Keto-polymethines: a versatile class of dyes with outstanding spectroscopic properties for in cellulo and in vivo two-photon microscopy imaging

The keto-heptamethine family has been expanded to various symmetrical and asymmetrical structures.


Introduction
Fluorescence microscopy has become a routine technique in biological research, and linear or non-linear microscopes are now standard equipment in most biological and biomedical research laboratories. 1 In particular non-linear two-photon microscopy (TPLSM) is generally considered as a non-invasive technique, which allows in vitro or in vivo cell imaging in tissues, detection of chemical analytes, or monitoring of biological processes with excellent spatial and temporal resolution. 2 The increasing development of microscopy techniques has invigorated the design of new luminescent probes featuring improved one-or two-photon brightness in biological media. 3 Among the broad category of commercially-available, synthetic uorescent molecules, polymethine dyes such as Cy5 (pentamethine) or Cy7 (heptamethine) derivatives, including the popular indocyanine green (ICG), are certainly the most commonly used for bio-medical applications (Chart 1). 4 The key advantages of polymethines are their large molar extinction coefficients (3) and moderate-to-good uorescence quantum yields (F), which result in excellent one-photon brightness (dened as the product of the extinction coefficient and the quantum yield, 3F) in the far-red or near-infrared regions. This spectral range is referred to as the "biological transparency window" (BTW: 680-1300 nm), where tissues absorption scattering and autouorescence are minimized, and thus where most research endeavors are concentrated. 3 However, polymethine dyes present several inherent drawbacks that may limit their broad application in future bio-imaging protocols. In particular, the cyanine-type electronic transitions are associated with very small Stokes shis (<1000 cm À1 ), which requires ltering of the incident or scattered excitation light and loss of most of the emission signal resulting in a decrease of the signal-to-noise ratio.
In 2005, Peng and co-workers reported the amino-heptamethine dye NCy7 (Chart 1) featuring a large Stokes shi (ca. 3500 cm À1 ) and a high uorescence quantum yield (ca. 40%) in the BTW. 5 This dye has since been adapted for bio-conjugation, bio-imaging, or FRET experiments and used for bio-sensing of metal ions like Zn 2+ , Cu 2+ , reactive oxygen species, or thiolcontaining residues like glutathione. 6 We recently demonstrated that it was possible to ne tune the electronic structure of heptamethine derivatives by simple nucleophilic substitution reactions on the halogenated central carbon of a common precursor, and to generate a variety of dyes with absorption bands covering the whole UV-visible spectral range. 7 These results were rationalized on the basis of an alteration of the initial cyanine to a bis-dipolar ground state electronic conguration. Among these dyes, we became particularly interested in polymethines functionalized by carbonyl moieties at the central carbon position, the so-called keto-polymethine that features excellent one-photon brightness in methanol (3F ¼ 24 500 L mol À1 cm À1 at 636 nm, D i D j ]O in Chart 1). 8 Keto-polymethine dyes were initially reported and studied (both theoretically and experimentally) in the early 2000s. 9 These chromophores were then used as pH sensors in integrated waveguide devices, 10 and as NIR electrochemical uorescence switches. 11 Surprisingly, the exploration of their potential for bio-imaging applications is scarce. Recently, Shabat and co-workers reported the quinone derivatives (QCy7, Chart 1) that enable intravital imaging of hydrogen peroxide. 12 In all other cases, keto-heptamethines were only reported as the reaction products of polymethine dyes involved in the detection of hydrazine, hydrogen sulde or cysteine both in vitro and in vivo. 13 Herein, we aim to expand the synthesis of keto-polymethine dyes featuring the classical bis-indolenine (D) electron-donating end-groups (D i D j ]O) where i, j designate identical or different substituents of the indolenine moieties. In addition we reported original symmetrically (AA]O) or asymmetrically (D i A]O) substituted keto-polymethine dyes containing anionic (A) tricyanofuran end-groups acting here as very strong donors (Chart 1). The linear and nonlinear photophysical properties of these new molecules were thoroughly investigated and the results show that these structural modications lead to major changes in their spectroscopic properties, which makes it possible to achieve very high one-or two-photon brightness in the BTW. In all cases, experimental and theoretical evidence indicates that hydrogen-bonding effects induce a strong increase of the uorescence quantum yield. The hydrophilic/ hydrophobic balance of these chromophores was then modulated by tuning the substituents on the indolenine fragment (alkyl chains, sulfonate moieties, hydrosoluble polymers), and the resulting water-soluble dyes were employed in bio-imaging experiments. We demonstrate that lipophilic dyes can be rapidly internalized into living cells and stain cytoplasm organelles, whereas amphiphilic derivatives are primarily localized in the plasma membrane. The most hydrophilic dyes are not internalized, but are advantageously used as two-photon probes for imaging the cerebrovascular structures of mice in a NIR-to-NIR conguration, thus expanding the scope of ketopolymethines as uorescent biomarkers.

Synthesis
The synthetic route towards all studied molecules is outlined in Scheme 1 and described in detail in the Experimental section. Briey, every keto-heptamethine derivative was synthesized following the general strategy reported by Strekowski et al. 9c relying on a key nucleophilic substitution step on the central position of the corresponding chlorinated heptamethine. For convenience, in what follows, chloro-heptamethines will be referred to as D i D j depending on the nature of the terminal substituents on both extremities of the molecule, and on the substitution of the indolenine fragment (D i with i ¼ 1-5, Scheme 1).
The starting chlorinated cyanines were classically obtained through a Knoevenagel reaction in basic anhydrous conditions between the chlorinated bis-aldehyde 1 and various indolenium salts (D 1 to D 5 ). For the latter, N-alkylations on the indolenium ring were used to obtain molecules with a range of different hydrophilic/lipophilic characteristics (see Scheme 1 and the Experimental section). The unsymmetrical chloro-heptamethine D 2 D 5 precursor featuring amphiphilic behavior (one sulfonate and one lipophilic extremity) was prepared in two successive Knoevenagel condensations: (i) the enol derivative 3 was obtained in 74% yield by reaction of bis-aldehyde 1 and the indolenium precursor D 5 at 80 C in the absence of base; (ii) the second indolenium D 2 was further added and the second condensation was achieved at 80 C in the presence of pyridine as a base. The water-soluble keto-heptamethine DD PHEA ]O, containing hydrophilic polymer chains was prepared following our previously reported methodology. 14 This functionalization was achieved through the D 4 D 4 intermediate that features bishydroxyl functionalities enabling the subsequent introduction of bromoisobutyryl groups (D 6 D 6 ). The corresponding keto-derivative, D 6 D 6 ]O, was prepared and the bromoisobutyryl moieties were nally used as initiators for the living Atom Transfer Radical Polymerisation (ATRP) reaction of 2-hydroxyethylacrylate monomer. 14,15 Aer purication of the resulting chromophore-polymer by dialysis, the 1 H NMR of DD PHEA ]O allowed us to calculate an average degree of polymerization of n ¼ 80 at both extremities of the chromophore. 2D diffusionordered NMR spectroscopy (DOSY) was used to estimate a polydispersity index (PDI) for this compound, following the methodology initially developed by Delsuc and collaborators. 16a A PDI ¼ 1.09 was calculated, highlighting the remarkable control provided by ATRP (see Fig. S1 †). The estimation of PDI using gel permeation chromatography techniques (GPC) resulted in a slightly higher value (PDI ¼ 1.39, see Fig. S2 †). By analogy with previous studies, 14,16 the observed discrepancies between the two techniques were attributed to the branched nature of the polymer chains that leads to an overestimation of polydispersity indexes by GPC. 17 A minimal length of 80 units was found to be necessary for the target biological application, as our attempts to produce shorter chains resulted in a non-water soluble material.
The synthesis of heptamethine AA was readily achieved by a Knoevenagel reaction between 1 and the tricyanofuran derivative A in acidic conditions, following published procedures. 18 Finally in the case of unsymmetrical heptamethines, the chlorinated precursors D i A were synthesized following our previously reported methodology, 19 consisting of two successive Knoevenagel reactions on 1, rst with the tricyanofuran withdrawing group in order to form 2, then with one of the indolenium salt derivatives (D i ). It is particularly remarkable that the formation of the ketone, which has been reported so far exclusively from cationic derivatives (D i D i ), was straightforwardly adaptable from both the neutral (D i A) and even anionic (AA) analogues, with identical reaction conditions and similar high yields. Using this approach, we were able to synthesize the entire series of chromophores depicted in Scheme 1. It is worth noting that compounds D i D j ]O (i, j ¼ 1-5) are neutral whereas D i A]O and AA]O become anionic and di-anionic, respectively as unambiguously demonstrated by mass spectrometry analysis. All target chromophores and new intermediates were fully characterized by 1 H and 13 C NMR and High-Resolution Mass Spectroscopy (HRMS). In the case of D i A]O heptamethines, broadening (and, in one case, splitting) of the 1 H and 13 C NMR peaks could be assigned to cis-to-trans isomerizations of the non-cyclic double bonds of the conjugated backbone that operates on the NMR time scale. 18 However, 1 H NMR experiments recorded at low (218 K) and high (368 K) temperatures did not allow observation of the coalescence of broad and split signals (see ESI †).

Photophysical properties
The spectroscopic properties of the lipophilic keto-heptamethine derivatives (D 1 D 1 ]O, D 3 A]O, AA]O) featuring different substituents were studied in a variety of solvents, in order to evaluate their potential as two-photon uorescent markers for biological applications. Representative absorption and emission spectra of D i D j ]O, D i A]O and AA]O keto-chromophores in methanol are presented in Fig. 1. The optical properties in different solvents are displayed in Fig. 2 and Table 1, while  Table 2 summarizes the data in methanol (or water, when available, vide supra) for all the new keto-derivatives.
Compared to the chlorinated parent molecules, which present a typical cyanine-type absorption band, all keto-derivatives exhibit a distinctive broad and almost structureless absorption band. These transitions clearly present a marked charge transfer (CT) character, typical for bis-dipole-type electronic structure that is a general hallmark for the keto-substitution. 7 As noted in ref. 7, the absorption band showed systematic hypochromic and hypsochromic shis compared to the corresponding chloro-heptamethine chromophores. Molar extinction coefficients are relatively similar within the series, with values in the range of 50 000-60 000 L mol À1 cm À1 . The transition energies vary in the order which is associated with the nature of the terminal substituents ( Fig. 1).
The absorption properties of keto-heptamethines are highly sensitive to the protic or aprotic nature of the solvent. As shown in Fig. 2, the position of the absorption band is only little dependent on solvent polarity; conversely a signicant red-shi is observed between aprotic and protic solvents. As an example, the absorption in octanol is red-shied compared to that in a more polar solvent like DMSO. This can be almost certainly correlated to the formation of hydrogen bonding between the chromophore and protic solvent molecules with the keto-heptamethine acting  as the hydrogen bond acceptor (vide supra). This hypothesis is discussed thoroughly and proved true in the theoretical calculation section.
The same dependence on the solvent acidity (protic or aprotic nature) is observed in the emission spectra. In nonprotic solvents, all compounds show a relatively classical emission prole evolution, i.e. a red shi of the emission upon increasing solvent polarity (positive emission solvatochromism), which comes with a loss of its ne vibronic structure. This is a typical signature of a uorescence emission arising from a CT excited state. In contrast, a different trend is observed in protic solvents: the emission becomes much narrower and is strongly red-shied by up to 124 nm (3810 cm À1 ) for D 1 D 1 ]O, even as compared to the most polar aprotic solvent (DMSO). It is also remarkable that the position of the band in protic solvents is only weakly affected by their polarity. Indeed, for both D 1 D 1 ] O and D 2 A]O, a comparatively modest 30 nm red-shi (805 and 660 cm À1 respectively) is seen between octanol and methanol. Again, hydrogen bonding constitutes the only reasonable explanation to account for the observed behavior and the hydrogen-bonding effect is discussed below. The most remarkable effect associated with protic solvents concerns the emission quantum yield and lifetime of the chromophores (Table 1). Whereas all of the molecules in this study are poorly uorescent in non-protic solvents (F < 5% for D i D j ]O), their quantum yields increase spectacularly up to 50% in protic ones, which constitutes an exceptionally high value, in the far-red (D i D j ]O) and near-infrared (AA]O) spectral range. The uorescence lifetimes follow the same trend: the mono-exponential decay is particularly short in non-protic solvents (s < 0.5 ns) and increased to 1.2-1.8 ns in protic solvents. These observations constitute a clear indication that the non-radiative relaxation pathways are disfavored in protic solvents, where hydrogen bonding with the solvent occurs.
. c Not measured because the lifetime is below our detection limit (0.5 ns).

Effect of hydrogen bonds
As stated above, the protic nature of the solvent plays a dramatic role on the photophysical properties of all keto-derivatives. We hypothesized that this effect results from a hydrogen bonding process, where the keto-chromophores and the protic solvent play the roles of hydrogen bond acceptor and donor, respectively, following the equilibrium depicted on top of Fig. 3.
In order to test this hypothesis, titration experiments were performed with the D 1 D 1 ]O, D 2 A]O and AA]O chromophores, following a classical spectroscopic titration protocol, as detailed in the ESI. † As shown in Fig. 3, a progressive decrease of the absorption and emission bands assigned to the "hydrogen bond-free" form (noted C) was observed. It was accompanied by a concomitant increase of the corresponding absorption and emission bands from the "hydrogen bonded" chromophore (noted C-S). The evolution of the intensity of the emission band associated to C upon addition of methanol could be used to calculate, at each point, the concentration of the free dye [C], the H-bond adduct [C-S] and free methanol [S]. A linear correlation was obtained upon plotting [C-S] versus [C] Â [S], for which the slope corresponds to the association constant K a (Fig. 3C, for more information about the methodology, see ESI †). In each case, the three titration experiments could be tted with a 1 : 1 binding isotherm with a very good accuracy (R 2 > 0.99) conrming that the spectral changes observed in protic solvents originate from hydrogen bonds between the solvent and the keto-chromophore. K a values of 2.61, 1.97 and 2.02 L mol À1 (i.e. 2.4, 1.65 and 1.75 kJ mol À1 , respectively) were found for D 1 D 1 ]O, D 2 A]O and AA]O, respectively. These association constants are rather low but support the occurrence of hydrogen bonding between the keto-heptamethines and protic solvents.
To further illustrate and provide a molecular scale description of the interaction of keto-polymethine dyes with the surrounding solvent molecules, we performed molecular dynamics (MD) simulations of the model DD Me ]O (i.e. with R i ¼ CH 3 ) in methanol and dichloromethane (see computational details in the Experimental section). From these simulations, we extracted the radial distribution functions for O*-H (O* refers to the oxygen of the keto-heptamethine carbonyl moieties) in the two solvents. These data are presented in Fig. S3. † The O*-H radial distribution function in methanol displays a sharp peak at $1.8Å. This conrms the suggested strong interaction between the methanol molecules and DD Me ]O that is much weaker in the dichloromethane solution. This is more clearly highlighted in Fig. 4 which shows the strong localization of hydrogen atoms in the vicinity of the O* atom. This conrms the involvement of hydrogen bonds between the chromophore and protic solvents.

pH effect
The strong dependency of the absorption and emission spectra of all keto-derivatives on the solvent's proticity and hydrogen bonding effects led us to investigate the behavior of these dyes upon protonation. Protonation of keto-polymethines have been reported already in the case of DD]O analogs 9a leading to the formation of the cationic hydroxyl derivative (DD-OH) featuring cyanine-type spectroscopic properties. 7 A similar study was rst performed on D 1 D 1 ]O and next generalized to As can be seen from the characteristic shape of the absorption and emission bands, the cyanine character of all molecules is fully restored upon protonation. As a result, both the absorption and emission maxima are strongly red-shied, e.g. to 900 nm for the absorption of AA-OH, and the Stokes shis are strongly decreased compared to the corresponding ketoderivatives. Unfortunately, all protonated forms (D 1 D 1 -OH, D 3 A-OH and AA-OH) exhibit a poor stability in solution, as illustrated by the relatively rapid decrease of the cyanine absorption band. This instability precluded further quantitative study on these molecules, such as pK a measurements and, a fortiori, their use as pH sensitive probes.

Nonlinear optical properties
Besides having signicant luminescence properties in the NIR, a chromophore should also display intense TPA to be considered as a relevant probe for two-photon microscopy in a NIR-to-NIR conguration. The TPA spectra of D 1 D 1 ]O, D 3 A]O and AA]O were mapped using the two-photon excited uorescence (TPEF) 20 and, due to the lack of a stable and reliable TPEF standard in NIR, the magnitude of cross sections were calibrated with open-aperture z-scan techniques. 21 As a general feature, all keto-heptamethines present similar TPA spectra composed of two transitions (Fig. 6). The lowest energy TPA band is weak and matches well the wavelength-doubled one-photon absorption spectra. On the other hand, the transition energy of an intense TPA band is signicantly blue shied   compared to one-photon transition energy. These spectral characteristics are generally associated with centrosymmetric quadrupolar molecules, in which selection rules for one-and two-photon absorption electronic transitions are mutually exclusive. In the case of non-centrosymmetric molecules, the transition energy of TPA and OPA generally overlap. In the present study, all keto-heptamethine dyes are non-centrosymmetric since they present either a quasi-C 2v symmetric CT-dipolar (D 3 A]O) or C 2v symmetric CT-bis-dipolar (D 1 D 1 ]O and AA]O) ground state electronic structure. Such molecules with C 2v or quasi-C 2v symmetries can be treated as bent quadrupolar molecules, of which the quadrupolar character results in intense, blue-shied TPA while the residual dipolar character gives a weak TPA band that generally overlaps with the one-photon transition. Similar observations were recently reported by Perry and co-workers, including diarylboryl dyes and extended Michler's ketone derivatives, a class of molecules that are closely related to keto-heptamethines. 22 In our case, remarkably large two-photon efficiencies were measured for the whole series, with two-photon cross section (s 2 ) values ranging from ca. 570 to 640 GM at 900 and 944 nm for D 1 D 1 ]O and D 3 A]O, respectively. In the case of the di-anionic ketone AA] O, a TPA cross-section of 1400 GM was recorded at 960 nm, which constitutes a high value that is comparable with other TCF-heptamethines reported previously. 21a

Water-soluble keto-heptamethines
The detailed spectroscopic study of the lipophilic probes allowed us to identify interesting candidates as two-photon uorescent probes. In particular, all measured keto-heptamethines present high two-photon brightness (s 2 F) of ca. 250, 345 and 460 GM in methanol for D 1 D 1 ]O, D 3 A]O and AA]O, respectively, with two-photon absorption and emission maxima both located in the red-NIR spectral range. Consequently, these chromophores are ideal candidates for two-photon imaging microscopy in the NIR-to-NIR microscopy conguration. In order to test these chromophores in the context of cellular imaging and in vivo vascular imaging, it was rst necessary to render the selected probes water-soluble and biocompatible. To that end, two types of functionalization of the indole moieties were considered using (i) sulfonate groups (D 5 ) or (ii) water-solubilizing polymers (D PHEA ). 14b For D i D j ]O derivatives (Scheme 1), functionalization could be introduced either symmetrically on both indolenium groups leading to the formation of D 5 D 5 ]O and DD PHEA ]O or selectively at one extremity providing the amphiphilic D 2 D 5 ]O and D 5 A]O derivatives. The photophysical properties of these chromophores in water were then investigated, except for D 2 D 5 ]O, which did not display sufficient water-solubility to allow this study. For all water-soluble dyes, the shapes and positions of the absorption and emission bands are very similar to those obtained in polar protic solvents, which indicates systematic occurrence of hydrogen bonding with water (Fig. 7). In the case of D 5 A]O, a severe drop in the emission efficiency was observed in water (F ¼ 7%, vs. 37% in MeOH). This may be due to an aggregative quenching of the emission related to the low solubility of the molecule in the medium. In contrast, the two hydrophilic molecules D 5 D 5 ]O and DD PHEA ]O conserved good quantum yields of ca. 30% in water, which constitute high values for red emitting water-soluble uorophores.

Cellular imaging
Human T24 cancer cells were imaged with keto-heptamethine dyes featuring various hydrophilic/hydrophobic groups in order to evaluate the inuence of the chromophore structure (charge, substitution) on the internalization and localization processes. All cells were imaged using either two-photon ( Fig. 8 and S4 †) or one-photon uorescence confocal microscopies ( Fig. 9 and S5 †). With both techniques, highly contrasted images were obtained for all probes.
It can be noted that for the highly hydrophilic dyes that are either neutral (DD PHEA ]O) or dianionic (D 5 D 5 ]O), no internalization in the cells was observed. The chromophores stayed in the extracellular medium and the cells appeared as dark spots surrounded by a uorescent medium ( Fig. 8 and S5 †). This feature is clearly detrimental for specic cellular imaging, but is potentially of interest for other imaging applications, as illustrated below. On the other hand, highly lipophilic dyes like neutral D 2 D 2 ]O or monoanionic D 2 A]O in DMSO solution were rapidly internalized in cells. The chromophores preferentially stained organelles in the cytoplasm and nucleoli inside the nucleus. Interestingly, AA]O presents the same behavior indicating that, in spite of its dianionic charge, this dye displays a strong lipophilic character (Fig. 9). This result can be explained by the strong charge delocalization over the entire C sp 2 -skeleton. 7,18 Similarly the anionic charge delocalization of  The emission proles were recorded within the cells incubated with keto-dyes (D 5 A]O and AA]O) in different channels from 570 to 720 nm (Fig. 10). Interestingly, in both cases, the shape of the spectra consists of two emission bands whose relative intensity depends on the local environment. Comparison with emission spectra recorded in methanol or acetonitrile clearly indicates that these two contributions correspond to emission of the H-bonded and H-bond free chromophores. In the case of D 5 A]O, a hydrogen-bond free form of the dye is exclusively seen in the cytoplasm organelles (point 1), with an emission peak around 580 nm (the shape of the signal coincides with the emission of D 2 A]O in acetonitrile reported in Fig. 2 and Table 1). In contrast, the signal extracted from the culture media (point 2) exhibits a behavior close to    600 and 660 nm, respectively. The emission proles match again the uorescence spectra of AA]O recorded in acetone and octanol. This dual emission imaging reveals the formation (or not) of hydrogen bonds in cellulo and suggests that ketoheptamethines can be considered as useful tools for probing the protic environment in cells or other biological media.
In vivo two-photon microscopy Finally, the two highly water-soluble keto-derivatives (D 5 D 5 ]O and DD PHEA ]O) were investigated in the context of intravital two-photon microscopy imaging of mice cerebral vascular networks. As these dyes exhibit both good solubility in water and high two-photon brightness in the biological transparency window, they could be potential candidates for in-depth NIR-to-NIR microscopy imaging.
As illustrated in Fig. 11, both probes stain the blood plasma with strong two-photon excitation efficiency at 900 nm. The images obtained were highly contrasted and allowed performing a deep brain vasculature imaging, up to depths of 500 mm. However the staining was not persistent for a long period. In the case of D 5 D 5 ]O, the diffusion of the dye across the blood brain barrier (BBB) occurred in the rst ve minutes. Blurry spots appeared on images (circle on Fig. 11a). Fig. 11b shows the diffusion of the dye within the brain tissue (dotted curved line). This leakage phenomenon increased over time as illustrated by the larger blurred area aer 25 min (Fig. 11b). It might be due to a photothermic effect, related to the large absorption coefficient of the molecule and to non-radiative dissipation of the absorbed energy, which can damage the vascular endothelium inducing micro-hemorrhage. In the case of DD PHEA ]O, which features large water-solubilizing polymeric chains, the release of the dye through blood vessels was not observed (Fig. 11d), yet the cerebral vasculature staining rapidly decreased over time.
As shown in Fig. 11e, the image of the cerebral microvasculature becomes almost unresolvable 40 min aer intra-venous injection. This difference in the probe behavior can be attributed to their difference in clearance mechanism. While the PHEA appended probe seems to be preferentially metabolized by the kidneys where it makes aggregates in tubule epithelial cells (Fig. 12c), D 5 D 5 ]O still circulates in the kidney tubules 2 hours aer intravenous injection (Fig. 12b). In contrast, the D 5 D 5 ]O derivative stains hepatocytes (Fig. 12e) whereas the PHEA probe keeps circulating in the liver vasculature (Fig. 12f). These results indicate that these two chromophores featuring identical conjugated skeletons behave very differently in vivo. The inuence of the pendant water-solubilizing groups thus turns out to be crucial regarding the probe properties. Sulfonate containing compounds diffuse through the blood vessels and tend to be more toxic due to liver accumulation. In contrast, the toxicity of water-soluble polymer containing dyes appears to be lower and no diffusion is observed. For these reasons, it seems that DD PHEA ]O is the best suited for in vivo applications in spite of its quite rapid excretion.

Conclusions
This article constitutes the rst systematic evaluation of the potential of keto-polymethine dyes for uorescence bio-imaging by confocal and two-photon microscopies. Its scope is not limited to the classical D i D j ]O derivatives, as our study involved original D i A]O and AA]O analogous species. Detailed spectroscopic investigations clearly demonstrated that this class of dyes provides a valuable alternative to some of the most efficient one-and two-photon uorophores known to date, including state-of-the-art heptamethines. Regarding confocal microscopy, their large quantum yield in the red-NIR region and their enhanced Stokes shi are substantial advantages over their parent polymethine dyes. For two-photon microscopy,  regardless of their symmetry and the nature of their end groups, all keto-heptamethines possess strong two-photon absorption in the 900-950 nm range, which makes them ideal candidates for NIR-to-NIR imaging applications.
The potential of these dyes in microscopy imaging was thus systematically investigated. We showed that the control of the hydrophilic/hydrophobic balance along with appropriate functionalization of D i D j ]O and D i A]O regulates the live cell internalization and localization. Finally, highly hydrosoluble dyes were successfully used for in vivo two-photon microscopy of mouse brain vascular networks.

Experimental section
Linear and nonlinear optical measurements UV-visible-NIR absorption spectra were recorded on a Jasco® V-670 spectrophotometer in spectrophotometric grade solvents (ca. 10 À5 mol L À1 ). Molar extinction coefficients (3) were precisely determined at least two times. The luminescence spectra were measured using a Horiba-Jobin Yvon Fluorolog-3® spectrouorimeter, equipped with a three slit double grating excitation and emission monochromator with dispersions of 2.1 nm mm À1 (1200 grooves per mm). The steady-state luminescence was excited by unpolarized light from a 450 W xenon CW lamp and detected at an angle of 90 for diluted solution measurements (10 mm quartz cuvette) by a red-sensitive Hamamatsu R928 photomultiplier tube. Spectra were reference corrected for both the excitation source light intensity variation (lamp and grating) and the emission spectral response (detector and grating). Fluorescence quantum yields Q were measured in diluted solution with an optical density lower than 0.1 using the following equation Q x /Q r ¼ [A r (l)/A x (l)][n x 2 /n r 2 ][D x /D r ] where A is the absorbance at the excitation wavelength (l), n the refractive index and D the integrated intensity. "r" and "x" stand for reference and sample. Excitation of reference and sample compounds was performed at the same wavelength. Short luminescence decay was monitored with the TC-SPC Horiba apparatus using Ludox in distilled water to determine the instrumental response function used for deconvolution. Excitation was performed using NanoLEDs, with models (peak wavelength; pulse duration) 495 (490 nm; <250 ps) and 740 (732 nm; 1.3 ns). The deconvolution was performed using the DAS6 uorescence-decay analysis soware. TPA spectra of keto-heptamethines were mapped with a TPEF setup and the TPA cross sections were characterized with an open-aperture z-scan technique. In both the TPEF and z-scan techniques, an amplied, femtosecond-pulsed Ti:sapphire laser (Solstice, Spectra-Physics, USA) was used. For TPEF measurements, laser excitation wavelengths were ranging from 760-1300 nm. Rhodamine 6G and LDS821 were used as uorescence standards. Quartz cuvettes with 1 cm pathlength were used and the optical densities of solutions were <1. Due to the poor stability of LDS821, TPA cross sections of keto-heptamethines were measured with an open-aperture z-scan technique and the TPA spectra were calibrated accordingly. For z-scan measurements, a near Gaussian beam at 900 and 950 nm with M 2 $ 1.15, beam waist u(HW 1/e2 ) $ 40 mm, and pulse width s p (HW 1/e ) $ 100 fs was used. The excitation irradiance ranges from 80-360 GW cm À2 . The optical pathlength of sample cuvettes for z-scan measurements was 1 mm.

Simulations details
Molecular dynamics simulations were performed using the self-consistent-charge density-functional based tight-binding method (SCC-DFTB), an approximated DFT scheme whose computational efficiency relies on the use of a minimal valence basis-set and parameterized integrals. 23 In the present work, we used the mio-set 23 for Slater-Koster tables for methanol based calculations and the halorg parameters to describe the dichloromethane based calculations. 24 To this end, we used an empirical dispersion term and a correction for the Coulomb interaction in which the Mulliken charges are replaced by the Class IV -Charge Model 3 developed in the context of DFT 25 and later introduced in the SCC-DFTB potential. 26 The importance of including such terms to accurately describe molecular aggregates was demonstrated by Simon and co-workers, in particular, these authors showed that they lead to accurate results for water molecule aggregates. 27 Jahangiri et al. also showed that the SCC-DFTB model and its extensions provide reliable results for the structure, energetics, charge distributions, and vibrational spectral of various hydrogen-bonded systems. 28 Details on the parameterization of the potential is provided in ESI. † In the present work, we used the mio-set for Slater-Koster tables and the halorg parameters to describe the halogen atoms. 24 The SCC-DFTB approach requires additional corrections to properly deal with weak intermolecular interactions. 26 To this end, we used an empirical dispersion term 29 and a correction for the Coulomb interaction in which the Mulliken charges are replaced by the class IV -Charge Model 3 developed in the context of DFT 25 and later introduced in the SCC-DFTB potential. 30 In all our calculations, the self-consistent process is stopped when the largest atomic charge uctuation is smaller than 10 À8 a.u. All the SCC-DFTB calculations were performed with the deMonNano code. 31 We performed MD simulations of the simple symmetrical keto-polymethine molecule, i.e., DD Me ]O where R i ¼ CH 3 into two different solvents namely methanol and dichloromethane. The simulations consisted of a unique molecule in a 25.0Å cubic box. To obtain a density that is close to the ambient temperature and pressure density of the systems, 213 and 134 molecules of methanol and dichloromethane were included in the simulation cells. All the simulations were performed in the canonical ensemble. The systems were equilibrated during 20 ps at 300 K using a Berendsen thermostat 32 following by 20 ps using a Nosé-Hoover chain thermostat 33 dened by a thermostat frequency and a number of thermostats in the chain of 800 cm À1 and 5, respectively. We then performed 80 ps of production run using the Nose-Hoover chain thermostat. In all calculations, we used a time step of 0.5 fs.

Cell culturing and treatment
We used the T24 human epithelial bladder cancer cell line (ATCC No. HBT-4). In our experiments, T24 cells were cultured in 25 cm 2 tissue-culture asks (T25) at 37 C, in a humidied atmosphere with 5% CO 2 . They were incubated in RPMI 1640 supplemented with 100 U mL À1 penicillin, 100 mg mL À1 streptomycin, and 10% fetal calf serum (complete medium). Cells were grown to near conuence in the culture asks and then suspended with 0.05% trypsin-EDTA solution (Sigma). Twentyfour hours before experiments, cells were placed on a LabTek I chambered cover glass (Nunc) at low cell density in complete culture medium. Living T24 cells were incubated for 30 min with solutions of the dyes in water or water/DMSO (98/2) resulting in an overall chromophore concentration in the medium of ca. 1 to 2 Â 10 À5 M and imaged without rinsing.

Confocal microscopy
All confocal experiments were performed using a LSM710 NLO (Carl Zeiss) confocal laser scanning microscope based on an inverted motorized stand (AxioObserver, Zeiss). The excitation was provided by a 561 nm DPSS cw laser in descanned detection mode. In the former case the pinhole was closed to 1 Airy Unit and in the latter one it was fully open. Spectral imaging was realized using an internal Quasar detector in the range 577-723 nm with the resolution of 9.7 nm.

In vivo two-photon microscopy
In accordance with the policy of Grenoble Institute of Neuroscience (GIN) and French legislation, experiments were done in compliance with the European Community Council Directive of November 24, 1986 (86/609/EEC). The research involving animals was authorized by the Direction Départementale des Services Vétérinaires de l'Isère -Ministère de l'Agriculture et de la Pêche, France and the Direction Départementale de la protection des populations -Préfecture de l'Isère-France (F. Appaix, PhD, permit number 38 09 39). All efforts were made to minimize the number of mice used and their suffering during the experimental procedure. CD1 mice were housed in cages with food and water ad libitum in a 12 h light/dark cycle at 22 AE 1 C. For in vivo two-photon microscopy, 4 months old CD1 mice (n ¼ 4) were anesthetized using isourane (5% for induction and 1-2% during experiments) in a 70% air, 30% O 2 gas mixture. Their body temperature was monitored with a rectal probe and maintained at 36 C using a heating blanket. A MouseOx system (STARR Life Sciences Corp.) for monitoring arterial O 2 saturation, heart and breath rate was used.
A catheter (Neoon™, BD, USA) was inserted in the tail vein for an intravenous (iv) injection of 0.1 mL keto-heptamethines (D 5 D 5 ]O and DD PHEA ]O) (5 mg mL À1 ) in saline just before the imaging experiments. A mixture of Hoechst 34580 and Fluorescein IsoThioCyanate-dextran (FITC-dextran, 70 kDa) was injected 2 h aer the rst iv injection (0.1 mL) to visualize both the vasculature and the nuclei of the two main organs of elimination. Ten minutes aerward, one kidney and the liver were removed for ex vivo imaging.
For intravital two-photon imaging of the cerebral vasculature, a craniotomy of 2-3 mm in diameter was performed with a dental drill above the motor cortex and lled with ultrasound gel. In some experiments, the skull was thinned instead of a craniotomy using the same drill. The head was xed in a homebuilt stereotactic frame.
For two-photon microscopy of ex vivo organs, the removed liver and kidney were put in a Petri dish and covered with ultrasound gel. Two-photon microscopy was performed using a LSM 7MP (Zeiss, Germany) equipped with a 20Â water-immersion objective (NA 1.0; Zeiss) and ZEN 2010 soware. Laser excitation at 950 nm was done using a Ti:sapphire laser (Chameleon Vision II; Coherent, UK). All the TPM images were obtained using a constant laser power around 60 mW. Fluorescence emissions were detected simultaneously by three non-descanned detectors with a 492/SP25 nm lter (Semrock, USA) for "blue" uorescence emission, a 542/50 nm lter (Semrock, USA) for "green" uorescence emission and a 617/73 nm lter (Semrock, USA) for "red" uorescence emission. Most 3D two-photon microscopic images were acquired as z-stacks with a 2 mm step size between each focus plane. The z-projections were performed with ImageJ soware 34 and Vaa3D soware was used for 3D images reconstruction. 35