Computational Basis for On-Demand Production of Diversified Therapeutic Phage Cocktails

The antibiotic resistance crisis has led to renewed interest in phage therapy as an alternative means of treating infection. However, conventional methods for isolating pathogen-specific phage are slow, labor-intensive, and frequently unsuccessful. We have demonstrated that computationally identified prophages carried by near-neighbor bacteria can serve as starting material for production of engineered phages that kill the target pathogen. Our approach and technology platform offer new opportunity for rapid development of phage therapies against most, if not all, bacterial pathogens, a foundational advance for use of phage in treating infectious disease.

therapeutic effects) (6). Therapeutic cocktails are generally composed of virulent phages (capable only of the lytic life cycle) isolated from environmental samples. Temperate phages (capable of both lytic and lysogenic life cycles) have typically been excluded because the resulting lysogenic target bacteria would survive to spread resistance to that phage; furthermore, temperate phages may carry cargo genes that promote antibiotic resistance and/or bacterial pathogenicity (7,8). However, there are far more genome sequences available for temperate phages (in the form of prophages integrated within bacterial genome sequences) than for virulent phage genome sequences at NCBI (9), and temperate phages can be converted into nonlysogenic phages using modern genome engineering tools (e.g., by knocking out the integrase gene) (10).
The definition of genomic islands (GIs) has been loosened over time to include any chromosomal DNA segment with evidence of horizontal transfer (11). We have introduced the term integrative genetic elements (IGEs) to mean that subset of GIs whose integration into the chromosome can be ascribed to a self-encoded integrase, either from the tyrosine or serine recombinase families (12). The IGE class includes prophages, integrative and conjugative elements (ICEs), and satellites. We have developed two complementary computational tools, Islander (13) and TIGER (12), that identify and precisely map IGEs within bacterial (and archaeal) genome sequences. This software reveals the bacterial host, complete sequence, and precise ends of each prophage. Knowledge of the prophage sequence enables its genome engineering to yield a lysogeny-disabled variant safe for therapy, for example through long amplicon PCR and Gibson assembly, with "rebooting" by introduction and growth in the target bacterium (14). Knowledge of the host enables identification of prophage-laden bacteria that are very close relatives of any given target bacterium, increasing the likelihood that the phages produced will be efficacious on the target. Our powerful computational prophage prediction software facilitates engineering with existing methodologies (3,14,15). This set of computational, molecular biology, and microbiology tools together constitute a technology platform for on-demand production of phage therapies against bacterial pathogens (Fig. 1). As an initial demonstration of this approach, we produced five engineered lysogeny-disabled phages that kill Pseudomonas aeruginosa PAO1 in liquid culture as well as in a waxworm model of infection. We foresee application of this platform to develop prepared or on-demand phage collections to target nearly any bacterial pathogen or to control undesirable components of environmental or clinical microbiomes.

RESULTS
Identification of prophage-laden close relatives of the target bacterial strain. Each IGE encodes an integrase that catalyzes recombination between an attachment site in the circular IGE (attP) and one in the bacterial chromosome (attB), yielding the IGE integrated into the chromosome flanked by left (attL) and right (attR) attachment sites. We have developed two complementary algorithms that identify IGEs, including prophages. Islander identifies IGEs encoding a tyrosine integrase with attB in a tRNA or tmRNA gene (13) (Fig. 1A, right); TIGER identifies IGEs encoding both tyrosine and serine integrases with no bias toward attB context (12) (Fig. 1A, left). Both methods are unique relative to competitors in their ability to precisely map attL and attR for each IGE; moreover, they have superior information retrieval properties. Prior to the publication of TIGER, Bertelli et al. (16) evaluated 20 genomic island (GI)-finding methods with two benchmarking systems: (i) GI-positive and GI-negative segments from 104 bacterial chromosomes and (ii) 80 "gold standard" GIs from six bacterial chromosomes. Islander was top ranked by both systems for precision (in the information retrieval sense), with values of 0.971 for GI-positive or -negative segments and 1.000 for gold standards. Using the benchmarks and evaluation tools of Bertelli et al., we measured the precision of TIGER as 1.000 in both systems (12). Thus, our two methods have better precision, by two measures, than any of the other 19 tested methods. Recall values were lower than for other methods by these benchmarking systems, which we attribute partly to the inclusion of many entries among the GI-positive segments and gold standards that are not associated with integrases and are therefore neither IGEs nor intact prophages. On our own set of 63 gold standard IGEs (not loosely defined GIs) from the same six chromosomes as Bertelli et al. (and one additional chromosome), we measured precision and recall for TIGER at 0.952 and 0.952, respectively, and for Islander at 1.000 and 0.349, respectively (12); the low recall of Islander is expected from its inability to find IGEs that integrate in sites other than tRNA or tmRNA genes (13). Aside from these superior information retrieval properties, TIGER and Islander are also unique in their mapping precision (in the sense of the exactness of the IGE genomic coordinates that they call); both methods identify the identity blocks at the attL and attR termini of the IGE. A prophage calling tool was developed along with TIGER and benchmarked against a set of temperate phage isolate genomes from NCBI as standards and against bona fide ICEs, mock GIs, and GI-negative chromosomal segments as negatives; it was shown to have 0.982 recall and 0.997 precision (12). With Islander, TIGER, and our prophage caller, we determined the prophage content of the 26 genome-sequenced Pseudomonas aeruginosa strains that were available from the American Type Culture Collection (ATCC), totaling 44 prophages. Two IGE-rich P. aeruginosa strains (referred to here as Pae5 and Pae1505) available from ATCC are close relatives of our target strain PAO1 (see Fig. S1 in the supplemental material). Pae5 has 12 IGEs, including one filamentous prophage and five additional prophages (41Z, 42argF, 44G, 52yheS, and 64L). Strain Pae1505 has 10 IGEs, including one filamentous and two additional prophages (43spxA and 52S) (see Table S1 in the supplemental material). IGE names indicate length (in kilobase pairs) and insertion site gene (a single letter representing the identity of a tRNA gene). Filamentous phages tend not to lyse bacteria and are therefore less suitable for therapy (17). All the prophages in these two strains encode only tyrosine integrases; however, our software is capable of finding prophages with serine integrases as well.
Ensemble validation of prophages. We sought to determine which of our suspected prophages were able to excise from the bacterial chromosome, produce phage particles, and infect our target strain PAO1. Strains Pae5 and Pae1505 were induced by treatment with mitomycin C; samples were collected at 0, 1, and 2 h, and DNA was prepared from three sample types: a cell pellet, sample from a supernatant filtrate (which should contain free phage particles), and a spot plaque of the filtrate grown on a lawn of PAO1. Precise mapping of the prophages by Islander/TIGER allowed design of PCR tests of the circular junction of each excised prophage; PCR products were not detected for the filamentous phages, but they were detected for the other seven prophages, in all three sample types of the 2-h time points ( Fig. 2A). We reexamined all cell pellet DNA samples using deep sequencing, first with our mobilome discovery software Juxtaposer (18) that confirmed the excision products attP and attB, for six of these seven prophages. We also applied our quantitation tool AttCt (18) to analyze attP and attB yields (normalized to attL and attR); higher levels of the former indicate postexcision replication. This analysis reveals (Fig. 2B) interesting biological differences in the onset of excision and replication rates among the prophages, even along this abbreviated time course. One wild-type (WT) phage from each source bacterium, 42argF and 52S, was isolated from each spot plaque by further plaque purification.
Although not all prophages are inducible by mitomycin C, the methods for validation are well suited to work with any number of alternative induction strategies, including pH and temperature shifts, UV induction, spontaneous induction, and chemical induction (19)(20)(21)(22). Genomic content of prophages. Dot plots of the prophage genomes allowed us to reject two prophages (43spxA and 52yheS) as being similar to others, leaving five unique prophages that we proceeded to develop for therapy ( Fig. S2; Table 1). Annotation of these prophage genomes ( Fig. S3) revealed genes for phage structural proteins (capsid, tail, terminase, and portal) and promoting lysis (endolysin and holin); only 42argF contained genes predicted to encode deleterious proteins (FimA, a known virulence factor [23], and a hic toxin-antitoxin system [24]). 42argF also encodes an endosialidase, which allows phage to recognize and degrade bacterial polysaccharide capsules (25), and an anti-CRISPR AcrF6 protein (Fig. S3).
Engineering of ⌬int lysogeny-disabled prophages. Conversion of a temperate phage into a lytic phage requires deletion of at least one key lysogeny determinant. Others have successfully targeted the repressor gene (3) or the "lysogeny control region" (integration and prophage maintenance gene cluster) (14) for this purpose. For high-throughput genome engineering of phages with diverse bacterial hosts, the repressor gene may be difficult to identify reliably amid other helix-turn-helix proteins. We chose to delete the integrase gene (int), which is essential for lysogeny and readily identifiable. Conveniently for deletion, int is typically located at one end of the prophage genome (12). The integrase gene and its surrounding regions, including the attL and attR sites, were deleted by using PCR primers to generate long, partially overlapping amplicons comprising all included prophage sequences, followed by Gibson assembly of the amplicons and transformation into the target host ( Fig. 1D) (14). Each amplicon was 8 to 15 kbp in length, producing terminal overlaps with neighboring amplicons of 40 to 60 bp (Table S2). Gibson assembly was performed with the three to five partially overlapping amplicons, and the products were transformed into electrocompetent (41Z, 42argF, 44G, and 52S) or chemically competent (64L) PAO1. The transformed bacteria were plated, and phages were recovered from plaques at 16 h postplating. PCR was used to verify Δint junctions in the phage genomes (Fig. S4).
Phage morphology. Transmission electron microscopy (TEM) was used to determine morphology for each phage (Fig. S5). All five phages had long, flexible, noncontractile tails characteristic of Siphoviridae. Four of the phages had icosahedral capsids, whereas 52S had a prolate head. Head and tail dimensions were measured from TEM images ( Table 1). Four of the five phages measured 200 to 300 nm in length, whereas 52S (both WT and Δint) measured 150 nm in length.
Engineered phages kill strain PAO1 in liquid culture. The Δint phages were tested for the ability to kill the target pathogen PAO1 in liquid culture. Each phage stock was added to a mid-log-phase PAO1 culture at a multiplicity of infection (MOI) of 10, and the culture was incubated at 37°C with shaking (250 rpm) for 24 h. Aliquots were removed from the culture at 1-h intervals for the first 4 h following phage addition in order to assess the degree and timing of bacterial death and phage proliferation in the culture. By 4 h following phage addition, all cultures exposed to Δint phage showed reduced bacterial titers relative to mock-exposed cultures (i.e., those receiving buffer only), indicating that each of the Δint phages is capable of killing PAO1 (Fig. 3 and Fig. S6). Additionally, phage titers increased at least 10-fold, and as much as 1,000-fold, in all of the cultures by 4 h (Fig. S7). Two mixtures (cocktails) of Δint phage stocks were designed: cocktail 3X, comprised of phages 44GΔ, 52SΔ, and 64LΔ, and cocktail 5X, comprised of the 3X phages plus 41ZΔ and 42argFΔ. These cocktails were added to PAO1 cultures at an MOI of 10 (MOI of 3.3 for each Δint phage in 3X, and MOI of 2 for each in 5X), aliquots were removed from the cultures at 1-h intervals over the course of 4 h, and the bacterial and phage titers within each aliquot were measured as described above. Both cocktails reduced bacterial titers by 10 6 -fold within 4 h postexposure.
⌬int phage therapy protects waxworms from PAO1 infection. We sought to determine whether these phages confer therapeutic effects in the context of a PAO1 infection model. The Galleria mellonella larva (waxworm) model is convenient, low cost, and well established for Pseudomonas and other bacterial infections and testing of antimicrobial therapies (26)(27)(28)(29)(30). Importantly, antimicrobial therapies, including phage therapy, that are efficacious in waxworm models of infection generally also show efficacy in mammalian models of infection (29,31).
Each larva was injected with 50 CFU of strain PAO1, and 30 min later injected with buffer (negative control), a single Δint phage (MOI of 10), or a phage cocktail (MOI of 10 for sum of phages in cocktail), with each therapy tested in 10 larvae (Fig. 4). This experiment was replicated twice more (Fig. S7). Waxworms treated with buffer alone showed only 30 to 40% survival by 3 days postexposure (black line, Fig. 4 and Fig. S7); mortality was due to PAO1 rather than the injection procedure in that mock-infected larvae (injected with buffer instead of PAO1 and then injected a second time to simulate treatment) showed 100% survival (gray line). In contrast, larvae treated with single Δint phage showed 40% to 80% survival. Larvae treated with phage cocktails showed further gains in survival, ranging from 70% to 90% (red and orange lines). Treatment with phage in the absence of infection had no deleterious effect on survival (gray line). These results indicate that Δint phages, particularly when combined as a cocktail, can protect waxworms against PAO1 infection.

DISCUSSION
This report validates a new technology platform that enables on-demand production of therapeutic phages against a bacterial pathogen of interest. The first step in our approach is to use two complementary computational tools (Islander and TIGER) to identify and precisely map the prophages present in genome sequences of bacteria closely related to the pathogen. We have carried this out for many genomes (12), and we plan to extend our analysis to all available genome sequences. Preemptive creation of prophage databases in this manner allows for rapid turnaround in phage therapy cases where time is limited-the physician can use the pathogen's genome sequence, 16S rRNA gene sequence, or multilocus sequence typing (MLST) sequences to place it on a phylogenetic tree, to identify close relatives bearing large numbers of prophages. These prophages then serve as starting material for construction of synthetic phages that are engineered for therapeutic use through deletion of genes (such as int) essential for lysogeny. Identification of numerous prophages from close relatives of bacterial targets has the potential to overcome many of the hurdles associated with environmental phage isolation, including defining optimal hosts, sequencing isolated phages, and annotation of deleterious gene products. Once prophages are identified for the host of interest, phage engineering can be completed by previously established methods (3,14,15).
In principle, our approach is pathogen-agnostic and could be extended to any pathogen group. Raw numbers of prophages are not generally a problem; Fig. S8 in the supplemental material shows the average prophage count per genome for diverse ESKAPE (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter species) and tier 1 select agent pathogen species. Ten pathogen species have double or more the average prophage count of all bacteria, although two (Burkholderia mallei and Francisella tularensis) do not. However, particular bacterial groups may present practical problems. Our study host, P. aeruginosa PAO1, has well-defined transformation protocols, which are not available for all pathogens (32)(33)(34), but circumventions are available (14). Many bacterial systems have phages with narrow host ranges (35,36); host range factors are still being investigated but have been linked to tail fiber mutations or host receptor mutations (36)(37)(38). While we did not encounter this problem, the validation methods described allow for prescreening of prophages that are inactive against the target pathogen prior to engineering. Further, Gibson assembly allows for additional engineering at other sites in the phage genome aimed at expanding its host range. Once such challenges are solved, both the computational and laboratory methods described will be suitable for high-throughput scale-up and can be further streamlined to support a phage factory that prepares therapeutic phage sets for any and all pathogens.

MATERIALS AND METHODS
Prophage detection. Two IGE discovery algorithms, Islander (13) and TIGER (12), were applied to 2,023 Pseudomonas genomes downloaded from GenBank in October 2017. Predicted prophage genomes were annotated. The multiPhATE pipeline (39), which calls Glimmer (40), Prodigal (41), and Phanotate (42), was used to predict open reading frames. Prokka (43) tFind, and rfind (13) were used for functional annotation. HHPRED (44,45) was used to further characterize phage functions to compare homology of each protein annotated against the Pfam-A v32.0 and PDB_mmCIF70_28_Nov databases. Prophage genomes were compared to each other to identify similar prophages. Gepard (46) was used to generate dot plots for the seven nonfilamentous prophage sequences. Phage genome maps were created using Easyfig (47).
Prophage induction. Prophage-laden strains P. aeruginosa 2192 (ATCC 39324; Pae5) and P. aeruginosa Boston 41501 (ATCC 27853; Pae1505) were purchased from ATCC. Strains were grown overnight in LB broth, diluted 1:100, and grown to an optical density at 600 nm (OD 600 ) of 0.4 to 0.5. Samples (1 ml) were collected at 0, 1, and 2 h after treatment with 1 g/ml mitomycin C. Cells were pelleted at 16,000 ϫ g for 2 min; supernatant was removed and filtered through a 0.22-m filter. Genomic DNA was isolated using the Qiagen DNeasy blood and tissue kit (Qiagen, catalog no. 69504). Filtrates were spotted onto a lawn of P. aeruginosa PAO1 (a kind gift from Annette LaBauve) with 0.5% LB soft agar and incubated overnight. Top agar was collected from the cleared spot regions and soaked in 0. Monitoring prophage activity through deep sequencing. P. aeruginosa (Pae5 and Pae1505) genomic DNA sequencing libraries were prepared using Nextera DNA library prep kit (Illumina, catalog no. FC-121-1031) and utilizing the Nextera DNA Sample Preparation Index kit (Illumina, catalog no. FC-121-1011) following the manufacturer's recommended protocol. DNA samples and libraries were quantified using Qubit high-sensitivity DNA assay kit (Thermo Fisher Scientific, catalog no. Q32854). Libraries were pooled in equal quantity and combined to multiplex to make a final library, and quality control (QC) was done again using Qubit quantification kit and Agilent bioanalyzer using high sensitivity DNA chip (Agilent Biotechnology, catalog no. 5067-4626). The final combined library was sequenced using Illumina technology on a NextSeq 500 sequencer using high-output 300-bp single-end read sequencing kit.
Sequencing reads were processed through BBDuk for quality filtering. The filtered reads were analyzed with Juxtaposer (18) to find mobile elements in the bacterial genomes. Briefly, Juxtaposer searches for recombinant reads relative to the reference genome sequence, which includes reads for the attP and attB products of IGE excision. Finally, attCt software (18) was used to determine counts for each IGE of attL, attR, attB, and attP, reporting attP/F and attB/F values, where F ϭ attL ϩ attR ϩ 2 attB).
Wild-type phage isolation. Soaked filtrates from mitomycin C (MMC) induction were serially diluted in SM phage buffer. One hundred microliters of P. aeruginosa PAO1 was infected with 100 l of each phage dilution, allowed to adsorb 20 min, and plated using 4.5 ml of 0.5% LB soft agar. Plaques were isolated in SM phage buffer, and PCR tests for the attP site were performed to determine which plaque was each phage (primers listed in Table S2 in the supplemental material).
Gibson assembly of ⌬int phages. Primers were designed to obtain long PCR fragments with overlapping joints suitable for Gibson assembly strategies (Table S2). Primers for the integrase deletion were created with artificial 40-bp overlaps. Briefly, a 20-bp primer was created around the integrase gene, and a 20-bp flanking region was added to the 5= end of the primer from the opposing end of the Δint circular junction. Long PCR was performed using the NEB Phusion High-Fidelity DNA polymerase master mix. PCR conditions were as follows: (i) 98°C for 2 min; (ii) 35 cycles with 1 cycle consisting of 98°C for 30 s, melting temperature (T m ) minus two degrees for 30 s, and 72°C for 1 min/kbp; and (iii) 72°C for 10 min. NEB Gibson Assembly master mix was used to ligate phage fragments together. Then, 0.3 pmol of each long PCR product was incubated at 50°C for 15 min for three-fragment phages (41ZΔ, 42argFΔ, and 44GΔ) and 60 min for four or more fragment phages (52SΔ and 64LΔ).
Competent cell preparation and transformation. Electrocompetent PAO1 cells were prepared as previously described (48). Briefly, an overnight culture of strain PAO1 was diluted 1:100 and grown to an OD 600 of 0.5. Cells were pelleted 2 min at 16,000 ϫ g, and supernatant was removed. Cells were washed twice with 300 mM sucrose and resuspended in 300 mM sucrose. For all phages except 64L, 5 l of the undiluted or 1:3 diluted Gibson assembly reaction was delivered into electrocompetent cells using a Bio-Rad electroporator (2.5 kV, 200 ⍀, 25 F, 2 mm). Cells were allowed to recover in 1 ml of LB medium for 1 h, shaking at 37°C. Cells were pelleted for 2 min at 16,000 ϫ g, 800 l of medium was removed, and cells were resuspended in the remaining medium. Transformations were diluted in SM phage buffer by 10-fold and 1,000-fold. Two hundred microliters of the transformed cells or dilutions was used to infect 100 l of mid-log PAO, incubating for 20 min, plating with 0.5% LB soft agar, and incubating overnight at 37°C.
Phage 64L did not produce plaques using electrotransformation protocols. Chemically competent (CC) PAO1 was prepared as previously described (49). Overnight cultures were diluted 1:100 and grown to an OD 600 of 0.8. Cells were chilled and harvested by centrifugation, washed with 100 mM MgCl 2 , and incubated on ice in 175 mM CaCl 2 for 20 min. The final cell pellet was resuspended in 100 mM CaCl 2 .
Fresh competent cells (200 l) were incubated with various amounts of Gibson assembly product DNA for 60 min on ice, heat shocked at 42°C for 1 min, chilled, and recovered in 1 ml of LB broth for 1 h prior to plating with 100 l of mid-log PAO1 and 4 ml of 0.5% LB top agar (TA).
Recovery of ⌬int mutant phages. Plaques recovered from the transformations were picked, put in 0.1 ml SM phage buffer, and confirmed through PCR with primers designed in the flanking region of the deletion (Table S2). Phage stocks were prepared by titrating on plates to obtain a web pattern of lysis. Plates were flooded with 8 ml of SM phage buffer overnight at 4°C, filtered through a 0.22-m filter. Phage stocks were stored in 0.3% sucrose.
Electron microscopy. Electron microscopy grids were prepared with fresh lysates at Ͼ10 9 PFU/ml. Ten microliters of lysate was added to a carbon grid (Ted Pella, catalog no. 1813) and allowed to incubate 10 min. The grids were washed twice with 10 l of water for 2 min. Grids were stained with 10 l of uranyl acetate alternative stain (Ted Pella, catalog no. 19485, using gadolinium acetate tetrahydrate) for 2 min and wicked off. Grids were allowed to dry at room temperature in a chemical fume hood for 1 h and stored in a grid box until imaging.
Transmission electron microscopy observations were conducted using a Themis Z transmission electron microscope (Thermo Fisher Scientific, Hillsboro, OR, USA) operated in scanning transmission electron microscopy (STEM) mode at an accelerating voltage of 300 kV. Image signals were collected using a high angle annular dark field (HAADF) detector, which is sensitive to the heavy elements (Gd) employed in the stain. In the micrographs presented here, the dark-field image contrast has been inverted, with the heavy metal stain showing as dark regions on the images.
Immediately prior to observation, specimens were plasma cleaned (Mobile Cubic Asher, IBSS Group, Burlingame, CA, USA) to reduce the build-up contamination under exposure to the focused electron beam. Plasma cleaning was conducted for 5 min using ambient air at a base pressure of 10 Ϫ4 torr and power setting of 34 W.
Liquid killing assays. Overnight PAO1 cultures were diluted 1:100 to an OD 600 of 0.02 to 0.04 in LB broth. Cultures were grown to an OD 600 of 0.4 to 0.5. Cells were pelleted by centrifugation at 3,000 rpm for 5 min. Bacterial pellets were resuspended in 1 ml PB containing a single phage stock at a multiplicity of infection (MOI) of 10 or a phage cocktail with a total MOI of 10; 5 ml of LB broth was added to each culture. Cultures were grown for 24 h, removing 0.5-ml samples at 0, 1, 2, 3, and 4 h after phage addition. Each sample was processed for PFU, CFU, and OD 600 measurements.
Bacterial cell counts were performed by serially diluting the sample in LB broth to 10 Ϫ8 and spotting 3 l of each dilution onto a solid LB agar plate and incubating at 37°C for 16 h. PFU were determined by centrifugation of each sample at 3,000 rpm for 5 min, removing 100 l of supernatant, serial diluting the supernatant to 10 Ϫ10 , followed by spotting 3 l of each dilution onto a 0.5% soft agar overlay plate. Bacterial and phage titers were calculated by counting the number of colonies or plaques, respectively.
Waxworm phage therapy. Galleria mellonella larvae were purchased from Timberline (Marion, IL) and stored at 15°C until use within 2 weeks. Only larvae weighing between 0.25 and 0.35 g were used for injection experiments. Larvae were equilibrated at room temperature for 4 h prior to injection. Five microliters of strain PAO1 (10 3 CFU/ml) was injected into the second to last right proleg using a 250-l, model 1725 LT syringe (Hamilton, catalog no. 81101) and PB600 repeating dispenser (Hamilton, catalog no. 83700). Thirty minutes after the PAO1 injection, 5 l phage (MOI of 10) (10 4 PFU/ml) was injected into the second to last left proleg, and the larvae were incubated at 37°C for 72 h. Death was assessed by melanization and lack of movement at 16, 20, 24, 48, and 72 h postinjection. Kaplan-Meier survival curves were calculated as described previously (50).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.

ACKNOWLEDGMENTS
This work was supported by the Laboratory Directed Research and Development program at Sandia National Laboratories, which is a multimission laboratory managed and operated by National Technology and Engineering Solutions of Sandia LLC, a wholly owned subsidiary of Honeywell International Inc. for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-NA0003525.