Safety and Accuracy of Matrix-Assisted Laser Desorption Ionization–Time of Flight Mass Spectrometry for Identification of Highly Pathogenic Organisms

ABSTRACT Matrix-assisted laser desorption ionization–time of flight mass spectrometry (MALDI-TOF MS) sample preparation methods, including the direct, on-plate formic acid, and ethanol/formic acid tube extraction methods, were evaluated for their ability to render highly pathogenic organisms nonviable and safe for handling in a biosafety level 2 laboratory. Of these, the tube extraction procedure was the most successful, with none of the tested strains surviving this sample preparation method. Tube extracts from several agents of bioterrorism and their near neighbors were analyzed in an eight-laboratory study to examine the utility of the Bruker Biotyper and Vitek MS MALDI-TOF MS systems and their in vitro diagnostic (IVD), research-use-only, and Security-Relevant databases, as applicable, to accurately identify these agents. Forty-six distinct strains of Bacillus anthracis, Yersinia pestis, Francisella tularensis, Burkholderia mallei, Burkholderia pseudomallei, Clostridium botulinum, Brucella melitensis, Brucella abortus, Brucella suis, and Brucella canis were extracted and distributed to participating laboratories for analysis. A total of 35 near-neighbor isolates were also analyzed.

Previous sample preparation studies (4)(5)(6)(7) have produced conflicting results with respect to their ability to adequately inactivate pathogens and may not have utilized manufacturer-recommended methods.
Indeed, validation and/or verification of MALDI-TOF MS software libraries poses another significant dilemma for clinical laboratories (8). The Clinical and Laboratory Standards Institute (CLSI) (9) has recently published some example end-user verification protocols, including a suggested list of organisms for testing, but the protocols still suggest that the final selection of organisms for verification should be compiled by the individual laboratory. This issue may be complicated by the choice of libraries (e.g., Food and Drug Administration [FDA] approved versus research use only [RUO]) that a laboratory elects to use. Most laboratories lack the resources and culture collection to verify every database entry but instead must verify the ability of their system to identify the clinical agents that they most commonly encounter. Due to Select Agent Program regulations, most clinical laboratories do not have access to BT agents and are unable to verify software performance for these agents. Therefore, this study was set up to evaluate the performance of the RUO and FDA-approved software packages offered by Bruker Daltonics (Billerica, MA) and bioMérieux (Durham, NC) using specimens prepared by the tube extraction method and tested in triplicate by eight participating laboratories. In addition, the Security-Relevant (SR) library available on the Bruker instrument was also tested.

RESULTS
Safety study. Overall results of the study are shown in Table 1. Eighty-nine percent of samples contained viable organisms after 1 l of drying of an organism suspension on a sterile coverslip ("Spot" samples). This suggests that whereas drying alone affects the viability of some organisms, it is insufficient to render samples nonviable in the time frame associated with routine sample preparation. Exposure to air for an extended period may also have contributed to the decreased viability of the Clostridium spp. The reagents used in sample preparation (Spot ϩ Matrix) for the direct and extended direct methods appeared to have little inhibitory effect, with 68% and 71% of the samples remaining viable, respectively. Only 11% of the samples that had been exposed to the tube extraction reagents contained viable organisms. Viable organisms were present on the target for 18% of the samples prepared using the direct and extended direct methods. No viable organisms were found following the tube extraction.
Accuracy study. Two experiments were performed to eliminate storage and pooling of extracts as potential sources of error. Isolates of Staphylococcus aureus ATCC 29213, Pseudomonas aeruginosa ATCC 27853, and Clostridium perfringens ATCC 13124 were extracted and tested in triplicate. The remaining extract was divided, stored at Ϫ20°C, and retested after 30 and 45 days in storage. The identification scores compared across time showed coefficients of variation of 3.9% for S. aureus, 2.8% for P. aeruginosa, and 1.3% for C. perfringens, indicating little deterioration of the extracts during storage. To demonstrate that pooling of extracts did not alter results, isolates of Streptococcus pneumoniae ATCC 49619, Burkholderia cepacia ATCC 17765, and Moraxella catarrhalis Since the only manufacturer-approved specimen preparation technique for Vitek MS system is the direct method, 50 random isolates submitted to the laboratory for identification were tested by the direct and tube extraction methods. Results showed that the two extraction methods yielded the same identification 96% of the time, with each sample preparation method providing one incorrect identification.
Identification accuracy results for the BT agents are shown in Table 2 and Table 3 for the Bruker and Vitek platforms, respectively. Results for near neighbors are shown in Table 4 and Table 5. Some participants failed to test the extracts a single time and then reanalyze the spectra using the other libraries; in some instances, the laboratories prepared new targets for each software library. Any result reported as representing no peaks or inadequate spectra was eliminated from data analysis.
The Bruker IVD and RUO software did not correctly identify any of the BT agents. This is to be expected since BT agents are not included in the software. However, the IVD and RUO libraries incorrectly identified 11.9% and 16.2% of the isolates, respectively. The Vitek IVD library did not correctly identify any of the BT agents but incorrectly identified 16.2% of the isolates. While several of the BT agents are in the RUO library, only 3.3% of extracts were correctly identified; F. tularensis was the only BT agent identified, with 11 of 45 (24.4%) extracts identified correctly. The RUO library incorrectly identified 7.5% of the extracts. Y. pestis was the most frequently misidentified organism, with 60.7% and 33.3% extracts being identified as Y. pseudotuberculosis by the IVD and RUO software, respectively. While the RUO software did not identify any of the Brucella extracts to the species level, it did correctly identify them to the genus level 56.9% of the time.
The IVD and RUO libraries misidentified 2.3% and 7% of the near-neighbor extracts, respectively. The RUO library incorrectly identified 55.6% of Francisella novicida extracts as F. tularensis.

DISCUSSION
MALDI-TOF MS presents clinical laboratories with a new tool that has the potential to rapidly and accurately identify organisms in a cost-effective manner; however, this technology also presents new challenges. Highly pathogenic organisms may present hazards to the laboratory staff during the preparation and testing of samples. Validation of identification systems also poses a challenge in that access to many highly patho-   Brucella melitensis (2) No identification Brucella spp.

(2)
Listeria seeligeri/Oligella ureolytica 43.2/56.7 (1) genic organisms is regulated by the Select Agent Program and, thus, these agents are not available to clinical laboratories to assess the limitations of the software libraries.
Use of MALDI-TOF MS for the rapid identification of naturally or intentionally released risk group 3 organisms in a biosafety level 2 (BSL2) environment makes inactivation a critical step to limit exposure risk for laboratorians. In addition to the sample preparation methods described by instrument manufacturers, several other methods have been proposed to inactivate highly pathogenic organisms, including the use of trifluoroacetic acid (TFA), ethanol, gamma irradiation, centrifugation, and filtration. Nonetheless, there are disadvantages associated with these methods. Treatment with 80% TFA for 30 min, for instance, has been shown to inactivate vegetative cells but failed to consistently kill spores of B. cereus and B. subtilis (5). The addition of centrifugation and filtration through a 0.22-m-pore-size membrane removed all remaining viable organisms and spores. However, the final preparation required a 1:10 dilution in water, which may decrease analytical sensitivity, and the high toxicity of TFA may also preclude its use in clinical laboratories. Gamma irradiation has been shown to successfully inactivate organisms (10, 11), but decreased peak intensities led to lower identification scores, and the availability of a ␥ source in clinical laboratories makes this approach untenable. Exposure to 70% ethanol for 5 min has been shown to inactivate non-spore-forming near-neighbor organisms but failed to inactivate B. cereus and C. sporogenes (4). TFA extraction and a tube extraction method utilizing ethanol-formic acid-acetonitrile rendered 14 of 15 bacterial strains nonviable; B. anthracis A100 survived, but all extracts were nonviable following the addition of centrifugal filtration through a 0.1-m-pore-size filter (6). Tracz et al. (7) showed that 3 of 31 Bacillus spp., including one B. anthracis strain and two B. thuringiensis strains, survived tube extraction, but the extracts were rendered nonviable following the addition of a filtration step.
This study showed that some of the BT agents survived the direct and on-plate formic acid sample preparation techniques widely used by clinical laboratories. These results differ from findings by Cunningham and Patel (4), who reported that all isolates tested were nonviable following treatment with 70% formic acid (on-plate sample preparation). However, the studies differed in the isolates tested. Vitek's on-plate formic acid sample preparation method utilizes 25% formic acid, whereas the present study used 70% formic acid as recommended by Bruker; thus, the results of organism inactivation using 70% formic acid may differ from those obtained using 25% formic acid. The operator's technique could also influence organism viability if the spotted organism is not completely covered by formic acid or the spot is not entirely encased by matrix. While none of the isolates tested in this study survived the tube extraction method, other investigators (6, 7) have shown that some isolates of B. anthracis and B. thuringiensis may survive the tube extraction procedure; those investigators recommended the use of a filtration step for added safety. The results of this and previous studies indicate that several inactivation procedures may be successful; however, intraspecies differences may make one strain more resistant to inactivation than others. The addition of a filtration step combined with the manufacturer's tube extraction procedure provides an increased margin of safety to ensure that samples contain no viable organisms. On the basis of this information, the American Society for Microbiology document "Sentinel Level Clinical Laboratory Protocols for Suspected Biological Threat Agents and Emerging Infectious Diseases" (www.asm.org/index.php/science -skills-in-the-lab/sentinel-guidelines) recommends that laboratories using MALDI-TOF MS for identification of suspect BT agents should use the tube extraction method followed by filtration through a Յ0.2-m-pore-size filter for suspected BT agents. Filtration of DNA preparations of B. anthracis spores for PCR through the use of a 0.1-m-pore-size filter prior to testing has been shown to render samples safe for testing outside BSL3 containment (12); this practice is widely used by state public health laboratories participating in the Laboratory Response Network (LRN) and should be extended to extracts of suspected highly pathogenic organisms prepared for MALDI-TOF MS.
Accurate assays for the identification of highly pathogenic organisms are critical for timely treatment, for decreasing laboratory exposures, and for instituting appropriate public health interventions that may be associated with an intentional release. In the United States, naturally occurring cases of brucellosis (115 in 2010), tularemia (314 in 2015), and plague (16 in 2015) reported to CDC pose additional hazards and diagnostic challenges for clinical laboratories. A European interlaboratory ring trial testing the ability of MALDI-TOF MS to identify six BT agents and four near neighbors showed an average accuracy of 77% (11). However, in 5 of the 12 participating laboratories that utilized Bruker software alone, the accuracies were 46.7% for six BT agents and 50% for the near neighbors. For the single Vitek participant, the accuracies were 66.7% for the BT agents and 100% for near neighbors. Another study (7) that looked at 57 isolates representing nine potential BT organisms showed an accuracy of 61.4% using the Bruker RUO and SR libraries. Those studies are in general agreement with the findings of the present study. In addition, both of those studies showed that the combination of the manufacturers' libraries and in-house libraries improved accuracy to Ͼ93% (11) and 100% (7). The results of those studies indicate the need for additional spectra in the commercial databases to improve identification accuracy.
Accurate results employing mass spectrometry require good sample preparation and a well-developed database. Several studies have looked at improving accuracy by optimizing specimen preparation and altering the manufacturer's criteria for genusand species-level identification. Studies have suggested scores of Ն1.7 for Grampositive organisms (13), Ն1.9 for enteric Gram-negative bacilli (14), Ն1.8 (15), and Ն1.9 (16) for anaerobic bacteria and even species-specific cutoff scores (17) to improve identification accuracy. The accuracy of identification reported in the present study might also increase if cutoff scores were optimized. The mean score for many the BT agents was near the cutoff value of Ն2.0, and a decrease to even Ն1.9 would have significantly improved identification to the species level. Identification accuracy can be improved by using phenotypic characteristics combined with MALDI-TOF results to make a final identification. CLSI recommends the use of Gram stain characteristics, colony morphology, rate of growth, culture conditions, and biochemical and/or antimicrobial susceptibility test (AST) results (9). For example, in this study, a Gram stain performed for the Vitek extracts would have detected 19% of the IVD misidentifications and 36.4% of the RUO misidentifications.
The sample preparation method may also have affected the accuracy of the study results. While our limited data suggest that the ethanol/formic acid extraction method employed here is compatible with the Vitek MS system, further studies to validate this extraction method are warranted. The interlaboratory effects of sample preparation technique were minimized in this study since all of the extracts were prepared in a total of four laboratories; however, storage and handling of the extracts could affect spectral quality. Our study showed no effect on identification scores for up to 45 days when extracts were stored at Ϫ20°C; however, some of the study participants analyzed extracts well beyond 45 days of storage. This may have affected spectral quality for some extracts, decreasing specimen scores and resulting in lower accuracy. However, it should also be noted that extracts for C. perfringens, C. septicum, C. sordellii, B. cepacia, and Y. enterocolitica were correctly identified by all Bruker participants and that B. megaterium, C. perfringens, and C. septicum were correctly identified by all Vitek participants regardless of the time between sample preparation and analysis. When the data for the identification of BT agents by the SR library were reanalyzed based on test date, we found that 13.8% (96/697) of the extracts were tested beyond 45 days.
Inclusion of only those extracts tested within 45 days increased the overall accuracy from 52.5% to 55.2%. While the identification accuracy for most agents increased, the accuracy for Clostridium botulinum and B. mallei decreased slightly. This suggests that testing beyond 45 days resulted in decreased spectral quality for some extracts whereas others were left unaffected. Additional studies conducted at a single laboratory are necessary to determine how storage time/temperature and genus/species affect spec-tral stability. These studies may have a significant impact on future multilaboratory studies and proficiency testing using prepared extracts.
The Bruker and Vitek IVD databases both exclude BT agents; Vitek covers some of the agents in the RUO database, and Bruker requires purchase of a separate database to identify these agents. While the Vitek RUO database failed to identify most of the agents to the species level, it provided genus-level (e.g., Brucella, Burkholderia), grouplevel (B. cereus group), or split-organism (B. thuringiensis/B. cereus/B. mycoides) identifications for some of the organisms. This level of identification may decrease exposure risks in clinical laboratories if they recognize software limitations and use appropriate supplemental testing procedures such as those outlined in the American Society for Microbiology (ASM) document "Sentinel Level Clinical Laboratory Protocols for Suspected Biological Threat Agents and Emerging Infectious Diseases." For example, 51 health care workers were exposed to B. melitensis in two incidents within 2 months in New York City (18), in part because both laboratories attempted identification using MALDI-TOF MS and the genus Brucella was not part of the instrument's database. Manufacturers should consider inclusion of the BT agents in their IVD/RUO databases for identification to the genus level or the species level or both together and specific instructions that results should be confirmed by other methods. In this study, the most frequently misidentified organism was Y. pestis. Differentiation from Y. pseudotuberculosis is problematic because Y. pestis evolved from Y. pseudotuberculosis only recently (19). Until that differentiation is possible, manufacturers may want to consider a disclaimer for the identification of both organisms. Until databases are updated, laboratories should clearly note limitations in their procedures and may want to consider the use of well-curated external databases like CDC's MicrobeNet. Currently the Bruker RUO library offers a "matching hints" disclaimer, which in some instances may assist a user in electing to follow the ASM recommended guidelines. However, the "matching hints" disclaimers also indicate the use of repeat testing with fresh material for Bacillus spp., which may increase exposure risk.
Implementation of MALDI-TOF MS in clinical laboratories poses some significant issues that should be addressed in a risk assessment and with validation studies. Laboratories should consider the hazards that preparing and testing potential BT agents and other agents easily transmitted by aerosol pose for health care workers. Since BT agents are not readily available for validation studies, laboratories should also be aware of software limitations and common misidentifications. Partial identifications or misidentifications resulting from the use of IVD (including unclaimed identifications) and RUO (20) with the following modifications: (i) to obtain uniform spotting, samples for the direct colony and on-plate extraction methods were prepared in high-performance liquid chromatography (HPLC)-grade water with turbidity equivalent to a 1 to 2 McFarland standard; (ii) samples for the tube extraction were prepared in HPLC-grade water with turbidity equivalent to a 3 to 4 McFarland standard; and (iii) 1-l aliquots were spotted onto sterile 15-mm-diameter no. 1 glass coverslips instead of the MALDI target.
A total of nine coverslips, representing a MALDI target, were prepared for each organism, and three were used for each extraction method at five participating laboratories. The coverslips were allowed to air dry. One coverslip was placed into 10 ml of brain heart infusion (BHI) broth supplemented or conditioned as needed to support organism growth. This coverslip (referred to as the "Spot" coverslip) served as a control to determine the effects of drying and air exposure (for anaerobes) on viability. A second coverslip was placed into a tube of BHI broth that contained all the reagents used in the extraction (for example, 1 l of 70% formic acid and 1 l of ␣-cyano-4-hydroxycinnamic acid [HCCA] matrix for the on-plate extraction samples). This coverslip (referred to as the "Spot ϩ Matrix" coverslip) was used to determine growth inhibition due to inadequate dilution of the extraction reagents in BHI broth. For the third coverslip, the extracted sample was overlaid with 1 l of HCCA, allowed to air dry, and then placed into 10 ml BHI broth. This coverslip (referred to as the "Target" coverslip) represented a sample ready for MALDI analysis. The tubes were incubated using appropriate temperatures and conditions for 7 days (21 days for Brucella). Any tube showing turbidity was subcultured and the growth identified by Gram stain and morphology.
Accuracy study. Whenever possible, the strains utilized for the study were clinically relevant organisms selected from the inclusivity and exclusivity panels approved by the AOAC International Stakeholder Panel on Agent Detection Assays (SPADA) (21-23; AOAC International, unpublished data). No SPADA panels were developed for Brucella species or Clostridium botulinum, so strains of these species were selected based on availability and clinical relevance. BT agents used for the study are listed in Table  S1 in the supplemental material along with their relationship to the SPADA panels and the presence of each genus and species in the software libraries tested.
Each isolate was prepared by performing Bruker's tube extraction in 10 replicates followed by filtering each extract through a 0.1-m-pore-size centrifugal filter (Millipore Ultrafree-MC-VV Durapore polyvinylidene difluoride [PVDF]) for 2 min at 7,050 ϫ g. The resulting extracts were pooled, mixed, divided into aliquots in 50-l volumes, and stored at Ϫ20°C. Ten percent of the final pooled volume or 100 l was tested to confirm sterility. Extracts were shipped on dry ice to the testing laboratories.
Participating laboratories were asked to test all extracts in triplicate on the same run using a freshly cleaned or disposable target within 45 days of extract preparation. A 1-l volume of extract was applied to the target, allowed to dry, and then overlaid with 1 l HCCA matrix. Spectra were generated using the run conditions programmed by the manufacturers. Six laboratories tested extracts on a Bruker MALDI Biotyper (Bruker Daltonics, Billerica, MA) equipped with one or more of the IVD (claim 1), RUO (claim 3; n ϭ 5,687), and Security-Relevant (claim 1; n ϭ 123) software libraries. Three laboratories tested extracts on the Vitek MS system (bioMérieux Inc., Durham, NC) equipped with the IVD (version 2.0) and RUO (version 4.12) software libraries.
Laboratories with IVD software were instructed to test extracts using each manufacturer's IVD protocol. Following completion of the run, the spectral data generated from the run were reanalyzed using all available software packages but performing the analysis with only one software package at a time. For laboratories with RUO software, spectra were generated in RUO mode. Each laboratory reported results using a spreadsheet listing the date tested, the software package used, the identification result, and the sample score. An identification result was considered accurate to the genus and species levels if the sample score was Ն2.0 for the Biotyper or the level of identification was Ն60% for the Vitek MS.