorgram has been recentlyeported in an implementation of SPR microscopy analysis of vesicleupture by virus mimetic attack (Chah and Zare, 2008).

The stages of this non-monotonous process depend on the con-entration of melittin injected, the level of lipid immobilized onhe sensor surface and on the quality of the chip. Surface recon-itioning determines a gradual decline of the L1 chip quality, butiscussion on optimization of surface regeneration conditions or onhe implications on lipid binding is out of the scope of the presentrticle.

.2. Assessment of melittin concentration based on characteristicime point in the non-monotonous process of interaction

In “classical” biosensing platforms, there is a monotonous timevolution of measured parameter(s) as a function of the targetnalyte, thus allowing for equilibrium analysis or use of arbitraryhosen time points to examine an interaction or for quantifica-ion. Since the interaction between a pore-forming compound (i.e.

elittin) and a lipid membrane is related to a non-monotonousffect determined by the target analyte, we propose a novel, sensingrocedure to quantify the concentration of melittin.

The pattern of evolution of the complex, non-monotonous pro-ess of peptide–lipid interaction provides a direct way to quantifyhe concentration of melittin. Specifically, the concentration of theest compound can be inferred from Tmax, the time point when theon-monotonous evolution of SPR signal due to peptide–lipid inter-ction reaches its summit. The timing of this characteristic value iselated to the ability of the pore-forming compound to insert intond destabilize the lipid matrix. It occurs during pore formation andestabilization of lipid membrane phase due to dynamic removalf lipids (and associated peptide) from the sensor surface once aestabilization threshold of peptide is reached.

As revealed in Fig. 2, one can construct “calibration curves” relat-ng melittin concentration to the inverse of the characteristic peakime: 1/Tmax that can be used in biosensing. Such a calibrationurve, based on Tmax, does not require any theoretical assumption orodeling. However, one should note the related drawbacks: (i) the

uration of an analysis should exceed occurrence of Tmax, obviouslyonger as concentration decreases, (ii) there is an inherent, theo-etical Limit Of Detection, LOD, in the �M range for concentrationf peptide, based on the condition of reaching characteristic pep-ide:lipid thresholds. We have limited the injection times to 60 min;

his corresponds to the maximum injection time, without sample

odification and bubble formation, in a direct inject configuration.ven if one takes advantage on the “BigInject” option of the Bia-ore 3000 instrument (Biacore Handbook, 2003a), the maximumxperimental time is limited to 150 min (for a 5 �L/min flux).

Fig. 2. “Calibration curve” based on the experimental data (�) and simulated (©)based on the model.

As such, the occurrence of the characteristic peak is a measure ofthe combined effect of both melittin concentration and lipid matrix,providing a powerful sensing platform for moderate concentrationsof pore-forming compounds. It also provides an important tool inplatform optimization and for mechanistic insight.

3.3. Influence of peptide:lipid ratio

An important parameter governing the interaction of pore-forming compounds and lipid platforms is peptide:lipid ratio.Different lipid membranes were employed to analyze the var-ious facets of melittin–lipid interaction. Specific ranges for thepeptide:lipid ratios were reported for the distinct steps in the inter-action process.

Fig. 3 shows characteristic sensorgrams and related data fit-ting for different (injected) peptide:(immobilized) lipid ratios.The sensorgrams for similar lipid coverages (Fig. 3A) reveal aclear dependence between the peak amplitudes and melittinconcentration, with a corresponding shift in time. The experimen-tal cases corresponding to unbalanced peptide:lipid ratios (largepeptide concentration and small lipid coverage and vice versa)are presented in Fig. 3B. This unbalance is revealed through-out melittin–lipid interaction: a faster attachment, insertion andappearance of the peak corresponding to lipid dissociation, occurfor the high peptide:lipid ratio (curve 1). A milder descent in the SPRdata corresponding to lipid dissociation is related to larger “stray”signal coming from direct attachment of melittin to L1 chip uponlipid uncoverage. In contrast, curve 2, shows more overlapped con-tributions of the attachment and insertion processes and a clearerlipid dissociation phase.

The highest peptide:lipid ratios, Fig. 3C reveal an earlier onsetof lipid dissociation prior to complete attachment of the availablemelittin. The smaller initial lipid coverage in sensorgram 2 enablesa larger direct melittin to chip binding and a subsequent alterationof the apparent dynamics in the dissociation phase.

3.4. Quantitative analysis

The proposed kinetic model (Eq. (1–4)) enables compact evalu-ation of these rather diverse SPR data, independent on the sensor

characteristics. All experimental data were fitted on the basis ofthis model and kinetic information was derived for each stage ofthe interaction process. The set of coupled differential equationsprovides the dynamics for each of the components within the mul-tilayer system in relation to the actual SPR data. As such, the model

M. Gheorghiu et al. / Biosensors and Bio

Fig. 3. (A) Sensorgrams corresponding to different melittin concentrations corre-sponding to similar lipid coverages (∼8550 RU) (1) 2.35 �M melittin for 8600 RUlipid, (2) 2.43 �M melittin for 8300 RU lipid, (3) 2.55 �M melittin for 8800 RU lipid.(B) Sensorgrams corresponding to unbalanced peptide:lipid ratios: (1) 3.62 �Mmelittin for 6500 RU lipid coverage and (2) 2.35 �M melittin for 8600 RU lipid cover-age. (C) Sensorgrams corresponding to high peptide:lipid ratios: (1) 3.62 �M melittinfor 6500 RU of lipids, (2) 2.81 �M melittin for 5800 RU of lipids.

electronics 24 (2009) 3517–3523 3521

allows a sufficient degree of flexibility with no a priori considerationof a molecular model of interaction.

Quantitative information on peptide:lipid interaction: kineticconstants for association and dissociation phases, threshold con-centrations corresponding to the phases of melittin insertion andlipid dissolution are derived, expanding the existing information.Based on limited monitoring, previous SPR studies (Mozsolits etal., 2001; Papo and Shai, 2003) have addressed the mechanism ofmelttin–lipid artificial membrane interaction process using differ-ent kinetic analysis models, allowing for limited characterization ofthe interaction process, without targeting quantitative detection ofthe pore-forming analyte.

The sets of model parameters derived, in different experimentalconditions, through quantitative, kinetic analysis of the SPR sen-sorgram, comprise: (a) kinetic data related to melittin interactingwith the lipid matrix: Ka1, Ka2 and Kd0 corresponding to association,insertion and dissociation constants, (b) kinetic data related to lipidmatrix destabilization due to melittin insertion and pore formationKdl0, (c) threshold values for melittin insertion m0, start and mi forthe end of insertion and lipid dissociation mL. As will be pointedout, the associated kinetic parameters have a great importancefor sensing and model elucidation. Moreover, the lipid dissoci-ation equation (Eq. (3)), not considered in any previous model,provides insight on the interaction model and peptide efficiency indestabilizing the lipid matrix. As revealed by data analysis, lipid dis-sociation is largely dependent on randomly associated rather thaninserted melittin, with no defined pore structure. This makes ourresults consistent with “toroidal pore”/“carpet like” models, and thenumerical solution might provide a new experimental support forthis model.

The two model compounds chosen in our study: the zwitterioniclipid (POPC) and melittin as a pore-forming peptide, have associatedcertain material parameters (rendering generality to our approach):v0m, related to peptide volume density (M−1), Kl (m2 M−1), Km (m2)and Ki (M−1) relating surface properties and volume concentrations.

Direct melittin attachment to the sensor surface, has associateddistinct kinetic and state parameters: Ka3 as apparent associationconstant, NC0 (M) and RR as the thresholds for direct association.

The initial lipid matrix is modeled based on Nl0 (M), the con-centration of lipid units on the chip and d2 + dm the associatedthickness.

The average values obtained, from model analysis, for the kineticparameters (rate constants and threshold parameters) are listed inTable 1, together with related standard deviations.

One should note that, except for ka1 and m0 (which are relatedto the electrostatic interactions with the immobilized lipid, and thespecific peptide:lipid ratio) these values are to be considered forsimilar lipid immobilization levels ∼8000 RU.

To cope with the actual experimental cases (corresponding toa wider distribution of peptide and lipid concentrations) one findsspecific dependencies of the ka2, mL or mi values on the peptide andlipid concentrations.

A simulation of the SPR response corresponding to a wider con-centration range of (injected) melittin and immobilized lipid hasbeen performed to illustrate the effect of peptide:lipid ratio (Fig. 4).

The simulation reveals the expected shift of Tmax occurrencetowards longer time intervals. As no mass transport limitationshave been considered in our model, there is a less pronounceddecrease of the corresponding amplitude in comparison with theexperimental data (Fig. 3A). For the time frame used in our exper-iments, concentrations below 1.6 �M did not lead to complete

evolutions with no peak visible in the sensorgram (for the simu-lated 1.6 �M concentration, the peak occurs at ∼7000 s). The actual0.35 �M, 0.7 �M and 1.4 �M (20 min) injections, data not shown,reveal a low amplitude, slow, association phase, with no insertionstep.

3522 M. Gheorghiu et al. / Biosensors and Bioelectronics 24 (2009) 3517–3523

Table 1Association (ka1, ka2, ka3) and dissociation (kd0, kdL0) rate constants and threshold concentration (m0, mi , mL) of melittin in POPC matrix determined for the complete interactionprocess.

Association rate constant (103 M−1 s−1) Dissociation rate constant (10−3 s−1) Threshold concentration (M)

ka1 1.46 ± 0.11 m0 0.024 ± 0.001ka2 0.05 ± 0.01 × 10−3 kd0

ka3 36.56 ± 6.67 kdL0

Fc

3i

amicrpbtoc

Fa

ig. 4. Simulated sensorgrams based on model parameters, different peptide con-entrations covering 1.4–3.50 �M range.

.5. Assessment of melittin concentration based on peptidensertion occurrence

Quantitative assessment of melittin–lipid interaction providesfaster and more sensitive detection avenue as compared with theerely experimental approach based on occurrence of character-

stic peak time, described earlier. While the “classical” approach inonstructing signal/concentration dependencies based on equilib-ium analysis, is not applicable to this non-monotonous complex

eptide–lipid interaction, we propose a novel detection approachased on the direct access, offered by our quantitative analysis, tohe threshold values and the corresponding time values for theirccurrence in different experimental conditions. Fig. 5 illustrates aalibration curve based on the inverse of Tins corresponding to the

ig. 5. Calibration curve based on the moment when the peptide insertion is initi-ted (m0 threshold according to the kinetic model).

0.67 ± 0.27 mi 0.354 ± 0.03510.58 ± 2.11 mL 0.277 ± 0.025

moment when the peptide insertion is initiated (when concentra-tion of associated melittin reaches the threshold m0 according tothe kinetic model).

Using this information, one can lower the analysis time intervaland eliminate the effect of lipid matrix variability, since the peptideassociation only depends on the association constant and peptideconcentration, while lipid coverage plays a limited role. Accordingto Fig. 5, the theoretical LOD for melittin concentration on a POPClipid matrix is 0; a concentration of 0.45 �M of melittin correspondsto a Tins of 1 h as compared to the 1.2 �M limit of detection for 1/Tmax

analysis for the same injection time.

4. Conclusions

Complete SPR monitoring was used, for first time, to reveal andassess the entire non-monotonous interaction process between apore-forming compound (melittin) and a lipid modified surface—L1sensor chip (Biacore). We quantitatively describe the interactionprocess based on a mathematical model encompassing the distinctstages involved in peptide–lipid interaction: association, insertionof melittin into lipid matrix, pore formation and destabilization oflipid membrane.

This model allows for qualitative and quantitative processdescription (via related rate constants and threshold concentra-tions) and provides relevant parameters for biosensing.

Two approaches for sensing melittin concentration via itseffect on lipid matrix have been advanced: (a) “calibration curve”relating melittin concentration to the inverse of the character-istic peak time: 1/Tmax. The detection limit, in the micromolarrange, is imposed by the characteristic peptide:lipid thresholdsdominating the complete interaction process and a specific exper-imental constraint: realistic injection times <3600 s; (b) basedon the threshold values and the corresponding time values fortheir occurrence in different experimental conditions provided byour quantitative analysis. Using this information, one can lowerthe analysis interval and eliminate the effect of lipid matrixvariability.

We envisage that this approach is able to support a compre-hensive detection and analysis platform, which could be furtherextended to membranes with different lipid compositions andother pore-forming compounds, as well. The analysis does notdepend on the structure of the peptide-induced pore. As SPRapproaches have been recently advanced in cell based biosensingformat we stress on the applicability of our approach for a widerclass of lipid matrices and even cells. The proposed model combinedwith appropriate design of the experimental protocol adds a newdepth to the classic SPR investigation of peptide–lipid interactionoffering a quantitative platform for understanding the manifoldfacets of the interaction and for supporting the controlled designof new, improved antimicrobial peptides.

Acknowledgment

Support of EU NMP3-SL-2008-214107 Nanomagma project isgratefully acknowledged.

nd Bio

A

t

R

BBA

A

AA

BB

CDF

FHHKK

L

Wessman, P., Stromstedt, A.A., Malmsten, M., Edwards, K., 2008. Biophys. J. 95 (9),

M. Gheorghiu et al. / Biosensors a

ppendix A. Supplementary data

Supplementary data associated with this article can be found, inhe online version, at doi:10.1016/j.bios.2009.05.007.

eferences

iacore, 2003a. 3000 Instrument Handbook. Biacore AB, Sweden.iacore, 2003b. Sensor Surface Handbook, Biacore AB, Sweden.nderluh, G., Besenicar, M., Kladnik, A., Lakey, J.H., Macek, P., 2005. Anal. Biochem.

344 (1), 43–52.nderluh, G., Dalla Serra, M., Viero, G., Guella, G., Macek, P., Menestrina, G., 2003. J.

Biol. Chem. 278 (46), 45216–45223.sthana, N., Yadav, S.P., Ghosh, J.K., 2004. J. Biol. Chem. 279 (53), 55042–55050.yela, C., Roquet, F., Valera, L., Granier, C., Nicu, L., Pugniere, M., 2007. Biosens. Bio-

electron. 22 (12), 3113–3119.erneche, S., Nina, M., Roux, B., 1998. Biophys. J. 75 (4), 1603–1618.orn, M., Wolf, E., 1980. Principles of Optics: Electromagnetic Theory of Propagation,

Interference and Diffraction of Light, 6th ed. Pergamon Press, Oxford/New York.hah, S., Zare, R.N., 2008. Phys. Chem. Chem. Phys. 10 (22), 3203–3208.empsey, C.E., 1990. Biochim. Biophys. Acta 1031, 143–161.eltis, B.N., Sexton, B.A., Glenn, F.L., Best, M.J., Wilkins, M., Davis, T.J., 2008. Biosens.

Bioelectron. 23 (7), 1131–1136.rey, S., Tamm, L.K., 1991. Biophys. J. 60 (4), 922–930.

abermann, E., Jentsch, J., 1967. Hoppe Seylers Z Physiol. Chem. 348 (1), 37–50.oa, X.D., Kirk, A.G., Tabrizian, M., 2007. Biosens. Bioelectron. 23 (2), 151–160.ass, R.S., 2005. J. Clin. Invest. 115 (8), 1986–1989.umbhat, S., Shankaran, D.R., Kim, S.J., Gobi, K.V., Joshi, V., Miura, N., 2007. Biosens.

Bioelectron. 23 (3), 421–427.adokhin, A.S., White, S.H., 2001. Biochim. Biophys. Acta 1514 (2), 253–260.

electronics 24 (2009) 3517–3523 3523

Lee, M.T., Hung, W.C., Chen, F.Y., Huang, H.W., 2008. Proc. Natl. Acad. Sci. U.S.A. 105(13), 5087–5092.

Lin, J.H., Baumgaertner, A., 2000. Biophys. J. 78 (4), 1714–1724.Lundquist, A., Wessman, P., Rennie, A.R., Edwards, K., 2008. Biochim. Biophys. Acta

1778 (10), 2210–2216.Matsuzaki, K., Yoneyama, S., Miyajima, K., 1997. Biophys. J. 73 (2), 831–838.Mozsolits, H., Thomas, W.G., Aguilar, M.I., 2003. J. Pept. Sci. 9 (2), 77–89.Mozsolits, H., Wirth, H.J., Werkmeister, J., Aguilar, M.I., 2001. Biochim. Biophys. Acta

1512 (1), 64–76.Nakajima, H., Kiyokawa, N., Katagiri, Y.U., Taguchi, T., Suzuki, T., Sekino, T., Mimori,

K., Ebata, T., Saito, M., Nakao, H., Takeda, T., Fujimoto, J., 2001. J. Biol. Chem. 276(46), 42915–42922.

Panchal, R.G., Smart, M.L., Bowser, D.N., Williams, D.A., Petrou, S., 2002. Curr. Pharm.Biotechnol. 3 (2), 99–115.

Papo, N., Shai, Y., 2003. Biochemistry 42 (2), 458–466.Popplewell, J., Freeman, N., Carrington, S., Ronan, G., McDonnell, C., Ford, R.C., 2005.

Biochem. Soc. Trans. 33 (Pt 5), 931–933.Popplewell, J.F., Swann, M.J., Freeman, N.J., McDonnell, C., Ford, R.C., 2007. Biochim.

Biophys. Acta 1768 (1), 13–20.Reitz, J.R., Milford, F.J., Christy, R.W., 1993. Foundations of Electromagnetic Theory,

4th ed. Addison-Wesley, Reading, MA/Wokingham.Shai, Y., 1999. Biochim. Biophys. Acta 1462 (1–2), 55–70.Vogel, H., Jahnig, F., 1986. Biophys. J. 50 (4), 573–582.Vornholt, W., Hartmann, M., Keusgen, M., 2007. Biosens. Bioelectron. 22 (12),

2983–2988.

4324–4336.Wilkop, T., Xu, D., Cheng, Q., 2008. Langmuir 24 (10), 5615–5621.Yang, N., Su, X., Tjong, V., Knoll, W., 2007. Biosens. Bioelectron. 22 (11), 2700–2706.Zhu, W.L., Nan, Y.H., Hahm, K.S., Shin, S.Y., 2007. J. Biochem. Mol. Biol. 40 (6),

1090–1094.