Self-Assembly of Metallo-Supramolecules under Kinetic or Thermodynamic Control: Characterization of Positional Isomers Using Scanning Tunneling Spectroscopy

Abstract

Coordination-driven self-assembly has been extensively employed to construct a variety of discrete structures as a bottom-up strategy. However, mechanistic understanding regarding whether self-assembly is under kinetic or thermodynamic control is less explored. To date, such mechanistic investigation has been limited to distinct, assembled structures. It still remains a formidable challenge to study the kinetic and thermodynamic behavior of self-assembly systems with multiple assembled isomers due to the lack of characterization methods. Herein, we use a stepwise strategy which combined self-recognition and self-assembly processes to construct giant metallo-supramolecules with 8 positional isomers in solution. With the help of ultrahigh-vacuum, low-temperature scanning tunneling microscopy and scanning tunneling spectroscopy, we were able to unambiguously differentiate 14 isomers on the substrate which correspond to 8 isomers in solution. Through measurement of 162 structures, the experimental probability of each isomer was obtained and compared with the theoretical probability. Such a comparison along with density functional theory (DFT) calculation suggested that although both kinetic and thermodynamic control existed in this self-assembly, the increased experimental probabilities of isomers compared to theoretical probabilities should be attributed to thermodynamic control.

Graphical Abstract

INTRODUCTION

Because of its intrinsic dynamic nature, self-assembly is able to organize suitable molecular components into more sophisticated supramolecular entities by virtue of noncovalent interactions.¹ Typically, self-assembly involves multiple reversible processes of dissociation, association, and reorganization in which the structures or properties can be tuned by physical conditions or external stimulus.² Studies have revealed that factors such as ligand geometry, metalions/anions, reaction time, temperature, concentration, and solvent can affect the final products of self-assembly.³ Beyond these factors, the assembled structures could also be determined by kinetic control or thermodynamic control.^2b In the majority of systems, the reversible and weak noncovalent interactions favor thermodynamically controlled structures with optimum conditions.⁴ In contrast, kinetically controlled self-assembly can form inert structures using less labile interactions⁵ or metastable products that are kinetically trapped at local minima^6,7 and may be further transformed to more stable products upon prolonged reaction time.⁷ In one of the pioneering works, Lehn and co-workers revealed that the triple helicate assembled by a tris-bipyridine ligand could progressively convert into the circulate helicate, corresponding to the switch from kinetic control to thermodynamic control.^7a In Fujita’s self-assembled M_nL_2n polyhedra,-thermodynamic control played a predominant role when n = 6, 12, or 24. While for n > 24, the kinetic trapping is no longer negligible.^6c As such, M₃₀L₆₀ and M₄₈L₄₉ assembled by the same ligand with Pd(II) were categorized as kinetically and thermodynamically controlled products, respectively.^7d It is worth noting that these assemblies generally gave distinct assembled structures regarding kinetic or thermodynamic control, facilitating detailed characterization.

Supramolecular isomerism has been well studied in coordination polymers and metal–organic frameworks which highly rely on crystal engineering to confirm the structure explicitly.⁸ In coordination-driven self-assembly of discrete metallo-supramolecules, more attention was focused on precise control over the shape and size of assemblies.⁹ Through careful design, a single metallo-supramolecule rather than a mixture of isomers was obtained in individual self-assembly.¹⁰ However, isomerism sometimes still occurred, for instance, positional isomers from the unbiased arrangement of dissymmetrical building blocks,¹¹ configurational isomers from the asymmetric chelating ligand arrangements around the metal centers,¹² social and constellation isomers through different arrangement/orientation of guests encapsulated in a confined host,¹³ and unidentical packing modes during the crystallization process.¹⁴ Such supramolecular isomerization at the molecular level is expected to increase the complexity and diversity of supramolecular chemistry and thus has resulted in applications in molecular devices and optics.¹⁵ However, supramolecular isomerism in discrete metallo-supramolecules remains less explored compared to MOFs and polymers, perhaps due to the lack of platforms with rational, precise design and insufficient characterization by NMR,^13,15 crystal engineering,^12,14 ion mobility-mass spectrometry (IM-MS),^10a and molecular modeling.^11a,16

In the present work, we demonstrated the construction of three giant trimetallic supramolecules through the self-recognition and self-assembly of a metal–organic ligand L. As a consequence of intramolecular and intermolecular complexation, three types of metal ions were successfully incorporated into the discrete metallo-supramolecules using terpyridine (tpy)-based¹⁷ coordination chemistry to form ⟨tpy–M–tpy⟩ connectivities in a stepwise manner (Figure 1A). Because of the dissymmetrical nature of the intermediate L·M after self-recognition, eight positional isomers were formed in the self-assembly of the final supramolecules in solution (Figure 1B). In terms of large molecular weight and high complexity, it remains a formidable challenge to address whether the self-assembly of those positional isomers is under kinetic and/or thermodynamic control with conventional characterization methods. Herein, we report the combined use of ultrahigh-vacuum, low-temperature scanning tunneling microscopy (UHV-LT-STM)¹⁸ and scanning tunneling spectroscopy (STS)¹⁹ to characterize the isomers at the atomic level in this complex supramolecular system. Using these methods, we investigated the contribution of thermodynamic and kinetic control in the outcome of isomer distribution.

Figure 1.

(A) Construction of three trimetallic supramolecules SA-SC through self-recognition and self-assembly. (B) Eight isomers generated solution after self-assembly.

RESULTS AND DISCUSSION

Synthesis of Trimetallic Supramolecules.

The metal–organic tpy ligand L was synthesized with ⟨tpy–Ru(II)−tpy⟩ connectivity using a Ru(II) end-capping approach by performing a Sonogashira coupling reaction on the Ru–tpy complex (Scheme S1).²⁰ L possesses six free tpy for self-recognition and self-assembly. In the introduction of the second type of metal ion, one equivalent of either Fe(II), Co(II), or Cr(II) was mixed with L at 80 °C to form an exclusive intermediate (L·M) through specific coordination or self-recognition. Conventional electrospray ionization-mass spectrometry (ESI-MS) first suggested formation of the proposed intermediates. The measured m/z agreed well with the molecular compositions of the intermediates, i.e., 849.81 for L·Fe, 850.56 for L·Co, and 853.06 for L·Cr at charge state 4+ (Figures S35 and S36). Here, Cr(II) was oxidized to Cr(III) by air bubbling, and an −OH group was attached to balance the increased valence.^{21 1}H NMR was also applied to confirm the specific coordination site within the structure of the intermediate L·Fe. Two sets of signals with metal and three sets of free tpy signals were observed in L, while four sets of coordinated tpy signals and three sets of free tpy signals were assigned in L·Fe because of the loss of molecular symmetry after complexation with an additional equivalent of Fe(II). The downfield shift of the tpy–E³′⁵′ signal and tpy–F³′⁵′ signal proved that Fe(II) was selectively coordinated through self-recognition (Figure 2). Detailed assignments here were based on 2D COSY and 2D NOESY spectra (Figures S19−S22). However, the ¹H NMR spectra of L·Co and L·Cr are broad (Figures S24 and S25), and unsatisfactory results were obtained for 2D NMR experiments because of the paramagnetic nature of Co(II) and Cr(III). Nevertheless, two sets of characteristic tpy signals spread out in a wide range (20–100 ppm) corresponding to the ⟨tpy–Co(II)−tpy⟩ connectivity²² were identified in the proton NMR, supporting formation of the proposed intermediate L·Co (Figure S24). We speculated that exclusive formation of the intermediate was energetically more favorable through intramolecular complexation rather than intermolecular complexation.²³ In addition, those intermediates are reminiscent of folding in protein structures, resulting from conformational regulation.²⁴ Without further purification or separation, two equivalents of Zn(II) as the third type of metal ion were added to the reaction mixture for self-assembly for 8 h at 50 °C. Among these metal ions, Zn(II) has the weakest coordination with tpy to facilitate self-assembly without disturbing the previous metal ion coordination.^25c After counterion exchange and washing, three targeted metallo-supramolecules with a fractal pattern^25,20b were obtained in high yield.

Figure 2.

¹H NMR spectra (500 MHz, 300 K) of L in CDCl₃ and L·Fe in CD₃CN.

Characterization of Supramolecules.

ESI-MS and traveling wave ion mobility-mass spectrometry (TWIM-MS)²⁶ were applied to provide structural information on the final heterometallic supramolecules. Only one set of continuous charged signals was observed from ESI-MS (Figure 3A and 3C; Figure S39a) corresponding to molecular weights of 28 136, 28 155, and 28 215 Da for SA, SB, and SC, respectively. A narrow distribution of drift time for each charge state in the TWIM-MS spectra (Figure 3B and 3D; Figure S39b) suggested clean formation of discrete metallo-supramolecules with high shape persistence.

Figure 3.

ESI-MS spectra of (A) SA and (C) SB; TWIM-MS plot (m/z vs drift time) of (B) SA and (D) SB.

Furthermore, the corresponding isotope pattern measured for each charge state of the supramolecules agreed well with the theoretical one (i.e., 1335.9 for SA, 1336.8 for SB, and 1340.0 for SC at charge state 19+), which strongly supports formation of the proposed structure (Figures S40−S42). The mass spectrometry data also excluded the possibility of disassembly and reassembly of ⟨tpy–Fe(II)−tpy⟩, ⟨tpy–Co(II)−tpy⟩, or ⟨tpy–Cr(III)−tpy⟩ within the targeted metallo-supramolecules. If disassembly and reassembly occurred, it was expected to observe multiple sets of signals in mass spectrometry corresponding to different numbers (0–18) of Fe(II), Co(II), or Cr(III) in the final assembled structures. Gradient tandem mass spectrometry (gMS²)²⁷ showed that the three supramolecules were all fully dissociated at +25 V, corresponding to a center-of-mass collision energy of 0.04 eV. This implies similar stability of these three metallo-supramolecules in the gas phase despite three different metal ions being incorporated (Figures S43−S45). Thus, the stability of the whole supramolecule is determined by the weakest ⟨tpy–Zn(II)−tpy⟩ among the three ⟨tpy–M–tpy⟩ connectivities.²⁸

Due to each L·M having two different orientations in the self-assembly with Zn(II), the final metallo-supramolecules would generate eight isomers (Figures S47 and S48) leading to very broad NMR spectra. Nevertheless, the characteristic signals in the ¹H NMR, 2D COSY, and 2D NOESY spectra of complex SA were still assigned and consistent with the proposed structure (Figures S28−S32). Diffusion-ordered spectroscopy (DOSY) NMR showed the diffusion coefficient (D, m²/s) of SA to be 1.43 × 10⁻¹⁰, which corresponded to an experimental hydro-dynamic radius (r_H) of 5.7 nm.^17e,20 This also agreed with the molecular diameter obtained from modeling (r = 5.5 nm) (Figure S34).

In addition to the structural characterizations described above, UHV-LT-STM was utilized to directly visualize the structures of the giant metallo-supramolecules. Working by the mechanism of quantum tunneling effects under a small bias voltage of ±1–2 V, STM effectively prevents molecular structural damage such as knock-on damage, heating effects, and radiolysis that can be introduced by other microscopy characterization methods.²⁹ Moreover, STM investigations can provide detailed information on the topography and electronic properties at the atomic scale, making it more powerful for characterization of self-assembled structures.³⁰ Since all three types of supramolecules are nearly identical in shape, SA was selected as a representative for STM characterization. For the STM experiments, a freshly prepared sample solution of SA (1 × 10⁻⁵ M) in acetonitrile was dropped on a Ag (111) surface, and then the sample was cooled down to ~4 K to reduce the thermal motion of the molecules, thereby a high resolution was achieved. Remarkably, intact supramolecules were observed directly in the scanned areas with some fragments absorbed on the surface (Figures 4A and S46). Note that the supramolecules can undergo disassembly via the weakest ⟨tpy–Zn(II)−tpy⟩ in a very dilute solution. With an applied positive voltage of +2 V, each ⟨tpy–M–tpy⟩ connection unit was visualized as a bright lobe in the STM image, while the overall organic backbone of the supramolecule was displayed at the negative bias voltage of −2 V (Figure 4E and 4F). The measured height (ca. 6 Å) and as well as the diameter (ca. 10.5 nm) perfectly matched the size and dimension predicted by theoretical modeling (Figure 4B–D).

Figure 4.

(A) Large-area STM image with supramolecule SA randomly anchored (scale bar 10 nm, scanning parameters V_t = 2 V, I_t = 130 pA). (B) Zoomed-in STM image for line-profile measurements(scale bar 2 nm, scanning parameters V_t = 2 V, I_t = 130 pA). (C and D) STM line profile measurements along green and red dashed lines inside the hexagon grid shown in B. Topography of SA at (E) bias voltage = +2 V (scale bar 2 nm, scanning parameters V_t = 2 V, I_t = 50 pA) and (F) bias voltage = −2 V (scale bar 2 nm, scanning parameters V_t = −2 V, I_t = 50 pA) with molecular modeling overlay.

Identification of Isomers.

With a well-visualized molecular structure by STM imaging, another important question regarding the existence of isomers still remained. As mentioned earlier, the intermediate L·M is dissymmetrical in chemical composition, and it could flip over during the self-assembly with Zn(II), leading to formation of 8 positional isomers with different orientations of Ru vs Fe in solution (Figures S47 and S48). Identification of those positional isomers is extremely challenging using conventional characterization methods, e.g., mass spectrometry, NMR, and X-ray diffraction. Therefore, scanning tunneling spectroscopy (STS) was employed here to probe the local density of states (LDOS) for each metal junction to differentiate these isomers at the atomic scale.¹⁹ Notably, 14 positional isomers are expected on the substrate (Figure 5A) because the two sides of each single molecule are no longer identical when depositing the sample solution on the surface of Ag (111) for STS measurement. To avoid confusion, those pairs of isomers with relatively the same positions of Ru vs Fe while two sides of molecules are considered the same type of isomers when doing the statistic calculation discussed below. In addition, the probability of the pair of isomers regarding the different sides of the same type of supramolecule was found to be equal on the substrate (Figure S61), further supporting our classification for isomer types as reasonable.

Figure 5.

(A) Fourteen isomers generated on substrate, and corresponding 8 isomers in solution, which were labeled a–h. (B, E, H, K) STM images of four isomers of SA determined by STS (scanning parameters V_t = 2.5 V, I_t = 48 pA). (C, F, I, L) dI/dV − V tunneling spectroscopy data of Fe(II) (blue) corresponding to the supramolecules shown in B, E, H, K. (D, G, J, M) dI/dV − V tunneling spectroscopy data of Ru(II) (red) corresponding to the supramolecules shown in B, E, H, K. (N) Histogram of experimental and theoretical probability (P_E and P_T) for eight isomers in solution after measuring 162 supramolecules. (O) Histogram of relative deviation δ of eight isomers in solution calculated from experimental and theoretical probabilities (δ = [(P_E − P_T)/P_T]).

We first measured the LDOS of the metal junctions of a reference molecule G5,^20b which contains only one type of metal ion Ru(II) in its outermost ring. The obtained STS data in a single molecule showed a distinct average band gap of 2.8 eV (Figure S49), which later served as the reference to identify ⟨tpy–Ru(II)−tpy⟩. As a following step, the STS data was collected for all 12 metal junctions in the outermost ring of a single SA supramolecule. After a careful analysis of the data and classifying the values into two groups, an average 2.8 eV band gap was observed for six metal junctions which could be assigned to ⟨tpy–Ru(II)−tpy⟩ according to the reference (Figure S49); meanwhile, an average band gap of 2.4 eV was found for the other six lobes, corresponding to ⟨tpy–Fe(II)−tpy⟩ (Figure 5B–M). As such, we are able to differentiate each Fe(II) from Ru(II) based on the band gap in a single molecule and successively determine the type of isomers out of the 14 possible on the substrate (Figure 5B, 5E, 5H, and 5K).

Such STS measurements were extended to a total of 162 supramolecules (Figures S50−S57), which allowed us to characterize all 14 isomers on the Ag (111) surface (Figures 5B–M and S58) for distribution analysis. A subsequent statistical study was performed to investigate the experimental and theoretical probability distributions. Overall, the experimental probabilities (P_E) basically followed the trend of the theoretical probabilities (P_T) (examined by chi-square test for independence, Tables S1−S3) but displayed variations from isomers (Figure 5N). It should be noted that P_T is calculated mathematically (Figure S59) and is primarily viewed as the result of self-assembly under kinetic control. To better present the variations between P_E and P_T, we calculated the relative deviation δ using the formula δ = [(P_E − P_T)/P_T] (Figure S60), and the results were plotted in a histogram as shown in Figure 5O. Isomers a and h have the same P_T of 0.0313, but experimental results showed an opposite trend, which leads to a negative δ value of −0.41 for isomer a and the highest δ value of +1.77 for isomer h. Meanwhile, isomers b, c, and f all showed negative δ values, and isomers d, e, and g showed positive δ values. These results implied that the second intramolecular complexation might not be dominated by kinetic control only; thermodynamic control also played an important role, which leads to an energy-biased isomer distribution.

Considering that the difference only came from the relative position of Ru vs Fe in the outermost ring in the chemical structure of isomers a–h, we proposed that the system could be energy favored when the same kinds of metals on the outer layer were neighboring on adjacent ligands. In other words, the total energy of the eight isomers should follow such a trend a > b, c, f > d, e, g > h based on the increasing number of the same neighboring metal ions on adjacent ligands (Figure S48). Our hypothesis was further supported by theoretical calculations, in which density functional theory (DFT) was applied to calculate the total energy of each isomer using the Vienna Ab initio Simulation Package (VASP) code with the Hubbard U (DFT +U) corrections.³¹ Note that only metal ions in the isomers were extracted to simplify the calculation (Figure S1). As shown in Table 1, isomer a has the highest total energy of −12.36 eV compared to the lowest one of −18.28 eV for isomer h, which is suggested as the thermodynamically controlled product among 8 isomers. Also, the energy levels of other isomers are also in agreement with our proposed trend. At this point, we can conclude that the generation of these positional isomers was determined by kinetic and thermodynamic controls simultaneously, while thermodynamic control might play a more predominant role for the isomers with increased probabilities compared to theoretical probabilities. We speculated that the ligands could initially self-assemble with Zn(II) to form the kinetically stable isomers based on mathematical statistics; then thermodynamic control enabled the self-assembly to rearrange the distribution and enhanced formation of the more energy-favored structures.³²

Table 1.

Total Energy (eV) of Eight Isomers Calculated from DFT

isomer	a	b	c	d	e	f	g	h
energy	−12.36	−13.17	−13.03	−16.87	−16.06	−12.71	−18.13	−18.28

CONCLUSIONS

In conclusion, we successfully constructed a series of trimetallic supramolecules in one pot using the strategy of self-recognition and self-assembly. Through self-recognition, Fe(II), Co(II), or Cr(III) selectively coordinated with the preorganized metal–organic ligand to form exclusive intermediates. The subsequent self-assembly with the weakly coordinating metal ion Zn(II) afforded the final supramolecules SA−SC with more diversity through formation of 8 positional isomers. The intermediates and the supramolecules were well characterized by mass spectrometry and NMR spectroscopy. SA was directly visualized using UHV-LT-STM, and all 8 isomers in solution (corresponding to 14 isomers on substrate) were successfully identified using STS to probe the local chemical environment with the LDOS at the atomic level. More importantly, the occurrence of supramolecular isomerism in our system was found to be determined by thermodynamic and kinetic control collaboratively using a statistical study and DFT calculation. Also, their contributions were quantitatively understood through the deviation between the theoretical probability and the experimental probability measured from the STS method we developed. Therefore, this study demonstrates that we are able to study the kinetic and thermodynamic features of supramolecular isomerism at the single molecular level in a complicated system. In addition, those trimetallic metallo-supramolecules may introduce metal-dependent properties and offer potential opportunities for future applications.