AT-specific DNA visualization revisits the directionality of bacteriophage λ DNA ejection

Abstract In this study, we specifically visualized DNA molecules at their AT base pairs after in vitro phage ejection. Our AT-specific visualization revealed that either end of the DNA molecule could be ejected first with a nearly 50% probability. This observation challenges the generally accepted theory of Last In First Out (LIFO), which states that the end of the phage λ DNA that enters the capsid last during phage packaging is the first to be ejected, and that both ends of the DNA are unable to move within the extremely condensed phage capsid. To support our observations, we conducted computer simulations that revealed that both ends of the DNA molecule are randomized, resulting in the observed near 50% probability. Additionally, we found that the length of the ejected DNA by LIFO was consistently longer than that by First In First Out (FIFO) during in vitro phage ejection. Our simulations attributed this difference in length to the stiffness difference of the remaining DNA within the phage capsid. In conclusion, this study demonstrates that a DNA molecule within an extremely dense phage capsid exhibits a degree of mobility, allowing it to switch ends during ejection.


INTRODUCTION
Bacteriophage has been used as a r epr esentati v e model system for the study of molecular biology and biophysics since Lederberg's first discovery ( 1 ). A phage is composed of a 63 nm icosahedral capsid head ( 2 ), and a 150 nm non-contractile tail ( 3 ). The diameter of the nanotube in the tubular tail is 4 nm, which is only twice the diameter of double-stranded DNA ( 4 ). The tiny capsid contains a linear DNA molecule comprising 48502 base pairs ( 5 ). This virus packs DN A unidirectionall y from the left-hand end to the right-hand end of the genome (6)(7)(8). A terminase, an ATPconsuming motor protein, binds the left ends of the viral DN A concatemer in Esc heric hia coli host cells and docks an empty viral procapsid to insert the DNA through ATP hydrolysis until it cleaves the next cos site ( 9 ). Such DNA packaging r equir es an extr emely high for ce (51 pN) to compr ess the long genomic DNA molecule into a nanometer space ( 10 , 11 ). Tight packaging leads to strong repulsi v e interactions due to the negati v e charges on the DNA backbone, and bending stress due to the persistence length (50 nm), resulting in internal capsid pr essur es r eaching 25 atm ( 12 ). This accumulated pr essur e in these tiny capsids is a fundamental force inducing viruses to eject the DNA genomes into the host cells during infection (13)(14)(15)(16).
When considering unidirectional packaging, a question arises which end ejects first? To date, most re vie w papers have stated that the last DNA end packaged would be the first end out, 'Last In First Out (LIFO)' ( 17 ). Three independent experimental studies supported the LIFO model (18)(19)(20). All of them used electron microscopic images of tail pr oteins cr oss-linked with the last-packaged DNA end (the right end of the genome). They determined the rightleft directionality by DNA melting maps since the right end of the DNA had an AT-rich region that could melt easily with increased temperature. Most bacteriophages are known to eject from the last packaged end (LIFO), except bacteriophage T4, which follows 'First In First Out (FIFO)' with the aid of a protein ( 18 , 21-25 ). LIFO seems reasonable because the capsid is extremely cro w ded with a near-crystalline density, which was determined by Xr ay diffr action and cryo-electr on micr oscopy. For example, Earnshaw and Harrison analyzed diffraction patterns from phage viral capsids to show 2.3-2.6 nm DNA-DNA interstrand spacings ( 26 ). This spacing matches the 2.4 nm DN A-DN A interstrand distance of the B-form DNA crystal reported by Franklin and Gosling ( 27 ). Using cryoelectr on micr oscopy, Lepault et al. visualized a DNA nematic liquid crystal structure in a capsid ( 28 ). Cerriteli et al. reconstructed a cryo-EM-based three-dimensional spool structure to explain the 2.5 nm spacing resolved by the Xr ay diffr action pattern ( 29 ). Ther efor e, it is expected that enca psulated DN A in a mature viral ca psid may be completel y tra pped in a glassy state, resulting in packed ultraslow mobility ( 30 ). Accordingly, LIFO is considered a standard model for phage ejection.
Howe v er, there hav e been contradictory observations reported in regard to this matter. Before three EM observations of a cross-linked tail protein and the right end of DNA wer e r eported (18)(19)(20), Sharp et al. published a paper titled 'Lack of polarity of DNA injection by Escherichia coli phage ' in 1971 ( 31 ). They used X-rays to break the DNA in the capsid and then infected E. coli host cells with the Xr ay-irr adiated phages. Their analysis of the ejected fragments containing phage DNA ends showed that half of the phages ejected their genomic DNA from the left end and the other half ejected their DNA from the right end. This result may be explained if the first entered left end can move relati v el y freel y within the capsid.
Notably, Liu et al. reported a fluid-like disorder ed cor e in the center of a crystal-like spooled configuration of stacked DN A ( 32 ). The DN A in the ca psid had a toroidal spool structure with near crystalline density, but it was ordered in concentric layers in the periphery of the capsid. A stiff DN A pol ymer with a persistence length of 50 nm should hav e a relati v ely low density in the center of the DNA spool in a 63 nm capsid. Evilevitch and his colleagues reported a series of papers regarding the solid-to-fluid transition in the core of a capsid that contained a cholesteric packing pattern (32)(33)(34)(35). Their characterization of the fluidic structure of the capsid core demonstrated different mobility regions in the phage capsid. A part of the DNA should be packed in the liquid crystal spool, but the other part of the DNA may mov e relati v el y freel y within the fluidic core. Evilevitch suggested a fluid-like DNA structure in the capsid core above 33 • C ( 35 ). Such a phase transition leads to higher disorder and a significant increase in genome fluidity. From the cryo-EM images, one can estimate the ratio of the disordered core to the phage capsid radius ( d 0 / R c ). For example, Lander et al. showed a cutaway view of the capsid, in which d 0 = 19 nm and R c = 32 nm ( 36 ). De Frutos et al. reported that the hexagonal crystal thickness was 11.1 nm for phage ( 37 ). Walker et al. reported that d 0 / R c was larger than 0.5 for most bacteriophages ( 38 ). Another notable observation was that both ends are located close to each other. For example, Haas et al. observed cross-linked DNA fragments between the left ends and the right ends, which implied that both ends were present in the near vicinity ( 39 ). Hohn et al. suggested a model in which both ends could be located in the vicinity of the portal v erte x after DNA packaging ( 40 ). Recently, Liu et al. reported a cryo-EM structure for the herpesvirus, in which both ends were located near the portal v erte x, although it was not a phage virus ( 41 ). If the left and right ends are closely located in the fluidic disordered core near the portal v erte x, either end can eject first because the left end of DNA may not be packed in the crystal. Accordingly, it remains unclear whether phage is ejected only from the right end.
In this paper, we visualized the directions of phage ejection using AT-specific DNA staining methods that we recently de v eloped ( 42 , 43 ). For in vitro phage ejection, we used three induction methods: these methods included the phage receptor protein of LamB ( 44 ), a cross-linking reagent of glutaraldehyde ( 45 ), and a high temperature of 65 • C ( 46 ). In a microfluidic flow cell, the ejected DNA molecules were immobilized on the surface by an airflow drying method ( 47 ) and AT-specific staining. Contrary to the status quo, we found that both ends could eject first, consistent with the result of Sharp et al. ( 31 ). Interestingly, the ejected DNA lengths were not the same: the ejected DNA lengths of LIFO were consistently longer than those of FIFO. To explain our experimental observations, we performed computer simulations to investigate the process involved in the ejection of DNA with a compact structure. Mor eover, our simulations r eproduced r esults consistent with the experimental observation that the ejection lengths of LIFO were longer than those of FIFO.

Bacteriophage pr epar ation
Bacteriophage ( BP5203) was obtained from ATCC. The phage pr eparation procedur e was based on a pr evious study ( 48 ). Briefly, E. coli K-12 MG1655 was grown in 5 ml of LB solution to reach an optical density of 0.6 at 600 nm (OD 600 ). Then 1 l bacteriophage solution was mixed with 200 l LB solution containing E. coli cells. This solution was mixed with LMP agarose containing TNT media (3.0 g tryptone, 1.5 g NaCl, 0.3 mg thiamine HCl, 1.5 g LMP agarose in 300 ml H 2 O). This medium was autoclaved and cooled to 37 • C before use. The mixture solution was poured onto an LB agar plate (2.5 g LB broth and 1.5 g agar in 100 ml H 2 O poured into Petri dishes) and placed into a 4 • C refrigerator for 30 min. After gelling the mixture solution in the agar plates, the plates were moved into a 37 • C incubator overnight. Then, 5 ml of LB solution was poured onto the agar plates and placed at room temperature for 1 h. Next, the solution was transferred to 50 ml conical tubes and centrifuged at 4500 rpm for 20 min. Finally, the supernatants from centrifugation were filtered using 200 nm pore syringe filters to obtain phage stock solutions.

LamB pr epar ation
LamB (maltoporin) was pr epar ed as described in Supplementary Information ( SI). Supplementary Figure S1 illustrates the LamB purification procedure ( 49 , 50 ). The Lamb gene (accession number: NP 418460.1) was isolated from the E. coli UT5600, cloned and expressed. The resulting LamB protein was purified and stored at −20 • C. A more detailed description of the cloning, expression, and purification methods can be found in the Supplementary informa tion Ma terials and Methods section.

ATTO647N-␤ 2 -p y 4 -␤-p y 4 -dp (AP8) synthesis
A computer-assisted Fmoc solid-phase synthesis was performed to produce H 2 N-P8 (1.3 mg, 1.0 × 10 −3 mmol), as previously described ( 42 ). ATTO647N was then added to the N-terminus using ATTO647N NHS ester. After the reaction was completed, the resulting product was purified by high-performance liquid chromato gra phy (HPLC), and the collected fractions were lyophilized to obtain AP8 as a blue po w der. A more detailed description of the synthesis and characterization of AP8 can be found in the Supplementary Informa tion (SI) Ma terials and Methods section, along with SI Supplementary Figures S2 ∼S4.

H-NS-mScarlet pr epar ation
Histone-like nucleoid-structural protein (H-NS) fused with mScarlet was pr epar ed as pr eviously described for H-NS-mCherry ( 43 ). To enhance the brightness, mCherry ( φ = 15.84 mM −1 cm −1 ) was replaced with mScarlet ( φ = 70.0 mM −1 cm −1 ) ( 51 ). The plasmid pET-15b was used to construct H-NS-mScarlet by adding the PCR-amplified mScarlet fragment to the C-terminus of H-NS. A more detailed description of the construction, expression, purification, and characterization methods can be found in the Supplementary Informa tion (SI) Ma terials and Methods section, along with Supplementary Figures S5 and S6.

Microscopy
The microscope system consisted of an inverted microscope (Olympus IX70, Tok yo , Japan) equipped with a 100 × Ol ympus UPlanSA po oil immersion objecti v e lens and illuminated LED light source (SOLA SM II light engine, Lumencor, Beaverton, OR). The light was passed through the corresponding filter sets (Semrock, Rochester, NY) to excite the fluorescent dye. Fluorescence images were captured using a scientific-grade complementary metaloxide-semiconductor digital camera (2048 × 2048, Prime sCMOS Camera, Photometrics, Tucson, AZ) and stored in 16-bit TIFF format as generated by the Micro-manager softw are. ImageJ w as utilized for image processing and measuring the lengths of ejected DNA molecules.

The PEG-biotin-coated glass surface pr epar ation
The PEG-biotin-surface was pr epar ed as described in our previous study ( 52 ). Briefly, glass coverslips were stacked in a Teflon rack and soaked in a piranha etching solution (30:70 v / v H 2 O 2 / H 2 SO 4 ) for 2 h. Then, the glass coverslips were rinsed with deionized water until the pH was neutral (pH 7), followed by sonication for 30 min. The cleaned coverslips were placed in a polypropylene container into which 200 ml of methyl alcohol was added. Then, 2 ml of N -[3-(trimethoxysilyl)propyl]ethylenediamine and 10 ml of glacial acetic acid were added. This container was shaken a t room tempera ture for 30 min, sonica ted a t 75 W for 15 min, and then shaken overnight at 100 rpm at room temperatur e. Then, the coverslips wer e washed once with methyl alcohol and twice with ethyl alcohol. For biotinylation, 50 l of PEG-biotin solution was dropped onto a glass coverslip, covered with PEG-biotin for 3 h, and then rinsed with water. The PEG-biotin solution was pr epar ed as 350 l containing 0.1 M sodium bicarbonate with 80 mg of mPEGsuccinimid yl valera te and 2 mg of biotin-PEG-succinimidyl carbonate. Figure 1 A shows the flow cell set up on a fluorescence microscope. The flow chamber was fabricated by placing an acrylic support on a PEG-biotin-coated glass coverslip with a height of 100 m, which was formed by a double-sided adhesi v e sheet as previously described ( 52 ). The flow chamber dimensions were a pproximatel y 5 × 10 × 0.1 mm ( L × W × H ). The channels were made with laser cutting. A yellow pipette tip was attached to the inlet port, and a tubing (inner diameter: 0.02 in.) was epoxidized to holes on the outlet port of an acrylic holder. The outlet port was connected to a NE-1000 syringe pump (New Era Pump Systems Inc., Wantagh, NY) with a flow rate of 100 l / min. Before each experiment, the PEG-biotin-coated glass coverslip was coated with neutravidin. Next, 20 g / ml of neutravidin in T50 buffer (10 mM Tris, 50 mM NaCl, pH 8.0) was added into a chamber, followed by incubation at room temperature for 5 min. released DNA from the NEB DNase I buffer (total volume = 5 l). This phage solution was loaded into a flow cell and then incubated for 10 min at room temperature to facilitate phage particles immobilization on the PEGbiotin-neutravidin-coated surface. To remove DNase I, the flow cell was washed with 300 l of 1 × TE buffer with a pH of 8.0 (10 mM Tris, 1 mM EDTA). For phage ejection, phage receptor protein, 25 g / ml of LamB diluted with NTE buffer (150 mM NaCl, 40 mM Tris-HCl at pH 8.0, 1 mM Na 2 -EDTA, 1% n -octyl-␤-maltopyranoside (OM), 1 mM DTT, and 30% glycerol), was loaded into the chamber. After 5 min of incubation, individual DNA was ejected. In the second method, phage particles were incubated with glutaraldehyde for 15 min to eject DNA from the capsid. The loaded glutaraldehyde concentration was 2.5% in 0.2 M sodium phosphate buffer (pH 7.25). In the third method, a high temperature was used to eject DNA from the capsid ( 35 ). Phage particles were incubated for 5 min a t 65 • C . For microscopic visualiza tion, H-NS-mScarlet or AP8 in 1 × TE buffer was added to stain the ejected DNA molecules. For sequence-specific visualization, the ejected and immobilized DNA should be incubated with H-NS-mScarlet or AP8 for at least 15 min. Alternati v ely, Supplementary Movie S1 shows the real-time observation of phage ejection stained by Y OY O-1. This is based on the fact that Y OY O-1 staining is fast, and Y OY O-1 is small enough to penetrate the phage capsid. More information can be found in the Supplementary Information (SI) Materials and Methods section, along with Supplementary Movie S1.

Langevin dynamics simulation
For the DN A sim ulations, we employed a generic coarsegrained model of a semifle xib le chain composed of 200 monomers of diameter and mass m. We performed Langevin d ynamics simula tions with a velocity-Verlet integrator and an integration time step of 0.005 τ , where τ is the reduced time unit, τ ≡ mσ 2 /k B T . The chemical bond between two neighboring monomers was described by the finite-e xtensib le nonlinear elastic potential (U b ), where K = 30 k B T / 2 and R 0 = 1.5 as in the standard polymer model of Kremer and Grest. = k B T is the unit of energy, where k B and T denote the Boltzmann constant and temperatur e, r especti v ely. The nonbonding interactions between monomers were described by the repulsi v e W eeks-Chandler-Andersen (WCA) potential. W e also imposed a bending potential ( U a ) between two consecuti v e chemical bonds, i.e. U a = K a (1 + cos ), where is the angle between two neighboring bonds. We set = 5 nm and K a = 10 k B T such that the persistence length ( l p ) of our polymer chain was comparable to l p = 10 = 50 nm of a double-stranded DNA. To investigate the effects of the rigidity differences (persistence length) between the first and the second halves of the DNA, we assigned the different values of K a to the first and the second halves, i.e. K F and K L , respecti v ely. We define the relati v e rigidity as r = ( K F − K L ) /K L and changes the value of r from 0 to 0.35 by setting ( K F , K L ) = (10, 10), (10.5, 9.5), (11,9) and (11.5, 8.5). Note that the average values of K F and K L were equal to 10 k B T in all cases. The spherical capsid was composed of 784 particles. We considered capsids of two sizes such that the packing fraction of the capsid (when the DNA is fully packaged) was either 0.4 or 0.6. When the packing fraction was as high as 0.6, the DNA was packed inside the capsid such that the conforma tional relaxa tion was quite slow. We attached a cubic box of 1.6 × 1.6 × 10 as a portal to the capsid. We pr epar ed 2000 initial configurations for each condition by placing the first 10 monomers of the DNA inside the portal and adding 190 monomers sequentially at random positions outside the portal. We followed three steps to mimic the ejection process: (i) packaging, (ii) resting and (iii) ejection. In the packaging step, we applied an external packaging force ( f p ) to the monomers inside the portal and package monomers into the capsid. The packaging force generated by the motor protein of phage was a pproximatel y 85 pN ( 53 ), which corresponds to approximately 8 k B T / in this simulation model. In this work, we employed a packaging force of 10 k B T / , which was slightly larger than that of phages. Then, we adopted a stalling step during the packaging process to mimic the packaging process in the viral capsid. Every time 50 monomers w ere packaged, w e turned off the packaging for ce, fix ed 10 monomers inside the portal at their positions and allowed the other parts of the DNA to relax their conformation during τ stall = 500 τ . After the packaging was completed with all 200 monomers inside the capsid, we blocked the entrance of the portal. Then, we initiated the resting step where the DNA chain inside the capsid was allowed to undergo conforma tional relaxa tion during the resting time ( τ rest ). We changed the value of τ rest from 0 to 1500 τ to investiga te the ef fects of conforma tional relaxa tion on the directionality of DNA ejection. Once the resting step was completed, we eliminated the blockage, introduced a capsid portal, and let the DN A spontaneousl y eject from the capsid.
In the ejection step, we introduced the resisting force ( f r = 0 . 35 k B T / σ ) at the end of the capsid portal to mimic the effect of the osmotic pr essur e outside the capsid. In the in vitro experiments , the whole part of the DNA did not eject, and a fraction of monomers remained inside the capsid due to the r esisting for ce induced by the osmotic pr essur e. When we applied a r esisting for ce of f r = 0 . 35 k B T /σ , 30% of the DNA did not eject from the capsid e v en after a sufficiently long time in the sim ulations, w hich is comparable to that in experiments. We checked the index ( i exit ) of the monomer ejected from the capsid for the first time. The probability of FIFO was estimated out of 2000 independent trajectories. Figure 1 illustrates the design of a single-molecule experiment using a fluorescence microscope to determine the orientation of DNA ejected from a phage capsid. The flow cell setup shown in Figure 1 A was the primary platform for performing multistep reactions by loading reagents through simply changing the pipette tips ( 52 ). Before loading phage particles into the flow cell, DNase I was used to remove prematur ely r eleased DNA molecules without induction. The loaded capsid was bound to the surface (Figure 1 B (i)) coated with PEG-biotin-neutravidin. Three different methods were used for in vitro phage ejection: these methods included LamB ( 44 ), glutaraldehyde ( 45 ) and 65 • C heating ( 46 ) (Figure 1 B (ii)). All three induction methods were successful for phage ejection. First, we used LamB since most in vitro phage ejection experiments have used this protein ( 57 ). LamB is a maltose outer-membrane porin and acts as a bacteriophage receptor in host E. coli cell membranes ( 58 ). Thus, we constructed an expression vector to obtain the LamB protein as described in the methods section (SI Methods and Supplementary Figure S1). Second, we used the fixation chemical glutaraldehyde because the fixation-induced phage ejection method has been used in many studies, such as those regarding T7 ( 45 ), T4 ( 59 ) and SPP1 ( 23 ). Although the detailed mechanism is still unknown, Kellermayer et al. suggested that the process ma y in volv e the le v er-like action of the tail fibers to cause a transition within the tail complex ( 60 ). The SI Movie S1 illustrates the real-time phage ejection process using glutaraldehyde induction and Y OY O-1 staining. Third, we used 65 • C heating, which was based on previous studies that showed that DNA could be ejected at a temperature between 50 • C and 75 • C ( 46 ). When optimizing the experimental conditions, the heat-induced ejection method was mainly used because it was simple, convenient, and compatible with subsequent procedures in terms of cleanliness. The DNA molecules ejected from phage capsids were fully elongated by a syringe pump-dri v en flow (100 l / min). Then, the DNA molecules were immobilized on the surface by airflow ( 47 ). Finally, the DNA molecules were stained with AT-specific ATTO647N-labeled octameric pyrroles (AP8) or AT-specific DNA binding fluorescent protein (H-NS-mScar let). The detailed char acteriza tion informa tion for the AP8 and H-NS-mScarlet can be found in the SI Supplementary Figures S2-S6. These figures provide data such as the synthesis scheme and molecular weight determined by mass spectrum (Supplementary Figure S2), binding affinity against AT0 (GC-only), AT4, AT7 and AT10 DNA oligomers determined by fluorescence resonance energy transfer (FRET) for AP8 (Supplementary Figure S4), and SDS-PAGE and gel filtration chromatography for H-NS-mScarlet (Supplementary Figure S5), as well as its binding affinity determined by FRET (Supplementary Figure S6). In a previous study, we reported the de v elopment of tetramethyl rhodamine (TAMRA)-labeled octameric pyrroles (TP8) for AT-specific DNA staining ( 42 ).

RESULTS AND DISCUSSION
Howe v er, since TAMRA has low brightness ( φ = 18.1 mM −1 cm −1 ), in this study we substituted it with the much brighter ATTO647N ( φ = 97.5 mM −1 cm −1 ) ( 61 , 62 ). The SI Supplementary Figure S3 Figure 2 demonstrates how to identify FIFO and LIFO occurrences based on DNA images. The genome contains a GC-rich (A T-poor: A T content less than 50%) region of up to 20 kb and an A T-rich (A T content more than 50%) region of 20-48.5 kb with some variations (Figure 2 A). The AP8-stained red DNA images matched well to re v eal detailed intensity profiles for AT abundance. The subsequent H-NS-mScarlet-stained pink DNA images were also good enough to determine the directionality of the DNA molecules, although they were noisier than the AP8stained DN A images, likel y due to the m ultimeric aggregations of H-NS-mScarlet. These four DNA images were produced by a pr estaining procedur e in which DNA was incubated with AP8 or H-NS-mScarlet in a test tube for 15 min and then loaded into the flow cell. Howe v er, to a ppl y the AT-specific staining method in the in vitro phage ejection experiments, it was necessary to consider poststaining because the phage should eject its DNA into the flow cell before staining. Ther efor e, as a control, we acquired the next four DNA images using a poststaining procedure. First, biotinend labeled DNA molecules were loaded and tethered to a neutravidin-coated surface. Second, DNA was immobilized on the surface with airflow. Finally, the staining reagent solution was loaded and incubated for at least 15 min. For the poststaining procedure, we had to reduce the AP8 concentration from 8 M to 27 nM, because 8 M was too high for poststaining in a flow cell, as it generated strong background noise. The DNA image quality obtained from the poststaining procedure was not as good as that obtained from the prestaining procedure. Howe v er, the orientation of the DN A was reco gnizable by the thin line demonstrated by the AT-poor region up to 20 kb and the thick line demonstrated by the AT-rich region after 20 kb.
Using this AT-specific staining, we performed singlemolecule DNA observations to determine the DNA ejection directionality, i.e. FIFO versus LIFO, during phage ejection. It would be ideal to capture a movie of phage ejection using AT-specific staining rea gents, b ut it is not possible due to the mismatch between the time r equir ed for phage ejection (a few seconds, as shown in SI Movie S1) and the incubation time for AT-specific staining (at least 15 min), which does not allow for the creation of a movie. Figure 2 B shows examples of FIFO and LIFO induced by the three ejection methods. In each case, the first two images were of DNA stained with AP8, and the last two images were of DNA stained with H-NS-mScarlet. We calculated Pearson cross-correlation coefficients to determine the direction of DNA. The positi v e and negati v e coefficients clearly determine FIFO or LIFO. We performed three sets of experiments for each induction method, and each set had 50 ejected DNA images. Figure 2 C shows the percentage of FIFO (LIFO = 100 − FIFO). The results of both the FIFO and LIFO percentages were a pproximatel y 50% for all data sets. Specifically, the FIFO percentage was 53.9 ± 7.0% and the LIFO percentage was 46.1 ± 7.0%.
To validate our single-molecule observations using an alternati v e method, we designed an experiment as shown in SI Supplementary Figure S7. Three methods, such as LamB, glutaraldehyde, and 65 • C, were used to induce DNA ejection from the capsid, followed by digestion with DNase to degrade any ejected DNA. After inactivating the DNase, we performed qPCR using fiv e primers targeting fragments at 0.5, 12.1 24.1, 36.2 and 47.7 kb. The PCR process involves heating the sample to 95 • C, which disrupts the phage capsid and releases the DNA remaining in the capsid for amplification. The resulting cycle threshold (Ct) values in SI Supplementary Figure S7 are the same in both ends and high at 24.1 kb, meaning that the 24.1 kb fragment had a much lower abundance compared to the others. Specifically, for the case of LamB, the Ct values were 13.33 (0.5 kb), 12.62 (12.1 kb), 28.92 (24.1 kb), 13.74 (36.2 kb) and 13.22 (47.7 kb) respecti v ely. To estab lish a control, we conducted the same qPCR experiment using NEB lambda (500 pg in 20 l). The resulting CT values were 11.3 (0.5 kb), 10.94 (12.1 kb), 10.9 (24.1 kb), 11.89 (36.2 kb) and 11.43 (47.7 kb), respecti v el y, w hich were e v en f or all positions. Theref ore, we can interpret these qPCR results as both ends of DNA remained after ejection at the same ratio, but the middle (24.1 kb) was mostly in the ejected DNA and degraded by DNase. These results support our single-molecule observations for both the FIFO and LIFO.
It is curious how 53.9% of the first-packaged left ends could be ejected first by passing through a DNA structure with near crystalline density. To understand such phage ejection at the molecular le v el, we performed Lange vin dynamics simulations ( 64 , 65 ). Our simulations suggested several factors that could determine the DNA ejection directionality: (i) how easily the DNA r elax es its conformation inside the capsid after packaging, (ii) whether the two ends of the DNA are randomized or not after conformational relaxation and (iii) how rigid the segment (to which each end of the DNA belongs) would be. Our simulation scheme consists of three subsequent steps: packaging, resting, and ejection (Figure 3 A). The DNA is allowed to relax its conformation during a designated resting time ( τ rest ) in the resting step. As shown in Figure 3 B, the probability of FIFO (P FIFO ) increases to 50% with an increase in τ rest depending on the packing fraction ( φ) of the ca psid w hen full y packaged. When the DNA is gi v en a sufficiently long resting time of τ rest > 1000 τ for conformational relaxation, P FIFO reaches a pproximatel y 50%. In the case of short resting times, the last-packaged end is more likely to be near the capsid portal than the first-packaged end. On the other hand, with a sufficiently long resting time, the positions of the two ends of the DNA are randomized such that the two ends are gi v en equal chances to be located near the capsid portal. We observed a similar trend for a smaller packing fraction of φ = 0.4: P FIFO increases and converges to 0.5 as τ rest increases. For a smaller packing fraction, P FIFO reaches 0.5 e v en for a shorter resting time because the DNA can relax its conformation more readily in a less densely packed capsid.
We also investigated the effects of a difference in the rigidity of the DNA segments. As shown in Figure 2 A, DNA has a GC-rich region on the left and an AT-rich region on the right of the genome. As expected, the GC-rich region is more rigid, and the AT-rich region is more fle xib le due to the number of hydrogen bonds. Chuang et al. reported a 14 nm difference in the persistence lengths depending on GC contents ranging from 30% to 60% ( 66 ). In their study, they developed statistical terpolymer models to predict persistence length ( l p ) of DNA as follows, l p = ( 23 + 23 γ + 26 γ 2 ) + c. Her e, γ and c r epr esent the fraction of GC content and a constant, respecti v ely. The difference in the persistence length between GC-rich and AT-rich regions is, then, l p = l p,GC − l p,AT = 23( γ GC − γ AT ) + 26 ( γ GC − γ AT ) 2 . As shown in Figure 2 A, the fractions ( γ ) of GC contents of GC-rich and AT-rich region are about 0.53 ( γ GC ) and 0.47 ( γ AT ), respecti v ely. Therefore, l p between two regions is estimated about 5.9 nm. Because the average persistence length of overall DNA is 50 nm, 0 . 45 l p,GC + 0 . 55 l p,AT = 50 considering that the portions of GC-rich and AT-rich r egions ar e about 0.45 and 0.55 as shown in Figure 2 A. Combining two equations, the persistence lengths of GCrich and AT-rich regions are estimated to be 53 and 47 nm, respecti v ely. To inv estiga te the ef fect of dif ference in rigidities of two regions, in this simulation, we set the persistence length of the first half of the DN A (w hich includes the firstpackaged end and is r epr esented by blue particles in Figure  3 A) as longer than that of the second half (which includes the last-packaged end and is r epr esented by red particles) by tuning the values of the force constants ( K F and K L ) of the bending potentials. K F and K L indicate longer and shorter persistence lengths of the first half (GC-rich region) and the second half (AT-rich r egion), r especti v ely. As shown in Figure 3 C, P FIFO after a sufficiently long resting time ( τ rest ) depends strongly on the difference in the persistence length (or K F > K L ). When the first and second segments are equally rigid with K F = K L = 10, P FIFO converges to a pproximatel y 50%. As the first half of the DNA becomes more rigid than the second half with K F > K L , P FIFO increases slightly beyond 50%, which agrees with our experimental observations of 53.9% for the FIFO (Figure 2 C).
Figur e 4 compar es the experimentally measur ed lengths of ejected DNA by FIFO and LIFO in the three induction methods. To estimate the ejection fraction, it is necessary to set the control length to 100%, as shown by the black bar in Figure 4 A. To determine the control DNA length in this experimental condition, we measured the length of DNA purchased from NEB, which was tethered to the surface by the hybridization of a biotin-tagged oligomer and then immobilized on the surface by airflow in the flow cell ( 47 ). Based on this control length of 17.25 m, the ejected DNA fractions were 60-75%. It is well-known that in vitro phage ejection cannot be complete (68)(69)(70)(71). Many in vitro The resting step where the entrance of the portal is blocked by a blockage (the green cubic box) and the polymer chain inside the capsid is allowed to relax its conformation during τ rest . (vii) The ejection step where monomers spontaneously eject through the capsid portal. We counted the index ( i exit ) of the monomer that ejects for the first time. We applied a r esisting for ce ( f r = 0 . 35 k B T /σ ) to the last two monomers at the end of the capsid portal (yellow shaded ar ea) to mimic the effects of the osmotic pr essur e outside the capsid. (B) Probability (P FIFO ) that the monomer (that enters the capsid first) ejects first as a function of the resting time ( τ rest ) for two values of the maximum packing fraction ( φ) of a polymer chain of the relati v e rigidity r = 0 with all 200 monomers packaged. Red circle and orange square symbols r epr esent the cases for φ = 0.4 and 0.6, respecti v ely. P FIFO conv erges to 50% when enough resting time ( τ rest ) is gi v en. ( C ) Simulation results of P FIFO as a function of the resting time ( τ rest ) for different sets of ( K F , K L ). The simulation scheme is identical (i-vii) while the parameters ( K F and K L ) for the angular potential energy are tuned. P FIFO converges to values about 50% as the resting time ( τ rest ) increases.
phage ejection experiments showed that a fraction of DNA remained in the viral capsid ( 12 , 13 , 44 , 54-56 ). When a phage ejects DNA on the membrane of an E. coli host cell, osmotic pr essur e facilitates the completion of phage ejection since the bacterial cytosol has higher osmotic pr essur e than the e xtracellular environment. Howe v er, in vitro DNA ejection cannot be complete because no osmotic pr essur e gradient forms. Evilevitch et al. demonstrated a gel electrophoresis result of fractional ejections by varying the concentrations of polyethylene glycol ( 55 ). Figure 4 A shows another intriguing fact that the ejected DNA lengths by LIFO were consistently longer than those by FIFO in all three induction methods, although they were within the error ranges. The histograms in Figure 4 C-E more clearly show that the ejected fractions of LIFO were longer than those of FIFO. All LIFO histograms peaked at a pproximatel y 12-14 m, but the FIFO histograms peaked a t approxima tely 9-11 m. To explain our observations, we conducted computer simulations, in which we found a plausible reason that the difference in the rigidity between the first half (GC-rich region) and the second half (AT-rich region) of DNA may lead to the difference in the ejected fractions by FIFO and LIFO. In our simulations, we quantified the extent of the ejection process by defining the ejection fraction ( φ e ) as the ratio of the number of ejected monomers to the total number of monomers. As shown in Figure 5 A, φ e = 0 and 1 correspond to the states before the ejection begins and when all the monomers eject, respecti v ely. During the ejection step in our simulations, we introduced a resisting force ( f r ) of 0.35 k B T / , which is r epr esentati v e of the osmotic pr essur e outside the capsid. This approach allowed us to reproduce the values of φ e in our experiments   11.5, 8.5). As the ejection step begins, φ e increases but converges to a value ( φ c ) of less than 1. The converged value ( φ c ) of the ejection fraction ( φ e ) depends on whether the ejection proceeds by FIFO or LIFO pa thways. W hen the ejection proceeds by the LIFO pathway, φ c ∼ = 0 . 75 such that a pproximatel y 25% of monomers do not eject. When the DNA proceeds by the FIFO pathway, φ c ∼ = 0 . 66. Therefore, the fraction of ejected DNA by FIFO is smaller than that by LIFO, consistent with our experimental observation (Figure 4 A). This simulation result also explains the results of Supplementary Figure S7, which shows the qPCR results of the ejected DNA degraded by DNase. The qPCR results indica te tha t DNA a t 12.1 kb (25% of the genome) and 36.2 kb (75% of the genome) were more abundant than the middle (24.1 kb) after DNase treatment, consistent with the predicted fractions remaining in the capsid.
A remaining question is why the length of ejected DNA is longer via the LIFO pathway. To answer this question, we investigated how the total energy ( E , the potential energy plus the kinetic energy) of the DNA changes during the ejection process in our simulations. If the DNA is allowed to relax its conformation for a sufficiently long time, the total energies ( E ) at two limiting states of φ e = 0 and 1 are independent of whether the ejection occurs via the FIFO pathway or LIFO pathway. Howe v er, when the rigidity of the first half (GC-rich region) of the DNA is higher than that of the second half (AT-rich region) ( K F >K L ), the DNA follows a different pathway in the energy landscape during ejection. Figure 5 C depicts the total energy ( E ) of the DNA of ( K F , K L ) = (11.5, 8.5) during the ejection processes. When the first half of the DN A, w hich is more rigid, ejects first and the more fle xib le second half of the DNA remains in the capsid (FIFO pathway), the total energy E is lower during the ejection process than when the DNA proceeds by the LIFO pathway. This is because the bending energy (which would create a high internal pressure inside the capsid) is relaxed more efficiently via the FIFO pathway. Such a difference in the total energy E between the FIFO and LIFO pathways is most prominent in the middle of the ejection process at φ e = 0.5. Since the total energy is higher along the LIFO pathway, the internal pr essur e inside the capsid should also be higher when the DNA proceeds by the LIFO pathway instead of the FIFO pathway. This makes the ejection of monomers via the LIFO pathway more efficient than that via the FIFO pathway. On the other hand, when the first and the second halves of the DNA are equally rigid with K F = K L (the dotted line in Figure 5 C), we do not observe any difference in E between these two pathways.
As the difference in the rigidity (or the persistence length) between the first and the second halves of the DNA increases, a higher fraction of monomers ejects more efficiently via the LIFO pathway, i.e. φ c becomes larger for LIFO than for FIFO. Figure 5 D depicts the values of φ c for the FIFO and LIFO pathways as a function of the relati v e rigidity r = ( K F − K L ) /K L . As r increases, the ejection via the LIFO pathway becomes more efficient than that via the FIFO pathway. Because the value of r of phage DNA in our study is approximately 0.13 ( 66 ), the difference in φ c between LIFO and FIFO pathways in the simulations is expected to be a pproximatel y 9%, w hich is consistent with our experimental results: LamB 7.1%, glutaraldehyde 7.4%, 65 • C 5.7% (Figure 4 A). The differences in lengths observed for in vitro phage ejections and computer simulations provide additional evidence of both FIFO and LIFO mechanisms in addition to the sequence-specific staining results.
Single-molecule observations and computer simulations have shown that phage ejection can initiate from either end, which raises the question of why the FIFO and LIFO models have a nearly 50% probability for the LIFO model, rather than a 100% probability. One possible explanation is related to the process of pha ge infection, where pha ge particles first bind to the outer membrane protein LamB via the viral tail fiber before the phage DNA is ejected into the periplasm ( 72 ). The periplasmic nuclease NucA is known not to degrade DNA ( 73 ), and instead, mannose transporter complex in the inner membrane facilitates the entry of DNA into the cytoplasm. This complex process takes about 5 min in vivo , including pauses and stalls ( 74 ). In contrast, in vitro phage ejection occurs within a few seconds at a rate of 60 kb / s ( 71 ). During this process, a partial genome is present in the cytosol, and the first DNA to enter may have a higher likelihood of being attacked by restriction enzymes. Ther efor e, the exposur e time of r eco gnition sites for DN Abinding endonucleases may differ depending on which end of the DNA enters the cytoplasm first, as predicted by the FIFO and LIFO models. Different E. coli strains have endonucleases with distinct DN A reco gnition sites. For instance, Eco RI ( E. coli RY13), Eco 72I ( E. coli RFL72) and Eco 110kI ( E. coli BKM B-2004) cannot recognize the GC-rich region of the first half of DN A, w hereas Eco 143I ( E. coli RFL143) and Eco 105I ( E. coli RFL105) cannot recognize the AT-rich region of the second half. Consequently, different endonucleases can degrade either the first or second half of DNA depending on the host strain. For instance, if phage infects the E. coli RY13 strain following the LIFO model, the last DNA end enters first and is degraded by Eco RI, while the FIFO model protects the DNA from degr adation. Similar ly, if phage infects other hosts such as the E. coli RFL143 strain, the LIFO model results in degradation by Eco 143I, while the FIFO model protects the DNA. Ther efor e, the nearly 50% probability of ejecting the first or last DNA end is an effecti v e survi val strategy for phage to infect various E. coli strains.
Another notable aspect of in vivo phage ejection is the possibility of multiple bacteriophages infecting a host cell sim ultaneousl y ( 74 ). This combination of FIFO and LIFO from multiple viruses would further enhance the protection of the entire viral genome against restriction enzyme attack. With both ends of the DNA secure in different orientations, more proteins can be produced to construct mature bacteriopha ges. Other bacteriopha ges, such as T2 and T4, have circularl y perm uted genomes (75)(76)(77), w hich likel y provide similar protection against restriction enzyme attack. Thus, the 50% probability of ejecting the first or last DNA end is an effecti v e survi val strategy for phage to infect various strains of E. coli as hosts. This bi-directional ejection allows the phage to evade detection and degradation by host restriction enzymes, as different parts of the genome are protected in different orientations.
In this study, we visualized phage-ejected DNA stained with AT-specific reagents. This visualization allowed us to determine the DNA directionality and the ejection lengths. Contrary to the popular model of only LIFO, singlemolecule data and computer simulation results show that the DNA ejection of phages starts from either the left end or the right end. An interesting observation is that the ejected DNA length by FIFO is shorter than that by LIFO. We reproduced the same result through our computer simulations by considering GC-rich first-half and ATrich second-half regions of DNA. The shorter length of the DNA ejected through the FIFO mechanism compared to the LIFO mechanism further supports this bidir ectionality. Furthermor e, the qPCR r esults obtained after DNase treatment of the ejected DNA also support the bi-directionality of the ejection process. The extremely condensed DNA within phage capsids is assumed to be immobile. Howe v er, our e xperiment and simulation studies demonstrate the possible mobility of the DNA with a liquid crystal density. This bi-directional ejection can be seen as a survival strategy for the phage, as it increases its chances of avoiding degradation by endonucleases present in its host cell. Additionally, this mobility of the DNA within the phage capsid, despite the extremely condensed state, highlights the potential importance of DNA mobility in viral survival and replication.

DA T A A V AILABILITY
The simulation codes and data underlying this article will be shared on reasonable request to the corresponding author.

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.