A diversity-oriented rhodamine library for wide-spectrum bactericidal agents with low inducible resistance against resistant pathogens

Antimicrobial resistance is a public health emergency and warrants coordinated global efforts. Challenge is that no alternative molecular platform has been identified for discovery of abundant antimicrobial hit compounds. Xanthene libraries have been screened for bioactive compounds. However, the potentially accessible chemistry space of xanthene dyes is limited by the existing xanthene synthesis. Herein we report a mild one-step synthesis, which permits late-stage introduction of a xanthene moiety onto i.e. natural products, pharmaceuticals, and bioactive compounds and construction of a focused library of rhodamine dyes exhibiting facile functional, topographical and stereochemical diversity. In vitro screening yields 37 analogs with mid-to-high bactericidal activity against WHO priority drug-resistant pathogens. These findings suggest that synthetic dye libraries exhibiting high structural diversity is a feasible chemical space combating antibacterial resistance, to complement the natural sources. Preparation of xanthene-containing compounds has been limited due to structural bias existing methods pose. Here, the authors developed a mild, diversity-oriented method for rhodamines synthesis, leading to the finding of compounds with antibacterial potency against a variety of bacterial species.

A ntibiotics have revolutionized medical practices, have dramatically increased the average life expectancy of humans and are among the greatest medical breakthroughs of the last century. However, antibiotics have only a limited period of clinical-utility before offset by the inevitable emergence of resistance 1 . For example, it took only a few years before resistance toward many antibiotics was discovered, such as penicillin, gentamicin, and ceftazidime 2 . Vancomycin lasted longer before resistance emerged, partly because it was typically used as the last resort. Nevertheless, vancomycin intermediate/ resistant pathogens are now widespread 3 . New antibiotics are desperately sought after to fight bacterial resistance 4 . Ironically, large pharmaceutical companies have dramatically shrunk or cut their antibiotic programs. The reasons for this situation are manifold. The golden mine for antibiotics, the secondary metabolites of soil actinomycetes, has run dry after decades of exploitation 5,6 . An alternative chemical space for abundant antibiotic hit compounds is yet to be identified 7 . Third, antibiotic development has a low expected profit return. As a result, the antibiotic pipeline has been running at an alarmingly slow pace in the past few decades 8 . In particular, among these recently approved few antibiotics, most are actually chemically modified derivatives of the existing classes of drugs, most of which are of natural origin [9][10][11][12] . The resistant strains may readily mutate to resist these analogs if their existing resistance mechanisms do not already exhibit partial cross-effectiveness [13][14][15][16] . The genes encoding resistance to the natural products that are present within the original organisms can also be horizontally transferred to pathogenic microbes enabling resistance to emerge. To prevent the post-antibiotic era becoming reality, alternative chemical space capable of producing abundant antibiotic scaffolds are warranted.
Previously untapped natural sources, i.e. uncultured soil microbes 17 , ocean bacteria 18 , unexpressed metabolites coded by silent operons 19 , and human commensals 20 , have recently been explored and yielded antibiotic compounds. At the same time, synthetic libraries should not be overlooked for antibiotic discovery 4 . Salvarsan for syphilis and prontosil for streptococcus were two classic antimicrobial drugs discovered from the synthetic dye libraries in the early 20th century, and other notable examples are quinolones and the recent addition of oxazolidinones 21 . Despite the fact that target-based antibiotic discovery since the 1990s has been largely unsuccessful 7,22 , synthetic library of unique chemical space exhibiting high structural complexity and diversity offers opportunities for formulating new paradigms for antibiotic development [23][24][25] . Recently, the diversity-oriented fluorescence library approach (DOFLA) has been proven to be a robust method for discovery of bioactive molecules via phenotypic screening 26 . Xanthene dyes, classic small-molecule fluorophores including fluorescein, rhodamine and rhodol, has recently attracted attentions for DOFLA. For example, Chang et al. and other research groups constructed a number of focused libraries based on various fluorescent scaffolds 27,28 , and further discovered useful imaging agents or fluorescent probes for smallmolecule targets, nucleic acids, proteins, lipid droplets, stem-cells and live animals [29][30][31] . Burgess et al. constructed a library of mitochondria-targeting rhodamine analogues exhibiting potent anti-proliferative effects toward tumor cells 32,33 . High structural diversity of dye libraries, the key to its success in the fluorescence library approach, is however the biggest obstacle to overcome. The ubiquitous structure feature of a dye is a two-dimensional conjugated system, to which some simple chemical groups may be attached. Therefore, dyes are typically not endowed with complex three-dimensional features and not ideal for discovery of bioactive compounds via phenotypic screening. Also, the conditions for dye synthesis are usually harsh and installation of chemical groups exhibiting exotic functional, topological and stereochemical features are difficult [34][35][36] . Diversity-oriented rhodamine library could not be constructed with the conventional method due to, extensive side-reactions, tedious purification, low yields and most important of all limited scope of compatible substrates. Take the xanthene dye library by Ahn et al. 37 as an example. It is a chemical library of 300+ compounds. However, all follow the general feature, i.e. a xanthene core with a small aromatic group at its C-9. Therefore, the library is not high in terms of diversity. A diversity index of only 3.228 was calculated for this library of rhodamine dyes with Extended-Connectivity FingerPrint (ECFP) [38][39][40] , which is a convenient estimate of the structural diversity of a chemical library (vide infra).
We herein report the construction of a diversity-oriented rhodamine library and discovery of potent and wide-spectrum bactericidal agents without inducible resistance. Phenotypic screening of the library yields two promising hit compounds, and RD22 and RD53 were found to be particularly noteworthy through further studies. They inhibit the growth of a widespectrum of pathogens, with a potent MIC of 0.5-2 μg mL −1 against the Gram-positive strains, and a MIC of 2-16 μg mL −1 against the Gram-negative strains. They are potent bactericidal agents and pathogens do not readily acquire resistance by genomic mutations toward RD22 and RD53. They exhibit low haemolytic activity with a Lysis20 over 100 μg mL −1 . The aforementioned properties of RD22 and RD53, along with other less potent analogs suggest that the rhodamine library is a viable source of antibacterial hit compounds.

Results
Construction of a library with 70 rhodamine analogs. A number of preparative methods are available for xanthene dyes. The earliest synthesis was by von Baeyer 34 in 1871 through the condensation of phthalic anhydride with phenol at high temperature, with Lewis acid or strong proton acid as the catalyst. To accommodate the growing demands for functional xanthene dyes, an alternative mild preparation was introduced by Tsien et al. in 1989, involving nucleophilic addition of a phenyl Grignard (or lithium) reagent with a xanthenone at low temperature 36 . This has enabled the preparation of the library of Ahn et al. 37 . Though robust, diversity of the xanthene dyes from this method is still limited by the availability and structural versatility of Grignard reagents. The limitation remains unsolved with the report of twostep Grignard/demethylation method by Yang et al. 41 . To address those difficulties with the classic xanthene synthesis, we have developed a mild one-step and high-yielding xanthene synthesis (Fig. 1a) via nucleophilic condensation of a dilithium reagent (2) with readily available esters, anhydrides, or even some amides (Supplementary Figure 1). This method is versatile and allows the preparation of xanthene dyes with exotic substituents at C-9, e.g. phenyls, aryls, alkyls, alkenyls and acyls (vide infra) (Fig. 1). Shortly after we filed a Chinese invention patent 42 of this method in March 2017, Grimm et al. 43 and Fischer and Sparr 44 also independently developed and reported this method in recent literature.
The capability of this method in preparation of rhodamines with substituted phenyl group at C-9 was first examined (Fig. 1b). As a proof-of-concept experiment, the dilithium reagent (1) was reacted with an alkyl benzoate, i.e. methyl benzoate (S1), to furnish RD1 in an 83% yield. Rhodamines with a methyl (RD2), hydroxylmethyl (RD3), formal (RD5) or a carboxyl group (RD6) installed at the ortho-position of the bottom phenyl ring was also smoothly prepared in excellent yields by reacting 1 with 2methyl-benzoate (S2), phthalide (S3), 3-methoxyphthalide (S4) or phthalic anhydride (S6). RD4, rather than RD5, could be obtained in an 84% yield if the reaction was quenched with saturated NH 4 Cl, rather than dilute HCl solution.
The one-step high-yielding synthesis of RD3 and RD5 shown here is superior to their literature preparations 45 Fig. 1 The rhodamine library via nucleophilic condensation of a dilithium reagent 1 with various substrates. a The general synthetic pathway. Structures of rhodamine dyes with b phenyl groups at C-9 (RD1-RD24); c polyaromatic or heteroaryl groups at C-9 (RD25-RD41); d alkenyl groups at C-9 (RD42-RD47); e alkyl groups at C-9 (RD48-RD59); f bioactive molecules at C-9 (RD60-RD69) and g oxazine (RD70  [1,3]dioxin-4-one (S10), benzo[d] [1,3]oxathiin-4-one (S11) or benzo[d] [1,3]oxazin-4-one (S13), methyl-2-nitrobenzoate (S15) respectively. Synthesis of rhodamine dyes with bulky groups at the ortho position of the bottom phenyl ring is not trivial. This method was powerful in this regard. Reactions between the dilithium reagent 1 and S16, S17, S18, S19 and S20 (Supplementary Figure 1) have produced a series of rhodamine dyes with such bulky groups as phenyls in RD16, RD18, RD19 and RD20, and diphenylphosphoryl in RD17, in a yield ranging from 65 to 70%. The carbonyl of S18 is involved in electronic push-pull system and therefore is not very reactive toward nucleophilic attack. Still, S18 was a viable substrate to react with 1 furnishing a rhodamine (RD18) with an electron-rich phenyl ring at the ortho-position, in a 70% yield. RD20 is a rhodamine modified with four freely rotatable phenyl groups and structurally analogous to an AIE-luminogen 47 , e.g. tetraphenylethene. Its potentials in AIE-based studies warrant further investigations. Rhodamines with phenyl diazo groups (RD21 and RD22) were also successfully prepared in yields over 76%. Unusual structures like RD23 or RD24, expected to exhibit high molar absorptivity and fluorescence brightness were successfully obtained, in good yields of 56% and 49%, respectively. Onto the C-9 position of the rhodamine core can be installed aryl groups other than phenyl, e.g. polyaromatic moities, heterocyclic aromatic moieties and the exotic sandwich/halfsandwich complexes (Fig. 1c). Rhodamines with naphthalene (RD25, RD26) or anthracene (RD27, RD28) were prepared from reactions of 1 with the corresponding ester or anhydride in good isolated yields (typically over 75%). Esters of pyridine, quinoline, furan, thiophene, pyridazine, pyrimidine, pyrazine, imidazole, oxazole, and thiazole (Supplementary Figure 1) could also react with the dilithium reagent 1 to prepare the corresponding rhodamine dyes (RD29-RD38). With the methyl ester of ferrocene, we prepared RD39, in which the ferrocene moiety is directly bonded to the C-9 position of the rhodamine core. We further showed that the half-sandwich complex could be directly attached to the rhodamine core (RD40). The potential applications of RD40 in sensing, imaging and catalysis are attractive. Substrates bearing two ester moieties could be used to template multiple rhodamine cores, to give RD41.
Alkenyl groups can also be installed onto the C-9 position of a rhodamine core (Fig. 1d). The double bond in such rhodamine dyes exhibits facile chemistry and may be useful in site-specific labeling 48 , chemosensing of nucleophiles 49 and reactive oxygen species 50 and optical switching 51 . With the ester of a coumarin, a rhodamine-coumarin dye pair (RD42) was synthesized as one example of this group of rhodamine dyes. Reactions between the dilithium reagent 1 and a range of substrates like maleic anhydride, cinnamates, or coumarins (Supplementary Figure 1) were all capable of furnishing such alkenyl substituted rhodamine dyes, i.e. RD43-RD47. We note that the double bond of RD43 adopts cis-conformation, as suggested by a coupling constant of 5.4 Hz between the two alkenyl H's. The double bonds in RD44-47 are all in trans-conformation, regardless the cis-or transconformation of the double bonds in the corresponding starting material (S44-S47, Supplementary Figure 1).
Alkyl esters, lactones and anhydrides are all feasible substrates for synthesizing rhodamine dyes and diverse range of complex alkyl groups are also conveniently installed to the C-9 of the rhodamine core (RD48-RD59) (Fig. 1e).
Till this point, we have demonstrated that this rhodamine synthesis is convenient, powerful and reliable. The range of substrates compatible for this method is exceedingly broad, including esters, anhydride, lactone, aryl lactam, and potentially more. To further promote the scope of this rhodamine synthesis, we are particularly interested in the possibility of using ester-/ lactone-/anhydride-containing bioactive molecules, e.g. pharmaceuticals and pesticides (S60-S69, Supplementary Figure 1), in rhodamine synthesis. The structures of such compounds are typically complex and there were no attempts to use them in dye synthesis. Though probably unexpected to the field, many of such compounds, e.g. telmisartan methyl ester (S60), noscapine (S61), adapalene (S62), a precursor of candesartan (S63), etomidate (S64), the methyl ester of flumazenil (S65), levofloxacin methyl ester (S66), vinpocetine (S67), methoxsalen (S68), ethyl chrysanthemate (S69) were indeed viable substrates for this rhodamine synthesis (Fig. 1f).
This method is also suitable for preparation of an oxazine dye RD70 (Fig. 1g) from isoamyl nitrite, which differs from its rhodamine analog in that it bears a nitrogen atom rather than a methine carbon.
Diversity index. The extended-connectivity fingerprint (ECFP) method was employed to assay the diversity of our library because of its wide-acceptance in drug discovery and has been integrated into various commercial softwares. ECFPs encode each atom and its molecular environment within a circle with a diameter of varying radius of chemical bonds. Depending on the chosen bond radius, different numbers of structural features having different sizes are produced. In this study, we introduce the definition of the total number of fingerprint features as a quantitative measure of the structural diversity of a library. We calculated the specified fingerprint for each molecule using ECFP_6 and counted the total number of unique fingerprint features collected over all the molecules. Then the Diversity Number Finger Print Features (DNFPF value) was defined as the total number of fingerprint features divided by the number of molecules. A DNFPF value of 18.875 was calculated for the obtained library of rhodamine dyes (RD1-RD69), while a DNFPF value of 3.228 was found for the Ahn's library 37 of rhodamine dyes (Supplementary Table 6). This verified that the synthetic method presented herein is feasible for construction of a rhodamine library with much enhanced structural diversity.  Table 1). Other solvents were not tested because we are typically concerned with the potentials of this library in bio-oriented applications. The maximal absorption wavelengths of rhodamine dyes with alkyl groups (RD48-RD59) at C-9 typically absorb at shorter wavelength region of 545-560 nm and ones substituted with heteroaryl groups (RD25-RD41) absorb at longer wavelength region of 551-585 nm, while others absorb in the range of 560-570 nm. Most dyes exhibit a small Stokes shift of ca. 15-30 nm, which is typical of a rhodamine, some have displayed a Stokes shift larger than 30 nm, e.g. 34 nm for RD31, 41 nm for RD33, 37 nm for RD35, and 34 nm for RD41, 32 nm for RD47, 39 nm for RD53, 35 nm for RD56, 42 nm for RD59. The fluorescence brightness of many dyes of this library is comparable or higher than that of the rhodamine B (RD6). They are useful for cell imaging- Antibacterial studies of the rhodamine library. The antimicrobial activity of (RD1-RD70) were preliminarily evaluated against two drug-resistant bacteria, which were a strain of Grampositive methicillin-resistant Staphylococcus aureus (MRSA, ATCC43300) and a strain of Gram-negative Acinetobacter baumannii (ATCC19606). The minimum inhibitory concentrations (MIC) of each compound were measured using the standard CLSI broth protocol (Fig. 3, Supplementary Table 2). Typically, antimicrobial potency of a compound is considered mid-to-high if its MIC is 8 μg mL −1 or lower. Out of the total 70 analogs, 37 were found to exhibit a MIC of 8 μg mL −1 or lower against MRSA, and 5 aganist A. baumannii. This is a result much to our delight as well as to our surprise considering the relentless yet largely futile efforts that the community has dedicated to the quest for a molecular platform for discovery of abundant compounds with antimicrobial activity. Further analysis revealed that these bioactive analogs are quite scattered in terms of the nature of substituent at the C-9 position of the rhodamine core. Among these bioactive analogs, 14 analogs bear a phenyl group at C-9 position, 12 a heteroaromatic moiety, 6 an alkenyl and 5 an alkyl group. Presumably, the rhodamine core is the pharmacophore and the substituent at the C-9 playing a subsidiary, yet critical role. These results strongly suggested that this diverse-oriented rhodamine library is a viable platform for discovery of potent antimicrobial agents.
Though all these 37 + 5 compounds with mid-to-high antibacterial activity worth further pursuing, we prioritized the ones with a higher activity for in-depth study of their antibiotic spectrum (Table 1) showed potential broad-spectrum Gram-positive antibiotic activity with a low MIC of 0.5-2 μg mL −1 , while RD1/15/31/45 were much less active toward E. faecalis compared to their activity against S. aureus. RD22/44/53, but not RD12/45, also exhibited antibiotic activity toward Gram-negative bacteria other than ATCC19606, with a relatively higher MIC of 8-32 μg mL −1 . From this study, RD22/53 stood out as the two most attractive hits. First, their MIC's against MRSA is 1 μg mL −1 , the lowest among this shortlist, and comparable or even lower than those of vancomycin and linezolid (Table 1), the last resorts for MRSA infections. Second, they were also equally potent toward both vancomycin-sensitive and vancomycin-resistant E. faecalis with a MIC of 2 μg mL −1 . Third, they were also active toward various clinically significant Gram-negative bacteria with MIC of 4-16 μg mL −1 . These encouraging preliminary results promoted us to explore the antibiotic profile of RD22/53 in a greater depth.
The antibiotic spectrum of RD22 and RD53 were further checked against a panel of standard reference and clinical-isolated Gram-positive and Gram-negative strains, which encompass most of the 12 deadliest drug-resistant bacteria 52 classified by the World Health Organisation (WHO) ( Table 2 The clinical significance of the first-line bacteriostatic antibiotics, e.g., vancomycin, linezolid, et al., warrants no further elaboration. However, in some serious clinical situations, such as endocarditis, meningitis, osteomyelitis, neutropenia, bactericidal action is necessary because of the reduced rates of metabolism and cell division, the poor immunologic competence or the poor drug penetration 53 . The bactericidal activity of RD22 and RD53 were compared to multiple first-line antibiotics using the time-kill assay (Fig. 4). Methicillin-resistant S. aureus, Vancomycinresistant E. faecalis, and polymyxin E-resistant A. baumannii were chosen as the tested Gram-positive and Gram-negative strains, respectively. MRSA (ATCC43300) were grown to early exponential phase and challenged with 10× MIC of vancomycin, linezolid, tigecycline, respectively (Fig. 4a). In the first 4 h, bacteria concentration remained steady in the presence of vancomycin and linezolid and then gradually dropped by twothree orders of magnitude within 24 h. Tigecycline exhibited a higher degree of bactericidal activity than vancomycin and linezolid. In comparison, RD22 (Fig. 4b) and RD53 (Fig. 4c) were more potent bactericidal agents than vancomycin, linezolid or tigecycline. Bacteria challenged by as low as 2.5× MIC lowered the bacteria concentration by four order of magnitude of bacteria within as short as 2 h. RD22 and RD53 also exhibits bactericidal activities to VRE (ATCC51299) (Fig. 4e, f). They lowered the concentration of VRE by over 4 orders of magnitude within 2 h by (5-10) × MIC. In comparison, linezolid, tigecycline did not exhibit strong bactericidal activity and bacterial concentration dropped by less than 2 orders of magnitude within 24 h. Daptomycin was able to reduce the cell concentration by over 4 orders of magnitude within 8 h (Fig. 4d). Polymyxin E-resistant A. baumannii (MIC = 256 μg mL −1 , stored in the lab) was obtained by serial passages of cells at sub-MIC concentration of polymyxin E during resistance acquisition of pathogens. RD53 at 10 × MIC was able to reduce the cell concentration by over 4 orders of magnitude within 6 h (Fig. 4i) while at the same dose of tigecycline and RD22 reduced the cell concentration by around 3 orders of magnitude (Fig. 4g, h). And RD53 at 8 × MIC could also reduce the cell concentration of H. pylori by more than 4 orders of magnitude within 2 h (Supplementary Figure 132). So, the time-kill assay proved that RD22 and RD53 are potent bactericidal agents. They can potentially complement bacteriostatic antibiotics in serious bacterial infections to hosts, whose immune response is compromised.
Resistance studies of RD22 and RD53. Ease of resistance acquisition toward RD22 and RD53 was evaluated using MRSA, VRE and A. baumannii as tested bacteria (Fig. 5). Serial passages of MRSA in the presence of sub-MIC levels of levofloxacin over a   period of 20 days raised the MIC by 128 folds, i.e. rendering the strain highly resistant to levofloxacin. In comparison, MRSA did not develop resistance toward vancomycin, exactly the reason why vancomycin serves as the last-line antibiotic against MRSA. It is very interesting to note that MRSA also did not acquire a resistance by genomic mutations toward RD22 and RD53 as well (Fig. 5a). In a separate experiment, VRE developed resistance toward linezolid, ciprofloxacin and daptomycin by serial passages in 20 days, with a raise of MIC's by 32-128 folds while no obvious resistance was observed with RD22 and RD53 by VRE (Fig. 5b). The Gram-negative A. baumannii readily acquired resistance by genomic mutations toward polymyxin E and ciprofloxacin with a 256 enhancement of MIC in as short as 7 days. Bacteria did develop resistance toward meropenem, tigecycline, RD22 and RD53 as well, but to a much reduced degree, i.e. 64-, 32-, 8-, and 32-fold enhancement of MIC within 20 days, respectively (Fig. 5c). It has become evident that either Gram-positive or Gram-negative pathogens were found not to readily acquire resistance by genomic mutations toward RD22 and RD53. Analogously, E. coli, E. faecalis and S. aureus were found not to acquire resistance toward RD53 (Supplementary Figure 133). Development of resistance toward an antibiotic by bacteria is an evolutionary inevitability and thereby imposes a limit on the lifespan of any given antibiotic. And resistance was involved in clinical failure of some antibiotic administration regimes. This is especially evident because the antibiotic levels at some points during treatment are in the sub-inhibitory range 54 . Therefore, the low tendency of resistance acquisition toward RD22 and RD53 by various pathogens greatly enhances their potentials for further drug development.
Probing mechanism of action. The mechanism of action was investigated, with RD53 as an example. First, the lack of noticeable morphological variation of MRSA and A. baumannii, as indicated by the SEM images (Fig. 6) with and without RD53, suggested that RD53 did not destruct either the cell wall or membrane. This observation was further corroborated by the fact that RD53 exhibits low haemolytic activity with a Lysis20 value of above 100 μg mL −1 . Second, a resistant strain of A. baumannii to RD53 (RD53-Ab-20th) was produced by serial passages of ATCC19606 exposed to sub-MIC concentration of RD53 over a period of 20 days, with the MIC of RD53 raised from 4 μg mL −1 to 128 μg mL −1 . Yet, the phenotypic profile of RD53-Ab-20th with respect to susceptibility to other antibacterial drugs, e.g., tigecycline (TGC), ciprofloxacin (CIP), polymyxin E (PmE), levofloxacin (LEV), and ampicillin-sulbactam (SAM), remains unchanged compared to ATCC19606, suggesting the lack of cross-resistance between RD53 and these antibiotics (Supplementary Table 9). Third, selective pressure on antibacterial targets could also be measured by genomic profile of mutational resistant bacteria. So, the genomic analysis was carried out for clues on the potential mechanism of action of RD53. Comparative analysis revealed the existence of eight nonsynonymous single nucleotide variants (SNV) and one nonframeshift deletion (Table 3), which belong to resistance/tolerance-associated genes encoding efflux pumps, metabolism, DNA replication or transcriptional regulation related proteins 55,56 . It is interesting to find the mutation in rsmJ which encodes 16S rRNA (guanine1516 -N2)methyltransferase. Notably, the mutation in 16S rRNA methyltransferase has been shown to cause kasugamycin resistance in E. coli 57,58 . These clues together have formulated a solid framework for further investigation of the mechanism of action of RD53.
Structure-activity relationship investigation. The structureactivity relationship of these rhodamine-type antibiotic compounds was examined. Out of the 70 members of this rhodamine library, 37 compounds exhibit antibiotic activity of varying potency (Fig. 3). This fact alone strongly suggested that their common structural motif is the pharmacologically active motif, which is the rhodamine core. To verify this hypothesis, we have prepared a few closely related analogs (C1-2, and RD71-74, Fig. 7). First, C1 and C2 differ from the RD53 in that C1 misses the oxygen atom of the rhodamine core compared to RD53, and C2 has one more aniline moiety removed compared to C1. Both C1 and C2 exhibited no appreciable antibiotic activity with their MIC's higher than 64 μg mL −1 toward both ATCC43300 and ATCC19606 (Fig. 7a). Second, the peripheral group of the rhodamine core or the adamantine moiety could be modified without inhibiting the antibiotic activity as long as the rhodamine core remains intact, as showcased with RD71-74 (Fig. 7b, c). The aforementioned hypothesis was further supported by the following experimental results that none of the starting materials (S1-S69) for preparations of these rhodamine analogs (RD1-RD69) display antibiotic activity (MIC > 64 μg mL −1 ) (Supplementary Table 3). This reiterates the necessity of the rhodamine core for antibiotic activity. Additional support to this hypothesis comes from the ten compounds listed in Fig. 7d. These compounds predominantly adopt the ring-closed form and therefore they do not actually have the rhodamine motif. And indeed, none of them have been found to exhibit antibiotic activity during the phenotypic screening. The discovery of the above structureactivity relationship is important as it provides a general guideline for further structural modification, if warranted in the future drug-development endeavors.

Discussion
Diversity-oriented synthesis of xanthene library attracted attention for discovery of original bioactive compounds through phenotypic screening. This has promoted search for a mild xanthene synthesis allowing facile installation of functional, topographical and stereochemical diversity. We have devised a mild, one-step, high-yielding and readily scalable xanthene synthesis, by reacting a rationally constructed dilithium species a b c d General procedures B was illustrated as the general synthetic route for the preparation of rhodamine RD1-72 via nucleophilic condensation of a dilithium reagent 1 with various substrates S1-S70. The UV-Vis absorption and fluorescence emission spectra in pH = 7.4 PBS with 1% DMSO was shown in Supplementary  Figures 2-69 Fig. 7 Structure-activity relationship investigation of the rhodamine library. a RD53 analogs without the rhodamine core; b RD53 analogs with the adamantine moiety modified; c RD53 analogs with the diethylamino group modified. d Rhodamine dyes mainly adopted the ring-close forms exhibit no antibacterial activity against ATCC43300 or ATCC19606, with MIC > 64 μg mL −1 4 × C n of RD22 or RD53. This serial passaging was repeated daily for 20 days. Experiments were performed with biological replicates. Results of absolute fold change of MIC of 30 colonies purified from the 5th and 10th passage were shown in Supplementary Table 7. Results of spontaneous frequency of resistance study were shown in Supplementary Table 8.
Scanning electron microscope (SEM) analysis for MRSA and A. baumannii. The cells were grown in the MHB broth medium to around 10 7 c.f.u. per mL and collected before drug treatment and 4 h after treatment of 8 × MIC RD53 (8 μg mL −1 for MRSA and 32 μg mL −1 for A. baumannii). The cells were fixed in 2.5% glutaraldehyde (prepared in PBS buffer) for 2 h, and washed with PBS buffer twice after fixation. Then the cells were dehydrated in a series of ethanol solutions which are 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100% ethanol, sequentially for 15 min each, and finally suspended in pure tert-butanol for 15 min. The cells were criticalpoint dried and coated with gold following by scanning electron microscope analysis (FEI Nova NanoSEM 230). The SEM was operated in the high vacuum at an accelerating voltage of 7.5 kV.
Haemolytic activity: red blood cell (RBC) lysis assay (Lysis20). Lysis20 values of RD22 and RD53 were measured as the concentration of compound that lyses 20% or less of sheep red blood cells (RBC). The therapeutic index (Lysis20/MIC) was over 100 when comparing the MIC against MRSA to the observed Lysis20. RBC lysis assays were performed on mechanically defibrinated sheep blood. 1.5 mL of blood was placed into a microcentrifuge tube and centrifuged at 8609 × g for 10 min. The supernatant was removed and the cells were resuspended with 1 mL of 5% glucose solution. The suspension was centrifuged as previously, the supernatant was removed, and cells were resuspended two more times. The final cell suspension was diluted twentyfold with 5% glucose solution. The twentyfold suspension dilution was then aliquoted into microcentrifuge tubes containing compound serially diluted in 5% glucose solution. Triton X (1% by volume) served as a positive control (100% lysis marker) and sterile 5% glucose solution served as a negative control (0% lysis marker). Samples were then placed in an incubator at 37°C and shaken at 200 r.p.m. After 1 h, the samples were centrifuged at 8609 × g for 10 min. The absorbance of the supernatant was measured with a UV spectrometer at a 330 nm wavelength.
Mammalian cytotoxicity of RD53. Cytotoxicity was determined on normal human kidney 2 cells HK-2 cells human renal proximal tubular cells (Cell Bank of Chinese Academy of Science, cat. no. SCSP-511), HUVEC Primary Umbilical Vein Endothelial Cells (American type culture collection, cat. no. PCS-100-013) and HeLa human cervical carcinoma cells (Cell Bank of Chinese Academy of Science, cat. no. TCHu 187). Briefly, HK-2 and HeLa cells were cultured in RPMI-1640 medium supplemented with FBS, and HUVEC were cultured in DMEM, high glucose medium supplemented with FBS. When cells reached~80% confluence, they were harvested and resuspended in the growth medium to~2.0 × 10 5 cells mL −1 . 100 µL cells were inoculated into 96 well plate and incubated at 37°C with 5% CO 2 for 4.5 h. Then, 0.25 µL of twofold serial compound dilutions in DMSO to 99.75 µL of media were added to the plate and incubated at 37°C with 5% CO 2 for 43 h. For cell viability test, 20 µL 0.5% MTT solution were added to the plate and the supernatant was discarded after 4 h incubation. Then 150 μL DMSO were added to dissolve formazan and the A490nm (OD490) was measured. 50% cell cytotoxicity (CC 50 ) was analyzed in GraphPad Prism. Results are shown in Supplementary Figure 134.
Whole-genome sequencing. A single colony of various bacteria was grown until the OD 600 value reaches 0.3. The Genomic DNA was extracted and subjected to paired-ends Illumina sequencing. The whole genome sequencing of RD53-Ab-20th and ATCC19606 was acquired on an Illumina HiSeq 2000 system (Illumina Inc; San Diego, USA). The manufacturer's protocol (NEBNext® Ultra™ DNA Library Prep Kit for Illumina®) was followed to construct the Next Generation Sequencing Library Preparations. The extracted genomic DNA (1 μg) was sonicated to random fragments containing 500 bp or less. The End Prep Enzyme Mix was employed to treat the resulting fragments for end-repairing, 5′-phosphorylation and dA-tailing in one step. T-A ligation was carried out to add adaptors to both ends. Adaptorligated DNA was then sized-selected with AxyPrep Mag PCR Clean-up to recover the fragments of ca. 410 bp, the approximate insert size of which is 350 bp. The resulting samples were subsequently amplified by PCR for 8 cycles. The PCR products were treated with the AxyPrep Mag PCR Clean-up kit, validated with Agilent 2100 Bioanalyzer, and quantified by Qubit2.0 Fluorometer. Then libraries with different indices were multiplexed and loaded onto an Illumina HiSeq instrument. Sequencing was performed using a 2x150 paired-end (PE) configuration. The image analysis and base calling were carried out with the HiSeq Control Software (HCS) + OLB + GAPipeline-1.6 (Illumina) on the HiSeq instrument. The sequences of adaptors, PCR primers, content of N bases more than 10% and bases with a quality lower than 20 were removed using Cutadapt (V1.9.1). The clean data was mapped to the reference genome with BWA (V0.7.12) and duplicate-removed by Picard (V1.119). The Unified Genotyper calls single-nucleotide variant (SNV), insertions/deletions (InDel) with the GATK (V3.4.6) software. Annotation for SNV/InDel was performed with Annovar (V21 Feb 2013). Pindel and CNVnator were employed to analyze genomic structure variation.
Data availability