Determination of Genotypes Using a Fully Automated Molecular Detection System

Tubes of EDTA anticoagulated blood samples (300 μL aliquots) were sent to the State University of New York (Buffalo) Center for Advanced Biomedical and Bioengineering Technology for DNA purification and bidirectional sequencing following standard operating protocols. Briefly, the QIAamp DNA Blood BioRobot kit (Qiagen, Valencia, California) was used for DNA purification followed by PicoGreen (Life Technologies, Carlsbad, California) double-stranded DNA quantitation. Target regions of interest were amplified (PCR primer sequences shown in Supplemental Table 2), and the amplicons were purified with the QIAquick PCR Purification Kit (Qiagen) and quantitated with PicoGreen using the Applied Biosystems 3730 DNA Analyzer (Life Technologies). Sequences were evaluated by 2 independent analysts (R.M. and S.G.) and 2 different software programs (SeqMan, DNASTAR, Madison, Wisconsin; and CLC Main workbench, CLC bio, Boston, Massachusetts). Twenty nucleotide regions flanking the individual SNPs (Table 1) were aligned for all sequences and evaluated for the allele. Any discrepancies observed between the 2 analysts were reviewed by a third analyst (G.S.) and, if necessary, resequenced.

A EncompassMDx workstation and disposable CARD cartridges (Rheonix, Inc, Ithaca, New York) (hereinafter, CARD) were used to process each sample. Briefly, the assays were initiated with the manual addition of 15 μL of EDTA anticoagulated, whole blood samples or the addition of controls into the sample or application port of the CARD (Supplemental Figure 1); each CARD is capable of simultaneously analyzing 4 separate samples. All other reagent additions were automatically performed by the software-driven robotics of the system, which is capable of automatically processing 16 samples per run. Within the sample preparation portion of the CARD (Supplemental Figure 1, zone 1), the cells were lysed, and nucleic acids were extracted and purified by a modified Boom et al¹⁹  method optimized for compatibility with the plastics and pump-driven microfluidics of the CARD. The DNA eluted from the silica filter column was then metered into 2 separate fluidic channels and allowed to flow into separate PCR reaction chambers (Supplemental Figure 1, zone 2), where PCR amplification was performed using primer pairs biotinylated at their 5′ termini. The PCR primer pairs and buffer conditions, optimized to amplify the CYP2C9*2 and CYP2C9*3 loci, were used in a single multiplex assay in one PCR reaction chamber, whereas the second chamber contained the primer pairs and buffer conditions, optimized to amplify the VKORC1 loci. Amplification took place within the onboard thermocycler, with both PCR chambers being subjected to the same thermocycling conditions. After completion of the PCR, hybridization buffer was pumped into the PCR reaction chambers, diluting the biotinylated, amplified products, which was followed by heat denaturation of the diluted mixes. Finally, both mixes were pumped to the DNA array portion of the CARD (Supplemental Figure 1, zone 3). Following hybridization of the single-stranded, biotinylated DNA, strands captured by immobilized DNA probes were detected via incubation with horseradish peroxidase (HRP)–conjugated streptavidin followed by the 3,3′,5,5′-tetramethylbenzidine (TMB) substrate.

The unique pumps and valves¹⁷  resulting from the manufacturing process²⁰  drive the fluidic movements to accomplish all steps, from lysis through color development, while keeping the fluids compartmentally sealed from the external environment and preventing any possibility of contamination. Spot color intensity from the precipitated product of the HRP-TMB reaction is captured by the optics system of the EncompassMDx workstation. The imaging software performed all required spot size and location verification, background subtraction, and intensity capture to generate relative intensities of the wild-type and mutant capture probes for each of the 3 alleles. The final readout of results is presented as the combined *2 and *3 genotype of the CYP2C9 gene, as well as the specific base calling for the VKORC1 allele.

Two separate PCR master mixes were delivered by the robotics of the EncompassMDx workstation to each of the 2 PCR chambers used to analyze each separate sample. One PCR master mix, containing primer pairs designed to amplify the CYP2C9*2 (430C>T) and CYP2C9*3 (1075 A>C) loci, along with all required amplification reagents, was introduced into one of the PCR chambers, whereas a second PCR master mix, containing all required reagents to amplify the VKORC1 (1173 G>A) loci, was introduced into the second PCR reaction chamber (Table 2). In addition, to avoid potential contamination from previously amplified PCR products, all PCR reactions were initially subjected to uracil-N-glycosylase (UNG) treatment by activation at 37°C for 10 minutes, followed by inactivation of UNG at 95°C for 2 minutes. Following UNG inactivation, PCR reactions proceeded through a total of 40 cycles, with 10 cycles programmed at 95°C for 30 seconds, 45°C for 30 seconds, and 72°C for 30 seconds; followed by 20 cycles at 95°C for 20 seconds, 45°C for 20 seconds, and 72°C for 20 seconds; followed by 10 additional cycles at 95°C for 15 seconds, 45°C for 15 seconds, and 72°C for 15 seconds.

All reverse dot blot incubation steps, except for color development and the final wash, were performed in a solution containing 150 mM NaCl, 10 mM NaH₂PO₄, 1 mM EDTA, and 0.1% sodium dodecyl sulfate. The amplified products generated in each of the 2 separate PCR reaction chambers for each sample or control were diluted with hybridization buffer, heat-denatured by incubation at 98°C for 3 minutes, pooled, and delivered to the DNA arrays for hybridization at 56°C for 10 minutes. Spotted on the arrays were specific probes (Table 3) designed to hybridize to the wild-type or mutant alleles at the 3 different loci. Probes were designed with locked nucleic acids in various positions to increase the differences in melting temperature between exact match and single-base pair mismatched DNA, thus allowing for the distinction of the SNP alleles.²¹  Following 2 washes at the hybridization temperature to remove unbound single-stranded DNA, the temperature of the chamber was decreased to 35°C and HRP-conjugated streptavidin was introduced into the chamber and incubated for 5 minutes. After 2 additional washing steps to remove unbound HRP, TMB was introduced and allowed to react at 35°C for 7.5 minutes. The microarray chamber was then washed to remove any unreacted TMB, and the chamber was filled with 20% ethanol in water to capture images without bubbles.

Each run of 4 CARDs (ie, 16 samples) contained one positive control and one negative control. The positive control (Maine Molecular Quality Controls, Scarborough, Maine) consisted of synthetic DNA sequences encoding both wild-type and mutant alleles of the CYP2C9*2, CYP2C9*3, and VKORC1 SNPs suspended in a proprietary, noninfectious, bloodlike matrix. The controls were processed the same as the authentic blood samples. Because they consisted of both wild-type and mutant alleles of each of the 3 SNPs of interest, a run was deemed valid if all spots were detected on the DNA array. The negative control consisted of a buffer blank applied to the CARD exactly as done with the test samples and was considered to pass if no spots, other than the spotting controls, were detected.

The ability to distinguish the various genotypes was based on the relative intensity of signals at the various locations on the DNA arrays. Briefly, the system's imaging hardware and software were designed to locate each of the 16 “regions of interest” (Figure 1, A and B), corresponding to the mutant and wild-type spots as well as the “spotting controls,” and then, to average the signal intensities (based on pixels) within the spots after subtracting the background intensity of 4 loci immediately adjacent to each of the 16 spots (ie, background signals) as noted in Figure 1, C. After completing the analysis for each individual region of interest, corresponding to the various probes, the system's software was programmed to compare the average intensities (beyond background) of the wild-type versus mutant spots and to make genomic “calls” based on the threshold angles determined in this study (see below).

The signal intensities were measured on a continuous scale; therefore, to classify samples by genotype, thresholds were derived, based on collected data. For a given patient sample, a pair of signal intensities was obtained, one corresponding to the wild type and one for the mutant, and such data were displayed via a 2-dimensional scatterplot. We hypothesized that genotypes determined by bidirectional DNA sequencing (used as the gold standard) could be differentiated by the automated system according to the standard position of the angle of the sample-specific intensity point. The classification rule assigned those samples that yielded angles less than Θ₁ as homozygous wild type and assigned those greater than Θ₂ as homozygous mutant. Those samples that fall between Θ₁ and Θ₂ were classified as heterozygotes. The diagnostic utility in discriminating genotypes under that working assumption was examined through the generation of the receiver operating characteristic surface, which is a graphic display of the rates of correct genotype classification under all possible angle thresholds. The receiver operating characteristic surface is a generalization of the classic receiver operating characteristic curve in the 3-dimensional case.²²  Pairs of optimal angle thresholds, Θ₁ and Θ₂, for discriminating the 3 possible genotypes for each loci were identified based on the Youden index and the generalized Youden index, under assumed normality, as described by Nakas and Yiannoutsos²³ ; the weights required for the generalized Youden index were taken to be the observed prevalence of the genotypes. Once established, the thresholds that permit distinction of the various genotypes were programmed into the system's software for automatic genotype calls. All statistical analyses were performed using SAS software, version 9.3 (SAS Institute, Cary, North Carolina).

All blood specimens (n = 254) used in the present study were obtained from the central laboratory of the Catholic Health System (Buffalo, New York) under an institutional review board–approved protocol. All samples were residual specimens provided after all required clinical testing had already been completed. All samples were deidentified with no linkage to the various individuals evaluated.

The automated determination of genotypes derived from blood samples from individuals was accomplished by allowing the fully automated molecular detection system to perform all necessary assay steps unattended. Although the system had previously been shown capable of automatically performing all necessary sample preparations, multiplex PCRs, and detections,¹⁷  it was necessary to “train” the software to discern and report the various possible genotypes at the loci under investigation. Cleary discernible hybridization spots were observed on the DNA arrays after the entire assay was performed automatically (Figure 1, D). In addition to the spots from the amplicon hybridization to the immobilized DNA probes, 3 spotting controls were also clearly visible on the right side of the array, along with one spotting control on the bottom left corner. The visualization and location of those spots not only ensured that the HRP reaction performed correctly but also allowed the imaging software to confirm the proper orientation of the DNA arrays, thereby mitigating the possibility of wrong calls. A “manual” read of the DNA array from the sample shown clearly revealed the hybridization at the wild-type and mutant probes for all 3 loci, resulting in a genotype call of CYP2C9 *2/*3 and VKORC1 AG. Because the goal of this study was to evaluate the fully automated process, including the final read out of the results, genotypes established by bidirectional sequencing were used as the gold standard to program the automated system's software to differentiate among the various possibilities. Although representative, reverse dot blots demonstrated the clarity of results (Figure 1, D), the software was programmed to automatically distinguish the various genotypes. When the relative intensity of the hybridization spots at the wild-type and mutant alleles were compared and plotted against each other (Figure 2, A through C), excellent discrimination among the 3 genotypes at each of the 3 loci was apparent. Ultimately, it was confirmed that the 3 possible genotypes at each loci could be discriminated by the standard position angle thresholds established.

Using the established threshold angles, the genotype calls by the system were compared with the bidirectional DNA sequencing data. A broad representation of genotypes was detected among the samples by bidirectional sequencing (Table 4). At each individual locus, all 3 possible genotypes (ie, homozygous wild type, heterozygous, and homozygous mutant) were detected. In addition, on a combined basis, with the exception of homozygous mutants at both CYP2C9 loci or either homozygous CYP2C9 mutation in combination with homozygous mutant at the VKORC1 loci, all other possible genomic combinations were observed. The system's software reported the same genotypes for each sample as determined by bidirectional sequencing. Moreover, with bidirectional sequencing results serving as the gold standard, no wrong genotype calls were made by the fully automated system.

In addition to the above studies, we independently analyzed 70 samples on each of 3 different EncompassMDx workstations and compared the genotype calls made by the optimized software on each of the 3 separate instruments (Table 5). In all cases, each of the 3 instruments yielded the same genotypes reported by bidirectional DNA sequencing.

Finally, once assured that the software was programmed to correctly call the various genotypes, it was further programmed to print out final reports describing the genotypes of the samples under investigation (Supplemental Figure 2).

The Centers for Medicare and Medicaid, as well as private and commercial payers, have argued that sufficient clinical utility has not been demonstrated for many pharmacogenomics tests. Although not implicitly stated, pharmacoeconomic considerations also have an important role in reimbursement decisions. For example, despite FDA recommendations that prescribing physicians take into consideration the genotypes of patients at the CYP2C9 and VKCORC1 loci before initiating anticoagulation therapy, many payers do not routinely reimburse such testing. When compared with an approximately $5.00 reimbursement for prothrombin time and INR testing routinely used to inform warfarin dosing, payers often argue that the high costs associated with many current molecular diagnostic tests are not justified. Taking into consideration the significant costs to the health care system resulting from ADEs associated with improper warfarin dosing,²⁴  however, the total cost to the system should be considered in reimbursement decisions. According to the Piazza et al²⁴  study, approximately $40,000 (inclusive of hospitalization and pharmacy costs) can be attributed to each ADE. Warfarin is one of the most widely prescribed drugs in the United States, with more than 30 million prescriptions filled per year,²⁵  resulting in more than 43 000 ADEs per year.^7,26  Therefore, the total added cost to the health care system from warfarin ADEs is approximately $1.7 billion/y. Factoring in the cost of human suffering and loss of productivity, the routine use of pharmacogenomics testing should be considered. A recent multicenter study²⁷  in the United States, comparing the clinical utility of genotype-guided versus clinically guided warfarin dosing, indicated no statistically significant improvement in anticoagulation control during the first 4 weeks of therapy, and there were no significant differences in major bleeding or thromboembolisms observed. In contrast, a similar study²⁸  performed in Europe revealed improved performance with a pharmacogenomics algorithm when evaluated during the first 4 weeks of therapy, although at the study's 12-week endpoint, no differences were observed. Moreover, a third study²⁹  reported no clinically significant bleeding events in a pharmacogenetic-guided warfarin study group, whereas 3 such events were noted in the control group. Taking those recent studies into consideration, a group from the FDA suggested that, in addition to intermediate endpoints, such as percentage of time in the therapeutic range, uncommon, but clinical meaningful, outcomes, such as ADEs, should also be considered when evaluating the use of pharmacogenetic testing.³⁰ 

Pharmacogenomics testing before the initiation of clopidogrel bisulfate (Plavix, Bristol-Myers Squibb/Sanofi Pharmaceuticals Partnership, Bridgewater, New Jersey) therapy, which must be bioactivated by CYP2C19 to provide effective antiplatelet protection, has also been recommended by FDA-approved labeling of the drug. Individuals harboring heterozygous mutations at either the CYP2C19*2 or CYP2C19*3 loci display significantly reduced bioactivation, and thus, dosing amounts need to be adjusted.³¹  Moreover, individuals who are homozygous mutant at either the CYP2C19*2 or CYP2C19*3 loci have virtually no ability to convert the drug to its active form, thus prompting the FDA to require a box label indicating that pharmacogenomics data should be considered before initiating drug therapy.³²  Obviously, physicians unaware of the CYP2C19 status of their patients could unknowingly provide no therapeutic benefit to their patients and possibly should have alternative therapies considered. Despite considerable data demonstrating the effect of pharmacogenomics on effective treatment by clopidogrel,^33–37  the reasons often cited for not performing pharmacogenomics testing are the high costs of molecular diagnostic testing, delayed turnaround times, lack of reimbursement, and physicians being unfamiliar with the underlying molecular biology basis of drug metabolism.^38,39 

To facilitate the use of pharmacogenomic data, we have designed a fully automated molecular detection platform that can dramatically simplify and reduce the cost of the molecular testing. A rapid-ACCE (analytic validity, clinical validity, clinical utility, and ethical, legal, and social implication) review⁴⁰  indicates that the cost of performing CYP2C9 and VKORC1 genotyping is $300 to $800 per test, not inclusive of the capital equipment required to perform the assays, which can be $100,000 to $250,000. Because the disposable CARDs are produced by inexpensive, plastic-injection molding methods, significant cost savings can be achieved with an estimated cost to users of $45 per test, inclusive of the equipment costs and all necessary reagents, thus eliminating that obstacle from payers' consideration. Moreover, the system's simplicity will dramatically reduce the training required and decrease the time until results to less than 3 hours. The automated system can also be used to collect large-scale data sets to support clinical utility claims, critically required by payers to support reimbursement. With such reimbursement in place, the automated system should also prove useful in expanding the use of pharmacogenomics data for clinical application, helping to ensure that specific drugs are only administered to those individuals genetically predisposed to obtain the most likely benefit. Not only will cost savings result from providing drugs only to receptive individuals but also additional savings will result from a reduction in hospitalizations from ADEs.

We thank James Jarnot, MBA, MT, of the Catholic Health System (Buffalo, New York) for providing the blood specimens used in this study. In addition, we thank Norma Nowak, PhD (Department of Biochemistry, State University of New York, Buffalo) and Sean Glenn, PhD (Genomics Shared Resource, Roswell Park Cancer Institute, Buffalo) for expert advice and the performance of the Sanger Sequencing, which was supported by Roswell Park Cancer Institute and by grant P30 CA016056 from the National Cancer Institute (Bethesda, Maryland). Financial support from the State University of New York (Buffalo) Center for Advanced Biomedical and Bioengineering Technology partially subsidized the cost of performing the required DNA sequencing.