Evaluation of the novel culture-based AtbFinder test-system for the selection of optimal antibiotics for critically ill patients with polymicrobial infections within 4 h

: Here we describe the validation of a new phenotypic culture-based AtbFinder method for rapid selection of antibiotics in vitro using specimens with mono- and polybacterial infections. AtbFinder, which can be applied to any type of non-blood tissue, does not require isolation of pure bacterial cultures. The method uses a novel TGV medium that allows more rapid bacterial growth of Gram-positive and Gram-negative monoisolates compared with that achieved with conventional laboratory media and demonstrated overall sensitivity, specificity, PPV, NPV values of 99.6%, 98.1%, 98.5%, and 99.4%, respectively, already after 4h. For polymicrobial infections, AtbFinder utilizes a novel paradigm of the population response to antibiotics, enabling bacterial growth in the form of a mixed microbial community and selecting the antibiotics targeting not only the principal pathogen, but also those bacteria that support their growth. TGV medium allowed culturing a more diverse set of bacteria from polymicrobial biospecimens, compared with that achieved with the standard media and enabled, already within 4h, accurate selection of the antibiotics that completely eliminated all cultivatable bacteria from clinical samples. In conclusion, AtbFinder system may be a valuable tool in improving antibiotic selection, enabling targeted empirical therapy and accurate antibiotic replacement, which is especially important in high-risk patients.


Introduction
Antibiotic therapy is typically started on an empirical basis, because the causative organism is not identified in an appreciable proportion of patients (1)(2)(3). However, empirical antibiotic therapy is inadequate in over 25% of cases, with 8-12% of patients receiving antibiotics that are ineffective (4).
For these patients, antibiotic therapy must be adjusted following antimicrobial susceptibility testing (AST) by culture-based and/or molecular biology methods. However, even following the switch of the empirically chosen antimicrobial, in over 10% of cases, the newly selected antimicrobial remains turnaround time to obtain the results and enabled characterization of polymicrobial infections (35,36). However, these assays also have a number of limitations. In particular, although the identification of bacteria helps to determine the therapeutic strategy, it does not really inform on the extent of resistance to antibiotics (6). Moreover, the methods based on 16S RNA sequencing are not very accurate as many bacterial species share similar or identical 16S rRNA sequences (37,38).
Some molecular biology methods are based on the detection of antimicrobial resistance genes (39,40). However, the reliance on the presence of resistance genes may be misleading with regard to accurate antibiotic selection as not all resistant genotypes result in resistant phenotypes (6).
Furthermore, these methods can only detect known antibiotic resistance genes but may omit those, which have not yet been discovered (39,41). The available resistance markers are not sufficiently comprehensive to provide clinically actionable results and may lead to the usage of either unnecessarily broad spectrum antibiotics or those without the sufficiently strong therapeutic effect.
Consequently, culture-based AST that predicts not only the resistance but also susceptibility to different drugs is still considered the gold standard for the identification of the most appropriate antibiotic for various infections (42).
Here, we evaluated a novel culture-based AtbFinder system that utilizes the recently developed TGV medium that supports simultaneous growth of a significantly more diverse set of bacteria from biological samples compared with that achieved with conventional laboratory media and uses a novel paradigm for rapid phenotypic-based antibiotic selection for monobacterial and polybacterial clinical isolates.
The AtbFinder approach is based on a novel paradigm for the selection of effective antibiotics that considers not only monobacterial infection but also polybacterial cooperative interactions at the infection site.

Study samples and laboratory settings
Study specimens were obtained from the Human Microbiology Institute (New York, NY, USA), maintained at about 4 °C (wet ice), but not frozen, and sent to TGV-Biomed within 8 h after sampling.
Bronchoalveolar lavage (BAL) or sputum samples obtained from patients with chronic obstructive pulmonary disease (COPD), community-acquired pneumonia (CAP), ventilatorassociated pneumonia (VAP), hospital-acquired pneumonia (HAP), cystic fibrosis (CF) were tested by using both state-of-the-art methods and AtbFinder system. Bacterial identification to the species level from biological specimens was done by using subcultures on AtbFinder or LB medium (Oxoid, UK). Isolates were examined for purity by light microscopy (Leica 2500DM, Leica Microsystems, Germany). To exclude the presence of mixed bacterial cultures, the isolates were assessed from at least 10 fields of view (58). After the subcultivation of every mixed bacterial culture up to monocultures, subsequent biochemical identification and matrix assisted laser desorption/ionization-time of flight (MALDI-TOF) mass spectrometry (Microflex LT, Bruker Daltonics, Germany) analysis were performed according to the manufacturer's instructions after 24 h of growth.
The complete 16S rRNA gene sequencing of bacterial colonies was also used to identify the isolates.
A PCR was performed with the general bacterial primers 27f (5-AGAGTTTGATCCTGGCTCAG-3) and 1492r (5-GGTTACCTTGTTACGACTT-3 Identifications to the genus and species-level were done on the basis of ≥97% and 99% identity of the 16S rRNA gene sequence to the reference, respectively (61).

AtbFinder system and interpretation of results
AtbFinder system is a multi-well (12,24  The presence of microbial growth was identified with the naked eye and confirmed with a stereoscopic microscope (Leica S6, Leica Microsystems, Germany). The signs of early bacterial growth, namely hemolysis, appearance of film, and microcolonies, in the wells with antibiotics were compared with signs of bacterial growth in control wells.
Microbial growth in any well meant that in the biospecimens, there were microorganisms resistant to the particular antibiotic in the nutrient medium in the well; therefore, the antibiotic was categorized as "ineffective". An absence of bacterial growth in the well meant that the antibiotic present in the well had killed or inhibited the growth of all bacteria in the biospecimens and was thus categorized as "effective". In rare cases, there was no growth in control wells after 4 h, so those cultures were not analyzed further as is separately described in the Results below.

Gold standard definition
Culture-based in vitro AST was selected as the standard of care method.

Data analysis
The results of antibiotic selection from the AtbFinder system were compared to those obtained with the culture-based gold standard microdilution method. For the purpose of data analysis, the following definitions were used. A true-positive result occurred when both the AtbFinder and standard-of-care methods indicated that the microorganism was resistant to the antibiotic. A truenegative result occurred when both AtbFinder and standard-of-care methods demonstrated sensitivity of the microorganism to the antibiotic. A false-positive result occurred when the AtbFinder system identified a strain as being sensitive to a certain antibiotic, but according to the standard-ofcare method, the organism was resistant to that antibiotic. A false-negative result was recorded when the AtbFinder system suggested that the microorganism was resistant to the antibiotic, whereas the standard-of-care method indicated that it was sensitive. We calculated the values of accuracy, sensitivity, specificity, as well as positive predictive value (PPV) and negative predictive value (NPV) as previously described (65).
The total category agreement (CA) was determined and the results which differed from those obtained by the state of care microdilution method were categorized as described previously (66).
Thus, the very major errors were recorded in the cases when the AtbFinder system indicated that an isolate was susceptible to the antibiotic, whereas according to the standard of care method, the isolate was resistant. The major errors were recorded in the cases when the AtbFinder system suggested that an isolate was resistant to the antibiotic, whereas the standard of care method indicated that the isolate was susceptible. The rate of very major errors was calculated by dividing the total number of very major errors by the total number of strains determined as being resistant and multiplied by 100%. The rate of major errors was determined by dividing the total number of major errors by the number of strains determined as being susceptible and multiplied by 100%.

Bacterial diversity analysis
The variety of bacteria that grew on the media used was characterized by the α-diversity indices such as non-parametric abundance-based coverage estimator (ACE) and Chao 1, indices which were managed and analyzed using R version 3.4.1 software (67).
Species richness (a count of different species), that gave growth on different media were represented on a dot plot, generated by package 'ggplot2' within R version 3.4.1 (67). All statistical analyses were conducted with a significance level of α = 0.05 (P < 0.05).

Statistics
Two-way ANOVA comparisons test was applied within the same data sets to test difference between microbial growth on different media at each time point. GraphPad Prism version 9 (GraphPad Sofware, CA, USA) or Excel 10 were applied for statistical analysis and illustration was used if not stated differently. P values < 0.05 were significant.

Comparison of bacterial growth rate on TGV medium to that on other media
We first performed a validation of TGV agar by using monomicrobial cultures to confirm that it enables faster bacterial growth compared with that afforded by the conventional LB agar, Columbia agar or brain heart infusion (BHI) agar.  pneumoniae strains out of 10 tested that gave visible growth at different time periods on TGV, LB, COL, or BHI.
We observed higher species richness after growth for 4 h on TGV medium as revealed by ACE and Chao 1 indices (28.07 and 33.5. respectively), compared with the values of these parameters after growth on LB agar (ACE = 25.6 and Chao 1 = 21.5) or Columbia agar (ACE = 25.5 and Chao 1 = 24; Fig. 2A, B). Therefore, TGV was the only medium that allowed to detect visible growth of monomicrobial cultures already after 4 h with high accuracy.

Estimation of the diversity of bacteria grown on TGV medium
Next, we applied TGV medium for the studies of polymicrobial infections. We used samples from patients with respiratory infections, which are usually associated with polymicrobial growth.
Bacterial growth on TGV medium was seen within 4 h in 20 out of 20 clinical specimens (100%), whereas microbial growth was detected only in 8 out of 20 samples (40%) grown on LB agar that was used as standard medium. Only subsequent growth for 24 h revealed clear bacterial growth in all samples cultured in LB media.
Next, we analyzed polymicrobial growth in each sample by using TGV and LB media at 4 h and 24 h of culturing. A detailed description of bacteria that gave growth on different media is provided in set of microorganisms from the biological specimens compared with that revealed by the standard medium after 24 h of culturing (Fig. 3). was used.
Notably, in the majority of specimens grown on TGV, we identified more than one well-known pathogen of respiratory tract infections, whereas the standard method allowed the identification of one and, only sometimes, two pathogens. The highest microbial diversity was observed in sputum of the patients with cystic fibrosis (CF) -a disease that is known to be characterized by a particularly complex, mixed lung microbiome (68). For example, from some tissue specimens, we isolated diseases whereby TGV enabled identification of a wider range of known respiratory pathogens compared to that revealed by the standard method.
Notably, by using TGV medium, we identified two previously unknown bacterial species Chryseobacterium mucoviscidosis sp. nov. and Bacillus obstructivus sp. nov. (75,76). These bacteria have not been identified by MALDI-TOF, as they had very low similarity to known species, and their further whole genome sequencing proved that they were previously unknown bacterial species highly enriched in genes encoding pathogenic and antibiotic resistance factors.

Antibiotic selection in monomicrobial cultures using AtbFinder
The  (Tables 2 and Supplementary table 2).   CA values for the AtbFinder were even higher when the period of culturing was extended.  Table 3).
At both time points tested, the level of major errors was very low, indicating that antibiotic-sensitive isolates did not grow on TGV medium with antibiotics added at chosen concentrations. In general, there was no discernible pattern of the errors that could be explained by the nature of microorganisms or antibiotics tested: AST errors occurred in various species and with various drugs. CA values were not statistically different between the antibiotics used.

Antibiotic selection in polymicrobial cultures using AtbFinder
When developing AtbFinder, we assumed that the and green circles represent susceptible determinations ("effective" antibiotics).
(D) Analysis of the discrepancies in the efficacy of antibiotics between the AtbFinder and standard AST methods. Red squares depict antibiotics that according to AtbFinder were predicted to be "ineffective" for specific biosamples but deemed "effective" by the standard AST method.
(E) Bacterial cultures from the discrepant cases outlined in that were suggested as ineffective against to be resistant to certain antibiotics with AtbFinder, and thus experienced growth in the wells with in the presence of thatcertain antibiotic, but were deemed "effective" against identified pathogens by the standard AST method were subcultured from the wells of AtFinder and identified. Bacteria marked with red letters were suggested to be the principal pathogen by the standard AST method and, when grown under the conditions modulated according to the standard AST method, were killed by this specific antibiotic (see Fig 4C). Bacteria marked in blue were not suggested to be the primary pathogen by the standard AST method.
To test this assumption, we used 10 randomly selected samples from the cases of respiratory infections tested before in this study and analyzed particularities of antibiotic selection with AtbFinder (after growth for 4 h) (Fig. 4B) and culture-based gold standard method (microdilution) (Fig. 4C). Notably, along with a broader diversity of bacteria from the specimens grown on TGV of AtbFinder compared to the microbial diversity on LB agar used in a standard method (Fig 4A), we identified a discrepancy between the antibiotics suggested as effective by the AtbFinder and standard methods (Fig. 4D). When the 10 biosamples were each tested against 10 antibiotics (100 evaluations in total), we found 18 cases of discrepancy in the conclusion regarding antibiotic efficacy. In these cases, antibiotics were deemed as effective by the standard AST, but they were suggested to be ineffective by AtbFinder, as the bacteria from those samples could grow in the wells of AtbFinder with those antibiotics (red squares in Fig. 4 D). Then, to identify the bacteria that grew in the wells of AtbFinder with these antibiotics, we performed their follow-up subculturing. As illustrated in Figure 4E, we were able to isolate bacteria that according to standard AST were suggested as the principal pathogen from 11 out of the total 18 discrepancy wells.
That result meant that in 11 out of 100 total antibiotic efficacy evaluations (11%) when a mixed bacterial community grew on AtbFinder, the principal pathogens resisted the antibiotics selected as effective in monobacterial cultures of these pathogens in conditions of the standard AST. Therefore, indicating complicated interspecies interactions in polymicrobial communities when one bacteria is required to support the growth of other microorganisms (Fig 4A). This is in agreement with some recently published data that have shown that the overall response to antibiotics of a mixed community is sometimes the opposite to that of individual bacteria (77).

Discussion
New diagnostic methods for the selection of antibiotics for tailored empirical therapy and for the change of antibiotic therapy that failed in immunocompromised patients are urgently needed (78). Our present experiments evaluated for the first time the performance of the novel AtbFinder system providing culture-based antibiotic selection within a short, 4 h period based on the novel principle to select antibiotics effective against polymicrobial communities from pathological material (Fig. 5). in the well means that the antibiotic present in the well leads to a complete killing or inhibits the growth of all bacteria in the biospecimens and such an antibiotic is categorized as "effective".
The AtbFinder method can be used for initial tailored empirical therapy and for the selection of an appropriate antibiotic to those patients who have not responded to the previous therapy and require change of antimicrobial treatment strategy. Our original hypothesis that the use of the AtbFinder system would provide a rapid and accurate selection of antibiotics for both monobacterial or polybacterial infections sampled directly from clinical specimens, was found to be confirmed experimentally. AtbFinder system allowed faster growth of monobacterial cultures compared with that achieved by the standard method used in clinical diagnostics and demonstrated an increased richness with higher ACE and Chao 1 indices for bacteria that gave growth within 4 h on AtbFinder medium than those recorded for bacteria grown on LB agar or Columbia agar (Fig. 2).
Our study also revealed that AtbFinder system provided accurate antibiotic selection compared to strains leads to the re-growth of other bacteria at the site of the infection unaffected by the initial antibiotic treatment, which may lead to disease progression and complications, e.g., due to the lack of sufficiently strong immune response. Moreover, as it has been shown before, some bacteria that are considered nonpathogenic in the lungs of non-immunosuppressed subjects may be pathogenic in immunosuppressed hosts (80,81). This is why, for some bloodstream infections, sterilization is widely used, particularly in the patients with impaired immune response. Therefore, the fact that AtbFinder system reliably detects both major and minor bacterial pathogens within the site of infection makes it promising for the use in immunocompromised patients. Second, our assessment of antibiotic sensitivity within polymicrobial communities showed that in some cases, antibiotics selected with the standard method did not eliminate even the dominant bacteria in mixed microbial communities. Such false-negative data might result in the selection of ineffective antibiotics and therapy failure (82). The reason for false-positive results generated by the gold standard method may be because this method enables growth of monobacterial cultures but does not take into consideration higher tolerance to antibiotics of bacteria within mixed communities (15). However, "real life" infections are predominantly polymicrobial by nature (10). The resulting survival of the lead pathogen leads to its re-growth and therapy failure. According to our in vitro data, AtbFinder is less prone to such limitations.
Furthermore, the use of the AtbFinder system for the treatment of polybacterial infections allowed selection of not only broad spectrum but also narrow spectrum antibiotics, confirming the presence of complexed interspecies interactions in mixed bacterial communities.
Notably, by using AtbFinder, we have isolated previously unknown bacterial species Chryseobacterium mucoviscidosis sp. nov. and Bacillus obstructivus sp. nov. that possessed several typical virulence factors, such as hemolysins and others, which are found in other respiratory pathogens (76, 76 83). Moreover, in these bacteria, we identified several antibiotic resistance genes that once found in endospore-forming Bacillus spp. are of concern because of the possible spread of antibiotic resistance among sporobiota members (84). The identification of previously unknown bacteria as well as some fastidious species, such as S. milleri, indicates that the developed TGV medium provides unique growth factors and cultivation environment that are not achievable with the standard media (19,85).
The use of the AtbFinder system enables to quickly select antibiotics that are effective for the treatment of each particular condition (i.e., those that kill or inhibit bacterial growth of all microorganisms present in the clinical specimen) without the need to identify precisely the causative agents of the infection or to determine MICs. Importantly, although the AtbFinder test does not allow immediate identification of bacteria on the species level, bacterial cultures that gave growth on TGV medium can be isolated and subsequently identified by using culture-based techniques and any other standard methods for the study of bacteria. AtbFinder system has all benefits of phenotypic culture-based methods, but is 30-54 h faster than standard culture-based diagnostics that requires time-consuming isolation of pure bacterial cultures.
Even compared with direct AST, the AtbFinder method allows antibiotic selection on a much shorter timescale (60). Like direct AST, AtbFinder system enables direct sampling of biological specimens without the need for culturing or time-consuming sample processing. Furthermore, the AtbFinder method can indicate the suitable antibiotic in only 4 h, whereas direct AST requires 18-36 h (and that is why direct AST is not applicable for tailoring empirical antibiotic therapy) (86).
Despite being as fast as some of the molecular biology methods, AtbFinder system lacks main disadvantages of molecular methods based on next-generation sequencing and 16S RNA sequencing, such as overestimation of antimicrobial resistance and inability to inform on antimicrobial susceptibility (37,38). Moreover, there is no need for either any specific equipment, except a thermostat, or for highly trained personnel, because the presence or absence of bacterial growth can be identified by the personnel with basic laboratory skills. In this study, we used AtbFinder system with a set of antibiotics used for the treatment of respiratory infections. However, it can easily be adjusted for the diagnosis of the infections of other parts of the body, such as urinary tract, skin, or soft tissues, by including the antibiotics used for the therapy of these infections in TGV medium.
Moreover, it can be used for the cultivation of bacteria with different oxygen requirements. The presence of bacterial growth was analyzed with naked eye and stereoscopic microscopy, but a more sophisticated device for visual monitoring can be used to increase the analysis accuracy. We believe that AtbFinder system may become a novel and valuable tool in improving antibiotic selection, with as little as 4 h turnaround time. In terms of the possible laboratory implementation, AtbFinder dramatically shortens the testing routine by allowing appropriate antibiotic selection on the same day. Therefore, AtbFinder enables more effective antibiotic selection for the targeted empirical therapy and accurate antibiotic replacement, especially in high-risk immunocompromised patients.
Future studies will be necessary to investigate clinical efficacy of the AtbFinder system and the clinical impact of its use alone and as an auxiliary method for standard diagnostics.

Conclusions
This section is not mandatory, but can be added to the manuscript if the discussion is unusually