Optical Monitoring of In Situ Iron Loading into Single, Native Ferritin Proteins

Ferritin is a protein that stores and releases iron to prevent diseases associated with iron dysregulation in plants, animals, and bacteria. The conversion between iron-loaded holo-ferritin and empty apo-ferritin is an important process for iron regulation. To date, studies of ferritin have used either ensemble measurements to quantify the characteristics of a large number of proteins or single-molecule approaches to interrogate labeled or modified proteins. Here we demonstrate the first real-time study of the dynamics of iron ion loading and biomineralization within a single, unlabeled ferritin protein. Using optical nanotweezers, we trapped single apo- and holo-ferritins indefinitely, distinguished one from the other, and monitored their structural dynamics in real time. The study presented here deepens the understanding of the iron uptake mechanism of ferritin proteins, which may lead to new therapeutics for iron-related diseases.

S3 centre-to-centre distance of 200 nm. A rectangle (3 nm × 40 nm) is placed in the middle of the two circles to bridge them. To create the DNH structure, we ran the focused ion beam (FIB) at an ion beam energy of 30 kV, and a beam current of 1 pA. The dwelling time for making circles was 1.25 μs, and for the box was 5 μs. These parameters were optimised so that the gap sizes of the DNH structures are mostly around 20 nm with no gold residuals inside the gap. Using scanning electron microscopy (SEM) in top-view and tilted modes, we captured images of the DNH structures ( Figure 1, and Fig. S1).
Optical Tweezer setup: All the optical components were acquired from Thorlabs as described previously. 1 The laser was focused into a spot with a diameter of about 1.2 μm using a 60× Plan Fluor objective with a numerical aperture (NA) of 0.85 (Nikon, Tokyo, Japan). A half-wave plate adjusts the polarisation of the laser to be perpendicular to the line that connects the centres of two holes. 2,3 The power density at the DNH sample was 19 mW/µm 2 due to the 32 mW incident laser power prior to the objective. The intensity of the transmitted light was detected by a silicon avalanche photodiode (APD120A, Thorlabs), which transformed the light intensity into a voltage signal. The voltage signal of the APD was recorded with a data acquisition card (USB-6361, NI) at a sampling rate of 1 MHz using a customised LabVIEW program.

Preparation of Protein and Iron Solutions:
Both apo-ferritin (from equine spleen, A3660) and holo-ferritin (from equine spleen, F4503), as well as other chemicals, were purchased from Sigma-Aldrich, United Kingdom. We used 0.5 μM apo-ferritin or 0.5 μM holo-ferritin in 0.1 M phosphate buffer (PB, pH 7.4) for S4 the trapping experiments. To make a 2 mM ferrous solution, we first made 2 mM Na 2 S 2 O 4 in the PB buffer to deoxygenate the media. 4 Then we added Ammonium iron (II) sulphate to this solution and magnetically stirred the mixed solution for about 10 min. For the PB buffer containing Fe 3+ that was used for the control experiment, we prepared the 2 mM ferrous solution, but this time we bubbled the solution with compressed air for at least five minutes to oxidise Fe 2+ to Fe 3+ . We observed an obvious colour change from green to yellow after bubbling. 5 All the solutions were prepared freshly and were filtered through a 0.22 µm pore size filter before every experiment.
Fluidic system: The flow cells used in this work are the same as previously reported. 1 We print the flow cells by using the FormLab 2 printer with Clear V4 resin at a resolution of 50 µm (Formlabs Inc., USA). A two-component silicone-glue (Twinsil, Picodent, Germany) was used to seal the samples in the flow cell with a cover glass with a thickness of 0.17 mm. The DNH sample and the cover slide were separated by a double-sided tape with a thickness of 50 µm (ARcare92712, Adhesive Research, Inc.), creating a fluidic channel with a volume of 3.5 µL. Through a 12-port valve (Mux Distrib, Elve Flow, France), a syringe pump (Harvard Apparatus, US) controlled the flow rate and flow direction. To get the buffer exchanged after trapping a protein, we infused the buffer into the flow chamber at a flow rate of 4.5 µL/min. According to the internal diameter of the tubing used in the flow controller system, the ferrous solution arrives at the flow chamber after 25 µL (6 min after injection) of the solution passing through the flow controller.

S5
Data Analysis: We used MATLAB scripts to analyse all the data presented in this work. All raw data were filtered using a zero-phase Gaussian low-pass filter to the desired cut-off frequency (1 kHz or 5 kHz) by using the filtfilt.m function. We calculated the probability density functions (PDF) by using the ksdensity.m function. To compare two trapping signals, we aligned the signal by subtracting the trace from its median value (Fig. 2c). All the normalised RMS ( Figure 3) were calculated by dividing the standard deviation of 1-s trace by its mean value.

SI-2 Characteristics of DNH structures and their influence on protein trapping
The DNH structure loses its effectiveness in capturing proteins after repeated uses, i.e. longer waiting time and shorter duration of trapping. 6 We attribute it to the rounded cusps and edges of DNH after use, which can be affected by the buffer and the laser power used in the experiments. Figure S1a provides the SEM images of six DNH structures used in this work, imaged before (only #2, #3 and #6 available) and after trapping experiments. In addition to the contaminations on the surface, the edges and cusps became smooth, and the gap sizes increased after being used for trapping. The gap sizes of #2, #3, and #6 changed from 15, 17, and 25 nm to 28, 34, and 37 nm, respectively (black circles in Fig. S1b).
The boxplot in Fig. S1b takes account of the gap sizes measured from all DNHs, including the structures that are not used in this work but were fabricated with the same parameters (SEM images in Fig. S1c). We S6 observed an increase in medium gap size from 24.4 nm to 35.6 nm after DNHs were used for two to three weeks, with a typical duration of 4-5 hours of laser illumination. In order to understand the influence of gap size on the trapping efficiency, we simulated the optical field distribution of the DNH structures by using COMSOL Multiphysics 6.1 based on finite-element simulation.
We consider an 852 nm laser with the polarisation perpendicular to the line that connects two holes. Figure   S2a-c demonstrate the electric field distribution of three structures with gap sizes of 15 nm, 17 nm and 25 nm, representing #2, #3 and #6 respectively. As indicated by the scale of colour bars, the field enhancement decreases with increasing the gap sizes. Considering the dimension of the Ferritin protein (12 nm), we aim for DNH with a gap size of 20 nm to ensure the narrow trapping well at the same time allowing the protein to enter the hotspot. When the polarisation of the laser beam is parallel to the connecting line of two holes, as shown in Fig. S2d, little to no field enhancement was observed in the gap of DNH. This dependency on laser polarisation allows differentiating the DNH from other structures, as recently demonstrated. 7 We note that the plasmonic optical trapping demonstrated here is based on the self-induced back-action, which operates well at the off-resonance wavelength. As shown in the transmission spectra in Figure S2e, the wavelength of the excitation laser at 852 nm is located at the off-peak of the localised surface plasmonic S8 resonance (LSPR) ( at around 700 nm). Furthermore, we expect another resonance located in near-infrared wavelength as reported by Ghorbanzadeh et al. 3 Despite this off-resonance position, the electric field is still significantly enhanced in the gap of DNH. This off-resonance detection demonstrated the robustness of the SIBA trapping -without strict limitation on the excitation wavelength, the DNH structure can provide field enhancement that is strongly confined within the hotspot to trap single proteins.

SI-3 Trapping individual apo and holo-ferritins by other DNH structures
Due to the grains in the gold film and the variation of the FIB situation, different DNHs might have different features. For comparison between the dynamic of apo-ferritin and holo-ferritin, we trapped individual proteins in six different DNH structures and produced the RMS, and probability density function (PDF) shown in Figure 3. Figure S3 shows the trapping events of both apo-ferritin and holo-ferritin in five other structures (#1, #2, #4, #5, #6, and results of #3 are presented in Figure 2). S11 Figure S3.

SI-5 Change in the transmission signal upon trapping single proteins
Table S1 summarises the changes in the transmission signal of six DNH samples upon trapping apoferritin and holo-ferritin. Here, V 0 is the mean value of the APD signal for the empty DNH, V 1 is the mean value of the APD signal for the DNH with protein trapped. The changes in the transmission signal (ΔT/T 0 ) were calculated by (V 1 -V 0 )/V 0 . Four out of six structures exhibit a higher ΔT/T 0 value for trapping holoferritins than that for apo-ferritins.  Figure S6 illustrates two consecutive trapping events of single apo-ferritin by using the same DNH structure (#2, other traces from this structure are shown in Fig. S3b). Both the RMS noise and ΔT/T 0 demonstrate the reproducibility and consistency of the dynamic characteristics of identical proteins by using the same structure.

S15
In this work, we observed mostly an increase in transmission upon trapping a single protein. Occasionally, trapping a protein induces a reduced transmission of DNH likely due to the protein being trapped in different trapping well (i.e. gold-water interface and gold-SiN x interface). Figure S7 gives two "down-trapping" events of apo-and holo-ferritin by using the same structure (#4). The dynamic information carried by such "down-trapping" is consistent with the "up-trapping" events. 1 We observed a larger change in transmission signal (ΔT/T 0 ) and a smaller RMS for trapping a holo-ferritin, compared to trapping an apo-ferritin.

SI-7 Waiting time for DNHs to trap a single protein
We summarised the time duration between tuning on the laser and trapping of a ferritin protein. Figure S8 plots the waiting time of 25 trapping events from different DNHs. The histogram indicates that ferritin is most likely to be trapped within 10-15 min after the laser turns on. We note that the duration required to trap a single protein is impacted by a variety of factors, including the features of the (DNH) structure, the size of the protein being trapped, and the laser power used in trapping. 3 The surface repulsion between the negatively charged protein and the negatively charged gold surface can also lead to a longer duration needed for trapping a single ferritin. 6,8 Figure S8. Histogram of the waiting time for trapping a ferritin protein after turning on the laser.

S17
For identifying the low RMS segment in Figure 4, we created a MATLAB program to identify the step based on the changes in standard deviation (std). The red lines in Figure S9 represent stable levels with similar std for Figure 4b-4e of the main text. The threshold for step searching is 0.5 × max(std). We considered the levels that last more than 500 ms as the validated lower RMS segments. This algorithm identified and marked the lower RMS segments in purple segments (Fig. S9). We then manually connected some segments that are very close to each other. and after 20 minutes of exposure to Fe 2+ solution (red). The asterisks indicate that these two data sets are significantly different with a p-value of 0.0002.

SI-11 Temperature in the trapping site
Laser illumination on the gold nanostructures can induce a heating effect due to the photon absorption and subsequent energy dissipation. 9 We calculated the temperature in the trapping site by finite-element simulation in COMSOL Multiphysics. 10 Figure S12 shows the temperature profile of a double nanohole at the laser power of 32mW used in this work. Considering the best scenario, i.e., all the absorbed laser power is converted to heat, the temperature in the DNH structure is estimated to be 49.8 °C.

SI-12 Exposing the single apo-ferritin to Fe 3+ (control experiment for iron loading)
As the iron binding sites in ferritin bind only to ferrous (Fe 2+ ) ions 11 , we performed a control experiment to expose apo-ferritin to ferric iron (Fe 3+ ) solutions. Figure S13a shows the transmission signal of the DNH with apo-ferritin trapped in the PB solution. We then replaced the solution in the chamber with a 2 mM ferric solution while the protein remained trapped. Figure S13b-S13d shows the optical trace after the apoferritin was exposed to the ferric solution for 8 min, 15 min and 20 min. No "on-off" patterns were observed. Figure S14 compares the RMS of the optical trace obtained before apo-ferritin was exposed to the ferric solution (trace shown in Fig. S13b) and obtained after the apo-ferritin was exposed to the ferric solution for 20 min (trace shown in Fig. S13e). No significant difference in the RMS between the two traces suggests S22 that apo-ferritin retained its relatively flexible conformation hence no iron loading happened. 12 Figures S13 and S14 again confirm that the "on-off" patterns observed in Figure 4 of the main text are due to the iron loading activity of ferritin.
S23 Figure S13. In-situ ferric iron loading into a trapped apo-ferritin, (a) Continuous transmission trace of a single apo-ferritin trapped in the hotspot of a DNH, then exposed to the ferrous solution for more than 20 min. After turning off the laser for 5 seconds, the transmission signal returns to baseline, indicating that protein was released. (b) 20-second transmission trace of trapped apo-ferritin before ferrous solution reaches the hotspot, (c and d) 20-second transmission traces after apo-ferritin was exposed to the ferric solution. (e) 20-second transmission trace after apo-ferritin was exposed to the ferric solution for more than 20 minutes. Figure S14. Root mean square (RMS) of 20 s transmission traces when a single apo-ferritin is trapped in PB solution and after it is exposed to Fe 3+ solution for more than 20 min. There is no significant difference (p = 0.119) between these RMS results from the two traces.