Dysfunctional accessory gene regulator (agr) as a prognostic factor in invasive Staphylococcus aureus infection: a systematic review and meta-analysis

The accessory gene regulator (agr) locus of Staphylococcus aureus is a quorum-sensing virulence regulator. Although there are many studies concerning the effect of dysfunctional agr on the outcomes of S. aureus infection, there is no systematic review to date. We systematically searched for clinical studies reporting outcomes of invasive S. aureus infections and the proportion of dysfunctional agr among their causative strains, and we performed a meta-analysis to obtain estimates of the odds of outcomes of invasive S. aureus infection with dysfunctional versus functional agr. Of 289 articles identified by our research strategy, 20 studies were meta-analysed for crude analysis of the impact of dysfunctional agr on outcomes of invasive S. aureus infection. Dysfunctional agr was generally associated with unfavourable outcomes (OR 1.32, 95% CI 1.05–1.66), and the impact of dysfunctional agr on outcome was more prominent in invasive methicillin-resistant S. aureus (MRSA) infections (OR 1.54, CI 1.20–1.97). Nine studies were meta-analysed for the impact of dysfunctional agr on the 30-day mortality of invasive S. aureus infection. Invasive MRSA infection with dysfunctional agr exhibited higher 30-day mortality (OR 1.40, CI 1.03–1.90) than that with functional agr. On the other hand, invasive MSSA infection with dysfunctional agr exhibited lower 30-day mortality (OR 0.51, CI 0.27–0.95). In the post hoc subgroup analysis by the site of MRSA infection, dysfunctional agr was associated with higher 30-day mortality in MRSA pneumonia (OR 2.48, CI 1.17–5.25). The effect of dysfunctional agr on the outcome of invasive S. aureus infection may vary depending on various conditions, such as oxacillin susceptibility and the site of infection. Dysfunctional agr was generally associated with unfavourable clinical outcomes and its effect was prominent in MRSA and pneumonia. Dysfunctional agr may be applicable for outcome prediction in cases of invasive MRSA infection with hardly eradicable foci such as pneumonia.


Results
The database search identified 286 articles, and we identified three more articles. Afterwards, 53 studies were removed due to duplication and 215 studies were excluded due to a lack of relevant information regarding our redefined outcome parameters and insufficient design. Finally, 20 studies were included in our analyses (Fig. 1). Among the 20 included studies, 18 studies collected cases of bacteraemia and their blood isolates, and two studies collected cases of lower respiratory infections and their respiratory isolates. Among 18 bacteraemia studies, seven studies used the isolates that were consecutively collected, and seven studies selected the cases and isolates by specific conditions other than the site of infection; infection with persistent bacteraemia, infection by MRSA with vancomycin MIC = 2 mg/L, or ICU setting. Four studies selected cases and isolates by specific sites of infection: catheter-related infection, infection with removed eradicable foci and without metastatic infection, endocarditis, or pneumonia.
Nineteen studies identified dysfunctional agr isolates by measuring delta-hemolysin production according to the method of Sakoulas G et al. 21 , and 1 study identified dysfunctional agr isolates by agr CAMP assay and the vesicle lysis test 34 .
Four studies included all cases of invasive S. aureus infections regardless of oxacillin susceptibility, 12 studies included only MRSA cases, and four studies included only MSSA. Twelve studies reported mortality as their main outcome, five studies reported the persistent bacteraemia as their main outcome, and 12 studies reported treatment failure (composite outcomes). Detailed characteristics of the included studies are provided in Table 1.
Overall unfavourable outcomes. For the association between dysfunctional agr and overall unfavourable outcomes of invasive S. aureus infection, data were available for 20 studies comprising 3426 patients (Fig. 2). To estimate the crude overall tendency, we conducted a meta-analysis using all included studies, although treatment failure was variably defined in each study 7,30,38,39,41,42,44,45,49 .
Pooled analysis showed that invasive S. aureus infection with dysfunctional agr was significantly associated with unfavourable outcomes (OR 1.32, 95% CI 1.05-1.66, I 2 = 0.27, Fig. 2A). The analysis of fourteen studies with MRSA demonstrated that invasive MRSA infection with dysfunctional agr exhibited an increased likelihood of unfavourable outcomes, and the effect was statistically significant (OR 1.54, 95% CI 1.20-1.97, I 2 = 0.10, Fig. 2B), while the analysis of seven MSSA studies demonstrated that invasive MSSA infection with dysfunctional agr did not increase the likelihood of unfavourable outcomes nor was it statistically significant (OR 0.71, 95% CI 0.47-1.09, I 2 = 0, Fig. 2C).

Mortality.
For the association between dysfunctional agr and the mortality of invasive S. aureus infection, data were available for 13 studies comprising 2659 patients, and pooled analysis showed that invasive S. aureus infection with dysfunctional agr significantly increased the likelihood of death in MRSA but decreased the likelihood of death in MSSA ( Supplementary Fig. 1). www.nature.com/scientificreports/ For the association between dysfunctional agr and 30-day mortality of invasive S. aureus infection, we included studies that reported or provided 30-day mortality as an outcome and that did not selectively include patients with specific sites of infection; data were available for nine studies comprising 2305 patients (Fig. 3). Pooled analysis showed that invasive S. aureus infection with dysfunctional agr was not associated with higher 30-day mortality in analyses that did not consider oxacillin susceptibility (OR 1.22, 95% CI 0.90-1.65, I 2 = 0.32, Fig. 3A). However, in the subgroup analysis of five studies with MRSA, invasive MRSA infection with dysfunctional agr showed significantly higher 30-day mortality than that with functional agr (OR 1.40, 95% CI 1.03-1.90, I 2 = 0, Fig. 3B). On the other hand, invasive MSSA infection with dysfunctional agr showed lower 30-day mortality than that with functional agr (0.51, 95% CI 0.27-0.95, I 2 = 0, Fig. 3C).
Mortality by the site of infection. We investigated the association between dysfunctional agr and 30-day mortality of invasive S. aureus infection by sites of infections for the five most common infection sites: centralline-associated bloodstream infection (CLABSI), pneumonia, skin and soft tissue infection (SSTI), bone and joint infection (BJI) and endocarditis (Supplementary Table 2).
Pooled analysis showed that dysfunctional agr was associated with higher 30-day mortality than functional agr in MRSA pneumonia (OR 2.48, 95% CI 1.17-5.25, I 2 = 0, Fig. 4A). BJI with MRSA with dysfunctional agr (OR 1.86, 95% CI 0.49-7.14, Fig. 4B) also had an increased likelihood of death within 30 days, while CLABSI with     Fig. 4E) also had an decreased likelihood of death within 30 days, and the result was not statistically significant. Pooled analysis showed that CLABSI with MSSA with dysfunctional agr had a decreased likelihood of death within 30 days, and BJI with MSSA with dysfunctional agr had an increased likelihood of death within 30 days, and both results were not statistically significant ( Supplementary Fig. 2). Persistent bacteraemia. Among the 18 SAB studies, five studies considered the persistent bacteraemia as an outcome, comprising 375 patients. Pooled analysis showed that SAB with dysfunctional agr was generally not associated with persistent bacteraemia (OR 1.54, 95% CI 0.78-3.04, I 2 = 0.39, Fig. 5A). In the subgroup analysis of three studies of MRSA SAB, we observed an increased likelihood of persistent bacteraemia, but the increase was not statistically significant (OR 2.15, 95% CI 0.74-6.19, I 2 = 0.57, Fig. 5B). In the subgroup analysis of 2 studies of MSSAB, we observed no difference in the rates of persistent bacteraemia between the dysfunctional agr group and the functional agr group (OR 0.91, 95% CI 0.40-2.08, I 2 = 0, Fig. 5C).

Discussion
To date, many studies have analysed the association between agr dysfunction and poor clinical outcomes of invasive S. aureus infection in various clinical settings 30,32,43,44,49,50 . Since Schweizer et al. reported that SAB with dysfunctional agr was associated with excessive mortality among severely ill patients 29 , subsequent studies have demonstrated that agr dysfunction was associated with higher mortality 31,32 , and the persistence of bacteraemia  www.nature.com/scientificreports/ in patients with SAB, especially in MRSA bacteraemia 7,30 . However, some studies reported that agr dysfunction of MRSA bacteraemia was not associated with treatment failure 39,41,42 , or mortality 51 . Despite the fact that various studies have addressed the association between agr dysfunction and the outcome of invasive S. aureus infection, data on its role in different infections and populations remained scattered. Despite the necessity for the integration of the information, this subject has never been systematically reviewed. Therefore, we systematically reviewed and performed a meta-analysis of 20 studies focusing on the association between clinical outcomes and dysfunctional agr in invasive S. aureus infections to investigate whether agr dysfunction can be a marker of poor clinical outcome. We performed data analysis using three different outcomes, namely  www.nature.com/scientificreports/ overall unfavourable outcomes, 30-day mortality and persistent bacteraemia. To reduce the complexity due to variably defined outcomes, we performed subgroup analysis by the site of infection using 30-day mortality. This meta-analysis demonstrated that invasive S. aureus infection with dysfunctional agr significantly increased the likelihood of unfavourable outcomes. However, the implication of agr dysfunction on the outcomes of invasive S. aureus infections was not prominent. The reason for the modest implication of dysfunctional agr on outcomes of invasive S. aureus infection could be that the studies dealt with various end points such as mortality, treatment failure (composite outcomes) and the persistence of bacteraemia. Oxacillin susceptibility, one of most important confounding factors, can affect the outcomes of invasive S. aureus infection 52 . Therefore, we performed a meta-analysis with the studies that reported or provided information on 30-day mortality according to the functionality of the agr locus of causative strains to minimize the effect of variances of end points of studies. The analysis demonstrated that invasive S. aureus infection with dysfunctional agr showed an increased likelihood of 30-day mortality in MRSA and a decreased likelihood of 30-day mortality in MSSA. It was interesting that dysfunctional agr could affect outcomes of invasive S. aureus infection differently according to the oxacillin susceptibility of isolated strains.
The difference in the impacts of dysfunctional agr on MRSA and MSSA infections can be speculated as follows. First, various experimental and clinical studies suggested that S. aureus acquires the ability to maintain persistent infection due to dysfunctional agr but loses virulence at the cost of dysfunctional agr 25,[52][53][54][55][56] . Second, the main therapeutic agents of studies with MSSA were anti-staphylococcal beta-lactam antibiotics. Antistaphylococcal beta-lactam antibiotics showed more rapid bactericidal effects than vancomycin in invasive S. aureus www.nature.com/scientificreports/ infection 57 . MSSA bacteraemia usually exhibits a shorter duration of persistence and less prevalently progresses to persistent bacteraemia 1,35 . The use of antistaphylococcal beta-lactam antibiotics may weaken the effect of persistent infection due to agr dysfunction. Therefore, it is possible that weakened virulence due to agr dysfunction nullifies the effect of persistent infections, and these effects may be more prominent in MSSA than in MRSA. A previous study demonstrated a relationship between isolates with reduced vancomycin susceptibility (RVS) and dysfunctional agr in MSSA bacteraemia 26 , and several studies demonstrated that infection caused by the RVS phenotype in MSSA isolates was associated with poor clinical outcomes regardless of antibiotics 37,[58][59][60][61] . However, we observed that dysfunctional agr was significantly associated with lower mortality in invasive MSSA infections. Therefore, dysfunctional agr might not be a marker of poor clinical outcomes of invasive S. aureus infection when the causative organism is susceptible to oxacillin. Depending on our findings, the 30-day mortality of invasive MRSA infection with dysfunctional agr was 1.4-fold higher than that of MRSA infection with functional agr. The implication of dysfunctional agr on the outcomes of invasive MRSA infections was statistically significant, but the magnitude of the effect of dysfunctional agr was not prominent. The lower prominence of the effect of dysfunctional agr on the outcomes of invasive MRSA infection in our meta-analysis can be explained by the fact that the included MRSA studies usually had other powerful confounding factors, such as a high proportion of community-acquired MRSA 38 , high vancomycin MIC 49 , hVISA phenotype 42 , different sites of infections (endocarditis or pneumonia) 41 , and the removal (or not) of foreign bodies 35 .
First, reduced vancomycin susceptibility (RVS) needs to be considered an important factor when we evaluate factors affecting unfavourable outcomes of invasive MRSA infections because intravenous administration of vancomycin is recommended as the standard treatment of patients with invasive MRSA infection; as a matter of fact, most of the included studies administered vancomycin as the main therapeutic agent for MRSA infections. A previous study reported that a high MIC of vancomycin is associated with high mortality and treatment failure in MRSA bacteraemia [62][63][64] , and agr dysfunction is associated with the attenuation of the bactericidal activity of vancomycin and the development of vancomycin intermediate/resistant S. aureus (VISA) 7,20,21,43 . This suggests that there is a possible intrinsic survival advantage of dysfunctional agr under vancomycin selective pressure [65][66][67] . However, most studies that used vancomycin as the main therapeutic agent showed a significant association between agr dysfunction and unfavourable outcomes regardless of vancomycin MIC [29][30][31][32] . Therefore, dysfunctional agr might be a marker of poor clinical outcomes of MRSA bacteraemia regardless of vancomycin MICs of causative organisms when patients were treated with vancomycin.
Second, the implication of agr dysfunction on unfavourable outcomes can be affected by specific conditions other than vancomycin MICs, such as the severity of infections and the proper removal of sources. Schweizer ML et al. demonstrated that agr dysfunction in SAB isolates was independently associated with high mortality among severely ill patients 29 . Hu HC et al. reported that S. aureus infection possessing dysfunctional agr, which was associated with the hVISA phenotype, exhibited markedly higher in-hospital mortality (12/13, 92.3%) than functional agr infection (23/35, 65.7%) in an ICU-setting study 43 . The study that limited the subjects to patients with adequate source control, which was another important factor for the outcome of invasive S. aureus infection, demonstrated that agr dysfunction is an independent risk factor for MRSA bacteraemia that persists despite the source control 30 . These findings suggested that dysfunctional agr can be a microbiological predictor of unfavourable outcomes in severe MRSA infection despite proper source control.
Third, the site of infection is one of most important factors in the outcomes of invasive S. aureus infection, and the implication of agr dysfunction on unfavourable outcomes can be affected by the site of infection. Therefore, we performed a subgroup analysis considering the site of infection. In the subgroup analysis by the site of infection, lower respiratory infection caused by MRSA with dysfunctional agr showed significantly higher mortality than MRSA with functional agr. BJI with MRSA with dysfunctional agr had an increased likelihood of mortality, although this result was not statistically significant. On the other hand, CLABSI and SSTI were not significantly affected by agr dysfunction. It is interesting that dysfunctional agr can affect the outcomes of invasive S. aureus infection differently according to the site of infection, although the reason for this difference is unclear. One probable explanation is as follows. First, patients with eradicable foci of infections, such as CLABSI and SSTI, might be more easily treated, and the magnitude of the effect of agr dysfunction on outcomes would be minimal. On the other hand, patients with foci that are difficult to eradicate, such as pneumonia and BJI, might be more difficult to treat, and the magnitude of the effect of dysfunctional foci would be more prominent 30 . Second, the vancomycin molecule is relatively large and penetrates poorly into the alveolar lining fluid and alveolar macrophages and into the bone and joints, further exacerbating the effect of RVS due to dysfunctional agr [68][69][70] .
The results of this meta-analysis should be interpreted with caution. First, the enrolled studies of our metaanalysis used different patient populations, different definitions of treatment failure, different definitions of the persistence of bacteraemia, and different time points of mortality. Except for seven studies that used consecutively collected isolates, the other enrolled studies used isolates from selected cases, such as specific sites of infection, persistent bacteraemia, a higher vancomycin MIC, and treatment in ICUs. The studies that used isolates from selected cases may have selected more virulent strains or those causing more difficult-to-treat infection, making it hard to compare the data with those of the studies that collected all consecutive isolates. To compensate for this heterogeneity of studies, we conducted subgroup analysis according to the sites of infection. We could extrapolate the effect of agr dysfunction on severity of disease based on the subgroup analysis by the site of infection. Moreover, the analysis was not adjusted for important factors that might affect clinical outcomes, such as source control, antibiotics used, time elapsed from infection onset to adequate therapy, incomplete vancomycin dosing, hVISA or strains with higher vancomycin MICs, severity of illness scoring, etc. Most of the included studies were retrospective, observational studies; therefore, publication bias is likely.
Second, most studies that investigated the association between agr dysfunction of MRSA bacteraemia and 30-day mortality were performed in Korea. In Korea, more than 70% of MRSA is SCCmec type II (mainly ST5),

Scientific Reports
| (2020) 10:20697 | https://doi.org/10.1038/s41598-020-77729-0 www.nature.com/scientificreports/ and 20-30% is SCCmec type IV/IVa (mainly ST72) 71 . Several molecular epidemiology studies reported that 41-90% of SCCmec type II-MRSA was associated with agr dysfunction, but 3-12% of SCCmec type IV/Iva-MRSA was associated with agr dysfunction in Korea 24,31,72,73 . Therefore, our results should be interpreted with caution because it is not clear whether the poor outcomes of invasive MRSA infection with dysfunctional agr in our results might be affected by the predominance of a specific clone, SCCmec type II (mainly ST5), which was associated with dysfunctional agr.

Methods
Search strategy and selection criteria. We searched for clinical studies reporting the proportion of dysfunctional agr and its association with outcomes of invasive S. aureus infections from database inception to 26th September 2018 in Medline and ISI Web of Science (Science Citation Index Expanded). The combination of the following keywords were used to search the studies: "Staphylococcus aureus", "bacteremia", "pneumonia", "endocarditis", "osteomyelitis", "arthritis" "Quorum sensing", "agr (accessory gene regulator)", and "delta-hemolysin". We excluded review articles, case reports and experimental studies. We also excluded colonization studies or epidemiological studies that did not report outcomes associated with agr functionality. Two authors (Shinwon Lee, Soon Ok Lee) independently performed the literature search and identified all studies potentially relevant for this review (Fig. 1).

Data analysis.
To analyse the association between the dysfunction of agr and outcomes of invasive S. aureus infections, we calculated odds ratios (ORs) comparing the odds of outcomes of S. aureus infection with dysfunctional agr with the odds of that with functional agr. We used meta-analysis to obtain estimates of the odds of outcomes of invasive S. aureus infection with dysfunctional agr and presented ORs and their 95% confidence intervals (CIs) in random-effects model analysis. Meta-analysis results are presented as forest plots, and funnel plots were inspected to judge potential evidence for publication bias. We applied a two-sided significance level of 0.05. R version 3.3.2 (R Foundation for Statistical Computing, Vienna, Austria) and R package 'meta' were used for all statistical analyses 76,77 .
The number of patients and events for the dysfunctional versus functional agr group were extracted by one author (Shinwon Lee). The results were independently validated by another author (Soon Ok Lee). We extracted information on study designs, settings, patient characteristics, method for isolate collection (or selection) and the year of data collection to assess potential heterogeneity in the study populations. A formal risk assessment of the individual studies was judged according to the Newcastle-Ottawa Scale (NOS) 78 . To investigate whether dysfunctional agr is associated with poor outcome among cases of invasive disease, we compared outcomes, mainly 30-day mortality and the persistent bacteraemia. The included studies reported various end points as their outcomes, such as 30-day mortality, in-hospital mortality, S. aureus bacteraemia-attributable mortality, persistent bacteraemia, treatment failure, and development of complications (Table 1). We analysed data using three different outcomes, namely overall unfavourable outcomes, 30-day mortality and persistent bacteraemia.