Photogeneration of Spin Quintet Triplet–Triplet Excitations in DNA-Assembled Pentacene Stacks

Singlet fission (SF), an exciton-doubling process observed in certain molecular semiconductors where two triplet excitons are generated from one singlet exciton, requires correctly tuned intermolecular coupling to allow separation of the two triplets to different molecular units. We explore this using DNA-encoded assembly of SF-capable pentacenes into discrete π-stacked constructs of defined size and geometry. Precise structural control is achieved via a combination of the DNA duplex formation between complementary single-stranded DNA and the local molecular geometry that directs the SF chromophores into a stable and predictable slip-stacked configuration, as confirmed by molecular dynamics (MD) modeling. Transient electron spin resonance spectroscopy revealed that within these DNA-assembled pentacene stacks, SF evolves via a bound triplet pair quintet state, which subsequently converts into free triplets. SF evolution via a long-lived quintet state sets specific requirements on intermolecular coupling, rendering the quintet spectrum and its zero-field-splitting parameters highly sensitive to intermolecular geometry. We have found that the experimental spectra and zero-field-splitting parameters are consistent with a slight systematic strain relative to the MD-optimized geometry. Thus, the transient electron spin resonance analysis is a powerful tool to test and refine the MD-derived structure models. DNA-encoded assembly of coupled semiconductor molecules allows controlled construction of electronically functional structures, but brings with it significant dynamic and polar disorders. Our findings here of efficient SF through quintet states demonstrate that these conditions still allow efficient and controlled semiconductor operation and point toward future opportunities for constructing functional optoelectronic systems.


S3
Transients were recorded as the static magnetic field was swept and continuous-wave microwave irradiation was applied. All DNA/PEN assemblies were measured with a microwave power of 0.63 mW. We expect the time resolution of both setups to be limited by the Q-factor of the critically coupled MD5 resonator and therefore do not expect any significant effect on the kinetics due to the detection method (diode vs IQ mixer). For analysis, EPR spectra were simulated using EasySpin. [2] Cryo-TEM. Using a standard vitrification procedure, samples were plunge-frozen into liquid ethane in a ThermoFisher Vitrobot Mk3 on Quantifoil TEM grids. The TEM imaging was carried out in low electron dose conditions in a ThermoFisher Krios G3i operated at 300 kV using a Falcon3 camera, controlling acquisition using EPU.
Cleavage of the oligonucleotides from the solid support and subsequent deprotection was achieved by exposure to concentrated aqueous ammonia solution for 1 h at room temperature, followed by heating in a sealed tube for 5 h at 55 °C.
All oligonucleotides were characterised by negative-mode electrospray using a UPLC-MS Waters XEVO G2-QTOF mass spectrometer and an Acquity UPLC system with a BEH C18 1.7 µm column (Waters). A gradient of triethylamine (TEA) and hexafluoroisopropanol (HFIP) in MeOH was used (buffer A, 8.6 mM TEA, 200 mM HFIP in 5% MeOH/H2O (v/v); buffer B, 20% v/v buffer A in MeOH). Buffer B was increased from 0-70% over 7.5 min or 15-30% over 12.5 min for normal oligonucleotides and 50-100% over 7.5 min for hydrophobic oligonucleotides. The flow rate was set to 0.2 mL/min. Raw data were processed and deconvoluted using the deconvolution software MassLynx v4.1.

B. Syntheses
Reagents and solvents were obtained in high-purity grades from commercial suppliers. DNA strands were purchased from IDT and diluted in PBS (20 mM phosphate buffer, 200 mM NaCl). Alkyne-terminating, 12-mer ssDNA, with all protecting groups on and on-resin, was purchased from IDT and used as received. Chemical syntheses were carried out in oven-dried glassware under an argon atmosphere unless otherwise stated. Reactions were followed by analytical thin layer chromatography on aluminium-backed silica gel plates (Merck, 60 Å, F254).) and visualised with ultraviolet irradiation (λmax = 254 or 365 nm) or permanganate staining. Flash chromatography purification was carried out with Acros Organics ultra-pure silica gel (60 Å, 40 -60 µm) under a positive pressure of air.
No chemical characterisation was performed due to the insolubility in common organic solvents.

Compound 2a
In accordance with literature. [3] Quinone 1a (583 mg, 1.51 mmol, 1 eq.) was dispersed in THF (20 mL) at -78 °C. In a separate flask, 2.5 M n-BuLi in hexanes (2.16 mL, 5.4 mmol, 3.6 eq.) was added dropwise to a stirred solution of (triisopropylsilyl)acetylene (1.17 mL, 5.26 mmol, 3.5 eq.) in THF (30 mL) at -78 °C under argon and stirred for 1 h. The lithiated mixture was transferred to a dispersion of 1a by cannula at -78 °C. The reaction was allowed to warm to room temperature and stirred overnight. 5 mL of saturated SnCl2·2H2O in degassed 10% HCl was added by injection to the crude. The crude mixture was degassed for a further 5 min, then left to stir for 3 h.

Compound 4a
Pentacene 3a (300 mg, 0.41 mmol, 1 eq.) and Et3N (86 μL, 0.615 mmol, 1.5 eq.) were mixed in 40 mL DCM under argon at 0 °C. Methanesulfonyl chloride (82 μL, 0.48 mmol, 1.1 eq.) was added by injection; the reaction was allowed to warm to room temperature and stirred for 1.5 h. The mixture was diluted with H2O (400 mL) and extracted with DCM. The combined organic phases were washed with brine, dried over MgSO4, filtered, and dried under reduced pressure. The blue solid was used immediately in the next step without further purification.

Compound 5a
Pentacene 4a (ca. 0.40 mmol, 1 eq.) was dissolved in DMF (30 mL) under argon at room temperature NaN3 (31.2 mg, 0.48 mmol, 1.1 eq.) was added as a solid in one portion and stirred overnight. The mixture was diluted with H2O (400 mL) and extracted with CHCl3. The combined organic phases were washed with brine, dried over MgSO4, filtered, and concentrated under reduced pressure. The solid was purified by column chromatography (silica gel, DCM/heptane 9:1) to yield 5a as a blue solid (174 mg, 0.23 mmol, 55 % over two steps).

Compound 6a
Resin-bound alkyne-terminating 12-mer ssDNA (15 nmol/mg, 2.1 mg, 31.5 nmol, 1 eq.) and 5a (2.5 mg, 3.15 µmol, 50 eq.) were mixed as solids in an argon-filled 300 µL Eppendorf® tube. The solids were degassed under gentle argon flow. In a separate flask, a 0.1 M tris(benzyltriazolylmethyl)amine solution in DCM was degassed and 0.5 µL (1.5 eq.) was added to the ssDNA-resin/5a mixture. Simultaneously, in another flask, 0.1 M CuBr in MeCN solution was degassed, and 0.34 µL (1.1 eq.) was added to the ssDNA-resin/5a mixture. The Eppendorf® tube was degassed under argon flow, sealed with PTFE tape and Parafilm®, and shaken for 2 d at room temperature. The crude was spun down to form a resin pellet. The supernatant containing residual 5a was extracted and separately purified by Chelex® to recover pure 5a. The blue resin was washed with DCM and the supernatant removed until the washings ran colourless. The blue resin was washed with MeCN three times. Finally, the blue resin was washed twice with ether and left to dry, yielding 6a as a blue-stained resin.
Illustration of the purification/washing steps of the resin-bound pentacene 6a.

Compound 4b
Pentacene 3b (81 mg, 0.1 mmol, 1 eq.) and Et3N (18 μL, 0.23 mmol, 2.3 eq.) were mixed in 40 mL DCM under argon at 0 °C. Methanesulfonyl chloride (28 μL, 0.2 mmol, 2 eq.) was added by injection; the reaction was allowed to warm to room temperature and stirred for 1.5 h. The mixture was diluted with H2O (400 mL) and extracted with DCM. The combined organic phases were washed with brine, dried over MgSO4, filtered, and dried under reduced pressure. The blue solid was used immediately in the next step without further purification. Rf = 0.9 (DCM/EtOAc 4:1)

Compound 6b
Compound 6b was prepared by mixing azide 5b and resinbound alkyne-terminating 12-mer ssDNA in a 300 µL Eppendorf® tube, and following the synthesis procedure and purification method described above for compound 6a. In three separate 300 µL Eppendorf® tubes: 1 M CuSO4·5H2O in ultrapure H2O, 1 M tris(3hydroxypropyltriazolylmethyl)amine in ultrapure H2O, and 2 M Na-ascorbate in ultrapure H2O. In another 300 µL Eppendorf® tubes 6b was kept under argon, and 3'-alkyne DNA (2 eq) in ultrapure H2O was added, and the suspension degassed for 5 mins. The solutions of CuSO4·5H2O, tris(3-hydroxypropyltriazolylmethyl)amine and Na-ascorbate were mixed in a 1:2:4 volume ratio and degassed, turning from blue to yellow. 10 eq. of this mixture were added to 6b and the suspension degassed for 5 mins. The Eppendorf® tube was degassed under argon flow, sealed with PTFE tape and Parafilm®, and shaken for 2 d at room temperature The crude was spun down to form a resin pellet. The supernatant was removed. The blue resin was washed with ultrapure H2O, shaken, spun down and the supernatant removed. This was repeated three times with ultrapure H2O, and three times with MeCN. Finally, the blue resin was washed twice with ether and left to dry, yielding compound 7 as a blue-stained resin.

C. DNA design
The ssDNA-based pentacene samples discussed in this study are constructed with a 12 nucleotide (nt) ssDNA linked via a phenylene and triazole bridge to the 2 position of the pentacene core (see SI, Section B for the synthesis and chemical structures).
The dsDNA-linked constructs in this study are formed from up to seven partially complementary ssDNA strands. Each position within the constructs corresponds to two specific 12 nt sequences (11 nt for terminal strands) on either side of the pentacene cargo -see base sequences below. Pairwise complementary DNA strands are colour coded.

D. Atomistic metadynamics molecular dynamics simulations Parameterisation
The parameters for pentacene cargo bound to one/two strands of DNA were developed separately. In each case, the structures were sketched using Marvin and hydrogen atoms were added using Chimera. [5] The PyRed server [6] was used to calculate the electrostatic potential of the structures at the HF/6-31G* level of theory using Gaussian09, and then perform a two-stage Restrained Electrostatic Potential (RESP) fit [7] to calculate the atomic charges. Antechamber [8] was then used to assign atom types to the semiconductor/linkers according to the GAFF2 parameter set. [9] Bonded and nonbonded interactions of the Silyl groups were modelled using the parameters of Dong et al. [10] ds-DNA and ss-DNA strands were built 6 basepairs long using Avogadro. [11] The xleap module of Amber16 [12] was then used to append pentacene to the DNA strands. The DNA strands were parameterized using the parmbsc1 [13] parameters. The system was solvated in an octahedral box of TIP3P [14] solvent and 0.15 M NaCl ions that used the parameters of Joung and Cheatham. [15] Finally, the amber parameters were converted to Gromacs [16] format using Parmed. [17] Simulation setup Models were first energy minimized until the maximum force on any atom was below 1000 kJ mol −1 nm −1 . The systems were then equilibrated using restraints on the DNA and semiconductors (force constants of 1000 kJ mol −1 nm −2 ) for 1 ns in the NVT ensemble followed by 1 ns in the NPT ensemble.
The simulations were performed starting with random velocities obtained from a Maxwell-Boltzmann distribution at 300 K and using a pressure of 1 bar. The temperature was kept constant using the V-rescale thermostat. [18] The pressure was maintained using the Parrinello-Rahman barostat. [19] Long-range electrostatics were calculated using the particle mesh Ewald (PME) algorithm [20][21] with a cutoff of 1.0 nm. The simulations used a timestep of 2 fs and were performed using Gromacs 2021 [16] patched with Plumed 2.7.2. [22][23] Trajectories were analysed using a combination of Gromacs and Plumed tools together with Python MDAnalysis [24] scripts.

Metadynamics simulations
Following equilibration, well-tempered metadynamics simulations [25][26] were performed to investigate the relative interactions of Pentacene molecules. Metadynamics simulations promote sampling along collective variable (ξ) space through the addition of an external biasing potential VE constructed as a sum of gaussians as where τG is the time interval at which the gaussians are added with height W, width δ and mean ξti. Here, we used two collective variables (i = 2) that determine the relative stacking orientations of the semiconductors and shown by our prior work [27] to promote sufficient sampling. ξ1 (also denoted as r) is defined as the Euclidean distance between the centres of the central aromatic rings of the two pentacene molecules. ξ2 (also denoted as θ) is defined as the angle between the longitudinal axes of the two pentacene molecules (SI, Figure S1a).
The metadynamics biasing potentials were deposited every 10 ps using a bias factor of 15 and gaussian widths of 0.002 nm and 0.005 radians along ξ1 and ξ2 respectively. The simulations were performed for 100 ns and convergence was assessed using the time-evolution of the free-energy profiles, which are invariant after ~50 ns (disregarding the timedependant constant) (SI, Figure S1b).

Normalized probability density calculations
The separation between pentacene molecules were calculated by projecting the distance between the centres of the central aromatic rings along long x and short y molecular axes (SI, Figure S1c). However, the addition of the history dependent bias potentials in metadynamics simulations precludes a direct ensemble averaging of the system's characteristics as simulation time is without physical meaning. Here, we used the methodology of Bonomi et al. [28] to S12 reweight trajectory frames and calculate unbiased equilibrium ensembles. Briefly, the probability distribution of a biased system along molecular coordinates r can be expressed as where U(r) is the underlying force field potential and β is the thermodynamic temperature. Disregarding the timedependent bias offset and assuming convergence along the collective-variable space, this biased probability density can be related to the unbiased probability density P0(r) as Thus, the offset projections between the two pentacenes (Δx and Δy) were first calculated for each trajectory frame and subsequently used to calculate a two-dimensional weighted histogram to obtain the normalized probability density (NPD). (c) Illustration of the points used to define the long (x, green dots) and short (y, yellow) molecular axes of the pentacene molecule.
(d) MD analysis of PEN2 (see Figure 2). Left: Normalised probability density (NPD) for displacements along the x and y molecular axes. Right: 1D representation of the NPD profiles for displacements along the x, y, and z molecular axes. S13 E. Self-assembled dimer (ssDNA-PEN)2 Figure S2. Synthesis and modelling of the self-assembled pentacene dimer.
(b) Molecular dynamics (MD) simulations predict the self-assembly of ssDNA-PEN into (ssDNA-PEN)2 dimers. In the MDoptimised geometry, the two ssDNA strands associate into a weakly bound double-helix and the pentacenes form a slipstacked dimer.
(c) MD analysis of the pentacene dimer geometries in (ssDNA-PEN)2. The map shows the normalised probability density for the lateral offset along the pentacene long (Δx) and short (Δy) molecular axes. The simulation indicates a range of probable configurations with slightly more or less slip-stacked geometry, which are likely to co-exist in the samples.
(d) In the MD-optimised geometry, the pentacenes form slip-stacked cofacial dimers with a stacking distance (Δz) of 3.6 Å and large longitudinal offset of 7.2 Å.
Attachment of one ssDNA to the pentacene semiconductor yields a highly amphiphilic compound with the non-polar, strongly hydrophobic TIPS-pentacene on one side and the charged, hydrophilic ssDNA on the other (SI, Figure S2a). When dissolved in aqueous buffer solution, this ssDNA-PEN self-assembles spontaneously into dimers where the hydrophobic pentacenes aggregate to minimise their interface with the surrounding polar medium. This spontaneous dimer formation is fully reversible when a solvent that can dissolve the pentacene moieties is added. In DMSO/buffer 95:5 (v:v), ssDNA-PEN is monomeric (see SI, Figure S7).
To study this dimer formation and assess the dimer geometry, we employed molecular dynamics (MD) simulations, as described in the SI, Section D. Starting from two separated ssDNA-PEN, the simulation converges to a dimer, in which the pentacenes arrange in a slip-stacked cofacial geometry and the two identical, non-complementary ssDNA form a weakly bound double helix (SI, Figure S2b, S3a). Since both ssDNA strands have the same base sequence and directionality, their association is much weaker than in a DNA duplex of two complementary strands.
Interestingly, the self-assembly of ssDNA-PEN generates dimers selectively and does not extend towards larger aggregates. When we initialise the simulation with three ssDNA-PEN, only two of them dimerise, while the third pentacene remains well separated (SI, Figure S3b). This finding is supported by electron microscopy studies (SI, Section G), where we observe homogeneous samples with objects whose size is fully consistent with the simulated (ssDNA-PEN)2 dimers, but no indication of more extended structures.
While the hydrophobic/hydrophilic contrast is the main driving force for the dimer formation, controlling the dimer geometry on a sub-molecular length scale requires additional interactions (SI, Figure S4a). The pentacene core is a flat aromatic unit that favours cofacial arrangements with typical π-π distances of around 3.5 Å. The triisopropylsilyl (TIPS) ethynyl substituents on the 6 and 13 positions are sterically demanding but positioned with sufficient distance from the core to allow the π-stacking of adjacent pentacenes. Due to their steric bulk, they enforce a substantial offset of 6-8 Å along the long pentacene molecular axis, while simultaneously limiting the offset along the short molecular axis to below 1 Å. This way, the molecular geometry generates a docking site, which locks the second pentacene in a very well-defined geometry. The effect of this docking site leads to very similar dimer geometries in the DNA-linked PEN2 and the selfassembled (ssDNA-PEN)2, which both closely resemble the packing motif in crystalline TIPS-pentacene (SI, Figure S4b).  (a-c) The self-assembled (ssDNA-PEN)2 appears as slightly elliptical objects that are about 4 nm long and 3 nm wide. This is in excellent agreement with the dimensions predicted by the MD simulations and confirms that the self-assembly exclusively generates dimers and no larger structures. Deconvolution of the spectra allows us to extract the fission rates of the DNA assemblies. We use a genetic algorithm (GA) to obtain species associated spectra and kinetics. [30] We use the characteristic absorptive feature at 450 nm, present in the pentacene singlet spectra of the monomeric ssDNA-PEN (Figure 2b and SI, Figure S7b) and isolated from absorptive triplet features, as an indicator of the singlet associated species. The other spectra, which exhibit a strong absorptive feature at 530 nm, are assigned to the triplet species (SI, Figure S8, left panels). We subsequently fit the kinetic decays and growths of the singlet and triplet associated species respectively to single exponential functions (SI, Figure S8, right panels), such that the rate of triplet growth matches that of singlet decay, using the following function: The outputted parameters are summarised in the

Time-evolution of ESR spectra
Here we show the full transient ESR data, the intensity of microwave emission and absorption as a function of magnetic field and time after the laser flash, for each sample. The DNA-assembled pentacene samples (SI Figure S11, columns 1-3 from left) show the evolution from predominantly quintet to triplet through time, while the TIPS-pentacene film shows the evolution from the initial triplet to secondary triplet spectrum with an additional central feature consistent with charges or charge-transfer states (SI Figure S11, column 4). We then take the quintet and triplet absorptive peak positions and plot the magnitude of the trESR signal as a function of time to extract a triplet and quintet lifetime. We note that the TIPS-PEN sample exhibits a rapid decay as the spectrum evolves into that shown at late times in SI Figure S11 (right). The fitted rise and decay times are given in SI Table S2 and all triplet states in the DNA-assembled samples are found to decay with a timescale of 1-2 µs. Figure S12. Kinetics for triplet and quintet absorptive peak transitions with dashed fitted lines and fit parameters given in Table S2 below.

Least-squares fitting parameters of trESR spectra
Quintet spectra are taken from the earliest time species for each DNA assembled structure. For (ssDNA-PEN)2 we observe significant spectral overlap between triplet and quintet throughout the early time evolution, and so we separate independent spectral components to extract the early time quintet spectrum based on its distinct time-evolution using the independent component analysis (ICA) function in scikit-learn library. [31] The only DNA assembled structure with a clear, measured long lived triplet signal with no quintet component is the (ssDNA-PEN)2 sample. We therefore use this long-time spectrum to estimate the zero-field splitting (ZFS) parameters, i.e., spin energy level splitting in the absence of a magnetic field, of the triplet on the DNA-assembled pentacene derivatives. We fit these spectra for each sample by combining the open-source Matlab-based ESR library Easyspin [2] and least-squares linear regression in SciPy. [32] The fitted D and E parameters define the ZFS parameters in MHz. The line broadening is taken to be Lorentzian in form and is reported as the full-width-half-maximum (FWHM) in Gauss.
The normalized spin-sublevel populations used to fit the spectra are further reported as follows. For S = 1 species the populations are either the high field populations of the m = 0,±1 sublevels quantized along the external magnetic field direction or the zero-field x,y,z populations where the x,y,z states are the eigenstates at zero magnetic field (indicated with an asterisk in Table S3 for (ssDNA-PEN)2). For S = 2 species, populations are reported as the m = 0,±1,±2 eigenstates quantized along the external magnetic field direction. The ordering in Table S3 is from lowest to highest energy eigenstates.
Where necessary an orientational ordering parameter is used to capture a degree of selectivity in spin polarization with orientation of the molecular axes relative to the external magnetic field. This parameter is defined in the Easyspin documentation. [2] We note that we expect spin populations may be orientationally selective due to the dipolar contribution to spin mixing between the singlet and quintet sublevels. The negative fitted orientational ordering parameter indicates a slight preferential ordering of the xy axis to the B0.
The best-fit parameters are summarized in the chart below with corresponding comparison to experimental spectra in SI Figure S13.  Figure S13. Fitted S = 2 quintet (a) and S = 1 triplet (b) spectra in black overlayed with corresponding experimental data from parameters in SI Table S3. Experimental time points for each spectrum are listed in SI Table S3 above.

Theoretical quintet (S=2) zero-field splitting parameters
We compare the fitted ZFS parameters , to the theoretical ZFS parameters taken from the geometry of the triplet pairs given by molecular dynamics simulations described in SI Section D. The theoretical quintet ZFS parameters from molecular dynamics ( ℎ , ℎ ) are determined following the theory laid out in Ref [33]. In short, in the limit of strong exchange coupling between triplets ( ≫ ) where is the inter-triplet exchange interaction and is the intratriplet dipolar interaction strength. The quintet zero field splitting is determined by the combination of intratriplet dipolar interactions and intertriplet dipolar interactions as follows.
The quintet zero-field splitting Hamiltonian interaction where 0 is the magnetic permeability of free space, the Bohr magneton, the g-factor, ��⃑ the centerto-center vector between triplet-bearing molecules, and � the unit vector between molecules a and b. We note that a more detailed derivation is given in Ref [33].
The input parameters we use are , = (1108,16) MHz (see section above) and � , � ,̂ axes defined along the long, short, and out-of-plane axes. [34] The relative orientations of each set of triplet axes and the inter-triplet dipolar interaction strength and unit vector we then take from molecular dynamics calculated geometry (see SI Section D for MD simulation details). In SI Table S4, we report the resulting theoretical quintet ZFS parameter ( ℎ , ℎ ) along with the inter-triplet center-to-center distance. Table S4. Calculated theoretical ZFS values for the experimental crystal structure of TIPS-pentacene, the MD-simulated dimer structure of DNA-linked TIPS-pentacene, and the fitted values from ESR described above. Measured (ESR) 350(5) 12(5) -/-*Geometry from reported crystal structure [29] , †Geometry described in Figure 1 and SI Section D,E.

S23
The systematic offset in the measured , parameters relative to the theoretical predictions could be due to strain in the dimer geometries in experiments versus theory. To determine whether this is possible, we simulate the degree to which the measured ZFS parameters vary with variation in intermolecular geometry using reasonable possible distortions. Namely we allow the inter-triplet distance to vary following a normal distribution of each component of ��⃗ = ( , , ) with standard deviation of 0.5 Å and variation in the rotation of each triplet ZFS axes with a standard deviation of 10° (rotation about a uniformly distributed random axis). We then compare the distribution of ℎ , ℎ to the measured values with least-squares fitted ZFS parameters shown in black and standard error from fitted spectra in grey. As shown in SI Figure S14 the experimentally measured , values fall within the distribution of ℎ , ℎ values from MD simulations including a random degree of strain. We note that , are sensitive to both (1) inter-triplet distance and (2) relative rotation of the pentacene cores and both factors likely contribute together to the observed systematic shift in measured , relative to the MD simulation. We also note that a pair residing on next-nearest neighbors with J > D would have reduced ZFS parameter values and this could also contribute to the distinct values compared to theory.
The function from intermolecular geometry to ZFS parameters is not one-to-one and so it is not possible to extract a single geometry from the measured D,E parameters. We conclude that while the measured ZFS parameters are not consistent with the exact intermolecular geometry given by MD simulations, they would be consistent with a small degree of strain in frozen solution compared to in silico molecular dynamics. Figure S14. Comparison between the experimental (black) and calculated distribution (blue) of the D,E parameters.