Origin of biological homochirality by crystallization of an RNA precursor on a magnetic surface

Homochirality is a signature of life on Earth, yet its origins remain an unsolved puzzle. Achieving homochirality is essential for a high-yielding prebiotic network capable of producing functional polymers like RNA and peptides on a persistent basis. Because of the chiral-induced spin selectivity effect, which established a strong coupling between electron spin and molecular chirality, magnetic surfaces can act as chiral agents and be templates for the enantioselective crystallization of chiral molecules. Here, we studied the spin-selective crystallization of racemic ribo-aminooxazoline (RAO), an RNA precursor, on magnetite (Fe3O4) surfaces, achieving an unprecedented enantiomeric excess (ee) of about 60%. Following the initial enrichment, we then obtained homochiral (100% ee) crystals of RAO after a subsequent crystallization. Our results demonstrate a prebiotically plausible way of achieving system-level homochirality from completely racemic starting materials, in a shallow-lake environment on early Earth where sedimentary magnetite deposits are expected to be common.


Syntheses of aminooxazolines from glyceraldehyde for direct crystallization
DL-glyceraldehyde (90.08 mg, 1 mol) was dissolved in 2 mL H2O and 2-aminooxazole (84.08 mg, 1 mol) was added. The resultant mixture was then stirred until a homogenous solution was obtained. This solution was then incubated at 40ºC for 24h and during which time it developed a yellow-brown coloration. The pH of the incubated mixture was measured to be 8.5. The incubated mixture was then analyzed by 1 H NMR spectrometry before the crystallization experiments and the relative ratios of aminooxazolines were determined by integration.

Fabrication of magnetite (Fe3O4) surfaces
We fabricated the magnetite films using the procedure demonstrated by Jubb and Allen (2010) (52). We first deposited iron films by the electron-beam evaporation of iron on a silicon substrate.
We used a 150 mm diameter wide and 625 µm thick boron doped P type Si <100> wafers from University Wafers (catalog ID number 3784). For the evaporation, we used SHARON EE-3 ebeam evaporation system located at the Harvard Center for Nanoscale Systems (CNS). A Fig. S3. 100 nm thick iron (Fe) evaporated on a silicon substrate (100 mm diameter, 500 µm thickness) gives a metallic, shiny gray surface (left). After the oxidation of the film in air at 175ºC for 4 hours, iron oxidizes to a smooth film of about 200 nm magnetite, Fe3O4, with a green-yellow-black color (right). cryopump and an oil-free roughing pump provide the evaporator vacuum. We used a FABMATE® crucible liner made out of high-strength graphite by Kurt J. Lesker (Part number:  to place the iron as an e-beam source. As the iron source, we used iron 99.95% pure iron pellets (1/8" diameter by 1/8" long) by Kurt J. Lesker (Part number: EVMFE35EXED). We filled the crucible liner with the iron pellets and placed it inside the e-beam evaporator, together with the silicon substrate as the target. We vacuumed the system down to 5 x 10 -6 Torr before we started the deposition process. We used the density of 7.86 g/cm³ and the z ratio of 0.349 for the iron evaporation. We turned on the e-beam source after the vacuum is stabilized and waited until a stable deposition rate of 0.5 Å/s is reached. We then opened the target shutter and started the iron deposition. We used a quartz crystal to monitor the in-situ thickness of the iron film and slowly deposited 100 nm of iron. We found that the slower the deposition rate the smoother the resultant iron film. We found an optimal deposition rate of around 0.2-0.7 Å/s. We then took the iron sample on the silicon substrate out of vacuum and placed it inside an oven (Binder Model FD 56 Laboratory Oven) set at 175ºC under ambient air. We oxidized the iron film for 4 hours to magnetite (Fe3O4) and observed that the shiny gray iron film turned into a smooth layer of greenish-yellow, black magnetite film, as seen in Fig. S3. According to the cited procedure (52), 100 nm of iron should give around 200 nm of Fe3O4 after the oxidation due to oxygen incorporation into the iron lattice.
After the magnetite films were fabricated, as a precautionary measure, we kept them in a vacuum desiccator to prevent the further oxidation to hematite and hydrous iron oxides. However, the films were stable under oxic, ambient conditions for weeks and we have not observed any physical damage or chemical degradation while doing experiments in water solutions.
6. Characterization of magnetite (Fe3O4) surfaces 6.1 AFM Fig. S4: AFM topography image (3D image on the left, 2D top view on the right) of a magnetite film prepared from 100 nm iron deposited on a silicon wafer, with a slow deposition rate of 0.5 Å/s. Peak-topeak surface roughness of 8.93 nm and an RMS roughness of 1.22 nm is measured for a randomly picked 1 x 1 µm area.
We characterized the surface properties of the magnetite films with a commercial AFM (Cypher S by Asylum Research) located at Harvard CNS facility. We used the AC mode (dynamic mode) and we imaged in air. We used a cantilever (Nanosensors Type SSS-NCH-10) with a nominal force constant of 42 N/m and a resonance frequency about 330 kHz. The cantilever oscillated at the auto-tuned resonance close to its frequency. The scan parameters were set at a scan rate of 1.00 Hz, an integral gain of 30.00, and a setpoint of 800.00 mV.
The tip's amplitude and phase were monitored with a photodetector by using the tip deflection. We took scan ranges of 1.0 x 1.0 µm, 5.0 x 5.0 µm, and 10.0 x 10.0 µm and calculated the root-mean-square (RMS) roughness from the 3D topography data.

FTIR
We characterized the vibration al spectra of magnetite films by Fourier transform infrared (FTIR) spectroscopy. For FTIR measurements, we used a potassium bromide (KBr) window by ThorLabs (25 mm diameter, 5 mm thick, Part number: WG10255). We chose this material due to its low absorption in the mid-IR range. We coated the KBr window with a 20 nm iron film using e-beam evaporation as described in Supplementary  to 350 cm -1 ). For the measurements, we used a resolution of 5 cm -1 , sample scan average of 100 scans, phase resolution of 16 in the Mertz phase correction mode. We used a MIR source and a KBr beam-splitter with a RT-DLaTGS detector. We took the measurements in the range of 700-350 cm -1 . We placed a KBr window into the background arm and the KBr/40 nm Fe3O4 sample into the sample arm of the spectrometer and used the background subtraction feature of the spectrometer. The measurement gave a prominent peak centered at 557 cm -1 (black vertical line in Fig. S5) with a full width at half maximum of around 50 cm -1 , as shown in Fig. S5. The central location of the peak agrees very well with the phonon mode splitting of magnetite, Fe3O4, as observed by Jubb and Allen (560 cm -1 ) and theoretically  Fig. S4). We were not able to observe the magnetite mode at 350 cm -1 due to high background KBr absorption and we attribute the small ripples in the background to the residual water IR absorption. The absorption intensity we measure is not physical due to the reflection losses from the shiny magnetite film. FTIR spectra shows that iron films are selectively oxidized into magnetite and not into other iron oxides like maghemite (much less magnetic than magnetite) and hematite (not magnetic).

UV-VIS
We measured the optical spectra of magnetite films by ultraviolet-visible (UV-Vis) spectroscopy. For UV-Vis measurements, we used an uncoated, broadband UV fused silica window by ThorLabs (25.4 mm diameter, 1 mm thick, Part number: WG41050). We chose this material due to its low absorption in the optical range. We coated the fused silica window with a 20 nm iron film and when we oxidized the sample we obtained a 40 nm magnetite film according to Jubb and Allen (2010) (52). We used a commercial UV-Vis spectrometer by Shimadzu (Model: UV-1900, 190 nm to 1100 nm). We took the measurements in the 190 -700 nm range with a resolution of 2 nm, using the slow acquisition mode. We placed a fused silica window into the background arm and the fused silica/40 nm Fe3O4 sample into the sample arm of the spectrometer and obtained the background subtracted UV-Vis spectra. The measurement gave a broad absorption feature, slowly rising as the wavelength decreased as seen in Fig. S6.

SQUID
We measured the magnetic properties of the magnetite samples by a superconducting quantum interference device (SQUID). We used the 200-nm-thick magnetite samples for the SQUID measurements on the silicon substrate. We took the measurements at room temperature by magnetizing the sample parallel to the surface normal. We separately measured the silicon substrate and subtracted its diamagnetic contribution. We measured the hysteresis curve of the sample with a magnetizing run from 0 Oe to 10 kOe, a first measurement from 10 kOe to -10 kOe, and a second one back to 10 kOe, ending with a demagnetizing run from 10 kOe to 0 Oe. We used a scan rate of 50 Oe/s. We measured the coercive field to be around 600 Oe, confirming the soft magnetic nature of magnetite. In addition, we observed that the sample does not saturate even at fields as high as 10 kOe, reaching a moment of above 1000 µemu. Around the fields the crystallization experiment were conducted (~ 3000 Oe) the moment was measured to be around 600 µemu.

General method for crystallization experiments
For the crystallization experiments we cut the magnetite wafers such that they could fit inside a 35 mm diameter Petri dish (Falcon Corning, Polystyrene, SN:351008, VWR#:25373-041) used for the experiments. We typically used samples with dimensions around 20 mm height by 15 mm width. Before every experiment we wiped the magnetite samples with acetone and then with ethanol to remove any organic residues and contaminants. We then secured the magnetite samples in a Petri dish horizontally by a double-sided kapton tape such that they are stable.
Before we added the solution, we blew the samples with clean air to get rid of any dust particles, which can act as nucleation sites for crystallization and reduce enantioselectivity. We then placed the Petri dishes containing the magnetite samples on top of a strong magnet (DIYMAG 40 x 40 x 20 mm N52 Neodymium magnet) such that the magnetization direction of the magnet is parallel to the magnetite surface normal. Before placing the surface on the magnet we marked the pole of the magnet so that we knew the spin direction for the particular experiment. North pole corresponds to surfaces polarized parallel to the surface normal and the South pole corresponds to surface polarized anti-parallel to the surface normal. We measured the magnetic field strength and the pole direction (North or South) with a Hall probe (Extech MF100 AC/DC magnetic field meter). For the reported experiments, we have used a magnetic field strength of 325 mT at the magnetite sample location as measured by the magnetic field meter. Afterwards, we placed the Petri dish with the magnet underneath into a fridge kept at 12ºC. After placing the magnets, we waited for about 10 minutes so the samples reached thermal equilibrium. Then we prepared a 2 mL racemic solution of RAO by fully dissolving the powdered material in warm water and slowly put the solution on the magnetite surface making sure that whole surface is covered with the solution, and no air gap is left. We have not filtered the solution before placing on the surfaces with a filter (e.g., 0.22 um pore size filter membrane), in order to decrease the number of possible nucleation sites due to the fact that a non-filtered solution is more representative of the natural prebiotic conditions. However, in principle, we think that using filtration paper or membrane can enhance the number of enantioselective crystallizations seeded on the magnetite surface. After we placed the solutions, we closed the Petri dish with its lid and did not disturb the experiment for about 24 hours, in order not to contaminate it. Typically, after 24 hours to 48 hours we stopped the crystallization and slowly collected the solution above the surfaces together with the floating racemic powder inside the water column. We then washed the surfaces with cold water several times to get rid of the racemic liquid and the floating crystals. We used cold water such that we could clean the surface without dissolving the crystals attached to the surface. After the surface has been cleaned, we waited for about 30 minutes such that the crystals dry and were easier to collect. We then collected the crystals individually under a stereomicroscope (Amscope 7x-45x, SM-2T-6WB-V331) using tweezers. We recorded how many crystals we collected for each surface.
We should mention that we collected the crystals formed on the magnetic surface and we discarded the ones forming on the plastic areas of the Petri dish and the crystals formed on the rough edges of the silicon wafer as shown in the schematic in Fig. S9. After collecting the crystals, we fully dissolved them in pure water for CD experiments. We occasionally discarded experiments if they had a contamination acting as a nucleation site such as visible pieces of lint or dust. We observed that if large pieces of dust are floating inside the solution they aggregate many crystals on top of each other. These aggregated crystals drop on the magnetic surface due to gravity and invalidate the experiment. Another crucial point is to find an optimal time to stop the crystallization. Because the enantioselective crystallization effect is a kinetic one, after a point increasing crystallization yields decreases selectivity. Furthermore, because the solution is not kept always racemic, if one enantiomer is depleted too much from the solution with selective Finally, yet importantly, once a crystal forms and grows, its face can act as a seed for the opposite enantiomer forming twinned racemic clusters nullifying the selectivity seeded by the surface. We found that the slower the crystallization and the earlier the crystals are collected the higher is the enantioselectivity. Optimal time may vary from experiment to experiment, but as a general guideline, collecting the crystals as soon as they are visible by eye and collectible by tweezers maximizes the measured enantioselectivity. Finally, if the experiments are started in a "fresh" room with fewer air seeds available, the optimal conditions may vary after a while when the room becomes saturated with the air seeds of the aminooxazoline. Although air seeding is an anecdotal phenomenon, we observed that the initial experiments required slightly higher (about 10-15% higher) concentrations for enantioselective crystallization compared to the final, optimal conditions. We attribute this to the fact that we started the experiments in a fresh laboratory in which RAO molecule had not previously been studied; therefore, no crystal seeds of RAO were available in the air.

Importance of using smooth magnetic surfaces
Before we started using the smooth thin films of magnetite made by e-beam evaporation, we experimented with magnetite powder and spin-coated magnetite nanoparticles and we have not observed the enantioselective effect. In the case of magnetite powder precipitated on the bottom of a cuvette due to gravity, the particles on the surface are still freely rotating. And therefore both sides of each particle surface are exposed although the magnetic field direction is pointing only one way. This is like the case described in the left panel of Fig. S10. When a chiral molecule approached a magnetic surface parallel or antiparallel to the direction of the magnetic field, no net selective effect is observed as on the upper face one enantiomer is selectively crystallized and on the lower face, the opposite enantiomer is crystallized.
Therefore, a loose, soft sediment is like a freely diffusing particle in the bulk in terms of being a seed for selective crystallization. For the case of a well-defined but a rough surface (middle column of Fig. S10) the situation is different. For rough surfaces with net spin alignment, depending on the spin vector's overlap with the "averaged" surface normal there is a net selective effect. However, for a rough surface, the surface normal is not uniform and one has to average the spin overlap for all locally flat surfaces. Therefore, the enantioselective effect is reduced although the chiral symmetry is still broken by the magnetic surface. For an ideal surface with perfect uniformity enantioselective effect is maximized as the spin vectors can perfectly overlap with the surface normal therefore with the molecular frame of chiral molecules. Of course, in a natural scenario, there maybe no such thing as a perfectly smooth surface. Therefore, the selectivity achieved with pristine surfaces should be large, as we have demonstrated, such that in a natural scenario the effect persists on rougher surfaces albeit with diminished magnitude.

Direct crystallization of DL-RAO from the mixture of aminooxazolines
We first synthesized a racemic mixture of aminooxazolines by incubating 0.5 M DLglyceraldehyde with 0.5 M 2-aminooxazole in 2 mL pure water at 40ºC for 24 hours. We then put the incubated solution on a magnetized magnetite surface and allowed for crystallization.
Typically, within a few hours we started to observe needle-shaped crystals appearing on the  On the left, a surface level ee is achieved due to the low number of twinned clusters. Individual needles are homochiral and they are seeded by the magnetic surface. On the right, no enantioselectivity is achieved due to the crystal twinning in the form of rosette shaped clusters. Note the powdery racemic material which can be removed along with the mother liquor by aspiration (mimicking the effect of drainage in a geochemical setting).
magnetite surface almost simultaneously with the rosette-shaped ones as seen in Fig. S12.
We stopped the crystallization as soon as the crystals were visible however, we could not reliably control the formation of rosette-shaped twinned clusters. We then collected the crystals one-by-one and dissolved them in 2 mL pure water. We first confirmed that they consists of only RAO by 1 H NMR, within the limits of NMR detection. For NMR studies, we added 10% D2O to our solutions.
We then checked the circular dichroism spectra of the solution. For direct crystallization, we could not reliably get a net CD signal. We then realized that the rosette-shaped crystal clusters forming from the racemic solution are racemic, twinned crystals of RAO with individual homochiral domains. Therefore, the twinning nullified the effect seeded by the surface, so no net enantiomeric excess was measured. We think that, if concentration is optimized and crystallization is slowed down, cluster formation can be circumvented and a net effect can be observed. For the control experiments, we crystallized the same starting solution on silicon substrates (without magnetite on them), in the presence of magnetic field.
For this case, we did not observe large needles and rosettes but a flaky, white powder of racemic RAO covering the silicon surface, as shown in Fig. S14. Under a high-magnification microscope, we have observed that the flaky powder actually consists of tiny needle-shaped crystals of RAO.

Re-crystallization of DL-RAO
We made a 65 mM solution of DL-RAO in 2 mL pure water. We then put the solution on magnetized magnetite and we typically started observing crystal formation in about 12 hours as shiny speckles on the surface. We collected the crystals one by one with tweezers usually in 24 to 48 hours depending on the existing crystal amount on the surface. We observed that the earlier we collected the crystals the higher the selectivity we obtained. There is typically a tradeoff between selectivity and the amount of crystals collected and we did not to collect less than 10 crystals for an experiment as due to small number statistics high ee's can be  deceptively obtained. After we collected the crystals, we fully dissolved them together in pure water and obtained their CD spectra. Table S2 shows the repeated experiments with different pole directions to accumulate statistics. The North pole (up spin) selects the crystals of D-RAO and the South pole selects the crystals of L-RAO. As one can notice, even for high number of individually homochiral crystals we collected (e.g., experiments # 1, 3, 7) we reliably observed spin-selective crystallization with high overall enantiomeric excess. CD spectra of the experiments displayed in Table S2 can be seen in Fig. 3 in the main manuscript and the raw data is shown on Fig. S16.

Re-crystallization of enantioenriched RAO
We prepared enantioenriched solutions of D-and L-RAO by mixing the racemic DL-RAO powder with the enantiopure D-and L-RAO powder. We measured the ee of the starting    Table S2. solutions to be -25% and 20% for D-RAO and L-RAO respectively. We then crystallized the solution enriched with D-RAO on the North pole (up spin polarized magnetite) and the solution enriched with L-RAO on the South pole (down spin polarized magnetite). After recrystallization, we found that the solution with -25% ee gave 26 crystals of enantiopure D-RAO. And for the solution with 20% ee we obtained an increased ee of 80% after the second crystallization. The raw data for the crystallization of the enantioenriched solutions can be seen in Fig. S17.

General method for CD measurements
After we collected the crystals, we fully dissolved them in 2 mL of pure water. We took the CD spectra of the dissolved solution in a quartz cuvette with a path length of 10 mm (Starna Cells Inc 1-Q-10-GL14-C). Before each measurement we took the CD and UV-Vis spectra of the water-cuvette background and subtracted this background from the sample measurements using the auto-baseline subtraction feature of the spectrometer. We also made sure that the absorption peak of the solutions is below OD = 1 for accurate measurements in the linear regime of the spectrometer. We took the measurements using a Jasco J-815 Circular Dichroism Spectropolarimeter with an active temperature control connected to a water bath with a temperature set to 20ºC. We performed the temperature feedback by a Jasco PFD-425S/15 controller with a Peltier control unit. We simultaneously measured the CD, and UV/Vis absorption spectra of the samples together with the photo-multiplier voltage to ensure that the spectrometer is not operating beyond its specified voltage range of 600 units. We took the measurements in the 185-210 nm wavelength range. We used a data pitch of 0.2 nm, a bandwidth of 1 nm, a data integration time of 1 second, and a scanning speed of 20 nm/minute. We averaged each measurement 5 times.  Table S2)

CD measurements of the starting racemic solutions
We obtained the CD spectra of the synthesized DL-RAO and the incubated solution synthesized from DL-glyceraldehyde before the crystallization experiments to make sure that the starting solution is racemic, within the limits of the CD measurement.  In order to calculate the ee of crystals we re-crystallized on magnetite we took CD calibration curves of enantiopure D-RAO. We prepared 5 solutions of D-RAO with varying concentrations at 40 µM, 55 µM, 70 µM, 85 µM, and 100 µM. We obtained the CD and UV-Vis spectra of these solutions. We then picked a wavelength to calibrate our samples. We picked 200 nm as the solvent and cuvette absorption is less at this wavelength compared to deeper UV wavelengths. Moreover, at this wavelength the slopes of the CD and UV-Vis curves are steeper, therefore the calibration is more sensitive at this wavelength compared to a plateau around 190 nm. Moreover, when we fitted the calibration lines (y = mx + c) to points taken at this wavelength, the constant term, c, was closer to 0 than other points. In an ideal scenario the constant term, c, should be 0 because if there is no absorption there should be no CD signal. Before we picked the points for fitting from the CD spectra, we smoothened the spectra using Savitzky-Golay filter with a window length of 101 and polynomial order of 3. Standard error of the slope from the least-squares fit to the data: 0.0002

Enantiomeric excess calibration curves
Standard error of the constant from the least-squares fit to the data: 0.012

Enantiomeric excess calculations
We used the slope (m = -19.187) of the linear fit shown in Fig. S19d to calculate the enantiomeric excess using the following formula: This formula calculates the ratio of the absorbance normalized CD signal of the sample and of enantiopure RAO. Due to the standard deviation of the linear fit 3% error in the ee is expected at minimum. Due to additional errors introduced by smoothening and background absorption we assume an error of about +-5% in the calculated ee.

General method for RAO crystallization with other aminooxazolines
We crystallized RAO with other aminooxazolines to understand the twinning properties of RAO crystal clusters and the effects of habit modifiers on crystallization.
We prepared 2 mL solutions of RAO with other aminooxazolines and crystallized them in a Petri dish for a few hours and investigated the shapes of crystals.

RAO crystallization with XAO and habit modification
We crystallized 50 mM D-RAO with 50 mM D-AAO, D-XAO, and D-LAO separately.
We observed that D-RAO forms needle shaped crystals in the presence of D-XAO and the other crystallizations gave regular diamond shaped crystals.
We then crystallized L-RAO with D-XAO (1:1, 50 mM) and did not observe the needle formation which indicates that the relative stereochemistry matters for the crystal habit modification. We also observed that D-XAO is embedded in the needle shaped crystals by 1 H NMR, as seen in Fig. S21. For this measurement, after the crystallization, we washed the crystals three times with pure water to make sure that We have also observed that the morphology of RAO crystals changes when XAO is in the solution and the two aminooxazolines are in differing proportions. See Fig. S26.
Last, at high concentrations we observed that in the presence of XAO, RAO forms twinned rosette-shaped clusters. We observed these clusters when we crystallized 250 mM of D-RAO in the presence of 62.5 mM D-XAO (4:1). We chose 4:1 as this mixing ratio is close to the one obtained when 2-aminooxazole adds to glyceraldehyde. We also observed that at high concentrations, these crystals form regardless of the ee of the solution. We observed rosette-shaped clusters when DL-RAO is crystallized with DL-XAO (4:1) as well.

10.3
Twinning properties of rosette-shaped RAO crystals DL-RAO crystals form rosette-shaped twinned crystal clusters made from individually enantiopure needle-shaped crystals. We confirmed this by analyzing each needleshaped arm of a cluster by X-ray diffraction. As seen on the right panel of Fig. S24, individual arms of a twinned cluster are enantiopure, yet the arrangement of the D and L crystals are randomly varying. Therefore, the whole cluster, if it consists of a non-stochastic number of needles, is on average racemic.   Similarly, at prebiotic mixing ratios of RAO and XAO (about 4:1) we have not observed the presence of XAO in the crystals unlike the case with a 1:1 mixing ratio (fourth row). This is likely due to the limits of NMR detection. The supernatant (liquid above the rosette-shaped crystals, third row) is depleted in RAO after the crystallization compared to before (first row).

Fig. S26: Crystal habit modification of RAO by XAO a.
RAO crystallization is modified with XAO. In the presence of XAO and at high concentrations, RAO forms rosette-shaped twinned conglomerates. b. Direct crystallization of RAO gives rosette-shaped twinned clusters, and as the concentration is decreased the conglomerate crystals become more angular and the rosette formation is suppressed. At lower concentrations for direct crystallization, the crystal morphology becomes identical to re-crystallized RAO in the presence of XAO. c. XAO modifies the crystal shape of RAO by inhibiting the growth of the side face and thus promoting the growth vertically. This makes the RAO crystals more elongated and eventually needleshaped with increasing XAO abundance. On the top-center, the ratio between D-RAO and D-XAO; on the bottom-right, the sketch of the crystal shape is shown.

Crystallographic Information Files
Crystallographic Information Files (CIF) representing the crystallographic properties of the crystals we obtained are attached as supplementary data.
Data S1. CIF file for D-RAO.