A novel flow cytometry assay based on bacteriophage-derived proteins for Staphylococcus detection in blood

Bloodstream infections (BSIs) are considered a major cause of death worldwide. Staphylococcus spp. are one of the most BSIs prevalent bacteria, classified as high priority due to the increasing multidrug resistant strains. Thus, a fast, specific and sensitive method for detection of these pathogens is of extreme importance. In this study, we have designed a novel assay for detection of Staphylococcus in blood culture samples, which combines the advantages of a phage endolysin cell wall binding domain (CBD) as a specific probe with the accuracy and high-throughput of flow cytometry techniques. In order to select the biorecognition molecule, three different truncations of the C-terminus of Staphylococcus phage endolysin E-LM12, namely the amidase (AMI), SH3 and amidase+SH3 (AMI_SH3) were cloned fused with a green fluorescent protein. From these, a higher binding efficiency to Staphylococcus cells was observed for AMI_SH3, indicating that the amidase domain possibly contributes to a more efficient binding of the SH3 domain. The novel phage endolysin-based flow cytometry assay provided highly reliable and specific detection of 1–5 CFU of Staphylococcus in 10 mL of spiked blood, after 16 hours of enrichment culture. Overall, the method developed herein presents advantages over the standard BSIs diagnostic methods, potentially contributing to an early and effective treatment of BSIs.

Bloodstream infections (BSIs) are severe diseases caused by the presence of microorganisms, mainly bacteria, in blood and are characterized by high morbidity and mortality 1,2 . The BSIs and associated organ dysfunctions (sepsis or septic shock) remain a life-threating disease due, partially, to the inability to rapidly detect and identify the responsible pathogens 3,4 .
Staphylococcus spp. are Gram-positive facultative anaerobic bacteria that frequently colonize the human body 5,6 . These pathogens are becoming increasingly resistant to antibiotics and are well-established in both community and healthcare environments, being commonly isolated in intensive care units (ICU) 6,7 . Staphylococcus aureus is a common cause of a variety of infections, from superficial skin infections to life-threatening diseases, including necrotizing pneumonia 8 , infective endocarditis 9 and BSIs 10 . Coagulase-negative staphylococci (CoNS) have also been described as harmful to humans, causing several infections, particularly in patients with implanted medical devices 6 .
The empirical antibiotic therapy remains the standard of BSIs treatments 11 and its correct use within the first hour after the recognition of the BSI is recommended by the Surviving Sepsis Campaign Guidelines 11 and was reported as having a great impact on the patient survival rate 12 . Nevertheless, the extensive use of broad-spectrum antibiotics and the large number of patients having negative blood culture samples and thus receiving unnecessary antibiotic treatment, are important contributors to the increase of antimicrobial resistance [13][14][15] . Thus, sensitive, rapid, cost-efficient and specific detection of pathogens in blood, followed by antimicrobial testing, is critical to de-escalate empirical antibiotic therapy and decrease the negative impact of BSIs 2,14,16 .
Blood cultures remain the reference standard for the detection of bacteria causing sepsis 17 . Generally, blood samples are collected and aseptically inoculated in bottles with specific media for aerobic and anaerobic 1 Centre of Biological Engineering, University of Minho, Campus de Gualtar, 4710-057, Braga, Portugal. 2 International Iberian Nanotechnology Laboratory (INL), Av. Mestre José Veiga s/n, 4715-330, Braga, Portugal. *email: carla. carvalho@inl.int microorganisms. These bottles are then incubated either in manual or in automatic systems that continuously monitor microbial growth 17 .
The conventional culture methods for diagnosis of BSIs involve sub-culturing and Gram staining upon blood-culture positivity, followed by phenotypic methodologies for bacterial identification and antibiotic susceptibility testing. These procedures can be accurate and reliable but are laborious and time-consuming 18 . In the last decade, other detection techniques have emerged as alternatives to conventional culture methods for the detection of BSIs, directly from positive blood cultures or from whole blood, and have been improved the time needed for pathogen identification. These include the Polymerase Chain Reaction (PCR) 19,20 , Peptide Nucleic Acid Fluorescence In Situ Hybridisation (PNA-FISH) 21,22 , Matrix-Assisted Laser Desorption Ionization Time-of-Flight Mass Spectrometry (MALDI-TOF MS) 23 and DNA microarrays 24 . However, these methods present some drawbacks, namely: PCR-associated amplification problems (such as PCR inhibitors) 25 , unspecific hybridization, which can be caused by the human DNA interference with primers and probes 25,26 , infidelity in DNA replication, interference of non-microbial material 17,25 , limited number of available probes 18 , the results obtained are complex and difficult to interpret 26 , and are unable to distinguish between live and dead cells leading to the occurrence of false positives 25,26 . Moreover, pathogen detection directly from blood samples remains a challenge due to the innumerable blood components that can interfere in the analysis 25,26 and to the low bacterial load normally present in the blood from patients with BSIs (1 to 100 CFU mL −1 ) 26,27 . Consequently, most of the detection methods for BSIs are dependent on blood cultures to increase the number of pathogens before the diagnostic test can be conducted 17 .
A promising approach for bacterial detection is the use of bacteriophages (phages) or phage-derived proteins as specific probing elements in conjugation with measurement techniques or biosensors. Phages are viruses that infect bacteria with high host specificity 28 . At the end of their life cycle, phages produce enzymes, called endolysins, to degrade the bacterial cell wall for the release of progeny virions. These proteins have been considered valuable tools to detect and control bacterial infections [29][30][31][32][33] . Endolysins from phages infecting Gram-positive bacteria present a modular structure composed of at least one enzymatic catalytic domain (ECD) and one cell binding domain (CBD) 29 . Most staphylococcal phage endolysins have three distinct domains: two ECDs, namely an N-terminal cysteine histidine-dependent amidohydrolase/peptidase (CHAP) domain and a central N-acetylmuramoyl-L-alanine amidase (Ami_2 or Ami_3) domain; and a C-terminal bacterial src-homology 3 (SH3b) domain as CBD 34,35 . Generally, the C-terminal CBD is responsible for directing the enzyme to its substrate by recognizing specific ligands on the bacterial peptidoglycan or other cell wall-associated molecules with extreme specificity and affinity 30,36 . Therefore, CBDs have been used as probing elements for the rapid detection or concentration of bacteria 32,37,38 , mainly foodborne pathogens 31 . The CBDs binding spectra are usually broader than the host ranges of the corresponding phage and can encompass entire bacterial genera 39 and their binding affinity to bacterial cells is equal or even higher than antibodies 36 . Moreover, phage proteins have shown increased stability to temperature and pH variations, less propensity to aggregation and a faster, easier and less expensive production in comparison with antibodies 40,41 .
Flow cytometry is a multiparametric high-throughput technique which enables the detection and quantification of cells or particles individually in a flow, allowing a rapid and effective analysis of their physical and chemical properties within a population 42 . The cells are often labelled with fluorescent markers, therefore allowing the discrimination of different populations in a sample 42 . In the past years, this technique has been widely used for pathogen detection purposes, although the assays that have been described present some drawbacks, such as being laborious and expensive 43 or lack the ability to detect pathogens in biological samples 44,45 .
In this study, we describe a novel flow cytometry assay based on phage-derived proteins for the detection of Staphylococcus in blood. For this, we exploited different domains at the C-terminus of the S. aureus phage endolysin E-LM12 46 (Amidase, SH3 and Amidase+SH3) in order to select the truncation that presents the higher binding affinity and specificity for Staphylococcus spp. to be used as a specific probe in the detection assay. To the best of our knowledge, this is the first detection method described in the literature, in which a CBD was used as a recognition molecule in a flow cytometry assay, enabling the sensitive and specific detection of Staphylococcus in blood.

Results
Bioinformatics analysis of E-LM12. After identifying the endolysin gene E-LM12 (gp159) from the S. aureus vB_SauM-LM12 phage genome 46 , a BLAST analysis revealed that E-LM12 shows high similarity with endolysins from several staphylococcal phages, including Staphylococcus phage MCE-2014 (100% coverage, 95% identity) 47 , Staphylococcus phage phiIPLA-RODI (100% coverage, 94% identity) 48 and Staphylococcus phage vB_SauM_0414_108 (99% coverage, 95% identity) 49 . Also, no similarity was found with endolysins from phages infecting other bacterial genera. The search for functional domains predicted the existence of three domains ( Fig. 1): an N-terminal CHAP catalytic domain, a central Amidase_2 catalytic domain and a C-terminal SH3b_5 cell binding domain 46 . In the same region of the Amidase_2, a peptidoglycan recognition protein (PGRP) domain was also predicted.
Functional analysis of the E-LM12 endolysin C-terminus. Three different fragments from the C-terminus of the E-LM12 endolysin 46 were cloned (AMI, SH3 and AMI_SH3) fused to a green fluorescent protein (GFP), originating the recombinant proteins GFP-AMI, GFP-SH3 and GFP-AMI_SH3 (Fig. 1). The binding ability of these proteins was assessed by epifluorescence microscopy through the observation of cells emitting green fluorescence. The results showed that the fusion proteins GFP-SH3 and GFP-AMI_SH3 were capable of binding to S. aureus cells, with the GFP-AMI_SH3 clearly displaying a higher binding ability than the GFP-SH3, as observed by the intensity of the green fluorescence at the microscope ( Fig. 2A,B). Conversely, no fluorescent cells were observed when the GFP-AMI protein (Amidase domain) was added to the host bacteria (Fig. 2C).
Specificity and sensitivity of GFP-AMI_SH3 and comparison with the lytic activity of LM12 phage and E-LM12 endolysin. Considering that the GFP-AMI_SH3 revealed a higher binding ability to S. aureus cells, the binding spectrum and specificity of this protein was evaluated by epifluorescence microscopy. For this, the protein was incubated with strains of Staphylococcus spp. and other Gram-positive and Gram-negative bacteria (Table 1). Simultaneously, bacterial cell samples without GFP-AMI_SH3 and the GFP individually were used as negative controls. All the strains tested and the respective results are summarized in Table 1.
The results demonstrated that the GFP-AMI_SH3 binding spectrum covered all staphylococcal strains tested including 12 S. aureus (representative example in Fig. 2D) and 10 other staphylococcal species (example in Fig. 2E), including CoNS. Some weak binding was visualized for Enterococcus faecium LMV-0-42, Enterococcus faecalis LMV-0-39 and Streptococcus pneumoniae R6st (see examples in Supplementary Fig. S1 online). In these cases, the fluorescence intensities were lower than the obtained for Staphylococcus species and its distribution was not homogeneous along the bacterial cell wall surfaces, with some intense dots randomly distributed along the cells. In contrast, no fluorescent cells were detected on the other 17 non-Staphylococcus spp. strains tested with GFP-AMI_SH3, including Gram-positive and Gram-negative bacteria (example in Fig. 2F). Also, control experiments with unlabelled cells and GFP individually did not display green fluorescence.
The binding affinity of GFP-AMI_SH3 was compared with the lytic activity of LM12 phage using the same strains ( Table 1). The phage showed a lytic effect against all Staphylococcus spp. strains tested, presenting "lysis from without" (LFW) 50 on five Staphylococcus strains, and was unable to lyse the S. haemolyticus strain. The phage also revealed LFW when tested against E. faecalis LMV-0-39 and E. faecium LMV-0-42 but did not lyse the other non-staphylococcal strains ( Table 1). The E-LM12 endolysin showed a lytic activity similar to the binding affinity of GFP-AMI_SH3, being able to lyse all the Staphylococcus strains tested (Table 1) when concentrations equal to or higher than 5 µM were used.

Flow cytometry assays. Exploitation of the C-terminus E-LM12 endolysin domains as recognition mole-
cules. In order to select the recognition molecule to be used on the flow cytometry assays, the three fusion proteins (GFP-AMI, GFP-AMI_SH3 and GFP-SH3) were incubated with S. aureus cells and samples were analysed using a flow cytometer equipped with a laser capable of exciting the GFP. The cells that were decorated with the proteins emitted green fluorescence, which was detected on the channel FL1, therefore allowing for the discrimination between labelled and unlabelled cells.
The results indicate that the GFP-AMI_SH3 showed the highest binding efficiency, decorating most of the S. aureus cells (about 99% of the population was in the positive quadrant) (Fig. 3A) when compared with the GFP-SH3 that stained only 39% of the population (Fig. 3B). Moreover, the intensity of the green fluorescence (FL1) of the positive population when using the GFP-AMI_SH3 protein was much higher compared with that using the GFP-SH3 protein, corroborating the results of the fluorescence microscopy assays. Conversely, the GFP-AMI revealed a negligible binding (about 16% of the population) (Fig. 3C), which was probably due to some sample residues or free protein that was not completely washed since no green fluorescent cells were detected by fluorescence microscopy (Fig. 2C).
Considering that the GFP-AMI_SH3 revealed a higher binding ability to S. aureus cells than the other fragments tested, this protein was chosen for further experiments.
Evaluation of detection parameters. Several detection parameters were assessed on the flow cytometry assays, using the GFP-AMI_SH3 as a recognition molecule. These comprised the specificity and sensitivity against staphylococcal and other Gram-positive and Gram-negative strains (Table 1) and the detection limit.
The assays clearly demonstrated that the developed method was able to specifically detect all the Staphylococcus spp. strains tested, including S. aureus, S. epidermidis, S. equorum, S. haemolyticus, S. hominis and S. warneri ( Table 1). As depicted in Fig. 4A, the histograms of the Staphylococcus spp. strains were in the right side with fluorescence values beginning in 100 a.u. while the histogram of a non-staphylococcal bacteria was shifted to the left. Moreover, the fluorescence intensity means were similar for S. aureus Sa12 and other staphylococcal strains (means of 2,142.4 ± 373.2 a.u. and 1,665.4 ± 241.2 a.u., respectively), but were considerable different from the non-Staphylococcus strains tested (mean of 7.7 ± 4.4 a.u.) (Fig. 4B). Also, mean fluorescence intensity from labelled Staphylococcus spp. strains was around 200-fold higher than the value obtained from unlabelled S. aureus cells (NL), used as negative control (mean of 7.4 ± 2.1 a.u.) (Fig. 4B).
For the determination of the detection limit, different concentration of S. aureus Sa12 cells (ranging from 1 × 10 4 CFU mL −1 to 1 × 10 8 CFU mL −1 ) incubated with GFP-AMI_SH3 were analysed through the flow cytometer. Based on these assays, the minimum concentration of bacteria for which it was possible to consider a positive signal, significantly different from the controls (unlabelled cells and labelled non-Staphylococcus strains), was 10 5 CFU mL −1 (Fig. 4B).
Detection of S. aureus in spiked blood samples. The flow cytometry assay was adapted and optimized in order to assess its ability to specifically detect S. aureus in artificially seeded blood samples. For this, as a proof-of-concept, horse blood was inoculated with 1 to 5 CFU mL −1 of S. aureus Sa12, followed by an enrichment for 16 hours. Samples were processed following an optimized procedure, which included treatment with water to deplete the red blood cells followed by centrifugations. Samples were then incubated with GFP-AMI_SH3 and analysed by flow cytometry. As a negative control, blood was mixed with medium without bacteria and submitted to the same procedure as the spiked samples, to evaluate unspecific binding of the fusion protein. In addition, unlabelled S. aureus Sa12 and a blood sample without protein were used as controls for the autofluorescence of bacterial cells and blood components, respectively. www.nature.com/scientificreports www.nature.com/scientificreports/ The results represented in Fig. 5 revealed that the GFP-AMI_SH3 combined with flow cytometry was successfully able to detect S. aureus cells in blood cultures. The protein was capable of decorating S. aureus cells, allowing the detection of about 99% of that population (Fig. 5A). On the contrary, blood samples incubated   www.nature.com/scientificreports www.nature.com/scientificreports/ with GFP-AMI_SH3 showed a negligible signal (about 0.35%) (Fig. 5B). In addition, the blood samples without GFP-AMI_SH3 (Fig. 5C) and the unlabelled S. aureus cells (Fig. 5D) did not emit a detectable fluorescence.
A specificity test was performed with K. pneumoniae 35 as negative control bacteria, following the same procedure used for S. aureus cells. The biparametric dot plot is displayed in Fig. 5E, demonstrating that only a minor number of events were recorded as positives (0.03% of the population), which was similar to the unlabelled K. pneumoniae cells (Fig. 5F). Moreover, since under the epifluorescence microscopy, GFP-AMI_SH3 showed a slight binding to E. faecalis LMV-0-39 and E. faecium LMV-0-42, these bacteria were further spiked in blood and the fluorescence was measured by flow cytometry. The results demonstrated that only a low percentage of the bacterial population (<5%) was plotted as positive (see Supplementary Fig. S2 online), which can be interpreted as a negligible signal.
Overall, the fluorescence signal obtained from the S. aureus samples decorated with GFP-AMI_SH3 was clearly distinguishable from the background signal (Fig. 6). The mean fluorescence intensity of labelled S. aureus cells (Fig. 6A) was 10-fold higher than the mean value obtained from the blood samples with GFP-AMI_SH3 (Fig. 6B) (means of 2,136.8 ± 145.6 and 204.6 ± 99.8 a.u., respectively). The weak fluorescence signal measured on  www.nature.com/scientificreports www.nature.com/scientificreports/ the labelled blood samples can be attributed to the presence of free protein that has not been completely washed or to the incomplete lysis of erythrocytes which are known to have some autofluorescence 43 . The blood sample (Fig.  6C) and the unlabelled S. aureus cells (Fig. 6D) displayed minimal fluorescence intensity means (6.7 ± 4.4 and 42.8 ± 6.2 a.u., respectively). Also, the signal obtained for K. pneumoniae (Fig. 6E) was significantly (p < 0.001) lower than the one obtained for S. aureus (Fig. 6A) and similar to the mean value obtained from unlabelled K. pneumoniae cells (Fig. 6F). Pairwise multiple comparisons were performed between all tested samples, and the results revealed that the mean fluorescence intensity of the signal emitted from S. aureus Sa12 cells labelled with GFP-AMI_SH3 was significantly different (p < 0.001) comparing with the mean value of fluorescence signal from the control samples namely blood, blood with GFP-AMI_SH3, unlabelled S. aureus and K. pneumoniae cells, and K. pneumoniae with GFP-AMI_SH3. Moreover, no statistically significant differences (n.s) were obtained when comparing these control samples (Fig. 6).

Discussion
BSIs and sepsis are the leading causes of human mortality among hospitalized patients and thus have a great impact on healthcare systems. Staphylococci are frequently isolated bacteria from BSIs and therefore their rapid identification from blood would allow the early diagnosis of a possible septicaemia 7,10 .
In this study, a new endolysin-based flow cytometry assay was developed for the early detection of Staphylococcus in blood. Firstly, different domains at C-terminus of S. aureus E-LM12 endolysin were cloned fused to GFP, expressed and functionally analysed. The results revealed that the GFP-SH3 protein was able to bind to S. aureus cells, which is in accordance with previous studies that reported that SH3 is the binding domain of endolysins in general 29,39,51 and of Staphylococcus phage endolysins in particular 39,52 . Interestingly, GFP-AMI_SH3 protein, which contains the amidase and SH3 domains, presented a higher binding efficiency to S. aureus cells than the GFP-SH3, suggesting that the amidase domain contributes to a more efficient binding of the SH3. This might be explained by the PGRP domain that was predicted in the same region of the Amidase_2, and that comprises conserved pattern recognition molecules that bind to bacterial peptidoglycan 53 . However, when the Amidase_2 domain (GFP-AMI) was used alone with S. aureus cells did not present any binding capacity. A previous study also showed that the amidase domain improved the binding affinity of a staphylococcal CBD to target bacterial cells, which resulted in an enhanced lytic activity of the endolysin 54 . The specificity and sensitivity assays performed with the GFP-AMI_SH3 revealed that this protein was able to bind to all Staphylococcus spp. tested. This constitutes an advantage of this biorecognition molecule once these pathogens have been frequently isolated from ICU infections 55 , being the CoNS and S. aureus responsible for 20.5% and 8.7% of ICU-acquired BSIs in Europe, respectively 55 . Other authors also reported a binding spectrum at the genus level for other CBDs from Staphylococcus endolysins 39,52,56 . Some studies suggested that CBDs recognize conserved binding ligands such as the glycine-rich inter-peptide bridge common to most staphylococcal strains [57][58][59] , as it has been reported for the SH3b-like cell wall targeting domain of lysostaphin 60 , which can explain their broad host range. Curiously, the GFP-AMI_SH3 protein demonstrated some binding affinity to two Enterococcus strains and one S. pneumoniae strain. Although the binding affinity of GFP-AMI_SH3 was not high and only few Enterococcus and Streptococcus cells were recognized, it can be hypothesized that the GFP-AMI_ SH3 binds to ligands of peptidoglycan in the bacterial cell wall that are conserved among these bacteria genera. In fact, although these bacterial species are members of two distinct phylogenetic orders namely Bacillales for Staphylococcus and Lactobacillales for Streptococcus and Enterococcus, they present peptidoglycan structures relatively similar apart from their cross-bridges that vary widely in composition and length 59 . Other authors also reported a broader binding affinity of endolysins and CBDs, showing recognition to other bacterial species 61,62 , such as the enterococcal PlyV12 endolysin which kills enterococcal, streptococcal and staphylococcal strains 62 and the Staphylococcus SH3b domain of endolysin LysF1 that binds to Staphylococcus and Streptococcus species 58 . We can also hypothesise that the amidase domain plays a role in the binding of SH3b to other bacterial species. Further investigations into the nature of the SH3 receptor molecules and the amidase domain are needed, which may be important for the understanding of the binding mechanisms of other endolysin CBDs as well.
The endolysin E-LM12 demonstrated capability to lyse all the Staphylococcus strains tested, including CoNS. These results are in agreement with other studies that report that Staphylococcus phage endolysins present a broad lytic activity covering all the Staphylococcus genus and commonly their spectrum is similar to the one from their CBD 56,63 . Also, the amino acid sequence of E-LM12 is very conserved, being closely related to several staphylococcal endolysins, namely LysK, which was described to be effective against S. aureus and CoNS 64 .
Regarding the LM12 phage, it exhibited a lytic effect against most of the Staphylococcus spp., although in some strains "lysis from without" was observed, which indicates the absence of phage infection 50 . On the other hand, the GFP-AMI_SH3 protein bound to all the staphylococcal strains tested, demonstrating its higher potential as a recognition probe for Staphylococcus spp. when compared with the phage particle.
Regarding the flow cytometry assays, the results indicate that this methodology was able to specifically detect all the Staphylococcus species tested and no false positives or false negatives were obtained. In this study, the developed assay revealed to be capable of detecting 10 5 CFU mL −1 of S. aureus cells without enrichment, which was expected since the minimal concentration of bacteria detectable by flow cytometry is reported to be in the www.nature.com/scientificreports www.nature.com/scientificreports/ range of 10 3 to 10 4 CFU mL −1 , using viability stains 65 . The detection around these limits affects the precision of the measurements due to the relatively poor analytical sensitivity 65 .
The flow cytometry assay had to be adapted and optimized for the detection of Staphylococcus in blood samples since the number of microorganisms present in circulation during a BSI ranges from 1 to 100 CFU mL −1 26,27 . Therefore, a sample preparation method was designed to include a pre-enrichment step, allowing the effective detection of 1-5 CFU of Staphylococcus in 10 mL of blood. The enrichment procedure also assures the detection of viable cells 66 , preventing the occurrence of false positives. Moreover, the developed method avoided the occurrence of positive signals derived from the autofluorescence of blood components 26,43,67 and revealed a good analytical reproducibility with results that were easily and objectively interpreted.
The conventional culture methods for the detection of Staphylococcus in blood are time-consuming and laborious. In fact, the mean time to blood culture positivity of S. aureus is reported to be between 12 to 16 hours [68][69][70] and after this, the phenotypic tests must be implemented to identify the specific pathogen which takes about 12 to 36 hours 17 . The assay described herein took less than 2 hours to perform after a pre-enrichment step, and therefore provides a rapid, highly sensitive and reliable methodology for the detection of pathogens in blood, outstanding the standard method of BSI diagnosis. Moreover, in this case, the extensive time required for bacterial enrichment was due to the initial bacterial load of 1 CFU per 10 mL of blood. Considering that the time to blood culture positivity is inversely proportional to the initially inoculated concentration 71 , the turnaround time of our assay can be significantly reduced if a higher concentration of Staphylococcus is present in patient blood samples.
In summary, in this study, we have identified the C-terminal fragment of a Staphylococcus endolysin that presents a high binding affinity and specificity, which is composed not only of the obvious SH3b domain but also of the Amidase_2 domain. Moreover, we developed a novel and highly reliable endolysin-based flow cytometry assay that allows the rapid and specific detection and identification of Staphylococcus spp. from blood cultures.
The designed flow cytometry assay presents several advantages over other bacterial detection methods described in the literature, such as the high specificity rendered by the use of specific phage proteins, simple and inexpensive sample preparation, and the possibility to be tailored to detect other pathogens by using target-specific CBDs. Furthermore, the method can be adapted to a multiplex assay, if specific CBDs are fused with different coloured fluorescent proteins, allowing the simultaneous detection of pathogens prevalent in BSIs. The developed method can also be easily adapted to different clinical samples and types of bacterial infections, by adjusting the sample preparation methodology.
This work is a proof-of-concept that this novel phage endolysin-based flow cytometry assay can be employed in blood samples for the specific detection of bacterial cells. We envisage its validation in blood samples from septic patients in order to be implemented in a clinical environment.

Material and Methods
Bacterial strains and growth conditions. The bacterial collection strains were obtained from the Spanish Type Culture Collection (CECT) and clinical isolates were provided by the Hospital of Braga (Portugal). To complete the binding spectrum, additional Gram-positive and Gram-negative species from private collections were used. This accounts for a total of 42 strains used (Table 1), including 12 S. aureus, 10 non-S. aureus staphylococcal strains, and 9 representative Gram-positive and 11 Gram-negative bacteria.
In Silico analysis of staphylococcal endolysin. The E-LM12 (NCBI Reference Sequence: AUV56903.1) was previously identified as an endolysin belonging to the S. aureus vB_SauM-LM12 phage (NCBI Reference Sequence: MG721208.1) 46 . The sequence was analysed through BLAST 79 to find possible homologous sequences. The search for functional domains was carried using Pfam 80 , HHpred 81 and InterProScan 82 . The molecular weight and isoelectric point of the proteins were calculated using the Compute pI/Mw program ExPASy 83 .
Cloning of E-LM12 endolysin domains. Primers containing specific restriction cloning sites (Table 2) were designed to amplify different fragments of the E-LM12 C-terminus (Fig. 1) and insert them into the pET_ GFP plasmid 32 (plasmid pET28a(+) from Novagen with the Aequorea coerulescens GFP gene inserted between the NdeI and BamHI restriction sites). Three different fragments (Supplementary Material S3) were amplified by PCR: a fragment containing only the Amidase domain (referred as GFP-AMI); another containing only the SH3 domain (GFP-SH3); and one containing both domains (GFP-AMI_SH3). Primer melting temperatures were calculated using OligoCalc 84 . The fragments were amplified with Phusion DNA Polymerase (Thermo Fisher Scientific) with vB_SauM-LM12 phage as template DNA. The PCR protocol consisted of an initial denaturation step at 98 °C for 30 s; followed by 35 cycles of: denaturation at 98 °C for 10 s, annealing at 60 °C for 30 s and elongation at 72 °C for 1 min; and a final extension step at 72 °C for 10 min. The products were then digested with the restriction enzymes (Table 2), inserted into the pET_GFP (in order to fuse them with the GFP upstream) and ligated with the T4 ligase (New England Biolabs) to obtain the different constructions. The ligation was transformed into competent E. coli BL21 (DE3). Colonies were screened through colony PCR and positives were used for plasmid extraction and further confirmation through Sanger sequencing.
Functional analysis of the E-LM12 endolysin C-terminus and specificity assays by epifluorescence microscopy. The binding ability of the different constructions of the E-LM12 endolysin C-terminus fused to GFP (GFP-AMI, GFP-AMI_SH3 or GFP-SH3) was inferred by fluorescence microscopy observations of S. aureus Sa12 cells after incubation with the fused proteins. Briefly, bacterial cells were grown in 10 mL of liquid LB or TSB at 37 °C until mid-log phase (OD 620nm = 0.6) and then the culture was centrifuged for 5 min at 6,000 × g, followed by resuspension in 10 mL of PB. A volume of 500 µL of each bacterial suspension was centrifuged at 6,000 × g for 5 min. The pellet was resuspended in 20 µL of purified GFP fused protein (at a concentration of 5 µM) and incubated for 30 min at room temperature. The cells were washed three times with PB by centrifugation (6,000 × g, 5 min) to remove the unbound protein. The washed pellet was resuspended in 10 µL of PB and observed at the epifluorescence microscope equipped with U-RFL-T light source (Olympus BX51, Magnification 1,000×) in bright field and under the FITC filter (Excitation BP 470-490 nm; Emission: LP 516 nm). Control samples of bacterial cells without addition of the recombinant proteins were prepared simultaneously. GFP alone was used as a negative control.
For the assessment of the specificity and sensitivity of GFP-AMI_SH3, bacterial species listed in Table 1 were incubated with this protein, following the same procedure described above and observed at the epifluorescence microscope.
Phage lytic spectra. The host range specificity of the LM12 phage was screened against all strains listed in Table 1. Bacterial lawns were made on Tryptic Soy Agar (TSA) plates by mixing 100 µL of exponential-phase cell cultures of each strain with 3 mL of TSA soft overlays (TSB with 0.4% (w/v) of agar). The bacterial lawns were spotted with 10 µL drops of the phage solution (10 9 PFU mL −1 ). LM12 phage lysis was assessed by adding 10 µL drops of serial 10-fold dilutions of phage stock to the bacterial lawns. After 16-18 hours incubation at 37 °C, plates were inspected for lysis zones and results were scored as: "positive" for clear lysis areas; "negative" for the absence of lysis areas; and "lysis from without" (LFW) when clear areas were observed with high phage titers but no plaques appeared when lower phage concentrations were spotted 50 . E-LM12 lytic activity. The E-LM12 lytic activity was carried out by the spot lysis test against all Staphylococcus strains listed in Table 1. Bacterial lawns were made on TSA plates by mixing 100 µL of exponential-phase cell cultures (OD 620 nm = 0.4) of each strain with 3 mL of TSA soft overlays. The bacterial lawns were spotted with 10 µL drops of E-LM12 at different concentrations (2, 5, 8, 10 and 15 µM).
Flow cytometry assays. For the flow cytometry assays, the three different C-terminus domains of E-LM12 (GFP-AMI, GFP-AMI_SH3 or GFP-SH3) were initially tested against S. aureus Sa12 cells. Bacterial cells were grown in 10 mL of liquid LB or TSB at 37 °C until mid-log phase (OD 620nm = 0.6). Briefly, 1 mL of each bacterial suspension was centrifuged at 6,000 × g for 5 min and the pellet resuspended in 20 µL of purified GFP fused protein at a concentration of 5 µM and incubated for 30 min at room temperature. The cells were washed twice with PB by centrifugation (6,000 × g for 5 min) to remove the unbound protein and cells were resuspended in 200 µL. The samples were analysed using an EC 800 flow cytometer equipped with a diode blue laser (excitation at 488 nm) (Sony Biotechnology). A total of 45,000 events were acquired with a sample flow rate of 10 μL min −1 . The fluorescence was detected through a 525/50 nm band-pass filter on the channel FL1. Data analysis was performed using FCS Express 6 RUO software (De Novo Software).