Residence time of singlet oxygen in membranes

Photodynamic therapy uses photosensitizers (PS) to kill cancer cells by generating reactive oxygen species – like singlet oxygen (SO) - upon illumination with visible light. PS membrane anchoring augments local SO concentration, which in turn increases photodynamic efficiency. The latter may suffer from SO’s escape into the aqueous solution or premature quenching. Here we determined the time constants of SO escape and quenching by target molecules to be in the nanosecond range, the former being threefold longer. We confined PS and dipolar target molecules either to different membrane monolayers or to the same leaflet and assessed their abundance by fluorescence correlation spectroscopy or membrane surface potential measurements. The rate at which the contribution of the dipolar target molecules to membrane dipole potential vanished, served as a measure of the photo-oxidation rate. The solution of the reaction–diffusion equations did not indicate diffusional rate limitations. Nevertheless, reducing the PS-target distance increased photodynamic efficiency by preventing other SO susceptible moieties from protecting the target. Importantly, our analytical model revealed a fourfold difference between SO generation rates per molecule of the two used PSs. Such analysis of PS quantum yield in a membrane environment may help in designing better PSs.

The electrical measurements were performed with the aid of silver-chloride electrodes that were connected to the aqueous compartments via agar bridges. Total electrode resistance did not exceed 30 kOhm. Membrane capacitance and conductance were continuously measured as previously described 13 . The difference of BLM boundary potentials, Δϕ b , was monitored by using the intramembrane field compensation (IFC) method 13,14 (see also reviews 15,16 ). IFC uses a variable dc offset to a sine wave input (300-700 Hz) to minimize membrane capacitance. We measured Δϕ b and determined the photo-oxidation rate of DT as previously described 17 . In brief, we placed the planar lipid bilayer into the focused beam of a monochromatic light source (semiconductor diode laser with a wavelength of 670 nm, optical power 1 mW). Subsequently, we added the PS into the distant aqueous compartment (also called cis compartment) with respect to the light source. DT was added either to the cis compartment or to the opposite (trans) compartment. Since both substances are membrane impermeable, their location remained well defined throughout the experiment".
The membrane surface densities, T and P, as well as the lateral diffusion coefficients, D DT and D PS of both DT and PS were assessed by placing the horizontal BLM formed by Montal-Mueller technique 18 into the focus of a laser scanning microscope (LSM 510 META/ConfoCor 3, Carl Zeiss, Jena, Germany) and exploiting fluorescence correlation spectroscopy, FCS 19,20 : DT PS d , 2 where r and τ d are the lateral focus radius of the confocal volume and the characteristic residence time of DT or PS in the confocal volume, respectively. τ d was obtained by fitting a one-component model for 2D translational diffusion 21 to the autocorrelation function G of the time, τ, dependent fluorescence intensity: where N is the average number of fluorescent molecules in the confocal volume.

Results
PS adsorption to the lipid bilayer. PS membrane adsorption alters bilayer boundary potential ϕ b . The introduced change Δϕ b depends upon PS' aqueous concentration, P a 5,13 (Fig. 1A). Generally, Δϕ b is a superposition of changes in membrane surface potential Δϕ s and membrane dipole potential Δϕ d 15,16 : ϕ s is accessible via measurements of the electrophoretic mobility of lipid vesicles. The mobility reflects the so-called ζ -potential that, in 0.1 M KCl, describes the electrostatic potential at 0.2 nm distance from the vesicle surface. Using the Gouy Chapman theory, ϕ s can be calculated from ζ. Upon AlPcS 3 and AlPcS 4 membrane adsorption, we measured Δϕ b values that can entirely be attributed to changes in ζ 5 , i.e. Δϕ b and Δϕ s are roughly identical. In contrast, for AlPcS 1 and AlPcS 2 negative ζ -values 5 (and thus negative Δϕ s values) have been observed for all aqueous PS concentrations P a , while Δϕ b was positive for some AlPcS 2 concentrations and all AlPcS 1 concentrations. The observation suggests that both AlPcS 2 and AlPcS 1 significantly alter Δϕ d . The underlying dipole moment of the phthalocyanine molecule is generated by the separation of the positive charge at the aluminum cation in the center and the negative charges of the sulfo-groups on its periphery. Thus, in contrast to AlPcS 4 , the charges in AlPcS 1 are distributed asymmetrically (Fig. 1B). The positive sign of Δϕ d indicates that AlPcS 1 inserts the positive Al-moiety into the lipid bilayer, whereas the negatively charged S-group faces the aqueous solution (Fig. 1B). AlPcS 2 is likely to adopt a similar orientation (Fig. 1B). Our result is in line with AlPcS n accessibility by fluorides 5 and molecular dynamic simulations with porphyrin 22 . Determination of membrane surface PS and DT densities. We performed FCS experiments to find T and P (Fig. 2). DT elicits Δϕ b changes that are proportional to its aqueous concentrations 17 . That is, DT's limited aqueous solubility prevents the linear relation between T a and T from breaking down as might be expected in case of a Langmuir adsorption isotherm. Since Δϕ b and the aqueous DT concentration T a are the proportional to each other 17 , there must be a linear relationship between T a , and T. It allowed calculating that T amounts to 30 molecules per μm 2 and per nM of T a . Thus, for T a = 60 nM we find T = 1800 molecules μm −2 ( Fig. 2A). Accordingly, DT's contribution to Δϕ b amounts to 0.003 molecules/nm 2 /mV.
From τ DT = 1.31 ms, we found D DT = 7.6 µm 2 /s (Eq. 2). We used FITC (fluorescein isothiocyanate) for calibration experiments. Its diffusion coefficient D FITC is equal to 565 µm 2 /s 23,24 . From the equation: we determined the confocal volume V = 0.22 fl. The structural parameter S (defined as the ratio of the long (ω z ) to the short radii (ω x = ω y = ω r ) of the ellipsoidal confocal volume and τ FITC were equal to 5 and 18 µs, respectively. Similarly, we obtained D PS = 3.4 µm 2 /s from PS residence time in the focal plane τ PS = 2.6 ms. At an aqueous AlPcS 2 concentration P a = 30 nM (Fig. 2B), we observed P = 1700 molecules/μm 2 . The corresponding adsorption coefficient amounts to 57 molecules/μm 2 per nM P a . The focal volume of the red laser was calibrated with Cy5. We took Cy5's diffusion coefficient and S to be equal to 280 µm 2 /s 25,26 and 8, respectively. Cy5's residence time τ Cy5 amounted to 31.6 µs.
The photodynamic activity of phthalocyanines with various numbers of sulfo-groups. When both DT and PS were adsorbed to the BLM surface, membrane illumination at 670 nm led to a drop in Δϕ b due to DT's oxidation by SO 17 . Δϕ b recovered in the dark due to the adsorption of intact DT molecules from the aqueous bulk solution (Fig. 3). The kinetics of photo damage depended on whether DT and PS were added into the solution to the same (cis configuration) side or to the opposite (trans configuration) side of the membrane. Dividing Δϕ b (t) by its value ϕ ads at time t = 0 we define the rate R of DT oxidation as 17 : enabling a quantitative analysis of the photodynamic effects. We found that R depended (i) on the number of sulfo-groups per PS molecule (Fig. 4A), (ii) the geometrical arrangement (cis or trans configuration), and (iii) the aqueous PS concentration. For PS molecules with two sulfo-groups, R depended non-monotonically on P a in trans configuration. Similar bell-shaped concentration dependencies were reported for SO-mediated gramicidin inactivation 5 . In contrast, R increased monotonically with P a in cis configuration, suggesting that SO quenching by PS may be involved. This would be in line with chemoluminescence-based observations of SO quenching by PS 27 .
In order to account for differences in AlPcS n membrane affinities, we replotted R as a function of P (Fig. 4B). For AlPcS 3 and AlPcS 4 , we calculated P from the increment in surface charge 5 , whereas we used FCS to determine P of the weakly charged AlPcS 2 and AlPcS 1 . R did not significantly vary with the number of sulfogroups in cis configuration: R was roughly proportional to P for P < 0.01 molecules/nm 2 , and it approached saturation at higher P. However, in trans configuration, we observed a 10-fold augmented R for AlPcS 2 and the absence of saturation for AlPcS 4 . The latter may be due to AlPcS 4 's low binding constant.  Mathematical model. Our counterintuitive observation that R is smaller if PS and DT are adsorbed to the same leaflet than to opposite ones has three possible explanations: (1) PS and DT molecules interact with each other thereby decreasing the quantum yield of SO, (2) PS molecules quench SO or (3) DT molecules possess two different moieties that may serve as a SO target: damage of only one of these moieties -of the aniline groupchanges membrane dipole potential, whereas targeting the second moiety -the unsaturated hydrocarbon chain in the middle of the molecule -remains electrically silent.
First, we ruled out the potential interaction by recording the effect of DT on the fluorescence spectra of liposomal PS (see Supplementary Fig. S1). Since there was none, a DT-PS interaction is unlikely. Second, we estimated the impact of PS mediated SO quenching. Since it should depend on the concentration of both molecules, we would expect the ratio R c /R t to vary with PS concentration. The invariance of that ratio at low PS concentrations (Fig. 4) excludes SO quenching by PS from being responsible for the difference between R c and R t . Third, to estimate the impact of SO quenching by DT molecules, we plotted 1/R against ϕ ads (Fig. 5). The slopes of the curves in "cis" configuration exceeded the slopes in "trans" configuration for all ϕ ads , thereby confirming DT's quenching effect. The difference between the two configurations was more pronounced when AlPcS 4 was substituted for AlPcS 2 , suggesting a deeper bilayer penetration depth of AlPcS 2 , and thus, a smaller PS to DT distance.
Differences in the oxidation rates of styryl dyes analogues 17 indicate that DT possess two different moieties that may serve as a SO target: damage of only one of these moieties -of the aniline group -changes membrane dipole potential, whereas targeting the second moiety -the unsaturated hydrocarbon chain in the middle of the molecule -remains electrically silent. This peculiarity generates the different R in cis and trans configurations: In cis configuration, DT's naphthalene ring is only ~ 1 nm farther away from the PS molecule than DT's double bond (Fig. 6). Consequently, the double bond may provide protection for the naphthalene ring. In trans configuration, the naphthalene ring is closer to the PS molecule, and thus, it may represent the primary SO target. This hypothesis only makes sense if τ r is very short -small enough to prevent SO from reaching distant targets, and definitely much shorter than suggested by the previously estimated τ l = 3 µs in a liposomal suspension.
We supported the hypothesis about 2 SO susceptible DT moieties with a mathematical model that takes into account all chemical reactions and membrane diffusion processes (see Theory). The numerical solution of the differential equations satisfactorily fits the experimental data (Fig. 5) for the parameter set displayed in Table 1. The model divides the lipid membrane into three layers and assumes that SO is generated by PS molecules in one of the outer layers. In contrast, DT molecules extend from the outer to the middle layer. The Δϕ d affecting moiety (designated by letter T in Fig. 6) -the naphthalene group at the end of the chromophore molecule -is immersed into the middle layer 17 . The other moiety that is susceptible to SO -the unsaturated hydrocarbon chain in the middle of the molecule (designated by letter S in the Fig. 6) -localizes to one of the outer membrane layers. SO may quench the S-group even when the T-moiety is oxidized and DT's contribution to Δϕ d has vanished.
We noticed that -contrary to the expectations -the limit of the oxidation rate at zero DT density depends on which side of the membrane contains DT (Fig. 5). This observation suggests that the DT preparation already contained oxidized molecules, which were capable of capturing SO, while being unable to contribute to Δϕ b .

Theory
Mathematical model of SO generation, membrane permeation, and quenching. The model describes the permeation of singlet oxygen through the membrane taking into account the non-uniform distribution of SO across the membrane. SO mirrors the bell-shaped distribution of ground-state oxygen with the concentration maximum in the middle of the membrane 10 . The steady-state concentration of SO not only depends on this distribution, but also on its lifetime. To simplify this distribution, we will consider it to be discrete. Let the membrane consist of three layers, two of them are the fields of the lipid bilayer lying near its interface with water, and the third one -the internal hydrophobic region of the membrane, is located in the layers of hydrocarbon chains of the lipids. We will refer to these layers as cis (the layer with the PS molecules), medium (the middle of the membrane) and trans (the layer opposite to cis). The concentrations of singlet oxygen in these layers will be designated as O c , O m and O t , correspondingly. The singlet oxygen is generated from the excited molecules of PS adsorbed on the cis-side of membrane. No matter which of the locations of the PS molecules is considered, ground state oxygen is present at saturating concentrations with respect to SO generation. Consequently, both AlPcS 2 and AlPcS 4 operate at a rate that is solely governed by light intensity (see for example The target of SO can be located either in the cis or trans layer. The model assumes that two parts of the target molecule react with singlet oxygen (Fig. 6). The first one, the aniline group is designated by "T". The oxidation of this group by singlet oxygen in the middle layer (O m in Fig. 6) nullifies their contribution to membrane dipole potential. The concentrations of this group in cis and trans layers are correspondingly designated, as T c and T t . The other part -the unsaturated hydrocarbon chain in the middle of the molecule (designated by "S" in Fig. 6) can react with singlet oxygen in the surface layers (O c , or O t ) without changing the dipole potential. The approach neglects the incremental diffusion span of no more than 0.5 nm that a SO molecule has to cross if born on the "wrong" (opposite to the direction of movement) side of a PS molecule. It corresponds to travel time of only ~ 1 ns, which is at least 10 times smaller than the most conservative estimate of τ dw = 12 ns from molecular dynamics simulations 10 .
The reactions and transfer of singlet oxygen in cis and trans photoeffects are shown in Fig. 6. The singlet oxygen can react with target molecules, be quenched by the medium, the molecules or photosensitizer and transfer to the adjacent layer. Once SO has left the membrane, the probability of its retrieval is negligible since (i) the volume of the aqueous bulk so much larger than that of the membrane, (ii) and the lifetime in aqueous solutions is very limited. The equations describe the damage of the target and the processes with singlet oxygen in the experiment, where the target molecules located at the cis side of the membrane are  Table 1. Model parameters. They represent the best fit of the system of differential equations (T4) to the experimental data (Fig. 5A,B).
To simplify the solution of these equations, we will consider that the processes described here are fast when compared with the processes of oxidation and diffusion of the targets described in 17 . If the concentration of the singlet oxygen changes much faster than that of the targets, it reaches the steady-state value before the concentration of the target significantly begins to change. Therefore, we can solve the equations assuming the steady-state distribution of singlet oxygen and equilibrium distribution of the targets.
In this case, the equations for the singlet oxygen concentration transform to the algebraic ones. By solving these equations and introducing the steady-state concentrations of the singlet oxygen into the first equations in (T1) and (T2) we only have single differential equations describing the kinetics of oxidation of the target either in cis or in trans positions. Thus we will instead solve these equations by determining the rate of oxidation of the target R c and R t at cis and trans positions, respectively, as a slope (time derivative) of the relative change of the potential at the beginning of illumination according to Eq. 5. The relation between the potential measured in the experiment and the surface density of the target molecules can be established experimentally by comparing the dipole potentials measured by IFC method at given concentration of the ANEPPS in the solution (the data in 17 ) and the surface density of these molecules measured by the FCS technique (Fig. 2). These values are proportional to each other: Relative potential can be derived through the ratio of current surface density of the target to the initial one corresponding to equilibrium of targets between the membrane and solution before the illumination For further simplification of equations (T1 and T2), we will neglect SO quenching by PS, putting the corresponding constant, k PS,O , equal to 0. It may be valid if we use very low concentrations of photosensitizers in the linear region of the dependence of the rate R on the concentrations of photosensitizers in the solution or on their surface densities (Fig. 4). It seems natural to assume that the surface densities of two parts of the same targets molecule, T and S, are equal. However, one cannot exclude the existence of a fraction of the target molecules, which are partially damaged and do not contribute to the dipole potential, but still can react with singlet oxygen and quench it. To take into account these "invisible" target molecules we will assume that surface density at each side of the membrane consists of two members: "visible" T and "invisible" S 0 Solving the equations (T1 and T2) with the above-mentioned simplifications gives the rates R c and R t as functions of the ANEPPS adsorption potential   We used equation T4 to fit the data in Fig. 5. k c-t,m was not used as a fitting parameter. but was obtained from the known membrane permeability P M = 80 cm/s of O 2 28,29 that we assume to be equal to P M of SO: Since we are not interested in the time τ that SO takes to cross the bilayer, but in the time that SO requires to diffuse from the middle of one of the outer layers to the middle of the membrane, we substitute δ and 1/τ for δ/3 and k c-t,m , respectively. This yields k c−t,m = 2.4 × 10 8 s −1 .
S 0 was treated as a constant parameter for each of PS and was found from the global fit as αS 0 = 30 mV, equal both for AlPcS 2 and AlPcS 4 .
One of the simplest consequences of the model is that the ratio of R c and R t is always less than 1. This ratio is equal to

Discussion
Our experiments revealed that SO in membranes is extremely short-lived: instead of spending microseconds in the membrane interior, SO does not survive tens of nanoseconds. In other words, τ r ≈ τ l . τ l is mainly governed by the time it takes SO to reach its nearest target. The conclusion is based on experiments, in which a SO molecule that was born in the membrane headgroup region was free to target molecules in (i) the same headgroup region, (ii) the inner hydrophobic layer, and (iii) the opposing headgroup region. Although the increment in distance between these targets was smaller than 2 nm, the probability of SO reacting with the targets sharply decreased with distance.
SCIENTIFIC REPORTS | (2018) 8:14000 | DOI:10.1038/s41598-018-31901-9 A lower limit of τ l maybe obtained by assuming a diffusion limited reaction. In our experiments, the DT molecules (for ϕ ads = 20 mV) were only 4 nm apart as indicated by FCS measurements. Thus, in cis configuration, the distance r between a PS and the nearest DT neighbor molecule did not exceed 2 nm. Taking into account SO's diffusion coefficient D = 5 × 10 −5 cm 2 s −1 7 , we find the SO travel time τ t is probably smaller in biological membranes since they are densely packed with SO scavengers that are represented by unsaturated lipids and membrane proteins containing aromatic residues [30][31][32][33] .
In order to obtain a more accurate estimate of τ l , we performed a quantitative analysis of our experiments by solving the differential equations for the combined system of chemical reactions and diffusion processes (see Theory). We found the simplified solution represented by Eq. (7): where T 0 , S 0 , R c , and R t denote T prior to illumination, the initial surface density of DT's oxidizable moiety that does not contribute to Δϕ b (Fig. 6), R in cis and trans configurations, respectively. The R t /R c ratio depends on the combined rate k W of SO membrane exit and quenching by the aqueous medium, the transfer rate k c-t,m from the outer membrane layer to the medium layer, and the rate k S,O of SO quenching by the S moiety of DT (Fig. 6). Eq. 7 assumes a non-negligible value S 0 , i.e. it does not work for T 0 ≫ S 0 . Equation (7) predicts that R c is always smaller than R t , which is in perfect agreement with the experiment. We first globally fitted the complete set of differential equations (see Theory) to individual sets of R c = f(T 0 ) and R t = f(T 0 ) that have been experimentally obtained for both AlPcS 4 and AlPcS 2 ( Table 1 and Fig. 5A,B) and then constructed the R t /R c = f(T 0 ) plot from the results of the fit (Fig. 5C). Solving the complete set of equations also allowed us to obtain the rate k Gen of SO generation. k Gen is fourfold higher for AlPcS 2 than for AlPcS 4 , indicating either an environmental effect on SO yield or immediate SO escape into the aqueous solution without ever entering the lipid membrane. Indeed, in contrast to AlPcS 4 's four negative charges, which ensure that it lies flat on the membrane surface, AlPcS 2 penetrates into the bilayer. Its chromophore ring inclines towards the hydrocarbon core as indicated by a substantial contribution of Δϕ d to Δϕ b . Thus, AlPcS 2 's ring is exposed to a four-fold higher oxygen concentration than that belonging to AlPcS 4 . The observed increment in k Gen agrees well with the previously reported increase of photosensitizer efficiency with increasing hydrophobicity 11,34,35 . Our mathematical model revealed the reaction rates of SO with targets: The rate k T,O of SO quenching by the T moiety of DT is roughly equal to k S,O ≥ 6.0·10 −10 m 2 s −1 . Taking into account the target density of about 0.1 molecule nm −2 , this value translates into a reaction time τ T,O ≈ 16 ns of the T moiety with SO. τ T,O is three orders of magnitude smaller than the estimate for τ l from experiments 36 with non-oxidizable lipids, underpinning the conclusion that the main determinant of SO lifetime is target density. Moreover, τ l is not diffusion limited, i.e. it is not governed by τ t : τ l ≈ τ T,O ≫ τ t .
The mathematical model also allows assessing τ dw . From k m,c−t = 5.5 × 10 7 s −1 and k w = (5.0 ± 0.9) × 10 7 s −1 ( Table 1) we find τ dw ~ (1/k m,c−t ) + (1/k w ) ≈ (40 ± 8) ns. This estimate agrees reasonably very well with the one derived from the literature values of SO's distribution coefficient and diffusion constant (see Introduction). τ l < τ dw indicates that most of the reactive oxygen species stay in the membrane long enough to react with the target, i.e. that τ r is determined by τ l . However, the values of τ l and τ dw are too close to each other to ensure a photodynamic efficiency of 100%. An increase in targeting efficiency can be realized by two different approaches: (i) by reducing the separation between PS and target so that other molecules susceptible to SO, like unsaturated lipids, cannot protect the proteinaceous target from encounters with SO and (ii) by increasing the quantum yield of PS molecules. Our model allows determining k Gen in a membrane environment, thereby offering an important tool for PS optimization.

Data Availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.