Citrate Coordination and Bridging of Gold Nanoparticles: The Role of Gold Adatoms in AuNP Aging

Gold nanoparticles used in many types of nanostructure are mostly stabilized by citrate ligands. Fully understanding their dynamic surface chemistry is thus essential for applications, particularly since aging is frequently a problem. Using surface-enhanced Raman spectroscopy in conjunction with density functional theory calculations, we are able to determine Au–citrate coordination in liquid with minimal invasiveness. We show that citrate coordination is mostly bidentate and simply controlled by its protonation state. More complex binding motifs are caused by interfering chloride ions and gold adatoms. With increasing age of stored gold nanoparticle suspensions, gold adatoms are found to move atop the Au facets and bind to an additional terminal carboxylate of the citrate. Aged nanoparticles are fully refreshed by removing these adatoms, using etching and subsequent boiling of the gold nanoparticles.

Since the gold chloride to citrate ratio of 1:1.33 was the same for both batches, this also means that K + and Na + have no influence on the reaction kinetics. The hydrodynamic diameter is ~49.7 nm with a standard deviation of ~17.6 nm. The pH of the freshly synthesised AuNP suspension is 3.5±0.1 as a result of the citrate to HAuCl4 ratio. The low pH is advantageous to keep citrate break-down products off the gold surfaces. With pKa=4.7 1 , acetate/acetic acid (the final by-product) will be fully protonated. Despite the low pH, the AuNPs are stable for months. Even after four months from the day of synthesis, no signs of aggregation are observed in the UV-vis spectra ( Figure S1c). The decrease in optical densities is caused by AuNPs deposited on the inside walls of the centrifuge tubes used to store the AuNP suspensions. During the ageing experiments, no deposition was noticed on the walls of the containers.
Ageing experiment for higher pH values and salted AuNPs. The ageing experiment was repeated by "salting" the AuNPs with trisodium citrate (TSC) and sodium chloride (NaCl) instead of using CB [5] as the nano-spacer. This is to rule out potential interference between CB [5] and citrate during the ageing process and to determine the coordination modes of citrate. The spectra of fresh and aged AuNPs contain peaks which are also present in the ageing experiments with CB [5] (Figure S2a,b). To observe and compare ageing at different protonation states of citrate, the pH values of the AuNP suspensions are adjusted to 3.8, 5.5 and 6.8 by adding citrate buffer (TSC/citric acid) with a final concentration of 5 mM. For the NaCl-salted AuNPs, both fresh and aged, the 1300-1320 cm -1 mode indicates monodentate binding of the terminal carboxylates possibly caused by chloride ions (see main text and Figure S13 for control measurements).
At pH=5.5, the c2 coordination mode (1020 cm -1 ) is more pronounced than for the lower pH=3.8 where c2-t2t2 (995 cm -1 ) dominates. This can be explained by a lower citrate surface coverage at lower pH which allows citrate ions to span the nanogap. The surface coverage is presumably lower at pH=3.8 (pKa1=3.13) because significantly more fully protonated citrate ions exist compared to pH=5.5.
Aged NaCl AuNPs show a decrease of the adatom-driven c2t2-t2 coordination with increasing pH. Conversely, a vibrational mode at 925 cm -1 increases with increasing pH. This mode v(CCOO -) of the uncoordinated central carboxylate exhibits a strong Raman activity and only occurs for t2t2 coordination.
The TSC salted AuNPs (~60 mM TSC concentration) show fewer coordination modes; the fresh AuNPs are clearly dominated by c2 for all pH values. The two higher pH values contain some c2-t2t2 modes as evident by the vibrations at 995 and 1535 cm -1 . Aged AuNPs exhibit the adatom-driven c2t2-t2. At pH=6.8, there is some indication of t2t2 with the emerging 925 cm -1 mode. A summary of the citrate coordination modes can be found in Table S1.   pH  fresh NaClsalted   aged NaClsalted   fresh TSCsalted   aged TSCsalted  3 Arrhenius energy calculation. The adatom-induced CC-stretch at 1080 cm -1 observed in the CB [5]:AuNP ageing spectra ( Figure S3a) with its linear increase over time is an excellent proxy to calculate the activation energy (Arrhenius energy) of the underlying gold adatom migration.  Accounting for the uncertainty in temperature of the samples, the calculation yields an activation energy = 1.1±0.2 eV = 26 kcal/mol.

Repeatability of CB[5]:AuNP.
Aggregates mediated by CB [5] with a final concentration of 2.5 µM produce highly repeatable SERS signals in terms of peak intensities and SERS background ( Figure S4a). This is due to the precise 0.9 nm gap size defined by CB [5] and the fractal morphology of the clusters.  The symmetric stretch vs(COO -) of aqueous citrate (Figure S5c, top) is located at ~1410 cm -1 . Upon bidentate coordination to two Au atoms (µ2), it is blue-shifted ( Figure  S7c, bottom) to ~1385 cm -1 . The COO --Au mode is not very useful for determining which of the three carboxylates are coordinated because of vibrational coupling between all three carboxylates. This makes it very difficult to disentangle each contribution. The vs(COO -) stretches are good markers to distinguish between bidentate bridging (µ2) and monodentate coordination(1κ:O 1 ) via a single oxygen.
The asymmetric carboxylate stretches va(COO -) are much more suitable for distinguishing which carboxylate is bound. In aqueous citrate ( Figure S5d, top), the COOstretches form a shoulder at ~1560 to 1630 cm -1 with a clear separation (seen in DFT) between the central (blue) and terminal carboxylates (red). Upon bidentate coordination, the vibrations are blue-shifted by ~40 cm -1 . Coordinated terminal carboxylates are translated to ~1530-1550 cm -1 and the central carboxylates (as in Figure S5b) to ~1560-1610 cm -1 . Intramolecular hydrogen bonding of the central OHgroup to the bound carboxylate (as observed in DFT) has a strong influence on the position of the COO --Au stretch. This results in an even stronger blue-shift by an additional ~20 cm -1 .
The carbon-carbon stretches ν(CC) of the CC bonds adjacent to the carboxylates ( Figure S7e) exhibit very distinct shifts when their attached carboxylate coordinates to Au. The vibration of unbound terminal carboxylate is located at 950 cm -1 (blue), whereas the unbound central carboxylate (red) is located at 1050 cm -1 . Upon coordination of a terminal carboxylate, the adjacent CC stretch is red-shifted to 960-985 cm -1 . The central carboxylate (Figure S5e) CC-stretch is blue-shifted to ~1020 cm -1 (strong red peak).
There are seven possible coordination modes that citrate can reach when trapped inside a nanogap ( Figure S6a) (excluding the hexadentate mode with all three carboxylates bound to the same facet, which we find from DFT is not possible to flat Au). The distinct positions of their CC-stretches in combination with the asymmetric COO --stretches are sufficient to distinguish these modes in a SERS spectrum. Compared to aqueous citrate, bidentate coordination of a terminal group (t2) results in the red-shift of the adjacent CC stretch from ~750 cm -1 to 775 cm -1 . For the central group, the C-C stretch is blue-shifted from 1050 cm -1 to 1020 cm -1 (1030 cm -1 DFT).
For citrate modes with structural symmetries the central and terminal CC-stretches are coupled and can coincide to form a single vibration. This occurs in c2-t2t2 (strong) and c2t2-t2 (weak) resulting in a joint vibration at 990-995 cm -1 . This is also the case for t2t2 where coupling between the two terminal groups with the unbound central carboxylate lead to a mode which is expected at ~1005 cm -1 (weak).
To distinguish between the triple bidentate modes c2-t2t2 and c2t2-t2, the CC-stretch at 1080 cm -1 (green) can be used. This mode becomes very active only if a terminal and the central carboxylate are bound on the same Au-facet. In measurements this mode is only observed when the terminal carboxylate coordinates to an adatom.
As mentioned above, the central hydroxide (-OH) group of citrate prefers hydrogenbonding to a carboxylate oxygen (intra-molecular H-bond). This affects the peak positions of the asymmetric carboxylate vibrations. The characteristic -OH···OOC bending mode is located at 1515 cm -1 in both measurement and DFT calculation, but only very weakly Raman active. Binding of the hydroxide group to gold (with or without cleavage of the hydrogen) is possible, but no stable geometry could be found using DFT calculations. If binding occurred via the -OH group, the C-OH stretching vibration (also weakly Raman active) located at 1140 cm -1 would undergo a shift and become much more Raman active, as is the case for the C-O-Au stretches of the carboxylate in monodentate binding.

Influence of cations (K + , Na + ) on ageing.
Concurrently to ageing experiments performed with TSC AuNPs, the same ageing protocol was applied to a batch of tripotassium citrate (TPS) AuNPs. The emerging peaks in the spectra (Figure S7, a-c) are identical for both cations. This suggests that the influence of cations on ageing and citrate coordination is negligible. Citrate decomposition products. Besides citrate, the supernatant of the AuNP suspension also contains reaction by-products (from the NP synthesis) and their thermal decomposition products. The typical decomposition pathway is: citrate, 1,3acetone-dicarboxylate (ADC), acetoacetate (AA), and finally acetate/acetone. The unstable reaction by-products AA and ADC (both as acids and salts) undergo rapid decarboxylation during the reaction (100 °C). Decomposition also carries on after the reaction whilst the freshly synthesised nanoparticle suspension cools down to room temperature (ca. 1h in our case).
The activation energy for the decomposition of ADC and AA are 23.9 kcal/mol=1.04 eV and 23.7 kcal/mol=1.03 eV. The rate constants at 25 °C for AA (acid) is = 1.64 × 10 −3 min −1 and for ADC (acid) = 0.99 × 10 −3 min −1 . Based on their SERS spectra (Figure S8, a-c) on plasma-cleaned AuNP films (top) and after the addition of excess amounts to fresh aqueous AuNPs (bottom), it is evident that observed ageing process is not related to citrate decomposition products. This is further supported by swapping the supernatant of aged and fresh AuNPs.

ADC AuNPs.
It is possible to synthesise gold nanoparticles by using disodium 1,3acetone-dicarboxylate (ADC) instead of trisodium citrate as the reducing agent. By doing so, the SERS spectra of citrate decomposition products, e.g. acetoacetate, without the presence of citrate can be observed. The synthesis yields mostly spherical AuNPs with a binary size distribution ( Figure  S9c). the typical ageing response observed in citrate-reduced AuNPs is not seen even when the particles are aged (Figure S9a). This confirms again that thermal decomposition of citrate by-products is not responsible for ageing. (at room temperature). Through the addition of HAuCl4 to the AuNP suspension, etching of adatoms is observed. The presence of citrate also leads to a slight growth of the AuNPs. However, the reaction rate for growth is very low (it normally requires 100°C). Therefore, the balance of these two is slow etching but only of adatoms off the facets. Zeta size measurements show that the hydrodynamic radius barely increases over 2h (likely not a significant variation) ( Figure  S10). Control experiment: Boiling of aged AuNPs without HAuCl4. Boiling aged AuNPs aggregates without addition of HAuCl4 (boiled after aggregation) increases and slightly broadens the adatom mode at 1080 cm -1 (Figure S11a) This is indicative of adatoms being formed. Figure S11. Boiling aged AuNP aggregates (a) Typical 'aged' CB [5]:AuNP spectrum (black) with adatom peak at 1080 cm -1 . Boiling of the CB [5]:AuNP aggregates leads to an increase of the adatom mode.

Effect of HAuCl4 on AuNP size
AuNP gap size. A geometry optimisation for citrate in the gap-spanning configuration c2-t2t2 with increasing distance between the Au slabs is performed (Figure S12a). For an Au-Au centre distance of nearly ~1 nm, the c2-t2t2 mode transitions to the non-spanning t2t2. This calculation gives a good intuition about the maximum gap size for the gap-spanning mode. The calculation is also consistent with what is seen in SERS measurements for fully deprotonated citratecoordination without the central carboxylate can occur. Surface chlorides and monodentate citrate binding. To prove that surface chlorides are preventing citrate from bidentate binding, AuNP aggregates are formed with a range of mono and divalent salts of which three (MgCl2, CaCl2, NaCl) are chlorides.
Consistent with our hypothesis, all three chloride-salted AuNP aggregates ( Figure  S13a) show the strong C-O stretch at ~1300-1320 cm -1 (monodentate, labelled Mono) as a result of only one carboxylate oxygen coordinated to the Au surface. Simultaneously, the decreasing intensity of the symmetric carboxylate stretch at ~1380 cm -1 (Bi) shows that bidentate binding is reduced. The dominant mode of the chloride-salted aggregates is clearly the gap-spanning mode c2-t2t2 with the strong 995 cm -1 and weak 1020 cm -1 vibration. Control measurements with two citrate salts (magnesium citrate and trisodium citrate) show typical citrate signals with strong c2 and some c2-t2t2 coordination (only for Mg citrate), but no evidence of monodentate binding. The SERS spectrum for sodium acetate-salted aggregates show typical citrate signals as obtained for the citrate salts or the citrate signals on AuNP films. There is some acetate adsorbed on the Au surface (black arrow) as revealed by the characteristic ν(C-C) of acetate at 930 cm -1 (see Figure S8c). Despite acetate being adsorbed, there is no evidence of monodentate binding of the adsorbed citrate species.
Identifying surface chlorides in AuNP nano-gaps. The presence of adsorbed chlorides on the gold surface can be confirmed via low-wavenumber Au-Cl vibrations. Upon addition of chloride salts (MgCl2, CaCl2, KCl, NaCl) and also HCl to freshly aggregated AuNPs (salted with 60 mM TSC), peaks at ~170 and ~240 cm -1 emerge ( Figure S14). Both peak positions are characteristic Au-Cl vibrations which have been studied in the past. 5,6 Figure S14. Trisodium citrate-salted AuNPs and chlorides. (a) AuNP aggregates formed by "salting" with TSC. After aggregation, chloride salts/HCl are added and low-wavenumber SERS is recorded. Background of spectra is shifted; no scaling is performed.
SEM close-ups of AuNP films. For the investigation of citrate binding in an environment free from citrate by-products, AuNP films are used (see methods section of main text for details). These AuNP films consist of one to three layers of closepacked 80 nm AuNPs as evident from the SEM images (Figure S15, a-d). A close-up image of the films prior to plasma treatment (Figure S15c, red arrows) clearly shows signs of electron irradiation damage ("halo-effect") of organic compounds deposited locally onto the AuNP surface. The O2 plasma cleaned AuNPs with all organics removed do not exhibit this effect (Figure S15d). In general, it can be stated that oxygen plasma etching used here preserves the shape of the nanoparticles and their facets. Surface chlorides and monodentate citrate binding in AuNP films. Adding sodium chloride to TSC AuNP films leads to an increase in monodentate binding of citrate. Exactly the same effect is observed in colloidal gold clusters aggregated with NaCl ( Figure S16a). The intensity of the symmetric carboxylate stretch at ~1380 cm -1 decreases whilst a peak at ~1300 cm -1 (C-O-Au stretch) emerges. Also clearly visible (black arrow) is the presence of the C=O mode of the monodentate carboxylate. Citrate-AuNP films ageing. To investigate the ageing behaviour of these AuNP films, a sample is kept in a citric acid solution (100 mM pH=3.5) for 87 days. The SERS spectra (Figure S17, a-c) show the expected c2 coordination mode for fresh films characterised by a strong 1020 cm -1 vibration (red arrow). With increasing age, citrate starts to span the gaps with the modes c2-t2t2 and c2-t2 as indicated by the 995 (dark blue) and 975 cm -1 (light blue) peaks. Binding of the terminal carboxylates is further confirmed by the COO --Au stretches at ~1530 which develop over time. For the oldest sample (87 days), a peak at 1050 cm -1 (black arrow) suggests the presence of uncoordinated central carboxylates. This means that also t2 modes are present. As discussed in the main text, the lack of the 1080 cm -1 vibrations suggests that the underlying ageing pathway of AuNP films is different to the colloidal case.