Introduction Identification of specific pathogens is essential for the selection of pathogen-specific treatments, minimizing the spread of infection, and monitoring for long-term complications. Currently, specific pathogen identification is achieved through available RT-PCR and antibody-based strategies. In the clinic, these tests are usually applied consecutively to evaluate for the presence of the most likely to least likely pathogen based on patient findings. An automated classification strategy that is less dependent on clinical knowledge but achieves rapid accurate identification of a single pathogen from among a group of possible pathogens is currently unavailable. 9 , 10 10 9 In this report, we describe a small-scale test of the use of a simple classification tree together with feedback-controlled FARA to identify four viruses. The feedback feature of online FARA is used to guide the selection of subsequent tests within the classification tree. Unnecessary tests are not performed, and each subsequent test becomes more specific for a single pathogen. These results establish FARA as a robust platform for classification of diverse pathogens. Materials and Methods Cells and Viruses l −1 −1 −1 1 1 1 , 3 , 6 Antibodies 2 , 12 , 13 1 Table 1. Antibody specificity. Antibody Protein specificity Reovirus strain T1L T3D T3SA+ 5C6 T1 σ1 + − − 9BG5 T3 σ1 − + + 8H6 T1 and T3 μ1 + + + 4F2 T3D σ3 − + − 2 Table 2. The five reaction chambers in FARA processing Chamber Description Solution Incubation time Chamber ID / Volume (mm / μl) 1 Block/Wash Filament a 15 min 2/235 2 Virus Incubation Unknown Virus 40 min 1/60  3 Wash PBS-T 1 min 2/235 4 Detecting Antibody Incubation Fluorescently Labeled Detecting Antibody 5 min 1/60  5 Wash PBS-T 1 min 2/235 a Filament Preparation 1 Figure 1. Schematic of a device for adsorption of capture antibodies to the filament. The filament is placed within the concave teeth of a PhastGel applicator to localize the spotted antibody to a small circumferential band around the filament.   Red fingernail polish (Poisonberry, Noxel Corporartion, Hunt Valley, MO) was used as a visible and fluorescent fiducial marker to identify the leading and trailing edges of capture-antibody regions during experiments and during laser scanning. The polish was applied by pipette between the teeth of the comb flanking the antibody region. A simple bar code system was used to identify each of the three capture-antibody regions. The first and second regions were preceded by a single fiducial marker, and the third region was preceded by two fiducial markers. Micro-reaction Chambers 2 9 12 −1 11 −1 −1 2 Filament Control Movement of the filament and, therefore, the antibody bands through the chambers, was achieved by using a rotary stage to wind or unwind the filament around a spindle. A small weight was attached to the opposite end of the filament to maintain a constant tension. Filament positioning to within several microns was achieved by using a rotary stage encoder from Yaskawa Instruments (Waukegan, IL) and a custom control algorithm written as a LabView Virtual Instrument (VI) (National Instruments, Austin, TX). Parameters including filament speed and residence times within chambers were controlled by using the LabView software interface. −1 −1 −1 Lasers and PMTs 2 6 5 −1 Figure 2. Schematic of lasers and photomultipliers (PMT) to detect online fluorescence of the filament. The filament is moved (arrow) through a rectangular sample chamber with a laser excitation source attached to either side. Photomultiplier tubes are attached to the top and bottom of the chamber.  Filters 2 Classification Algorithm 3 Figure 3. Decision-tree algorithm to classify virus strains. The filament control program enters the decision tree from the top at level 1 and, based on the type of virus found at each level, follows different branches of the decision tree. When reovirus is detected at level one, the filament is moved forward to the next antibody region for level 2 testing. If a serotype 3 reovirus strain is detected (T3D or T3SA+ ) at level 2, testing continues to level 3 to distinguish between these strains.   The filament-control program was designed to find each peak, calculate its location along the filament, and make a decision about additional tests. Spatially localized fluorescence from the filament was measured as a 0–14 V signal from the transimpedance amplifier. Voltage data from the scan were used as input data by a peak-detection function in LabView. The LabView peak detector fit a quadratic polynomial to sequential sets of points depending on a width parameter entered by the user. Data were then compared to a threshold parameter, also entered by the user, to identify peaks. A binary decision to stop or continue testing was based on the location and number of peaks found. Peak-detection threshold parameters were defined as 0.3 V (approximately three times background) with a width of 30 data points, which corresponded to a width slightly larger than the physical width of the comb tooth. 10 2 −1 −1 −1 −1 −1 3 If no peaks were found between the fiducial markers, a message was generated indicating that no virus was found. Similarly, if a peak was detected in the negative control region of the filament, a warning message was generated. Although theoretically possible, this situation did not arise during testing. Results 2 3 4 Figure 4. Fluorescence intensity in volts as a function of filament position for M13K07 phage (A) and reoviruses T1L (B), T3D (C), and T3SA+ (D). The immobilized capture antibody positions on the filament are labeled within each panel. Captured virus was detected by a fluorescently labeled second antibody (anti-M13K07 [red], 8H6 [red], or 4F2 [green]). Corresponding images of the scanned filaments are shown at the top of each graph. Each panel shows a representative experiment of three performed.   4 4 4 4 4 Discussion In this study, we demonstrated the feasibility of a FARA pathogen-classification approach using a well-characterized virus system. Three reovirus strains and one phage were successfully classified by performing sequential antibody-binding assays directed by the decision tree. The structure of the decision tree was based on published characteristics of each virus strain and the known specificity of each antibody. Because the capture antibodies were arrayed at known locations along the filament, and the fluorescence of the detection antibodies was measured as a function of filament location, an increase in filament fluorescence at a particular location indicated antibody interaction with the virus. For each virus strain tested, the expected fluorescence peaks were observed, and the correct classification was made. Moreover, successful transit of all decision tree branches was demonstrated. 10 4 14 11 4 , 7 , 8 12 12 5 12 1 7 10 9 9 Attachment of the capture antibody to a mobile filament in FARA makes accurate positioning of the filament and correct identification of the capture antibody regions essential for virus strain classification. These parameters are particularly important in multilevel testing of the type used in our study. A simple bar code system based on fluorescent fiducial markers was used to identify the leading edge of each of the capture antibody regions and position the filament in the reaction chambers. The pattern of the fluorescent marks also was used to identify the level of testing. Since our simple test system involved three levels, and testing always began in level 1, only searches for patterns containing one or two fluorescent marks corresponding to levels two and three were required. However, this simple approach allows for the incorporation of more complex bar codes using additional spots or patterns of spots. This study demonstrates the feasibility of an automated system for diagnosing specific virus strains. Although the scheme we report is a relatively simple implementation of this system, more complex designs are possible. For example, each antibody capture region could contain additional antibodies. In addition, there is essentially no limit to the overall length of the filament, which would allow the incorporation of many additional testing regions. The sensitivity of FARA could also be increased by the use of quantum-dot labeled detection antibodies as well as a more sensitive fluorescence detection system. The most important requirement is the availability of a wide range of antibodies to the specific pathogens of interest. Such a strategy may have applications for detecting specific pathogens from complex mixtures. This approach would have both clinical and environmental applications.