Coding the Assembly of Polyoxotungstates with a Programmable Reaction System

Chemical transformations are normally conducted in batch or flow mode, thereby allowing the chemistry to be temporally or spatially controlled, but these approaches are not normally combined dynamically. However, the investigation of the underlying chemistry masked by the self-assembly processes that often occur in one-pot reactions and exploitation of the potential of complex chemical systems requires control in both time and space. Additionally, maintaining the intermediate constituents of a self-assembled system “off equilibrium” and utilizing them dynamically at specific time intervals provide access to building blocks that cannot coexist under one-pot conditions and ultimately to the formation of new clusters. Herein, we implement the concept of a programmable networked reaction system, allowing us to connect discrete “one-pot” reactions that produce the building block{W11O38} ≡ {W11} under different conditions and control, in real time, the assembly of a series of polyoxometalate clusters {W12O42} ≡ {W12}, {W22O74} ≡ {W22} 1a, {W34O116} ≡ {W34} 2a, and {W36O120} ≡ {W36} 3a, using pH and ultraviolet–visible monitoring. The programmable networked reaction system reveals that is possible to assemble a range of different clusters using {W11}-based building blocks, demonstrating the relationship between the clusters within the family of iso-polyoxotungstates, with the final structural motif being entirely dependent on the building block libraries generated in each separate reaction space within the network. In total, this approach led to the isolation of five distinct inorganic clusters using a “fixed” set of reagents and using a fully automated sequence code, rather than five entirely different reaction protocols. As such, this approach allows us to discover, record, and implement complex one-pot reaction syntheses in a more general way, increasing the yield and reproducibility and potentially giving access to nonspecialists.


■ INTRODUCTION
The chemistry of molecular metal oxide polyoxometalate (POM) clusters is a rapidly expanding field because of the vast number of structures, the diversity of their properties, 1−5 and a wide range of potential applications. 6,7 However, the use of one-pot reaction methods 8 for the synthesis of POM clusters 8−11 presents an important problem because it is very hard to understand the crucial steps taking place in solution giving rise to the building blocks, both transient and isolatable, for the assembly of such clusters. In fact, there is currently no general route for exploring one-pot reactions, especially for molecular metal oxides in which real-time kinetics are hard to obtain by nuclear magnetic resonance (NMR) and mass spectrometry. All of these issues mean that it is virtually impossible to devise testable hypotheses regarding the assembly of giant clusters, especially because many of the assembly processes are controlled by complex nonequilibrium kinetics. 12−14 Indeed, it could be that this problem is even more daunting because the one-pot system effectively masks a vast and interconnected range of intricate discrete reactions that underpin the self-assembly process. Furthermore, these issues are also relevant in the solution assembly of other complex molecules such as coordination frameworks 15 and discrete giant metallo-supramolecular 16 architectures.
In this respect, we have recently described the potential for new discoveries by the programmed manipulation of the assembly of some common building blocks in solution 17−19 in an automated fashion. However, with the previous focus being on discovery, there is now a pressing need to develop, and validate, a well-defined platform that incorporates programming and feedback control. Such a setup would allow us to explore different kinetic regimes, reagent combinations, and the control of variables like pH, ionic strength, and redox potential. 20−22 Further, sequential, batch one-by-one conventional one-pot reactions for systematically exploring the POM systems 23,24 and screening the parameter space are very time-consuming and limited in scope, allowing the reactions to be probed only in a comparatively narrow time domain. Most importantly, it is practically impossible to keep a self-assembled system "off equilibrium" or to trap and utilize transient building blocks in batch synthesis.
However, networking reaction systems 25 applied in this case to the assembly of polyoxometalate clusters allow us to test the hypothesis that it is possible to "program" one-pot reactions 26 in both the time and space domains simultaneously (see Figure  1). In this way, the one-pot reaction compositions (r) can be integrated across several different reaction vessels (R N ), simultaneously allowing the convolution of reaction variables (r → R N ) as a function of fundamental programmable operations (Op) combining fully automated configurable reactors and externally attached in situ analytical techniques [pH, ultraviolet (UV), etc.]. Following these principles, the addition of reactants is denoted by S m , transferring reaction solutions between reactors by P N−N , and output of reaction solutions by O N according to the operation sequence (see eq 1).
This means that, by controlling the reaction variables, we should be able to devise new ways to explore and probe onepot reactions within networked reactor systems, ultimately leading to the discovery and programmable synthesis of molecules using algorithmic control. Indeed, continuous-flow systems have already been used to regulate reaction conditions under pressure, 27−29 and we have also shown how flow-based methods led us to trap a reactive intermediate responsible for the assembly of a gigantic molecular inorganic {Mo 154 } wheel cluster. 30,31 Because these flow reactor designs were first employed, we have expanded the concept to a new generation of automated digital programmable linear and networked reactor systems; these have led to the discovery and synthesis of gigantic macrocyclic POM architectures such as {Mo 96 } and {W 200 }. 26, 32 The purpose of this work was to develop a programmable system that allows the deconvolution of complex systems and compartmentalization of the fundamentally incompatible (with regard to the experimental conditions) individual one-pot reactions for the manipulation of the formed building block libraries, with the ultimate goal being the discovery of new reactivity, mechanism, and compounds. More specifically, we show how a four-connected networked reactor system (a 3,1-NRS) can be used to validate this approach by the synthesis of five different polyoxometalate clusters using a common set of reagents. We also show it is possible to control, in real time, the self-assembly of the common building blocks present in all five clusters using in situ analytical techniques that allow us to isolate compounds {W 11

■ RESULTS AND DISCUSSION
We investigated and integrated the synthesis of a set of isopolyoxotungstates (iso-POTs) 35 in a single automated system giving Na 12 Figure 2a), synthesized by a programmable pumping sequence, whose reaction conditions were transiently formed, and linked to yield the final cluster  The process starts from a virtual library of the simplest building blocks to form {W 11 } and {W 12 } subunits during the acidification process by means of programmable reaction conditions. Two configurations of the reactor system (b and c) correspond to the reaction connectivity of the filling and recycling steps in our approach, respectively. Also, the fundamental programmable operations (P N−N , S m , and O N ) are specifically assigned to each of the syringe pumps.

Inorganic Chemistry
Article compounds as outlined above. As described in the literature, 33,40 sodium tungstate dihydrate (Na 2 WO 4 ·2H 2 O) and sodium sulfite (Na 2 SO 3 ) solutions are required to synthesize these clusters using five distinct one-pot reactions. Using the NRS, we aimed to simplify syntheses of these compounds under (R 1 → R 2 → R 3 ) n operation or R 1→3 ( Figure  2b).
The experiments were performed in the NRS using three aqueous solutions (I 1 is 6 M HCl, I 4 is 2.4 M Na 2 SO 3 , and I 5 is 2.8 M Na 2 WO 4 ·2H 2 O) as the main reagents. Moreover, two other stock solutions of amine-based hydrochloride salts [I 3 is 1 M triethanolamine (TEA)·HCl or 1 M dimethylamine (DMA)· HCl] and potassium chloride salt (I 6 is 4.5 M KCl) were also prepared to extend the family to {W 36 O 120 } 25 and {W 11 O 38 } clusters, using the same networked system (see Table 1), and finally, a H 2 O reservoir (I 2 ) was attached to control the total volume of the final solution.
We wondered whether it is possible to take three separate one-pot reactions, which use same initial reagent composition but produce different cluster building blocks as they are being acidified at different rates. Indeed, we were able to exploit the fact that the {W 12 } cluster has a stability higher than that of the {W 11 } cluster in the presence of sodium cations, 40 and these polyoxometalate clusters are formed at different pH values as the {W 12 } cluster forms at pH >4 and the {W 11 } cluster forms at lower pH values. 39 Thus, by monitoring the pH and controlling the addition of acid, via simultaneous programmable transfer of reagents and acidification, we could observe a gradient of "local" pH changes according to the equation pH(R 1 ) < pH(R 2 ) < pH(R 3 ), which has been previously simulated for three interconnected continuously stirred tanks (see Figure S2). This approach allows us to alternate between near and far-from equilibrium thermodynamic states to determinate populations of the building blocks (BBs) in each reactor, and the hypothesis can be defined by the applied pumping pathways such as BB(R 1 ) → BB(R 2 ) → BB(R 3 ), according to the pumping regime between the reactors.
The key point in the synthesis of compounds 1−3 is controlled by the BB distribution, which is time-dependent, in each R 1→3 , which is related to the space-dependent (only in R 1 ) acidification operation (see Figure 3). Such compounds 1−3 were obtained only after a second transfer stage, named the linkage step in R 4 , where the pH successively decreases for each compound [1 at pH(R 4 ) 3.4, 2 at pH(R 4 ) 2.4, and 3 at pH(R 4 ) 2.2]. The isolation of 3 additionally requires the use of organic cations, in this case TEA·HCl. 34 Alternatively, the NRS has the ability to integrate more chemical components because multiple inputs can be attached to the sequentially connected one-pot approach (Figure 4a).
Then, the hypothesis is specifically formulated by following the BB populations: 1 and 2 by R 1 11 } is always expected because the acid is being added uniquely into R 1 as giving {W 11 } as the kinetically controlled cluster product. On the other hand, {W 12 } is always produced when the pH is >3.4 because of its high thermodynamic stability under these reaction conditions. To investigate the validity of our hypothesis, we performed a real-time monitoring in situ analysis (pH and UV) (Figure 4c) by sequentially ordered operations (see the diagram in Figure 4b). The pH in the three reactors (R 1 −R 3 ) was monitored in parallel by three pH sensors over the 40 min reaction mixture transfer process. Figure 4c shows that pH measurements indicate different local pH values in each reactor, with the same stepped profiles, but shifted in time only because of different acidification rates. The values of pH(R 2 ) and pH(R 3 ) can be monitored from the beginning of the process, while pH(R 1 ) can be measured only ∼30 min after initiation of the experimental procedure, because of the formation of the white precipitate that redissolves after further acidification. In sequence, the addition of acid stops and each of the R 1→3 reactors have different "local" pH values leading to different re-equilibrated BB population distributions. By converging the R 1→3 "one-pot" reactors in the central one (linkage step), we adjusted a suitable "global" pH or pH(R 4 ) for compounds 1−3 when all the components were homogeneously mixed in R 4 , heated, and, finally, collected in an open flask.
UV−vis spectrometry was also used to monitor the reaction during the pumping sequence between reactors, showing how the UV bands evolve with the acidification process (Figure 5a). The band at 330 nm, assigned to {W 11 } cluster formation (see Figure S6), was monitored in R 2 and R 3 simultaneously (Figure  5b), and the UV profiles are shifted in a time-dependent fashion. Hence, by dipping UV probes into every single "onepot" reactor, we observed the formation of {W 11 } on a different

Inorganic Chemistry
Article time scale, for instance comparing R 2 to R 3 (Figure 5c,d). The mechanistic studies are an interesting aspect to consider because reactor R 2 may be affected and gradually populated by the {W 11 } clusters, allowing the synthesis of 2 by reaction of {W 12 } and 1 (see Figure 4d). Indeed, the study of the selfassembly processes has allowed us to reconsider the role of the {W 11 } as a key building block. As described previously, subunit {W 11 } is commonly found within these structures despite it is being difficult to isolate. Lehmann and Fuchs described this cluster in 1988, but the procedure is very hard to reproduce. 39 However, using mechanistic observations explained above, we found that the {W 11 26 Other aspects of this particular reaction system are efficiency and reliability, which we can change with the counterions; this is very interesting because it is often very difficult, if not impossible, to change the counterions in the normal one-pot synthesis of a polyoxometalate without affecting the synthesis or structure formed. For instance, by using the same programmable reaction protocol that was applied while the TEA was replaced by a stock solution of DMA, we isolated (DMAH) 9 KNa 2 3a·22H 2 O 5 (anions 3a are isostructural, although they crystallize in distinct unit cells). 34 These results suggest that the compounds presented in this work are related by a network of underlying reactions between the {W 11 } and

Inorganic Chemistry
Article {W 12 } BBs. The formation of these compounds has previously been justified entirely by the existence of a "virtual combinatorial library" of tungstate aggregates based on crystallographic observations. It is has long been suggested that large clusters must be assembled in solution from a series of smaller so-called building blocks accessible from a virtual combinatorial library of common building blocks. 41 The use of automated control of networked "one-pot" reactor systems allows us to test this idea of far-from equilibrium reaction conditions operating in volume transfer, flow rates, and pumping frequency basis ( Figure 6). In this sense, the future potential of the NRS is focused on the exploration of not only multiconnectivity and configurability between the formation of transient N ≥ 5 "one-pot" reactors but also different pumping regimes and reaction conditions in a programmed and sequential way (giving access to new compounds with the same "space synthesis" but with a different "time synthesis"). This feature makes the NRS unique and potentially very useful in mechanistic studies and selfassembly of labile coordination or organic clusters 42 to achieve only-NRS accessible exotic products (kinetically unstable units co-trapped giving dynamic kinetic stability), such as giant supramolecular architectures. 43−45 Finally, we postulate that it would be interesting to utilize the NRS to explore n times the same one-pot reactions as a function of time (r t ), where aliquots of r t+1 could be used to seed r t−1 , thereby allowing a future "template" to be reseeded to previous or "past" reaction conditions. This would lead to a real prospect that novel compounds could be generated by a pumping frequencydependent reseeding approach using future information intrinsic to the system to disturb a past version of the system with a higher concentration of unlinked or exploited building blocks.

■ CONCLUSIONS
In conclusion, the NRS is used to integrate distinct syntheses of structurally related clusters, by multiple one-pot connected reactions. The common reaction variables were explored because of the NRS, which facilitates the automation of the syntheses of compounds 1−5. The NRS allowed us to modulate chemical compositions and an efficient pH adjustment during the application of a reaction protocol, thus producing different experimental conditions in each reactor within the system. As such, we propose a rational synthetic explanation ( Figure 6) for the isolation of such isopolyoxometalates based on {W 11 } and {W 12 } building blocks (BBs) formed in separate reactors. The hypothesis has been probed by using pH and UV real-time control to identify and monitor populations of transient species or building clusters, validating in a very basic sense the idea of a POM-based dynamic library.
In future work, we will extend the NRS approach to generate "sequence codes" for the synthesis of new compounds and reactive species from a given NRS connectivity with a predefined reagent set. We will also devise a standard reactor format suitable for cost-effective open-source duplication. This raises the real possibility of vastly expanding the exploration of one-pot synthesis, across a variety of chemistries, for the isolation of industrially useful and functional materials that can be reliably synthesized on a large scale without highly specialized know-how from sequence codes and an opensource reactor format.
■ EXPERIMENTAL SECTION Synthesis. All experiments are based on the operation of the networked reactor system, followed by regular analytics to fully characterize compounds 1−5. Crystal data and structure refinements are available for the following compounds. K 8 (7)°; V = 30890(3) Å 3 ; Z = 8; ρ = 4.063 g cm −3 ; λ(Mo Kα) = 0.71073 Å; 131486 reflections measured; 30320 unique reflections (R int = 0.1252) that were used in all calculations; 1682 refined parameters; final R 1 = 0.0540 and wR 2 = 0.1241 (all data). Data collection and reduction were performed using the CrysAlis software package, and structure solution and refinement were performed using the SHELXS-97 and SHELXL-97 46,47 program package of WINGX. 48 Corrections for incident and diffracted beam absorption effects were applied using empirical absorption correction. CCDC 982262 (4) and 982261 (5) contain the supplementary crystallographic data for this paper. These data can be obtained from The Cambridge Crystallographic Data Centre via http://www.ccdc. cam.ac.uk/.
UV-Integrated Networked Reactor System. The NRS presented here is assembled from 13 pumps; six of these are used to introduce the initial stock solutions of each reagent, and the other six are used for transferring the reaction mixtures and interconnecting the four stirred reactors. Finally, one pump is used to extract the final reaction mixture for collection. The current setup uses programmable syringe pumps (C3000 model, Tritontinent Ltd.) and a LabVIEWbased PC interface for the real-time monitoring and control of the setup.
In Situ Monitoring. The pH control was performed in the three primary reactors simultaneously using SevenMulti Mettler Toledo double-instant pH-meters. UV monitoring was performed simultaneously in R 2 and R 3 using a second LabVIEW PC interface connected to the Avantes Avaspec-2048 spectrometer equipped with a DH-2000

Inorganic Chemistry
Article halogen light source (Oceanoptics) that was connected further through a fiber-optic cable to a TP300 fiber probe.

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10

Funding
This work was supported by the EPSRC (Grants EP/ H024107/1, EP/I033459/1, and EP/J015156/1), the University of Glasgow, and the ERC (Project 670467 SMART-POM). L.C. thanks the Royal-Society Wolfson Foundation for a Merit Award.

Notes
The authors declare no competing financial interest.

■ ACKNOWLEDGMENTS
We thank Dr. Craig J. Richmond and Niall Mclean for useful discussions.