Structure and luminescence of DNA-templated silver clusters

DNA serves as a versatile template for few-atom silver clusters and their organized self-assembly. These clusters possess unique structural and photophysical properties that are programmed into the DNA template sequence, resulting in a rich palette of fluorophores which hold promise as chemical and biomolecular sensors, biolabels, and nanophotonic elements. Here, we review recent advances in the fundamental understanding of DNA-templated silver clusters (AgN-DNAs), including the role played by the silver-mediated DNA complexes which are synthetic precursors to AgN-DNAs, structure–property relations of AgN-DNAs, and the excited state dynamics leading to fluorescence in these clusters. We also summarize the current understanding of how DNA sequence selects the properties of AgN-DNAs and how sequence can be harnessed for informed design and for ordered multi-cluster assembly. To catalyze future research, we end with a discussion of several opportunities and challenges, both fundamental and applied, for the AgN-DNA research community. A comprehensive fundamental understanding of this class of metal cluster fluorophores can provide the basis for rational design and for advancement of their applications in fluorescence-based sensing, biosciences, nanophotonics, and catalysis.


Introduction
Metal "nanoclusters" are the smallest of nanoparticles, consisting of only 2 to 10 2 metal atoms and possessing remarkable properties which are very nely tuned by cluster size, shape, and charge. Bare metal clusters have been studied for decades in order to understand how single atoms with quantized energy levels transition into the continuous properties of bulk materials. 1 Because the majority of unprotected metal clusters are unstable at ambient conditions, fundamental studies of metal clusters previously necessitated interrogation under ultra-high vacuum, 2 which limited practical applications of these nanomaterials. This challenge has been overcome by the use of stabilizing ligands and supporting surfaces to bring metal clusters into the "real world" for applications such as catalysis, photonics, and electronics. 3 In the past two decades, advances in synthetic chemistry have produced a "zoo" of different stable metal clusters passivated by molecular ligands, with cluster sizes that can even be tuned to atomic precision for especially ne control of their emergent properties. 4 This review concerns an especially unusual type of ligand-stabilized metal cluster, the DNA-templated silver cluster (Ag N -DNA), which combines the atomic precision of cluster science with the programmability of DNA nanotechnology.
Ag N -DNAs are relatively new entrants into the diverse zoology of metal clusters, with unique properties that arise from their polynucleic acid ligands. Following work by the Dickson group on silver clusters stabilized in dendrimers 5 and silver oxide lms, 6 in 2004, Petty, Dickson, and co-authors reported formation of uorescent silver nanoclusters exhibiting 400-600 nm electronic transitions by chemically reducing an aqueous mixture of single-stranded cytosine-rich DNA and AgNO 3 . 7 They then found that certain Ag N -DNAs exhibit very bright uorescence 8 and signicant photostability and can be harnessed as biolabels. 9, 10 Gwinn, et al., showed that the uorescence colors of Ag N -DNAs depend sensitively on nucleobase sequence and that Ag N -DNAs prefer to form on single-stranded (ss) DNA rather than double-stranded (ds) DNA, 11 motivating the important role played by silver-nucleobase interactions in Ag N -DNA formation. In the next few years, Ag N -DNAs were shown to be effective sensors for toxic metal ions, 12 polynucleic acids, [13][14][15] and other biomolecules. 16 Together, these and other early studies generated considerable interest in harnessing DNA's sequence programmability for custom design of Ag N -DNA uorophores tailored for precise sensing, uorescence microscopy of cells and tissues, and direct integration into DNA nanotechnology schemes. [17][18][19] The most remarkable characteristic of Ag N -DNAs is their sequence-dependent uorescence. By employing DNA template strands with wide-ranging nucleobase sequences, a diverse color palette of Ag N -DNAs with uorescence emission colors of 450 nm to 1000 nm has been developed 22,23 (Fig. 1A), with quantum yields as high as 93% (ref. 24) and Stokes shis as large as 5893 cm À1 . 25 Ag N -DNA uorescence may be excited by at least two pathways, either directly at the cluster's size-, shape-, and charge-dependent excitation peak or universally via the DNA bases (Fig. 1B). 21,26 Ag N -DNAs also exhibit unusual photophysics, 27 intriguing dark states which can be harnessed for background-free uorescence microscopy, [28][29][30][31] light-up or colorswitching behavior induced by various stimuli, 13,32-41 and catalytic activity. 42,43 Most well-studied ligand-stabilized metal clusters are protected by monolayers of small molecules such as thiolates 44 and phosphines, with sizes smaller than or comparable to the metal clusters themselves. 45 In contrast, Ag N -DNAs and their lessstudied counterparts, Ag N -RNAs, 46 are protected by bulky polynucleic acids much larger than the silver cluster. The structure and properties of these and other metal clusters stabilized by large macromolecular ligands, including proteins 47 and dendrimers, 48 are less understood than for monolayer-protected clusters, in part because bulky ligands can obscure resolution of cluster(s) and challenge crystallization, a necessary step for "solving" structure by X-ray crystallography. However, macromolecular ligands can also endow functionalities without the need for ligand exchange, adding a degree of versatility to applications of Ag N -DNAs and other macromolecule-stabilized nanoclusters.
Ag N -DNA synthesis is facile and is typically carried out by borohydride reduction of a solution of Ag + and ssDNA in neutral pH aqueous solution (Fig. 1C). This method is robust to varying solution compositions, stoichiometries, and specic mixing/ heating. 7,8,11,[49][50][51][52] In contrast to the simplicity of synthesis, achieving compositionally pure solutions of Ag N -DNAs is more challenging because reduction forms a heterogeneous mixture of silver-bearing DNA products containing varying numbers of silver atoms, N tot , and numbers of DNA strands, n s . The majority of these products are nonuorescent 53 and include clusters, Ag + -DNA complexes, and larger silver nanoparticles. 54,55 It is also possible for a given DNA template to stabilize multiple different emissive cluster species, 56 as has been observed for up to 25% of randomly selected DNA template sequences. 57 Due to characterization of as-synthesized Ag N -DNAs without purication and/or due to fragmentation during mass spectrometry (MS), early reports underestimated Ag N -DNA sizes 8,11 or found no correlation of uorescence color with silver cluster size. 58 A lack of awareness of this heterogeneity continues to hinder accurate characterization of Ag N -DNAs, and the assumption that the composition of Ag N -DNAs is uncorrelated to the optical properties of these nanoclusters still persists. 59 The challenge of heterogeneity has been overcome by the use of reversed-phase high performance liquid chromatography (HPLC) 53,60 and size-exclusion chromatography (SEC) 61,62 to isolate a uorescent Ag N -DNA of interest prior to compositional and spectral characterization. Additionally, development of gentle electrospray ionization (ESI) MS now enables compositional analysis without fragmentation of the Ag N -DNA product. 24,53,63 Using tandem HPLC-MS with in-line UV/Vis and uorescence spectroscopy, Schultz, et al. determined the compositions of several uorescent Ag N -DNAs with uorescence emission wavelengths, l em , ranging from green to near infrared (NIR), nding that these clusters contained N tot ¼ 10-21 Ag atoms stabilized by n s ¼ 1-2 copies of the templating DNA strand. 53 This ability to isolate and characterize compositionally pure solutions of Ag N -DNAs has enabled numerous future studies, leading to dramatic advances in our understanding of the structure-property relations of these nanoclusters, which we discuss in Section 3, and of their photophysical properties, which we discuss in Section 4.
This review focuses on the recent advances in fundamental understanding of Ag N -DNAs, with a particular emphasis on the recent detailed studies of compositionally pure Ag N -DNAs. We note that this review is timely because previous reviews which primarily focused on fundamental structure and properties [64][65][66][67] are several years old and do not discuss recent breakthroughs, including the rst reported Ag N -DNA crystal structures. 22,25,68,69 Readers may also nd a comprehensive list of DNA sequence/ structure and optical properties for a large number of Ag N -DNAs by New, et al., 70 as well as previous reviews focused on the emerging applications of Ag N -DNAs as sensors and biolabels. 64,[71][72][73][74][75] Here, we summarize what is known about the connections among DNA sequence, Ag N structure, and photophysical properties. We rst review current understanding of the Ag + -mediated DNA base paired structures that are the synthetic precursors of Ag N -DNAs (Section 2). Next, we discuss current models for the structures of Ag N -DNAs, which have rapidly advanced due to detailed studies of compositionally pure Ag N -DNAs and a few breakthrough crystal structures (Section 3). 22,25,68,69 Then we review current understanding of the excited state processes which lead to uorescence in Ag N -DNAs and the unusual dark states exhibited by Ag N -DNAs (Section 4). We then discuss recent work to decode how DNA sequence selects Ag N -DNA properties by combining high-throughput experimentation and machine learning (Section 5). Finally, we review work on merging structural DNA nanotechnology with Ag N -DNAs for Fig. 1 (A) The fluorescence colors of Ag N -DNAs, which are selected by DNA sequence, span a large spectral range from visible to NIR wavelengths and are correlated with cluster size. 20 (B) Ag N -DNA excitation spectra exhibit a dominant peak in the visible to NIR spectral range as well as a UV excitation band corresponding exactly to the DNA template strand. Fluorescence spectra excited via the DNA bases (inset, purple) have the same lineshapes as spectra excited at the cluster's unique visible to NIR transition. 21  ordered arrangement of these nanoclusters (Section 6) and comment on opportunities and challenges facing the eld of Ag N -DNA research (Section 7). It is our intent to provide a comprehensive and current picture of the properties of Ag N -DNAs which is accessible to researchers from many backgrounds, in order to aid others in developing applications of these unique nanoclusters and to inspire new experimental and computational studies of their fundamental properties.
2. Silver-mediated base pairingprecursors to Ag N -DNAs A complete understanding of Ag N -DNA structure and sequencedependent properties naturally begins with an understanding of Ag + -DNA complexation. This is because (i) Ag N -DNAs are formed by chemical reduction of Ag + -DNA complexes, 76 (ii) high-resolution MS of HPLC puried Ag N -DNAs shows that usually about half of the silver atoms within Ag N -DNAs remain cationic, 24 meaning that Ag + -DNA interactions play a key role in determining Ag N -DNA structure, and (iii) Ag + -DNA interactions are highly sequence-dependent, 54,55 which may lead to the sequence dependence of Ag N -DNA size and uorescence properties. Here, we review recent advances in fundamental understanding of Ag + -nucleobase interactions and secondary structures of Ag + -DNA complexes, with a focus on properties relevant to the formation and sequence-dependence of Ag N -DNAs. We note that this topic is a small part of the rich eld of metal-mediated nucleobase pairing, an area of great interest as a route to expanded base-base interactions, DNA-based electronics, and sensing. We do not attempt to review this entire eld here and point to excellent comprehensive reviews elsewhere on metal-mediated pairing of both natural and articial bases [77][78][79][80] and in the specic case of Ag and Au for natural DNA. 81

Watson-Crick base pairing
The four natural nucleobases of DNA are adenine (A), cytosine (C), guanine (G), and thymine (T). In canonical Watson-Crick (WC) pairing of dsDNA in B form, which is the most common structure of DNA in vivo ( Fig. 2A), two complementary DNA strands join by hydrogen bonds ("H-bonds") between A and T and between C and G, forming the familiar antiparallel double helix. C and G are held together by three H-bonds between the O2, N3 and N4 positions of C and the O6, N1 and N2 positions of G. In like manner, T and A are H-bonded through the O2 and N3 positions of T, and the N6 and N1 from A (Fig. 2C). This difference in the number of H-bonds between nucleobase pairs results in a weaker A-T WC bond as compared to C-G. The WC B-form double helix is further stabilized by hydrophobic stacking interactions between neighboring nucleobases. Additional less common DNA structures also exist, including WC paired A-DNA 84 and Z-DNA 85 and Hoogsteen base pairing. 86 The extensive scientic understanding of DNA structure and thermodynamics has enabled the birth of DNA nanotechnology, which exploits DNA as a fundamental materials building block, 82 engineering DNA sequence to achieve self-assembled predened shapes, 18,87 tuned colloidal interactions, 88-90 and molecular computation. 91,92 2.2 Ag + -nucleobase interactions of homobase strands Silver cations (Ag + ) are well-known to prefer binding to DNA nucleobases over the phosphate backbone at neutral pH. 93 (Hg 2+ possesses similar preference, 79,93-95 but its signicant toxicity prohibits applicability). This preference enables Ag + intercalation into single base mismatches in WC-paired dsDNA, typically by interactions with nucleobase ring nitrogens. 93,96,97 Cytosine (C) is especially well-known for affinity to Ag + , and this has been harnessed to expand the interactions among DNA oligomers, enabling Ag + -paired C-C mismatches, 96 Ag + -folded imotif secondary structures in C-rich DNA, 98 Ag + -crosslinked DNA hydrogels, 99 and DNA nanotubes. 100 More recently, the study of Ag + -mediated nucleobase pairing has been extended to consider DNA that is unconstrained by WC base pairs. These studies show that silver can completely rearrange canonical DNA structures, as opposed to simply intercalating within base pair mismatches. Here, we review these recent advancements to provide context for the sequence-property connections that govern Ag N -DNAs (Section 5).
To understand how Ag + complexes with DNA in the case where the DNA does not form WC base pairs, Swasey, et al., investigated interactions of Ag + with homobase DNA strands. 54 Aer solvent-exchanging DNA oligomers to remove any residual salts from oligomer synthesis, DNA was mixed with AgNO 3 in an aqueous solution of ammonium acetate, followed by thermal annealing at 90 C. Resulting products were analyzed by highresolution negative ion mode ESI-MS to determine absolute composition by resolving the isotopic distribution (discussed in Section 3.1). Fig. 3A shows the compositions of all observed products for 11-base homobase strands. While C is best-known for affinity to Ag + and was shown by Ritchie, et al., to form Ag +mediated duplexes, 52 G was actually found to associate the greatest number of Ag + , with order of affinity: G > C > A > T. While the 4 types of natural nucleobases all formed Ag + -bearing single homobase strands, Ag + also mediates formation of homobase duplexes for C and G. When two different single homobase strands are mixed, Ag + only mediates the heteroduplex A-Ag + -T, completely replacing the WC A-T duplex. Ag + also disrupts WC-paired C-G duplexes to instead form C-Ag + -C and G-Ag + -G homobase duplexes. Fig. 3B summarizes all observed pairing between homobase strands. C-Ag + -C and G-Ag + -G homoduplexes are remarkably stable, with C 6 -Ag + -C 6 and G 6 -Ag + -G 6 homoduplexes remaining intact at 90 C, while C 6 -G 6 WC duplexes melt below 20 C (Fig. 3C). 54 Quantum chemical calculations support greater stability of Ag + -mediated homoduplexes for C and G than for A and T. In the absence of steric factors, (base-Ag + -base) N duplexes have higher bond energies than (base-Ag + ) N structures. Because C-Ag + -C and G-Ag + -G are nearly coplanar, with dihedral angles of 171.9 and 181.2 respectively, while T-Ag + -T and A-Ag + -A are nonplanar, with dihedral angles of 140 and 101.6 , respectively, C-Ag + -C and G-Ag + -G homoduplexes are expected to be signicantly more stable (Fig. 3D). The A-Ag + -T bond is also non-coplanar, but its stability could be explained by the difference in size between A and T, which still allows adenine stacking interactions. 54 The nucleobase sites with which Ag + interacts differ from WC pairing. Simulations by the Lopez-Acevedo group have determined that pyrimidines C and T interact with Ag + at the N3 position, 54,101 which is deprotonated for thymine, while purines A and G coordinate with Ag + at the N7 position. 54 These binding sites correspond to the Hoogsteen region (Fig. 2C). However, these positions might change depending on the other nucleobase of the Ag + -bridging bond, as is the case for the C-Ag + -G bond, reported by Kondo, et al., where the interaction with the purine base is through the N1 position, which is deprotonated. 102 2.3 Ag + mediates parallel strand orientation of highly stable homobase duplexes Quantum chemical and hybrid quantum mechanics/molecular mechanics (QM/MM) calculations by the Lopez-Acevedo group predicted that, unlike the antiparallel strand orientation of natural WC duplexes where one 5 0 -3 0 strand pairs to a complementary 3 0 -5 0 strand, 82 C-Ag + -C duplexes and G-Ag + -G prefer a parallel orientation, with 5 0 ends aligned. 101,103,104 These helical duplexes align Ag + along the helix axis and are stabilized not only by Ag + -nucleobase interactions but also by novel interplanar H-bonds ( Fig. 4A and B). 101,103,104 Calculated electronic CD spectra of C 2 -Ag + -C 2 tetramers agree well with experimentally measured CD spectra, further supporting a parallel arrangement. 101 However, other experimental studies report varying behavior. One study of the conductivity of C-Ag + -C duplexes achieves antiparallel duplex formation of strands conned at ends to a metal surface and scanning probe tip. 105 As  Recent study of unconstrained homobase strands conrms parallel duplex structure for C-Ag + -C and G-Ag + -G by utilizing Förster Resonance Energy Transfer (FRET) experiments to determine DNA strand orientation and ion mobility spectrometry (IMS) MS coupled with density functional theory (DFT) calculations to elucidate structure. 108 Variations in FRET efficiency between donor and acceptor dyes coupled to ends of two DNA strands support parallel Ag + -paired C homobase duplexes and G homobase duplexes (Fig. 4C). This parallel orientation was further demonstrated by IMS-MS experiments coupled with DFT calculations of collision cross sections (CCS), which support high aspect ratios for both guanine and cytosine duplexes, consistent with rigid, wire-like structures (Fig. 4D). Based on CCS values and their agreement to calculated values, the G-Ag + -G duplex is found to be more rigid because nucleobases form additional H-bonds with the phosphate groups in the backbone, whereas the C-Ag + -C duplex lacks these extra bonds and is more exible.

Ag + -nucleobase interactions of mixed base strands
The vast majority of reported Ag N -DNA nanoclusters are stabilized by DNA strands with mixed base sequences. To understand how heterobase strands recruit Ag + , Swasey and Gwinn examined ten noncomplementary 11-base DNA strands, determining composition of Ag + -DNA complexes by ESI-MS (HPLC-MS was employed to analyze very heterogenous samples). 55 Interestingly, strands with sequences formed by single-base "mutations" of C 11 increase the distribution of the number of Ag + attached to duplexes, and inclusion of mutations in G 11 homobase strands can signicantly increase the average number of Ag + by up to 7 or 8 Ag + per duplex (Fig. 5A). Both homobase and heterobase Ag + -mediated duplexes were found to be stable in various solution conditions, signicant Mg 2+ concentrations, and high concentrations of urea (a strong denaturant). While the chemical structures adopted by these heterobase duplexes are not known, the differences in Ag + recruitment have important implications for the origins of Ag N -DNA sequence dependence, which we discuss in Section 5.
Kondo, et al., recently developed remarkable uninterrupted Ag + -DNA "nanowires" and solved their 3D structure, determining formation of consecutive Ag + -paired duplexes with antiparallel orientation. 102 The DNA strand used to form the Ag + wires, GGACT( Br C)GACTCC, is a near-complement which forms a WC-paired homodimer with one C-C mismatch at room temperature in biologically relevant salt concentrations (determined using UNAfold soware 109,110 ). C-Ag + -C, G-Ag + -G, T-Ag + -T, and C-Ag + -G bonds were observed in the nanowire, and interestingly, not all nucleobases in the strand participate in the principal linkage between strands. A's protrude outwards (Fig. 5B) and contribute to crystal-packing through formation of AT-Ag + -A triplets and AA stacking interactions. Thanks to the near reversibility of the sequence used, and because A's do not participate in duplex bonding, most nucleobases are bonded to a like base in the partner strand, with only two C-Ag + -G pairs (C) Emission spectra of Ag + -mediated C 20 and G 15 duplexes labeled with donor (green dot) and acceptor (red dot) dyes at 5 0 end and 3 0 end, respectively (orange curve) or with both dyes at 3 0 ends (blue), compared to emission of the donor-bearing strand alone (blue dotted curve). Excitation is at 450 nm, which directly excites the donor only. Significant quenching of donor emission with concomitant acceptor emission (high FRET efficiency) clearly demonstrates that Ag + -mediated pairing of homo-duplexes arranges strands in a parallel orientation. 108 (D) DFT-optimized structures of Ag + -DNA duplexes of G 20 and C 20 compared to WC duplexes of observed. Pairing between the two strands occurs with a oneposition shi, enabling formation of nanowires up to 0.1 mm long. Despite the antiparallel orientation and the C-Ag + -G pairs, in which the G is bonded through the N1 position, the system clearly does not obey WC pairing because the main interaction sites lie in the Hoogsteen region. Furthermore, the propeller twist angles obtained are larger than in WC pairing, which can be explained by repulsions between amino and carbonyl groups of opposite bases. 102 Liu, et al. solved the 3D structure of another Ag + -paired mixed base strand, 5 0 -GCACGCGC-3 0 , which forms curved dimers attached by one G-Ag + -G bond and one C-Ag + -C bond, with parallel strand orientation (Fig. 5C). 106 In this structure, it is only like bases which participate in Ag + -mediated pairing, and these base pairs are less planar (Fig. 5C) than the nearly coplanar angles predicted by previous DFT calculations. 54 This suggests that mixed base strands can accommodate a wide range of Ag + -mediated base interactions beyond just linear wires. This 8-base sequence was also uncovered in an unrelated study using machine learning methods to design templates for Ag N -DNAs with uorescence emission in the 600 nm < l em < 660 nm window. 111 This surprising coincidence suggests that some Ag N -DNAs are formed by chemical reduction of nontrivial Ag + -DNA complexes.
Very recently, the Kohler group reported evidence for a parallel oriented Ag + -mediated duplex of C 20 with signicant "propeller" twist of the C-Ag + -C base pairs, as has been reported in the studies above. This evidence was based on strong agreement between experimentally measured and calculated CD spectra. 113 The authors note that such twisting has been associated with reduced exibility of DNA, 114 and this enhanced rigidity agrees with the past IMS studies of C-Ag + -C duplexes described in Section 2.3. 108 2.5 Relevance of Ag + -mediated base pairing for Ag N -DNAs As synthetic precursors of Ag N -DNAs, 76,115 Ag + -DNA complexes are the scaffolds that reorganize into the cluster-stabilizing cage of an Ag N and, at least in part, provide the Ag + "fuel" to grow the Ag N upon reduction. Early studies which found that Ag N -DNA do not form on completely dsDNA templates 11 have led to the false assumption that the Ag N can always be conned within singlestranded regions of WC-paired DNA structures such as hairpins 116,117 or other dsDNA structures with ssDNA regions, 118 based on the assumption that WC DNA secondary structure is preserved in the presence of Ag + . The dramatic rearrangement of DNA homobase and heterobase strands by Ag + , together with the signicant thermal and chemical stabilities of Ag + -mediated DNA duplexes, 54,55 call into question whether this assumption is accurate. It is more likely that Ag + can invade and unravel WC dsDNA under appropriate conditions, rearranging secondary and tertiary structures which then further evolve upon chemical reduction. This has been suggested by several careful studies, 49,[119][120][121] and Ag + has also been shown to rearrange the well-known G-quadruplex structure 108 and i-motif structure. 113 Further studies will be needed to determine to what degree DNA secondary structure is preserved aer Ag N -DNA synthesis, especially when Ag N -DNAs are incorporated into the larger DNA structures discussed in Section 6.  Protruding adenines foster assembly of multiple wires into 3D lattices. 102 Silver atoms are shown in gray and potassium atoms in purple. Image created from PDB ID 5IX7 with NGL Viewer. 112 (C) Structure of a dimer of 5 0 -GCACGCGC-3 0 (orange, green) paired by two Ag + (grey). The third Ag + (bottom right of structure) supports supramolecular assembly of the structure during crystallization. 106 Image created from PDB ID 5XJZ with PyMOL.
advancements have been enabled by compositionally pure Ag N -DNAs isolated using HPLC 53 or SEC. 37,38,62 These techniques separate different DNA complexes by exploiting variations in size and polarity that are induced by different silver products on the DNA template strands. (Methods for isolating Ag N -DNAs using HPLC have been reviewed in detail previously. 66 ) Purication prior to characterization is crucial because assynthesized solutions contain multiple dark and uorescent products, including Ag nanoparticles, Ag N -DNAs and Ag + -DNA complexes, as supported by LC-tandem MS. 24,57 Even though one would naively expect Ag N -DNA properties to be similar in the as-synthesized and puried states, a recent report by Gambucci, et al., showed different rotational correlation times, indicating that synthesis fragments could be attached to the Ag N -DNAs, e.g. by Ag + -mediated interactions. 122 Compositional analysis methods that only infer average stoichiometry of the entire heterogeneous as-synthesized solutions may misjudge the number of silver atoms within an Ag N -DNA and cannot resolve the number of DNA strands n s that stabilize a single cluster, and MS performed directly on as-synthesized samples makes it challenging to identify the uorescent Ag N -DNA of interest from the other products formed during synthesis. Here, we primarily review structural studies of HPLC-isolated Ag N -DNAs with bright visible or NIR uorescence, which have thus far been found to contain N tot ¼ 10-30 Ag atoms, 23,24,53,57,123 as opposed to earlier reports of dimers or trimers of Ag. 8, 11 We then discuss other studies that focus on inference of the conformation of the DNA template strand(s) around the Ag N . Unless indicated, all Ag N -DNAs discussed are compositionally pure.

Mass spectrometry to determine Ag N -DNA composition
Prior to the breakthrough crystallographic structures of Ag N -DNAs solved in 2019, 22,69 efforts to discern Ag N -DNA structure mainly employed correlations of experimentally measured absorption, excitation, and/or emission for Ag N -DNAs of known composition with computational studies or simple models. These past studies do not provide the same level of structural detail as the recent crystal structures but do provide a more comprehensive picture of the structure-property relations of Ag N -DNAs in general, with detailed studies on about 20 different HPLC-puried Ag N -DNAs as compared to the smaller number of crystal structures currently available. 22,25,68,69 Since metal cluster size, charge, and geometry strongly determine properties, accurate characterization of composition is a key step towards building a fundamental understanding of metal clusters. 45 It is well-established that other ligandstabilized metal clusters are only partially reduced because a fraction of the metal atoms in the cluster are bound to the surrounding ligands and that the number of remaining effective valence electrons in the cluster core is a major determinant of the electronic properties of the cluster. [124][125][126][127] Partially oxidized Ag N -DNAs were proposed by Ritchie, et al., based on the oxygen and chloride dependence of the uorescence of C 12 -stabilized clusters. 128 An experimental method to not only count the numbers of silver atoms N tot and DNA strands n s in a puried sample of Ag N -DNA but also to separate N tot into neutral (N 0 ) and cationic (N + ) silver content can yield insights into how Ag N clusters are ligated to the DNA and enable computational studies of their electronic properties. 126,127,129 High resolution mass spectrometry (HR-MS) is an ideal tool to achieve this goal because HR-MS can be used to determine both ion mass, M, and charge, Z, rather than just the ratio M/Z, by resolving the isotope pattern that arises due to natural variation in isotopic abundances of elements. Koszinowski and Ballweg determined the charge of an Ag 6 4+ -DNA by comparing the experimentally measured isotope pattern to the calculated distribution of this cluster. 63 To characterize the properties of uorescent Ag N -DNAs, this approach has been developed in conjunction with chromatographic purication by the Gwinn and Petty groups. 24,130 Because DNA is easily deprotonated, negative ion mode ESI-MS is suitable to resolve weakly bound, noncovalent DNA complexes 131,132 and has been used to size a variety of silverbearing DNA complexes. [53][54][55]57,60,63,108 (While more sensitive, positive mode ESI-MS can oxidize encapsulated clusters during electrospray, hindering full determination of composition. 133 ) The mass spectrum of an Ag N -DNA product may be collected either by tandem HPLC-MS ( Fig. 6A) or by direct injection into the MS following previous purication. Determination of N 0 and N + for an Ag N -DNA from its mass spectrum is illustrated in Fig. 6. First, the charge state Z-of a M/Z (mass to charge ratio) peak is determined by the spacing between adjacent peaks of the isotope pattern: these peaks are spaced by 1/Z (Z is dened as a positive integer). For example, Fig. 6B shows the 7-charge state (Z ¼ 7, minus sign is due to negative ion mode MS) of an Ag 30 -DNA, with individual isotope peaks separated by 1/7. 23 (The product shown in Fig. 6B has a charge of -7e, where e is the fundamental unit of charge). The total charge of the complex corresponding to this M/Z peak is equal to the charge of the number of silver cations, eN + , minus the charge of the number of protons removed from the DNA, en pr , to reach the total charge of ÀeZ: Note that as number of silver cations, N + , increases, more protons must be removed from the complex to reach charge state Z. Then, because n pr protons have been removed from the Ag N -DNA complex, the measured total mass M (in amu) is: where m DNA is the DNA template strand mass, n s is the number of DNA strands in the complex, and m Ag is the silver atom mass (the mass of a proton is treated here as 1 amu). In the case of well-resolved patterns, N + and N 0 may be determined by calculating the isotope distribution pattern for varying values of N + , and thus n pr , to determine the charge which best matches the isotope pattern (Fig. 6B). 24 If signal is too low to precisely resolve the isotope pattern, charge may be inferred by comparing Gaussian ts of the calculated and experimentally measured isotope patterns. 55 Using this method, Schultz, et al., determined that approximately half of the silver atoms within Ag N -DNA are cationic in nature. 24 HR-MS is advantageous for determination of n s , N 0 , and N + without ambiguity, provided that gentle enough ESI is applied. Inductively coupled plasma-atomic emission spectroscopy (ICP-AES) has also been used to determine the composition of puried Ag N -DNAs, 37 although n s cannot be determined by this method. This has led to underestimates of the sizes of Ag N -DNAs with n s > 1, which were later characterized by HR-MS. 24

Experimental evidence for elongated cluster geometry
The rst experimental evidence that Ag N -DNA cluster geometry differs from globular (or quasi-spherical) arose by comparing the absorption spectra of compositionally pure Ag N -DNAs (whose N + and N 0 were determined by HPLC-MS) to the experimental and computed spectra of bare Ag N in the gas phase which have similar numbers of effective valence electrons (equal to N 0 ). The electronic properties of ligand-stabilized metal clusters depend on the number of effective valence electrons in the cluster core, not only the total number of atoms N tot , and these valence electrons can delocalize to form "superatomic" orbitals. 125 Thus, it is most appropriate to compare the properties of Ag N -DNAs with bare silver clusters having like numbers of effective valence electrons. Due to ligation with the nucleobases, 24 not all Ag atoms in an Ag N -DNA will contribute to the valence electron count. To determine the effective valence electron count of an Ag N -DNA, we subtract the charge of the cluster, N + , from the total number of atoms in the cluster, N tot , nding that the number of effective valence elec- Schultz, et al., found that the numbers and locations of peaks in the optical spectra of Ag N -DNAs differ markedly from their globular bare cluster counterparts. Naked Ag N with cluster sizes N ¼ 2 to 20 exhibit globular geometries and absorption spectra with multiple UV transitions in the 3 to 5 eV spectral range. 134,135 In contrast, puried Ag N -DNAs have much simpler spectra with single dominant peaks in the visible to NIR range <3 eV, whose locations strongly depend on N 0 , 24,53 and an additional UV absorption band due to the DNA ligand (Fig. 1B). 21 The energies of the visible to NIR absorbance peaks of Ag N -DNAs with varying N 0 can be described by quantum chemical calculations by Guidez and Aikens for linear atomic chains of silver (Fig. 7A). 136 Based on these results and on the signicant degree to which Ag N -DNA emission is polarized, as observed by spectroscopy of single Ag N -DNAs, a rod-like structure for Ag N -DNAs was proposed by the Gwinn group. 24 Following this model, Ramazanov and Kononov used DFTcalculated electronic excitation spectra to argue that threadlike clusters show better agreement with experimental data than planar clusters. 137 A rod-like geometry is also supported by the magic N 0 numbers of Ag N -DNAs. The energetic stability of many ligandstabilized metal clusters can be described by the "superatom" In this illustration, the initial sample (yellow tube) is a mixture of products including multiple dark Ag-DNA complexes, one green-fluorescent Ag N -DNA species, and one red-fluorescent Ag N -DNA species. The as-synthesized Ag N -DNA solution is injected into an HPLC outfitted with a core-shell C 18 column for reverse-phase, ion-pair (IP) HPLC. Products are separated due to slight variations in column affinity with a water-methanol gradient and a triethyl ammonium acetate (TEAA) IP agent. By monitoring both absorbance at $260 nm, which correlates to the absorbance of DNA, and fluorescence emission (e.g. UV-excited fluorescence 21 ), correlation of absorbance and fluorescence chromatogram peaks indicate elution of a fluorescent Ag N -DNA species. We note that the chromatogram schematics are simplified for illustration; real chromatograms are more complex. 53 Products of interest can either be sized by in-line negative-ion mode ESI-MS or collected for subsequent ESI-MS. A mass spectrum for a previously studied 30-atom NIR-emissive product is shown in the bottom right. 23 Both monomeric and dimeric (labeled "D") products are visible, with spacing of the isotopic peaks indicating the charge state of each product (labeled as superscript of "D") for dimeric products. (B) Experimental mass spectrum of the Ag 30 -DNA product at the 7À charge state dimeric product (labeled D À7 in (A)) is shown in black, with the calculated mass distribution (green bars) for a product with 2 DNA strands, N 0 ¼ 12 Ag 0 , and N + ¼ 18 Ag + . 23 Inset: compares the experimental spectrum 23 with the calculated distribution for a product with no charged silvers (2 DNA strands and 30 Ag 0 ), illustrating how the shift between the experimental and calculated isotopic finger distribution can be used to accurately determine the numbers of Ag 0 and Ag + in an model, which states that the effective valence electrons in the cluster core are characterized by an electronic shell structure, similar to the shell structure of the atomic nuclei. 124,125 For spherical metal clusters, closed shells are expected for N 0 ¼ 2, 8, ., resulting in enhanced abundances of clusters of these sizes due to their signicantly enhanced stabilities (the same behavior is observed for gas phase bare metal clusters 2 ). Copp, et al., performed a large-scale study of Ag N -DNAs stabilized by $700 different DNA templates, nding enhanced abundances of Ag N -DNAs with even numbers of neutral silver atoms: greenemissive Ag N -DNAs with N 0 ¼ 4, red-emissive Ag N -DNAs with N 0 ¼ 6, and larger NIR-emissive N 0 ¼ 10-12 Ag N -DNAs (Fig. 7B); the spherical magic numbers of 2 and 8 were not especially abundant (Fig. 7C). This behavior is consistent with clusters that are signicantly aspherical, for which additional energy stability is primarily conferred by pairing of electron spins, resulting in enhanced stabilities for even values of N 0 . 57 Chiroptical properties of Ag N -DNAs have been well-modeled by a thread-like cluster structure. Because circular dichroism (CD) spectroscopy is extremely sensitive to specic geometrical structure and can be calculated using rst-principles methods, CD allows a direct interface with theory. Swasey, et al., measured the CD spectra of four Ag N -DNAs spanning the visible to NIR color palette. Quantum chemical calculations for bare atomic Ag chains with a chiral twist agree well with the experimental spectra. 138,139 Similarity between CD spectra of Ag N -DNAs and their unreduced Ag + -DNA precursors was also observed, pointing to the role played by the Ag + -DNA complex in dictating nal cluster structure 138 (we note that recent studies suggest the Ag N itself is not the cause of the CD signal observed for Ag N -DNA but that the DNA-silver interaction of the intrinsically chiral DNA plays a crucial role in generating chiroptical properties of these clusters 127 ).
Past studies have found that classical theories which describe collective electronic excitations of colloids, 140 such as Mie-Gans theory, 141,142 show surprising agreement with the optical properties of small metal clusters, 134,143 particularly for longitudinal plasmonic modes. 144 Copp, et al., examined whether Ag N -DNAs can also be described by classical models, applying Mie-Gans theory to HPLC-puried Ag N -DNAs with 400-850 nm cluster excitation wavelengths and numbers of effective valence electrons, N 0 , determined by HR-MS in order to elucidate the aspect ratios of these clusters. Application of Mie-Gans theory to this experimental data predicted prolate cluster geometry, with aspect ratios of 1.5 for N 0 ¼ 4 up to $5 for N 0 ¼ 12. (The currently reported crystallographic structures for Ag N -DNAs do not yet have determined charges, 22,25,68,69 so these aspect ratios remain unconrmed by solved structures.) Ag N -DNAs with N 0 $ 6 displayed shis in peak excitation wavelength dependent on solvent dielectric, as is expected for a collective electronic excitation 145 and observed for larger metal nanoparticles; 146 such sensitivity may be useful for applications. The increase in peak excitation wavelength and extinction coefficient with increasing cluster core size N 0 (ref. 147) is a characteristic shared by the longitudinal collective electronic excitation sustained by rod-like metal clusters 136,144,148 and larger metal nanoparticles. 145,149,150 While the proper denition of a plasmon versus a collective electronic excitation at the cluster scale remains debated, other molecular-scale systems have also been shown to exhibit plasmon-like behavior. [151][152][153][154] The Sánchez group simulated toy model Ag N -DNAs with magic number sizes, 57 nding that a neutral silver cluster rod surrounded by nucleobase-bound Ag + is generated when a partial charge is placed on the cluster. When excited, these clusters supported longitudinal plasmon-like modes. 129 Intriguingly, single Ag N -DNAs studied at temperatures below 2 K exhibit surprisingly broad spectral linewidths. 155 For larger nanoparticles, surface plasmon resonance broadening is understood to arise from dephasing processes for multiple delocalized electrons, 156 but such effects are less well understood at the cluster scale. As silver cluster rods, Ag N -DNAs may provide a unique platform to investigate these important questions. It remains to be determined whether the optical transitions in Ag N -DNAs are collective or plasmonic-like, and further experimental and theoretical studies are needed to reveal which models are most suitable to represent the behavior of Ag N -DNAs. 3.3 X-ray and IR spectroscopy of solution-state Ag N -DNAs Several groups have applied X-ray spectroscopy, nuclear magnetic resonance (NMR), and infrared (IR) spectroscopy to probe the structures and silver-DNA interaction in puried Ag N -DNAs. To interrogate stoichiometry, oxidation state, ligand environment, and structure of a violet-absorbing Ag N -DNA, Petty, et al. used ESI-MS, X-ray absorption near edge structure (XANES), and Extended X-ray Absorption Fine Structure (EXAFS). 130 This dimly uorescent cluster has absorbance peaked at 400 nm but converts into a NIR-emissive species upon perturbation of its DNA template strand. 33 MS data (Fig. 8A) and Ag L 3 edge XANES spectra establish the SEC-puried violet cluster to be an Ag 10 6+ . CD spectra of this Ag 10 6+ remain stable above 70 C, pointing to the temperature stability of the DNA-Ag interaction (Fig. 8B). Ag K-edge EXAFS was used to probe organization of Ag atoms and Ag-nucleobase interactions. The experimental EXAFS trace (black) was tted to three individual scattering paths (Fig. 8C) to infer specic bond lengths and coordination numbers. Based on these results, the authors proposed an octahedral cluster structure (Fig. 8D). While creating an accurate model from EXAFS data is nontrivial given the vast number of possible geometries in such a complicated system, the model contains several structural elements later found in the crystal structure of an Ag 16  This altered cluster has the same oxidation state as the "violet" cluster above and can be reversibly converted by manipulation of the hairpin region. The altered cluster is highly uorescent and has red-shied absorbance. Using differences in EXAFS data between the two clusters, the altered cluster is proposed to have a more extended and distinct metal-like core, presumably due to variations in coordination with the DNA ligand. These variations are supported by later studies using activated electron photodetachment MS. 121 Volkov and co-workers used X-ray photoelectron spectroscopy (XPS) to study an HPLC-puried Ag N -DNA. 158 The oxygen spectra are similar with and without Ag + , supporting that Ag + prefers to bind to nitrogen when no reducing agent has been added. For the puried Ag N -DNA, binding of silver to oxygen atoms was present, suggesting that the interacting oxygens belong to the sugar moiety and/or phosphodiester bond. The crystal structures by Cerretani, et al., found Ag atoms bound to the phosphate group, conrming this observation. 25,68,69 In addition, Ag 3d core-level spectra were measured for various species containing both Ag(0) and Ag + . The 3d 5/2 Ag peak shis to higher binding energies (x0.6 eV) as one goes from Ag(0) nanoparticles to Ag N -DNAs to Ag + -DNA complexes (Fig. 9), supporting an Ag N -DNA with a positive charge which is neither purely cationic nor fully reduced, in agreement with MS studies by others. 24,57, 130 Schultz, et al., recently studied an HPLC-puried Ag N -DNA 159 emissive at 670 nm with a previously measured high quantum yield of 0.75. 24 By combining analytical centrifugation with NMR and MS, it became apparent that despite HPLC isolation, the emissive product was a mixture of Ag 15 and Ag 16 . Thus, even rigorous chromatographic separation may not always fully separate Ag N -DNAs into compositionally pure solutions when two or more species have very similar compositions/ conformations. IR spectroscopy combined with MD simulations provided insights into the DNA binding sites of Ag + . The experimentally measured IR spectra of the Ag N -DNA and bare DNA only show marked shis between 1350-1500 cm À1 aer cluster formation (Fig. 10A). These shis correspond to the nucleobases (Fig. 10B), not phosphate backbone, conrming that the Ag N ligates primarily to DNA through Ag-nucleobase interactions. 54,55 The Kohler and Petty groups very recently reported femtosecond time-resolved IR (TRIR) spectroscopic studies of two Ag 10 6+ clusters stabilized by very similar 18-base DNA strands, C 4 AC 4 TC 3 XT 3 , where X represents either guanosine or inosine (an articial nucleoside lacking the exocyclic C2-NH 2 of natural guanosine). 160 These two DNA strands stabilize products with nearly identical spectra but dramatically differing quantum yields and uorescence decay times, suggesting that the X nucleoside inuences the excited state processes of the Ag 10 6+ .
Following excitation of the clusters by a 490 nm femtosecond laser pulse, the TRIR spectra are collected in the 1400-1720 cm À1 range, corresponding to spectral features from the nucleobases. While individual nucleobases are excited in the UV, TRIR spectra show that 490 nm excitation of the clusters results in bleaching of the vibrational modes of select nucleobases, most notably cytosine. Thymine appears unperturbed by cluster excitation, supporting the many past studies which show that silver has low affinity for this nucleobase at neutral pH and suggesting that this base does not coordinate with the cluster. Slight differences in the TRIR spectra of X ¼ G and X ¼ I Ag N -DNAs suggest that this method may enable precise probing of the electronic coupling of the Ag N and surrounding nucleobases, a topic which remains poorly understood for Ag N -DNAs. 160

Electron microscopy
Many reports of transmission electron microscopy (TEM) to characterize Ag N -DNAs report 2-20 nm particles, which have been attributed to the uorescent clusters of interest. [161][162][163][164][165][166][167][168][169][170] However, due to the much smaller sizes of Ag N -DNAs established by HPLC-MS and recent crystallographic studies (Section 3.5), it is highly likely that the particles observed in TEM are silver nanoparticles formed as byproducts during chemical synthesis.  silvers not directly bound to the Ag 8 promote crystal packing (blue spheres in Fig. 11). Ag-Ag distances in the pentameric core are x2.9Å, comparable to Ag-Ag bond lengths in bulk silver. 174 In this cluster core, adenines interact with Ag via N1 and N6, whereas cytosines are coordinated through N3 and N4 (Fig. 11).

Ag
The exocyclic nitrogens (N4, N6) are hypothesized to be deprotonated. The zipper region is characterized by C-Ag + -C base pairs with parallel strand orientation and twisted base pairs, as observed elsewhere. 108,113 Every Ag interacts with the N3 site of one cytosine on each strand (Fig. 11C), as established previously for Ag N -DNAs stabilized by C 12 strands 52 and for C-Ag + -C duplexes. 54,101 All base-Ag interactions have distances of x2.1Å.
Unlike the Ag N -DNA studied by Schultz, et al., 159 signicant interactions between adenines and Ag were found in the Ag 8 -DNA (Fig. 11D). It is notable that one Ag atom of the pentamer portion of the Ag 8 is stabilized by a neighboring strand's adenine (Ag atom in orange in Fig. 11D), which may explain why the cluster could not be formed in solution without modications to the DNA template strand. 22 Six crystal structures have also been reported by Cerretani, et al., for NIR Ag 16 -DNAs stabilized by DNA templates that differ by only one nucleobase. 25,68,69 The rst reported crystal structure is for an Ag 16 -DNA stabilized by two strands of a DNA decamer, 5 0 -CACCTAGCGA-3 0 , previously identied by Copp, et al., 175 with  an unusually large Stokes shi. 176 The second Ag 16 -DNA is stabilized by two copies of a 9-base sequence corresponding to removal of the A 10 at the 3 0 -end of the decamer (Fig. 12A). Clusters formed on these two templates are nearly identical, and removal of the terminal A 10 has no discernable impact on the wavelength of the absorbance peak but causes a slight redshi in uorescence emission. 25 Similarly, mutations of position 5 in the DNA sequence allow one to produce and crystalize a similar NIR emitter. 68 The latter study showed also that certain nucleotide positions in the DNA sequence, while not relevant for binding to the Ag N , could be mutated in order to promote or alter crystal packing interactions. This concept could enable in the near future to re-engineer DNA sequences to promote crystallization and determine the structure of the emissive Ag N . 68 It also demonstrates that, especially when the Ag N is stabilized by multiple strands, the 3D organization of the nucleotides is more relevant than the sequential 5 0 to 3 0 order. Unlike the Ag 8 -DNA investigated by Huard, et al., which did not perform NaBH 4 reduction before the crystallization process, the Ag 16 -DNAs were synthesized in aqueous solution and then HPLC-puried prior to crystallization.
The clusters comprise 16 Ag atoms with occupancy of 1, along with additional silvers with lower occupancy (Fig. 12A and B). All bases, except thymine and one of the adenines in position 2, interact with Ag atoms, with the thymine ensuring strand exibility and promoting crystal packing interactions (Fig. 12C-G). Most of the Ag-Ag distances are between 2.7 and 2.9Å, similar to or shorter than their metallic radius. Nevertheless, the cluster charge cannot be elucidated by these distances alone and as mentioned previously, ample HR-MS data suggests that Ag N clusters are generally highly cationic in nature. 23,24,57,130 Similar to the crystal structure published by Huard, et al., 22 Ag atoms interact with cytosines via N3, and with adenines via N1. Interestingly, additional interacting sites were discovered, consistent with Schultz, et al. 159 silvers coordinate O2 of cytosines, as well as N1, N7 and O6 of guanines, and N7 and the oxygens of the adenine phosphate group. Ag-N distances are 2.2-2.5Å, mostly shorter than the Ag-O coordinate bond lengths 2.4Å to 2.9Å. The Ag-N bond lengths suggest that G 9 is deprotonated at N1 (2.3-2.4Å). 25,69 Some Ag N -DNA crystal structures contain Ag + which are not attached to the central cluster but do participate in non-WC base interactions and crystal packing. 22,25,68,69 It is possible that such "accessory" Ag + also exist in solution-phase Ag N -DNAs, as recently suggested by Gambucci, et al. 122 If sufficiently tightly bound, these Ag + would be counted by HR-MS as part of the Ag N -DNA but may not be part of the silver cluster itself and, thus, may not contribute signicantly to the cluster's electronic properties. HR-MS results have not been reported for the Ag 8 reported by Huard, et al., 22 nor the multiple Ag 16 species reported by Cerretani, et al., 25,68,69 so it remains unknown whether all of the accessory Ag + are present in solution. Studies which compare the MS-determined sizes of Ag N -DNAs with their crystallographic sizes are needed in order to probe the existence and role(s) of accessory Ag + in Ag N -DNAs and, more generally, to what degree HR-MS measurements of puried Ag N -DNA species can discern the size of the emissive cluster. It will also be important to clearly state the assumptions made when assigning the cluster size N of an Ag N -DNA, particularly in light of the aforementioned evidence that observed optical properties are strongly correlated to the numbers of neutral silver atoms N 0 determined by HR-MS and not necessarily the total silver atom number N tot .

Alternate possible cluster geometries and higher-order structures
We have primarily reviewed compositional and structural studies of Ag N -DNAs which are synthesized by chemical reduction and, notably, are stable under HPLC purication to enable accurate characterization. Smaller Ag 2 and Ag 3 clusters intercalated between base pairs of dsDNA can be synthesized by electrochemical means and exhibit $300 nm uorescence emission, 177-180 which supports smaller size and/or different cluster geometry than the HPLC-puried Ag N -DNAs discussed here. The versatility of macromolecular cluster ligands like DNA may permit multiple classes of metal clusters of the same metal species, even for Ag N -DNAs synthesized by chemical reduction. Thus, it is likely that other cluster sizes and geometries than the HPLC-puried ones discussed here may exist which may be unstable under the solvent and high-pressure conditions requisite for chromatographic separation.

Conformation of the DNA template strand
In addition to cluster structure, the secondary/tertiary structures of the cluster's DNA templates are of interest. An understanding of this structure is also critical for schemes which integrate with DNA nanotechnology. 181 As before, we primarily review studies of puried samples or which employ techniques which may lead to isolated Ag N -DNA species, including micro-uidic capillary electrophoresis, 49,182 gel electrophoresis, 59,116 and SEC. 38 The Petty group combined SEC and other analytical methods to discern tertiary structures of their developed Ag N -DNA sensors, which signal binding of DNA analytes by transforming nonuorescent Ag 11 clusters on ssDNA templates into NIRemissive clusters of twice the size. 38,183 SEC separates complexes by molecular size and shape, with larger products eluting more quickly. A difference in retention time between two products indicates differences in molecular size. To count the number of DNA strands, n s , which scaffold the NIR cluster, 10-thymine tails were appended to one end of the target DNA analyte. SEC shows that a 1 : 1 mixture of DNA analytes with and without tails splits the chromatogram into three peaks. This splitting can be interpreted as complexation of two DNA probes to form the NIR Ag N -DNA (Fig. 13A). This was one of the rst demonstrations of formation of Ag N -DNAs stabilized by template strand dimers, 38 apart from HR-MS. 53 For a modied sensor scheme, alignment of thymine tails was further used to probe alignment of the two DNA strands stabilizing the NIR Ag N -DNA. 183 These clever experiments provide an alternate technique for inferring n s , which is especially useful for larger DNA complexes hosting an Ag N -DNA, which may not be stable even under gentle negative mode ESI-MS. Thymine tails were later used by Del Bonis-O'Donnell, et al., to separate a set of Ag N -DNA-based probes for Hepatitis A, B, and C in a single microcapillary electrophoresis protocol. 184 The Petty and Brodbelt groups recently determined and compared binding sites of two different Ag 10 clusters to their DNA templates using activated electron photodetachment (a-EPD) MS. 121 One Ag 10 was stabilized by a 20-base strand which is single-stranded in the absence of silver (the subject of Fig. 8), 130 and the other by a 28-base strand which forms a hairpin in the absence of silver (Ag N -DNAs were studied without subsequent purication). 157 The DNA templates with and without Ag N were analyzed by a-EPD, using 193 nm irradiation to induce DNA fragmentation, followed by MS. Fig. 13B-E compares mass spectra of the fragmented DNA host strands with and without Ag N , showing that certain fragments are suppressed in the presence of the Ag N . The suppression of fragmentation for certain regions of the DNA templates was associated with binding of the nucleobases to the Ag N in these suppressed regions. For the ssDNA template, a remarkably short 4-base segment of CCTT was suppressed (Fig. 13C); in comparison to available crystal structures, 22,25,68,69 it is reasonable that this segment represents only part of the silver-ligated nucleobases. For the hairpin DNA template, a much longer 13-base segment was suppressed, most of which is in the WC paired hairpin stem in the absence of silver (Fig. 13E). This provides credence to the notion that Ag + can signicantly reorganize DNA secondary structure. Future a-EPD studies could yield insights into other Ag N -DNAs.

Photophysical studiesprobing excited luminescent and dark states of Ag N -DNAs
Compared to the current growing understanding of Ag N -DNA structure, the luminescence process of Ag N -DNAs remains less understood. Ag N -DNAs most certainly luminesce through an allowed uorescence-like process, as supported by 1-4 ns uorescence decay times and quantum yields >0.1 for most puried Ag N -DNAs. 24,123,176,185,186 In contrast, phosphorescencelike emission of other metal clusters is characterized by much longer decay times and lower quantum yield values due to less allowed/forbidden transitions. 4,187 However, the Ag N -DNA uorescence process does differ from the simple Jablonski diagram of organic uorophores. 188 Ag N -DNAs lack the characteristic vibronic shoulders of organic molecular uorophores, 21,147 and their solvatochromic behavior is not well- described by Onsager-based methods used to model many organic uorophores. 189 Certain Ag N -DNAs retain surprisingly high quantum yields into the NIR, 190 while quantum yields of organic dyes diminish rapidly in this region. 191 Ag N -DNAs also have highly polarized excitation and emission due to well-dened transition dipole moments. 192 Finally, the process of indirect uorescence excitation via the DNA bases, which produces the same color of uorescence as direct excitation in the visible or NIR excitation band of the Ag N -DNA (Fig. 1B), 21,26 remains poorly understood. Here, we review spectroscopic studies of the photophysics of Ag N -DNAs, with a focus on puried Ag N -DNAs in more recent years. In order to ensure that measured photophysical properties are not affected by the presence of byproducts, such as Ag nanoparticles and nonuorescent Ag N -DNAs, purication is essential to preparation and analysis of these uorophores.

Ultrafast studies of the Franck-Condon state
A limited number of experimental studies have probed the ultrafast dynamics that occur upon excitation of Ag N -DNAs to the initial excited state (Franck-Condon state). 26,27,193,194 Patel, et al., proposed the rst phenomenological model describing the excitation process (Fig. 14), based on ultrafast transient absorption experiments performed on three unpuried red and NIR Ag N -DNAs (Fig. 15). 193 It was observed that a fraction of the population in the Franck-Condon state returned to the ground state with a time constant in the hundreds of fs, as seen by the ground-state recovery (Fig. 15A). Additionally, a rise component of similar time scale was attributed to formation of the emissive state. This emissive state then decays back to the ground state on a nanosecond timescale, as witnessed by similar time-scales of the ground-state recovery and the typical ns uorescence decay times measured by time-correlated single photon counting (TCSPC). 35,176,185 In addition, nanosecond transient absorption spectroscopy (Fig. 15B)   . The depletion appears at negative DOD, but is plotted in its absolute value. It has been corrected for the spectral overlap by subtracting the contribution from the transient absorption, which is based on the kinetics at 775 nm calibrated to the expected value based on the peak curve fittings. The data was collected by exciting with a 100 fs Ti-sapphire laser at 1 kHz, then probing with a white light continuum generated from the same laser. The excitation wavelength was tuned to the peak of the ground state absorption. (B) Normalized femtosecond and nanosecond transient absorption spectra for Ag680. The sample was excited by 100 fs pulsed excitation, except for the long delay time curve, which was generated from excitation by a 7 ns pulsed laser. The dip in the spectrum around 800 nm is an instrumental artifact. 193  Recently, Thyrhaug, et al., performed 2D electronic spectroscopy experiments 27 on a previously sized NIR-emitting Ag 20 -DNA. 53,196 Excitation into the Franck-Condon state led to ultrafast evolution of the Franck-Condon state into the emissive state, which then decayed on a nanosecond time-scale observed from TCSPC measurements. The transfer from the initially populated state to emissive state occurred in about 140 fs, in line with the order of magnitude reported by Patel,et al. 193 Additionally, the Ag N -related absorption feature appeared to consist of two closely lying transitions, and a coherent excitation of both states occurred due to the short pulse width of the laser. Interestingly, for this particular Ag N -DNA, coherence was transferred to the emissive state and can be seen by oscillatory quantum beating features that dephased with a time constant of $800 fs. Thus, aer a few ps, all ultrafast processes were complete, and the only remaining process is the $ns uorescence. The dominant quantum beating mode frequency of 105 cm À1 is similar to Ag-Ag vibrational modes. 197

Dark states
The presence of a ms-lived dark state in Ag N -DNAs was rst reported by Vosch et al. 8 Ag N -DNAs were immobilized in a polyvinyl alcohol (PVA) lm and uorescence intensity was recorded as a function of time. Autocorrelations of the uorescence intensity trajectories revealed ms blinking. A similar ms correlation time was observed by uorescence correlation spectroscopy (FCS) in solution. FCS experiments are not only useful for determination of the decay time of the dark state and the quantum yield of dark state formation 198 but also for estimation of the molar extinction coefficient by determining the number of emitters in a certain volume identied from a reference measurement. 8,199 While nontrivial to determine or suggest the exact nature of the dark state, dark states have been reported in other studies 60,196,200 and may be common for most Ag N -DNAs. The quantum yields of dark state formation have been estimated to range from a few up to 25 percent. 8,60,196,201 When removal of molecular oxygen from the environment results in a lengthening of the dark state decay time, this is oen a good indicator that the dark state is a triplet state. 202 For Ag N -DNAs, the DNA scaffold around the silver cluster might act as a physical barrier for this type of Dexter-type triplet state quenching, resulting in minimal or no effect of removal of oxygen on the dark state decay time. 8 Richards, et al., demonstrated that the dark state formed by a primary excitation laser can be optically excited with a secondary NIR laser, resulting in depletion of this long-lived state and an overall increase in uorescence intensity. 31 It was recently proven that optical excitation of the dark state can transition the Ag N -DNA to the emissive state, resulting in optically activated delayed uorescence (OADF). [28][29][30]195 This process is similar to typical reverse-intersystem crossing processes observed in organic dyes. 203,204 OADF combined with timegating provides background-free signal because the delayed uorescence is on the Anti-Stokes side (lower wavelength side) of the secondary excitation laser, allowing any Stokes-shied auto-uorescence from the secondary laser to be suppressed with a short-pass lter in the detection path. Fig. 16 shows an example by Krause, et al., that demonstrates the OADF imaging concept. 30 Additionally, Krause, et al., showed that the use of the secondary NIR laser only (blocking the primary excitation laser) yielded similar uorescence which was linearly dependent on the excitation intensity. This process is termed upconversion uorescence (UCF), 29,30 in analogy to the well-established upconversion processes in lanthanide based emitters. 205

Emissive state
While one would expect a single emissive species to exhibit mono-exponential uorescence decay, several HPLC-puried Ag N -DNAs with long DNA template strands (19-30 bases) exhibit multi-exponential uorescence decay. 35,185,186 Because solutions are puried prior to characterization and no shi is present in the steady-state emission as a function of excitation wavelength, a heterogeneous mixture of Ag N -DNA species can be excluded as the cause of this multi-exponential decay. The multi-exponential decay behavior can instead be explained by relaxation of the emissive state on a time-scale similar to the uorescence decay. This effect, termed "slow" spectral relaxation, can be conrmed by time-resolved emission spectra (TRES) which show a gradual red-shi of the emission maximum on the nanosecond time scale (Fig. 17A). We note that the "slow" spectral relaxation is a minor part of the overall Stokes shi, with the majority of relaxation occurring on a timescale below the IRF. A consequence of "slow" spectral relaxation is that the average decay time increases as a function of emission wavelength (Fig. 17B). Furthermore, the decay associated spectra (DAS) usually lead to spectra where the fastest decay time component tends to have positive amplitudes at shorter wavelengths and negative amplitudes (rise) at longer wavelengths (Fig. 17C). 185 Only two processes can cause this effect: energy transfer or "slow" spectral relaxation. 188 Energy transfer can be excluded since, as stated above, there is no evidence for multiple independent emitters in the HPLC-puried Ag N -DNA solutions.
Unlike small solvent molecules which rearrange on picosecond timescales, the DNA template and its structurally bound water molecules require much longer, up to a few nanoseconds, to adapt to the new charge distribution of the Ag N -DNA in the emissive state. A similar effect was observed when a coumarin dye was embedded in an abasic site of dsDNA. 206 Emission spectral shis could be observed from the femtosecond time scale up to tens of nanoseconds. Other parameters, e.g. changes to solvent viscosity or temperature, also affect the "slow" spectral relaxation. 176,185 If spectral relaxation occurs entirely within the time-scale of the IRF, the observed decay time will be mono-exponential. This is the case for Ag N -DNAs stabilized by short, 9-10 base DNA strands, whose "slow" spectral relaxation is negligible at room temperature and in low viscosity solvents, 25,176,190 most likely because multiple short strands are more exible and rearrange faster than one long oligomer. Spectral relaxation could be a useful tool to establish the rigidity of the DNA scaffold and its effect on the excited state of Ag N -DNAs. 35,51,185,186

Excitation and emission transition dipole moments
Another interesting spectroscopic feature of Ag N -DNAs is their parallel excitation and emission transition dipole moments. Hooley, et al., employed defocused wideeld microscopy to investigate the transition dipole moments of a C 24 -templated Ag N -DNA immobilized in PVA. 192 By defocusing a common wideeld image, the emission of a single emitter displays a bilobed shape that depends on the orientation of its emission transition dipole. In order to determine both excitation and emission transition dipoles simultaneously, defocused wide-eld microscopy was combined with rotating the polarized excitation light. Then, the intensity of each emitter is directly correlated to the excitation efficiency of the Ag N -DNA. Maximum emission intensity was observed when excitation light was aligned with the emission transition dipole, indicating that the excitation and emission transition dipole moments lie along a similar direction.
The Vosch group has also observed further evidence of the alignment of excitation and emission transition dipole moments by time-resolved anisotropy measurements. 122,176,190 Three different Ag N -DNAs, two NIR emitting and one red,  displayed limiting anisotropy values close to 0.4, which indicates that the excitation and emission transition dipole moments are parallel (one example in Fig. 18).

Coherent two-photon excitation
Patel, et al., rst reported two-photon excitation (800 to 1000 nm range) of Ag N -DNAs in 2008, in a study of four non-puried Ag N -DNAs with emission maxima at 620 nm, 660 nm, 680 nm and 710 nm. 207 For the 660 nm, 680 nm and 710 nm emitters, the two-photon emission exhibited quadratic dependency on excitation intensity, as expected, and one-photon and two-photon uorescence decay times were similar, indicating that emission occurred from the same emissive state. The one versus twophoton excitation spectra of the 620 nm emitter indicated that cross-section maxima occurred at different wavelengths. The reported two-photon cross-sections ranged from 33 900 to 50 000 GM, roughly two orders of magnitude higher than typical organic uorophores (e.g. 210 GM at 840 nm for Rhodamine B). 208 Yau, et al., reported a two-photon cross-section of $3000 GM at 800 nm and quadratic dependence of emission on excitation intensity for an unpuried 650 nm emitter. 194 This 650 nm emitter was made by rst creating an Ag N -DNA using ssDNA, followed by addition of an excess of a complementary strand with a guanine-rich section. Because few studies have probed two-photon excitation of Ag N -DNAs, future investigations on puried Ag N -DNA could shed light on the origin of the very high two-photon cross-sections.

Informed designdecoding the sequence-color connection for Ag N -DNAs
The fascinating sequence dependence of Ag N -DNAs results from the nucleobase-specic interactions of DNA with silver (Section 2). The ability of DNA sequence to select for the sizes and optical properties of metal nanoclusters has attracted great interest due to the promise of highly customized uorophores. 199,209 To date, it is likely that thousands of different DNA template strands have been reported, corresponding to Ag N -DNAs with wideranging uorescence colors, Stokes shis, quantum yields, chemical yields, photostabilities, and chemical stabilities. 70 Yet the connection between DNA sequence and Ag N -DNA properties has remained obscure. Most studies select Ag N -DNAs by experimentally testing small numbers of DNA template strands rich in C or G. 32,46,128 One large-scale study by the Dickson group used DNA microarrays to identify uorescent Ag N -DNAs, but only a few of the DNA template sequences were reported. 199 To fully realize Ag N -DNAs as programmable materials, it is crucial to "decode" the connection between DNA sequence and Ag N -DNA properties.
Rational design of Ag N -DNAs is especially challenging due to an astronomical number of possible DNA template sequences and a complex connection between Ag N -DNA color and DNA sequence. Ag N -DNA templates are typically 10-30 base oligomers. Because a sequence of the four natural nucleobases can have 4 L distinct L-base sequences, Ag N -DNA templates must be chosen from 4 30 ($10 18 ) possible sequences. While in some cases subtle sequence changes can dramatically shi uorescence, 65,169 in other cases different DNA sequences can stabilize Ag N -DNAs with the same emission wavelength. 57 To make matters more complex, some DNA sequences can stabilize different types of uorescent Ag N clusters, 57 with yields of each cluster species possibly depending on synthesis method and/or Ag:DNA stoichiometry. First-principles computational methods have not yet matured sufficiently to model the structures of realistic Ag N -DNAs, let alone their accurate electronic properties. Small-scale studies of DNA sequences with constrained patterns 119,169,196,[210][211][212] have been useful for developing a few Ag N -DNAs with well-controlled properties but are limited in their applicability to the majority of reported Ag N -DNAs. Here, we review large-scale experimental studies of the Ag N -DNA sequence-color connection for 10 3 DNA strands, in which machine learning enables predictive design and provides new physical insights.

Large-scale studies of sequence dependence
The combinatorial nature of DNA makes data science wellsuited to study how DNA sequence selects Ag N -DNA properties. Copp, et al., have pioneered high throughput experiments together with supervised machine learning (ML) to understand how DNA sequence selects for Ag N -DNA uorescence emission and to predict new templates for optimized Ag N -DNAs. The methods described here have uncovered Ag N -DNAs which are the subjects of later detailed studies. 23,69,176,190 For readers seeking to learn more about ML, we recommend a tutorial review by Domingos 213 and a review of ML for so matter by Ferguson. 214 To train a ML algorithm to output Ag N -DNA uorescence properties (or whether any uorescent product can be stabilized) given an input DNA template sequence, one must rst amass a data library connecting DNA sequence to Ag N -DNA uorescence spectra for hundreds to thousands of sequences. This data cannot be mined from the literature because (i) synthesis and characterization methods vary widely, prohibiting isolation of the effects of DNA sequence from other experimental parameters, and (ii) while $75% of DNA sequences are unsuitable for templating uorescent Ag N -DNAs, these "negative" DNA sequences are rarely reported. 57 The absence of negative sequences from the literature is problematic because to effectively learn what makes a suitable DNA template for brightly uorescent Ag N -DNAs also requires knowledge of what does not make a suitable template.
To enable ML for Ag N -DNAs, Copp, et al., developed highthroughput Ag N -DNA synthesis and characterization in well plate format using robotic liquid handling followed by rapid uorimetry, 57 via universal UV excitation of all Ag N -DNA products through the nucleobases (Fig. 19 part I). 21 Because uorimetry is performed one day, one week, and four weeks aer synthesis, this data set allows ML studies to focus only on timestable Ag N -DNAs. Experiments are normalized using a wellstudied Ag N -DNA control, 24,35,122,185 for direct comparison of uorescence wavelength and intensity among all data in the library. To date, we have reported on >3000 distinct DNA template sequences, most 10 bases long, for Ag N -DNA synthesis in 10 mM NH 4 OAc aqueous solution of neutral pH. 23,111,175,215 Effective ML requires appropriate choice of "feature vectors," which are the parameterizations of training data provided as inputs to the ML classier(s). For Ag N -DNAs, feature vectors should represent the salient properties of a DNA sequence which determine how sequence is mapped onto Ag N -DNA uorescence. Because these properties are not well-known (otherwise ML would be unnecessary), this feature engineering process is a critical step in the ML workow 213 and has led to new physical insights into Ag N -DNAs. Early work used training data for 684 randomly generated 10-base DNA sequences to learn to predict Ag N -DNA uorescence brightness given an input template strand sequence. 215 Using integrated uorescence intensity, I int , as a metric of brightness, sequences with the top 30% of I int values were dened as bright and the bottom 30% of I int values dened as "dark." Then, a ML algorithm called a support vector machine (SVM) was trained to distinguish bright and dark sequences ( Fig. 19 part II). It was found that the SVM most accurately predicted a sequence's class if feature vectors were engineered to quantify the occurrence of certain DNA subsequences called "motifs" which were identied by bioinformatics approaches to be correlated with one class but not the other. 216 The resulting trained SVM's classication accuracy was 86%, as determined by crossvalidation (a process which trains on most of a training data set and reserves a small $10% portion as a "test set" to assess SVM performance on data which the ML classier has not yet encountered). New DNA templates for bright Ag N -DNAs were designed using the bright-correlated motifs as building blocks and then screened by the trained SVM to choose those predicted as most likely to be bright. 78% of designed DNA templates stabilized bright Ag N -DNAs, as compared to 30% of the initial random sequences. 215 This early work pointed to the role of certain DNA base motifs in stabilizing Ag N -DNAs, which agreed with later ndings that not all DNA bases coordinate the Ag N . 22,25,69 While predicting Ag N -DNA uorescence intensity increases the likelihood of selecting uorescent Ag N -DNAs by three-fold, this simple method also prefers red-uorescent Ag N -DNAs over green Ag N -DNAs. 215 It is ideal to instead predict both brightness and color from an input DNA sequence. To achieve this, Copp, et al., used physically motivated Ag N -DNA classication based on the known correlation between Ag N -DNA color and cluster size. The multi-modal distribution of Ag N -DNA uorescence colors in the visible spectrum was shown to arise due to the magic numbers of these clusters: Ag N -DNAs in the 500-570 nm abundance have N 0 ¼ 4 neutral Ag atoms, while Ag N -DNAs in the 600-670 nm abundance have N 0 ¼ 6 ( Fig. 20A). 57 Because "Green" and "Red" Ag N -DNAs have distinct cluster sizes, there is likely a fundamental difference between template sequences for these two cluster sizes. To learn to distinguish between DNA sequences based on cluster structural differences, a training data set of $2000 10-base DNA sequences was separated into four color classes: the three shown in Fig. 20A ("Very Red" is dened as the high wavelength histogram shoulder, which may signal a different cluster structure) and a "Dark" class similar to the one previously dened. 215 Because the numbers of sequences in these classes are unequal, with far more Dark sequences than Green sequences (Fig. 20B), it is critical to apply subsampling to balance classes prior to ML, ensuring training on equal numbers of sequences from each class. 217,218 Feature vectors were then constructed using DNA motif mining to identify color-correlated motifs, followed by feature selection 219 to reduce the list of selected motifs to those most important for classication; this critical step reduces problems which can arise from overtting. We note that both data balancing and feature selection should generally be applied when using ML for real-world materials systems.
Because SVMs are inherently binary classiers, a "one-versusone" approach was used to distinguish the four color classes. Six different SVMs were trained to discriminate between the six possible pairs of classes (cross-validation scores, which represent the accuracy of classication, in Fig. 20C). To experimentally test the performance of the trained classiers, new DNA template sequences were designed for the two least abundant classes, Green and Very Red. First, color-correlated DNA motifs for the desired class were selected from a probability distribution weighted by intensity and placed into an initially empty DNA sequence. Second, designed candidate DNA templates were screened by the trained SVMs to estimate the probability of falling within the desired color class. Finally, templates corresponding to the top 180 probabilities were selected for experimental testing. With this method, the likelihood of selecting a Very Red Ag N -DNA increased by nearly 330%, and the likelihood of selecting a Green Ag N -DNA was increased by >80%. 175 This method was later modied to enable design of Ag N -DNA templates of any strand length, and it was found that training data of only 10-base sequences still enabled effective prediction of Ag N -DNA color for other lengths of DNA templates, up to the maximum 16-base length tested (Fig. 20D). 111 This suggests that there exist certain DNA motifs which are selective of cluster type and thus color for a range of DNA template lengths, making ML design approaches for Ag N -DNAs much more promising. We note that thus far, all Ag N -DNAs stabilized by DNA templates of <19 bases have been found to be "strand dimers" which contain two template strands per cluster; 23,24,53,57 it is possible that longer DNA templates, which have not been designed by ML, may have some different DNA sequence rules for Ag N -DNA color selection.
In addition to improving design efficiency, ML provides key insights into how DNA sequence selects silver cluster size, and thus uorescence wavelength. Fig. 21B shows average base composition of the motifs identied by feature selection to be most predictive of Dark, Green, Red, and Very Red sequences. 175 To summarize, thymines are strongly correlated with no uorescence. Adenines show preference for smaller and Green Ag N -DNAs, while guanines, particularly consecutive guanines, are correlated with long wavelength uorescence (associated by MS with clusters containing more Ag atoms). Cytosines are strongly selective for uorescence brightness but less selective of color  than A or G. To better understand these correlations, we compare to HR-MS studies of DNA-Ag + complexes (Section 2), which are the precursors of Ag N -DNAs prior to reduction by NaBH 4 . Fig. 3A shows the distribution of Ag + attached to single DNA homobase strands or pairs of strands, and Fig. 5A shows the same distribution for Ag + -mediated dimers of C or G strands with central base mutations. 54,55 Because homo-thymine strands only weakly associate with Ag + , thymine-rich DNA sequences may be unsuitable (at neutral pH) to host uorescent Ag N due to (i) too few Ag atoms recruited prior to reduction, resulting in insufficient silvers to form a cluster and/or (ii) little to no coordination with the cluster. The greater occurrence of T's in green Ag N -DNA templates further supports this notion, since these clusters are smaller in size and may require fewer nucleobase coordination sites. Adenine homobase strands bind to a few Ag + , which may support formation of smaller N 0 ¼ 4 clusters with green emission. In comparison to A and T, C-and G-rich homobase strands can form Ag + -mediated duplexes with $1 Ag + per base pair, providing more Ag atoms during cluster growth and supporting nucleobase-silver binding in the Ag N -DNA. Interestingly, duplexes of G homobase strands with a single central A, C, or T base mutation can harbor $60% more Ag + than G homobase polymers with no mutation. This significant increase in Ag + attachment as compared to C-rich strands, and the structural differences in the DNA secondary/tertiary structures supported by IMS-MS of these strands, 108 could explain why consecutive G's are strongly associated with Very Red Ag N -DNAs. 175 6. Supra-cluster assemblytowards applications in photonics and sensing Structural DNA nanotechnology harnesses DNA as a programmable building block for self-assembled nanostructures. 220 It is promising to combine sequence-controlled Ag N -DNAs with DNA nanotechnology for realization of precise metal cluster arrays, which could be envisioned as functional sensors and photonic devices. These achievements will require robust strategies to effectively embed Ag N -DNAs into larger WC-paired architectures. Here, we review efforts to harness DNA self-assembly for multi-Ag N -DNA organization (many groups have incorporated single Ag N -DNAs within WC paired DNA structures to build biomolecular sensors, 13,14,33,37,73,183 which were recently reviewed elsewhere 74 ).
O'Neill, et al., rst reported decoration of a DNA nanostructure with multiple Ag N -DNAs. 100 A mixture of green and red clusters were synthesized onto ssDNA hairpin protrusions programmed into DNA nanotubes (Fig. 22A). Without hairpins, nanotubes did not foster cluster growth, consistent with early ndings that dsDNA is an unsuitable Ag N -DNA template. 11 The authors noted that Tris buffers typical of DNA self-assembly schemes were unsuitable for chemical synthesis of uorescent Ag N -DNAs; this incompatibility is commonly faced in supracluster assembly of Ag N -DNAs. Orbach, et al., demonstrated Ag N -DNA synthesis on mm-scale DNA wires with hairpin protrusions. 118 Resulting uorescence colors depended on salt concentration, pointing to the complexity of controlled cluster synthesis on complex DNA scaffolds. The authors then incorporated Ag N -DNA-stabilizing hairpins into a hybridization chain reaction (HCR), with wire formation only aer addition of an additional DNA strand (Fig. 22B). Ag N -DNAs have also been incorporated into DNA hydrogels. 221 Only two works have assembled puried Ag N -DNAs in order to approach atomic precision over cluster size in multi-cluster assemblies. Schultz, et al., developed DNA "clamps" for dualcolor Ag N -DNA pairs which exhibited Förster resonance energy transfer (FRET) between donor and acceptor Ag N -DNAs. 222 DNA clamps were designed by appending complementary tails of A and T bases 13 to templates for a green-emissive Ag N -DNA donor (Fig. 23A) and a red-emissive Ag N -DNA acceptor (Fig. 23B), which have a 6 nm Förster radius. 188 Aer HPLC purication of individual Ag N -DNAs, various geometries of clamps were formed by WC pairing. For clamps where donor and acceptor were held within <6 nm, donor excitation produced acceptor emission (e.g. Fig. 23C), with >60% FRET efficiency estimated by donor quenching (Fig. 23D) and assuming no isolated donor is present 186 (use of excess of acceptor increased the likelihood of all donors in the paired state). FRET could be repeatedly cycled by heating and cooling, corresponding to cyclic melting and reforming of the DNA clamp (Fig. 23E). The clamp design is somewhat general and was demonstrated with a different acceptor cluster. Notably, Schultz, et al., found that HPLC purication was essential to observing FRET due to low chemical yield of Ag N -DNA synthesis; without purication, very few clamps contain both donor and acceptor clusters. 222 Recently, Zhao, et al., observed FRET between donor and acceptor Ag N - DNAs without prior purication by synthesizing Ag N -DNAs within surfactant reverse micelles with 5-10 nm diameters. 223 By this method, a fraction of the micelles contained both donor and acceptor clusters conned together within a "nanocage" whose size is of the length scale of the Förster radius of the pair. This method also enabled spectroscopy-based measurement of micelle diameter in agreement with more laborious cryoelectron microscopy, suggesting that Ag N -DNA-based FRET may be a promising route to size measurement of biological "nanocage" structures. 223 Copp, et al. presented a modular design strategy for multifunctional DNA templates with distinct Ag N -DNA stabilizing regions and single-stranded "linker" regions. 200 This strategy exploits large data libraries 57 to identify 10-base DNA strands which do not foster uorescent Ag N -DNA growth. These strands are candidate linkers to extend an Ag N -DNA template strand while leaving the cluster unchanged. Candidate linkers are appended to the DNA sequence of an HPLC-stable Ag N -DNA (Fig. 24A) and experimentally screened to determine if linkers leave Ag N -DNA optical spectra unshied, signalling little to no change in cluster geometry. A complementary "docker" site is then engineered onto a DNA nanotube (Fig. 24A). Following HPLC purication of the Ag N -DNA, multi-cluster assembly occurs by WC pairing of linker and docker strands (Fig. 24B). Ag N -DNAs with atomically selected sizes are unperturbed aer binding to the nanotubes, as supported by unchanged spectral shapes aer assembly (Fig. 24C). Future studies are needed to conrm labelling efficiency. The method is general to multiple sizes of Ag N -DNAs and linker sequences 200 and could generalize to many types of DNA scaffolds, for precise control over both cluster geometry and orientation.  Recently, Yourston, et al., thoroughly studied Ag N -DNAs formed on RNA nanorings with DNA "arms." 224 Because in situ synthesis was used to decorate arms with Ag N -DNAs, it is uncertain how many Ag N were harboured on a given nanoring. Interestingly, placement of the ssDNA region on which Ag N presumably formed affected not only uorescence spectra of the clusters, indicating variations in size/shape and possibly rigidity, but also time stability: clusters formed within the nanoring were much more time stable, perhaps due to enhanced protection from redox reactions which can blue-shi Ag N -DNA emission over time. 35,225 Studies such as these will be important for assessing the practicality of DNA-based Ag N -DNA arrays as functional materials.
The heterogeneous mixture of products and low chemical yield of Ag N -DNA synthesis can prohibit precise Ag N -DNA arrays by direct synthesis onto a DNA nanostructure. Additionally, we pointed out in a previous section that one should also not a priori assume that the envisioned WC base pairing of the DNA nanostructure will be maintained once silver is introduced.
Assembly methods which instead rely on WC pairing aer purication have their own limitations due to the limits of purity aer HPLC and due to labelling efficiency of the DNA nanostructure by binding of Ag N -DNA linkers to each docker site; this may be overcome by adding an excess of Ag N -DNAs. Much more work is required to realize precise cluster arrays by either method.

Future directions and challenges
Signicant recent progress has been made in understanding the structure-property relations of Ag N -DNAs and achieving their rational design. These advances were enabled by new experimental and computational strategies to purify and size Ag N -DNAs, to select new DNA templates for especially uorescent Ag N -DNAs, and to crystallize Ag N -DNAs for structure determination, as discussed in this review. Here, we discuss outstanding challenges in this eld and areas of especial promise, which we hope will catalyze new research directions in this important eld.

Near-infrared emissive Ag N -DNAs
Nearly all reported Ag N -DNAs exhibit l em in the 500-750 nm range. 57 The most well-studied NIR Ag N -DNAs have been developed by Petty and coauthors. 15,33,60,183,196 High-throughput studies by Copp, et al., uncovered additional NIR emissive clusters, 175 and the Vosch group has characterized several of these recently discovered NIR Ag N -DNA, including one with an unusually large Stokes shi 176 and one with an impressively high 73% quantum yield. 190 These quantum yields are competitive with organic uorophores, making Ag N -DNAs promising for development of biolabels in the NIR tissue transparency windows. 191 Until recently, only two Ag N -DNAs with l em > 800 nm were reported, 196 and it was assumed that NIR Ag N -DNAs are inherently rare compared to their visibly emissive counterparts. However, because Ag N -DNA studies employ UV-Vis optimized photodetectors commonly used for spectroscopy in the chemical and biological sciences, for which sensitivity is poor above $800 nm, it is possible that many NIR-emissive Ag N -DNAs have simply gone undetected. Swasey and Nicholson, et al., developed a custom NIR well plate reader equipped with an InGaAs detector to search for NIR uorophores in high-throughput (Fig. 25A). 226 Using this tool to scan $750 Ag N -DNA samples, 161 previously unidentied NIR-emissive Ag N -DNAs were uncovered (Fig. 25B). This huge abundance of NIR products was unexpected because the scanned Ag N -DNAs were stabilized by randomly selected DNA template sequences 57 or by oligomers previously designed for visible uorescence. 175,215 Among the newly discovered Ag N -DNAs were found the longest wavelengthemissive Ag N -DNA to date, with 999 nm peak uorescence emission 23 (Fig. 25C) and the largest Ag N -DNA to date, an Ag 30 with 12 Ag 0 and 18 Ag + (Fig. 25D). A directed search for NIR Ag N -DNAs, using the informatics methods described in Section 5, is highly promising for discovery of new NIR Ag N -DNAs.

Ag N -DNA photophysics
Both experimental and computational efforts are needed to further our understanding of the uorescence process in Ag N -DNAs, including the nature of the initial excited state, the relaxation process(es) leading to the origins of Stokes shis for these emitters, and the roles of both the Ag N and the surrounding nucleobases in governing excited state properties. While a zoo of Ag N clusters stabilized by different ligands have been described in literature, their optical properties largely differ from the distinctive features of Ag N -DNAs described in Section 4. Zeolite stabilized Ag N clusters display mainly strong UV absorption bands, with emissive excited-state decay stretching from picoseconds to the microsecond range. 227,228 Similarly, the Mak group recently reported an intriguing octahedral silver cluster with 95% uorescence quantum yield and microsecond-scale uorescence decay times caused by thermally activated delayed uorescence. 229 Such microsecond-scale uorescence decay times have not yet been observed for puried Ag N -DNAs. Only microseconds-lived dark states of Ag N -DNAs have been reported.
The unusual rod-like geometry of HPLC-stable Ag N -DNAs, 24 which has been conrmed in recent crystal structures of NIR Ag N -DNAs, 25,68,69 makes Ag N -DNAs particularly interesting experimental systems for the study of collective electronic excitations in molecular-like materials. 148,[151][152][153] With only N 0 ¼ 4-12 effective valence electrons in Ag N -DNAs characterized thus far by HR-MS, 23,24,57 Ag N -DNAs lie well below the atomic size identied as the onset of plasmonic excitations in monolayerprotected gold clusters. 230 However, the high aspect ratios of some identied Ag N -DNAs 147 may make certain Ag N -DNAs better approximated as atomic silver rods, which computational studies have shown to exhibit plasmonic-like excitations. 136,139,144 Future studies probing the ultrafast excited state dynamics of Ag N -DNAs are needed to better understand whether, or to what degree, collective electronic excitations are involved in the luminescence process of Ag N -DNAs.
Another feature of Ag N -DNA photophysics which remains poorly understood is the exact nature of the UV excitation process, which for the case of pure Ag N -DNA solutions leads to the same uorescence spectral shapes as visible/NIR excitation (Fig. 1B). 21 Due to DNA's complex and elegant excited state dynamics, 231 it is possible that DNA imbues Ag N with similar properties. Berkadin, et al., have computationally examined the UV excitation process in Ag N -DNAs using molecular dynamics (MD) to simulate thread-like silver clusters in a DNA duplex, followed by DFT-based tight binding to calculate the electronic dynamics of the relaxed structure. Interesting, UV excitation results in a net negative charge transfer to the cluster, due to promotion of electrons from the localized p state of the DNA to the cluster. 232 Such simulations performed on the recently reported crystal structures would be of great interest. 22,25,68,69 Furthermore, recent experiments by the Kohler group on Ag +nucleobase complexes are also promising for enhancing our understanding of this aspect of Ag N -DNA photophysics, 113,233 with their very recent study nding evidence for an extremely long-lived, $10 ns excited state in a C 20 -Ag + -C 20 duplex.

Rational sensor design
Many chemical and biomolecular sensing schemes employing Ag N -DNAs have been developed, such as NanoCluster Beacons, 32,39 ratiometric sensors, [234][235][236] and microRNA sensors 14,119 (more complete list in a past review 74 ). Designing these sensors is extremely challenging, and designs may not generalize because silver clusters are not conned only to the expected regions of a probe. 234 Further, the mechanisms underlying the function of these sensors remain uncertain in most cases, although color and brightness changes are likely due to restructuring of Ag N -DNAs. 34,62,138 Recent efforts have focused on strategies to improve sensitivity and selectivity of Ag N -DNA sensors, such as by addressing the low chemical yield of these clusters. 237 Due to the complexity of designing Ag N -DNA sensors, we propose that high-throughput experimentation combined with machine learning approaches may be a useful path forward. The Yeh group has recently pioneered highthroughput screening of Ag N -DNA sensors, which may signicantly expedite progress in this area. [238][239][240] Puried Ag N -DNAs could also serve as sensitive nanophotonic sensors. The photophysical properties of pure Ag N -DNAs have also been shown to exhibit sensitivity to temperature, 176,185 refractive index, 147,189 and viscosity. 176 Combined with advancement in the ability to pattern DNA nanostructures with Ag N -DNAs, it may be possible to design sensors which colocalize Ag N -DNAs with analyses of interest for nanoscale measurements.

Non-natural polynucleic acids as cluster templates
While DNA has been well-studied as a template for silver clusters, and RNA to a lesser extent, 46 much less is known about the suitability of non-natural polynucleic acids to template silver clusters. 241 Because RNA is less exible than DNA, it has been noted that RNA may be less suitable as a scaffold for Ag N if signicant exibility of oligomer ligands is required for a given cluster geometry. 69 Synthetic polynucleic acids could expand cluster structures and geometries, enhance stabilities, and imbue added functionalities. In addition to the four natural nucleobases, numerous articial nucleobases have well-studied affinity for silver and other metals. 77-80 A large range of uorescent nucleotide analogues are currently available, with more being actively developed continuously. [242][243][244] These bases could shi the universal UV excitation peak into the blue region of the visible spectrum, and FRET experiments could help elucidate the energy transfer processes in Ag N -DNAs and even unravel distances and proximity of selected nucleobases to the Ag N cluster. Also of interest are chemical modications developed for therapeutics to reduce enzymatic nucleotide digestion, 245 which have been reported to template Ag N -DNAs, 246 and other backbone modications which would inuence ligand conformation and, therefore, possible stabilized cluster geometries. Future studies are needed in this promising area.

Uncharacterized toxicity and biocompatibility of Ag N -DNAs
Ag N -DNAs are oen touted as nontoxic and biocompatible uorophores, 64,65,70-74 but very few studies have established these properties. 247,248 While Ag + is certainly less toxic than other heavy metals that compose luminescent nanoparticles such as quantum dots, 249 it is also a common environmental metal pollutant. 250 Ag nanoparticles can break down in the body in a range of different ways, resulting in toxicities due to either the nanoparticles themselves or to Ag + and silver salts. Ag nanoparticles are also nding use as anti-cancer therapeutics, 251 adding further complexity to our understanding of the toxicity of Ag N -DNAs. In-depth studies of the toxicities of Ag N -DNAs and their uptake and possible clearing from tissues and organisms are needed to advance their applications in the biomedical sciences and to ensure environmentally responsible use and disposal.

Enhancing stability
While Ag N -DNAs are oen touted as extremely photostable and/ or chemically stable, degradation of Ag N -DNAs in biologically relevant solutions and in the presence of living cells is a significant hindrance to their practical use in bioimaging. 252 To overcome this challenge, Jeon, et al., encapsulated Ag N -DNAs within silica nanoparticles, signicantly increasing cluster chemical stability. 253 The encapsulated Ag N -DNAs can also be used to monitor the stability of their silica nanoparticle hosts in various biological media. 254 In situ synthesis of Ag N -DNAs within DNA hydrogels can improve photostability, likely by shielding clusters from oxidative degradation. 255 Lyu, et al., reported signicantly enhanced chemical stability and increased cellular uptake of Ag N -DNAs modied by cationic polyelectrolytes. 252

Conclusions
Ag N -DNAs lie at the unique intersection of metal cluster science and DNA nanotechnology, combining the atomic precision of ligand-stabilized metal clusters with the sequence programmability of DNA nanomaterials. Their photophysical properties also provide a window into the regime between behavior associated with single small molecules and behavior associated with nanoparticles. For these reasons, the study and engineering of Ag N -DNAs are both extremely challenging and extremely promising. Here, we have reviewed recent advances in the fundamental understanding of these nanoclusters, with a focus on studies of puried Ag N -DNAs with chromatographically selected sizes. The latest (and evolving) ndings of Ag N -DNA structure and the nature of the DNA-silver interaction have been discussed. Photophysical studies, particularly of puried Ag N -DNAs, have been summarized. The current understanding of how DNA sequence selects for cluster size and optical properties has been reviewed, as have emerging methods for predictive design of Ag N -DNAs and their larger organization into multi-cluster arrays. We provide perspectives on emerging areas of interest and signicant unanswered questions related to these uorescent clusters in the hopes of stimulating researchers to explore these fascinating nanomaterials.

Conflicts of interest
There are no conicts to declare.