Elizabethkingia anophelis bacteremia is associated with clinically significant infections and high mortality

Unlike Elizabethkingia meningoseptica, the clinical importance of E. anophelis is poorly understood. We determined the clinical and molecular epidemiology of bacteremia caused by Elizabethkingia-like species from five regional hospitals in Hong Kong. Among 45 episodes of Elizabethkingia-like bacteremia, 21 were caused by Elizabethkingia, including 17 E. anophelis, three E. meningoseptica and one E. miricola; while 24 were caused by other diverse genera/species, as determined by 16S rRNA gene sequencing. Of the 17 cases of E. anophelis bacteremia, 15 (88%) were clinically significant. The most common diagnosis was pneumonia (n = 5), followed by catheter-related bacteremia (n = 4), neonatal meningitis (n = 3), nosocomial bacteremia (n = 2) and neutropenic fever (n = 1). E. anophelis bacteremia was commonly associated with complications and carried 23.5% mortality. In contrast, of the 24 episodes of bacteremia due to non-Elizabethkingia species, 16 (67%) were clinically insignificant. Compared to non-Elizabethkingia bacteremia, Elizabethkingia bacteremia was associated with more clinically significant infections (P < 0.01) and positive cultures from other sites (P < 0.01), less polymicrobial bacteremia (P < 0.01), and higher complication (P < 0.05) and mortality (P < 0.05) rates. Elizabethkingia bacteremia is predominantly caused by E. anophelis instead of E. meningoseptica. Elizabethkingia bacteremia, especially due to E. anophelis, carries significant morbidity and mortality, and should be considered clinically significant unless proven otherwise.

accurate diagnosis is important to guide appropriate antibiotic regimens which often consist of a combination of ciprofloxacin or rifampicin with piperacillin-tazobactam or vancomycin.
In contrast to E. meningoseptica, the epidemiology and pathogenicity of E. anophelis and E. miricola were less well understood. E. miricola, originally named C. miricola when first isolated from condensed water obtained from the Russian space station, Mir, only rarely causes nosocomial infections in human 7,8 . On the other hand, E. anophelis was first isolated from midgut of the mosquito Anopheles gambiae in 2011 9 . Soon after its discovery, it was reported to cause neonatal meningitis in the Central African Republic and a nosocomial outbreak in an intensive-care unit in Singapore 10,11 . The first discovery of E. anophelis from mosquito gut has raised suspicion on mosquitoes as the source of neonatal meningitis cases in Africa 10 . However, our recent report on E. anophelis meningitis in two neonates and chorioamnionitis in a neonate's mother in Hong Kong suggested that mosquitoes were unlikely the vehicles of transmission 12 . Since the transmission route was initially obscure, draft genome sequencing was performed and showed evidence for perinatal vertical transmission from a mother to her neonate 12 . The ultimate resolution power of genome sequencing also enabled species confirmation and discrimination from the phenotypically similar species, E. meningoseptica 12 .
Since E. anophelis was commonly misidentified as E. meningoseptica in previous reports [10][11][12][13] , we hypothesize that many previously described E. meningoseptica isolates were actually E. anophelis and that E. anophelis may account for a significant proportion of Elizabethkingia infections. To better understand the epidemiology and clinical disease spectrum of E. anophelis and Elizabethkingia as a whole, we determined the clinical and molecular epidemiology of bacteremia caused by Elizabethkingia-like species from five regional hospitals in Hong Kong. All bacteremia episodes caused by Elizabethkingia-like species identified by conventional phenotypic tests from 2004 to 2013 during the study period were included. For the 45 episodes of Elizabethkingia-like bacteremia identified, 16S rRNA gene sequencing was performed for species identification and clinical characteristics and outcomes were analyzed.
Clinical characteristics of patients with Elizabethkingia-like bacteremia. The clinical characteristics of the 45 episodes of Elizabethkingia-like bacteremia were summarized in Tables 1-3. Of the 17 episodes of E. anophelis bacteremia, the male:female ratio was 11:6, with a median age of 58 years (range, 1 day to 104 years) ( Table 2). Most patients had underlying diseases. Except two patients with pseudobacteremia, most cases were associated with clinically significant bacteremia, with 12 cases being hospital-acquired and three community-acquired. The most common diagnosis was pneumonia (n = 5), followed by catheter-related bacteremia (n = 4), neonatal meningitis (n = 3), nosocomial bacteremia (n = 2) and neutropenic fever (n = 1). Among the five cases of pneumonia, three were community-acquired, including one in a 51-year-old previously healthy man (case EA10). All three cases of neonatal meningitis were hospital-acquired. Details of two (HKU36 and HKU38) of the three neonatal meningitis cases have been reported previously 12 . They were included because they fell into the inclusion criteria during the study period. The source of E. anophelis in the two cases of nosocomial bacteremia and one case of neutropenic fever was obscure, although severe mucositis was present in the latter. Three of the 17 patients had multiple positive blood cultures. E. anophelis was concomitantly isolated from other body sites in seven patients, including cerebrospinal fluid (n = 3), respiratory samples (n = 3) and catheter tip (n = 1). Most patients survived, but four patients died despite appropriate antibiotics given in two of the four patients. Complications as a result of E. anophelis bacteremia included acute pulmonary edema, congestive heart failure, multi-organ failure, disseminated intravascular coagulation, septic shock, haematemesis, acute renal failure, metabolic acidosis and intraventricular haemorrhage. Various antibiotic regimens as single drug or combinations were used for empirical or definitive treatment, including quinolones, penicillins, cephalosporins,   carbapenems, vancomycin and rifampicin. Removal of catheter was often required in catheter-related bacteremia cases in addition to antibiotic treatment. All the four patients with E. meningoseptica or E. miricola bacteremia also had underlying disease and clinically significant infections (Table 2). Two patients with E. meningoseptica bacteremia had biliary tract infections, while one had nosocomial bacteremia. Interestingly, the only episode of E. miricola bacteremia (case EMI1) was detected in the same patient with E. anophelis nosocomial pneumonia (case EA1) which occurred more than one month previously during the same admission and was complicated by heart failure and acute pulmonary edema requiring intubation and inotropes. Five weeks after successful treatment with piperacillin-tazobactam, the patient developed another episode of bacteremia caused by E. miricola which was also isolated from central venous catheter. The patient recovered with levofloxacin and catheter removal.
In contrast, 16 of the 24 patients with bacteremia due to non-Elizabethkingia species had pseudobacteremia ( Table 3). The other eight patients were diagnosed to have nosocomial bacteremia (three cases), transfusion-related bacteremia (two cases), catheter-related bacteremia (one case), neutropenic fever (one case) and sepsis (one case). Except for a case of community-acquired sepsis in a 90-year-old man (case P1), all other cases were hospital-acquired. Only one episode of catheter-related bacteremia was complicated by septic shock, where C. arthrosphaerae was isolated from six blood cultures. All the eight patients survived.
Comparison between Elizabethkingia and non-Elizabethkingia cases showed that clinically significant infections were more common in patients with Elizabethkingia bacteremia than those with non-Elizabethkingia bacteremia (i.e. fewer pseudobacteremia cases were observed) (P < 0.01). Moreover, patients with Elizabethkingia bacteremia had more positive cultures from other sites (P < 0.05) and lower incidence of polymicrobial bacteremia (P < 0.01). They also showed higher complication (P < 0.05) and mortality (P < 0.05) rates than those with with non-Elizabethkingia bacteremia (Table 1).
MALDI-TOF MS using Reference Library Biotyper v3.1.2.0 (Bruker Daltonik, Germany) failed to identify the 17 E. anophelis isolates (10 misidentified as E. meningoseptica with score 2.073 to 2.403, two identified as Elizabethkingia species with score 1.952-1.971 and five unidentified with score 1.32 to 1.42 to E. meningoseptica). When the database was expanded with inclusion of mass spectra from seven E. anophelis isolates, all the other 10 E. anophelis isolates were correctly identified as E. anophelis with score 2.321 to 2.634. The three E. meningoseptica and one E. miricola strains were correctly identified by the Bruker reference library (see Supplementary Table 2). Hierarchical cluster analysis showed that the protein mass spectra of E. anophelis and E. miricola were clustered together but formed a distinct branch from E. meningoseptica (Fig. 2).

PFGE typing of E. anophelis.
To determine the genetic relatedness of the 17 E. anophelis strains isolated from five different hospitals, PFGE was performed using the bacterial genomic and DNA restriction endonuclease XbaI. In general, the PFGE patterns among the 17 E. anophelis strains were distinct from each other and different from that of the type strain R26 T . Dendrogram constructed using the PFGE images showed that some adjacently clustered strains were isolated from the same hospital (Fig. 3). For example, strains EA2, EA3 and EA4 were from hospital 5, while strains EA13 and EA15 were from hospital 6. However, no clear epidemiological linkage could be identified in terms of time and place of blood culture isolation. Specifically, strains EA2, EA3 and EA4 were isolated from different years. Strains EA13 and EA15 were isolated four months apart and the two patients were hospitalized in different wards of hospital 6.

Discussion
E. anophelis bacteremia should be considered clinically significant unless proven otherwise and should prompt appropriate antibiotic therapy. In contrast to the traditional belief that E. meningoseptica was the most important Elizabethkingia species associated with bacteremia, the present study showed that the majority of Elizabethkingia bacteremia cases were caused by E. anophelis. In particular, neonatal meningitis was associated exclusively with E. anophelis in this study. In the present series, E. anophelis bacteremia mainly occurred in neonates or adults with underlying medical illnesses, and was most commonly associated with pneumonia and hospital-acquired infections such as catheter-related bacteremia and neonatal meningitis. Notably, community-acquired pneumonia occurred in three patients including a healthy middle-aged adult, where the infective source remained obscure. Apart from the two previous reported cases of neonatal meningitis (HKU36 and HKU38) which were most likely acquired from maternal-to-infant transmission based on comparative genomics 12 , no obvious epidemiological linkage or genetic relatedness was identified among the other cases. Nevertheless, some potentially genetically related strains upon PFGE analysis may have circulated in the same hospital. Further studies should be performed to better understand the disease spectrum and transmission routes of E. anophelis. Staphylococcus aureus; NA, not applicable; PDA, patent ductus arteriosus; PTB, pulmonary tuberculosis; PVD, peripheral vascular disease; RAS, renal artery stenosis; RDS, respiratory distress syndrome; RPC, recurrent pyogenic cholangitis. E. anophelis bacteremia carries significant morbidity and mortality. Various complications were observed and four patients died, giving a mortality rate of 23.5%. This is in line with previous reports of E. anophelis neonatal meningitis being associated with poor outcomes 10 . Nevertheless, all the three present cases of neonatal meningitis were cured with early use of appropriate antibiotics 12 . Although E. anophelis usually confers resistance to multiple antibiotics such as ceftazidime, imipenem and aminoglycosides 12 , all the 17 isolates were susceptible to ciprofloxacin, cefoperazone-sulbactam and vancomycin, which should be considered in empirical treatment while awaiting susceptibility results. In cases of catheter-related bacteremia, infected catheters should be removed in addition to antibiotic treatment. Future prospective studies with population-based data should be performed to determine the prevalence or incidence of E. anophelis bacteremia.
E. meningoseptica and E. miricola appeared to be much less prevalent than E. anophelis, although similar studies in other countries are required to more accurately assess their relative importance. Similar to E. anophelis, E. meningoseptica and E. miricola bacteremia were associated with clinically significant infections. Besides one case of E. meningoseptica nosocomial bacteremia and one case of E. miricola catheter-related bacteremia, biliary tract infections were also noted in two cases of E. meningoseptica bacteremia. It remains to be determined if E. meningoseptica may have the propensity to cause biliary tract infections among the genus. Given their similar antibiotic susceptibility profiles to that of E. anophelis, ciprofloxacin, cefoperazone-sulbactam and vancomycin should also be included in treatment regimens for E. meningoseptica and E. miricola bacteremia.
In contrast to Elizabethkingia species, isolation of non-Elizabethkingia species from blood cultures should raise suspicion of their clinical significance. In this study, non-Elizabethkingia bacteremia is associated with higher incidence of pseudobacteremia and polymicrobial bacteremia than Elizabethkiniga bacteremia. In particular, all six isolates of C. indologenes and the three potentially novel species were associated with pseudobacteremia. Bacteremia caused by non-Elizabethkingia species is also associated with lower incidence of complications and mortality than Elizabethkingia bacteremia, suggesting that these bacterial species may be less virulent than Elizabethkingia. Moreover, these environmental, non-Elizabethkingia bacterial species may contaminate blood cultures when aseptic techniques during blood taking are breached. Careful clinical assessment is required to determine the clinical significance and the need for antibiotics when these bacteria are isolated from blood cultures.
Although the two E. anophelis strains, HKU36 and EA14, are phylogenetically genetically close to E. endophytica strain JM-87 T , they should belong to E. anophelis instead of E. endophytica or a novel species. The two strains possessed 99.9% nucleotide identities to both the newly proposed species, E. endophytica strain JM-87 T and E. anophelis R26 T in their 16S rRNA gene sequences. Although strain JM-87 T was reported to possess 51-52% similarities to E. anophelis R26 T by DNA-DNA hybridization 3 , our previous study showed that the draft genome of strain HKU36 possessed 78.3% nucleotide identity to the genome of E. anophelis R26 T by estimation of intergenomic distance 12 . Given the close relatedness of strains HKU36, EA14 and JM-87 T in their 16S rRNA genes (Fig. 1), it is likely that the genome sequences of strain JM-87 T and EA14 may also possess > 70% identity to that of E. anophelis R26 T . Since genome-based comparison can offer ultimate resolution for species delineation which is superior to traditional DNA-DNA hybridization methods, genome sequencing of strain JM-87 T and related strains such as EA14 should be performed to more accurately define their taxonomic positions.
The present results confirmed our suspicion that E. anophelis was a previously under-reported bacterium which can be easily misidentified as E. meningoseptica 12 . Although E. anophelis was first discovered from mosquito gut 10 , we previously showed that maternal chorioamnionitis, instead of mosquitoes, was more likely the source of neonatal meningitis 12 . Similarly, mosquitoes are unlikely the route of transmission in other E. anophelis infections. We speculate that contaminated environments, such as infected catheters, are the source of infection in most cases, as in the case of previously described "E. meningoseptica" infections. Given their similar phenotypic characteristics, E. anophelis isolates from previous reports were often mistaken as E. meningoseptica initially [10][11][12]14 . Phenotypic tests, such as acid production from cellobiose and citrate utilization, previously reported as potentially useful for species discrimination 12 , were unlikely to be reliable. In our previous study, the three E.  anophelis strains were also initially misidentified as E. meningoseptica even with MALDI-TOF MS, owing to the absence of E. anophelis spectra in commercial databases 12 . The 16S rRNA genes of E. anophelis possessed > 98% nucleotide identity to those of E. meningoseptica and E. miricola, which should offer sufficient discriminative  Results showed that the 17 isolates possessed distinct PFGE patterns. In Panel (B), dendrogram was constructed with PFGE data by similarity and clustering analysis using the Dice coefficient (1% tolerance and 0.5% optimization) and unweighted pair-group method using average linkages with GelCompar II. power 12 . However, some "E. meningoseptica" strains with 16S rRNA sequences deposited in GenBank, such as strains G3-1-08 and 502 15,16 , should belong to E. anophelis based on phylogenetic analysis 12 . These "misidentified" strains in GenBank may confuse the interpretation of 16S rRNA gene sequencing results and should be rectified 12 . While E. anophelis can be distinguished from E. meningoseptica by MALDI-TOF MS when the database is expanded with mass spectra from E. anophelis strains, E. anophelis and E. miricola appear to be indistinguishable from each other. With an expanded database using E. anophelis isolates, MALDI-TOF MS is the method of choice for rapid and accurate diagnosis of E. anophelis infections, which is crucial to better understand its epidemiology and clinical disease spectrum.

Methods
Ethics statement. The use of blood culture isolates and anonymous clinical data were approved by Institute Review Board, The University of Hong Kong/Hospital Authority, Hong Kong (reference UW 04-278 T/600). The methods and all experimental protocols were carried out in accordance with the approved guidelines. Since this study does not involve experimentation on human subjects or the use of tissue samples from human subjects, written informed consent has been waived by our institutional review board.

Settings and Patients.
Patients were hospitalized in five regional hospitals located in different areas of Hong Kong from 2004 to 2013. To identify potential cases of Elizabethkingia bacteremia, all bacteremia episodes caused by oxidase-positive, non-glucose fermenters that were identified as Elizabethkingia, Flavobacterium or Chryseobacterium species by conventional phenotypic tests during the study period were included with clinical data analyzed 17 . Bacteremia was categorized as clinically significant or pseudobacteremia (contamination of blood culture) by clinical and laboratory criteria 18 . The criteria include the patient's clinical presentation, physical examination findings, body temperature at the time of the blood culture, leukocyte and differential cell counts, imaging or operative results, histopathological findings, number of positive blood cultures out of the total number performed, and response to treatment.
Bacterial isolates. Collection of clinical specimens, bacterial cultures and conventional phenotypic identification were performed according to standard protocols 19 . Two of the 45 Elizabethkingia-like isolates from two neonates (HKU36 and HKU38) have been reported previously 12 . The same isolate recovered from the same patient was counted only once.
16S rRNA gene sequencing for species identification. The 45 blood culture isolates were subject to 16S rRNA gene sequencing according to previously published protocols with modifications, using primers LPW57 (5′ -AGTTTGATCCTGGCTCAG-3′ ) and LPW58 (5′ -AGGCCCGGGAACGTATTCAC-3′ ) 12,20 . The sequences of PCR products were compared to known gene sequences in GenBank by multiple sequence alignment using CLUSTAL_W in MEGA version 6 21 . Phylogenetic tree was constructed by maximum likelihood method using MEGA version 6 21 .
Statistical analysis. A comparison of characteristics was made between patients with Elizabethkingia and non-Elizabethkingia bacteremia using Chi-square test (IBM SPSS Statistics version 19). P < 0.05 was regarded as statistically significant.

Phenotypic characterization and matrix-assisted laser-desorption ionization-time-of-flight mass-spectrometry (MALDI-TOF MS) of Elizabethkingia isolates. The 21
Elizabethkinigia isolates were characterized by phenotypic tests, Vitek 2 GNI system (bioMérieux, France) and MALDI-TOF MS. MALDI-TOF MS was performed by ethanol formic acid extraction method as described previously, using Bruker Daltonics microflex LT system with Reference Library Biotyper v3.1.2.0 (Bruker Daltonik, Germany) 22,23 . Since E. anophelis is not included in the Bruker reference, mass spectra generated from seven E. anophelis strains with identity confirmed by 16S rRNA gene sequencing were later added to the database. Obtained spectra were subject to hierarchical cluster analysis as described previously 24 . Antibiotic susceptibility testing was performed by Kirby Bauer disk diffusion method with results interpreted according to Clinical and Laboratory Standards Institute for Staphylococcus aureus (vancomycin) and Pseudomonas aeruginosa (other drugs), because of the lack of interpretative criteria for Elizabethkingia 25 .
Pulsed-field gel electrophoresis (PFGE). The 17 E. anophelis isolates were characterized by PFGE using CHEF Mapper XA system (Bio-Rad, CA, USA) and restriction endonuclease XbaI as described previously 12,26 . After PFGE, the gel was stained with ethidium bromide (1 μ g/ml) for 30 minutes and the patterns of the genomic DNA digest were visualized with a UV transilluminator. Digital images were stored electronically as TIFF files and analyzed visually and with GelCompar II (version 3.0; Applied Maths, Kortrijk, Belgium), and represented by UPGMA method.
Nucleotide sequence accession number. The 16S rRNA gene sequences of the 45 blood culture isolates have been deposited in the GenBank sequence database under accession no. KP875383 to KP875427.