Conceived and designed the experiments: MAP AJB CLB LM KN DEJ SAT BL EA AH JD SFD DH-FT KMR. Performed the experiments: MAP LM KN DEJ SAT KMR. Analyzed the data: MAP AJB CLB LM DEJ SAT TR BL EA AH JD DH-FT KMR. Contributed reagents/materials/analysis tools: JD CLB TR. Wrote the paper: MAP AJB CLB TR.
Current address: Department of Pathology, University of Utah School of Medicine, Salt Lake City, Utah, United States of America
Current address: MolecularMD, Portland, Oregon, United States of America
I have read the journal's policy and have the following conflicts: Several authors are current or former employees of Idaho Technology. PCT Applications WO2006/121997 and WO2008/140568 have been filed on aspects of the work described here. AJB, JD, AH and CLB collaborate with Idaho Technology, Inc., on several National Institutes of Health and Centers for Disease Control-funded projects (see funding). There are no other relevant declarations relating to employment, consultancy, patents, products in development or modified products. This does not alter our adherence to all the PLoS ONE policies on sharing data and materials.
The ideal clinical diagnostic system should deliver rapid, sensitive, specific and reproducible results while minimizing the requirements for specialized laboratory facilities and skilled technicians. We describe an integrated diagnostic platform, the “FilmArray”, which fully automates the detection and identification of multiple organisms from a single sample in about one hour. An unprocessed biologic/clinical sample is subjected to nucleic acid purification, reverse transcription, a high-order nested multiplex polymerase chain reaction and amplicon melt curve analysis. Biochemical reactions are enclosed in a disposable pouch, minimizing the PCR contamination risk. FilmArray has the potential to detect greater than 100 different nucleic acid targets at one time. These features make the system well-suited for molecular detection of infectious agents. Validation of the FilmArray technology was achieved through development of a panel of assays capable of identifying 21 common viral and bacterial respiratory pathogens. Initial testing of the system using both cultured organisms and clinical nasal aspirates obtained from children demonstrated an analytical and clinical sensitivity and specificity comparable to existing diagnostic platforms. We demonstrate that automated identification of pathogens from their corresponding target amplicon(s) can be accomplished by analysis of the DNA melting curve of the amplicon.
The ability to rapidly detect and distinguish multiple infectious organisms is critical for the accurate diagnosis of seasonal and sporadic outbreaks, emerging pathogens and agents of bioterrorism
Standard microbiological testing can require several days for initial identification of a pathogenic organism, and many organisms cannot be recovered using conventional techniques
Although it has many advantages, the introduction of PCR into the standard clinical microbiology and virology laboratory has been associated with practical challenges
PCR assays for infectious disease range from the relatively simple, in which a pathogen is identified by the detection of a single positive amplification product, to multiplex assays for groups of pathogens
Despite the limitations, multiplex PCR strategies have already demonstrated clinical utility, particularly for detection and identification of pathogens causing respiratory tract infection
Sensitivity of detection is another important consideration when developing diagnostic tests, as even very low levels of pathogen can cause disease. Nested PCR is an exquisitely sensitive methodology in which a target is amplified in a two-step process. In the first stage, a template is amplified using a pair of “outer” primers. This PCR product is diluted and subjected to a second stage amplification using primers located within the first PCR amplicon. The second stage product can be detected by real-time or end-product analysis. Nested PCR increases sensitivity over conventional PCR due to the ability to perform up to 50 or 60 total cycles of PCR. Specificity of nested PCR is similar to that of probe-based assays, as all 4 primers must match the template
An integrated system that can interrogate a clinical sample for a broad range of pathogens is highly desirable in both diagnostic laboratory and clinical settings
Each FilmArray pouch is comprised of an injection molded polypropylene reservoir (the “fitment”, 120 mm long, 10 mm wide, 25 mm high, “A” in
(
The second stage PCR array is manufactured from 0.5 mm thick black polycarbonate plastic. 102 wells of 1 µl volume each are drilled into the array (“I” of
The layers of film and adhesive attaching the array to the pouch are separated to show the flow of liquid into the cells of the array (figure is not to scale). From the top the layers are: 2nd pouch film, 1st pouch film, array adhesive layer (orange), pricked cover film, array (black, with wells), and array cover film. All of the actual layers are transparent except for the array itself. Second stage PCR primers are spotted into the cells during manufacture and air-dried (Methods). Arrows show the flow of PCR master mix (without primers) entering the array through a hole cut in the 1st pouch film.
All of the other biochemical reagents are freeze-dried into the 12 wells of the fitment. Moving from left to right in
Well 1: Process control material (
Wells 2, 3, 4, 5: Wash buffer
Well 6: Nucleic acid elution buffer.
Well 7: Reverse transcription/first stage PCR master mix:
Well 8: Dilution buffer
Wells 9, 10: Second stage PCR master mix: containing LCGreen® Plus+ (ITI),
Well 11: empty
Well 12: Overflow reservoir for the second stage PCR mix.
After these reagents are loaded into pouches a “plunger tree” (“B” in
The FilmArray instrument is 39.1 cm long×25.4 cm wide×16.3 cm high, weighs 8.2 kg (
The movement of liquid through the pouch is controlled by three pneumatic elements within the instrument. Pistons (located behind “B” in
Organism related sequence information (complete genomes, gene sequences and partial gene sequences) was obtained from NCBI (
Organism | Gene Target(s) | Strain |
LOD95 |
AV | Hexon | Type 1 | 300 |
BoV | NP-1 | Clinical Sample | 4000 |
|
Toxin | A639 | 4,000 |
|
ompA | TW183 | 3000 |
CoV 229E | Polymerase | VR-740 | 4 |
CoV HKU1 | Nucleoprotein | PCMC 6123 | 1.9×106 |
CoV OC43 | Nucleoprotein | VR-759 | 600 |
CoV NL63 | Nucleoprotein | NR-470 | 5 |
EV | 5′ UTR | Echovirus 6 | 30,000 |
hMPV | Nucleoprotein | hMPV-16/IA10-2003 Type A1 | 2 |
HRV | 5′UTR | 1A | 1 |
Flu A (H1N1) | Matrix |
A/Brisbane/59/07 | 200 |
Flu A (H1N1- 2009) | Matrix |
A/SwineNY/03/2009 | 100 |
Flu A (H3N2) | Matrix |
A/Wisconsin/67/2005 | 5 |
Flu B | Hemagglutinin | B/FL/04/06 | 60 |
|
Toxin | M129 – Type 1 | 30 |
PIV 1 | Hemagglutinin | Type 1 | 500 |
PIV 2 | Fusion | Type 2 | 10 |
PIV 3 | Fusion | Type 3 | 10 |
PIV 4 | Fusion | Type 4a | 5,000 |
RSV | Matrix | RSV Type A | 2 |
See
LoD concentrations are expressed in CFU/ml and TCID50/mL for bacteria and viruses respectively except for
The LoD for Enterovirus (30,000 TCID50/ml) is based on positive results for the Entero1 or Entero2 assays. A final result of Human Rhinovirus/Enterovirus based on the combination of 6 different assays (HRV1–4, Entero1 and Entero2) can be obtained at much lower concentrations (∼300 TCID50/mL).
The Flu A Matrix and NS1 gene assays are referred to as “pan1” and “pan2” respectively in the text.
AV, Adenovirus; B. per, Bordetella pertussis; BoV, Bocavirus;
1st and 2nd stage assays were initially tested separately to ensure that they produced the expected amplification product, and that RNA assays were dependent on the presence of reverse-transcriptase in the reaction mix. An additional criterion was that all 1st and 2nd stage primers must function well at the same annealing temperature. First and 2nd stage assays were then combined to form a singleplex, nested assay and tested for efficiency, sensitivity and specificity using quantification cycles (Cq,
The freeze-dried reagents in the fitment are resuspended with hydration solution using a 3 ml syringe fitted with a blunt metal cannula. The cannula is inserted into the hydration port (“X” in
Sample injection into the pouch is performed in a biosafety cabinet following the appropriate biohazard guidelines for working with potentially infectious samples. For the RP pouch, the FilmArray instrument may be operated on a laboratory bench or inside a biosafety hood.
Well 1 of the fitment contains the sample together with
Total nucleic acid in the sample is isolated by moving the sample lysate across the silica-magnetic beads particles present in well “E”. A retractable permanent magnet (located behind blister “E” in
Reverse transcription and first stage PCR occur in blisters “F” and “G”. PCR master mix (containing the reverse transcriptase) from well 7 of the fitment is pushed into blister “F”. Bladders push both blisters against a Peltier device behind the pouch. A mechanical hot start is achieved by holding the contents of the two blisters separate (using a hard seal between them) until they reach 54°C. Reverse transcription occurs during an initial 3 minute hold at 54°C. The first stage PCR consists of 26 cycles of 94°C for 4 seconds followed by 60°C for 19 seconds. During reverse transcription and PCR cycling the contents of the reaction are mixed by moving the liquid between blisters “F” and “G”. At the end of cycling the reaction is diluted approximately 225-fold into second stage PCR master mix by two successive dilution steps, first with TE buffer from well 8 and then with PCR master mix from wells 9 and 10.
The second stage PCR occurs in the wells of the array (“I” in
After the final PCR cycle the sample is held at 63°C for 5 sec followed by a linear ramp in temperature from 68°C to 95°C at a nominal rate of 0.5°C/second. Images are acquired 10 times per second.
The FilmArray instrument is capable of collecting fluorescence images and corresponding temperature data during the temperature ramp performed after the second stage PCR. The melt curve, defined as the average fluorescence intensity of each well as a function of temperature, is the basis for the automated organism detection algorithm described below. During the development of the system, the instrument was also programmed to acquire images once per PCR cycle in order to generate conventional real-time PCR amplification curves and corresponding Cq values. However the amplification data are not used in the automated organism calls for the commercial FilmArray system (see Results).
For the automated analysis, a hierarchy of calls is made: first for individual wells, then for individual assays (when specific primers are replicated in multiple wells of the array) and finally for each organism. This analysis is first performed on the control assays. If the controls return positive results, the analysis proceeds to the pathogen assays and the results are reported. If controls assays return a negative result, the run is declared ‘Invalid’ and no organism results are reported.
For each well, curve shape and peak location analyses of the melt curve are used to make a “Positive” (amplicon present) or “Negative” call. If two or more replicate wells are Negative for any one assay, then assay is called “Negative”. Next, if two of the melting temperatures (Tms) for positive replicates are within assay-specific limits (see Results) then the software assigns a positive call to the assay. For organisms with a single associated assay the final test result of ‘Detected’ or ‘Not Detected’ is based on the assay call. For influenza A and Rhinovirus/Enterovirus, the final test result is based on the integration of all associated assays.
Viruses and bacteria used in this study are indicated in
Residual clinical NPA specimens (stored frozen at −80°C) came from children younger than 18 years who had NPA collected for respiratory viral testing by direct fluorescent antibody (DFA) and culture at Primary Children's Medical Center (PCMC), Salt Lake City, UT between 2006 and 2008. Approximately half of the NPA specimens chosen for analysis were negative by DFA and viral culture. FilmArray testing was performed at both PCMC and ITI. PCR results were not used to inform clinical management or reported to microbiology technicians performing DFA and viral culture.
FilmArray data used for tuning the melt calling algorithm were acquired at sites performing beta testing of the instrument. The data used to validate the algorithm were acquired during clinical trials of the FilmArray system and RP pouch at the Medical University of South Carolina (Frederick S. Nolte, PhD), Detroit Medical Center (Hossein Salimnia, PhD), and Children's Medical Center of Dallas (Beverly Rogers, M.D.).
The institutional review boards of the University of Utah and PCMC approved this study and granted a waiver of informed consent because the patient samples were de-identified. All external clinical studies were performed with appropriate IRB approval. Data from these sites were de-identified before being sent to Idaho Technology.
The PCMC microbiology laboratory performs DFA for seven respiratory viruses: Influenza A (FluA), Influenza B (FluB), Respiratory Syncytial Virus (RSV), Parainfluenza viruses 1–3 (PIV 1–3) and Adenovirus (AV) using a panel of DFA assays (Simulfluor respiratory screen, Light Diagnostics, Temecula, CA) with reflex to viral culture. Human metapneumovirus (hMPV) is detected with a specific hMPV monoclonal antibody (Diagnostic Hybrids, Athens, OH).
Viral cultures are performed using a single cell line (R-Mix-Too; Diagnostic Hybrids) with an exit stain at 72 hours. The sensitivity of DFA testing, compared with viral culture, was 90% for FluA, 72% for FluB, 99% for RSV, 77% for PIVs, and 92% for hMPV, with specificity of 90% for all of the viruses in the PCMC laboratory
McNemar's test is used to compare the DFA results to the FilmArray results
FilmArray RP | DFA | Positive Percent Agreement | Negative Percent Agreement | Discordance P Value |
||
Pos |
Neg |
(95% CI) |
(95% CI) |
|||
AV | Pos | 22 | 32 | 84.6 (65.1–95.6) | 89.4 (85.4–92.6) | <0.001 |
Neg | 4 | 270 | ||||
hMPV | Pos | 4 | 10 | 66.7 (22.3–95.7) | 96.9 (94.4–98.5) | 0.021 |
Neg | 2 | 312 | ||||
Flu A | Pos | 14 | 2 | 100 (76.8–100) | 99.4 (97.7–99.9) | 0.50 |
Neg | 0 | 312 | ||||
Flu B | Pos | 1 | 8 | 100 (2.5–100) | 97.6 (95.2–98.9) | 0.008 |
Neg | 0 | 319 | ||||
PIV1 | Pos | 6 | 0 | 54.5 (23.4–83.3) | 100 (98.8–100) | 0.063 |
Neg | 5 | 317 | ||||
PIV2 | Pos | 10 | 8 | 90.9 (58.7–99.8) | 97.5 (95.1–98.9) | 0.039 |
Neg | 1 | 309 | ||||
PIV3 | Pos | 19 | 28 | 95.0 (75.1–99.9) | 90.9 (87.1–93.9) | <0.001 |
Neg | 1 | 280 | ||||
RSV | Pos | 37 | 9 | 94.9 (82.7–99.4) | 96.9 (94.2–98.6) | 0.035 |
Neg | 2 | 280 |
Positive or Negative test result comparing FilmArray RP (new test) to DFA (reference standard subject to error). (N = 328)
Clopper-Pearson 95% confidence Interval.
McNemar test, comparing discordant cells (FilmArray positive, DFA negative) vs (FilmArray negative, DFA positive).
If the sensitivity and specificity of the two methods are equal, the two off diagonals should be approximately equal and the estimated probability of being the same should be high. If one method is more sensitive or specific than the other, one of the off diagonal cell counts would be larger than the other and the estimated probability that the two methods have the same sensitivity and specificity would be low. The test does not provide the user with the information to determine that one method is more sensitive or the other is more specific, rather only gives them the power to say that they are different.
Initial development of the system was performed using DNA and RNA targets from
To maximize the sensitivity of the system, we determined the number of cycles in the first stage PCR that are needed to enter the plateau phase of the reaction
The completed first stage PCR mixture is diluted and then mixed into fresh PCR reagents. We determined empirically that two successive dilutions of ∼15 fold were necessary and sufficient to minimize primer carryover from the first stage PCR. Dilution of less than 100 fold generated nonspecific amplification products in the second stage PCR. Dilution of more than 300 fold caused a reduction in sensitivity.
The second stage PCRs are performed in individual wells of a high-density polycarbonate array (“I” in
The FilmArray pouch incorporates three controls to assess the performance of key steps in the system. As an RNA process control, the yeast
In parallel with the development of the FilmArray instrument and pouch we designed a panel of assays to detect viruses and bacteria known or suspected to cause upper respiratory tract infection in humans. Pathogens were chosen in consultation with pediatric infectious disease experts (AJB, CLB). Primers were designed to amplify conserved regions of the targets using standard software and alignment tools. The assays were initially optimized in conventional PCR instruments, using as template either organism from reference collections or pediatric NPA samples that tested positive for respiratory viruses by conventional (DFA) testing or the assays described here. Sequencing was used to confirm target identity. A successful set of assays for 23 targets from 21 pathogens was transferred to the pouch. The final FilmArray RP pouch contains 61 primers in the outer multiplex (of which four are for controls) and 31 inner second stage PCR assays (spotted in triplicate on the array with nine empty wells as negative controls). Pathogens and gene targets in the optimized RP pouch are listed in
Typical amplification and melt curves generated using a research version of the FilmArray instrument and RP pouch are shown in
Respiratory Pathogen pouches were injected with viral transport medium spiked with 200 TCID50 FluA H1-seasonal (panels
FilmArray can detect multiple targets in a single assay, and in particular, detect a low-copy target in the presence of a different, high-copy target.
To determine whether the FilmArray system would detect organisms in patient samples, we performed a study using pediatric NPA samples previously tested for respiratory infection at PCMC by DFA. Pre-clinical testing was performed at ITI and also during a 2-month placement of an instrument within the PCMC microbiology laboratory. Positive organism calls were made by expert users examining the amplification and melt curves. Three hundred and twenty eight samples were tested by both DFA and FilmArray. The results were compared for those viruses identified by both testing methods. When analyzed separately, similar results were obtained from both the research and clinical laboratories (data not shown) and thus combined data is presented.
The FilmArray, with 21 respiratory pathogen assays, identified significantly more pathogens than DFA in these pediatric samples (
Pediatric NPA samples (N = 328) were tested either by DFA at PCMC (yellow bars) or on the FilmArray (Blue bars). The percent of samples in which no virus (Negative) or one of the indicated viruses was detected is shown. The viruses are grouped into those in which both DFA and FilmArray assays are available or only the FilmArray assay is available.
A diagnostic system that automates the technically demanding steps of nucleic acid isolation and PCR amplification would benefit from automated analysis of the PCR results. FilmArray runs generate large amounts of data in the form of real time amplification curves and the associated melt curves. In similar systems the properties of the amplification curve are used to make a positive or negative call for that assay
Respiratory Pathogen pouches were injected with viral transport medium spiked with 1 TCID50 of the FluA- H1 seasonal virus used in
To compare the sensitivity of a detection algorithm based on melt curve analysis to one based on amplification curves, we had expert users annotate a total of 18,156 amplification and melting curves as positive or negative. Automated analysis of the data using both the amplification and melt profiles produced a sensitivity of 94.7%, a specificity of 99.95%, and an error rate of 4.15% compared to the expert user's annotations. By comparison, analysis of the melt curves alone produced a sensitivity of 97.49%, a specificity of 99.6%, and a total error rate of 2.92% compared to the expert calls. The higher error rate of the combined analysis is explained by false negative calls for weak amplification curves. Therefore we proceeded to develop an automated analysis of the FilmArray data using only the melt curves.
To maximize the specificity of melt curve analysis, we determined the range of possible Tms for amplicons from each different organism assay. The theoretical melting temperature of a DNA sequence on the FilmArray instrument was calculated using the model (modified from
Histograms of the theoretical or observed Tms of the hMPV assay are shown. Tm data for the FilmArray runs includes each of the three replicates of the second stage PCR.
To maximize sensitivity and specificity of the melting curve detection algorithm we optimized it using a large training dataset comprising 1566 RP pouch runs performed both at Idaho Technology (900 runs) and at external sites (666 runs) (Methods). The majority of the data generated at Idaho Technology was derived from contrived samples spiked with dilution series of the various target organisms (
Expert Interpretation | Melt Detector Call | Percent agreement (95% CI) | ||
Positive | Negative | |||
Training Set | Positive | 37,614 | 231 | 99.39 Pos (99.31–99.47) |
Negative | 141 | 108,529 | 99.87 Neg (99.85–99.89) | |
Validation Set | Positive | 8,153 | 30 | 99.63 Pos (99.48–99.75) |
Negative | 17 | 39,323 | 99.96 Neg (99.93–99.97) |
Using the same training and validation data set, the FilmArray automated analysis was compared to expert calls for the assay results (
Expert Interpretation | RP system Assay Call | Percent agreement (95% CI) | ||
Positive | Negative | |||
Training Set | Positive | 12,596 | 34 | 99.73 Pos (99.62–99.81) |
Negative | 4 | 35,912 | 99.99 Neg (99.97–100.00) | |
Validation Set | Positive | 2,713 | 8 | 99.71 Pos (99.42–99.87) |
Negative | 1 | 13,119 | 99.99 Neg (99.96–100.00) |
To determine at what level the PCR assays in the FilmArray system could correctly and consistently identify organisms, titered viral and bacterial respiratory pathogens were spiked into negative NPS sample matrix collected from healthy individuals or into a simulated NPS matrix consisting of viral transport medium (VTM) and a human epithelial cell line. Serial dilutions of the viruses and bacteria were spiked into NPS samples, both singly and in combinations of up to 5 organisms per sample. The spiked NPS samples were then tested on the FilmArray instrument. Quantification by TCID50 is a measure based on infectivity or cytotoxicity rather than number of organisms or copies of nucleic acid. LoD determined in TCID50/mL may not be an accurate indicator of the relative sensitivity of detection between different organisms.
An initial estimate of the system Limit of Detection (LoD95, or the concentration of organism that can be reliably detected in 95% or more of the samples tested) was based on the serial dilutions. Additional samples were then prepared and tested at the estimated LoD concentration and 10-fold lower to confirm that the correct organism was detected in at least 95% of the samples at LoD and in less than 95% of the samples containing 10-fold less organism. A positive organism detection was determined according to the automated analysis performed by the FilmArray software. For multi-assay organism calls such as FluA subtypes, all relevant assays were required to be positive at the LoD95 level. Column C in
The sensitivity of detection was comparable between samples containing a single organism and those containing up to five different organisms. Subsequent clinical evaluations determined that the sensitivity of each assay was appropriate for accurate detection of clinically relevant pathogen levels in NPS specimens. It is worth noting in this regard that the LoD concentration for Coronavirus HKU1 (1.9×106 RNA copies/mL) is below the published viral load detected in acute Coronavairus HKU1 infection (8.5–9.6×106 RNA copies/ml during the first week of the illness
In the last decade advances in diagnostic testing have led to changes in clinical laboratory evaluation that have translated into improved clinical care
The FilmArray is a realization of a “Lab-on-a-Chip” or μTAS system (micro total analysis system,
Although the FilmArray pouch manipulates relatively large volumes of liquid for a μTAS system, it shares several advantages with such systems. The steps of the system are automated which reduces operator work-load and error. The process is rapid: the time lag between one step of the chemistry and the next is measured in seconds. The physical separation of reagents between the fitment and the blisters also enables a hot start for both the first and second stage PCRs (see Methods). This eliminates the expense and inefficiency associated with using chemical or biochemical means of inhibiting Taq. A physical hot start also has the additional, unique, advantage that the reverse transcriptase is prevented from interacting with primers below the desired temperature of the reaction. This minimizes the formation of primer dimers or other nonspecific products in the deep multiplex of the first stage PCR.
The FilmArray lab-in-a-pouch is also an efficient solution to the “sample to assay” problem that many microfluidics systems must solve
In addition to its increased sensitivity and specificity, nested PCR simplifies the development of complex multiplex PCR panels. In order to detect viruses with great sequence diversity (e.g. AV and HRV) the first stage PCR contains moderately degenerate primers, or multiple primer sets. Unlike the common observation with single-stage multiplex PCR reactions
Whiley et al
nmPCR is also highly resistant to target competition (
The automated analysis of FilmArray data is robust to sequence variation in the target amplicon as well as potential instrument and pouch variation. The combination of melt detection in individual wells of the array together with the redundancy provided by the well-to-well comparison of replicate melt curves results in exceptionally sensitive and specific organism detection.
We have observed that melting curve analysis is more sensitive than amplification curves analysis for the detection of input material. The amount of data collected during the gradual temperature ramp of the amplicon melt greatly exceeds that collected during the relatively dynamic temperature cycling of the second stage PCR. The resulting melt curves have a higher signal to noise ratio than that of the amplification curves. For several reasons (e.g. loss of resolution at the high end of dynamic range, lack of a standard curve on a single sample instrument) the FilmArray Cq is not a meaningful measure of organism load in the sample. For this reason, amplification curve data are not reported in the commercial version of the instrument.
The initial testing of the FilmArray RP pouch with clinical samples demonstrates a successful real-world application of this technology. When compared to DFA using pediatric NPA samples, the platform showed high percent agreement. The most common reason for discordance was the detection of pathogens by FilmArray in DFA-negative samples. We believe this is due to the increased sensitivity of PCR when compared to DFA. In addition, the ability to test for a much larger panel of pathogens led to a decrease in the number of negative samples when compared to conventional testing, and increased the number of instances in which more than one pathogen was detected in a sample. Other multiplex PCR-based studies have reported similar findings
The FilmArray instrument and a subset of the assays in the RP pouch have recently been cleared by the FDA for IVD use and an initial comparison of the FilmArray instrument with the xTAG RVP (Luminex Corporation, Austin TX) and conventional detection methods have been reported
The advent of diagnostic platforms with the capability of medium level multiplexing
(TIF)
The authors thank FilmArray chemistry technicians, engineers, programmers (ITI), Trenda Barney for running samples on the FilmArray (PCMC) and Hsin-Yi Weng and Greg Stoddard (University of Utah School of Medicine) for statistical analysis, Carl Wittwer and our colleagues at ITI for editorial review of the manuscript and Mark Kessler and Kent Moyle for help with the figures. We thank the beta and clinical trials sites for the FilmArray data used for melt detection tuning and validation.