Targeted metagenomic assessment reflects critical colonization in battlefield injuries
ABSTRACT
Current diagnostics and clinical management strategies for combat wounds are based on decisions made by expert clinicians. However, even in the hands of experienced surgeons, wounds from combat injuries can exhibit failed healing and complications related to limitations in the rapid and comprehensive generation of diagnostic information. Previous studies have demonstrated the possible use of genomic sequencing approaches to detect microbial signatures involved in combat casualty care. While effective, whole metagenome sequencing is limited by the depth required to confidently detect all relevant signatures. To address this, we developed a targeted capture sequencing panel to detect microbial signatures relevant to wound healing. These targets include known microbial nosocomial pathogens, wound colonizers, and genes involved in virulence and antimicrobial resistance. A bioinformatics pipeline was built to identify genomic regions of interest and over 8,000 oligonucleotide probes were designed for capture. The panel was synthesized and validated using control reference genomes in human background and on wound-effluent samples from a cohort of combat-injured U.S. service members. Our panel was sensitive against wound-colonizing species, Acinetobacter baumannii and Pseudomonas aeruginosa, and was specific in detecting corresponding virulence and antimicrobial-resistance genes as well as other pathogenic species present in microflora mixtures. Random forest feature permutation confirmed the prevalence of Acinetobacter and Pseudomonas in critically colonized wounds and wounds that failed to heal, respectively. Our results demonstrate the capability of targeted sequencing tools and analysis platforms to profile and deliver information on pathogenic factors influencing wound progression, thereby guiding therapeutic intervention.
IMPORTANCE
Microbial contamination in combat wounds can lead to opportunistic infections and adverse outcomes. However, current microbiological detection has a limited ability to capture microbial functional genes. This work describes the application of targeted metagenomic sequencing to profile wound bioburden and capture relevant wound-associated signatures for clinical utility. Ultimately, the ability to detect such signatures will help guide clinical decisions regarding wound care and management and aid in the prediction of wound outcomes.
INTRODUCTION
Metagenomics and corresponding informatics analyses continue to prove valuable for exploring the interplay of microbial factors in human health and infection. Such techniques can provide information with utility in trauma and corresponding infections, including combat casualty care (1, 2). The severely invasive nature of combat trauma creates massive regions of injury, colonization, and infection, requiring specialized diagnostic and aggressive therapeutic approaches. However, the lack of comprehensive microbiological detection assays has hindered early detection of potentially detrimental microbial factors, resulting in wounds that fail to heal and the occurrence of other complications (3 – 5).
Despite the impact of infection on combat wounds, the functional gene profiles of associated bioburden are challenging to thoroughly measure, creating obstacles in applying such data for prognostic and clinical management purposes. This is especially crucial as multidrug-resistant microorganisms are consistently observed in injured service members with reports throughout the recent conflicts in Iraq and Afghanistan (6). Similarly, an assessment of isolates from the Department of Defense health system between 2009 and 2015 identified 27,000 multidrug resistant Gram-negative isolates. In addition to the assessment of resistance, the examination of virulence genotypes is also important in revealing functional bioburden profiles that are likely to expand into clinically problematic infections capable of impeding wound healing. The Trauma Infectious Disease Outcomes Study has examined in detail the importance of evaluating such factors in the bioburden of wound infection (7). Ultimately, a more timely, informative evaluation of wound infection promises to reduce morbidity, shorten hospital stays, and improve rehabilitation for combat-wounded service members.
One strategy to evaluate bioburden profiles of wound samples is through genomic sequencing. Microbial metagenomic sequencing, in particular, can provide this information through the detection and analysis of various genomic factors. Metagenomic sequencing for wound pathogen detection has previously demonstrated concordance with quantitative bacteriology analyses in combat injuries (2). However, untargeted whole genome sequencing is limited by sequencing depth, resulting in insufficient coverage yield required for sensitively detecting and analyzing individual microbial genes, largely due to the human genomic sequence background in clinical samples. Therefore, novel targeted sequencing approaches are needed to enable DNA sequencing as a novel paradigm for wound infection management.
Targeted sequencing platforms that leverage amplification or hybridization capture of specific nucleotide sequences have already demonstrated clinical utility for the detection and analysis of numerous human genetic conditions, including cancer, cardiovascular disease, immune deficiency, and a range of inherited diseases (8 – 10). Furthermore, efforts to identify sequencing targets and apply them for the detection of microbial pathogens, such as respiratory syncytial virus (11), Ebola virus (12), and drug-resistant Mycobacterium tuberculosis (13), have been routinely explored. Previous efforts identifying genomic signatures for relevant targets, for instance, influenza virus (14), can be leveraged for targeted amplification or capture of such regions. Accordingly, an amplification panel targeting 518 signatures has been used to assess antimicrobial-resistance factors present in the International Space Station (15). The lower limit of detection and elevated specificity that could be delivered by an approach employing targeted capture sequencing would facilitate the detection of microbial genomic signatures in clinical samples. The integration of targeted capture and sequence-based readout techniques can detect low-abundance nucleotides, which are applicable for the detection of early-stage signatures before detrimental phenotypes start to emerge in patients. Sequencing captured regions facilitates both the detection of relevant genomic regions and identification of novel sequence variation, which is not possible with existing qPCR-based approaches for AMR detection. Dynamic evolution of resistance genotypes has been observed for numerous wound-relevant pathogens, which may not be detectable using assays reliant on static gene signatures (16, 17). Sequence-based approaches can be used to survey emerging and naturally occurring mutations that could confer novel or enhanced biological functions relevant to clinical treatment. Furthermore, the ability to use millions of capture probes on such platforms enables massively multiplexed diagnostic applications.
To assess the utility of such approaches for wound infection biomarkers, a targeted panel for the capture and sequencing of microbial genomic signatures relevant to wounds from combat injuries was developed. This panel selectively sequences thousands of microbial genomic regions relevant to bioburden in traumatic wound injuries, thereby facilitating high-confidence detection of critical microbial signatures that are otherwise difficult or impossible to assess using current standards of care. These microbial signatures include genus and species-level identification of pathogens, antimicrobial resistance, and virulence. The resultant panel was validated for sensitivity and specificity using reference genome controls. Subsequent evaluation in wound samples derived from combat extremity injuries in U.S. military service members was carried out to demonstrate the clinical utility of these targeted signatures. Overall, we concisely reported the utility and feasibility of using targeted panels in profiling wound bioburden and capturing relevant wound-associated microbial signatures. Our results suggest that such applications could guide the clinical management of wounds and influence subsequent therapeutic interventions with opportunities for personalized treatments.
RESULTS
Selection of genomic regions for taxonomic, virulence, and antimicrobial-resistance targets and panel design
The targeted panel was designed to detect relevant pathogen genera/species and genes that encode for antimicrobial resistance and virulence factors (Fig. S1). First, a subset of microbial genera/species known to be relevant to wound healing and hospital-acquired infections was selected for inclusion (Table 1). DNA sequences conserved within, and unique to, these species were identified based on genomic loci previously employed for a microbial detection microarray (18, 19). To achieve species and/or genus-specific detection, probes with maximal predicted breadth and sensitivity were identified. Probes were selected via optimization of target hybridization probabilities, screening against control and relevant reference sequence, and for human cross-hybridization.
TABLE 1
| Bacteria | Fungi |
|---|---|
| Achromobacter sp. | Absidia sp. |
| Acinetobacter baumannii | Aspergillus sp. |
| Enterobacter cloacae | Bipolaris sp. |
| Enterococcus faecalis | Candida sp. |
| Enterococcus faecium | Cunninghamella sp. |
| Escherichia coli | Fusarium sp. |
| Klebsiella pneumoniae | Lichtheimia sp. |
| Proteus sp. | Mucor sp. |
| Pseudomonas aeruginosa | Pythium sp. |
| Pseudomonas putida | Saksenaea sp. |
| Staphylococcus aureus | |
| Stenotrophomonas maltophila |
Next, an amplification-based targeted sequencing platform for the detection of AMR was designed. This AMR panel was derived from AmpliSeq-based probes designed for loci with confirmed genotypes from previous efforts (15), in addition to supplementation with antifungal genomic signatures. A collection of 702 AMR genes, both bacterial and fungal, were selected for inclusion and 3,059 probes were designed to capture AMR genomic loci (supplemental material). Over 3,000 multiple sequence alignment files were retrieved from the Virulence Factor Database (VFDB; October 2018) and 3,217 probes were designed for the detection of these virulence signatures (20). These virulence factors comprise genes across 32 pathogenic species, which are involved in various virulence-associated activities such as adherence, invasion, motility, and biofilm formation.
Multiple single and polymicrobial reference genomic DNA sequences were selected for panel testing and validation. First, Acinetobacter baumannii and Pseudomonas aeruginosa references were selected due to their prevalence in combat wound infection. As traumatic wounds are prone to polymicrobial bioburden, a mixed microbial pathogen gDNA reference from the American Type Culture Collection (ATCC) containing 10 human pathogenic species commonly observed in infections was also selected (Table 2). Furthermore, as clinical samples will also contain non-pathogenic commensal microbes, an ATCC gDNA mixture of six commensal skin bacterial species (Table 3) was tested. This latter mixture represented a polymicrobial “negative control,” assessing the detection response elicited from naturally occurring skin microbes. Reference gDNA for each species and microbial mixture were spiked into a human reference control background and subjected to targeted hybridization capture via the designed panel prior to sequencing. Validation on single and polymicrobial reference genomes was carried out and 58 effluent specimens from wounds in combat-injured U.S. service members were analyzed.
TABLE 2
| Species present in ATCC MSA-4000 | % of mixture | Probes designed? | Genus reads detected (% of total reads) | Species reads detected (% of total reads) |
|---|---|---|---|---|
| Neisseria meningitidis | 28.9 | No | Yes (1.27) | Yes (0.24) |
| Streptococcus pneumoniae | 28.9 | No | No | No |
| Klebsiella pneumoniae | 14.4 | Yes | Yes (39.28) | Yes (37.85) |
| Staphylococcus aureus | 15.1 | Yes | Yes (46.95) | Yes (46.93) |
| Streptococcus pyogenes | 7.2 | No | No | No |
| Streptococcus agalacitae | 2.9 | No | No | No |
| Escherichia coli | 1.4 | Yes | Yes (3.94) | Yes (3.94) |
| Enterococcus faecalis | 0.7 | Yes | Yes (5.41) | Yes (5.34) |
| Pseudomonas aeruginosa | 0.3 | Yes | Yes (0.68) | Yes (0.68) |
| Acinetobacter baumannii | 0.1 | Yes | Yes (0.65) | Yes (0.58) |
a
Boldface indicates a species from which sequence was positively detected, and for which probes were specifically designed.
TABLE 3
| Species present in ATCC MSA-1005 | % of mixture | Probes designed? | Genus reads detected (% of total reads) | Species reads detected (% of total reads) |
|---|---|---|---|---|
| Acinetobacter johnsonii | 16.7 | No | Yes (52.55) | Yes (32.89) |
| Corynebacterium striatum | 16.7 | No | No | No |
| Micrococcus luteus | 16.7 | No | No | No |
| Cutibacterium acnes | 16.7 | No | No | No |
| Staphylococus epidermidis | 16.7 | No | No | No |
| Streptococcus mitis | 16.7 | No | No | No |
Performance validation for microbial taxonomic identification
Assessment of read-level taxonomic assignment was performed for pre-defined, experimentally spiked validation samples containing either A. baumannii, P. aeruginosa, or a mixed microbial reference spiked into 50 ng of human control reference background DNA (14,500 copies). A. baumannii and P. aeruginosa were spiked at a range of 10–100,000 copies in 10-fold titrated quantities. From metagenomic sequences of wound-effluent clinical specimens (2), the ratio of Acinetobacter and Pseudomonas classified reads to human reads are on average, 0.007 ± 7 × 10−4 and 0.0009 ± 2 × 10−5, respectively. At spike-ins of 10 copies, the molecular mass ratio of Acinetobacter baumannii and Pseudomonas aeruginosa to human gDNA is 8.78 × 10−7 and 1.47 × 10−6, respectively, which would correspondingly be reflective of very low bioburden conditions. At spike-ins of 100,000 copies, the mass ratio of Acinetobacter baumannii and Pseudomonas aeruginosa to human gDNA is 0.00878 and 0.0147, respectively. Therefore, the tested copy number spike-ins fit within the range of expected pathogenic bioburden in clinical samples while also allowing for evaluation of the panel’s detection limits.
In the single microbial spike-in samples, sequence reads were correctly detected at the genus level for all tested copy numbers (Fig. 1A). No sequence reads were assigned to microbial genera other than Acinetobacter and Pseudomonas. However, a small number of reads were assigned to other Acinetobacter species in A. baumannii spike-in samples at 100,000 copies. These results suggest that genus and species-level detection of microbes by the panel is sensitive to at least 10 genomic copies of A. baumannii and P. aeruginosa in the presence of human background gDNA, while the specificity of A. baumannii probes was reduced at high copy numbers.
Fig 1
The performance of the targeted capture sequencing panel was further evaluated using a polymicrobial pathogen mix of genomic DNA. The polymicrobial pathogen mixture was spiked at 100,000 copies to simulate a high bioburden load. Each species was present at different abundances as indicated by the manufacturer (Table 2). The mixture included several previously determined wound-associated pathogenic species for which probes were designed for capture. The panel did not include capture probes for the detection of Neisseria and Streptococcus species. Using the targeted sequencing panel, reads corresponding to the correct genus and species were detected for all microbes for which probes were designed, even for species that were present at low levels. However, reads were assigned to Neisseria (1.04% of total reads) and Neisseria meningitidis (0.24% of total reads) despite no designed capture probes for these taxa.
Taxonomic assignment of the captured sequence was also examined with gDNA mixtures from six skin commensal microorganisms (Table 3). The commensal mixtures were spiked at 10,000 copies to simulate commensal background whereby reads were only assigned to Acinetobacter johnsonii. These results suggest the possibility of shared target regions between A. baumannii and A. johnsonii, thereby contributing to A. johnsonii detection. Further analyses revealed that 12 A. baumannii-specific probes (out of 1,031 probes) have an alignment identity score of ≥80% to at least two A. johnsonii RefSeq genomes, suggesting possible capture of A. johnsonii gDNA with those probes (Table S1). Aside from A. johnsonii, the targeted panel did not detect a pathogen signal from otherwise normal, commensal skin microflora present in the skin polymicrobial sample.
Performance validation for virulence signature detection
In addition to taxonomic assessment, each of the reference genome spike-in samples was analyzed for virulence detection. Detection of virulence genes in control genomes was compared to the expected detection based on in silico analyses of exact reference sequences of genomes found in the polymicrobial mix. Correct virulence gene calls were made from samples containing as few as 10 genome copies, with an increasing number of positive hits at higher spike-in copy numbers (Table 4). At 100,000 copies, 100% (46/46) of expected genes was detected for A. baumannii, while 97% (103/106) of expected genes was detected in P. aeruginosa. There were instances whereby genes absent in in-silico reference genomes were experimentally detected.
TABLE 4
| Sample copy number | No. of genes expected based on reference | No. of genes detected and present in reference | No. of genes detected but not present in reference | No. of genes undetected but present in reference |
|---|---|---|---|---|
| Acinetobacter baumannii | ||||
| 10 | 46 | 40 | 0 | 6 |
| 100 | 46 | 44 | 2 | 2 |
| 1,000 | 46 | 45 | 2 | 1 |
| 10,000 | 46 | 45 | 1 | 1 |
| 100,000 | 46 | 46 | 3 | 0 |
| Pseudomonas aeruginosa | ||||
| 10 | 106 | 18 | 3 | 88 |
| 100 | 106 | 54 | 1 | 52 |
| 1,000 | 106 | 97 | 4 | 9 |
| 10,000 | 106 | 99 | 7 | 7 |
| 100,000 | 106 | 103 | 8 | 3 |
| Polymicrobial pathogen mix | ||||
| 524 | 506 | 3 | 18 | |
| Polymicrobial skin mix | ||||
| 7 | 4 | 1 | 3 | |
a
Comparisons are made between experimentally detected genes to genes expected in the reference sequence of the corresponding strain via informatic analysis.
Virulence specificity was also assessed on targeted capture sequence data obtained from the separate polymicrobial mixed gDNA of skin pathogens and commensals (Fig. 2A; Table 4). The panel detected 510 virulence genes in the pathogen mixture. Based on the in silico reference genome analyses, the successful detection rate of expected virulence genes was 96.6% (506/524) (Table 4). The capture panels also detected three (allB, narH, and fliA) genes that were not expected from the in silico reference genomes (Fig. 2A). In silico screening of the polymicrobial skin commensal mix reference genomes resulted in seven virulence genes (lytB, pavA, pce_cbpE, sdrH, spaF, piuA, and eno). Sequence data of the skin commensal sample were mapped to five virulence genes (sdrH, piuA, sdrF, spaF, and eno), four of which were detected in silico (Table 4). As expected, the pathogen polymicrobial mix had a higher number of virulence detection events compared to the skin commensal mixture. The proportion of off-reference virulence hits over the total number of detected genes in the pathogen and skin commensal mixture is 0.006 (3/506 genes) and 0.2 (1/5 genes), respectively (Table 4).
Fig 2
Performance validation for AMR signature detection
Signatures of AMR genes were also assessed in single and polymicrobial reference samples by mapping reads to probes designed for the capture of AMR-associated gene sequences (Fig. 1B and C, Fig. 2B and C). Detection of AMR genes in reference genomes was assessed at varying coverage thresholds (50%, 75%, and 100%) as multiple probes were used per gene target sequence. Results were compared to expected detection based on reference sequence analysis. For A. baumannii, 11 out of 12 AMR genes predicted to be present in the reference were detected with as low as 100 genomic copies present, with only four AMR genes detected at 10 copies (Fig. 1B). While an expected blaADC-12 gene was not detected, a related beta-lactamase gene, blaAC1, was captured instead. A putative polymyxin-resistant gene, eptA, was detected at all thresholds despite being absent in the reference genome.
A minimum of 1,000 copies of P. aeruginosa gDNA was required for the detection of all four AMR-associated genes identified from its reference genome. One P. aeruginosa-associated AMR gene was detected at 10 copies, and none were detected at 100 genome copies (Fig. 1C). In particular, additional non-reference fluoroquinolone genes were detected, gyrA and parC. Similar to virulence gene detection, the number of positive gene calls increased with higher spike-in copy numbers of reference gDNA. Overall, the coverage threshold had minimal impact on gene calling, with a slight decrease in positive hits at a stringent threshold of 100%.
AMR-associated gene detection results were also assessed from the reference polymicrobial control mixtures of pathogenic and commensal species (Fig. 2B and C). At the highest coverage threshold of 100%, the panel detected 92.39% (85 out of 92) of expected genes in the pathogenic mixture (Fig. 2C). Thirteen additional genes [aac(6)-Iib, aad(A1), aad(A1b), aadA8b, blaADC1, cat(TC), cmlA, cmr(Ec), mac(B), mupA, parE (Ecloacae), penA, tet(A)] not found in silico were also detected. Analysis of skin commensal reference genomes identified the presence of eight AMR-associated genes [blaPC, blaZ, dfr(A), qacA/B, qacC/D, pbp2a, vga(A), vga(A)lc] (Fig. 2B). Four [blaPC, blaZ, dfr(A), pbp2a] out of eight of these genes were expected in silico, while three additional genes [gyrA-QRDR (Acineto), parC-QRDR (Vc), tet(K)] not anticipated to be present were detected. As anticipated, and in agreement with virulence detection results, a stronger AMR signal was observed from the pathogen mixture relative to the skin commensal mixture. The proportion of off-reference AMR hits over the total number of detected genes in the pathogen and skin commensal mixture is 0.15 (13/85 genes) and 0.42 (3/7 genes), respectively.
Detection of microbial signatures in wound samples and association with clinical outcomes
The targeted panel was applied to 58 effluent samples obtained from combat-injured service members across various sampling times: initial (earliest sample after arriving at the continental U.S.), intermediate, and final (pre-closure). This application allowed for the evaluation of panel utility for assessing polymicrobial infections in acute traumatic combat wounds. The number of samples collected from a single individual varied according to treatment progression. To facilitate consistent comparative analyses of functional genomic signatures, virulence genes were binned into 14 functional pathways based on the VFDB (21), and AMR-associated genes were binned into 22 antimicrobial categories to which they are anticipated to confer resistance towards (Fig. S2). Eighteen targeted microbial genera were detected across samples with Acinetobacter being the most frequently detected.
Hierarchical clustering was applied to assess whether targeted genomic signatures were associated with wound outcomes typically observed in clinical practice (Fig. 3). These outcomes were determined by clinicians at the time of sampling and two primary wound metrics were examined; overall wound closure outcome (successful delayed wound closure or failed healing, defined by requiring additional surgical debridement intervention after delayed wound closure was attempted) and critical colonization (bacterial load of >105 per g of tissue/µL fluid). The definition of critical colonization used here does not imply deleterious signs of infection or host response, but only total microbial quantification exceeding a defined quantity of microorganisms measured per unit of specimen investigated (22, 23). Final timepoint samples were used based on sample availability, according to the patient standard of care treatment, representing local wound microenvironments proximal to the time of delayed closure. Hierarchical clustering demonstrated that critically colonized samples harbored a high number of targeted microbial signatures, resulting in distinct clusters of specimens corresponding to colonized outcomes. The prevalence of these microbial targets was less exclusive in samples from wounds that failed delayed closure versus those that healed successfully (Fig. 3A). Multidimensional analyses of these samples also revealed distinct profiles between critically colonized and non-critically colonized samples (Fig. 3B and C). Overall, the panel detected 215 antimicrobial resistance genes, 383 virulence genes, and 12 genera in clinical specimens. These results demonstrate successful detection of highly relevant targeted microbial signatures from complex clinical specimens derived from combat wounds and the utility of targeted sequencing panels to evaluate bioburden profiles.
Fig 3
Random forest classification and feature importance permutation
Random forest models were developed using the abundance of either microbial genera, antimicrobial resistance genes, or virulence genes as features to classify critical colonization and overall wound outcome (Fig. 4). Model performance was robustly evaluated based on the area under the receiver operating characteristic curve (AUROC) from 100 random data splits (Fig. 4A). Models built using virulence and antimicrobial-resistance genes as features performed significantly better (FDR < 0.05) at predicting critical colonization compared to overall outcome across testing and training data sets. Subsequent feature selection based on random permutation revealed that Acinetobacter was important in classifying critically colonized samples, while Pseudomonas was important in distinguishing failed from healed wounds (Fig. 4B). These features were determined to be important based on a mean difference between the test and permuted AUROC that was greater than zero in 70% of the data splits. Additional signatures that were important in classifying critically colonized samples include a beta-lactam resistance gene, ampC, a polymyxin-resistant gene, eptA, and two virulence genes, hblA and adeG, which are associated with toxin and biofilm formation, respectively (Fig. 4B).
Fig 4
DISCUSSION
Previous reports indicate an estimated occurrence of wound infections in 18%–25% of combat-related injuries (24, 25). Extreme wound infections are associated with substantial morbidity with patients facing long-term complications and subsequent increased burden on wound care and healthcare systems (7). Impaired wound healing is also attributed to the presence of underlying co-morbidities and critical colonizing microorganisms (26, 27). Such delayed healing provides an opportunity for further shifts in the profile of contaminating microbes resulting in treatment difficulties (26). Thus, military healthcare professionals are continuously in need of improved methods and biomarkers that can be used to manage the treatment of combat wounds and improve diagnostics (4). Examples of such efforts include the use of autofluorescent imaging to detect bacterial load in traumatic wounds during debridement and the use of a rapid, label-free pathogen identification system to identify antimicrobial-resistant bacteria in the battlefield (28, 29).
However, combat wound management remains challenging due to the severity of traumatic wounds, simultaneous injury across body sites, and wound exposure to environmental contaminants. Often, combat wounds are prone to infectious complications caused by pathogenic and multidrug-resistant organisms (30). Therefore, microbiological signatures serve as important biomarkers for guiding wound therapy and portending wound outcomes. Standard microbiological screening in wound management involves plating sample swabs onto microbiological media to obtain visible isolates (31). While this method allows for the identification of known targeted microbes, their underlying pathogenic factors remain unknown without subsequent enrichment and further testing. Depending on the time of sampling and pathogen load, isolates might not be recoverable prior to the presentation of adverse complications. Conversely, the use of targeted sequencing platforms as described here is advantageous due to the potentially shorter turnaround time for rapid diagnostics, relative to traditional clinical microbiology techniques, and the ability to identify a wide range of potential functional attributes of detected pathogens.
Advancements in sequencing and informatics technologies have resulted in the development of high-resolution microbial profiling methods. While whole genome shotgun sequencing can capture infectious microbial signatures, it is limited by its sequencing depth, resulting in an inability to detect microbial targets that are low in abundance. To overcome this limitation, we described the design and implementation of a targeted sequencing panel to detect microbial signatures for clinical diagnostic applications specific to wound care. The advantages of this panel include (i) low limit of detection, (ii) high specificity and sensitivity, (iii) high coverage of target regions, (iv) enhanced depth for the analysis of gene variation, and (v) low cost and analysis burdens relative to untargeted whole genome sequencing.
Microbial targets were carefully curated based on consultation with clinical experts and previously derived data (15, 19). We first demonstrated that the panel could detect relevant and specific taxa, AMR genes, and virulence genes across a wide range of concentrations in the presence of other bacterial species and host DNA. In the polymicrobial mixtures, the targeted species and genes were still identified with high sensitivity and specificity, even though the overwhelming quantity of total DNA material corresponded to non-microbial and non-target sequences. In particular, the panel was able to detect the presence of two major wound-infecting species, A. baumannii and P. aeruginosa, even at low concentrations in the presence of host material. Previous reports have highlighted the importance of monitoring these species as they harbor multidrug-resistant and virulent properties (25, 32, 33). The sensitivity of detection at the gene level was lower than at the genus and species level for these species, requiring at least 100 genome copies of A. baumannii and 1,000 genome copies of P. aeruginosa.
At high skin commensal mixture spike-in levels, we did observe the detection of untargeted A. johnsonii. This was attributed to the hybridization of A. johnsonii genomic DNA to A. baumannii-specific probes. While unexpected, the detection of A. johnsonii at high concentrations could be beneficial as previous reports have shown that A. johnsonii is capable of causing hospital-acquired infections, possibly harboring antibiotic-resistance genes as a result of horizontal gene transfer events (34). Similarly, the detection of Neisseria species in the pathogenic mix is likely due to the elevated content of this species in the mixture. In further iterations, especially those intended for diagnostic use, such detection events could be filtered through application of a probe or read count threshold, according to the needs of any given application. Further investigation is warranted to determine probe specificity against other species that are not present in the polymicrobial mixes and will be the subject of future panel iterations. As AMR and virulence probes were designed for the capture of a comprehensive number of targets, including targets beyond the employed validation species, unexpected detection of off-reference events could be attributed to the capture of nucleotide sequences that were highly similar to probe targets. Noise generated during sequencing is also expected and can interfere with probe mapping events.
As expected, the screening of genomic virulence and antimicrobial signal in polymicrobial mixtures clearly indicates the specificity of the panel toward pathogen-derived signatures with minimal signal obtained from avirulent commensals. This is overall indicative of a low risk for false-positive detection events. We also observed a few instances of discrepancies between expected genes in silico and genes detected by the panel. This is likely a result of the inclusion of a comprehensive set of probes to detect AMR and virulence genes across species and alludes to opportunities for the discovery of genes with putative relevant functions.
We further applied the panel on military service member-derived combat wound specimens and evaluated the association of targeted microbial signatures with clinically relevant parameters. The primary assessed parameters included critical colonization (as defined by total bacterial load per unit specimen) and overall outcome of wound healing. Results from hierarchical clustering and multidimensional analyses indicate that samples from wounds with clinically detrimental outcomes yield an elevated quantity of pathogenic taxa, AMR genes, and virulence factors. Furthermore, critically colonized samples were observed to share similar microbial profiles, suggesting that the targeted microbial signatures could be potential predictors of risk from future infections. Our previous efforts suggest that wounds with adverse outcomes tend toward demonstrating reduced microbial diversity, as assessed via metagenomics (2). If microbial profiles converge toward a consistent set of taxonomic and functional features reflecting critical colonization and/or failed healing, detection of such features could be clinically informative, and optimization of the described panel toward this end will be the subject of future study.
The random forest models were also able to better classify the presence or absence of critical colonization compared to wound failure outcomes. Other important contributors include a beta-lactamase gene, ampC and a biofilm-producing gene, adeG. Previous studies have reported high prevalence of ampC-producing A. baumannii and P. aeruginosa isolates in burn wounds, which drive beta-lactam resistance, posing potential challenges in administering effective treatments, increasing risk of outgrowth of resistant microbial sub-populations, and prolonging hospital stays (35, 36). In addition, a study on chronic burn wound infections discovered an association between biofilm-producing genes such as adeG with antibiotic resistance and subsequent prolonged persistence of A. baumannii infection (37). As critical colonization events could lead to infection with downstream impacts on inflammation and wound resolution, the ability to use these signatures as biomarkers to predict such outcomes will help inform clinical decision support. Indeed, a larger sample size will be required to confidently profile critically colonized wound-associated microbiomes.
The design and integration of the targeted sequencing panel described here have multiple far-reaching implications. First, such implementation will have an immediate and profound impact in the management of combat wound infection by supporting comprehensive assessment of wound bioburden. The ability to predict clinically relevant outcomes with such targeted resolution can critically influence decisions to improve outcomes in combat injuries. Consequently, the corresponding interventions would have the potential to reduce morbidity, shorten hospital stays, and improve rehabilitation for service members with traumatic injuries or other conditions associated with microbial infections. Furthermore, we demonstrated the panel’s utility on effluent samples, which are relatively easier to collect in a clinical setting compared to tissue samples.
The technology described here also has broad societal impacts as infections resulting from multidrug-resistant nosocomial pathogens represent a recurrent and tremendous burden on the U.S. healthcare system (38 – 40). The microbial signatures selected for inclusion in the panel are relevant to infections treated in civilian settings such as those derived from non-healing diabetic ulcers (41). This panel could, therefore, be applicable to the surveillance of a broad range of infections according to gene specificity and further validation. In this study, panel validation was limited to the use of available wound-effluent samples. Future iterations should include the use of wound swabs for comprehensive profiling of wound microbiomes. It is also important to note that while the panel provides a quicker and extensive expansion to culturing methods, a 24 hour time window is currently ideal for sample preparation, processing, and analysis. Furthermore, while wound profiling is important for monitoring and predicting potential outcomes, the presence of biomarkers may not be indicative of infection and other related symptoms. Lastly, DNA-based detection assays as described here do not imply the presence of viable microorganisms or microbial metabolic activity. This is a limitation shared by metagenomic sequencing and existing qPCR-based approaches for pathogen detection and characterization. Future efforts to infer comprehensive metabolic activity based on sequence could be evaluated through RNA-based methods. Given the historical and near irreproducible nature of these unique specimens, RNA-based and phenotypic assessment was not feasible within the current study; however, these results will inform and facilitate comparable evaluation in civilian and laboratory-based studies.
Currently, evaluation of specific infectious parameters related to species identity, virulence, and antimicrobial resistance simultaneously is challenging and often not readily available. The findings described in this manuscript lay the foundation for the expansion of the utility of targeted sequencing throughout both military and civilian healthcare diagnostic infrastructures with the goal of improving care through precision approaches. Data obtained from such panels can also be used to survey community-wide gene variations and their impact on clinical decision-making, which is particularly important for accurate assessment of antimicrobial susceptibility profiles (42, 43). Therefore, future efforts should consider similar targeted panels for enabling gene- and mutation-level analyses in rapid point-of-care applications. In addition, approaches for gene quantitation from panels that account for gene size and gene copy number should be considered for diagnostic utility. It is possible that amplification of the assessed feature space with such quantification could improve the performance of future downstream models. Overall, the ability to detect relevant microbial signatures at the sensitivity demonstrated here could allow for early detection of clinically impactful factors, facilitating a more precise and individualized approach to patient care.
MATERIALS AND METHODS
Reference microbial genomes employed for panel testing
All reference microbial genomes were obtained from the ATCC. Single reference genomes include A. baumannii (17978D-5) and P. aeruginosa (27853D-5). The polymicrobial pathogen standard (ATCC MSA-4000) comprised Acinetobacter baumannii, Enterococcus faecalis, Escherichia coli, Klebsiella pneumoniae, Neisseria meningitidis, Pseudomonas aeruginosa, Staphylococcus aureus, Streptococcus agalactiae, Streptococcus pneumoniae, and Streptococcus pyogenes. The polymicrobial skin commensals standard (ATCC MSA-1005) comprised Acinetobacter johnsonii, Corynebacterium striatum, Micrococcus luteus, Cutibacterium acnes, Staphylococcus epidermidis, and Streptococcus mitis. All reference genomes were spiked into 50 ng of human reference control gDNA (ATCC CRL-2322).
Target identification and panel design
A list of 12 bacterial and 10 fungal species that are associated with wound healing was identified by consulting experts in combat wound management and literature review specific to combat extremity wounds (2, 22, 44, 45) (Table 1). Probes for these relevant species were then obtained from a previously designed high-throughput pathogen microarray (18, 19). These probes were extended in either direction against reference genomes to satisfy Illumina’s length requirements of 80 base pairs.
Virulence targets were identified from the virulence factor database (25 October 2018) whereby maximally conserved regions were located for all variants of each gene through multiple sequence alignments (MSAs) to create a consensus sequence. MSAs are generated using muscle 3.8 (46) and are readily available from VFDB (20). Different dominant base frequency thresholds were iteratively tested, and the highest frequency threshold yielding at least one region of a predetermined window size (free of indeterminate bases) was used to generate a consensus sequence. The window size was set to 80 corresponding to the exact length of probes on the panel. When conserved regions could not be identified due to a large number of SNPs, a phylogenetic approach was used to identify two consensus sequences per gene. The Phylo package from Biopython was used to transform each MSA into a phylogeny tree using the neighbor-joining method. The tree was rooted at its midpoint and used to split the MSA into two sequence clusters. Subsequently, if consensus generation with at least one 80-mer window failed, one variant of each gene was randomly selected for probe design. The phylogenetic approach was used to split MSA built from 460 genes. Of those 460 genes, 225 genes did not have a consensus sequence and a variant was selected for probe design. Probes for virulence targets were then designed from consensus sequences using compact aggregation of targets for comprehensive hybridization (47). Parameters were set to probe length of 80, coverage of 0.01, and mismatch of 0 with the identify flag specified.
AMR targets were selected from a previously designed amplification-based targeted sequencing platform for the detection of AMR genes in the International Space Station (15). This list of 518 genes was modified to incorporate genes that were experimentally shown to confer antimicrobial resistance, and a final list of 702 AMR genes was used for hybridization capture probe design by Illumina’s Sequencing Assay Designer. For a subset of genes that harbored multifunctional domains, a gene-tiling approach was used to identify capture regions. For genes that did not require tiling, a minimum of two probes were selected for each gene whereby probe targets were spread apart across the gene.
Identification of genomic loci in reference microbial species
Reference genomes in single and polymicrobial controls were screened for virulence and AMR genomic targets from the designed panel to create a “ground truth” table, indicating expected genes in validation samples. In rare cases where the sequence for the exact strain was not available, a closest neighbor strain was used. Given the potential for gene variants between strains, capture target regions were used for mapping instead of strain reference sequence as target regions were designed to be conserved and gene-specific. For AMR genes, probe mapping was performed via BLAST, as regions corresponding to AMR signatures were longer (~275 bp) than the probes themselves (80 bp). Virulence mapping was performed at the probe level via bowtie to obtain improved specificity for shorter sequences (80 bp).
Experimental testing of panel employing reference microbial species
Reference genomes for both A. baumannii and P. aeruginosa at 10, 100, 1,000, 10,000, and 100,000 genomic copies were spiked into 50 ng of human reference control background. Mixed polymicrobial controls containing either pathogen (100,000 copies) or skin commensal (10,000 copies) microbe gDNA were similarly spiked into 50 ng of reference human background. Samples were subjected to hybridization capture and targeted sequencing. Libraries were prepared according to the Illumina Nextera Flex for Enrichment workflow, target sequences were captured using the custom panel, and sequencing was performed (Illumina NextSeq 500). The Illumina Nextera Flex for Enrichment workflow uses bead-bound transposons for tagmentation, biotinylated probes for hybridization and capture, and streptavidin magnetic beads for enrichment. PCR is used for amplification of the enriched library.
Sequence data alignment and analysis
Sequence data were analyzed via several pipelines. Reads for each sample were aligned to panel probe targets corresponding to either genus/species, virulence, and AMR content using either the capture probe sequence (for genus/species and virulence) or the full gene target sequence (for AMR). This alignment was performed via BLAST for genus/species and virulence targets and via the BWA aligner for AMR gene targets.
For genus/species targets, reads were analyzed using the Livermore Metagenomics Analysis Toolkit, a previously developed LLNL pipeline for scalable metagenomic classification of DNA sequence (48), which assigns reads to appropriate taxonomic identifiers. To assign detection events for AMR, the percentage of the normalized gene length covered at >20× sequence read coverage was examined. A gene was called as detected if >50% of the normalized gene length was found to be covered; however, thresholds of 75% and 100% were also tested. For purposes of this analysis, the normalized gene length was defined as (probe count × 80 bp). The percentage of the normalized length covered could exceed 100%, as probes can capture regions outside their 80 bp length.
To assess the specificity of probes designed for the detection of A. baumannii, 1,031 A. baumannii-specific probes were aligned across 46 A. johnsonii Refseq genomes using bbsplit in the bbmap tool suite (49). Probes with an alignment identity of ≥80% against two or more A. johnsonii genomes were reported along with the number of genome hits (Table S1).
Application of targeted sequencing panel to combat injury wound samples
Wound specimens were previously collected from injured U.S. service member patients. This study was performed in compliance with all federal regulations governing the protection of human subjects and informed consent and was approved by the Institutional Review Boards of Walter Reed National Military Medical Center and Lawrence Livermore National Laboratory.
Genomic DNA from 58 wound-effluent samples was extracted as previously described (2). Library preparation and capture were performed using the Nextera Flex for Enrichment protocol (Illumina) and 2 × 150 bp sequencing was performed (Illumina NextSeq 500). Pre-processing of the resultant sequence data was performed as described above to identify the presence of microbial signatures.
Statistical analyses and random forest models
Downstream statistical analyses and visualization were performed in R (v4.2.0) with the following packages: readxl (50), ggplot2 (51), reshape2 (52), caret (53), ranger (54), tidyverse (55), pheatmap (56), ggsci (57), vegan (58), ggpubr (59), and mikropml (60). Statistical significance was determined using the Wilcoxon rank-sum test for bivariable comparisons with ggpubr (59). Random forest models and feature permutation were implemented using mikropml (60) for each microbial feature set; taxa, AMR genes, and virulence genes, and for two clinically determined outcomes; critical colonization and overall wound outcome. Data sets were split into 100 unique 50:50 train/test splits using random seeds, and identical train/test splits were used for each microbial feature set. A permutation importance test was carried out to identify the top contributing features used for classification in the random forest models.
ACKNOWLEDGMENTS
The authors wish to thank Dr. Gary Vora, Ph.D., for his intellectual contributions to the antimicrobial resistance gene content of the designed panel.
This study was supported by the Department of Defense U.S. Army Medical Research and Development Command through the Accelerating Innovation in Military Medicine program (Award No. CDMRPL-18-0-DM171034), and by the Lawrence Livermore National Laboratory’s Laboratory Directed Research and Development Program. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. LLNL Disclaimer: This document was prepared as an account of work sponsored by an agency of the United States government. Neither the United States government nor Lawrence Livermore National Security, LLC, nor any of their employees make any warranty, expressed or implied, or assume any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represent that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States government or Lawrence Livermore National Security, LLC. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States government or Lawrence Livermore National Security, LLC, and shall not be used for advertising or product endorsement purposes. USU-WRNMMC Surgery and HJF Disclaimer: The contents of this manuscript are the sole responsibility of the author(s) and do not necessarily reflect the views, opinions, or policies of Uniformed Services University of the Health Sciences (USUHS), The Henry M. Jackson Foundation for the Advancement of Military Medicine, Inc., the Department of Defense (DoD) or the Departments of the Army, Navy, or Air Force. Mention of trade names, commercial products, or organizations does not imply endorsement by the U.S. Government.
SUPPLEMENTAL MATERIAL
Supplemental figures and tables - spectrum.02520-23-s0001.pdf
Fig. S1 and Fig. S2; Table S1.
- Download
- 126.31 KB
Supplemental data set - spectrum.02520-23-s0002.xlsx
Lists of antimicrobial resistance genes and virulence factors as described in the manuscript.
- Download
- 84.85 KB
ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.
REFERENCES
1.
Be NA, Allen JE, Brown TS, Gardner SN, McLoughlin KS, Forsberg JA, Kirkup BC, Chromy BA, Luciw PA, Elster EA, Jaing CJ. 2014. Microbial profiling of combat wound infection through detection microarray and next-generation sequencing. J Clin Microbiol 52:2583–2594.
2.
Avila-Herrera A, Thissen JB, Mulakken N, Schobel SA, Morrison MD, Zhou X, Grey SF, Lisboa FA, Unselt D, Mabery S, Upadhyay MM, Jaing CJ, Elster EA, Be NA. 2022. Metagenomic features of bioburden serve as outcome indicators in combat extremity wounds. Sci Rep 12:13816.
3.
Forsberg JA, Potter BK, Wagner MB, Vickers A, Dente CJ, Kirk AD, Elster EA. 2015. Lessons of war: turning data into decisions. EBioMedicine 2:1235–1242.
4.
Hahm G, Glaser JJ, Elster EA. 2011. Biomarkers to predict wound healing: the future of complex war wound management. Plast Reconstr Surg 127:21S–26S.
5.
Hawksworth JS, Stojadinovic A, Gage FA, Tadaki DK, Perdue PW, Forsberg J, Davis TA, Dunne JR, Denobile JW, Brown TS, Elster EA. 2009. Inflammatory biomarkers in combat wound healing. Ann Surg 250:1002–1007.
6.
Hospenthal DR, Crouch HK, English JF, Leach F, Pool J, Conger NG, Whitman TJ, Wortmann GW, Robertson JL, Murray CK. 2011. Multidrug-resistant bacterial colonization of combat-injured personnel at admission to medical centers after evacuation from Afghanistan and Iraq. J Trauma 71:S52–7.
7.
Tribble DR, Murray CK, Lloyd BA, Ganesan A, Mende K, Blyth DM, Petfield JL, McDonald J. 2019. After the battlefield: infectious complications among wounded warriors in the trauma infectious disease outcomes study. Mil Med 184:18–25.
8.
Sung JS, Chong HY, Kwon N-J, Kim HM, Lee JW, Kim B, Lee SB, Park CW, Choi JY, Chang WJ, Choi YJ, Lee SY, Kang EJ, Park KH, Kim YH. 2017. Detection of somatic variants and EGFR mutations in cell-free DNA from non-small cell lung cancer patients by ultra-deep sequencing using the ion AmpliSeq cancer hotspot panel and droplet digital polymerase chain reaction. Oncotarget 8:106901–106912.
9.
Kuznetsova MV, Trofimov DY, Shubina ES, Kochetkova TO, Karetnikova NA, Barkov IY, Bakharev VA, Gusev OA, Sukhikh GT. 2017. Two novel mutations associated with ataxia-telangiectasia identified using an ion AmpliSeq inherited disease panel. Front Neurol 8:570.
10.
Bailie C, Kilner J, Maxwell AP, McKnight AJ. 2017. Development of next generation sequencing panel for UMOD and association with kidney disease. PLoS One 12:e0178321.
11.
Schobel SA, Stucker KM, Moore ML, Anderson LJ, Larkin EK, Shankar J, Bera J, Puri V, Shilts MH, Rosas-Salazar C, Halpin RA, Fedorova N, Shrivastava S, Stockwell TB, Peebles RS, Hartert TV, Das SR. 2016. Respiratory syncytial virus whole-genome sequencing identifies convergent evolution of sequence duplication in the C-terminus of the G gene. Sci Rep 6:26311.
12.
Mbala-Kingebeni P, Pratt CB, Wiley MR, Diagne MM, Makiala-Mandanda S, Aziza A, Di Paola N, Chitty JA, Diop M, Ayouba A, Vidal N, Faye O, Faye O, Karhemere S, Aruna A, Nsio J, Mulangu F, Mukadi D, Mukadi P, Kombe J, Mulumba A, Duraffour S, Likofata J, Pukuta E, Caviness K, Bartlett ML, Gonzalez J, Minogue T, Sozhamannan S, Gross SM, Schroth GP, Kuhn JH, Donaldson EF, Delaporte E, Sanchez-Lockhart M, Peeters M, Muyembe-Tamfum J-J, Alpha Sall A, Palacios G, Ahuka-Mundeke S. 2019. Ebola virus disease outbreak in équateur province, democratic republic of the Congo: a retrospective genomic characterisation. Lancet Infect Dis 19:641–647.
13.
Park J, Jang W, Kim M, Kim Y, Shin SY, Park K, Kim MS, Shin S. 2018. Molecular drug resistance profiles of mycobacterium tuberculosis from Sputum specimens using ion semiconductor sequencing. J Microbiol Methods 145:1–6.
14.
Nelson MI, Stucker KM, Schobel SA, Trovão NS, Das SR, Dugan VG, Nelson SW, Sreevatsan S, Killian ML, Nolting JM, Wentworth DE, Bowman AS, Schultz-Cherry S. 2016. Introduction, evolution, and dissemination of influenza A viruses in exhibition swine in the United States during 2009 to 2013. J Virol 90:10963–10971.
15.
Urbaniak C, Sielaff AC, Frey KG, Allen JE, Singh N, Jaing C, Wheeler K, Venkateswaran K. 2018. Detection of antimicrobial resistance genes associated with the international space station environmental surfaces. Sci Rep 8:814.
16.
Hamidian M, Maharjan RP, Farrugia DN, Delgado NN, Dinh H, Short FL, Kostoulias X, Peleg AY, Paulsen IT, Cain AK. 2022. Genomic and phenotypic analyses of diverse non-clinical Acinetobacter baumannii strains reveals strain-specific virulence and resistance capacity. Microb Genom 8:000765.
17.
Chen Y, Yang R, Guo P, Liu P, Deng J, Wu Z, Wu Q, Huang J, Liao K. 2023. Dynamic evolution of ceftazidime-avibactam resistance due to interchanges between bla(KPC-2) and bla(KPC-145) during treatment of Klebsiella pneumoniae infection. Front. Cell. Infect. Microbiol 13:1244511.
18.
Jaing CJ, Thissen JB, Gardner SN, McLoughlin KS, Hullinger PJ, Monday NA, Niederwerder MC, Rowland RRR. 2015. Application of a pathogen microarray for the analysis of viruses and bacteria in clinical diagnostic samples from pigs. J Vet Diagn Invest 27:313–325.
19.
Thissen JB, Be NA, McLoughlin K, Gardner S, Rack PG, Shapero MH, Rowland RRR, Slezak T, Jaing CJ. 2019. Axiom microbiome array, the next generation microarray for high-throughput pathogen and microbiome analysis. PLOS ONE 14:e0212045.
20.
Chen L, Zheng D, Liu B, Yang J, Jin Q. 2016. VFDB 2016: hierarchical and refined dataset for big data analysis-10 years on. Nucleic Acids Res 44:D694–7.
21.
Liu B, Zheng D, Zhou S, Chen L, Yang J. 2022. VFDB 2022: a general classification scheme for bacterial virulence factors. Nucleic Acids Res 50:D912–D917.
22.
Lisboa FA, Forsberg JA, Brown TS, Gage FA, Potter BK, Elster EA. 2013. Bilateral lower-extremity amputation wounds are associated with distinct local and systemic cytokine response. Surgery 154:282–290.
23.
Shiota J, Ohura N, Higashikawa S, Yamato T, Kasahara H, Itatani K, Tagawa H. 2016. Presepsin as a predictor of critical colonization in CLI hemodialysis patients. Wound Repair Regen 24:189–194.
24.
Tribble DR, Conger NG, Fraser S, Gleeson TD, Wilkins K, Antonille T, Weintrob A, Ganesan A, Gaskins LJ, Li P, Grandits G, Landrum ML, Hospenthal DR, Millar EV, Blackbourne LH, Dunne JR, Craft D, Mende K, Wortmann GW, Herlihy R, McDonald J, Murray CK. 2011. Infection-associated clinical outcomes in hospitalized medical evacuees following traumatic injury-trauma infectious disease outcome study (TIDOS). J Trauma 71:S33–42.
25.
Mende K, Akers KS, Tyner SD, Bennett JW, Simons MP, Blyth DM, Li P, Stewart L, Tribble DR. 2022. Multidrug-resistant and virulent organisms trauma infections: trauma infectious disease outcomes study initiative. Mil Med 187:42–51.
26.
Nahid MA, Griffin JM, Lustik MB, Hayes JJ, Fong KSK, Horseman TS, Menguito M, Snesrud EC, Barnhill JC, Washington MA. 2021. A longitudinal evaluation of the bacterial pathogens colonizing chronic non-healing wound sites at a United States military treatment facility in the Pacific region. Infect Drug Resist 14:1–10.
27.
Gjødsbøl K, Christensen JJ, Karlsmark T, Jørgensen B, Klein BM, Krogfelt KA. 2006. Multiple bacterial species reside in chronic wounds: a longitudinal study. Int Wound J 3:225–231.
28.
Chen Y, Chau J, Yoon J, Hladky J, Sousa C. 2022. Rapid, label-free pathogen identification system for multidrug-resistant bacterial wound infection detection on military members in the battlefield. PLOS ONE 17:e0267945.
29.
Blumenthal E, Jeffery S. 2018. Autofluorescence imaging for evaluating debridement in military and trauma wounds. Mil Med 183:429–432.
30.
Mende K, Stewart L, Shaikh F, Bradley W, Lu D, Krauss MR, Greenberg L, Yu Q, Blyth DM, Whitman TJ, Petfield JL, Tribble DR. 2019. Microbiology of combat-related extremity wounds: trauma infectious disease outcomes study. Diagn Microbiol Infect Dis 94:173–179.
31.
White BK, Mende K, Weintrob AC, Beckius ML, Zera WC, Lu D, Bradley W, Tribble DR, Schnaubelt ER, Murray CK, Infectious Disease Clinical Research Program Trauma Infectious Disease Outcome Study Group. 2016. Epidemiology and antimicrobial susceptibilities of wound isolates of obligate anaerobes from combat casualties. Diagn Microbiol Infect Dis 84:144–150.
32.
Tipton CD, Wolcott RD, Sanford NE, Miller C, Pathak G, Silzer TK, Sun J, Fleming D, Rumbaugh KP, Little TD, Phillips N, Phillips CD. 2020. Patient genetics is linked to chronic wound microbiome composition and healing. PLOS Pathog. 16:e1008511.
33.
Hawley JS, Murray CK, Griffith ME, McElmeel ML, Fulcher LC, Hospenthal DR, Jorgensen JH. 2007. Susceptibility of Acinetobacter strains isolated from deployed U.S. Antimicrob Agents Chemother 51:376–378.
34.
Montaña S, Schramm STJ, Traglia GM, Chiem K, Parmeciano Di Noto G, Almuzara M, Barberis C, Vay C, Quiroga C, Tolmasky ME, Iriarte A, Ramírez MS. 2016. The genetic analysis of an Acinetobacter johnsonii clinical strain evidenced the presence of horizontal genetic transfer. PLOS ONE 11:e0161528.
35.
Gupta V, Garg R, Garg S, Chander J, Attri AK. 2013. Coexistence of extended spectrum beta-lactamases, AmpC beta-lactamases and metallo-beta-lactamases in Acinetobacter baumannii from burns patients: a report from a tertiary care centre of India. Ann Burns Fire Disasters 26:189–192.
36.
Kumar V, Sen MR, Nigam C, Gahlot R, Kumari S. 2012. Burden of different beta-lactamase classes among clinical isolates of AmpC-producing Pseudomonas aeruginosa in burn patients: a prospective study. Indian J Crit Care Med 16:136–140.
37.
He X, Lu F, Yuan F, Jiang D, Zhao P, Zhu J, Cheng H, Cao J, Lu G. 2015. Biofilm formation caused by clinical Acinetobacter baumannii isolates is associated with overexpression of the AdeFGH efflux pump. Antimicrob Agents Chemother 59:4817–4825.
38.
Nelson RE, Stevens VW, Jones M, Khader K, Schweizer ML, Perencevich EN, Rubin MA, Samore MH. 2018. Attributable cost and length of stay associated with nosocomial gram-negative bacterial cultures. Antimicrob Agents Chemother 62:11.
39.
Stone PW. 2009. Economic burden of healthcare-associated infections: an American perspective. Expert Rev Pharmacoecon Outcomes Res 9:417–422.
40.
Khan HA, Ahmad A, Mehboob R. 2015. Nosocomial infections and their control strategies. Asian Pac J Trop Biomed 5:509–514.
41.
Schmidt BM, Erb-Downward J, Ranjan P, Dickson R. 2021. Metagenomics to identify pathogens in diabetic foot ulcers and the potential impact for clinical care. Curr Diab Rep 21:26.
42.
Liu F, Zhu Y, Yi Y, Lu N, Zhu B, Hu Y. 2014. Comparative genomic analysis of Acinetobacter baumannii clinical isolates reveals extensive genomic variation and diverse antibiotic resistance determinants. BMC Genomics 15:1163.
43.
Su M, Satola SW, Read TD. 2019. Genome-based prediction of bacterial antibiotic resistance. J Clin Microbiol 57:e01405-18.
44.
Radowsky JS, Brown TS, Lisboa FA, Rodriguez CJ, Forsberg JA, Elster EA. 2015. Serum inflammatory cytokine markers of invasive fungal infection in previously immunocompetent battle casualties. Surg Infect (Larchmt) 16:526–532.
45.
Brown TS, Hawksworth JS, Sheppard FR, Tadaki DK, Elster E. 2011. Inflammatory response is associated with critical colonization in combat wounds. Surg Infect (Larchmt) 12:351–357.
46.
Edgar RC. 2004. MUSCLE: a multiple sequence alignment method with reduced time and space complexity. BMC Bioinformatics 5:113.
47.
Metsky HC, Siddle KJ, Gladden-Young A, Qu J, Yang DK, Brehio P, Goldfarb A, Piantadosi A, Wohl S, Carter A, Lin AE, Barnes KG, Tully DC, Corleis B, Hennigan S, Barbosa-Lima G, Vieira YR, Paul LM, Tan AL, Garcia KF, Parham LA, Odia I, Eromon P, Folarin OA, Goba A, Viral Hemorrhagic Fever Consortium, Simon-Lorière E, Hensley L, Balmaseda A, Harris E, Kwon DS, Allen TM, Runstadler JA, Smole S, Bozza FA, Souza TML, Isern S, Michael SF, Lorenzana I, Gehrke L, Bosch I, Ebel G, Grant DS, Happi CT, Park DJ, Gnirke A, Sabeti PC, Matranga CB. 2019. Capturing sequence diversity in metagenomes with comprehensive and scalable probe design. Nat Biotechnol 37:160–168.
48.
Ames SK, Hysom DA, Gardner SN, Lloyd GS, Gokhale MB, Allen JE. 2013. Scalable metagenomic taxonomy classification using a reference genome database. Bioinformatics 29:2253–2260.
49.
Bushnell B. 2014. BBMap: a fast, accurate, splice-aware aligner. Available from: https://www.osti.gov/servlets/purl/1241166
50.
Wickham H, Bryan J, Kalicinski M, Valery K, Leitienne C, Colbert B. Package ‘readxl’. Version 13. 2019;1.
51.
Wickham H. 2011. ggplot2. WIREs Computational Stats 3:180–185.
52.
Wickham H. Reshape2: Flexibly reshape data: a reboot of the reshape package. R package version. 2017;1(3).
53.
Kuhn M. 2008. Building predictive models in R using the Caret package. J Stat Softw 28:1–26.
54.
Wright MN, Ziegler A. Ranger: a fast implementation of random forests for high dimensional data in C++ and R. arXiv preprint arXiv:150804409. 2015.
55.
Wickham H, Averick M, Bryan J, Chang W, McGowan L, François R, Grolemund G, Hayes A, Henry L, Hester J, Kuhn M, Pedersen T, Miller E, Bache S, Müller K, Ooms J, Robinson D, Seidel D, Spinu V, Takahashi K, Vaughan D, Wilke C, Woo K, Yutani H. 2019. Welcome to the Tidyverse. JOSS 4:1686.
56.
Kolde R. Package ‘Pheatmap'. R package. 2018;1(10).
57.
Xiao N, Li M, Xiao MN, LazyData T. Package ‘ggsci’. 2016.
58.
Dixon P. 2003. VEGAN, a package of R functions for community ecology. J Vegetation Science 14:927–930.
59.
Kassambara A. Package ‘ggpubr. R package version 01. 2020;6(0).
60.
Topçuoğlu BD, Lapp Z, Sovacool KL, Snitkin E, Wiens J, Schloss PD. 2021. mikropml: user-friendly R package for supervised machine learning pipelines. J Open Source Softw 6:61.
Information & Contributors
Information
Published In
Microbiology Spectrum
Volume 11 • Number 6 • 12 December 2023
eLocator: e02520-23
Editor: Paul M. Luethy, University of Maryland School of Medicine, Baltimore, Maryland, USA
PubMed: 37874143
Copyright
This is a work of the U.S. Government and is not subject to copyright protection in the United States. Foreign copyrights may apply.
History
Received: 16 June 2023
Accepted: 18 September 2023
Published online: 24 October 2023
Keywords
Data Availability
The data underlying this study are not publicly available due to sensitivities regarding their generation from injured military service member cohorts. Data from the corresponding author are available on reasonable request and in accordance with applicable regulations and data usage agreements.
Contributors
Editor
Paul M. Luethy
Editor
University of Maryland School of Medicine, Baltimore, Maryland, USA
Metrics & Citations
Metrics
Note:
- For recently published articles, the TOTAL download count will appear as zero until a new month starts.
- There is a 3- to 4-day delay in article usage, so article usage will not appear immediately after publication.
- Citation counts come from the Crossref Cited by service.
Citations
Citation
2023.
Targeted metagenomic assessment reflects critical colonization in battlefield injuries. Microbiol Spectr
11:e02520-23.
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. For an editable text file, please select Medlars format which will download as a .txt file. Simply select your manager software from the list below and click Download.