N-Heterocyclic carbene-based C-centered Au(I)-Ag(I) clusters with intense phosphorescence and organelle-selective translocation in cells

Photoluminescent gold clusters are functionally variable chemical modules by ligand design. Chemical modification of protective ligands and introduction of different metals into the gold clusters lead to discover unique chemical and physical properties based on their significantly perturbed electronic structures. Here we report the synthesis of carbon-centered Au(I)-Ag(I) clusters with high phosphorescence quantum yields using N-heterocyclic carbene ligands. Specifically, a heterometallic cluster [(C)(AuI-L)6AgI2]4+, where L denotes benzimidazolylidene-based carbene ligands featuring N-pyridyl substituents, shows a significantly high phosphorescence quantum yield (Φ = 0.88). Theoretical calculations suggest that the carbene ligands accelerate the radiative decay by affecting the spin-orbit coupling, and the benzimidazolylidene ligands further suppress the non-radiative pathway. Furthermore, these clusters with carbene ligands are taken up into cells, emit phosphorescence and translocate to a particular organelle. Such well-defined, highly phosphorescent C-centered Au(I)-Ag(I) clusters will enable ligand-specific, organelle-selective phosphorescence imaging and dynamic analysis of molecular distribution and translocation pathways in cells.

S ub-nanoscale gold clusters with atomic precision are promising miniature nanoscale materials. Gold clusters often exhibit photoluminescence properties such as phosphorescence, in addition to the unique molecular structures and aurophilicity 1,2 . To date, several excellent protocols have been developed to improve the photoluminescence performance of gold clusters. Examples include alloying by metal kernels, supramolecular networking by self-assembly of clusters, surface hardening by additives, chemical modification by capping ligands with electron-donating/withdrawing groups, and so on [3][4][5][6][7][8][9][10] . The structure, electronic state, and reactivity of the gold cluster moiety can be greatly affected by the protective ligands and different metals that additionally bind to the gold atoms.
The octahedral hexagold(I) cluster with a hyper-coordinated carbon center (CAu I 6 ), first developed by Schmidbaur et al. using phosphine ligands, is one of the most classical models of Au I clusters 11,12 . These clusters emit bright yellowish-green light in the solid-state, but not in solution. Wang et al. reported a heteronuclear metal cluster capable of emitting light even in solution by the pyridyl-phosphine bidentate ligand 13,14 . This method has made it possible to construct a series of heterometallic Au I clusters that exhibit strong red phosphorescence in solution 15,16 . This luminescence is thought to be due to the formation of a solid sphere in which the surface of the cluster is completely protected 17 . It has also been reported that such cluster complexes are a promising group of compounds that emit light at specific locations in the cell 18 .
More recently, N-heterocyclic carbene (NHC) ligands have been developed as one of the most promising organic ligands for Au 0 clusters with high designability [19][20][21][22] . Carbene ligands have strong electron-donating properties and are known to enhance the stability of metal surfaces, metal nanoparticles, and metal nanorods [23][24][25] . Therefore, NHC ligands have been used in the synthesis of Au 0 clusters [26][27][28][29][30] , and their interfacial structure, stability, and catalytic activity have been elucidated. In this context, we successfully applied the NHC ligand to the CAu I 6 cluster in 2018 [31][32][33] . We found that when an imidazolylidene carbene ligand was attached to each gold atom of the C-centered CAu I 6 cluster, the phosphorescence emission was red-shifted in the solid-state compared to clusters protected by phosphine ligands 31 . On the other hand, when a benzimidazolylidene ligand with one benzene ring fused to the imidazole ring was used, the phosphorescence emission showed a large blue shift of about 60 nm 32,33 . These two examples indicate that such a simple chemical modification can significantly change the photochemical properties of the clusters. Phosphorescent metal cluster complexes have the potential to precisely control the structure and electronic state of the metal cluster part by ligand design, and are expected to contribute not only to the creation of structurespecific photochemical functions but also to live-cell imaging and elucidation of intracellular molecular behaviors.
In this study, we design and synthesize bidentate ligands consisting of an NHC ligand linked to a pyridyl ligand, and clarify the detailed structure and photochemical properties of heteronuclear clusters of Au I and Ag I , which exhibit very high phosphorescence quantum yields even in solution. Furthermore, the strong phosphorescence emission of these clusters is used to elucidate the cellular uptake and organelle-selective translocation pathways of the cluster complexes (Fig. 1). Interestingly, the NHC ligand-protected heterometallic clusters are translocated to a specific intracellular organelle in a ligand-specific manner. This is in sharp contrast to phosphine-protected clusters with the same metal core structure, which are non-selectively dispersed in the cytosol. Thus, the control of intracellular translocation of C-centered Au I clusters is achieved by a slight modification of organic ligands. These results not only show that NHC ligands can improve the phosphorescence emission efficiency of C-centered Au I clusters, but also guide the development of functions for metal cluster-based functional molecules in chemical synthesis approaches.

Results
Synthesis and characterization. To further polynucleate the Au I cluster with different metals, a nitrogen donor was introduced into the wing-tip portion of each NHC ligand (N-isopropyl-N'−2-(5methylpyridyl)benzimidazolylidene (1a), N-isopropyl-N'−2-pyridylbenzimidazolylidene (1b), N-isopropyl-N'−2-(5-methylpyridyl) imidazolylidene (1c), N-isopropyl-N'−2-pyridylimidazolylidene (1d); Supplementary Figs. 1-4). Specifically, CAu I 6 complexes [(C) (Au I -L) 6 ](BF 4 ) 2 (2a-d, L = 1a-d) in which only the carbene ligand is coordinated to Au I , were synthesized from unsymmetrical bidentate ligands 1a-d. The isolation yields of 2a-d were 8-52% based on the amount of Au I used according to previously reported literatures [11][12][13][14][15][16][17][18] , and their molecular structures in the solid-state were determined by single-crystal X-ray diffraction (ScXRD) (Fig. 2 Fig. 3a, the molecular structures of CAu I 6 Ag I 2 clusters, [(C)(Au I -L) 6 Ag I 2 ](BF 4 ) 4 (3a-d, L = 1a-d), were determined by ScXRD, and all clusters were found to have a bicapped octahedral core. The two Ag I ions in each cluster are located on two opposite sides of the octahedron, each anchored by three pyridyl groups and interacting with three neighboring Au I atoms. These CAu I 6 Supplementary Fig. 47). The structures are similar to those of the previously reported [(C)(Au I -dppy) 6 Ag I 2 ](BF 4 ) 4 (4, dppy = 2-pyridyldiphenylphosphine) 14 , but differ in the following two points. First, because the C − Au I bond distances in 3a-d with NHC ligands are shorter than those of the P − Au I bonds in 4 with phosphine ligands, and also because the Au I -Ag I distances are shorter in 3a-d, the overall structures of the CAu I 6 Ag I 2 parts of 3a-d are significantly smaller than that of 4 (Fig. 3b). The Au I -Ag I distances in 3a-d are in a very similar range (Supplementary  Table 13). Second, there are no intramolecular C − H • • • Au interactions in 4.
Upon photoexcitation, 3a-d showed strong yellow luminescence in the solid-state with λ em max lying in the range from 559 to 578 nm, which significantly red-shifted as compared to 2a-d (from 482 to 490 nm), and is similar to 4 (553 nm, Fig. 3c and Supplementary Figs. 29 and 48). The two Ag I atoms do not simply bind to the CAu I 6 core but may change the electronic structure of the whole clusters [13][14][15][16][17] .
The CAu I 6 Ag I 2 clusters 3a-d were further characterized by solution-phase NMR spectroscopy (Supplementary Figs. 49-64). Two interesting features were observed in the 1 H NMR spectra. First, in the 1 H NMR spectra of 3a-d, the septet signals of the secondary hydrogen atoms of the isopropyl groups were shifted to the high field by about 0.5 ppm compared to 2a-d, which is a change similar to that observed when AgBF 4 was added to 2a-d . This result confirmed the existence of C − H • • • Au interactions in 3a-d even in solution 36 . The second feature is that split signals of methyl of isopropyl groups were observed in the both 1 H and 13 C NMR spectra of 3a-d. For example, the methyl groups of 2b showed only one set of signals in the 1 H (1.49 ppm) and 13 C NMR (21.9 ppm) spectra, whereas two sets of signals were observed in the 1 H NMR (1.44 and 1.33 ppm) and 13 C NMR (22.6 and 22.1 ppm) spectra for 3b. It was inferred from the molecular structures of 3a-d that this signal change was caused by the helical arrangement of the ligands in the bicapped octahedral CAu I 6 Ag I 2 and the two opposing CAu I 3 Ag I moieties showing different helical directions 14,15 . Heterometallic species, [(C)(Au I -L) 4 Au I Ag I ](BF 4 ) + and [(C) (Au I -L) 6 Ag I ](BF 4 ) 2 + (L = 1a-d), were also observed in the MS spectra ( Supplementary Fig. 65). These results strongly suggest that all of 3a-d maintain the bicapped octahedral structures even in solution.
Solutions of CAu I 6 Ag I 2 clusters 3a-d in CH 2 Cl 2 or CH 2 Cl 2 / CH 3 OH showed multiple optical absorption bands in the range of 300-450 nm (Fig. 3d). These optical absorption modes were consistent with those observed when AgBF 4 was added to the solutions of 2a-d. Their molar absorbance coefficients of 3a and 3b, protected with benzimidazolylidene ligands, were significantly higher than those of 3c and 3d, protected with imidazolylidene ligands.
Importantly, the use of the bidentate NHC ligands and the coordination of Ag I ions have a synergistic effect, and the heterogeneous metal clusters emit very strong phosphorescence in solution ( Fig. 3e and Supplementary Fig. 66). In comparison, the metal clusters 3a-d with NHC ligands emit yellow light in solution, whereas 4 with phosphine ligands emit red phosphorescence, with a difference in wavelength of about 90 nm. The phosphorescence quantum yields of 3a and 3b were determined to be 0.88 and 0.86 in CH 2 Cl 2 (λ em max = 562 nm), respectively, the highest values among the reported Au I clusters (Fig. 3g). In contrast, single crystals of 3c and 3d with imidazolylidene ligands showed similar yellow luminescence, but the phosphorescence quantum yields in CH 2 Cl 2 and CH 2 Cl 2 /CH 3 OH were remarkably low, 0.14 and 0.01, respectively. In addition, the phosphorescence lifetimes of 3a and 3b protected with benzimidazolylidene ligands were 1.85 and 1.66 μs, respectively, which were significantly longer than those of 3c and 3d protected with imidazolylidene ligands, 0.32 and 0.16 μs, respectively (Fig. 3f). The radiative (k r ) and non-radiative rate constants (k nr ) shown in Fig. 3g indicate that clusters with NHC ligands have significantly higher radiative emission rates than clusters with phosphine ligands. Furthermore, the benzimidazolylidene ligand could dramatically improve the quantum yield and microsecond-level lifetime of the phosphorescence of the CAu I 6 Ag I 2 clusters, 3a and 3b, by significantly suppressing the non-radiative relaxation pathways.
Theoretical calculations of CAu I 6 Ag I 2 clusters. In order to mechanistically clarify how the photochemical properties change depending on the type of carbene ligands, the absorption and phosphorescence properties of 3b with benzimidazolylidene ligands and 3d with imidazolylidene ligands were theoretically calculated and comparatively analyzed (Fig. 4). As a result, the calculated absorption spectra agreed well with the experimental observations in the calculated energy range. The trend of the energy difference was reproduced; the first peaks for 3b and 3d were calculated to be 387 and 388 nm, respectively. The order of the molar absorbance coefficients (3b > 3d) was also well reproduced. The distributions of SOMO and SOMO − 1, which characterize phosphorescence, are similar to those of LUMO and HOMO, respectively (Fig. 4c, d). The imidazolylidene and benzimidazolylidene ligand moieties are involved in the SOMO − 1 orbitals. The calculated phosphorescence energies of 3b and 3d are 2.09 and 2.08 eV (592 and 596 nm, respectively), which are in close agreement with the experimental values of 2.21 and 2.17 eV, respectively, within error 37 (Supplementary Table 21).
It is worth noting that the molecular structure of the clusters in the singlet state is very different from that in the triplet state 8,9,38,39 (Supplementary Fig. 71 and Supplementary  Table 23). For example, the Au I -Au I distances of 3b in the . a Molecular structures of 3a-d (50% probability for 3a, 3c, and 3d; 25% probability for 3b), with the anions BF 4simplified. b Comparison of the key structure parameters of 3a-d and 4. c Emission spectra of 3a-d and 4 in the solid-state, with λ em max being 559, 573, 578, 570, and 553 nm, respectively, and corresponding photos at room temperature. d UV-vis absorption spectra of 3a (ε 336 = 8.4 × 10 4 , and 4 (ε 321 = 1.8 × 10 4 M −1 cm −1 ) in CH 2 Cl 2 (293 K). e Emission spectra of 3a-d and 4 in CH 2 Cl 2 at 293 K, with λ em max of 562, 562, 564, 571, and 650 nm, respectively, and the corresponding photos taken at room temperature. f Emission decay of 3a-d and 4 in CH 2 Cl 2 at room temperature, with τ of 1.85, 1.66, 0.32, 0.16, and 3.74 µs, respectively. g Quantum yields, radiative rate constants (k r ) and non-radiative rate constants (k nr ) of 3a-d and 4 in CH 2 Cl 2 at room temperature. Note that a mixed solvent CH 2 Cl 2 /CH 3 OH (9:1, v:v) was used for 3d 42 . singlet state were found to be in the range of 3.090−3.203 Å, while in the triplet state they were found to be in the range of 2.931−3.443 Å. Therefore, we quantitatively analyzed the orbital compositions to evaluate the ligand effects. Mulliken partition analysis confirmed that even though the core of CAu I 6 Ag I 2 is mainly involved in the MOs, the ligands make significant contributions (Supplementary Tables 24-26). Unlike the phosphine ligand, the carbene ligand reduced the involvement of the ligand in the frontier orbitals and suppressed the involvement of the ligand in the charge transfer processes (see Supplementary Information for the details).
We then directly investigated the phosphorescence lifetimes and the radiative rate constants of 3b and 3d using the ZORA method with spin-orbit interaction in a perturbative way 40 implemented in the ADF program package 41 . The results are compared to the experimental values, which approximate the calculated k r as k r = 1/τ (Supplementary Table 27). The phosphorescence is attributed to the three lowest, nearly degenerate spin-orbit states, which are about 0.1 eV lower than other states, with T 1 being the dominant contribution. Although the order of the absolute values of τ differs from the experimental values, the trend of k r (3b > 3d) agrees with each other. Importantly, the k r values of imidazolylidene-and benzimidazolylidene-protected clusters were almost quantitatively reproduced in the calculations. Note that the low k r of 3d with imidazolylidene ligands is due to the solvent effect of CH 3 OH 42 . In addition, we analyzed the wavefunction and spin-orbit coupling of the low-lying spin-orbit states (Supplementary Tables 27 and 28). These states are described mainly by the T 1 , T 2 , and T 3 components, with small contributions from the S 1 and S 3 components. It can be seen that the NHC ligand significantly changed the main component of each state and the coupling of the spin-orbit states. This is thought to be the origin of the different k r values of these compounds.
Finally, in order to evaluate k nr , the minimum energy crossing point (MECP) between S 0 and T 1 states was calculated using the Harvey method 43,44 . The results showed that the core of CAu I 6 Ag I 2 is significantly deformed at the MECP. For example, the shape of CAu I 6 Ag I 2 of 3d is significantly changed compared to the shape of the T 1 state (Supplementary Fig. 72). The energy barrier from the minimum of T 1 state to MECP were calculated as 11.6 and 10.8 kcal/mol for 3b and 3d, respectively, which qualitatively agrees with the k nr of these complexes (Supplementary Table 29). However, the energy difference of the MECP for 3b and 3d is small, so there may be another cause. Overall, although NHC ligands are less involved in the electronic structure of the clusters than phosphine ligands, they can accelerate radiative decay by enhancing the spin-orbit coupling of the lowlying spin-orbit states. On the other hand, when the benzimidazolylidene ligand was used, non-radiative decays did not occur preferentially, resulting in 3a and 3b having very high quantum yields.
Ligand-specific translocation in cells. Luminescent metal complexes have been used as promising bioimaging reagents for the past two decades [45][46][47] Table 30), we further examined their behavior in living cells to determine if there are differences in subcellular distribution and cell renewal pathway depending on the organic ligands. Figure 5 shows the optimized bioimaging results for 3a, 3b, and 4. At the concentration of 1.0 or 2.0 μM, where no cytotoxicity was observed by confocal microscopy analysis ( Supplementary  Figs. 75-82), 3a, 3b, and 4 entered HeLa cells within 10 min. Clusters 3a and 3b were suggested to accumulate in specific organelles, whereas cluster 4 was uniformly distributed in the cytosol. HEK293T and COS7 cells labeled with 3a or 3b showed a similar distribution pattern. To identify the accumulated structures, the cells labeled with 3a or 3b were further labeled with several organelle markers. Confocal microscope analysis showed that 3a and 3b were selectively located in the ER. The time-lapse imaging revealed that clusters 3a and 3b were transported into the cells within 5 min after accumulation on the surface. 3a or 3b had accumulated in the ER after 10 min, and the longer the incubation time, the more it accumulated in the nucleus region ( Fig. 5d and Supplementary Figs. 78, 80). The accumulation of 3a in the ER, induced by a 10-min incubation, was maintained for 36 h (Supplementary Fig. 84). Thus, clusters with the NHC ligands can be selectively translocated to an intracellular organelle in a manner specific to the ligand structure. Furthermore, since these clusters have long emission lifetimes on the order of microseconds, phosphorescence lifetime imaging (PLIM) was conducted. The calculated lifetime of 3a was 0.15 μs, which was in good agreement with the in vitro result. Thus, 3a can be used as an ER-selective targeting reagent, and its phosphorescence can be well distinguishable from autofluorescence.
It is noteworthy that CAu I 6 Ag I 2 clusters protected by the NHC or phosphine ligands showed different distributions in the cells, even though they have almost the same core structure. To clarify the cellular uptake pathway of 3a and 4, cells were incubated with 3a and 4 at a lower temperature (4°C), where endocytosis was expected to be non-specifically inhibited. As shown in Fig. 6a, the cells labeled under the condition did not show any luminescence signals. Then, route-specific inhibitors such as wortmannin, sucrose, and genistein were used at 37°C to specifically inhibit macropinocytosis, clathrin-and caveolindependent endocytosis, respectively. For cluster 3a with benzimidazolylidene ligands, the genistein treatment effectively inhibited the ER accumulation. These results suggest that the uptake of cluster 3a is due to a caveolin-dependent endocytosis process. In contrast, the accumulation of cluster 4 with phosphine ligands in the cytosol was blocked by all inhibitors. Thus cluster 4 was taken up into the cells and dispersed in a nonspecific manner. Based on these results, we proposed a possible route of cellular uptake and intracellular translocation of NHC-and phosphine-protected CAu I 6 Ag I 2 clusters (Fig. 6b). These clusters first accumulate on the surface of the cell membrane and are then taken up into the cells by endocytosis. For phosphine-protected 4, the uptake is an energy-dependent, nonspecific process that is associated with macropinocytosis, clathrin-mediated, and/or caveolin-dependent endocytosis. In the case of NHC-protected 3a-d, caveolin-dependent endocytosis is the main pathway for cellular uptake. NHC-based clusters 3a and 3b were sequentially translocated to ER after intracellular uptake, but phosphine-based cluster 4 was released from intracellular compartments to cytosol within 10 min. Prolonged incubation allowed the nuclear accumulation of the clusters 3a, 3b, and 4, which in turn increased the permeability of their membranes. In addition, high concentration of clusters accelerated these processes.

Discussion
In this study, we have established a rational design and synthesis method for a series of phosphorescent Au I -Ag I clusters, and revealed that the organic ligands bound to the surface of the metal clusters are important for their photochemical property and subcellar distribution. The NHC ligands were found to remarkably shift the emission wavelength of the CAu I 6 Ag I 2 clusters, and significantly affecting the phosphorescence quantum yield and lifetime, compared to the same cluster protected by phosphine ligands. More importantly, the CAu I 6 Ag I 2 clusters were taken up into the living cells and showed structure-specific translocation pathways and distribution. This suggests that molecular design may be able to control molecular behaviors in the cell. Therefore, this study is expected to provide a strategy to create metal clusters with tunable functionalities, and also to lead to clear guidelines for designing active and applicable metallodrugs.

Methods
Synthesis of 1a·HI and 1b·HI. Under a nitrogen atmosphere, a Schlenk tube was charged with benzimidazole (5.00 g, 42.3 mmol), K 2 CO 3 (3.90 g, 28.2 mmol), and the corresponding bromopyridine derivatives: 2-bromo-5-methylpyridine (2.43 g, 14.1 mmol) for 1a·HI and 2-bromopyridine (1.37 mL, 14.1 mmol) for 1b·HI. The reaction mixture was heated at 200°C for 12 h and then allowed to cool to room temperature. After it was diluted with water (50 mL) and extracted with CH 2 Cl 2 (50 mL × 3), the combined organic phase was washed with sat. Na 2 CO 3 aqueous solution (50 mL × 3), and brine (50 mL), and then dried over anhydrous MgSO 4 before filtration. Concentration under reduced pressure gave a light red oil. This intermediate was then transferred into a Schlenk flask and dissolved in CH 3 CN (50 mL) under a nitrogen atmosphere. 2-Iodopropane (1.51 mL, 15.2 mmol) was added to the solution, and the reaction mixture was heated at reflux for 12 h. After cooling to room temperature, the mixture was concentrated to ca. 5 mL under reduced pressure. When diethyl ether (50 mL) was added to this residue, a pale yellow precipitate was obtained as a crude product. Colorless crystals 1a·HI and 1b·HI were obtained by layering diethyl ether on CH 2 Cl 2 /CH 3 CN (9:1, v:v) solution containing the crude product. Yields: 2.30 g (43%, based on 2-bromo-5methylpyridine) for 1a·HI; 1.90 g (37%, based on 2-bromopyridine) for 1b·HI. Synthesis of 1c·HI. Under a nitrogen atmosphere, a Schlenk tube was charged with imidazole (2.04 g, 30.0 mmol), K 2 CO 3 (2.76 g, 20.0 mmol), and 2-bromo-5methylpyridine (1.72 g, 10.0 mmol). The reaction mixture was heated at 190°C for 12 h and then allowed to cool to room temperature. After it was diluted with water (50 mL) and extracted with CHCl 3 (50 mL × 3), the combined organic phases were washed with sat. Na 2 CO 3 aqueous solution (50 mL × 3), and then dried over anhydrous MgSO 4 before filtration. Concentration under reduced pressure gave a colorless oil, which was then transferred into a Schlenk flask and dissolved in After cooling to room temperature, the mixture was concentrated to ca. 5 mL under reduced pressure. When diethyl ether (50 mL) was added to this residue, a pale yellow precipitate was obtained as a crude product. Colorless crystals of 1c·HI were obtained by layering diethyl ether on CH 2 Cl 2 /CH 3 CN (9:1, v:v) solution containing the crude product. Yield: 1.71 g (52%, based on 2-bromo-5methylpyridine). 1 (10 mL). Tht-AuCl (96.0 mg, 0.30 mmol) was added and then the solution was stirred for 5 min, which was followed by adding K 2 CO 3 (828 mg, 6.00 mmol). After the mixture was stirred for 12 h in the dark, it was filtered through a thin layer of Celite. The solvent was then removed under reduced pressure using a rotary evaporator. After adding NaBF 4 (165 mg, 1.50 mmol) and CH 3 OH (10 mL), the suspension was stirred for 5 min. CH 2 Cl 2 (5 mL), a solution of KOH (28.0 mg, 0.50 mmol) in CH 3 OH (3 mL), a solution of AgBF 4 (58.5 mg, 0.30 mmol) in CH 3 OH (1 mL), and H 2 O (50 μL) were sequentially dropwise added into the mixture under stirring, which leads to a brown suspension. After another 5 min stirring, the suspension was again filtered through Celite and evaporated to dryness. The solid was then transferred to a Schlenk flask with a nitrogen atmosphere, and dry CH 2 Cl 2 (5 mL), Et 3 N (30.0 μL, 0.20 mmol), and a 2.0 M solution of Me 3 SiCHN 2 in hexanes (48.0 μL, 0.10 mmol) were added. The resulting mixture was stirred for another 1 h. After filtration into a tube, a layer of dry Et 2 O was added to the CH 2 Cl 2 solution, which gave the products colorless block-like crystals within 2 weeks. Yields: 59.5 mg (40%, based on tht-AuCl) for 2a; 74.3 mg (52%, based on tht-AuCl) for 2b; 31.3 mg (24%, based on tht-AuCl) for 2c; 11.6 mg (9%, based on tht-AuCl (1d·HBr)) for 2d; 10.0 mg (8%, based on tht-AuCl (1d·HI)) for 2d.  4 AuAg](BF 4 ) + ).
X-ray crystallography. Intensity data of compounds 1a-c·HI, 2a-d, and 3a-d were collected on a Rigaku XtaLAB Synergy-DW system (CuKα) at 93 K. The structures were solved by direct methods, and non-hydrogen atoms except for the disordered BF 4and CH 2 Cl 2 in 2b were refined anisotropically by the least squares on F 2 using the SHELXTL program. The hydrogen atoms of organic ligands were generated geometrically; no attempt was made to locate the hydrogen atoms of disordered dichloromethane molecules in 2b and water molecules in 2d.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of this study are available within the manuscript and its supplementary information and from the corresponding author upon request. The