A method for structure prediction of metal-ligand interfaces of hybrid nanoparticles

Hybrid metal nanoparticles, consisting of a nano-crystalline metal core and a protecting shell of organic ligand molecules, have applications in diverse areas such as biolabeling, catalysis, nanomedicine, and solar energy. Despite a rapidly growing database of experimentally determined atom-precise nanoparticle structures and their properties, there has been no successful, systematic way to predict the atomistic structure of the metal-ligand interface. Here, we devise and validate a general method to predict the structure of the metal-ligand interface of ligand-stabilized gold and silver nanoparticles, based on information about local chemical environments of atoms in experimental data. In addition to predicting realistic interface structures, our method is useful for investigations on the steric effects at the metal-ligand interface, as well as for predicting isomers and intermediate structures induced by thermal dynamics or interactions with the environment. Our method is applicable to other hybrid nanomaterials once a suitable set of reference structures is available.

Apart from the rather specific and empirical nature of the approach, perhaps its main limitation is the fact that it requires the positions of the Au atoms in the NP to be known prior to a search. In addition the method relies on a reference set of experimentally refined ligated-NP structures. In other words the searches only consider different ligand positions on known NP core structures biased by experimental knowledge of similar systems. Thus the method is limited in the scope of its predictive potential. Are their many cases where the positions of the metal atoms are well known in detail but where the X-ray positions of the ligands are not AND for which a number of experimental structures of ligated NPs are already known? If such scenarios are not common what is the main application of the method? The restriction of using a fixed known NP core also greatly limits the configurational search space and thus wonders if a simple force field based global optimisation approach of for finding low energy ligand binding positions on a fixed NP structure would be more appropriate and direct. What is the advantage of the current method over such direct, less empirical, approaches? The authors claim that their empirical ranking of obtained structures correlates well with that of energetic stability but that confirmation of this is outside the scope of the current work. I strongly feel that to help provide evidence that the ranking employed is reliable this correlation should be reported in the present work.
Predictive structure searches have been widely employed in many fields of chemistry and nanoscience. Although, the application of a bottom-up search algorithm for ligated NPs appears quite novel, the reported method is limited to known NP structures and, for each search, relies on a number empirical principles and tailored parameters. Although perhaps useful for the reported system, it is not convincing that the approach is as general or predictive as suggested in the title of the manuscript. Considering the above, I feel that the current manuscript would be better suited for publication in a more specialised journal.
Reviewer #2 (Remarks to the Author): This report describes the use of computationally inexpensive algorithms for predicting the structure of gold monolayer protected clusters (MPCs) based on a chemical basis set of known structure solutions. The authors were able to successfully generate the ligand structure of five clusters capped with different thiolate ligands by starting with the known gold core structure; this included clusters ranging from 36 to 279 gold atoms. The structural features of double and/or triple bridged Au(I)thiolate (staples and hemirings) were observed in all of the MPCs with single bridge ligands only being observed occasionally in some clusters. These results are expected for this class of MPCs and demonstrate the validity of their algorithms and choice of input parameters.
This work is of great significance as a means for general structure prediction, although it has only been demonstrated on Au-S based MPCs here. The adaptability of this method to include partial structure information from experimental data will be especially helpful in cases where total structure determination by XRD is challenging. Overall, this work is of the appropriate caliber and scope for publication in this journal.

Questions/Points of Clarification:
The algorithm success rate for Au38(SR)24 was dramatically higher than for any other cluster, even compared to Au36(SR)24, which used the same basis set (albeit a different ligand). Can the authors comment on this?
Page 7, Au36(SR)24 section: the ligand is listed as tert-butyl benzene (TBBT) instead of tert-butyl benzene thiol. I'm assuming this is a typo.
Page 10, Discussion: paragraph 2, Au-S "bolding configuration" -please define this, or is it supposed to be bonding configuration?
Page 14, 1. Local Environments: "extending the NN distances to 3-4 takes into account" I am unclear if this means 3-4 angstroms or if kNN = 3-4? Can you clarify this part?
Reviewer #3 (Remarks to the Author): The authors presented a computational method to predict the ligand-metal layers of thiolateprotected gold nanoclusters. A few successful examples were explored by the algorithm. It is always attractive to obtain the structure of a nanocluster without growing single crystals, however, I do not think the manuscript is suitable to publish on Nat. Comm., unless the authors can address the following concerns: 1. The manuscript focuses on the thiolate-protected nanoclusters, but in fact, other types of ligands, such as alkyne, phosphine, amine, pyridine, or a mixture of them, have also been used to protect the metal core. Since other types of ligands have very different coordination environment than thiolate (different staples, or no staples at all), I wonder whether the method presented here can be applied to other systems easily as the authors claimed.
2. In order to predict the structure of metal-ligand layers successfully, it seems that all gold atoms, including the ones that forms the staples, have to be fixed before adding thiolate ligands. However, the identification of surface gold atoms or even core gold atoms has already troubled the experimental research a lot. I think the significance is not as great as it looks, because people will still have to spend enormous amount of time to obtain the location of metal atoms.
3. There are several examples of polymorphism thiolate-protected gold nanoclusters, such as Au30, Au38, Au44, Au52. I think they might be better candidates to be explored theoretically, because they show that different types of thiolate ligands can cause different metal-ligand interface, as well as different core structures.
1. The manuscript by Malola et al. describes a method to search for likely thiolate ligand arrangements on gold nanoparticles (NPs). The work is well presented and, as far as I am aware, this is the first example of an attempt to do this. The algorithm employed bases its primary search for sulphur positions on comparison with a set of reference ligated NP structures. Second, it proceeds with a stochastic search restricted by chemical heuristics (e.g. coordination, atomic distances). Finally, the ligated NP structures produced are ranked according to two empirical criteria. The method is applied to five ligated Au NPs and the method is shown to be quite successful in finding structures close to the experimental structure, when the latter is known. This approach to structure prediction depends on a number of specific choices, which although perhaps chemically reasonable, are not clearly shown to be applicable to other ligated NP systems, as suggested. As such, the generality of the approach is not established. In fact, to show only five cases of a successful application of the method to one type of system does not provide a particularly large set of confirming examples for general application to Au-thiolate NPs. This potential lack of generality is underlined by the fact that each example uses specifically tuned parameters (e.g. error margins for distances and angles, number and type of structures in reference set) which are reported without justification. The dependency of the success of the method on such particular choices should be provided. 2. Apart from the rather specific and empirical nature of the approach, perhaps its main limitation is the fact that it requires the positions of the Au atoms in the NP to be known prior to a search. In addition the method relies on a reference set of experimentally refined ligated-NP structures. In other words the searches only consider different ligand positions on known NP core structures biased by experimental knowledge of similar systems. Thus the method is limited in the scope of its predictive potential.
Response: We agree that the predictive power of the method concerning "novel" metal-ligand interfaces structures is ultimately defined by the extent by which the already known experimental structures have been able to catch the essential local chemistry around the metal atoms and protecting ligands. We dare to argue that this most probably is the case at least in the best-studied gold-thiolate systems, where around 100 crystallographically determined cluster structures exist in the literature. 3. Are their many cases where the positions of the metal atoms are well known in detail but where the X-ray positions of the ligands are not AND for which a number of experimental structures of ligated NPs are already known? If such scenarios are not common what is the main application of the method? The restriction of using a fixed known NP core also greatly limits the configurational search space and thus wonders if a simple force field based global optimisation approach of for finding low energy ligand binding positions on a fixed NP structure would be more appropriate and direct. What is the advantage of the current method over such direct, less empirical, approaches?

University of Jyväskylä Faculty of Mathematics and Science
Response: Despite the great recent successes of atom-precise experimental structural characterizations of ligand-stabilized metal nanoclusters by single-crystal X-ray crystallography, there are still many cases (and more expected) where the atom-precise total structure determination, including the organic ligand layer, may not succeed. In fact, the initial motivation for us to start developing the method described in this work came from the collaboration with the group of R.D. Kornberg (Science 345, 909 (2014); ACS Nano 11, 11866 (2017); ACS Nano 11, 11872 (2017) Figures 4 and 6 show a strong correlation of the CSE-ranked best model structures to the true crystallographic structure of several gold and silver clusters, respectively, and Figure  7 shows that the best CSE-ranked model structures of Au36(SR)24 also have the lowest calculated DFT total energies. Please note that our definition of the CSE is straightforward to generalise to include other desired criteria, such as "closeness" of a simulated powder XRD function of a model structure to experimentally measured XRD data, etc.
5. Predictive structure searches have been widely employed in many fields of chemistry and nanoscience. Although, the application of a bottom-up search algorithm for ligated NPs appears quite novel, the reported method is limited to known NP structures and, for each search, relies on a number empirical principles and tailored parameters. Although perhaps useful for the reported system, it is not convincing that the approach is as general or predictive as suggested in the title of the manuscript. Considering the above, I feel that the current manuscript would be better suited for publication in a more specialised journal.
Response: As already discussed above, we have extended the analysis in the revised MS to include significantly more and different systems as compared to the original version of the MS. It should be noted that, in principle, our method is not limited to clusters or nanoparticles having metal-ligand interfaces. It can be used as well for planar metal-molecule interfaces, such molecular SAMs, or complicated 3D systems such as molecule-capped metal nanowires, metal-organic networks, etc.

Referee 2
1. This work is of great significance as a means for general structure prediction, although it has only been demonstrated on Au-S based MPCs here. The adaptability of this method to include partial structure information from experimental data will be especially helpful in cases where total structure determination by XRD is challenging. Overall, this work is of the appropriate caliber and scope for publication in this journal.
Response: We thank the referee for his/her positive evaluation of the significance of our work.