Near-infrared co-illumination of fluorescent proteins reduces photobleaching and phototoxicity

Here we present a method to reduce the photobleaching of fluorescent proteins and the associated phototoxicity. It exploits a photophysical process known as reverse intersystem crossing, which we induce by near-infrared co-illumination during fluorophore excitation. This dual illumination method reduces photobleaching effects 1.5–9.2-fold, can be easily implemented on commercial microscopes and is effective in eukaryotic and prokaryotic cells with a wide range of fluorescent proteins.


Photophysical model of the RP effect
We constructed the simplest photophysical model accounting for the observed reduced photobleaching of EGFP under continuous dual visible and NIR illumination (Fig. 1a,b) and its dependence on the illumination intensity INIR in the NIR (Fig. 1c), the illumination intensity I470 at 470 nm (Fig. 1d) and the wavelength λNIR of the NIR illumination (Fig. 1e).
We first considered a basic RISC model (Supplementary Fig. SD1, black arrows).In this model, the EGFP chromophore is initially excited by absorption of a visible photon (wavelength 470 nm) from the ground-state S0 to the highly fluorescent singlet excited state S1 (1).The excited molecules which do not relax back to the ground state via fluorescence or internal conversion (2) evolve via intersystem crossing (ISC, 3) to the longer-living triplet excited state T1 which can undergo irreversible chemical changes to non-fluorescent (bleached) states D (4) or relax back to S0 (5).Re-excitation of triplet T1 by the NIR illumination (wavelength λNIR, 6) to a higher-lying triplet state Tn decreases the photobleaching rate by depopulating T1.Once formed, Tn can convert back to S1 by reverse intersystem crossing (RISC, 7) or relax to T1 by internal conversion (8).The role of the triplet state T1 as an intermediate in photobleaching and the existence of RISC in EGFP are both supported by the action spectrum of the RP effect (Fig. 1e) which is similar to the published absorption spectrum of the EGFP triplet. 1 Supplementary Fig. SD1: Photophysical model of the RP effect (similar to Fig. 1f).Black arrows correspond to the basic RISC model while purple arrows show the additional processes we had to include to satisfactorily reproduce the experimental intensity-dependencies.
Simulations based on this basic model led however to a linear dependence of the RP effect on I900, in contrast with experimental observations (Supplementary Fig. SD2, black line and red circles, resp.).To obtain a saturation of the RP effect at high I900 (Supplementary Fig. SD2, red line), we had to add a second photobleaching pathway starting from Tn (Supplementary Fig. SD1, 9).Moreover, simulations based on a model including only processes 1 to 9 led to a RP effect independent of I470, in disagreement with the data (Supplementary Fig. SD3, red line and blue circles, resp.).To reproduce the decrease of the RP effect at high I470 (Supplementary Fig. SD3, blue line), we had to take the non-zero absorption of T1 at 470 nm into account (Supplementary Fig. SD1, 10).The latter leads to saturation of the T1→Tn transition by 470-nm light at high I470, so that the NIR co-illumination does not provide any additional benefit.Higher singlet Sp (p > 1) and triplet Tq (q > n) excited states were neglected in our model due to their fast depopulation (ps) to lower-lying states of the same spin multiplicity.
Our final model is close to that of Ringemann et al. who first came up with the idea of exploiting RISC in fluorophores, showing that the process can significantly increase the fluorescence signal of small organic fluorophores excited with very high intensities (0.1-10 MW/cm 2 ). 2 Due to our bottom-up approach of starting from a basic model and gradually adding photophysical processes until we account for the experimental intensity dependencies, and our much lower excitation intensities (1-10 W/cm 2 for visible excitation and up to a few kW/cm 2 for NIR co-illumination), our model is however simpler.In particular, it does not take singlet excited states above S1 into account, nor photobleaching from S1, nor stimulated emission, which could all be neglected in our experimental conditions.

Simulation of the RP effect in EGFP
The differential equations governing the evolution of the concentrations in our final model (Fig. 1f and Supplementary Fig. SD1 with all arrows) are given below (Supplementary Equations SD1-SD5).The photon fluxes F at 470 and 900 nm (in mol.s -1 .cm - ) were calculated from the corresponding power densities I (in W/cm 2 ; Supplementary Equation SD6).The notations used are listed in Supplementary Table SD1.
Supplementary Table SD1 a Values of the extinction coefficient of T 1 at other wavelengths used for the simulation of the action spectrum (Supplementary Fig. SD5) were deduced from  T 1 900 using the T 1 spectrum shape from Ref. 1.
The equations were solved numerically starting from the initial condition [S0](t=0) = 2×10 -5 mol.L -1 and [S1](t=0) = [Sp](t=0) = [T1](t=0) = [Tn](t=0) = 0 in order to obtain the evolution of the concentrations of the different states over time.The fluorescence signal at time t is proportional to the concentration of the bright S1 state.The simulated RP effect for a given set of 470-nm and 900-nm intensities was therefore calculated as the ratio of the integral of [S1](t) in presence of NIR light to its integral in absence of NIR light.We fixed the values of the different photophysical parameters (Supplementary Table SD1) either based on the available literature or in such a way as to minimize the difference between the simulated and measured RP effect at the different 470-nm and 900-nm light intensities (Fig. 1c,d).
The simulated concentration profiles of S0, S1, T1, Tn and D under 470-nm illumination alone (32 W/cm 2 ) or combined with saturating 900-nm co-illumination (2 kW/cm 2 ) are shown in Supplementary Fig. SD4.At the onset of 470-nm illumination, a small population is transferred in a few ns from S0 to S1 (panel a), leading to the fluorescence signal.In the absence of NIR light, S1 and S0 populations which are in rapid equilibrium then partially convert to T1 on the ms timescale (b and e, dotted lines), due to intersystem crossing from S1 to T1.This results in a significant T1 population and a new equilibrium between S0, S1 and T1.These three populations finally decay on the timescale of a few minutes due to conversion of T1 to D (c, d and f, dotted lines).By promoting RISC, NIR light drastically reduces T1 population (e and f, solid orange line) in favor of S0 (solid blue line).S1 and Tn populations are also enhanced (b, c, g and h, solid cyan and red lines).The increase in Tn population makes photobleaching from Tn dominant.With [Tn]/[T1] = 4.1 × 10 -7 and rate constants kDn = ΦDn × kTn = 4.9 × 10 5 s -1 and kD = ΦD × kT1 = 1.4 × 10 -2 s -1 (Supplementary Table SD1), the ratio of the photobleaching rates from Tn and T1 is in fact vDn/vD = kDn/kD × [Tn]/[T1] ≈ 14.The photobleaching rate from Tn in these conditions is however smaller than the photobleaching rate from T1 in the absence of NIR light (Supplementary Fig. SD4d, green lines) due to the much smaller population of Tn than T1.
Our model enabled us to satisfactorily simulate the experimental dependence of the RP effect on I470 and I900 (Fig. 1c,d).We also used it to generate a contour plot of the RP effect as a function of the two light intensities in the range of 1-10 4 W/cm 2 (Fig. 1g).This plot shows that the effect is at maximum (≥ 3-fold increase of the time-integrated emission of EGFP) in the rectangle defined by I470 ≤ 100 W/cm 2 and I900 ≥ 500 W/cm 2 .Finally, we simulated the NIR wavelength dependence of the RP effect (action spectrum; Supplementary Fig. SD5) using values of the molar extinction coefficient of T1 at the different wavelengths deduced from its value at 900 nm (Supplementary Table SD1) and the T1 spectrum shape from Ref. 1.The simulation appropriately predicts the correspondence with T1 spectrum at low NIR intensity as well as the observed broadening at high intensity.The latter comes from the saturation of the RP effect at high NIR intensity, a regime in which the RP effect depends less on the absorbed light intensity, and therefore on the variation of the extinction coefficient with the wavelength.
Although it predicts the dependencies of the RP effect on visible and NIR intensities and NIR wavelength, our model does not fit the experimental photobleaching kinetics well (Supplementary Fig. SD6).Indeed, a satisfactory fit of the data requires three exponential components (Supplementary Fig. SD7), while the simulated photobleaching kinetics contain only one (Supplementary Fig. SD8)neglecting fast components due to S0 to S1 and S1 to T1 conversions (Supplementary Fig. SD4, a and b).This suggests that the photophysics of EGFP involves more states than assumed in our model.In agreement with previous reports, 8-10 we observed a small reversible photobleaching component in EGFP (Supplementary Fig. SD9) indicating the formation of low-yield, long-lived dark states under illumination.2][13] While such states could lead to additional components in the fluorescence decay, they are not expected to be sensitive to NIR light.Moreover, the action spectrum of the RP effect (Fig. 1e) indicates clearly that it originates from re-excitation of the triplet state.Supplementary Fig. SD4: Simulated concentration changes of S1 (a, b, c), D (d), S0 and T1 (e, f) and Tn (g, h) under continuous illumination at 470 nm alone (32 W/cm 2 , dashed lines) or combined with a saturating co-illumination at 900 nm (2 kW/cm 2 , solid lines), shown on different timescales.Note that S1 nanosecond rise due to 470-nm illumination is independent of the presence of NIR light.S1 concentration profiles with and without NIR are therefore superimposed on this timescale (panel a).Supplementary Fig. SD8.Mono-exponential fits of the simulated concentration profiles of S1 under illumination at 470 nm (32 W/cm 2 ; blue and yellow lines) alone or combined with 900-nm co-illumination (2012 W/cm 2 ; red and green lines).The bottom panel shows the residuals of the fits.The obtained lifetime is 193 s for 470-nm illumination and 406 s for dual illumination.Fast components due to S0 to S1 and S1 to T1 conversions were neglected.
Supplementary Fig. SD9.Reversible photobleaching of EGFP.A sample of purified EGFP immobilized in a PAA gel was first continuously illuminated with 32 W/cm 2 of 470-nm light until fully photobleached (1).In a 2 nd phase (2), the intensity was reduced to 3.2 W/cm 2 which led to a slight recovery of the fluorescence on the time scale of 1000 s, as highlighted in the inset.In a 3 rd phase (3), the intensity was increased back to 32 W/cm 2 .The partial fluorescence recovery during the 2 nd phase as well as the higher level of fluorescence at the beginning of the 3 rd phase than at the end of the 1 st phase show the presence of a small component of reversible photobleaching in EGFP.Note that the instantaneous changes in fluorescence intensity at the boundaries between the different phases simply reflect the changes in excitation intensity.
1.3.Sensitivity of the RP effect to the fluorophore's photophysical parameters We show in Fig. 1h that beyond EGFP, the RP effect can be obtained in a wide range of FPs, with however variable amplitudes.In addition, RISC is not specific to FPs but has been demonstrated in a number of organic dyes such as cyanines 14 or xanthenes, 15 suggesting that the RP effect could also occur in small fluorophores.Interestingly, large increases in instantaneous fluorescence assigned to RISC have been reported for several small fluorophores subjected to dual illuminations at extremely high intensities (0.1-10 MW/cm 2 ). 2 In order to understand, or predict, the variations of the RP effect from one fluorophore to another, we examined its sensitivity to the photophysical parameters of fluorophores (molar absorption coefficient of T1, quantum yield of RISC, quantum yields of bleaching from T1 and from Tn, quantum yield and lifetime of T1; Supplementary Fig. SD10-SD12), in the framework of the model of Fig. 1f.The wavelength used to excite the ground-state is noted here λ1 (with intensity I1) and that used to excite the triplet state λ2 (with intensity I2).Each figure panel illustrates the sensitivity of the RP effect to one parameter, with all other parameters set to the values in Supplementary Table SD1.
Changing the value of  T 1 ( 1 ) impacts the dependence of the RP effect on I1 (Supplementary Fig. SD10a), where with a higher  T 1 ( 1 ) value the RP effect is observable only at a lower range of I1.Logically, when  T 1 ( 1 ) = 0 the RP effect is independent on I1.Similarly, the value of  T 1 ( 2 ) impacts the dependence of the RP effect on I2 (Supplementary Fig. SD10b).The higher  T 1 ( 2 ), the less I2 is required to observe the RP effect.
The values of the quantum yield of RISC ( RISC ), bleaching from T1 ( D ) and bleaching from Tn ( D n ) affect the magnitude of the RP effect but not its dependence on the intensities at λ1 and λ2 (Supplementary Fig. SD12).The higher the values of  RISC and  D , the greater the RP effect.Conversely, increasing the value of  D n decreases the effect.Some combinations of values of these three parameters can lead to a RP effect < 1, that is an acceleration of the photobleaching under dual illumination.This could in particular explain the increase in photobleaching observed for bacteria labelled with mCherry or mRFP1 and grown on PBSagarose pads (Supplementary Fig. 3c-d).In order to achieve a RP effect > 1,  D n must be at least 4-5 orders of magnitude lower than  RISC and 2 orders of magnitude lower than  D .Supplementary Fig. SD12: Simulation of the RP effect for different values of a) the quantum yield of RISC ( RISC ), b) the quantum yield of bleaching from T 1 ( D ) and c) the quantum yield of bleaching from T n ( D n ).All other photophysical parameters were set to the values in Supplementary Table SD1.
In conclusion, the variations in amplitude of the RP effect among green and yellow FPs (Fig. 1h) are most likely due to slight variations in several of the photophysical parameters discussed above.
As far as small fluorophores are concerned, we expect the RP effect to take place in a higher intensity regime than for FPs (more precisely, over a wider range of I1 intensities and for higher I2 intensities) because their triplet lifetimes are ~1000 times shorter (microsecond scale).For instance, for fluorescein whose triplet lifetime is 33 μs, 16 we expect an onset of the RP effect for I2 intensities of a few kW/cm 2 , and a maximum effect from several 100 kW/cm 2 (Supplementary Fig. SD13a).We checked this prediction experimentally by exciting fluorescein at 470 nm (35 W/cm 2 ) and re-exciting its triplet in its absorption tail at 700 nm 17 to induce RISC.In agreement with our prediction, we observed an increase in time-integrated emission with 2.5 kW/cm 2 of 700-nm light (maximum intensity achievable with our setup) while 250 W/cm 2 proved insufficient to obtain an effect (Supplementary Fig. SD13b).It is interesting to note that unlike FPs for which the time-integrated emission increases due to a slowing down of photobleaching, here the increase is mainly due to an instantaneous increase in fluorescence at t=0.This observation is in agreement with the work of Ringemann et al. who reported large increases in instantaneous fluorescence for several small fluorophores under dual illumination at very high intensities (0.1-10 MW/cm 2 ). 2 Supplementary Fig. SD13: RP effect in fluorescein.a) Simulated intensity-dependence of the RP effect for a fluorophore with a triplet lifetime of 33 μs.All other photophysical parameters were kept to their EGFP values listed in Supplementary Table SD1.b) Fluorescence photobleaching curves of fluorescein under continuous illumination at 470 nm (35 W/cm 2 ) alone or combined with 700 nm.Fluorescein was immobilized in a polyvinyl alcohol (PVA) film obtained by drying a drop of a 20-μM dye solution containing 1 wt% PVA. 18he measurement was repeated three times and the temperature rise inside the NIR beam never exceeded 0.1 °C.Note that this measurement was carried out under continuous illumination by the NIR beam, and that the temperature rise would be even lower for discontinuous illumination as used in time-lapse microscopy.We therefore expect no difficulties for wide-field microscopy applications with medium to high magnification (40x, 60x and 100x objectives).The temperature issue may however need to be reassessed if the total incident power of the NIR beam were to increase, for example to cover a field of view >200 μm with a NIR power density >1 kW/cm 2 .

Reduction of photobleaching and phototoxicity in primary mouse neutrophils expressing LifeAct-GFP and co-illuminated with NIR light:
We show the results obtained for two cell preparations from different mice.Neutrophils were confined in PDMS (polydimethylsiloxane) microchambers to promote their movement in 2D, and their photobleaching and motility were assessed for 5 min at 37°C under different illumination conditions: weak 470 nm (0.8 W/cm 2 ), strong 470 nm (20 W/cm 2 ) or strong 470 nm (20 W/cm 2 ) combined with 900-nm co-illumination (2 kW/cm 2 ) over part of the field of view.Control cells were exposed to weak 470 nm for 200 ms every 1 s.Other cells were exposed to strong 470 nm with or without 900-nm co-illumination for 400 ms every 1 s (first cell preparation) or for 200 ms every 1 s (second cell preparation).a-c, Fluorescence images of neutrophils at the start, middle and end of illumination with strong 470 nm.Neutrophils indicated by the arrows were co-illuminated at 900 nm and show reduced photobleaching.Scale bars: 10 μm.d, Mean squared displacement (MSD) of neutrophils illuminated with weak (control, black) or strong 470 nm, alone (blue) or combined with 900-nm co-illumination (red).e, Fluorescence photobleaching curves of neutrophils illuminated with weak (control, black) or strong 470 nm, alone (blue) or combined with 900-nm co-illumination (red).All curves were normalized to 1 at t=0.For the first cell preparation, values are mean ± s.e.m. (n=32 cells from 6 microchambers) for 470-nm illumination, mean ± s.e.m. (n=14 cells from 6 microchambers) for dual illumination and mean ± s.e.m. (n=44 cells from 5 microchambers) for control.For the second cell preparation, values are mean ± s.e.m. (n=35 cells from 8 microchambers) for 470-nm illumination, mean ± s.e.m. (n=18 cells from 8 microchambers) for dual illumination and mean ± s.e.m. (n= 44 cells from 5 microchambers) for control.
As can be seen in panels d, the MSD of 470-nm illuminated cells (blue) was significantly lower from the onset than that of control cells (black) due to the phototoxicity of strong 470-nm light.In contrast, cells illuminated at both 470 nm and 900 nm (red) maintained MSD values similar to the control for 60-100 s.Despite a departure from control at longer times, co-illuminated cells kept on migrating further than cells illuminated only at 470 nm.These experiments show a protective effect of NIR light against phototoxicity due to visible excitation.

Growth of EGFP-labelled (a) and unlabelled (b, c) E. coli colonies subjected to different illuminations:
The growth was followed by phase-contrast imaging.We show here the temporal evolution of the binary logarithm of the area of the microcolonies normalized with respect to their initial area.a, EGFP-labelled colonies (BL21) on a LB-agarose pad.Unexposed colonies (black) were subjected to minimal exposures of 100 ms of low intensity white light every 3 min, for the sole purpose of recording images.Exposed colonies were illuminated for 5 s every 3 min with 490-nm light (0.5 W/cm 2 , blue), 885-nm light (0.8 kW/cm 2 , orange) or both (red).Values are mean ± s.e.m. (n=4 colonies from 1 sample).b, Unlabelled colonies (MG1655) on a LB-agarose pad.Unexposed colonies (black) were subjected to minimal exposures of 100 ms of low intensity white light every 3 min, for the sole purpose of recording images.Exposed colonies were illuminated for 5 s every 2 min with 490-nm light (0.5 W/cm 2 , blue), 885-nm light (0.8 kW/cm 2 , orange) or both (red).The circles and crosses show the results for two colonies from the same sample in each illumination condition.c, Unlabelled colonies (MG1655) on a MMcasa-agarose pad.Unexposed colonies (black) were subjected to minimal exposures of 100 ms of low intensity white light every 4 min, for the sole purpose of recording images.Exposed colonies were illuminated for 2 s every 4 min with 517-nm light (1.2 W/cm 2 , blue) alone or combined with 885-nm light (0.8 kW/cm 2 , red).The circles and crosses show the results for two colonies from the same sample in each illumination condition.
Note that the experiments with EGFP-labelled and unlabelled bacteria exposed to 490-nm light were performed using different E. coli strains.They should therefore not be directly compared.The control on unlabelled bacteria is here only intended to illustrate the toxicity of blue light to bacteria, which is well documented in the literature.
Simulations of the expected RP effect dependence on 900-nm intensity for the basic RISC model including only processes 1-8 (black line) and for the first extended model including in addition photobleaching from Tn (process 9; red line).Values are mean ± s.d.(n=3 samples).Simulations of the expected RP effect dependence on 470-nm intensity for the first extended model including only processes 1-9 (red line) and for the final model including in addition T1 absorption at 470 nm (process 10; blue line).Values are mean ± s.d.(n=3 samples).

:
Parameters involved in the simulation of the RP effect of EGFP: notations and values used.Process numbers are as defined in Supplementary Fig.SD1.kX denotes the reciprocal of the lifetime of species X.
Three-exponential fits of the photobleaching kinetics of purified EGFP in PAA gel under continuous illumination at 470 nm (32 W/cm 2 ; blue circles, yellow line) alone or combined with 900 nm (2012 W/cm 2 ; red circles, green line.Data are mean ± s.d. (n=4 different samples).The bottom panel shows the residuals of the fits.The obtained lifetimes are 93 s (84 %), 267 s (15 %) and 1103 s (1 %) for 470-nm illumination and 49 s (9 %), 243 s (27 %) and 559 s (64 %) for dual illumination, where percentages indicate the relative weight of each component.