The structural schematic diagram of the fiber optic biosensor is shown in Figure 1A . A semiconductor laser (Shanghai Fiblaser Technology Co., Ltd., Shanghai, China) with wavelength of 532 nm and a modulation frequency of 1 kHz was selected as the excitation light source, because the excitation wavelength of QDs is ≤565 nm. The laser was linked to one input of an optical fiber coupler (2 × 2, couple ratio: 20:80) through its pigtail fiber. The combination tapered fiber probe was connected to the 20% end of the coupler using the fluorescence optical fiber connector. The input laser light was transmitted to the fiber probe to excite QDs that bound with antibody specifically captured on the fiber probe surface ( Figure 1B ). The fluorescence signal with 605 nm emission wavelength of QDs was also collected by the fiber probe, and sent back to the other input end of the fiber coupler. The fluorescence signal was filtered through a long-pass filter (FEL0600, Thorlabs, USA) with a cut-on wavelength of 600 nm in order to remove the excitation light (532 nm). Then, the fluorescence signal was injected into an avalanche photo-diode (APD, LCSA500-01, Laser Components, Germany) for photo-electronic conversion, and detection in data acquisition (DAQ) system. To enhance signal resolution, digital signals from DAQ system were further processed by a correlation detection system based on phase-locked filtering algorithm, which was performed with LabVIEW (Liu et al., 2014). Finally, the fluorescence signal was displayed on the computer screen. The fiber optic immunofluorescence biosensor prototype was shown in Figure 1C .

In this study, the immune sandwich principle of antibody–antigen specific capture of S. aureus was used to improve the sensitivity and specificity of S. aureus detection. We had selected two different kind of antibodies, among them, the rabbit polyclonal anti-S. aureus antibody (ab20920) immobilized on the probe surface was used to capture as many S. aureus as possible in order to improve the sensitivity of the biosensor. When the S. aureus strains were captured, the mouse monoclonal anti-S. aureus antibody (ab37644) specifically binds to the captured S. aureus again to form a double-antibody sandwich structure. Therefore, the choice on the two antibodies could ensure that the probe has both high sensitivity and high specificity in capturing and detecting S. aureus. As shown in Figure 1B , the activated fiber probe immobilized with rabbit polyclonal antibody to S. aureus (ab20920) defined as capture antibody was firstly used to capture S. aureus. Secondly, the fiber probe bound with bacteria was incubated in the mouse monoclonal antibody (ab37644), which was specific for S. aureus and defined as detecting antibody to bind to S. aureus. Thirdly, the biotin labelled goat-anti-mouse IgG antibody (GaMIgG) was introduced to bind to the mouse monoclonal antibody (ab37644). Finally, avidin-labelled QDs (605 nm emission wavelength) was employed for fluorescence labelling by the specific binding between biotin and avidin. Moreover, the signal cascade amplification principle of the biotin–avidin–QDs system was used to enhance the fluorescence signal and improve the sensitivity and stability of S. aureus detection.

The fiber optic immunofluorescence biosensor was developed by the Department of Precision Instruments, Tsinghua University. The fiber probe was made of a step-index multimode silicon optical fiber (NA = 0.22, 105 μm core/125 μm cladding, Beijing Scitlion Technology Co., Ltd., Beijing, China), and the composition of the silicon is mainly Silicon dioxide (SiO2). The rabbit polyclonal anti-S. aureus antibody (ab20920) and mouse monoclonal anti-S. aureus antibody (ab37644) was purchased from Abcam (USA). Biotin labeled goat-anti-mouse IgG antibody (GaMIgG) and avidin-labeled QDs (QDs-605, its emission wavelength is 605 nm and its excitation wavelength of the laser in this study is 532 nm) were purchased from Wuhan Jiayuan Quantum Dots Co., Ltd., (Wuhan, China). The QDs labeled to avidin applied in this study is a kind of silicon (Si) nano quantum dot. Sodium borohydride (NaBH₄) was purchased from Shanghai McLean Biochemical Technology Company, and Trypticase Soy Broth bacterial culture medium (TSB) was purchased from Beijing Aokexing Biotechnology Company. Bovine serum albumin (BSA, P1621), phosphate buffer (1×PBS, B1203), Phosphate Tween buffer (1×PBST, A1007), and other non-special reagents were purchased from Beijing Pulilai Gene Technology Corporation in China.

All experimental bacterial strains were derived from the clinical microbiological laboratory of our hospital (Chinese PLA General Hospital), including the S. aureus standard strain (strain number ATCC 29213), Pseudomonas aeruginosa (ATCC 27853), Acinetobacter baumannii (ATCC 19606), E. coli (ATCC 25922), Staphylococcus epidermidis (clinically isolated from a patient), Klebsiella pneumoniae (clinically isolated from a patient) and five clinical S. aureus strains isolated from sputum specimen of patients with chronic obstructive pulmonary disease complicated with pulmonary infection, the five clinical S. aureus strains were respectively named as SA-1, SA-2, SA-3, SA-4 and SA-5. Among the five clinical S. aureus strains, the two S. aureus strains (SA-4 and SA-5) were from patients with severe lung infection, and both the two S. aureus strains were sensitive to vancomycin according to the antimicrobial susceptibility results. The two patients were also treated with vancomycin for anti-infective treatment, but the clinical effect was evaluated to be ineffective after excluding the influence of other clinical factors. It was speculated that the two S. aureus strains might have produced the biofilm in the infection in body, resulting in ineffective antimicrobial treatment.

Bacterial culture was performed in a biosafety cabinet in the clinical microbiological laboratory. The stored S. aureus standard strain (ATCC 29213 strain) was cultured on solid medium for 24 h by streaking to grow single colonies. A single colony was picked and inoculated in 20 ml of TSB liquid medium for 16 h to the logarithmic growth phase of the bacteria. The bacteria in the logarithmic growth phase were centrifuged at 5,000 revolutions per minute (rpm) for 5 min, and the supernatant was aspirated. The remaining bacteria were washed twice with PBS, and the bacterial concentration was determined using a haemocytometer. This method is to take a certain volume of cell suspension of the sample and place it in the counting chamber of a hemocytometer, and then observe and count by microscope. Since the volume of the counting chamber is constant, the number and the concentration of bacteria in the sample can be calculated according to the number of bacteria in the counting chamber of the hemocytometer. This method is simple and easy to implement, and the result can be obtained immediately. This method is not only suitable for cell and bacterial counting, but also for fungi and fungal spores. The concentration of the bacterial suspension was adjusted to 1 × 10⁸ CFU/ml with PBS and then serially diluted 10-fold six times for subsequent experiments, producing 10⁸, 10⁷, 10⁶, 10⁵, 10⁴, and 10³ CFU/ml concentrations.

Clinical isolates of the five S. aureus strains mentioned above were derived from sputum specimen of chronic obstructive pulmonary disease (AECOPD) patients by bacterial culture, and identified by automatic microorganism identification instrument (VITEK 2). Subsequently, the clinical isolated strains and other strains used for this study were also cultured in the same way as the S. aureus standard strain (ATCC 29213).

A taper-and-cylinder combination optical fiber was prepared as the fiber probe. It was fabricated with a simple static tube etching method with 40% hydrofluoric acid (HF) as the etchant (Liu et al., 2014; Zhang et al., 2014). The fiber probe (inset in Figure 1A ) consisted of a tapered section and a sensing region, whose length were approximately 0.03 and 1 cm, respectively. The diameter of tapered section was etched from 125 μm to about 45 μm within the length of 300 μm, and the diameter of sensing region was maintained about 45 μm for matching the V number of fiber (Anderson et al., 1994).

As shown in Supplemental Material Figure S1 , the fiber probe was immersed in a piranha solution to remove the residual HF. And then, it was activated by hydroxylation, silanization, glutaraldehyde cross-linking, and covalent conjugation of the rabbit polyclonal anti-S. aureus antibody (ab20920, 20 × 10⁻³ g/ml in Tris–HCl buffer, which is the optimal concentration proved in the followed exploratory research) to capture the S. aureus specifically. After antibody immobilization, the fiber probe was immersed in NaBH₄ solution to reduce the interference of background fluorescence caused by aldehyde group (Clancy and Cauller, 1998). And the remaining aldehyde groups and non-specific binding cites of the fiber probe were blocked by BSA solution. The activated fiber probes were stored at 4°C for future use.

As the same as the rabbit polyclonal anti-S. aureus antibody (ab20920) immobilized on the probe surface, the biotin-labeled goat anti-mouse IgG antibody used in this study was also a polyclonal antibody and it was labeled with biotin. Therefore, it was used as an analog to explore the optimal concentration of the rabbit polyclonal antibody (ab20920) used to immobilize onto the probe surface. The schematic diagram of experimental principle and the results were exhibited in Supplemental Material Figure S2 . The processes of the research were as followed.

The biotin-labeled goat anti-mouse IgG antibody was diluted with Tris–HCl buffer to different antibody concentrations (2, 5, 10, 20, 50, 100 × 10⁻³ g/ml) as shown in the Supplemental Material Figure S2 and the Tris–HCl buffer was as the control group without antibody. Firstly, the probe was incubated in the biotin-labeled goat anti-mouse IgG antibody in various concentrations mention in the Supplemental Material Figure S2b at 37°C overnight to allow the aldehyde groups on the probe to covalently bind amino groups of the antibody and made the antibody immobilized onto the probe surface. Secondly, after antibody immobilization, the fiber probe was immersed in NaBH4 solution to reduce the interference of background fluorescence caused by aldehyde group. And the remaining aldehyde groups and non-specific binding cites of the fiber probe were blocked by BSA solution. Thirdly, the probes were immersed in a biotin-labeled goat anti-mouse IgG antibody dilution solution (1:100 dilution) for 30 min at 37°C. Fourthly, the probes were washed with PBST and were immersed in the avidin-labeled quantum dots solution (QDs-605, 1:100 dilution) for 30 min at 37°C. Finally, the probes were washed with PBST for three times and then were connected to the biosensor and turn on the laser. After running 200 s, the fluorescence signal was collected and displayed on the computer. As proven in the exploratory experiment of the fluorescence signal-time response trace mentioned below, the effective response interval was defined from 50 s to 200 s and the average values of the fluorescence signal of the effective time interval calculated by the computer could be the optimal detection values. Each of the above experimental groups was repeated three times and their optimal detection values of the fluorescence signal were collected for statistical analysis of the differences between each group.

As shown in Supplemental Material Figure S2 , the results had confirmed that the 20 × 10⁻³ g/ml was the optimal concentration of the antibody used to immobilize onto the probe surface.

The SEM images of the S. aureus captured on the optical fiber probes were performed to evaluate the capture capacity of the prepared antibody-immobilized optical fiber probes for S. aureus. There were six groups in the experiment. Among them, the five experimental groups (five groups) were the prepared S. aureus standard strain (ATCC 29213) with a concentration gradient of 10⁸–10⁴ CFU/ml. In addition, the control group (one group) was the undiluted S. aureus suspension (1 × 10⁸ CFU/ml) and the non-antibody-immobilized optical fiber probe, which was used to detect whether the non-antibody-immobilized optical fiber probe could capture non-S. aureus strains. Each probe was immersed in the 1 ml S. aureus suspension of the various concentrations for 30 min at 37°C, and then the probe was removed and washed three times with 1 ml of PBST to remove non-specific adsorption. All the probes were imaged by the scanning electron microscope (SEM, Hitachi S-5500). In addition, during the process of the SEM imaging of the probe surface, we also analyzed the energy spectrums of the SEM image of S. aureus captured on the surface of the fiber probe. Energy spectrums were recorded by the Energy dispersive X-ray spectrometry (EDX) combined with SEM (Hitachi S-5500). The principle of the EDX is as follow.

The research purpose of the SEM and EDX images is to evaluate whether there were captured bacteria and to exclude some non-bacterial false light spots in the SEM images using EDX. In all SEM images of various groups, there may be areas where bacteria are captured and blank areas where bacteria are not captured. The result exhibited in Figure 2 was just used to prove that the EDX technology during SEM image analysis can ensure that SEM image analysis can effectively identify bacteria captured on the probe.

Using the S. aureus standard strains, we exploratory studied the fluorescence signal responses corresponding to bacterial concentrations and the fluorescence signal-time response trace of the biosensor in the detection process. The exploratory experiment only included two samples. One sample was the S. aureus standard strains at the 1 × 10⁷ CFU/ml concentration, the other sample was PBS control without any bacteria (the concentration of the S. aureus was 0). The detection process of the samples was as follows.

Firstly, two prepared antibody-immobilized optical fiber probes were immersed in the two samples solutions above for 30 min at 37°C for S. aureus capture. Secondly, both the probes were washed three times with PBST to remove non-specific adsorption, and then the probes were immersed in the mouse monoclonal anti-S. aureus antibody dilution solution (ab37644, 100 × 10⁻³ g/ml) for 30 min at 37°C for formatting the double antibody sandwich structure. Thirdly, after washed with PBST, the probes were immersed in a biotin-labeled goat anti-mouse IgG antibody dilution solution (1:100 dilution) for 30 min at 37°C. Fourthly, after washed with PBST, the probes were immersed in an avidin-labeled quantum dots solution (QDs-605, 1:100 dilution) for 30 min at 37°C. Finally, the probes were washed with PBST and were connected to the biosensor. Then, turn on the laser with the 532 nm wavelength excitation light. The emitted fluorescence signal was collected at 605 nm wavelength for 200 seconds (200 s) or longer and the fluorescence signal-time response trace was displayed on the computer.

According to the signal-time response trace, the effective response interval could be defined from 50 to 200 s and the average values of the fluorescence signal of the effective time interval calculated by the computer could be the optimal detection values applied in this study.

The sensitivity experiment was performed to evaluate the detection limit of S. aureus provided by the optical fiber probe-based fluorescence biosensor. There were seven groups in the experiment: the control group was PBS, and the other six experimental groups were the prepared S. aureus suspensions with the concentration gradients of 10⁸–10³ CFU/ml. The detection process of all the samples by the biosensor was as the same as the detection process in exploratory experiment of the fluorescence signal-time response trace mentioned above. And each of the above experimental groups was repeated three times and their optimal detection values of the fluorescence signal were collected for statistical analysis of the differences between each group.

To evaluate the specificity of the biosensor, five most common clinical pathogenic non-S. aureus strains were employed in the experiment. The bacterial strains were S. epidermidis, K. pneumoniae, P. aeruginosa, A. baumannii and E. coli. All the strains were cultured to the logarithmic growth phase, and the bacterial suspension concentrations were all adjusted to 1 × 10⁷ CFU/ml with PBS. The experiment consisted of eight groups, including the five experimental groups (the five non-S. aureus strains mentioned above), two positive control groups (1 × 10⁴ CFU/ml and 1 × 10⁷ CFU/ml S. aureus suspensions) and 1 blank control group (PBS). The detection process of all the samples by the biosensor was as the same as the detection process in exploratory experiment of the fluorescence signal-time response trace mentioned above. And each of the above experimental groups was repeated three times and their optimal detection values of the fluorescence signal were collected for statistical analysis of the differences between each group.

To verify the sensitivity (detection limit) of the biosensor for clinically isolated S. aureus strains, five S. aureus strains (named SA-1 to SA-5) isolated from different patients in our hospital were detected by the biosensor. Each of the 5 S. aureus strain suspensions above was adjusted to two concentrations (1 × 10³ CFU/ml and 1 × 10⁴ CFU/ml). A S. aureus-free PBS solution was used as a negative control group. Thus, a total of 11 groups were in the experiment. The detection process of all the samples by the biosensor was as the same as the detection process in exploratory experiment of the fluorescence signal-time response trace mentioned above. And each of the above experimental groups was repeated three times and their optimal detection values of the fluorescence signal were collected for statistical analysis of the differences between each group.

SPSS 19.0 software (IBM Corp., Armonk, NY, USA) was used for statistical analysis of the experimental data in this study. All data are tested for normal distribution and homogeneity of variance. All data conforming to the normal distribution were analyzed by two independent sample t-tests for statistical differences between groups. On the contrary, the rank sum test is used to analyze the statistical differences between groups for non-normally distributed data. For the linear regression relationship between the fluorescence intensity and the concentration gradient of S. aureus, the linear regression equation was established and the correlation coefficient (r) was calculated. All statistical analyses were performed using two-sided tests, and P <0.05 was considered to be statistically significant.

As exhibited in Figure 3 , the SEM images of the optical fiber probes capturing S. aureus show that the higher the concentration of the S. aureus suspension is, the more S. aureus can be captured by the probes. However, with the increase of the S. aureus suspension concentration (from 1 × 10⁴ to 1 × 10⁸ CFU/ml), the capture amounts of the S. aureus strain on the surface of the probes gradually increased in S. aureus suspensions at various concentrations.

In addition, the energy spectrums were recorded by the energy dispersive X-ray spectrometry (EDX) combined with SEM images. The comparation of the SEM and EDX image of between S. aureus and the blank area on the surface of the fiber probe were shown in Figure 2 . The precise EDX analysis is shown in Supplemental Material Table S1 . And the results suggested that the S. aureus strain contain Carbon (C) atoms and Sodium (Na) atoms, and the mass percentage (Weight %) and atomic percentage (Atomic %) of carbon (C) and sodium (Na) were higher than the blank area of the fiber probe. By this way, the presence or absence of S. aureus and other bacteria on the surface of the probe can be distinguished.

The fluorescence signal-time response trace during a complete test cycle (200 s) was shown in Figure 4 . The time response curve shows that the signal reached the maximum value instantly after the laser turning on and then decreased back gradually. This was caused by photo-bleaching (Liu et al., 2014; Zhang et al., 2014). After decreasing back, the fluorescence signal gradually tended to be stable. The burr phenomenon on the curve may be the electrical signal interference during the instrument operation. The results of the trace showed that the effective response interval could be defined from 50 to 200 s and the average values of the fluorescence signal of the effective time interval calculated by the computer could be the optimal detection values applied in this study. In addition, the results also showed that the signal response of the S. aureus at the 10⁷ CFU/ml concentration was significantly higher than that of 0 CFU/ml (PBS control without any bacteria) and the biosensor might be used to distinguish S. aureus at different concentrations.

The signal corresponding to fluorescence intensity and significant difference between the S. aureus concentration groups are shown in Figures 5A, B . Signal-time response trace during a complete test cycle (200 s) with the S. aureus concentration gradients from 0 to 1 × 10⁸ CFU/ml is shown in Figure 5A . The histogram in Figure 5B shows the average signal measured in triplicate at different concentrations of S. aureus. Compared with the signal of the blank control group, the fluorescence intensity of the 1 × 10³ CFU/ml concentration group was higher but not significantly (P >0.05). However, the fluorescence intensity of all concentrations ≥1 × 10⁴ CFU/ml was significantly higher than that of the blank control (P <0.05). Moreover, in the concentration range of 1 × 10⁴ CFU/ml to 1 × 10⁷ CFU/ml, the fluorescence intensity gradually increased with increasing S. aureus concentration, and there was a dose-effect relationship. Surprisingly, the fluorescence intensity of the 1 × 10⁸ CFU/ml group was lower than that of the 1 × 10⁷ CFU/ml experimental group.

To evaluate the linear correlation between fluorescence intensity and S. aureus concentration, a linear regression curve of fluorescence intensity and S. aureus concentrations was established ( Figure 5C ). The results showed that the logarithm of the S. aureus concentration had a significant positive linear correlation relationship with the fluorescence intensity in the range of 10³–10⁷ CFU/ml (correlation coefficient R ² = 0.9731, P = 0.011). The linear regression equation is y = 0.0348x − 0.0156. Therefore, this biosensor could achieve quantitative detection of S. aureus. Correspondingly, the approximate concentration of S. aureus in the samples to be tested might be estimated by the fluorescence intensity.

The fluorescence intensity and the significant difference between S. aureus and five non-S. aureus strains are shown in Figure 6 . Signal-time response trace during a complete test cycle (200 s) with the different kinds of pathogenic bacteria are shown in Figure 6A . Histogram in Figure 6B suggests that the fluorescence intensities of all non-S. aureus groups were not significantly different from that of the blank control group (P >0.05), but all were significantly lower than those of both S. aureus groups (1 × 10⁴ CFU/ml and 1 × 10⁵ CFU/ml, P <0.05).

The results of the biosensor sensitivities for five clinically isolated S. aureus strains are shown in Figure 7 . The results showed that compared with that of the blank control group (PBS), the fluorescence intensity of the five S. aureus groups (1 × 10⁴ CFU/ml) was significantly higher (P <0.05). However, although the fluorescence intensity of the five S. aureus groups (1 × 10³ CFU/ml) was higher than that of the blank control group, there were no significant differences between three (3/5, 60%) of the S. aureus strains (SA-2, SA-4, SA-5) and the blank control (P >0.05).