Solution-based self-assembly represents a powerful strategy for spontaneous organization of individual nanoparticles into well-defined assemblies (e.g., small clusters and ordered superstructures) that exhibit distinct collective properties [1–4]. Particularly, owing to strong surface plasmon coupling and light-matter coupling, plasmonic nanoparticle assemblies [5] have demonstrated immense potential in functional applications such as optoelectronic devices including surface enhanced Raman scattering (SERS) [4], biosensing and medical diagnostics [6, 7], chiral catalysis [7], and plasmonic metamaterials [8]. Since the properties of the nanoparticle ensembles are remarkably dependent on the shape, orientation, and spatial arrangement of the nanoparticles, precise control over the architectural configuration as well as the structural complexity is necessitated for achieving desired applications [9, 10]. In this regard, self-assembly of anisotropic nanoparticles is receiving increasing attention because the particle geometry plays a key role in engineering hierarchical assemblies with particular structures [11, 12]. For example, anisotropic nanocrystals with different morphologies, such as rods [4, 13, 14], plates [15], polyhedral [16–19], dumbbells [20], arrows [21], and multipods [22, 23], have been successfully employed for assembling superstructures with tunable and complicated architectures. Nevertheless, it remains challenging to program the self-assembly of colloidal nanoparticles into desirable structures with predetermined configurations [24, 25].

Programmable self-assembly of colloidal particles into sophisticated superstructures requires site-selective directional bonding or interactions [25, 26]. Generally, the shape anisotropicity in combination with regioselective functionalization can endow anisotropic nanoparticles with desirable bonding directionality needed for programmable self-assembly. As prototypical anisotropic nanoparticles, gold nanorods (GNRs) have been widely employed for site-selective functionalization to achieve end-to-end and side-by-side self-assembly selectively [11, 27, 28]. Notably, the local curvature is a crucial parameter in determining the site-selective adsorption of functional ligands on anisotropic nanocrystals. For example, thiolated ligands tend to preferentially bond to the two ends of cetyltrimethylammonium bromide- (CTAB-) capped GNRs at low concentrations because of the smaller packing density of CTAB at the ends with a larger curvature [29], which can lead to the end-to-end assembly of GNRs with the ends functionalized with either molecular ligands like cysteamine [30] or polymers like thiol-terminated polystyrene [31]. Interestingly, the preferential adsorption of 2-naphthalenethiols on the tips of gold triangular nanoprisms and subsequent attachment of polystyrene-b-poly-(acrylic acid) resulted in tip-patched nanoprisms that can be assembled into unusual twisted dimers [32]. Besides small molecules and polymers, DNA has been the most frequently used as powerful ligands to enable programmed self-assembly of nanoparticles into structures with increased complexity and diversity [33]. It is noteworthy that unprecedented complexity of DNA-programmed nanoparticle assemblies has been achieved based on two strategies: DNA-mediated self-assembly and DNA as a structure-defining element [34]. As an outstanding example of the DNA-programmed self-assembly, DNA-modified triangular gold bipyramids were successfully assembled into the most sophisticated clathrate architectures [17]. Considering that the nonselective binding of the terminal thiol of DNA to the different facets of anisotropic nanoparticles could result in a mixture of nonuniformly encoded nanoparticles, a regioselective surface encoding method based on selective blocking of particle surfaces with a polymer was recently developed [25]. Despite progress toward programmable self-assembly, the anisotropic nanoparticles employed as the building blocks are still largely limited to the particles with a convex geometry, and the complexity of the realized nanoparticle assemblies remains to be increased. Since information-rich building blocks are generally required for the self-assembly of complex structures [24], the unique nanoarrows with a distinct convex-concave structure and clearly demarcated surface curvatures [21, 35] would represent a promising building block for programmable self-assembly.

Herein, we demonstrate that regioselective adsorption of 1,10-decanedithiol dictated by the local curvature allows for programmable self-assembly of gold nanoarrows (GNAs) with five distinct assembly modes. Furthermore, a two-step self-assembly process involving regioselective blocking with 1-dodecanethiol followed by adsorption of 1,10-decanedithiol leads to six assembly modes including three novel wing-engaged modes. This approach reveals how the geometric morphology of nanoparticles affects their regioselective functionalization and drives their self-assembly, which represents a significant step toward programmable self-assembly of nanoparticles with designed shapes into diverse and complicated superstructures.

The unique [001]-elongated GNAs consist of two pyramidal heads exhibiting four {111} facets together with four truncated {100} facets, which are connected by a four-wing shaft with each wing parallel to {110} facets [21]. The well-defined convex-concave structure endows the GNAs with a variety of distinct surface curvatures; particularly, the local curvature decreases from positive for vertexes and edges through zero for facets to negative for the outer surface of wings [35]. The general self-assembly strategies based on regioselective adsorption on GNAs are schematically illustrated in Scheme 1. Initially, the GNAs are capped by CTAB bilayers that can be readily replaced by thiol-terminated molecules due to the strong Au-S bonding. Notably, thiolated ligands tend to replace the CTAB molecules on a location with a higher curvature due to a lower packing density of CTAB [29], indicating that there exists a preferential order for the regioselective adsorption: vertexes, edges, facets, and wings. In the one-step dithiol-induced self-assembly, 1,10-decanedithiol is introduced as a divalent linker, which would exclusively bond to the vertexes of GNAs at a low concentration, and sequentially bond to the edges and facets with increasing concentration, leading to a variety of assembly modes ranging from vertex-to-vertex to facet-to-facet. Alternatively, a two-step thiol-dithiol-induced self-assembly involves regioselective blocking of GNAs with the monovalent ligand 1-dodecanethiol and subsequent regioselective adsorption of the divalent linker 1,10-decanedithiol, leading to a variety of assembly modes ranging from edge-to-edge to wing-to-wing.

Well-defined GNAs ~125 nm in length were used as nanoscale building blocks for self-assembly, and their spatial orientation can be easily determined based on the orientation of the central four-wing shaft as well as the apparent vertex angle of the pyramidal heads of the GNAs lying on the substrate (Figure S1). Through one-step dithiol-induced self-assembly, five distinct assembly modes can be achieved based on regioselective adsorption of the dithiol linker (Figure 1). Introduction of 5  μM 1,10-decanedithiol to the GNAs dispersion induced self-assembly of the GNAs into regular GNA assemblies (termed as GNA-dithiol-5), as evidenced by the time evolution of the absorption spectra (Figure S2a), which revealed a gradual redshift of the longitudinal plasmon peak from 934 to 958 nm with increasing time to 2 h. Typical transmission electron microscopy (TEM) images suggested that the GNA-dithiol-5 exclusively adopted a vertex-to-vertex assembly mode (Figure 1(a1) and Figure S3), which basically consisted of two contact ways: top vertex-to-top vertex (Figure 1(a2)) and top vertex-to-side vertex (Figure 1(a3)). As the 1,10-decanedithiol concentration was increased to 20  μM, in addition to the vertex-to-vertex mode, the obtained GNA-dithiol-20 exhibited two new assembly modes: vertex-to-edge (Figure 1(b)) and edge-to-edge (Figure 1(c)), and the vertex-to-edge became the dominant assembly mode (Figure S4). If the 1,10-decanedithiol concentration was further increased to 40  μM, two novel assembly modes appeared for the resultant GNA-dithiol-40, i.e., edge-to-facet (Figure 1(d)) and facet-to-facet (Figure 1(e)), with the edge-to-facet as the most frequently observed assembly mode. It is noteworthy that the predominant number of the GNAs constituting the assemblies was generally increased from 2 for GNA-dithiol-5 (Figure S3) to 4-6 for GNA-dithiol-40 (Figure S5). This result is consistent with the time evolution of the absorption spectra of GNA-dithiol-40 (Figure S2b), which showed a larger redshift of the longitudinal plasmon peak from 934 to 984 nm, indicating a longer chain-like assemblies [29]. To quantify the yield of each assembly mode at different dithiol concentrations, a number of GNA assemblies involving more than 200 assembly sites connecting adjacent GNAs were statistically analyzed based on TEM observations (Figure 1(f)), which indicated a general change in the assembly mode from vertex-to-vertex through vertex-to-edge to edge-to-facet with increasing the dithiol concentration. Considering the possible perturbation of the preparation of TEM samples to the assembly configuration, the GNA assemblies were stabilized by silica coating in solution [36]. While the assembly modes at different dithiol concentrations were essentially unchanged, some complicated three-dimensional (3D) configurations could be captured for the silica-coated assemblies (Figure S6-9), indicating that the normal sample preparation did not considerably change the contact way of GNAs due to the strong bonding of the thiol group to gold despite possible deformation of the 3D configuration upon drying on copper grids. It is noteworthy that the SiO₂ coating can well preserve the structure of nanoparticle assemblies formed in solution and endow them with long-term stability and solution processability [36, 37], which would allow for exploring useful applications of the stabilized GNA assemblies.

The observed assembly modes of the one-step dithiol-induced self-assembly of GNAs can be rationalized in terms of regioselective adsorption of 1,10-decanedithiol dictated by the local curvature (Figure 2). Since the vertexes of GNAs have the highest local curvatures, the dithiol linker would be preferentially adsorbed on the top and side vertexes at a low concentration (e.g., 5  μM). Therefore, the thiol end group of the dithiol ligand attached to a vertex of a GNA would tend to bond to an available vertex of another GNA, where the dithiol ligand was not adsorbed or was unsaturatedly adsorbed, resulting in the exclusive vertex-to-vertex assembly mode (Figure 2(a) and Figure S10). The observation that the GNA dimmers were the predominant product indicated that the amount of the dithiol linker could be too low to occupy all the vertexes, and hence, the further aggregation would be prevented owing to the lack of dithiol molecules. When the dithiol concentration was increased to 20  μM, most of the vertexes were occupied by dithiol and a few sites on the edges were also attached by dithiol, leading to occurrence of the vertex-to-edge mode (Figure 2(b)) and the edge-to-edge mode (Figure 2(c)). For the edge-to-edge mode, the bonding of a thiol end group of the dithiol adsorbed on an edge to a blank edge site of another GNA would be accompanied by subsequent bonding of the neighbouring dithiol linkers to the blank sites on the edge, resulting in a stable edge-to-edge connection. With further increasing the dithiol concentration to 40  μM, most of the vertexes and edges were occupied by dithiol and a few sites on the facets were also adsorbed by dithiol, leading to the appearance of the edge-to-facet mode (Figure 2(d)) and the facet-to-facet mode (Figure 2(e)). It may be noted that the adsorption of the dithiol linker and the self-assembly of the linker-attached GNAs occurred almost simultaneously; the coexistence of several assembly modes seemed inevitable for the self-assembly at a high dithiol concentration.

If monocovalent 1-dodecanethiol was firstly introduced to the GNA dispersion before the divalent 1,10-decanedithiol was added, regioselective blocking of the GNAs occurred and subsequent adsorption of the dithiol linker could take place only on the unblocking areas. This two-step thiol-dithiol-induced self-assembly of GNAs could bring about novel assembly models. The concentration of 1-dodecanethiol was varied to change the blocking area while the concentration of the dithiol linker was kept at 10  μM. When 20  μM 1-dodecanethiol was introduced, the obtained GNA assemblies (termed as GNA-thiol-20-dithiol) exhibited three assembly modes, namely, edge-to-edge, edge-to-facet, and facet-to-facet (Figures S11 and S12). All these three modes had been observed in the GNA assemblies obtained by the one-step dithiol-induced assembly. Notably, the vertex-engaged modes (i.e., vertex-to-vertex and vertex-to-edge) did not appear at all because the vertexes had been completely blocked by the monothiol ligand, as schematically illustrated (Figure S13). With increasing the 1-dodecanethiol concentration to 40  μM, considerable edge areas were blocked and the edge-to-wing emerged as the main assembly mode for the resultant GNA-thiol-40-dithiol (Figure 3(a) and Figure S14). Interestingly, this edge-to-wing mode basically contained two contact ways, namely, one pyramid edge contact one wing on its face (Figure 3(a2)) and one edge contact two wings on their edges (Figure 3(a3)). When the 1-dodecanethiol concentration was further increased to 60  μM, the edges were largely blocked and the facet-to-wing mode (Figure 3(b)) emerged as the main assembly mode for the resultant GNA-thiol-60-dithiol, which was accompanied by the appearance of the wing-to-wing mode (Figure 3(c)).

It may be noted that for each of the three wing-engaged assembly modes, there existed two contact ways: one-wing-engaged and two-wing-engaged (Figures 3(a)–3(c) and Figure S15). The statistical analysis of the overall fractions of the two contact ways suggested that while the one-wing-engaged way was slightly preferred for the facet-to-wing and wing-to-wing modes, the two-wing-engaged way was significantly favored for the edge-to-wing mode (Figure 3(d)). Based on the proposed assembly mechanisms (Figure 3(e) and Figure S16), a pair of parallel GNAs adopting the edge-to-wing mode showed a high degree of shape complementarity and the two GNAs had a denser packing compared with the other two wing-engaged modes. It is known that for two nanoparticles approaching each other at a small separation distance, the closer they are, the higher the van der Waals (vdW) attraction energy becomes [1, 35]. Since the neighbouring GNAs with the edge-to-two-wings manner had a closer contact than the edge-to-one-wing manner, the strong vdW attraction would considerably contribute to the edge-to-two-wings assembly manner. Generally, with increasing the monothiol concentration in the two-step thiol-dithiol-induced self-assembly, the main assembly mode evolved from edge-to-facet through edge-to-wing to facet-to-wing (Figure S17). In summary, based on regioselective adsorption of thiolated ligands, five distinct assembly modes were achieved by one-step dithiol-induced self-assembly of GNAs, and six assembly modes including three novel wing-engaged modes were realized via two-step thiol-dithiol-induced self-assembly (Figure 4).

The obtained diverse GNA assemblies represent unique plasmonic nanostructures showing notable structure-dependent field confinement. The electric field distribution of GNA dimers with five representative assembly modes was simulated with a finite element method (FEM), which displayed a variety of hot spots generated from strong localized surface plasmon resolution (LSPR) coupling among adjacent GNAs (Figure 5(a)). It is noteworthy that the contact location of the assembly adopting the vertex-to-edge mode exhibited the highest normalized electric field intensity among the five assemblies probably owing to the existence of strong tip-edge-tip plasmon coupling. It is known that the hotspots located at the nanogaps between plasmonic nanoparticles as well as the sharp corners of the particles contribute greatly to SERS [38]. The distinct nanogaps between adjacent nanoarrows with various contact manners together with plenty of sharp tips and edges of each nanoarrow would endow the GNAs with strong SERS response toward probe molecules. Crystal violet (CV) was employed as a probe molecule to evaluate the SERS property of the different GNA assemblies dispersed in DMF. The CV molecules exhibited a strong peak at 1620 cm^-1 attributed to the aromatic C–C stretching vibration [39], and its intensity displayed significant difference for the GNA assemblies obtained under different conditions (Figure 5(b)). Particularly, the intensities for GNA-dithiol-5, GNA-dithiol-20, and GNA-dithiol-40 were 2.8-fold, 3.8-fold, and 6.0-fold higher than those for dispersed GNAs, respectively. This result indicated that the SERS performance of the GNA assemblies was generally increased from vertex-to-vertex through vertex-to-edge to edge-to-facet and facet-to-facet, which was associated with increased contact area and augmented hotspot amount. Among the GNA assemblies obtained by the two-step thiol-dithiol-induced self-assembly, the GNA-thiol-20-dithiol with mainly the edge-to-facet and facet-to-facet modes showed the strongest Raman intensity, confirming the importance of the contact area or the hotspot amount [40] in determining the SERS performance of GNA assemblies. These results suggested that tunable plasmonic characteristics and SERS properties could be readily achieved by manipulating the self-assembly mode via simply changing the ligand concentration, indicating potential applications in optoelectronics, biosensing, and biomedicine.

In conclusion, regioselective adsorption of 1,10-decanedithiol on the vertexes, edges, and facets of GNAs allowed for programmable self-assembly of GNAs with five distinct assembly modes, and regioselective blocking with 1-dodecanethiol followed by adsorption of 1,10-decanedithiol gave rise to programmable self-assembly with six assembly modes including three novel wing-engaged modes. It was unraveled that the assembly mode was essentially determined by regioselective adsorption of dithiol linkers dictated by the local curvature together with the shape complementarity of GNAs. The fact that multiple complex assemblies were readily obtained by introducing simple small molecular ligand demonstrated the superiority of nanoarrows with multiple distinct surface curvatures as unique building blocks. This approach revealed how the geometric morphology and regioselective functionalization of nanoparticles drove their self-assembly, which represents a significant step toward programmable self-assembly of nanoparticles with designed shapes into diverse and complicated superstructures.