N-Cyanorhodamines: cell-permeant, photostable and bathochromically shifted analogues of fluoresceins

Fluorescein and its analogues have found only limited use in biological imaging because of the poor photostability and cell membrane impermeability of their O-unprotected forms. Herein, we report rationally designed N-cyanorhodamines as orange- to red-emitting, photostable and cell-permeant fluorescent labels negatively charged at physiological pH values and thus devoid of off-targeting artifacts often observed for cationic fluorophores. In combination with well-established fluorescent labels, self-labelling protein (HaloTag, SNAP-tag) ligands derived from N-cyanorhodamines permit up to four-colour confocal and super-resolution STED imaging in living cells.


Introduction
One of the most important advantages of small molecule uorescent probes over genetically encoded uorescent proteins is their superior photostability, 1 which becomes essential under the demanding conditions of super-resolution microscopy (nanoscopy). 2 Recent advances in uorescence nanoscopy have prompted the rational design of small molecule uorescent labels. 3 Achieving live-cell compatibility with high target speci-city, 4 and the ability to control the net charge of a uorescent probe, 5 while maintaining high brightness, chemical stability and low photoreactivity of the uorophore 6 remain considerable challenges.
Fluorescein (resorcinolphthalein) is a green-emitting (l exc ¼ 498 nm, l em ¼ 517 nm, uorescence quantum yield F ¼ 0.90 in 0.1 M NaOH) 7 uorescent dye from the phthalein family of triarylmethane dyes, rst reported by von Baeyer in 1871. 8 While its pK a value of 6.3 makes it brightly uorescent at cytosolic pH ¼ $7.2, 9 the anionic form does not cross the cell membranes of mammalian and plant cells. 10 On the contrary, its non-uorescent O-acyl esters, 10 in particular uorescein diacetate, enter living cells and are hydrolyzed by cell esterases into free uorescein. Employing self-labelling protein tags such as Hal-oTag, 11 CLIP-tag 12 and SNAP-tag, 13 uorescein ester-derived cellpermeant and uorogenic live-cell probes targeting specic fusion proteins 14 have been developed. Fluorescein itself, however, undergoes comparatively rapid photobleaching with a quantum yield of F bl ¼ 3 Â 10 À5 in water. 15 Its 2 0 ,7 0 -diuoro derivative (Oregon Green) is somewhat more photostable and has found wider use in uorescence microscopy. Besides Oacylation, photocleavable 16 and enzymatically cleavable 17 Oprotecting groups have been used to render uorescein and Oregon Green dyes and their derived probes cell-permeant. Replacing the oxygen bridge (X ¼ O) in the xanthene core of uorescein affords its analogues (thiouorescein 18 (X ¼ S), carbouorescein 19 (X ¼ CMe 2 ) and Si-uorescein 20 (X ¼ SiMe 2 )) with lower LUMO energies and red shis of both uorescence excitation and emission maxima (Fig. 1). Their photostabilities, however, have not been systematically studied.
Rhodamine dyes are among the most widely used uorescent labels in super-resolution live-cell microscopy 21 due to their ability to cross intact mammalian cell membranes. Their cell permeability has been attributed to the equilibrium between the colourless spirolactone form prevailing in non-polar, aprotic media and the coloured, uorescent zwitterionic form dominating in aqueous environment. This behaviour masks the potential off-targeting of lipophilic rhodamine-based ligands to cell membrane structures such as the endoplasmic reticulum (ER), but makes these probes unsuitable for labelling transmembrane proteins (e.g. b-adrenoceptors 22 ). Thus, uoresceins or negatively charged sulfonated rhodamine labels are oen preferred in this context. 23 Moreover, the use of positively charged rhodamine probes in live-cell imaging is severely limited by their off-targeting to mitochondria. 24 Unlike uoresceins, rhodamine labels remain uorescent in acidic media and will appear as bright granules whenever accumulated in lysosomes (pH 4.6-5.0) and endosomes (pH 5-6), 9 which prompted the development of more complex self-quenching uorogenic rhodamine-tetrazine conjugates 25 for wash-free imaging. 25b,c The superior brightness and photostability of rhodamines, especially those lacking N-alkyl substituents 26 prompted us to develop their analogues bearing a negative charge delocalised across the xanthene core. Indeed, a recent report by Sharma et al. 27 demonstrated that rhodols and rhodamines with electron-decient N-(2,2,2-triuoroethanesulfonyl) substituents, unlike colourless N-acylrhodamines, 28 retain uorescence in neutral and basic aqueous solutions. In the pioneering work of Wang et al., 29 N-cyanoamide group was employed to increase the NH-acidity of rhodamine amides so as to ne-tune their spirocyclization behaviour. Here, we describe the synthesis and characterization of N-cyano-and N,N 0 -dicyanorhodamines as red-shied, cell-permeant and biocompatible analogues of uorescein dyes with excellent photostability, and demonstrate their application in stimulated emission depletion (STED) nanoscopy in living mammalian cells.

Results and discussion
Synthesis of N-cyanorhodamines and N,N 0dicyanorhodamines Unsubstituted cyanamide is a weak acid (pK a ¼ 16.9 in DMSO), comparable to sulfonamides (pK a ¼ 10-18) but much stronger than most primary amides (pK a ¼ 17-25). 30 It is ambiphilic and, as a nucleophile, less reactive than primary or secondary amines, 31 which makes it a problematic cross-coupling partner. Based on a literature precedent, 32 our initial attempts aimed at Pd-catalyzed Buchwald-Hartwig arylation of cyanamide. We have found, however, that this transformation is only successful with rhodol triate 1 as substrate, but fails for uorescein ditriate 2 ( Fig. 2A). This prompted us to look into Ullmann coupling for alternative reaction conditions 33 using halo-uorans 3, 4 as starting materials. 26 Employing the catalytic system rst reported by Ding et al., 34 we were able to obtain the asymmetric N-cyanorhodamines 5a and 5b from the corresponding monoiodides as well as the target N,N 0 -dicyanorhodamines (6a-c) from known 3 0 ,6 0 -diiodouorans (Fig. 2B). The use of a less active catalytic system (1,10-phenanthroline/CuI) resulted in formation of the mono-cyanamide product 7 in modest yield (see Table S1 † for optimization details).
Regioisomerically pure 6-carboxyuorescein analogue CR1, ready for derivatization to uorescent ligands, could also be prepared under these conditions; however, the transformation proved difficult for 6-carboxylated 3 0 ,6 0 -diiodocarbo-and Si-uorans. An alternative strategy based on the base-induced degradation of 1-aryltetrazoles into N-arylcyanamides was therefore designed for these uorophores (Fig. 3A). The unprotected 6-carboxyrhodamine 110 8a and its carbo-and Sirhodamine analogues 8b and 8c were condensed with sodium azide and orthoformate ester to yield 3 0 ,6 0 -bis-(1-tetrazolyl)uoranes 9a-c in moderate yields (except for 600SiR 5 ). u-Chloroalkane group (HaloTag ligand) was then introduced under peptide coupling conditions followed by treatment with KOH to unmask the N-cyanamido substituents. In the case of CR1, the direct coupling with the suitable self-labelling protein ligands was possible (albeit with lower yield) in the presence of free cyanamido substituents, but it is generally to be avoided because of the side reactivity of cyanamido groups.

Photophysical characterization
With the synthetic access to the diverse N-cyano-and N,N 0dicyanorhodamines established, we next studied their optical and physicochemical properties. We rst determined their absorption and uorescence emission spectra ( Fig. S1 and S2 †),  Table 1 and Fig. 4A-E. According to our expectations, in neutral phosphate buffer N,N 0 -dicyanorhodamine 6a demonstrated absorption and emission maxima nearly exactly matching those of tetramethylrhodamine (TMR) with a 10 nm larger Stokes shi, showing $60 nm bathochromic shis compared to uorescein (Table 1, Fig. 4B). For the corresponding N-cyanorhodamine analogue 5a, the observed $30 nm bathochromic shi (as compared to N,N-dimethylrhodol 10) represented the effect of introducing a single Ncyanamido substituent (Fig. 4C).
For the N,N 0 -dicyanocarbo-and Si-rhodamines CR2, CR3, a similar trend was noted providing potential labelling options for multicolour uorescence imaging, with uorophore emission maxima reaching 690 nm for CR3 (Fig. 3B). The uorescence quantum yield and lifetime values of 6a were lower than for uorescein and TMR, but related well to those of rhodol 10 ( Table 1). Chemical stability of the N-cyanamido group in the context of N-cyanorhodamines within the biologically relevant pH 5 to 10 range was also conrmed for 6a (Fig. S7 †).
The dye 6a (pK a ¼ 5.1) was found to be signicantly more acidic than uorescein (pK a ¼ 6.1) and N,N-dimethylrhodol 10 (pK a ¼ 5.8), corresponding to 6a existing nearly completely in its colourless protonated form at pH # 4 (Fig. 4D). Mono-N-cyanorhodamines 5a-c demonstrated an even higher acidity (pK a ¼ 4.6-4.9). The molecular brightness of N-cyanorhodamines  could be increased by the introduction of an N-azetidinyl auxochromic group (5c) reported to suppress the transition into a twisted internal charge transfer (TICT) excited state, 6 which undergoes non-radiative relaxation. Following the alternative strategy of rigidizing the dialkylamino substituent within the julolidine context of 5b (ref. 35) led to a similar improvement.
There is a general consensus that intact cell membrane permeability of rhodamine-type uorophores depends on their propensity to reversibly close into colourless and non-uorescent spirolactone forms 21 (Fig. 1). The electrophilicity of the C-9 atom of the uorescent xanthylium form of rhodamines and their analogues, responsible for the lactone ring closure, generally increases with the introduction of electronwithdrawing substituents at the amino groups. For every dye, the spirolactone-zwitterion equilibrium can be characterized by the D 0.5 value, 36 which is easily tested by recording a series of absorption or uorescence emission spectra of the uorophores in 1,4-dioxane/water mixtures with varying water content ( Table  1, Fig. 4E). All studied N,N 0 -dicyanorhodamines demonstrated D 0.5 $ 50, approaching the high value of the iconic Sirhodamine dye (D 0.5 ¼ 64.5), 36 widely appreciated in biological imaging for its uorogenicity and far-red uorescence emission (l max ¼ $660 nm). 37 For N-cyanorhodamines (except for the electron-rich 5b), lower D 0.5 values were obtained ($40), similar to carborhodamine dyes 580CP and 610CP. 36 The stability of the cyanorhodamine dyes against photolysis was tested by continuous irradiation of their dilute (3.3 mM) solutions in basic sodium borate buffer (pH 9.9) under air with intermittent recording of absorption spectra. Under these conditions, uorescein dyes are known to rapidly decompose into intractable mixtures of low-molecular weight polar products, while rhodamines are signicantly more photostable. For benchmarking, the solutions of 5a and 6a were irradiated with a 530 nm light-emitting diode (LED), and their photobleaching rates were compared with those of TMR, N,N-dimethylrhodol (10) and carbouorescein (accounting for the varying optical densities at the excitation wavelength, Fig. S8 †). As evident from these data (Fig. 4F), the anionic forms of 5a and 6a are as photostable as the cationic TMR, while the rhodol and especially uorescein dyes undergo rapid photodegradation.

Fluorogenicity and binding kinetics of HaloTag ligands
The HaloTag protein, 11 an engineered version of the Rhodococcus rhodochrous dehalogenase DhaA, forms a covalent ester bond between its active site aspartate residue and linear uchloroalkanes in an S N 2-type reaction. Since the uncatalyzed reactivity of simple chloroalkanes is very low, using HaloTag fusion proteins allows for selective bioorthogonal labelling with only a small molecular weight adjunct. As HaloTag enzyme variants were initially optimized for the TMR chloroalkane (TMR-Halo, Table 2) substrate for uorescent tagging, their reaction rates with cationic and zwitterionic rhodamine-type ligands are unprecedentedly high for self-labelling covalent tags, approaching the diffusion limits for ligands derived from the dyes 610CP and abberior LIVE 580. 39 On the contrary, the apparent second-order reaction rates reported for uncharged small molecular weight chloroalkanes and the negatively charged sulfonated rhodamine Alexa Fluor 488 were $1000 times lower (k app 10 4 -10 5 M À1 s À1 with HaloTag7). 39 We therefore considered important to evaluate the labelling kinetics of N,N 0 -dicyanorhodamine HaloTag ligands, negatively charged at  (Table 1) Fig. S9 †), and chemoselectivity of the labelling reaction was veried by mass spectrometry (Fig. S10 †). Fortunately, all three uorescent substrates CR1-Halo-CR3-Halo demonstrated k app ¼ $10 6 M À1 s À1 with the HaloTag7 protein ( Table 2, Fig. S11 †), a kinetic behaviour comparable to the SNAP-tag protein/O 6benzylguanine ligand system 13 widely used in live-cell uorescent labelling. As we anticipated, the k app values for N,N 0dicyanorhodamine-derived ligands laid between those of TMR (bearing a cationic 3,6-diaminoxanthylium core in its uorescent zwitterionic form) and uorescein (with anionic 3-hydroxy-6-uorone chromophore). Numerous carborhodamine and Si-rhodamine-based Hal-oTag ligands have demonstrated a uorogenic response, i.e. an increase in uorescence intensity upon covalent binding to the HaloTag protein. 3 In the rhodamine series, only uorinated dyes with electron-withdrawing N-substituents (2,2,2-tri-uoroethyl, 3,3-diuoroazetidinyl) or with a lactone-to-lactam modication 3,36 showed a similar behaviour. Accordingly, only the CR3-Halo ligand demonstrated moderate uorogenicity, comparable to the Si-rhodamine HaloTag ligand. 36, 37 The two other cyanorhodamine ligands and Fluorescein-Halo showed no uorogenic response upon binding to HaloTag7 (Fig. S12 †). It has been previously noticed that this response correlates with an increase in emission intensity of uorescent HaloTag ligands bound non-specically to bovine serum albumin (BSA) in buffered solutions upon addition of anionic surfactant sodium dodecyl sulfate (SDS). 36 Conversely, upon addition of cationic cetyltrimethylammonium bromide (CTAB) detergent, the uorescence intensity of uorogenic rhodamine ligands decreases. We hypothesized that this relation would be reverted for negatively charged N,N 0 -dicyanorhodamine derivatives. Indeed, the uorescence of BSA-bound CR2-Halo and CR3-Halo markedly decreased upon addition of SDS to the medium, likely due to unfavourable electrostatic interactions forcing spirolactonization under local environment conditions (Fig. S13 †). However, virtually no response to the presence of SDS or CTAB was observed for CR1-Halo and Fluorescein-Halo. Since zwitterionic rhodamines, rhodols and uoresceins have previously been developed into a series of transmembrane potential sensors, 40 cell-permeant and non-uorogenic N-cyanorhodamines with distinct environmental sensitivity may provide additional options for the synthetic design of similar uorescent reporters for functional imaging.

Biocompatibility and cellular imaging
In several reports describing the cytotoxicity of rhodamine dyes (e.g. Rhodamine 123, 41 Rhodamine 6G 42 ), cytotoxic effects have been attributed to the cationic form accumulating in mitochondria and disrupting the synthesis of adenosine triphosphate (ATP), the primary renewable energy source of the mammalian cell. For this reason, we rst evaluated the effects of N-cyanorhodamine core compounds 5a, 6a on human bone osteosarcoma epithelial (U-2 OS) cell viability and noted the absence of cytotoxicity at concentrations up to 100 mM in the medium over 24 h (Fig. 5A). These dye loadings far exceed the usual #5 mM concentrations employed in cellular imaging; indeed, the toxic concentrations were comparable to that of the DMSO vehicle and did not surpass the tolerated concentration   limits for TMR. The cell morphology and proliferation rates of living U-2 OS cells, monitored by means of holographic imaging cytometry in the presence of 5 mM of 5a, 6a or various CR1-CR3 derivatives in the culture medium, were unaffected over at least 48 h (Fig. 5B-D and S14 †).
Off-target labelling artefacts, most commonly observed as diffuse uorescent staining of mitochondria, lysosomes and/or plasma membrane structures such as the ER, are the main deterrent in the development of selective uorescent labels for living cells. To compare the off-target affinity of N,N 0 -dicyanorhodamines with commonly used TMR-based probes, living U-2 OS cells were treated overnight with identical (5 mM) concentrations of uorophores 6a, TMR and the corresponding Hal-oTag ligands CR1-Halo and TMR-Halo. While neither 6a nor CR1-Halo demonstrated any intracellular staining, TMR-treated samples showed diffuse uorescence of membrane structures including the plasma membrane (Fig. 6A), and TMR-Halo predominantly accumulated in the ER (Fig. 6B), which was conrmed by the successful colocalization with ER-Tracker Blue-White DPX probe (average Pearson's correlation coefficient 0.62 AE 0.05 (N ¼ 9), Fig. 6C and D).
Having veried the absence of both uorophore-directed offtargeting and cytotoxicity for N-cyanorhodamines, and good photostability and fast labelling kinetics for our N-cyanorhodamine HaloTag probes, we nally performed multicolour confocal and STED uorescence microscopy in living mammalian cells in combination with previously established live-cell and STED-compatible uorophores. To this end, several new label combinations permitting up to 4-colour imaging (2Â confocal, 2Â STED) were proposed and evaluated in genetically modied U-2 OS cell lines expressing fusion proteins of suitable cellular structures targeted with a HaloTag. In one example, the HaloTag-fused nuclear pore complex protein Nup96 was labelled in living U-2 OS-NUP96-Halo cells 43 (engineered with the clustered regularly interspaced short palindromic repeats (CRISPR) technique) with the CR1-Halo ligand (5 mM, 6 h), costained for nuclear chromatin with the far-red label SiR-Hoechst, 44 for microtubular cytoskeleton with abberior LIVE 510 tubulin, and for ER with ER-Tracker Blue-White DPX commercial probes at submicromolar concentrations (Fig. 7A). Signicantly improved resolution of individual nucleoporin clusters was achieved with 775 nm STED nanoscopy with little to no diffuse background in the cells (Fig. 7B). The density of SiR-Hoechstlabelled chromatin in the nucleus could be simultaneously evaluated with subdiffraction precision (Fig. 7C). Furthermore, fused HaloTag-vimentin protein in living CRISPR-engineered U-2 OS-Vim-Halo cells 35 was tagged with the CR2-Halo probe (1 mM, 5 h; spectrally identical with a widely utilized carborhodamine uorophore 610CP), 36a,45 together with the abberior LIVE 550 tubulin probe and Hoechst 33342 for co-staining nuclear DNA (Fig. 7D). Both vimentin (Fig. 7E) and tubulin laments (Fig. 7F) were resolved with subdiffraction resolution and without cross-talk between the two red-uorescent labels. On the contrary, while the CR3-Halo ligand provided good quality confocal images of vimentin in living U-2 OS-Vim-Halo cells and was STED-compatible ( Fig. S15 †), it was impossible to record continuous signal from individual laments under STED conditions, since the majority of CR3 uorophore population remained in the colourless spirolactone form under the physiological conditions of the live cell sample.
For uorescence imaging with the SNAP-tag ligand CR1-BG, U-2 OS cells were transiently transfected to achieve overexpression of a SNAP-tag fusion with the promyelocytic leukemia protein (PML), 46a which, aer a series of posttranslational modications and oligomerization, forms distinct nuclear subcompartments (nuclear bodies up to 1 mm in diameter). 46b Upon labelling these with CR1-BG (5 mM, 5 h) and co-staining with GeR-tubulin, 36b Mito-Tracker Green FM (for mitochondria) and Hoechst 33342 (for DNA; Fig. 8A), the hollow-spherical structure of PML-nuclear bodies was resolved with 775 nm STED nanoscopy (Fig. 8B). Subdiffraction resolution of individual microtubules labelled with GeR-tubulin was simultaneously achieved (Fig. 8C).
In all of the multicolour imaging examples above, very little to no background or appreciable off-targeting artefacts were observed with N,N 0 -dicyanorhodamine uorescent ligands. We consider this selectivity remarkable given the high uorophore concentrations (1-5 mM) in the medium, relatively long incubation times and the absence of innate uorogenic behaviour of the new HaloTag labels. These observations support our alternative approach to high-contrast live-cell labelling, employing the negatively charged cell-permeant xanthene uorophores instead of uorogenic rhodamine amides with decreased content of the uorescent zwitterionic form in the equilibrium. 3b,29

Conclusions
Rhodamine uorophores, bearing a net negative charge at neutral pH values due to the presence of anionic groups (sulfonate 47a or carboxylate 47b ), have been previously recognized as cell-impermeant and utilized solely in the development of specic probes for extracellular targets or for labelling xed cells. While signicant efforts have focused on the development of medium-polarity-sensitive uorogenic probes, 3,29,36 the unselective binding of these probes is simply masked by their reduced uorescence in non-polar environments such as lipidrich membrane structures. The main drawback of this approach is that the uorescence of the zwitterionic form of rhodamines is maintained or even enhanced in acidic compartments, and that accumulation of lipophilic rhodamines in their non-uorescent spirolactone form may at least partially be responsible for their cytotoxicity. These effects become the main limiting factor determining the labelling conditions such as probe concentration and incubation time, and may easily become prohibitive for uorescent ligands with lower binding affinities.
On the other hand, the live-cell application of anionic uorescein-based probes is free from these drawbacks but requires chemical protection (usually in the form of acetate ester or acetoxymethyl ether) and has to rely on enzymatic cleavage to recover the uorescent label. In addition, all reported uoresceins and rhodols show hypsochromic absorption and emissions shis and poor photostability as compared to the corresponding rhodamine dyes.
In our work, we have proposed a class of N-cyanorhodamine uorophores, which maintain live cell permeability despite being negatively charged within the physiological pH range. In particular, N,N 0 -dicyanorhodamine dyes demonstrated absence of toxicity, high photostability and sufficient spectral diversity in the orange-to far-red emission range permitting their use in long-term labelling and multicolour super-resolution microscopy. We have performed the initial evaluation of N-cyanorhodamine label combinations for three-and four-colour livecell imaging, and anticipate the development of photoactivatable and/or enzymatically activatable uorogenic probes designed around these original core structures. Conflicts of interest S. W. H. owns shares of Abberior GmbH and Abberior Instruments GmbH whose dyes and STED microscope, respectively, have been used in this study. MR) for a HaloTag7 protein sample, Dr Elisa D'Este and the Optical Microscopy facility (MPI MR) for access to an Abberior expert line STED microscope, and the Department of Chemical Biology (Prof. Kai Johnsson, MPI MR) for access to a Quantaurus-QY instrument. We appreciate the helpful discussions with PD Dr Jochen Reinstein (MPI MR).