Comparison of four automated microbiology systems with 16S rRNA gene sequencing for identification of Chryseobacterium and Elizabethkingia species

Chryseobacterium and Elizabethkingia species have recently emerged as causative agents in life-threatening infections in humans. We aimed to evaluate the rates at which four common microbial identification systems identify Chryseobacterium and Elizabethkingia species in clinical microbiology laboratories. Based on the results of 16S rRNA gene sequencing, a total of 114 consecutive bacteremic isolates, including 36 (31.6%) C. indologenes, 35 (30.7%) E. anophelis, 22 (19.3%) C. gleum, 13 (11.4%) E. meningoseptica, and other species, were included in this study. The overall concordance between each method and 16S rRNA gene sequencing when identifying Chryseobacterium and Elizabethkingia species was 42.1% for API/ID32, 41.2% for Phoenix 100 ID/AST, 43.9% for VITEK 2, and 42.1% for VITEK MS. Among the 22 C. gleum isolates, only one (4.8%) was correctly identified using VITEK 2 and Phoenix 100 ID/AST, and none were accurately recognized using API/ID32 or VITEK MS. Except for two isolates that were not identified using API/ID32, all E. anophelis isolates were misidentified by all four identification systems as E. meningoseptica. Our results show that these approaches have low accuracy when identifying Chryseobacterium and Elizabethkingia species. Hence, we recommend amending the discrimination rate of and adding non-claimed pathogens to databases of microbial identification systems.

example, matrix-assisted laser desorption ionization-time of flight mass spectrometry; MALDI-TOF MS). Occasional studies have reported misidentifying Chryseobacterium and Elizabethkingia species when using conventional phenotypic identification systems and the VITEK 2 Automated Identification System (bioMérieux, Marcy l'Etoile, France) 6,10,13 . However, 16S rRNA gene sequencing has been shown to be a reliable method of identifying Chryseobacterium and Elizabethkingia species 14,15 . In this study, we used 16S rRNA sequencing to analyze Chryseobacterium and Elizabethkingia species isolated from patient blood samples. We compared the accuracies of the following four bacterial identification systems that are commonly used to identify Chryseobacterium and

Isolates and claimed microorganisms in databases of identification systems. A total of 114
consecutively non-repeated isolates that were initially identified as Chryseobacterium and Elizabethkingia species by a clinical microbiology laboratory were included in this study (Table 1). According to BLAST results based on 16S rRNA gene sequencing, 36 isolates were C. indologenes (31.6%), 35 were E. anophelis (30.7%), 22 were C. gleum (19.3%), 13 were E. meningoseptica (11.4%), 3 were C. culicis, 2 were C. bemardetii, 2 were Candidatus Chryseobacterium massilia, and 1 was E. miricola. The coverage rates of these Chryseobacterium and Elizabethkingia species that were obtained using the API/ID32, Phoenix 100 ID/AST, VITEK 2, and VITEK MS v2.0/v3.0 identification databases are shown in Table 1. Chryseobacterium indologenes and E. meningoseptica were claimed in the databases of all biochemical systems and MALDI-TOF MS systems. C. gleum was built into all systems except API/ID32. Elizabethkingia miricola was included only in the Phoenix 100 ID/AST database. However, these identification systems did not contain identification data for Candidatus C. massilia, C. bemardetii, C. culicis, and E. anophelis.  (2) C. indologenes (2) No 0

Discussion
In this study, we compared the accuracies of four commercial microbial identification systems to that of 16S rRNA gene sequencing for identifying Chryseobacterium and Elizabethkingia species. The overall concordance between each of these four commercial methods and 16S rRNA gene sequencing for identifying Chryseobacterium and Elizabethkingia species were as follows: API/ID32, 42.1%; Phoenix 100 ID/AST, 41.2%; VITEK 2, 43.9%; and VITEK MS, 42.1%. After taking the coverage of each database into account, the overall concordance between 16S rRNA gene sequencing and API/ID32, Phoenix 100 ID/AST, VITEK 2, and VITEK MS was 98%, 75.8%, 70.4%, and 67.6%, respectively. Chryseobacterium gleum is rarely reported to cause infection in humans 16,17 . However, our data reveal that C. gleum accounts for 33.8% of Chryseobacterium bacteremia cases in humans. Lo et al. 6 reported that 15 clinical isolates of C. gleum that were confirmed by 16S rRNA gene sequencing were misidentified by VITEK 2 as C. indologenes (14/15; 93.3%) and E. meningoseptica (1/15; 6.7%). When submitted to a Bruker Microflex LT MALDI-TOF MS System using Biotyper database 3.0 (Bruker Daltonics, Bremen, Germany), 2 (13.3%) and 13 (86.6%) of these 15 isolates were identified as C. gleum species and probable species, respectively. In our study, 81.8% of C. gleum isolates were misidentified as C. indologenes by all four commercial identification systems. Chryseobacterium gleum was included in the Phoenix 100 ID/AST, VITEK 2, and VITEK MS (v2.0 and v3.0) databases but not in the API/ID32 database. Among the 22 C. gleum isolates in our study, only 1 (4.5%) was correctly identified by VITEK 2 and Phoenix 100 ID/AST, and none were accurately recognized by VITEK MS. MALDI-TOF MS systems have become popular in clinical microbiology laboratories because they rapidly, highly accurately, and cost-effectively identify different microorganisms. However, despite the fact that C. gleum was included in the spectral database, none of the C. gleum isolates were accurately identified by VITEK MS. The four microbial identification systems used in our study are widely used by clinical microbiology laboratories all over the world. The inability of these techniques to distinguish C. gleum from C. indologenes may result from false impressions that have led to the notion that there is a low prevalence of C. gleum and an overestimation of the prevalence of C. indologenes infections in humans.
Recent studies have shown that E. anophelis is frequently misidentified as E. meningoseptica [8][9][10] . Lau et al. 10 reported 17 patients in Hong Kong who were diagnosed using 16S rRNA gene sequencing with infection with E. anophelis. However, all 17 E. anophelis isolates were recognized as E. meningoseptica by VITEK 2, and the Bruker MALDI-TOF MS Biotyper also failed to correctly identify E. anophelis 10 . Similar to a report by Lau et al., Han et al. 13 found in their study performed in South Korea that none of the tested 51 E. anophelis isolates was correctly identified by a Bruker MALDI-TOF MS Biotyper. A VITEK MS research-use-only (RUO) system coupled with a SARAMIS SuperSpectra database successfully identified all 51 E. anophelis isolates 13 , but this system is not available to clinical microbiology laboratories. In our study, 55.5% (27/49) of previously identified E. meningoseptica were revealed to be E. anophelis based on the results of 16S rRNA gene sequencing. Our results show that using VITEK MS with the v2.0 and v3.0 Knowledge Bases or any of the other three commonly used biochemical systems discussed here resulted in the failed identification of E. anophelis. We suggest that many previously reported E. meningoseptica infections might in fact have been identified as E. anophelis if they were analyzed using commercial identification systems. The prevalence of E. anophelis infections in humans could therefore be dramatically underestimated.

Conclusions
Being able to correctly identify microorganisms is extremely important in clinical practice and microbiologic research. However, the results of our study show that four microbial identification systems that are widely used in clinical microbiology laboratories are highly inaccurate when identifying Chryseobacterium and Elizabethkingia species. Specifically, the extremely low rates at which these methods identify the life-threatening pathogens C. gleum and E. anophelis may cause the prevalence of these species to be substantially underestimated. We recommend amending the method used to discriminate C. gleum from C. indologenes and adding a database for E. anophelis to the microbial identification systems discussed here.

Ethics and experimental biosafety statements. This study was approved by the Institutional Review
Board of E-Da Hospital (EMRP-105-134). The need for patient informed consent was waived by the Institutional Review Board of E-Da Hospital because the retrospective analysis of routine blood cultures posed no more than a minimal risk of harm to the subjects. The experiments in this study were approved by the Institutional Biosafety Committee of E-Da Hospital. All experiments were performed in accordance with relevant guidelines and regulations. Study design. year retrospective study was conducted at a 1,000-bed university-affiliated hospital that serves more than 2 million people in southern Taiwan. A clinical laboratory database was searched to identify blood cultures that were identified as containing Chryseobacterium and Elizabethkingia species between January 2005 and December 2015. The isolates were initially identified as Chryseobacterium and Elizabethkingia species by a clinical microbiology laboratory that first used API/ID32 Phenotyping Kits (2005-2013) and then used a VITEK MS MALDI-TOF MS System (2014-2015) after upgrading the microbial identification system. All isolates were stored as glycerol stocks at −80 °C until used. Chryseobacterium indologenes BCRC 17271 (ATCC 29897) and Elizabethkingia meningoseptica BCRC 10677 (ATCC 13253) were used as quality controls. The 16S rRNA gene sequencing method was considered the reference method for bacterial identification.
Identification of microorganisms using microbial identification systems. For re-identification, the thawed bacteria were inoculated on tryptic soy agar with 5% sheep blood after they were removed from the freezer. The plates were then incubated in a 5% CO 2 atmosphere at 35 °C for 15 to 24 hours. All isolates were re-identified using API/ID32 Phenotyping Kits, Phoenix 100 ID/AST Automated Microbiology System, VITEK 2 Automated Identification System, and VITEK MS MALDI-TOF MS System. The isolates were identified according to each manufacturer's instructions. For the API/ID32 Phenotyping Kits, an ID 32 GN card and database version 3.1 were used to identify microorganisms according to ATB Expression. The results obtained using the Phoenix 100 ID/AST System were analyzed using database version 5.51 A. A confidence level of ≥90% was defined as acceptable for the Phoenix System 20 . The identifications yielded by the VITEK 2 system were obtained using a GN ID card and database version 7.01. The quality of bacterial identification was assessed using VITEK 2 Advanced Expert System. The results were defined as acceptable at a confidence level of 96-99% (excellent identification) or 93-95% (very good quality) 21 . The mass spectral fingerprints generated by the VITEK MS System were analyzed using Knowledge Base v2.0 and repeatedly tested using Knowledge Base v3.0. A confidence value of ≥90% (reliable identification) or 85%-90% (acceptable identification) was regarded as a successful identification. A value was defined as no identification if the VITEK MS confidence value was <85% 22 . Data Availability. The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.