How to Prepare Kinetically Stable Self‐assembled Pt₁₂L₂₄ Nanocages while Circumventing Kinetic Traps

Abstract

Supramolecular coordination‐based self‐assembled nanostructures have been widely studied, and currently various applications are being explored. For several applications, the stability of the nanostructure is of key importance, and this strongly depends on the metal used in the self‐assembly process. Herein, design strategies and synthetic protocols to access desirable kinetically stable Pt₁₂L₂₄ nanospheres are reported, and it is demonstrated that these are stable under conditions under which the palladium counterparts decompose. Descriptors previously used for palladium nanospheres are insufficient for platinum analogues, as the stronger metal–ligand bond results in a mixture of kinetically trapped structures. We report that next to the dihedral angle, the rigidity of the ditopic ligand is also a key parameter for the controlled formation of Pt₁₂L₂₄ nanospheres. Catalytic amounts of coordinating additives to labilise the platinum‐pyridyl bond to some extent are needed to selectively form Pt₁₂L₂₄ assemblies. The formed Pt₁₂L₂₄ nanospheres were demonstrated to be stable in the presence of chloride, amines and acids, unlike the palladium analogues.

Avoiding the unwanted: Robust Pt₁₂L₂₄ assemblies were selectively formed by a combination of thermodynamic tuning, achieved by rigidifying the ligands, and kinetic tuning with catalytic amounts of additives. The described methodology can be applied to prepare a variety of functionalised platinum assemblies that were shown to be kinetically stable against biologically and catalytically relevant substrates.

Introduction

Supramolecular coordination‐based self‐assembled nanostructures have a potentially rich diversity of applications including drug delivery,^[1] containers for catalytic transformations,^[2] separation vehicles^[3] and many more.^[4] In general, the properties of supramolecular structures are unique in comparison to the components they are based on, most often because of the three dimensional structure that is generated, creating specific spaces and organising molecular components.^[5] Therefore, it is important that the coordination‐based supramolecular nanostructures are stable under the conditions required for applications such as catalysis, guest binding and separation and in vivo applications.[^2c, ^2d, ^2e, ^4b, ⁶] Strategies to enhance the stability of coordination‐based self‐assemblies are generally based on increasing the metal‐ligand bond strength, which can be achieved by changes in the ligand system or by using a different metal site. Examples that led to more robust assemblies by ligand design include changes in the electronical properties of the donation site and changes in the steric properties.[^1c, ⁷] Changing the metal site might lead to more robust assemblies when known building blocks are used, without changing too much the overall structure. Prominent examples include the change of iron to nickel in tetrahedral M₄L₆ assemblies,^[8] the change from copper to rhodium in M₄L₄‐type assemblies^[9] and the change of palladium to platinum in M₂L₄ assemblies (Figure 1).^[10]

Figure 1.

Schematic representation of different types of assemblies whose stability was greatly enhanced by changing the metal (ligands displayed in light blue, metal centres as spheres).

The power of the formation of coordination‐based supramolecular self‐assemblies is that it is based on reversible bonds and as such it proceeds through repair of wrongly formed bonds to produce the thermodynamically controlled desired product. This is particularly important if the assembly is based on a large number of components, which is the case for M₁₂L₂₄ nanospheres. The quest for more stable structures by using stronger bonds between the metal and ligand therefore comes with the challenge of selective formation of the thermodynamically most stable assembly, while preventing the formation of kinetically trapped species and undesired intermediates.^[6] Motivated by this challenge, we studied how robust Pt₁₂L₂₄ nanospheres can be generated based on Pd₁₂L₂₄ assemblies, taking both thermodynamic and kinetic arguments into account. We introduce the ligand flexibility parameter that can be used in the design, by which kinetic traps can be overcome.

Palladium‐based M_nL_2n assemblies are formed by complexation of a building block containing two pyridine moieties that are oriented with a certain dihedral angle. Four pyridine moieties coordinate to the metal site in a square planar geometry.^[11] This quantitively yields Pd_nL_2n spheres as the thermodynamic most favourable structure. This type of Pd_nL_2n assemblies were shown to provide highly versatile materials as they can be easily functionalised on both the outside (exo) or the inside (endo). Endo functionalisation can lead to preorganisation of up to 24 functional groups inside of Pd₁₂L₂₄ spheres, yielding ideal systems to study cooperativity effects.[^2c, ^2d, ^2e, ¹²] Through the implementation of different functional groups on the outside of Pd_nL_2n assemblies, functional materials and even gels can be prepared.^[13] Some of these structures were shown to be able to interact with biological components (such as proteins, enzymes and DNA),^[14] providing an opening for applications in biology. Given the huge variety of applications of Pd_nL_2n spheres, the extension to more stable analogues is desirable as it might lead to an even broader scope of applications. The most obvious strategy is the preparation of platinum‐based analogues, and indeed some examples have been reported.[^2d, ^2e, ^4b, ⁶, ^10a, ¹², ¹⁵] Next to Pt₆L₁₂, which was reported and crystalised earlier by Fujita et al.,^[6] we recently have shown that Pt₁₂L₂₄ assemblies are formed through a set of kinetically stable intermediates with the general formula Pt_nL_2n for n=6–11 (Figure 2).^[15a] Using only the design principles developed for palladium‐based assemblies, that is taking the dihedral angle between the pyridine donors as a guide for the prediction of the formation of the nanosphere, was insufficient for platinum nanospheres. Building blocks that were predicted to provide Pt₁₂L₂₄ nanospheres yielded a mixture of different assemblies such as Pt₈L₁₆, Pt₉L₁₈ and the desired Pt₁₂L₂₄ assembly, even when reacted at high temperature (150 °C). The kinetically trapped states were too stable and could not be converted to the desired Pt₁₂L₂₄ nanosphere. These trapped states were destabilised by introducing a bulky group or a positively charged moiety at the endo site of the building block (Figure 1), which resulted in selective formation of the Pt₁₂L₂₄ nanospheres. Based on the successful destabilisation strategy of intermediates, we were motivated to develop a general ligand approach using design principles, rather than functionalisation, to achieve selective formation of self‐assemblies with strong metal‐ligand bonds.

Figure 2.

Example of trapped Pt₈ L^OMe ₁₆ and Pt₉ L^OMe ₁₈ and the averaged dihedral angle requirements.

Results and Discussion

During the formation of platinum‐based Pt₁₂L₂₄ assemblies, some Pt_nL_2n assemblies for n=6–11 have been recently identified as trapped intermediates.^[15a] A comparison of the modelled structures Pt₈ L^OMe ₁₆, Pt₉ L^OMe ₁₈ and the desired Pt₁₂ L^OMe ₂₄ assembly shows a difference in dihedral angle of the ditopic ligand building block for various structures that deviates from the natural ligand dihedral angle (Figure 2). Whereas the dihedral angle between the pyridine donors is close to the preferred 120° arrangement in the free building block and in the Pt₁₂L₂₄ assemblies, it is distorted to 110° or 115° for the analogue Pt₈L₁₆ and Pt₉L₁₈. Destabilisation of these intermediate structures can therefore be achieved by increasing the deformation energy required to bend the building block from the preferred 120°, in a similar manner as reported previously for Pd₁₅L₃₀ assemblies.^[11b] For this reason, we propose not only to calculate the dihedral angle, but also the flexibility range of a building block, which we define as accessible dihedral angle window within 2 kcal/mol energy penalty. It is anticipated that rigid building blocks, with a small flexibility range, will destabilise the intermediates to a larger extend. To investigate this, the dihedral angle and flexibility range for three electronically similar building blocks were evaluated using DFT (Section SI6): the acetylene‐linked L^OMe building block, a building block that has no linker in between the central benzene ring and the pyridine donors (L^SOMe ) and a building block with a 1,4‐phenylene linker in between (L^BnPy ). The flexibility range of L^OMe is within 2 kcal/mol from 95°–146° (Figure 3). In comparison to that, the two other building blocks L^BnPy and L^SOMe allow only access to dihedral angles ranging from 106° to 133° (145°; Figure 3). We therefore identified L^SOMe as the most rigid building block and L^BnPy as moderately rigid and L^OMe as flexible.

Figure 3.

Structure of the building blocks investigated. Energy surface for vertical bending of the pyridine donors computed by relaxed scan of the dihedral angle at the B3LYP/def2TZV level of theory.

To evaluate how the flexibility range influences the relative stability of the Pt₁₂L₂₄ compared to previously found intermediates Pt_nL_2n for n=6–11, modelling was performed using molecular mechanics (Section SI6). The thermodynamically most favoured structure was found to be the Pt₁₂L₂₄ assembly in all cases. Because the thermodynamic energy of different sized assemblies is relatively similar for the flexible building block L^OMe selective formation of the desired assembly might be not possible under thermodynamic control (as supported by previous experiments).^[15a] The 1,4‐phenylene linked building block L^BnPy shows a similar energy profile as the ethynyl linked building block L^OMe , with even a smaller difference between differently sized assemblies (Figure 4). This building block was found thus to be not suitable for selective sphere formation as similar results as obtained for L^OMe are expected. For the most rigid building block L^SOMe , the relative energy of small sized spheres for n=6–11 is significantly higher than for the Pt_n L^OMe _2n analogues (Figure 4). In the case of L^OMe and L^BnPy , the Pt₉L₁₈ spheres are energetically close to the Pt₁₂L₂₄ assembly with a difference of 1.5–1.6 kcal/mol. In contrast to that, for L^SOMe , the Pt₁₁L₂₂ assembly is energetically the second most favoured structure, with a difference of 6 kcal/mol with respect to the Pt₁₂L₂₄ assembly. Having demonstrated on the basis of theoretical models that small sized assemblies are thermodynamically destabilised for the rigid L^SOMe building block due to a higher deformation penalty, we decided to build on these promising results in the context of selective Pt₁₂L₂₄ sphere formation in subsequent experimental investigations.

Two building blocks L^SOMe and L^BnPy were investigated experimentally, and compared to sphere formation experiments with L^OMe . Sphere formation experiments were performed by mixing 1 equiv. of building block with 0.6 equiv [Pt(BF₄)₂(MeCN)₄] in acetonitrile, stirring the solution at 150 °C for a certain amount of time. Sphere formation experiments under these conditions using L^BnPy led to formation of ill‐defined material with broad ¹H NMR signals (Figure S34 in the Supporting Information). Mass spectroscopy (MS) analysis showed no presence of spherical structures, in line with formation of dominantly oligomeric or polymeric material. In a similar experiment using L^SOMe , a downfield shift of the pyridine protons of L^SOMe and a general broadening of the ¹H NMR signals were observed, in line with cage formation (Figure 5, below). To obtain insights into the structures formed for L^SOMe , the reaction mixture was analysed by high‐resolution electrospray ionisation mass spectrometry (HRMS‐ESI). The sample shows characteristic signals attributed to [Pt_n L^SOMe _2n ]^x+ for n=8–12 with matching isotopic pattern, indicating that a mixture of sphere structures has been formed under these conditions (Figures S14 and S15, calibration Section SI7). The relative intensity of unique signals attributed to the five assemblies is similar to what has been observed for L^OMe (Figure 5, below). Characteristic differences in comparison to the more flexible L^OMe ligand are that the rigid building block L^SOMe yields more of the larger spheres (such as Pt₁₀ L^SOMe ₂₀ and Pt₁₁ L^SOMe ₂₂) and none of the small assemblies (such as Pt₆ L^SOMe ₁₂; Figure 5, below). This observation is in line with the hypothesis that more rigid linkers have higher deformation energies to form the smaller spheres, thereby destabilising these assemblies. As the calculations predict a large thermodynamic difference between the Pt₁₂ L^SOMe ₂₄ nanosphere and the smaller sized assemblies, these smaller structures most likely represent kinetically trapped species (Figure 4). A selective formation of the Pt₁₂ L^SOMe ₂₄ nanosphere can therefore be possible by using conditions in which the conversion of smaller structures to the Pt₁₂ L^SOMe ₂₄ nanosphere is kinetically possible. This was explored by changing the reaction conditions such as temperature, reaction time, solvent and the use of additives. When sphere formation was carried out at higher temperature (e. g., at 170 °C; Figures S12 and S14), the signals in ¹H NMR spectrum of the reaction mixture were broader and analysis of the mixture by in HRMS‐ESI indicated that the Pt₁₂ L^SOMe ₂₄ sphere was present in lower concentration, suggesting decomposition of the structures before the thermodynamic equilibrium has reached. Longer reaction times at 150 °C did not change the distribution (up to 10 days). Also sphere formation experiments carried out in other solvent showed to be unproductive due to the instability of the solvents at high temperature (DMSO, DMF) or just because the solvent did not change the ligand exchange dynamics (MeNO₂, CDCl₃). As shown previously, platinum‐nitrogen bonds can be destabilised by using UV radiation.^[16] Unfortunately, due to a conjugation of the pyridine donors to an aromatic ring, alterative decay pathways without breaking the platinum‐pyridine bond are possible. Because we did not want to change the ligand design, an alternative strategy was envisioned based observations made for palladium‐based Pd₁₂L₂₄ nanospheres. These types of structures appear to show poor stability in the presence of biologically relevant components found in the cytoplasm of living cells, such as amines, thiols, acids and halides.[⁷, ¹⁷] These biological substrates are able to coordinate to the palladium centre and when present in higher quantities they can destroy the self‐assemblies. Whereas this property is typically not desired for palladium assemblies, when used as additive they may just labilise the platinum–nitrogen bond to the extent that the kinetic intermediates can be converted to the thermodynamic product. We therefore anticipated that the same components relevant to living cells that destroy palladium‐based self‐assemblies might enhance the dynamicity of platinum‐assemblies when added in catalytic amounts, thus potentially driving the selectivity towards the desired thermodynamically favoured structures.

Figure 5.

Top left: General conditions of sphere formation using L^OMe and L^SOMe. Bottom left: Distribution of different sized assemblies for the ligands determined by ESI‐MS analysis. Right: Representative ¹H NMR spectra (in MeCN) of L^SOMe sphere formation using different types of additives (7 mol % with respect to Pt). All sphere formation was performed at 150 °C except for the literature TFE/DMSO system, which was performed at 85 °C.

Figure 4.

Schematic representation of different types of Pt_nL_2n assembly and relative energies of different sized Pt_nL_2n assemblies for L^OMe,^[x] L^SOMe and L^BnPy.

We included in our studies the effect of water on the self‐assembly process as well as typically found biological functional groups (amine, carboxylic acid, imidazole, chloride, thioI) in catalytic amounts (Figure 5). As a reference we also included the previously reported TFE/DMSO system that was also reported to be a good combination for the preparation of Pt‐based assemblies.^[6] The additive must labilise the metal‐ligand bond (to a larger extend than the solvent), but should not coordinate too strongly as then it becomes competitive for pyridyl coordination, leading to ill‐defined materials. First, the effect of the water content was studied. After heating the ligand L^SOMe and platinum precursor at 150 °C for 4 days in [D₃]MeCN/D₂O (9 : 1), free ligand was still in the solution (as indicated by the ¹H NMR; Figure 5). Apparently, the presence of water hampers sphere formation, and as such all other reactions were performed in high pressure tubes under N₂ with dry degassed solvent. The sample prepared in TFE/DMSO at 85 °C was precipitated in Et₂O and redissolved in CD₃CN for further analysis. The TFE/DMSO sample showed a similar broad ¹H NMR spectrum as the sample prepared in pure MeCN (Figure 5) indicating the presence of oligomeric material in solution. Furthermore, HRMS‐ESI analysis of the TFE/DMSO sample showed no typical signals associated with nanospheres or kinetically trapped intermediates (Section SI7 and Figure S19). Reactions performed in MeCN at 150 °C in the presence of catalytic amounts of imidazole, amine or thiol displayed ill‐defined ¹H NMR spectra with some free ligand still present (e. g., Figures 5 and S23). Only small amounts of Pt₁₂ L^SOMe ₂₄ nanospheres and kinetically trapped intermediates were formed according to the HRMS‐ESI spectra taken from the samples, in a similar ratio as in the absence of these additives. The rather broad ¹H NMR signals together with the low intensity of spherical compounds in HRMS‐ESI suggest that under these conditions also some ill‐defined oligomers are formed.

In contrast, reactions carried out in the presence of chloride or acetic acid as additive resulted in much better resolved ¹H NMR spectra of the mixture (Figure 5). HRMS‐ESI analysis of the resulting solutions showed that the desired Pt₁₂ L^SOMe ₂₄ nanosphere was formed in high selectivity (Figure 6). TBACl addition showed the best resolved ¹H NMR spectrum with well‐defined signals and excellent selectivity towards the Pt₁₂ L^SOMe ₂₄ assembly based on HRMS‐ESI, ¹H NMR and DOSY (Figure 6 and S49). Whereas signals that are attributed to the β‐pyridine proton and all central benzene protons (Figure 5, right, b–d) are well resolved, the α‐pyridine protons (Figure 5, right, a) are split into multiple sets of doublets. Because no other species could be identified using HRMS‐ESI, we attribute this splitting to different rotation orientations of the α‐pyridines around the platinum centre (Figure S11 and ref. [18]). Interestingly, high purity nanospheres can be formed by either using TBACl or alkyl acids. Analogues experiments using the more flexible L^OMe in the presence of catalytic amounts of TBACl showed not much difference in the distribution of sphere products that were formed. The ¹H NMR spectra were identical as the experiment in absent of additives and HRMS‐ESI analysis showed no significant increase in selectivity for the Pt₁₂ L^OMe ₂₄ assembly (Figure 5). The difference in relative energy between the different nanospheres based on L^OMe is sufficient for palladium‐based spheres to generate pure Pd₁₂L₂₄ spheres. This similar energy difference observed for the platinum‐based analogue does not lead to pure Pt₁₂ L^OMe ₂₄ sphere formation, even if chlorides additives are used. This suggests that labilisation by additives alone is not sufficient for these structures to convert the kinetic traps into the Pt₁₂ L^OMe ₂₄ sphere, but that the thermodynamic destabilisation of the intermediates is required too. This can be achieved by using rigid ligands as demonstrated with L^SOMe .

Figure 6.

Optimised conditions for the synthesis of exo‐functionalised Pt₁₂L₂₄ spheres. ¹H NMR of the assembly and the applied building block L^SOMe. HRMS‐ESI of the Pt₁₂L^SOMe ₂₄ assembly, top: obtained spectra, bottom: simulated patterns.

To evaluate the functional group‐independency of this combined method of chloride‐catalysed assembly of Pt₁₂L₂₄ cages with rigid ditopic ligands, we applied L^Fu (nonfunctionalised), L^PEG (exo‐functionalised) and L^SOBn (endo‐functionalised; Figure 7) as building blocks in sphere formation. Sphere formation using catalytic amounts of TBACl in [D₃]MeCN and high temperatures (150 °C), yield for both rigid building blocks L^Fu and L^PEG pure Pt₁₂L₂₄ nanospheres as judged by the well‐resolved ¹H NMR spectrum (Figures S25 and S29). Also, HRMS‐ESI confirms the high selectivity for the desired Pt₁₂ L^Fu ₂₄ and Pt₁₂ L^PEG ₂₄ nanospheres as signals corresponding to other spheres were absent (Figures S28 and S30).

Figure 7.

Left: Selective sphere formation with 7 mol % TBACl using differently functionalised building blocks. Right: Formation of self‐assemblies using different linker core structures.

In order to introduce endo functionalities on spheres which are made up from building blocks without any linkers between the central benzene ring and the pyridine donors, mixing of two building blocks, one without any endo functionalisation and one with endo functionalisation is required. When the L^SOBn building block was mixed in a one‐to‐one ratio with L^SOMe , selective formation of the Pt₁₂ L^SOMe _x L^SOBn _24−x assemblies was achieved based on ¹H NMR and HRMS‐ESI analysis (Figures S31 and S32). The HRMS‐ESI show a statistical distribution of the different spheres (Pt₁₂ L^SOMe _x L^SOBn _24−x ) in agreement with a system that shows no self‐sorting (Figure S33). Importantly, also for these nanospheres the protocol in the presence of TBACl additive leads to pure Pt₁₂L₂₄ nanosphere formation, and none of the smaller spheres were observed in the MS spectra. These experiments provide evidence for the functional‐group independency of the design principles. It is important to mention here that the general applicability of this methodology is only ensured if an applied building block posses the following characteristics: 1) it allows solubility of intermediates and the resulting sphere in MeCN, and 2) it destabilises the oligomers and kinetic traps to the same or higher extent as L^SOMe . The methodology works worse if one or both of these criteria are not fulfilled. When we attempted sphere formation under identical conditions as described for L^SOMe with L^Naph (Figure 7; which does not dissolve well in MeCN and leads to precipitation of some intermediates), a lower selectivity was achieved for the desired Pt₁₂ L^Naph ₂₄ assembly (80 % as judged by MS analysis, Figure S37). When L^Thio was used for sphere formation (Figure 7; which energetic profile was not considered herein), selective formation of the desired Pt₂₄ L^Thio ₄₈ was achieved as judged by MS analysis (Figures S40 and S41). However, the relatively broad ¹H NMR spectrum (which is in accordance with analogous previously reported Pd₂₄L^Thio ₄₈ assemblies^[11e]) does not allow for judgement of the bulk purity (Figure S38). Even though, the presence of oligomeric material cannot be excluded, the desired Pt₂₄ L^Thio ₄₈ assembly was formed selectively as judged by MS analysis. A general applicability for other ligand systems requires in general in‐depth analysis of the energetic states and design that allow good solubility. However, with the L^SOMe type presented herein, any functionality can be placed on the in‐ or outside of Pt₁₂L₂₄ assemblies, providing an advantage of this method over previously reported design principles.

With a series of Pt₁₂L₂₄ nanospheres in hand made based on a new protocol, their kinetic stability was investigated. Because better solubility of the building block and the resulting spheres was expected for L^PEG , we chose the nanosphere based on this building block for the stability experiments. A set of different substrates (Figure 8), which represent the additives used to increase dynamicity, were added to a solution containing either Pt₁₂ L^PEG ₂₄ or Pd₁₂ L^PEG ₂₄. The solutions were stirred for 24 h at room temperature and the stability was studied using ¹H NMR. Signals corresponding to the palladium‐based systems disappeared and the signals of the free building block L^PEG appeared under all applied conditions (Figure S54). As such, the Pd₁₂ L^PEG ₂₄ nanosphere is not stable in presence of any coordinating substrates. In contrast, the platinum spheres remained intact in presence of the additives. The peaks corresponding to the Pt₁₂ L^PEG ₂₄ nanosphere in the ¹H NMR did not change, and signals that are typical for the free building block were not visible (Figure S55). As such platinum‐based spheres are kinetically stable in the presence of base (DMAP), acetates (benzoic acetate), amines (benzylamine) and halides (NaCl in D₂O) at room temperature, but the same compounds can facilitate cage formation at 150 °C as they somewhat labilise the inert platinum‐pyridyl bond.

Figure 8.

Stability studies performed on Pt₁₂L^PEG ₂₄ and Pd₁₂L^PEG ₂₄. Stability was measured by using ¹H NMR spectroscopy after 24 h.

Conclusion

In summary, we have demonstrated the successful preparation of robust platinum‐based nanospheres by using a new protocol. The challenge in making M₁₂L₂₄ platinum spheres is that kinetic trapped intermediates can form, prohibiting the formation of pure structures. Such unwanted structures should be avoided. We therefore introduce here the flexibility angle, next to the previously introduced dihedral angle for palladium spheres, to allow the rational design of ligand building blocks that form Pt₁₂L₂₄. For some ligands that differ in flexibility range, possible intermediates were computationally analysed to confirm that the rigidity of the applied building blocks leads to larger energy differences between the thermodynamically most stable Pt₁₂L₂₄ structure and the smaller intermediates. Under standard conditions a mixture of spheres was still formed, thus indicating that the strong metal‐ligand bond does not allow the repair mechanisms required for the formation of the most stable structure. These bonds can be labilised by the presence of additives such chloride and alkyl acids, converting the kinetically trapped intermediates to the desired Pt₁₂L₂₄ structures. This was demonstrated for a series of rigid ligands with a small flexibility range. Whereas these additives lead to sufficient dynamicity at elevated temperature to form the thermodynamically most stable structure, the formed assemblies are kinetically stable in presence of these additives at room temperature. The possibilities of how such coordinating additives can influence the outcome of the formation of self‐assemblies kinetically together with thermodynamic tuning by using computational investigation yields an ideal starting point for the investigation of other self‐assemblies for which enhancement of their stability is demanded.

Conflict of interest

The authors declare no conflict of interest.

Supporting information

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

Supporting Information

E. O. Bobylev, D. A. Poole, B. de Bruin, J. N. H. Reek, Chem. Eur. J. 2021, 27, 12667.