1 Introduction

Metal alkoxides are salts of metal cations and alkoxide anions, RO, derived from alcohols that are extremely weak organic acids [1, 2]. This results in strong Brönsted and Lewis basicity of this class of compounds broadly used as basic catalysts [3]. The extreme basicity of the alkoxide ligands leads in many cases to formation of oligonuclear aggregates even for homoleptic alkoxides, i.e. derived from solely one kind of RO ligand.

Majority of oxo-alkoxides, where part of the ligands are replaced by an even more basic oxide ligand, O2−, are complexes comprising several metal centers. Major input into bonding in these compounds is, just as in the structures of metal oxides, done by electrostatic interactions between cations and the oxygen atoms, permitting thus to consider these molecular aggregates as fragments of oxide structure terminated by alkyl groups [4].

Another factor shaping the oligonuclear metal alkoxides is the general trend for minimization of surface energy, leading to spheroidal or ellipsoidal topology [5]. It is rather important to note that the structures of metal (oxo)alkoxides are very often closely resembling those of polyoxometalate ions (POMs), following the Pauling’s principles of mineral oxide network construction [6] and obeying the minimum surface energy trend. Oligonuclear metal complexes are often called “clusters”, but this term is not correct in applying to oligonuclear species not featuring metal-metal bonds [7] such as metal alkoxides and POMs and thus a new term Metal-Oxo-Paperbags (MOP) has been established [8] and will be used further in this review. MOP resemble metal oxide nanoparticles (NP) and at present are probably most interesting as molecular level defined models of NP surface and reactivity, including redox properties, incorporation of single atom catalytic centers, ligand attachment to surface, ligand influence on the bandgap structure etc.

2 Analysis and discussion

2.1 General structural principles

The structures of oligonuclear alkoxide complexes (OAC) and MOP in general can be seen in the light of Bradley’s structure theory [9] and Pauling’s principles of mineral construction [6], both underlying the importance of stable coordination polyhedrons, resulting from critical radius ratios between cation and surrounding anions. The other important point identified by Pauling is the need to maximize the distance between highly charged cations in the structure, resulting in lower stability of edge sharing fragments compared to corner sharing and even less – for the face sharing ones. In fact, the face sharing in metal alkoxides has only been proved to exist in true clusters containing metal-metal bonds, such as, for example, Re2O3(OMe)6 [10] or ReMoO2(OMe)7 [11] (Fig. 1A, B). The other trend is, of course, the densest possible packing of metal cations and ligands in the structure [12]. This results in two octahedra sharing an edge, a M2O10 core, being the most common structure for binuclear OAC, exemplified by [M(OiPr)4(iPrOH)]2 (M = Zr, Hf, Sn(IV), see Fig. 1C [13, 14]) or M2O2(OMe)8 (M = Mo [15], W [16], see Fig. 1D). Three octahedrons sharing edges in a triangle, a M3O13, is the most common building block in both OAC and POM structures and is often observed in oxo-alkoxides of Ti, Zr, Hf [17, 18] (Fig. 1E).

Fig. 1
figure 1

Molecular structures of Re2O3(OMe)6 (A) [10], ReMoO2(OMe)7 (Re in blue and Mo in turquoise, B) [11], [M(OiPr)4(iPrOH)]2 (M = Zr), Hf, Sn(IV) (C) [13, 14], M2O2(OMe)8, M = Mo [15], W [16] (D) and Ti3O(OiPr)10 (only central C-atom provided for clarity, disordered iPr bridging group not displayed, E) [17]

Its extension into an infinite flat layer analogous to the layered double hydroxide (LDH) structure is the case of M(OR)2 with primary alkoxide ligands, where low-charged and relatively small (!) M2+-cations can occupy all the neighboring places [2] (Fig. 2).

Fig. 2
figure 2

Molecular structure of the Ni(OH)2 LDH complex with Nb6O198− anions [19]

The molecular construction of LDH formed on hydrolysis of nickel precursors has been confirmed in the work of Ira Weinstock where it was complexed on the surface by POM species [19]. If we look closer at general trends in OAC structure build-up, it can be noticed that the M3O13 triangle is typical as a building block for high-valent cation structures at lower hydrolysis ratios (oxo-ligand content) in Zr(IV) and Hf(IV) [20] (Fig. 3A), Nb(V) such as in oxo-alkoxides derivatives of linear chain alcohols, Nb8O10(OR)20 [21, 22] (Fig. 3B) or Mo(VI) and W(VI) alkoxides and Lindquist and Keggin POMs [23, 24] (Fig. 3C). Additional factor seen apparently in the species with relatively low hydrolysis ratio/oxo-ligand substitution is the already mentioned trend to minimization of surface energy and steric hindrance, leading to placement of the alkyl chains on the surface of the aggregate, converting it into a hollow spheroid/ellipsoid. One of the most impressive examples of this trend is revealed in the structure of (H6[Ti423-O)60(OiPr)42(OH)12)] [25], which can be described as a layer of anatase structure rolled in into a sphere (Fig. 3D). At higher hydrolysis/oxo-substitution ratios the structures “collapse” and the core becomes more like a 3D oxide structure [23, 26]. For titanium near-complete hydrolysis at room temperature or in boiling water results in formation of uniform nanoparticles with most commonly anatase core structure that can be confirmed by either electron [27] or even X-ray diffraction [28].

Fig. 3
figure 3

Molecular structures of Zr4O(OnPr)14 (A) [20], Nb8O10(OR)20 (B) [21, 22] only first carbon atom of the ligand displayed for clarity, Lindqvist-W5TiO18(OiPr)3− anion [24] (C) and (H6[Ti423-O)60(OiPr)42(OH)12)] (D) [25] (no carbon atoms were actually securely located in the reported model)

Less charged cations permit denser packing in the core with square pyramidal M5O building blocks for trivalent metal cations such as Al3+ [29], Fe3+ [30] and even Rare Earths (RE), Ln3+, as in Ln5O(OiPr)13 [31] (Fig. 4A). It is interesting to note that this kind of packing looks as resembling that in NaCl structure type, not related to the oxide or hydroxide structures for these elements. For large cations of divalent elements the structures of OAC can be resembling that of the corresponding oxide for Ca2+ as in [Ca64-0)23-0Et)4({0Et})4]·14EtOH, featuring a NaCl type packing (as in CaO, see Fig. 4B) [32], or display a fluorite type packing with even larger cations such as Sr2+, Ba2+ and even Sn2+ (Fig. 4C) [33].

Fig. 4
figure 4

Molecular structures of Fe5O(OEt)13 (A) [30], [Ca64-0)23-0Et)4({0Et})4]·14EtOH (B) [32]—only first carbon atoms of the ethyl group retained for clarity, and Sn6O4(OEt)4 (C) [33]

It can thus be concluded, that the analogy between the structures of oligonuclear alkoxides and those of the corresponding oxides not always can be traced. The case of lucky coincidence, serendipity, is observed, however, for one of the most important oxide materials, titanium dioxide (as anatase), opening a perspective for experimental molecular modeling of its surface and even bulk properties [34].

2.2 Modeling of redox properties

Visualization of redox transformations of metal oxides was actually the topic of the very first publication, formulating the concept by Malcolm Chisholm et al. [4]. The focus was on understanding of transformations in metal-metal bonding, especially the fate of a localized multiple bond [35] (see Fig. 5).

Fig. 5
figure 5

Transformations of a triple W≡W bond in reactions of the tungsten alkoxides (after [35]), M = Mo, W; X = CO, HCCH; L = Py, HNMe2

An exciting challenge would be to model the redox properties of a metal oxide NP, applying an OAC as model. Attention in the recent years has been set on protein corona phenomenon [36] and oxidation of specific amino acids. Possible hinder lies in facile oxidation of the alkoxide ligand itself, reducing the choice of possible meaningful substrates and the sensitivity of alkoxides to hydrolysis in the view that the reactions of interest occur usually in aqueous media. The research in this direction is on-going in our group and at least POM species also belonging to the MOP family have already successfully been applied for visualizing of, in particular, specific tryptophan oxidation pathways [37]. It was possible to demonstrate that both POM and NP with relatively high oxidation potentials were reacting with simultaneous proton and electron transfer, traced via interactions in the crystal structure between the tryptophan molecule and the POM and confirmed by a number of spectroscopic techniques and theoretical calculation (see Fig. 6) [37].

Fig. 6
figure 6

Fragment of a crystal structure of POM-tryptophan complex, indicating the tracks of electron and proton transfer (A) and the discovered mechanism of direct specific tryptophan oxidation by NP (B) [37]

2.3 Modeling of single atom catalytic centers

Strong attention during the recent years has been attracted by the effects of single atom catalysts implanted on the surface of oxide matrices. Bringing understanding into placement of the doping atom and construction of its coordination sphere is challenging. Commonly applied techniques are indirect, using either photo luminescent characteristics of the inserted atom (being limited to structures with relatively good crystallinity/order and focusing very much on doping with RE elements [38]), or X-ray absorption spectroscopy approaches, such as EXAFS, requiring actually an in advance known reliable model for successful treatment of the data [39].

Principal breakthrough in providing such models has been made by the group of Dominic Wright at the University of Cambridge. They investigated interaction between transition metal halides and titanium and zirconium alkoxides under solvothermal conditions, leading to multiple examples of surface-implanted titanium oxide related OAC structures. The most typical representatives of these species featured halide terminated penta-coordinated both early (as Cr(III) and Mn(II) and late transition element (Co(II), Ni(II), Cu(II), Zn(II)) atoms either capping a Ti3O triangle as in [Ti4MO(OEt)15Cl] (M = Co, Zn, Fe, Cu, see Fig. 7A), or replacing one of the peripheral Ti atoms in the Anderson POM like Ti7O4(OEt)20 resulting in [Ti7MIIO5(OEt)18Cl2)] (MII = Co, Fe, see Fig. 7B) structures [40].

Fig. 7
figure 7

Molecular structures demonstrating the [Ti4MIIO(OEt)15Cl] (A) and ([Ti7MIIO5(OEt)18Cl2)] (B) arrangement types [40], MII = Co, Ni, Fe, Zn. Only first carbon atoms of the ethyl groups are retained for clarity

The observed structures permit to explain the enhanced reactivity of the surface-implanted transition metal atoms in the catalysts, because they feature lower coordination and are terminated by a halide atom that should be easy to substitute in surface reactions.

More recent attempts to use RE nitrate hydrate salts, Ln(NO3)3·xH2O, for implanting RE elements showed strong dependence on the Ti : RE molar ratio and thermal conditions. At lower Ti : RE ratios (about 2 : 1 and just above) at room temperature the crystallization of non-oxo species [Ti2Ln(OiPr)9(NO3)2] took place [41]. This structure is just the result of replacement of one Ti atom in the triangular aggregate, but with an alkoxide, not oxide ligand for charge compensation. Face-sharing octahedra with two Ti or zirconium atoms are stabilized in the presence of electropositive cations with large number of representatives of this structure type reported in literature [2]. More excitingly, at higher Ti : RE molar ratios, the result of interaction was dependent as noted above on the temperature conditions: at room temperature a salt of oligonuclear cation and anion, [Ti12(OiPr)183-O14)][La3(NO3)11(HOiPr)6], was forming with a large yield, while on reflux conditions in the parent alcohol, iPrOH, a substituted MOP with the formula [Ti11La(OiPr)173-O)14(NO3)2] was formed [42].

Getting insight into transition metal cation attachment to silica has also been investigated using alkoxide model species. In this case, the silsesquioxane molecules were used as models for silica surface and applied as ligands for metal alkoxides. Interestingly, it was discovered that even in the excess of such ligands the transition metal fragments did not dissociate to single atoms, but remained a small group of even only two cations [43, 44]. These observations provided new insights into possible mechanisms of action of peroxidation catalysts in double C = C bond degradation (see Fig. 8).

Fig. 8
figure 8

Formation and thermal behavior of alkoxytitanasilsesquioxanes. Reprinted with permission from [43]

2.4 Modeling of ligand surface attachment

Application of additional organic ligands in order to modify reactivity of metal alkoxides has gained popularity in 1980-es in connection to active development of metal oxide sol-gel for preparation of functional coatings and films [45]. Additional interest to these compounds emerged due to quest for single-source precursors of high-K dielectrics and High-Temperature Superconductors (HTSC), involving elements not producing soluble or volatile alkoxides, such as Cu, Pb etc. Sol-gel procedures to prepare HTSC and high-K dielectric layers were employing derivatives of other organic ligands than alkoxides for these elements together with alkoxides of Ti, Zr and RE elements, which resulted in isolation and structural characterization of a broad spectrum of so-called heteroleptic complexes [46]. Attachment modes of hetero ligands have in fact attracted a lot of attention from the very beginning, but were seen just in light of their binding into mixed-ligand complex molecules [47].

The interest in modeling the surface attachment of ligand to oxide NP emerged with growing applications of hybrid organic-inorganic materials and understanding of importance of insight into the activity of organic functions on the surface [48]. First steps were taken in the chemistry of silicon, where strong progress in characterization of oligomeric silsesquioxane compounds took place [49].

Most attractive goals in modeling with OAC turned to be the interaction of NP with biomolecules for, on one hand, biological and, specifically, drug delivery applications, and, on the other—understanding of possible interaction of NP with living cells and with viruses. Attachment modes and geometric parameters of bonding were obtained for many important classes of biological ligands. In particular, to understand the bonding and release mechanisms from titania NP the structures of titanium complexes with phenol family antibiotics like triclosan [50] were investigated. In order to understand self-assembly of surface-capped titania NP for encapsulation of time-temperature food indicators (Fig. 9A), the structures of carboxylate complexes were determined (Fig. 9B) [51]. Insight into bonding of oxy-carboxylate ligands such as lactate brought understanding for its stronger attachment to titania surface, hindering its interaction with proteins [52]. Among the most interesting results in application of OAC was discovery of the bonding mode of titania NP with proteins. It turned out that it occurs normally for not surface-protected TiO2 NP via inner-sphere complexation with a carboxylic residue in a side-chain of an alpha-helix and requires to be supported by several strong hydrogen bonds to protonated amino groups of basic side chain residues (see Fig. 10) [53].

Fig. 9
figure 9

The cartoon showing how long-chain heteroligands guide self-assembly of titania NP in an inverted Pickering emulsion (A) and the molecular structure of Ti12O12(OAc)6(OnPr)18 polyhedral view (B). Reprinted with permission from [51]

Fig. 10
figure 10

Model showing the docking of the metal-oxygen core of (H6[Ti423-O)60(OiPr)42(OH)12)] with the structure of nsp1 taken from [62]. The possible interactions are show in cyan: carboxylate of D48 with titanium, forming an inner sphere complex, and green: hydrogen bonding of OH-groups from the nanoparticle to E41, Q44 and H45. The amino acids of the other amides showing chemical shift perturbation are marked. Reprinted with permission from [53]

Very important class of biomolecules are those bearing phosphate residues that have extremely high affinity for metal oxide NP. Among examples of such classes are phospholipids, teichoic acid and nucleic acids, RNA and DNA. Interaction of phosphate residues with NP surface can effectively be modeled by investigation of molecular structures of complexes resulting from reactions of phosphonic acids with metal alkoxides. It was demonstrated that under near-neutral pH conditions the attachment of phosphate occurs in a tripodal fashion, binding to 3 different metal atoms, while under extremely acidic conditions one P-O-Ti bond can be broken due to protonation, resulting in a dipodal attachment. Connection of a phosphate/phosphonate residue via only one oxygen atom has never been observed in the OAC models [54, 55].

2.5 Ligand effects on bandgap and hinders in modeling photocatalytic activity

One of the most exciting properties of metal oxide NP is their ability to act as photo catalysts producing reactive oxygen species for a broad spectrum of applications. The most addressed photo catalyst is titania, TiO2, in its anatase phase [56]. It has, unfortunately, just a little too broad bandgap about 3.2 eV, which limits its efficiency to successfully using only the UV component of the visible light. It has been demonstrated, however, that ligands adsorbed/bound on the surface of titania NP can significantly change/reduce the bandgap. This can have many implications – from the use of surface-capped oxide NP in solar cells [55] and in photocatalytic water splitting [57] to possibilities in regulation of drug release from NP-containing composites [58]. OAC bearing relevant ligands can, via investigation of their optical characteristics, shed light on the influence of the attached molecules on the energy states in related titania NP. An important contribution to this field has been made by the group of Philip Coppens [34].

It is however, very important to bear in mind that OAC are not clusters, but paperbags. Attempts to measure their photocatalytic activity or adsorbent properties in aqueous medium result normally in measuring the activity of titania NP, to which they are converted by hydrolysis [55]. The literature contains quite a number of misinterpretations of this kind [59,60,61].

3 Conclusions

The idea of using metal alkoxides as models for metal oxides has through the four decennia after being originally formulated, developed into a powerful tool in modeling different characteristics of metal oxide NP. It offered insight into their bulk and especially surface structure, redox properties, surface reactivity etc. by visualizing the ligand attachment, and even providing clues to their electronic structure. A lot in this development has been driven by materials technology and especially nanotechnology.