Constitutionally Selective Dynamic Covalent Nanoparticle Assembly

The future of materials chemistry will be defined by our ability to precisely arrange components that have considerably larger dimensions and more complex compositions than conventional molecular or macromolecular building blocks. However, exerting structural and constitutional control in the assembly of nanoscale entities presents a considerable challenge. Dynamic covalent nanoparticles are emerging as an attractive category of reaction-enabled solution-processable nanosized building block through which the rational principles of molecular synthetic chemistry can be extended into the nanoscale. From a mixture of two hydrazone-based dynamic covalent nanoparticles with complementary reactivity, specific molecular instructions trigger selective assembly of intimately mixed heteromaterial (Au–Pd) aggregates or materials highly enriched in either one of the two core materials. In much the same way as complementary reactivity is exploited in synthetic molecular chemistry, chemospecific nanoparticle-bound reactions dictate building block connectivity; meanwhile, kinetic regioselectivity on the nanoscale regulates the detailed composition of the materials produced. Selectivity, and hence aggregate composition, is sensitive to several system parameters. By characterizing the nanoparticle-bound reactions in isolation, kinetic models of the multiscale assembly network can be constructed. Despite ignoring heterogeneous physical processes such as aggregation and precipitation, these simple kinetic models successfully link the underlying molecular events with the nanoscale assembly outcome, guiding rational optimization to maximize selectivity for each of the three assembly pathways. With such predictive construction strategies, we can anticipate that reaction-enabled nanoparticles can become fully incorporated in the lexicon of synthetic chemistry, ultimately establishing a synthetic science that manipulates molecular and nanoscale components with equal proficiency.


General experimental procedures
Unless stated otherwise, all reagents were purchased from commercial sources and used without further purification. Dry N,N-dimethylformamide was purchased from Acros Organics. Prior to use, traces of dimethylamine were removed by evaporation under vacuum with gentle heating, followed by cooling to room temperature under vacuum then storage under a nitrogen atmosphere. All other dry solvents were obtained by means of a MBRAUN MB SPS-800TM solvent purification system, where solvents were passed through filter columns and dispensed under an argon atmosphere. Flash column chromatography was performed using Geduran® Si60 (40-63 µm, Merck, Germany) as stationary phase. Thin-layer chromatography (TLC) was performed on pre-coated silica gel plates (0.25 mm thick, 60F254, Merck, Germany) and observed under UV light (max 254 nm) or visualized by staining with acidic ceric ammonium molybdate solution, followed by heating.
Scanning transmission electron microscopy (STEM) was performed on an FEI Titan Themis operated at 200 kV and equipped with a CEOS DCOR probe corrector and a SuperX energy dispersive X-ray spectrometer (EDX). High angle annular dark field (HAADF) images were acquired with a probe convergence angle of 21.2 mrad and inner/outer collection angles of 56.3 and 200 mrad, respectively. Transmission electron microscopy (TEM) was also performed using a JEM 2010 microscope. High-resolution scanning electron microscopy (SEM) was performed using a Scios dualbeam microscope. Samples for electron microscopy were prepared by deposition of one drop of nanoparticle suspension on holey carbon films supported on a 300 mesh Cu grid (Agar Scientific®). Nanoparticle diameters were measured automatically using the software ImageJ. The images were first converted to black and white images using the "Threshold" function. The area of each nanoparticle was measured using the "Analyze particles" function. Particles on edges were excluded. UV-vis absorption spectroscopy was performed using a quartz cuvette (10 mm path length) on a Thermo Scientific Evolution 220 UV-Visible Spectrophotometer or an Agilent Technologies Cary 100 Series UV-vis Bio Spectrophotometer. Dynamic light scattering (DLS) measurements were performed using a glass cuvette (10 mm path length) on a Malvern Zetasizer μV or a Zetasizer Nano-ZS instrument. Each data point is the average of three independent measurements made in series. For each sample, three independent measurements were taken in series, and the results averaged. In turn, each measurement is the average of 10-17 sequential scans. The solvodynamic sizes are reported as the mean size for distributions expressed as % volume of particulate material. Size distributions were calculated by the instrument from the recorded intensity data. Equations reported in the literature were used to estimate appropriate values for viscosity, refractive index, and dielectric constant of 9:1 v/v DMF/D2O from the reported values for the neat solvents at 25 °C. 1-2 3 1 H, 13 C, 19 F NMR spectra were recorded on Bruker Avance 300, Avance II 400 and Avance III 400, 500 and 700 MHz instruments, at a constant temperature of 25 °C. 1 H Chemical shifts are reported in parts per million (ppm) from high to low field and referenced to the literature values for chemical shifts of residual non-deuterated solvent, with respect to tetramethylsilane. 4

PdNP-1(1)
Using a modified version of the synthetic procedure originally developed by Stucky and co-workers, 6 palladium(II) acetylacetonate (0.25 g, 0.50 mmol) was weighed into a 100 mL round bottom flask and dissolved in CHCl3 (40 mL). Hexanethiol (0.151 mL, 0.125 g, 1.06 mmol) was added and the reaction mixture heated to 55 °C. tert-Butylamine borane complex (0.44 g, 5.05 mmol) was added and the reaction was held at 55 °C for 1 h before being allowed to stir at room temperature for 5 h. The solution was transferred to a 250 mL flask to which MeOH/MeCN (4:1 v/v, 140 mL) was added before standing in the freezer overnight. The supernatant was removed, the residue dissolved in the minimal volume of CH2Cl2 and transferred to a vial, then the solvent was removed under a stream of compressed air. The residue was dispersed in MeOH with sonication, then subjected to centrifugation (1312 g rcf, 20 min). The supernatant was removed, and the pellet redispersed in fresh solvent. The nanoparticles were washed repeatedly in this manner with MeOH (3), then MeCN (2), after which no further impurities were observed in the supernatant by TLC. The black residue was then dried under vacuum overnight. The hexanethiyl-stabilized nanoparticles (65 mg) were then S5 suspended in CH2Cl2/DMF (4:1 v/v, 8 mL) and the mixture sonicated to ensure good dispersion of the nanoparticles. Thiol 1H (0.750 g, 1.38 mmol) was added and the mixture was stirred at room temperature in the dark. After 2 days, despite being still darkly coloured, the solution appeared cloudy with some solid precipitate at the bottom of the flask. Complete precipitation was achieved by addition of EtOH (20 mL), followed by centrifugation (1312 g rcf, 20 min). The supernatant was removed and discarded. In the same manner, the nanoparticles were subjected to a further two cycles of washing in Et2O/EtOH (1:1 v/v, 15 mL). After the final wash, traces of volatiles were removed under gentle air flow. Analysis by 1 H NMR spectroscopy following oxidative ligand desorption of a small sample indicated residual surface-bound hexanethiyl ligands. The nanoparticle solid was therefore suspended in DMF (8 mL) and the mixture sonicated. A further portion of thiol 1H (1.00 g, 1.84 mmol) was added and the solution was stirred at room temperature in the dark for a further 3 days. Nanoparticles were then precipitated upon addition of 6:1 v/v Et2O/EtOH (25 ml), followed by centrifugation (1312 g rcf, 20 min). The colourless supernatant was carefully discharged, before the black solid obtained was washed using the following procedure: nanoparticles were dispersed in Et2O/EtOH/CH2Cl2 (3:1:0.1 v/v, 15 mL), sonicated for 15 min, then centrifuged (1312 g rcf, 20 min). This operation was repeated a further 3 times. At this stage, no unbound molecular species were detected in the supernatant by TLC or NMR analysis. Traces of volatile solvents were removed from the purified residue under a stream of compressed air, to provide PdNP-1(1) (58 mg).

PdNP-1(2)
Using a modified version of a two-step synthetic procedure, 7 a solution of PdCl2 (0.250 g, 1.41 mmol) in H2O (25 ml) and HCl (5 ml) was extracted with a solution of tetraoctylammonium bromide (1.54 g, 2.82 mmol) in N2 purged toluene (150 mL). To the resulting orange organic solution, dioctylamine (8.52 mL, 28.2 mmol) was added. The mixture was vigorously stirred under N2 for 1 h. After cooling the mixture to 0 °C, a solution of NaBH4 (1.54 g, 2.48 mmol) in H2O (5 mL) was added rapidly. After 1.5 h of stirring at 0 °C, the aqueous layer was removed. To the obtained nanoparticle solution, thiol 1H (1.53 g, 2.82 mmol) in DMF (5 mL) was added rapidly. The reaction mixture was stirred for 15 min at room temperature before complete nanoparticle precipitation occurred. The colourless supernatant was carefully discharged, then the black solid obtained was washed using the following procedure: nanoparticles were dispersed in EtOH/toluene (1:1 v/v, 12 mL), sonicated for 10 min, then centrifuged (1312 g rcf, 10 min, 4 °C). This operation was repeated a further two times. Traces of volatile solvents were removed from the purified residue under a stream of compressed air. The solid obtained was dissolved in DMF (10 mL) and thiol 1H (0.780 g, 1.41 mmol) was added. The solution was stirred at room temperature for 3 h, before nanoparticles were precipitated upon addition of Et2O/EtOH (6:1 v/v, 50 ml). The colourless supernatant was carefully discharged, while the black solid obtained was washed using the following procedure: nanoparticles were dispersed in Et2O/EtOH/CH2Cl2 (3:1:0.1 v/v, 10 mL), sonicated for 15 min, then centrifuged (1312 g rcf, 10 min, 4 °C). The operation was repeated a further three times. At this stage, no unbound molecular species were detected in the supernatant by TLC or NMR analysis. Traces of volatile solvents were removed from the purified residue under a stream of compressed air, to provide PdNP-1(2) (45 mg).

'Electrophilic' nanoparticles AuNP-2
Two batches of gold core AuNP-2 were prepared, isolated and purified as previously described. 5 Structural and compositional details (Table S1) were assessed by electron microscopy (Figures S7, S9) and thermal gravimetric analysis ( Figure S8, S10). Although differing in mean core size, each batch had a similar monolayer density.     Samples were kept at room temperature throughout the experiment time course.
The concentration of colloidally stable material was monitored by UV-vis absorbance spectroscopy without any further dilution. The first absorption spectrum (t = 0) was recorded prior to addition of acid (where present), and subsequent spectra recorded at 24 h intervals.
Solvodynamic size distribution was monitored by dynamic light scattering (DLS). Samples were prepared by collecting 150 μL of solution previously subjected to UV-vis absorbance spectroscopy, then diluting to a final volume of 2.00 mL, using fresh DMF/H2O (1.85 mL, 9:1 v/v).
Samples of colloidally stable nanoparticles were prepared for electron microscopy by depositing one drop of nanoparticle suspension (solution previously subjected to UV-vis absorbance spectroscopy) onto a TEM grid sitting on a lint-free tissue. To image insoluble material, the sample was first sonicated for 5 min before withdrawing a drop of the resulting suspension, which was deposited on the TEM grid. All grids were left to dry at ambient pressure and temperature before imaging.        Solutions A and B were prepared volumetrically to give accurately known concentrations of CF3CO2H:
Concentrations of surface-bound hydrazones were assessed by 19 F{ 1 H} NMR spectroscopy relative to 4-fluorotoluene as internal standard, which was added at a known concentration (ca. 5 mM).
The following mixtures were prepared to contain equimolar concentration of the two NP-bound hydrazones ( Table S11. Samples were kept at room temperature throughout the experiment time course. Changes in the concentration of colloidally stable material were monitored by UV-vis absorbance spectroscopy without any further dilution. The first absorption spectrum (t = 0) was recorded prior addition of acid catalyst (where present), and subsequent spectra recorded at 24 h intervals.
Solvodynamic size distribution was monitored by dynamic light scattering. Samples were prepared by collecting 150 μL of solution previously subjected to UV-vis absorbance spectroscopy, then diluting to a final volume of 2.00 mL, using fresh 9:1 v/v DMF/H2O (1.85 mL) or fresh anhydrous DMF (1.85 mL, for experiments in absence of water).
For analysis by electron microscopy, insoluble extended aggregates formed as a result of complementary (complete precipitation of NP material) and selective (significant quantity of suspended solid) assembly were first isolated and purified from any unbound NPs. The precipitates were collected by centrifugation. The recovered black solid was then washed using the following procedure: solid was dispersed in DMF (4 mL), sonicated (10 min, 20 °C), then recollected by centrifugation (1935 g rcf, 15 min, rt). The same operation was repeated a further twice. Finally, each sample was suspended in fresh DMF (2 mL), sonicated for 5 min before withdrawing a drop of the resulting suspension and spotting onto a TEM grid. All grids were left to dry at ambient pressure and temperature before imaging.
Aggregate material composition was assessed by EDX spectroscopy at a minimum of three distinct areas of the grid. Spectra were acquired over 120 s. Atomic percentages were determined by fitting the emission lines for Pd(L) and Au(L). The relative atomic percentages were then converted to a ratio of nanoparticles using the average number of metal atoms in each particle as determined from (S)TEM size analysis (Table S1).
Experiments C1, C2: Co-assembly Figure S21. Experiment C1 representative HAADF images for assemblies produced at day 6 from the complementary assembly of PdNP-1 with AuNP-2 (from initial state [1]: [2] = 58:42; PdNP:AuNP = 50:50). EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S3. . EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S4. Complementary co-assembly from a homomaterial mixture PdNP:AuNP = 50:50). EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S5.  Figure S25. Experiment LN2 representative HAADF images for assemblies produced at day 5 from the nucleophilic linker-driven assembly of from a mixture of PdNP-1, AuNP-2 and 4 (initial state [1]: [2]:[4] = 1.4:1:6; PdNP:AuNP = 50:50) in the absence of water. EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S6.  S22 Figure S27. Experiment LN3 representative HAADF images for assemblies produced at day 5 from the nucleophilic linker-driven assembly of from a mixture of PdNP-1, AuNP-2 and 4 (initial state [1]: [2]:[4] = 1:1:6; PdNP:AuNP = 41:59). EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S7.  Figure S28. Experiment LN4 representative HAADF images for assemblies produced at day 5 from the nucleophilic linker-driven assembly of from a mixture of PdNP-1, AuNP-2 and 4 (initial state [1]: [2]:[4] = 1:1:6; PdNP:AuNP = 41:59) in the absence of water. EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S8.   Table S9, with accompanying HAADF images in Figure S30 Figure S30. Experiment LN5 representative HAADF images for assemblies produced at day 6 from the nucleophilic linker-driven assembly of from a mixture PdNP-1(2), AuNP-2(2) and 4 (initial state [1]: [2]: [4] = 1:1:3; PdNP:AuNP = 59:41). EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S9. Table S9. Experiment LN5 heteromaterial aggregate composition and enrichment factor (E.F.) determined by EDX mapping on three distinct sample regions of aggregates produced at day 6 from the nucleophilic linkerdriven assembly of from a mixture of PdNP-1(2), AuNP-2(2) and 4 (initial state [1]: [2]: [4]   Experiments LE1-LE3: Electrophilic linker selected assembly Figure S32. Experiment LE1 representative HAADF images for assemblies produced at day 10 from the electrophilic linker-driven assembly of from a mixture PdNP-1, AuNP-2 and 3 (initial state [1]: [2]: [4] = 1:1:6; PdNP:AuNP = 41:59). EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S10. -0.28 Figure S33. Experiment LE2 representative HAADF images for assemblies produced at day 10 from the during the electrophilic linker-driven assembly of from a mixture PdNP-1, AuNP-2 and 3 (initial state [1]: [2]: [4] = 1:0.7:4.3; PdNP:AuNP = 50:50). EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S11.   Table S12, with accompanying HAADF images in Figure S35)  Figure S35. Experiment LE3 representative HAADF images for assemblies produced at day 10 from the during the electrophilic linker-driven assembly of from a mixture PdNP-1, AuNP-2 and 3 (initial state [1]: [2]: [4] = 1:1:3; PdNP:AuNP = 59:41). EDX spectra were acquired from a region in the centre of each of these areas to give the results reported in Table S12.  Figure S36.

Kinetic models of selective assembly processes
During each multicomponent assembly process, several reaction pathways, involving numerous reaction substrates located in either bulk solution or nanoparticle-bound environments, are in competition. Nevertheless, even simplified kinetic models that focus on the reactions likely to dominate during the initial stages of each assembly experiment provide useful insight on the overall outcome. Each of these initial reactions generate new nanoparticle-bound and bulk-solution reactive species that can initiate further downstream reaction pathways. Eventually, inter-nanoparticle reactions produce colloidally unstable aggregates, after which precipitation and solution-solid reactions introduce further kinetic complexity. We proposed that the kinetic behavior of each system would be in large part controlled by the reactions of the initial nanoparticle-bound and bulk-solution species.
Making the further simplifying assumption that reactions of nanoparticle-bound species are not affected by neighboring ligand identity (and therefore rates of each particle-bound reaction are independent of the overall evolution of the system) allowed us to make reasonable estimates for the rate constants of each reaction based on our earlier kinetic studies (see below for further details). 5 In turn, this allowed us to probe computationally the effects of parameter variation by simulating the speciation profiles over time under different input conditions. Simulations were run in Copasi 4.33 (Build 246). 8 It should be noted that, given the simplifying approximations and uncertainties in estimated values for some reaction parameters, we interpret these simulations only as representing the qualitative evolution of the assembly processes; absolute values of time and concentration are note expected to be directly comparable to the experimental results.

Selective assembly of electrophilic DCNPs using a nucleophilic linker
The reactions possible during the initial stages of nucleophilic linker-driven assembly are represented by equations (1)-(4) (Scheme S2). Our estimates for the kinetic parameters for reactions (1-4) ( Table S13) suggest that the fastest process should be nucleophilic attack of bulk solution bis(hydrazide) 4 on the electrophilic AuNP-2 (reaction 4). Subsequent reaction of the nucleophilic nanoparticle product AuNP-L4-NHNH2 with another AuNP-2 produces the desired homomaterial inter-nanoparticle cross-links. However, either bulk solution or nanoparticle-bound hydrazides can also react with PdNP-1, to scavenge the 4-fluorobenzylidine unit (reaction 3 and equivalent reaction between AuNP-L4-NHNH2 and PdNP-1), thus producing nucleophilic PdNP-NHNH2. These highly nucleophilic nanoparticle-bound species can subsequently react with AuNP-2 or AuNP-CHO to produce heteromaterial linkages, serving to lower the selectivity of the assembly process. To a first approximation, therefore, selective linker-driven assembly of AuNP-2 will be achieved by maximizing the concentration of AuNP-L4-NHNH2, while minimizing the concentration of PdNP-NHNH2. A predictive parameter for assembly selectivity can thus be defined as  Table S13. Table S13. Estimated rate constants for reaction processes dominating during the initial stages of nucleophilic linker-driven assembly from a mixture of nucleophilic PdNP-1 and electrophilic AuNP-2 (Scheme S2). The measurement and estimation of rate constants is discussed in the following section. Building a kinetic model using reactions (1)-(4) and our estimated rate constants (Table S13) allowed us to simulate the speciation and evolution of SAu over time ( Figure S38). The simulations reveal that selectivity is maximal at the very start of the process, rapidly decreasing as the concentration of PdNP-NHNH2 increases. Thus, the maximum in [AuNP-L4-NHNH2] during the first few hours of the simulation is key to achieving high kinetic selectivity, indicating that a balance must be struck between aggregate yield and selectivity. In the absence of water, reactions (1) and (2) are shut down, with the consequence that absolute levels of selectivity are increased and extending the timeframe for which homomaterial-favoring species dominate ( Figure S38d-f). Simulating the influence of linker concentration revealed that both the rate of production and absolute maximum concentration of key species AuNP-L4-NHNH2 increase with increased [4]0 ( Figure S39a, solid lines). Increasing linker concentration also affects the rate of generation of PdNP-NHNH2 but less strongly ( Figure S39a, dashed lines). The consequence for the overall selectivity ( Figure S39b) is that higher linker concentrations correspond to higher absolute selectivity (t  0) and steady state selectivity (t  ∞), but also cause more rapid erosion of selectivity at intermediate time points. Crucially, therefore, the optimum linker concentration depends on the time point at which the aggregate will be probed ( Figure S39b, inset). This is consistent with our observation that the highest experimental selectivity was observed with 3 molar equivalents of linker under the input conditions of experiment LN5 (main text Figure 8 and Table S2).

Selective assembly of nucleophilic DCNPs using an electrophilic linker
Selective assembly of nucleophilic DCNPs (Scheme S3) first requires hydrolysis of PdNP-1 to reveal nanoparticle-bound nucleophile PdNP-NHNH2 (reaction 1). Although an essential intermediate en route to the necessary linker-capped PdNP-L3-CHO (reaction 5), the key nanoparticle-bound nucleophiles can also initiate pathways that are deleterious to assembly selectivity, via condensation with AuNP-CHO (reaction 6), or direct transamination with AuNP-2 (reaction 10). As we have no experimental system that is able to provide a quantitative measure of reaction rate between two nanoparticle-bound species, kinetic parameters for these reactions are poorly known. The latter reactiona transimination between two nanoparticle-bound speciescan be predicted to be relatively slow; however, we observe that hydrazide-aldehyde condensation reactions are minimally affected by surface confinement of only one component, so we must assume that the analogous nanoparticle-nanoparticle analogues (e.g. reaction 6) may be kinetically significant. Conversely, the formation of inter-particle linkages can be expected to be highly cooperative, so hydrolysis of interparticle linkages is likely to be considerably kinetically inhibitedeven when one linkage is cleaved, this is unlikely to result in a change to the nanoparticle aggregation state.
Based on the considerations above, the concentration of PdNP-NHNH2 is not an adequate predictor of assembly selectivity. Therefore, we must consider a second level of reactions involving the initially generated intermediates. Thus, including reaction 6, as discussed, and reaction 7scavenging of the linker by the released nucleophile 6. The resultant monotopic molecular aldehyde 6-L3-CHO can in-turn cap the nucleophilic DCNPs PdNP-NHNH2 to generate capped PdNP-L3-6 (reaction 11). This 'dead-end' intermediate can also be produced by direct reaction of molecular nucleophile 6 with key intermediate PdNP-L3-CHO (reaction 9); so, both of these processes reduce the concentration of key reactive nanoparticle intermediates. A transimination mode of attack between nucleophilic PdNP-NHNH2 and either linker-activated DCNPs PdNP-L3-6 or monotopic linker 6-L3-CHO is also available, giving alternative pathways to Pd-selective NP aggregates PdNP-L3-PdNP or linkeractivated PdNP-L3-CHO, respectively (Reactions 13 and 12). It can be predicted that each of these transamination processes is significantly slower than the alternative hydrazone condensation mode of attack at the highly reactive aldehyde site in each case.
Scheme S3. Reaction pathways available during electrophilic linker-driven assembly from a mixture of nucleophilic PdNP-1 and electrophilic AuNP-2. The dominant role of nanoparticle-bound intermediates in favoring (solid boxes) or disfavoring (dashed boxes) selectivity is indicated. Rate constants for hydrolysis (khydr), condensation (kcond) and transimination (ktrans) reactions are labelled; estimated values are given in Table S14. In this model, the linker molecule cannot react with the nanoparticles in their starting state; the action of water is first required to produce a nanoparticle-bound nucleophile. Thus, the influence of both water and the concentration of linker on the assembly process should be expected to be quite different to the selective assembly of electrophilic DCNPs considered above. Selective assembly of nucleophilic DCNPs will be favored by maximizing the concentration of linker-activated PdNP-L3-CHO, which should react in a relatively fast nanoparticle-nanoparticle condensation reaction with PdNP-NHNH2 to produce homomaterial assemblies. Meanwhile, minimizing the concentration of AuNP-CHO, which can undergo a condensation with PdNP-NHNH2 to give heteromaterial assemblies, should also be key to selectivity. One intrinsic factor favoring the desired selectivity should be the faster reaction of electron-poor PdNP-L3-CHO in reaction 8 compared to electron-rich nanoparticle-bound electrophile AuNP-CHO in reaction 6.
A measure of selectivity for this assembly process should be provided by the number of homomaterial versus heteromaterial inter-particle linkages: SPd = [PdNP-L3-PdNP]/[PdNP-AuNP] (main text Figure 7). However, it should be noted that this parameter is derived from a model that includes poorly known interparticle reaction rates. As a measure of the extent of aggregation, we also considered the number of interparticle linkages as a proportion of the total number of nanoparticle-bound reactive units (%NP-NP link).

S35
In contrast to the nucleophilic linker-driven assembly of electrophilic DCNPs, simulations reveal that selectivity for nucleophilic DCNPs in the presence of an electrophilic linker reaches a maximal steady state at long time periods (main text Figure 7). Furthermore, the small number of interparticle linkages predicted at the steady state ( Figure S41) is in line with our experimental observations that a large proportion of material remains in solution.
Examining the evolution of activated nanoparticle species over time, it is apparent that as well as the hydrolyzed electrophilic AuNP-CHO, linker-activated and linker-capped nucleophilic DCNP species PdNP-L3-CHO and PdNP-L3-6 dominate in solution at later time points (Figure S40a, c). Increasing the rate of PdNP-1 hydrolysis in an attempt to produce more nucleophiles that may capture PdNP-L3-CHO, however, will concurrently increase the formation of heteromaterial linkages through reaction with AuNP-CHO, as well as generating higher concentrations of molecular capping agent 6, thus producing more dead-end PdNP-L3-6. Therefore, the parameter most likely to positively influence aggregate selectivity is the concentration of linker 3. The simulations predict that employing higher relative concentrations of 3 increases selectivity (main text Figure 7), but at the expense of further reducing aggregate yield ( Figure S41), suggesting that moderate molar excesses of linker are optimal. It is notable that at low concentrations of linker 3, SPd < 1 at all time points (Figure 7a), corresponding to more heteromaterial than homomaterial linkages. This is consistent with our experimental results where the experiment with the lowest linker concentration (LE3) showed no selectivity for assembly of PdNP.

Parameter estimation for kinetic models
Rate constants used in the kinetic models were estimated by direct experimental measurement where possible or by extrapolation from values measured using monotopic (i.e. non-aggregating) modifier units as described below. All measurments were made on AuNP cores. In recogntion of the uncertainty in the extrapolated values, we only consider values to an accurarcy of 1 significant figure. Experimental measurements were made by reaction tracking using 19 F NMR as previously reported for AuNP-1 and AuNP-2. 5

Condensations and hydrolyses involving molecular fragments
Rate constants were estimated (Tables S15 and S16) from previously reported experimental measurements (Scheme S4). 5 Based on our observations of these and analogous reactions on molecular substrates, two trends were applied to estimate rate constants for reactions that could not be measured directly:  Hydrolysis rates depend strongly on the electronic nature of the benzylidine portion and are only moderately influenced by the electronic nature of the leaving group. Nanoparticle attachment via the bezylidine end inhibits hydrolysis by a factor of ca. 0.5 and via the hydrazide fragment by a factor of ca. 0.8. 5  Condensation rates strongly depend on the electronic nature of the aldehyde and are moderately influenced by the electronic nature of the hydrazide. Condensations are minimally affected by the regiochemistry of the reactive site with typical surface inhibition factors ca. 0.9 of the solution phase value. 5 Scheme S4. Experimentally determined rate constants used for estimating values for hydrolyses and condensations involving at least one molecular fragment. Values were determined as previously described by tracking reactions using 19 F{ 1 H} NMR spectroscopy then fitting to a bimolecular reversible rate equation. 5 Conditions: CF3CO2H (20 mM), D2O/DMF (10% v/v). Table S15. Estimation of rate constants for hydrolyses involving at least one molecular fragment.