Methicillin-Resistant Staphylococcus aureus Bacteremia Relapses Show Diverse Genomic Profiles but Convergence in Bacteremia-Associated Genes

Abstract

Background

Recurrence of methicillin-resistant Staphylococcus aureus (MRSA) bacteremia is a high risk complication for patients. Distinguishing persistent lineages from new infections is not standardized across clinical studies.

Methods

We investigated factors contributing to recurrence of MRSA bacteremia among subjects in Philadelphia, Pennsylvania. Subject demographics and clinical history were collected and paired with whole-genome sequences of infection isolates. Recurrent bacteremia episodes were recorded and defined as relapse infections (same lineage) or new infections by genomic criteria, where a relapse contained isolates ≤ 25 single nucleotide polymorphisms (SNP) different, and by clinical criteria. All isolates were assessed for pairwise SNP distances, common mutations, and signatures of within-host adaptation using the McDonald-Kreitman test. Clusters of transmission between relapse-associated isolates and other subject lineages were identified.

Results

Among 411 sequential subjects with MRSA bacteremia, 32 experienced recurrent bacteremia episodes, with 24 subjects having exclusively relapse infections, 6 with infections exclusively from a new strain, and 2 patients with both relapse and new infections. No concordance between a genomic and a clinical definition of relapse was evident (Cohen κ = 0.18; confidence interval, −0.41). Recurrence-associated lineages exhibited signatures of positive selection (G test, < 0.01). Genes with SNPs occurring in multiple relapse lineages have roles in antibiotic resistance and virulence, including 5 lineages with mutations in mprF and 3 lineages with mutations in rpoB, which corresponded with evolved phenotypic changes in daptomycin and rifampin resistance.

Conclusions

Recurrent infections have a diverse strain background. Relapses can be readily distinguished from newly acquired infections using genomic sequencing but not clinical criteria.

Whole-genome sequencing differentiates recurrent MRSA bacteremia into new and relapsing infections, showing poor correlation with clinical criteria. Daptomycin and rifampin may select for resistant strains in relapsing infections. Relapse-associated strains may contribute to MRSA spread in health care settings.

Staphylococcus aureus bacteremia is a complex clinical syndrome with high mortality, which is often complicated by endocarditis and other metastatic infections [1]. Recurrence of S. aureus bacteremia, when patients are diagnosed with bacteremia again after assumed resolution of a previous infection, is an ongoing clinical challenge. Global data demonstrate that 5%–15% methicillin-resistant S. aureus bacteremia (MRSAB) episodes result in recurrence, and the risks associated with recurrence are similarly heterogenous to bloodstream infections overall [2–9]. Known risk factors for recurrence of S. aureus bacteremia in adults include younger patient age, presence of a foreign body, hemodialysis dependence, valvular heart disease, liver cirrhosis, and endocarditis [6, 9].

Distinguishing new infections from bacteria persisting within host tissues is challenging for clinicians, especially if the infection focus is unknown. Recurrence has been heterogeneously defined using strain genotypes (eg, spa typing or multilocus sequence typing [MLST]), time intervals, or clinical history of previous infections [4, 6–9]. Without including genomic similarity criteria, it has not been possible to distinguish relapse from new infections accurately. Consequently, misidentification of the 2 types of recurrence includes, for relapses, not identifying persistent infections, and for new infections, that a patient may have a new risk factor for S. aureus infection.

Colonization by S. aureus has been associated with higher risk of developing bacteremia. The transition from colonization to invasive infection, and the subsequent treatment regimen, can result in bacterial adaptive evolution [10–12]. Convergent mutations observed in MRSAB alter virulence regulation [13–25] and antibiotic targets [13–15, 26, 27], and result in the development of small-colony variants [28, 29]. Emerging mutations can confer cross-resistance to first-line antibiotic treatments or multidrug resistance, further complicating prevention of persistent infections in patients with deep-seated foci [10]. Several loci appear to increase the risk for persistent infections and possible recurrence, some of which suggest adaptation during bacteremia. However, it is not known which, if any, are associated with relapse versus new infection. Further information is also needed regarding the influence of strain background on forward evolution leading to relapse. Knowledge of genetic change patterns in S. aureus during bacteremia episodes may help differentiate new infections from relapses.

To define genetic differences and risks for MRSAB recurrence, we examined a cohort of patients in Philadelphia, Pennsylvania experiencing MRSAB during a 3.5-year period. We performed whole-genome sequencing (WGS) on single isolates from single and recurrent episodes, and split recurrent MRSAB into “relapses” caused by the same genetic lineage, and “new infections” in which a different S. aureus lineage was responsible. We examined the associations between host clinical factors and bacterial genetics to characterize recurrent MRSAB, and defined the transmission and adaptive history of infections that resulted in a relapse.

METHODS

Subject Cohort and Isolate Collection

This study was considered exempt by the University of Pennsylvania Institutional Review Board. Subject MRSAB isolates were included from a cohort of adult patients admitted to at least 1 of 2 hospitals in Philadelphia, Pennsylvania and diagnosed with MRSAB between July 2018 and February 2022. A single colony isolate of the first positive culture was stored for each MRSAB episode. Clinical and demographic information was collected through medical record reviews, including age at time of isolate collection, race, ethnicity, sex at birth, death within 30 days of infection diagnosis, comorbidities, MRSAB complications, antibiotic treatment, suspected source site of infection, and health care-associated epidemiological category. Antibiotic resistance phenotypes were collected from clinical microbiology records associated with the unique MRSA isolates. Antibiotic resistance was assessed using the Vitek 2 automated system, and among a subset of isolates the minimum inhibitory concentrations were determined using plate or broth dilutions, or E-tests. Susceptibility/resistance was assigned in accordance with Clinical and Laboratory Standards Institute protocols [30].

Clinical Distinction of Relapse and New Infections

Subsequent MRSAB events per subject (a recurrent episode) were classified as a relapse or new infection according to the following criteria, as outlined in Supplementary Figure 1: MRSAB episodes included all isolates from the same subject from the first known episode and for 30 days after. Per discrete MRSAB medical encounter, if subjects with 1 or more MRSA isolate from blood were cultured greater than 30 days after isolation of all previous MRSAB isolates during the study period, they were considered to have a recurrent episode. Recurrent episodes were then categorized separately into new infections and relapse infections using clinical criteria and genomic criteria. Clinically, subjects with new infections were considered cured of their previous infection and had to fulfill all of the following: the first positive culture from the episode was 30 days or more following the last positive MRSA blood culture; all symptoms at the source and metastatic sites of the previous infection were resolved; no new antibiotics were prescribed after completion of definitive therapy for the index MRSAB episode aside from antibiotics aimed at suppression of an infected focus; if the subject was administered suppressive antibiotics and the infection site was different from the previous infection; if a central venous catheter was exchanged at the time of the previous MRSAB, it was not exchanged over a wire but instead was resited [31]; and the infection source was clinically ruled different from the previous infection. Otherwise, if any 1 or more criteria were not true, the bacteremia episode was considered a relapse.

Sequencing Quality and Phylogenetics

Genomic DNA was extracted from S. aureus isolates and sequenced at the Penn/Children's Hospital of Philadelphia Microbiome Center, using paired-end short-read WGS as previously described [32]. Reads were processed using Bactopia (version 3.0.0) [33]. Adaptor sequences were removed, and reads were de novo assembled using Shovill (version 1.1.0). Sequences were characterized using Bactopia summary and included if they were bronze level and above: reads with an average per-read quality score of Q12, mean read length of 49 bp, a genomic assembly with 20 × coverage, and fewer than 500 contigs (Supplementary Table 1). Multilocus sequence type (ST) and clonal complex (CC) were assigned by calling Ariba (version 2.14.6) within Bactopia, which utilized the S. aureus-specific ST scheme from PubMLST [34]. For novel STs or STs without a defined CC, the CC was manually defined using a maximum likelihood (ML) tree and relative position to the next isolate with a defined CC. A core genome alignment was created among all MRSAB isolates and methicillin-susceptible S. aureus strain Newman (GCF_020985245.1) as an outgroup with the Bactopia subworkflow “pangenome” using PIRATE [35]. Areas of likely recombination were identified and masked using ClonalFrame ML (version 1.12) [36]. A ML tree of all isolates was created from the masked alignment with IQTree (version 2.1.2) [37], and ModelFinder [38] determined the best fit model was a generalized time reversible model with Empirical base frequencies plus the FreeRate model.

Genomic Criteria for Relapsed and New Infections and Cluster Identification

Snp-dists (version 0.8.2) was used to calculate single-nucleotide polymorphism (SNP) pairwise distances between aligned isolates. Subjects with recurrent MRSAB episodes were categorized as having a relapsed or new infection based on pairwise SNP differences from the isolate of the most recent previous episode relative to the next subsequent episode. A SNP threshold of ≤ 25 SNPs between episodes defined a relapse based on previous observations among persistent MRSA lineages [39]. The phylogenetic branch patterns were also investigated between recurrent episode isolates: relapse isolates must also share their most-recent common ancestor with another isolate from within the same subject and share the same ST. Cohen κ [40, 41] was calculated between the clinical and genomic relapse definitions to assess method agreement, sensitivity, and specificity.

Association of Clusters With Demographic and Clinical Data

Demographic and clinical characteristics were compared between subjects with recurrent MRSAB episodes and single MRSAB episodes using Kruskal-Wallis test for continuous variables and Pearson χ² test or Fisher exact test for categorical variables. Demographic and clinical history were first compared between MRSAB subjects who died within their index MRSAB episode window (30 days) and those who survived their index infection (Supplementary Table 2). Decedents from an index MRSAB episode were excluded from the analysis assessing risk factors associated with MRSAB recurrence.

Genomic similarity between isolates was also compared to identify potential transmission clusters. For all relapse-associated isolates ≤ 25 SNPs from different subjects' isolate genomes, subtrees of the larger ML tree were examined for branch structure and days since the earliest collection date in the cluster to identify the role of relapse isolates in putative onward transmission.

Variant Calling and Gene Detection in Relapse Lineages

Bactopia's Snippy subworkflow (Snippy version 4.6.0 [42]) was used to call SNPs of all isolates against a previously generated ancestrally reconstructed S. aureus genome [43]. Variant calling was conducted for each set of identified relapse infections by using the first known isolate (index) in each lineage as the reference. Index isolates were annotated using Bakta (version 1.9.3) [44]. The –snippy-core option was run on all isolates to identify variants in core coding regions across all genomes to produce an alignment and predicted changes to amino acid structure. To identify nonsynonymous changes within each recurrence lineage, the –snippy-core option was also used to create SNP alignments for each set of isolates from the same subject to identify common variants and changes to amino acid structure. SNPs in coding regions across recurrence lineages were concatenated and summarized. We also examined changes in the presence and absence of noncore genes between sequential relapse isolates to identify potential mobile genetic elements. We used gene families detected and their corresponding synteny groups from the Bactopia subworkflow “pangenome” PIRATE [35] and calculated the number of gene families that occurred variably (ie, a change from present to absent between sequential isolates) in relapse isolates over sequential bacteremia episodes. For gene families simultaneously newly present or absent, their synteny groups were compared.

Creating a Database of Bacteremia-Associated Mutations

A literature review was conducted to collect previously identified genes and/or mutations in S. aureus genomes associated with bacteremia. PubMed was searched in December 2023 using the query (((staphylococcus aureus)) AND ((bacteremia) OR (bloodstream infection))) AND (genetic mutation). Only peer-reviewed articles were scanned for evidence. Genes or mutations in genes were considered if the study reported that they occurred in the S. aureus genome, samples were derived from S. aureus bacteremia, and genes or mutations were associated with characteristics relevant to virulence expression, host survival, or phenotypic changes related to survival within the bloodstream. A listing of citations identifying these genes is in Supplementary Table 3.

Calculation of the Index of Neutrality

McDonald-Kreitman tests [45] were conducted on all MRSAB isolates and relapse-associated isolates to assess selection on the whole genome or on known bacteremia-associated genes. A core genome alignment against the ancestral reference was created using snippy-core for all isolates. Fixed sites were determined by counting the nucleotide sites universally conserved from the reference among all isolates or the subset of relapse isolates. Polymorphic sites were determined if a nucleotide site differed between individuals within the isolate groups, and statistically assessed with G tests.

RESULTS

Recurrent MRSAB Are From Common Clinical and Phylogenetic Backgrounds Among Circulating Strains

We included 456 S. aureus isolates from MRSAB episodes in 411 subjects, of whom 32 had at least 1 recurrent MRSAB episode in addition to the index infection (77 isolates). Recurrence-associated isolates occurred across major S. aureus CCs including CC5 (n = 29), CC8 (n = 41), CC30 (n = 3), CC78 (n = 3), and CC72 (n = 1) (Figure 1). CC distribution was similar between recurrence and nonrecurrence isolates (Fisher exact P = .90).

Figure 1.

Recurrent methicillin-resistant S. aureus bacteremia (MRSAB) isolates share a similar phylogenetic distribution to nonrecurrent MRSAB episodes. A core pangenome tree of 411 MRSA bloodstream isolates was constructed and rooted using methicillin-susceptible S. aureus strain Newman (GCF_020985245.1). Whether or not an isolate was associated with a recurrent episode and molecular characteristics are shown in the heat map including, from left to right, clonal complex, mutlilocus sequence type, the presence or absence of mecA, and SCCmec type. Blue dots indicate nodes where Shimodaira-Hasegawa approximate likelihood ratio test and ultrafast bootstrap values were > 70.

Surviving subjects (n = 311) were assessed for risk factors for recurrent MRSAB (Table 1). When subjects with recurrent episodes were compared to those without recurrence, we found no significant differences in clinical or demographic characteristics. Subjects epidemiologically categorized as having health care-associated community onset infections were more frequent among recurrent subjects (84%) compared to nonrecurrence patients (66%), although this was not statistically significant.

Table 1.

Clinical and Demographic Characteristics of 311 Subjects Surviving the First (Index) MRSAB Episode, Comparing Those With a Recurrent and Those Without a Recurrent MRSAB Episode

Patient Attribute	Overall (n = 311)	No Recurrent Episode (n = 279)	Recurrent Episode (n = 32)	P Value (No Recurrences vs Recurrence)
Patient demographics
Age at diagnosis, y, median (IQR)	54 (41–65)	54 (41–65)	52.5 (33–63)	.40a
Sex				.40c
Male	182 (58.5)	166 (59.5)	16 (50)
Female	129 (41)	113 (40.5)	16 (50)
Race				.21c
Asian	5 (2)	4 (1)	1 (3)
Black	131 (42)	121 (43)	10 (31)
White	146 (46)	128 (46)	18 (56)
More than 1 race	2 (< 1)	1 (< 1)	1 (3)
Other race	9 (3)	8 (3)	1 (3)
Unknown	18 (6)	17 (6)	1 (3)
Ethnicity				.64c
Hispanic or Latino	14 (4.5)	12 (4)	2 (6)
Not Hispanic or Latino	297 (95)	267 (96)	30 (94)
Patient comorbid conditions
Chronic skin diseased	32 (10)	29 (10)	3 (9)	1c
Diabetes mellitusd	105 (46)	92 (33)	13 (41)	.50b
Cancerd	40 (13)	37 (13)	3 (9)	.88c
Respiratory diseased	68 (22)	59 (21)	9 (28)	.58b
Cardiovascular diseased	148 (47.5)	125 (45)	19 (59)	.17b
Liver diseased	30 (10)	26 (9)	4 (12.5)	.58c
Kidney diseased	83 (27)	72 (26)	11 (34)	.41b
Hemodialysis in the last 12 md	59 (19)	51 (18)	8 (25)	.35c
Current intravenous drug used	74 (24)	67 (24)	7 (22)	1c
Complication of bacteremia
Infective endocarditisd	50 (16)	46 (16.5)	4 (12.5)	.82c
Epidural abscessd	18 (6)	15 (5)	3 (9)	.47c
Soft tissue abscessd	41 (13)	36 (13)	5 (16)	.63c
Osteomyelitisd	67 (21.5)	60 (21.5)	7 (22)	1c
Septic pulmonary embolid	41 (13)	40 (14)	1 (3)	.19c
Other metastatic complication(s)d	43 (14)	40 (14)	3 (9)	.59c
Involvement of foreign body in the bacteremiad	69 (22)	62 (22)	7 (22)	1c
Health care acquisition type of index infection				.14c
Community associated	38 (12)	36 (13)	2 (6)
Health care associated	61 (20)	58 (21)	3 (9)
Health care associated, community onset	212 (68)	185 (66)	27 (84)
Suspected source of index bacteremia episode				.71c
Arteriovenous graft	11 (4)	11 (4)	0 (0)
Central venous catheter infection	44 (14)	40 (14)	4 (13)
Device infection	8 (3)	7 (3)	1 (3)
Other	12 (4)	9 (3)	3 (9)
Respiratory source	17 (5)	16 (6)	1 (3)
Skin site	83 (27)	72 (26)	11 (34)
Surgical site	27 (9)	25 (9)	2 (6)
Urinary source	8 (3)	8 (3)	0 (0)
Unknown	101 (32)	91 (33)	10 (31)
Median No. of antibiotics administered during index MRSAB episode (IQR)	2 (1–2)	2 (1–2)	2 (1–3)	1a
Received vancomycind	295 (95)	264 (95)	31 (97)	1c
Vancomycin duration, d, median (IQR)e	28 (10.5–42)	28 (11–42)	28 (6–42)	.57a
Received daptomycind	123 (40)	105 (38)	18 (56)	.064b
Daptomycin duration, d, median (IQR)e	33 (12–42)	32.5 (12–42)	38 (21–42.5)	.43a

Data are No. (%) except where indicated.

Abbreviations: IQR, interquartile range; MRSAB, methicillin-resistant Staphylococcus aureus bacteremia.

^aKruskal-Wallis test.

^bχ² test.

^cFisher exact test.

^dCompared to respondents of the category without exposure.

^eAmong subjects with exposure.

New Infections Are Genetically Distinct From Relapse Infections Among Recurrent Episodes

Using genomic criteria we classified 24 subjects as having only relapses and 6 with infections only from a new strain; 2 subjects experienced both relapse and new infections. We identified 45 subsequent recurrent episodes among the 32 subjects. Relapse episodes occurred sooner after the prior infection episode (n = 37, median 97 days; interquartile range [IQR], 69–200 days) compared to new strain episodes (n = 8, median 311 days; IQR, 79–568 days), although this was not statistically significant (P = .17) (Figure 2A). Most relapses fell well below the 25 SNP threshold; only 3 episodes required additional review, which showed they were most closely related to an isolate from the same subject compared to all other isolates. All isolates classified as new infections were distantly related to the subject's previous infection and frequently belonged to a separate clade of S. aureus (Figure 2B).

Figure 2.

Relapsing infections and new infections within a patient are genomically distinguishable. A, Subsequent isolate pairs among recurrences episodes (n = 45) were separated into relapse-associated (n = 37) or new infections (n = 8) based on pairwise single nucleotide polymorphism (SNP) distance and distance to last common ancestor between isolates within the subject. Difference in time between subsequent episodes for genomically new infections and relapse associated infections were compared using the Kruskal-Wallice test (P = .17). B, Counts of the pairwise SNP distances between all isolate pairs from the same host were quantified (n = 73). The inset display demonstrates counts where the SNP distance between isolates was between 0 and 350 SNPs. The dotted line represents the 25 SNP threshold.

When clinical and genomic definitions were compared, the overall concordance was poor (Cohen κ = 0.18; 95% confidence interval, −0.41), with the genomic definition predicting that 82% of subsequent infections were related to the previous infection, and the clinical definition predicting that 50% were related (Figure 3). Compared to genomic relapse, identifying clinical relapse had a sensitivity of 55% and a specificity of 75%. When clinical outcomes of the index infection were compared between patients who uniquely experienced a relapse (n = 26) or a new infection (n = 6), relapse patients had longer hospital stays and were exposed to more antimicrobials (Supplementary Table 4). The genomic definitions of relapse and new infections were used for the remainder of analyses.

Figure 3.

Clinical and genomic definitions of relapse are discordant. Pairs of isolates from all recurrent infections were compared and identified as relapse (filled black rectangle) or new infections (white rectangles) based on a genomic definition or clinical definition, for a total of 45 pairs. The suspected source type of each infection within the episode was identified, with the least recent isolate associated with the “first source” and the most recent isolate associated with the “second source.” Clinical source type was additionally assessed to determine if the first and second source were anatomically from the same source.

Relapse Infections Demonstrate Distinct Adaptation to the Host

Evolutionary changes between relapse episodes may reveal unique underlying mechanisms for within-host adaptation. Therefore, we calculated neutrality indices for all isolates and relapse isolates. We grouped genes into bacteremia-associated and not bacteremia-associated and calculated separate neutrality indices for each group. Across all isolates, there was evidence of neutral evolution. Comparatively, relapse-associated lineages showed positive selection in the whole genome, and in genes classified as bacteremia-associated or nonbacteremia-associated (Table 2). Although few bacteremia-associated genes were synonymously mutated, there were no synonymous mutations at polymorphic sites for the relapse isolates. Together, this suggests that relapse-associated lineages likely undergo selection after the index infection. We also examined variable presence and absence of genes over the course of relapse episodes. Isolates in relapse infections showed variable changes in the presence or absence of genes between subsequent bacteremia episodes, and most genes gained or lost together over time did not exhibit patterns of shared synteny suggestive of large mobile genetic element shifts (Supplementary Table 5).

Table 2.

Neutrality Indices of All Methicillin-Resistant Staphylococcus aureus Bacteremia and Relapse-Associated Isolates at the Whole-Genome Level and for Bacteremia-Associated Genes

Group	Fixed, Nonsynonymous	Fixed, Synonymous	Polymorphic, Nonsynonymous	Polymorphic, Synonymous	McDonald-Kreitman Value	P Value, G Test
All isolates, whole genome	17275	24530	1451	1825	.88	< .01
All isolates, bacteremia genes	232	330	24	22	0.64	.153
All isolates, nonbacteremia genes	17043	24200	1427	1803	0.89	< .01
Relapse, whole genome	2344	4912	115	24	0.10	< .01
Relapse, bacteremia genes	26	70	10	0	…	…
Relapse, nonbacteremia genes	2318	4824	105	24	0.11	< .01

Bold text indicates a significant P Value for the G Test.

Antibiotic Resistance Genotypes and Phenotypes in Relapses Correspond With Patient Exposures

Eleven genes had unique SNPs with mutations in 2 or more relapse lineages (Figure 4A). Mutations in these genes occurred regardless of CC, indicating that clade background alone did not contribute to mutations in these genes. Genes in which multiple relapse lineages showed nonsynonymous mutations were implicated in known virulence traits and antibiotic resistance. The genes most commonly mutated were mprF, in 5 separate subject lineages, and rpoB, in 4 subject lineages. Changes in both mprF and rpoB are associated with drug resistance to antibiotics commonly used for bacteremia; we therefore investigated amino acid impact, isolate minimum inhibitory concentration, and treatment course. Multiple amino acid changes were detected between patients for proteins encoded by both genes. The same amino acid change, Ala477Asp, occurred for rpoB products in 2 subjects with rifampin resistance; for 1 subject the resistance phenotype was present in their first infection before relapse, a change at position 477 and position 471 (Asp471Tyr), which corresponded with a phenotypic change from susceptible to resistant (Figure 4B). Three subjects with mprF mutations emerging in relapses demonstrated acquired daptomycin resistance with changes at Ser337Thr, Ser337Leu, and Leu291Ile (Figure 4C). All recurrence-associated subjects with mprF mutations had exposure to daptomycin prior to mutation regardless of daptomycin resistance phenotype. No other relapse lineages gained or lost resistance to rifampin or daptomycin without emergence of mprF or rpoB mutations.

Figure 4.

Commonly mutated genes among relapse lineages are associated with antibiotic resistance phenotypes. A, Unique nonsynonymous (NS) mutations in coding regions of the genome were quantified by gene from isolates of relapse-associated infections and summarized relative to the subject from which the isolate was collected. Each unique mutation was annotated with the clonal background of the lineage from which the set of relapses were derived. For lineages with rpoB mutations (B) and mprF mutations (C), a timeline (days since index infection) was created for each set of relapsing infections by the subject experiencing that set of relapses. Individual episodes were annotated with the amino acid changes detected in the respective genes, the clinical assay used to assess minimum inhibitory concentration (MIC) and the corresponding MIC, and whether the patient was exposed to rifampin (RIF) (B) or daptomycin (DAP) (C). Dots are colored based on the clinical assay determination of drug susceptibility to RIF or DAP.

Relapse Isolates Cluster With Other Subject Isolates, But Do Not Contribute to Onward Transmission

Nine clusters across subjects including at least 1 subject who had a relapse were identified and occurred across CC5 and CC8 clades (Figure 5). If relapse infections were contributing to onward transmission, or if patients were reinfected with a closely related circulating strain, we might expect that infections in unique hosts would cluster within relapse lineage clades. Across all 9 clusters, isolates from different subjects clustered significantly outside of the relapse lineage isolates. This suggested that transmission occurred before the onset of relapsing infection and the unique lineages within a host were highly specific to the individual.

Figure 5.

Nonrelapse associated isolates cluster separately from relapse-associated isolates. A–I, A tree based on core genes was generated for each of the 9 nonrelapse isolate genome clusters using the unequal transition/transversion rate plus empirical base frequencies model to investigate branching positions. Clusters were investigated when at least 1 relapse-associated isolate genome differed by 25 SNPs or fewer from an isolate from a different subject. Subtrees were extracted based on the most-recent common ancestor shared by relapse subject isolates and the clustered additional subjects. Nodes denoted with a blue dot indicate ultrafast bootstrap values and Shimodaira-Hasegawa approximate likelihood ratio test values that are greater than 70. Bolded tip labels indicate an isolate that is part of a relapse. Tips are annotated with the patient subject IDs and the number of days at which the isolate was collected relative to the earliest isolate in the cluster. Branch lengths are scaled in substitutions per site.

DISCUSSION

Our study showed that recurrent MRSAB is well differentiated into new and relapse infections by WGS. Because consensus of a time and genomic definition of relapse is still not standard, it is imperative to document and compare genotypic and phenotypic patterns among recurrent isolates to better refine these molecular definitions. For infections during our investigation period most isolates from the same person were separated by fewer than 25 core genome SNPs. We defined 3 additional isolates from episodes as relapses upon inspection of their phylogenetic relatedness to intrahost isolates. Previous studies have identified a similar pattern in distinguishing relapses from new infections even using less-precise technologies, including pulsed-field gel electrophoresis, finding that approximately an equal number were classified as relapse and new infections [9]. Our SNP-based definition revealed a much larger percentage of relapse infections than previous studies would suggest. We observed unique phenotypic traits (ie, daptomycin resistance) within relapse lineages, which would suggest emergence and likely persistence of a strain within the host. Although external reservoirs cannot be entirely ruled out, the nesting of the infections and the accumulation of genetic changes suggest a high likelihood of specific host association. Furthermore, while phylogenetic histories of relapse-associated MRSAB isolates do occur in transmission clusters, they do not appear to be directly contributing to ongoing spread leading to additional MRSAB cases in health care settings.

We identified recurrent MRSAB among the most commonly circulating backgrounds causing bacteremia in the United States [6, 39], making it especially important to identify risk factors at substrain or host-factor levels. We did not identify a significant association between antibiotic treatment duration and recurrence. Other reports show a negative association between antibiotic duration and relapse [4, 46]. These discrepancies may be explained by the type and duration of antibiotic treatment. The number of recurrent infections we detected is from a single hospital group and may not be generalizable to other populations. Larger population sizes will be needed to identify variables with small effect sizes.

We found evidence that antibiotic therapy may have selected for strains with genetic mutations known to confer daptomycin and rifampin resistance. Several distinct amino acid changes were detected in isolates resistant to rifampin and daptomycin at the index infection or acquired over time. Changes to some amino acid sequences corresponded with phenotypic changes observed in other studies such as the changed asparagine at position 481 of the rpoB transcription product resulting in a rifampin-intermediate resistance profile [47], and the point mutations observed in mprF that resulted in changes at the same amino acid site (Ser337Thr and Ser337Leu) corresponding to daptomycin resistance. Multiple amino acid changes corresponding with phenotypic changes identified in this study demonstrate wide diversity of mutational profiles of rpoB and mprF even within a single study population, and convergence of traits within infected patients rather than single, resistant strains spreading between individuals. Heteroresistance or antibiotic cross-resistance may also contribute to the persistence of certain nonsynonymous changes. The convergence of mutations and the signatures of positive selection across relapse lineages suggest opportunities to further explore new evolutionary trajectories for infections that defy cure.

Management of MRSAB involves a comprehensive assessment of patient history, physical examination, and source identification [1]. Delays removing or draining an infection focus can increase the risk of persistent bacteremia [1] or metastatic spread [48]. When central venous catheters or other foreign bodies are suspected to be the source of SAB, removal of the foreign body is considered, although it is not always possible when it increases risk of patient harm. We found genomically similar MRSAB episodes among patients with foreign body infections, suggesting the presence of foci reseeding the blood. There is nearly a 5-fold greater odds for MRSA infections associated with foreign bodies among patients who had a previous MRSA infection within a year compared to those with no previously reported infection [49]. Clinicians should maintain a high index of suspicion of implanted foreign bodies as a source of recurrent MRSAB and advise patients accordingly during MRSAB recovery.

WGS provided additional support for case identification of new and relapsing infections in the context of the clinical history. The complexity of host factors, within-host selection pressures, strain background, and type of antibiotic treatment all play interacting roles in relapse. It is not possible to determine if recurrent MRSAB is caused by the same strain without performing WGS. Persistent populations undergo positive adaptation to the host, and convergently mutating genes are consistent with long-term usage or high exposure to antibiotics. Other traits necessary for survival in the body may still play an important role in persistence. Combining the frequency of genetic mutations, genetic relatedness, and known clinical risk factors for recurrence will lay the groundwork for better prediction of new MRSAB and may contribute to therapeutic strategies to prevent relapsing infections.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online (http://jid.oxfordjournals.org/). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.

Notes

Author contributions. B. M. T. conceptualized the project, designed analysis, collected data, analyzed data, and drafted initial manuscript. N. F. J., T. A., and K. B. collected data. K. H. collected data and reviewed analysis. M. Z. D. conceptualized the project, designed analysis, collected data, and drafted the initial manuscript. T. D. R. conceptualized the project, designed analysis, and drafted the initial manuscript, All authors reviewed and approved manuscript.

Acknowledgments. The authors thank Pam Tolomeo, MPH, of the University of Pennsylvania for assistance with clinical data management and the Penn/Children's Hospital of Philadelphia Microbiome Center for whole-genome sequencing services.

Disclaimer . The funders had no role in the study design, data collection, and analysis.

Financial support. This work was supported by the National Institutes of Health, National Institute of Allergy and Infectious Diseases (grant numbers 1R01AI139188-01 and 1R01AI158452-01A1 to T. D. R. and M. Z. D.); and the Office of Advanced Molecular Detection, Centers for Disease Control and Prevention (cooperative agreement number CK22-2204 through contract 40500-050-23234506 from the Georgia Department of Public Health to B. M. T. and T. D. R.).

Data availability. Raw sequence reads for this study are publicly available in the Sequence Read Archive under the project ID PRJNA751847. Metadata and analysis code used to generate the figures and tables are available at https://doi.org/10.5281/zenodo.15066414.