Genotyping accuracy of high-resolution DNA melting instruments

High-resolution DNA melting is a closed-tube method for genotyping and variant scanning that depends on the thermal stability of PCR-generated products. Instruments vary in thermal precision, sample format, melting rates, acquisition, and software. Instrument genotyping accuracy has not been assessed. Each genotype of the single nucleotide variant (SNV) (c.3405-29A>T) of CPS1 (carbamoyl-phosphate synthase 1, mitochondrial) was amplified by PCR in the presence of LCGreen Plus with 4 PCR product lengths. After blinding and genotype randomization, samples were melted in 10 instrument configurations under conditions recommended by the manufacturer. For each configuration and PCR product length, we analyzed 32-96 samples (depending on batch size) with both commercial and custom software. We assessed the accuracy of heterozygote detection and homozygote differentiation of a difficult, nearest- neighbor symmetric, class 4 variant with predicted <Tmof 0.00 °C.

Genotyping Accuracy of High-Resolution DNA Melting Instruments Mei Li,1,2 Luming Zhou,1 Robert A. Palais,3 and Carl T. Wittwer1* BACKGROUND: High-resolution DNA melting is a closed-tube method for genotyping and variant scan-ning that depends on the thermal stability of PCR-generated products. Instruments vary in thermal pre-cision, sample format, melting rates, acquisition, and software. Instrument genotyping accuracy has not been assessed. METHODS: Each genotype of the single nucleotide vari-ant (SNV) (c.3405-29AT) of CPS1 (carbamoyl-phosphate synthase 1, mitochondrial) was amplified by PCR in the presence of LCGreen Plus with 4 PCR product lengths. After blinding and genotype random-ization, samples were melted in 10 instrument configu-rations under conditions recommended by the manu-facturer. For each configuration and PCR product length, we analyzed 32-96 samples (depending on batch size) with both commercial and custom software. Weassessed the accuracy of heterozygote detection and homozygote differentiation of a difficult, nearest-neighbor symmetric, class 4 variant with predicted Tm of 0.00 °C. RESULTS: Overall, the heterozygote accuracy was 99.7% (n 2141), whereas homozygote accuracy was 70.3% (n 4441). Instruments with single sample detection as opposed to full-plate imaging better distinguished homozygotes (78.1% and 61.8%, respectively, 2 P 0.0005). Custom software improved accuracy over commercial software (P 0.002), although melting protocols recommended by manufacturers were better than a constant ramp rate of 0.1 °C with an oil overlay. PCR products of 51, 100, 272, and 547 bp had accura-cies of 72.3%, 83.1%, 59.8%, and 65.9%, respectively (P 0.0005). CONCLUSIONS: High-resolution melting detects hetero-zygotes with excellent accuracy, but homozygote ac-curacy is dependent on detection mode, analysis software, and PCR product size, as well as melting temperature differences between, and variation within, homozygotes. © 2014 American Association for Clinical Chemistry High-resolution DNA melting is a simple method of genotyping that uses dyes instead of probes. PCR prod-ucts are directly melted without additional processing. Advantages include low contamination risk, high speed, and high analytical sensitivity (1-4). High-resolution melting is widely used in research and clin-ical diagnostics for detecting DNA sequence variants, either known (genotyping) or unknown (scanning). Most single nucleotide variants (SNVs),4 as well as most small deletions or insertions, are easily geno-typed, often resulting in reduced costs and sequencing burden for many analyses (5-9 ). Heterozygotes intro-duce large shape changes in the melting curves and are easy to detect, with accuracies approaching 100% (10 ). In contrast to heterozygous variants, detecting homozygous variants typically depends on melting temperature (Tm) shifts and usually requires high-resolution melting. Class 1 and 2 SNVs exchange A:T and G:C base pairs with a Tm difference (Tm) of about 1 °C in small PCR products (2 ). These constitute about 84% of human SNVs and are easily detected on high-resolution melting instruments. However, about 16% of SNVs (class 3 and 4) simply switch bases while re-taining the same base pair, having predicted Tm val-ues of 0.0-0.4 °C between homozygotes. About one quarter of class 3 and 4 SNVs (4% overall) are "nearest neighbor symmetric" or "base-pair neutral" homozy-gotes with predicted Tm values of 0.0 °C (2, 11 ). These SNVs and some small insertions or deletions with Tm values close to 0.0 °C are the most difficult to differentiate by melting analysis (12 ). Even with pre-dicted Tm values near or at zero, some of these can be resolved by high-resolution melting, possibly because 1 Department of Pathology, University of Utah Medical School, Salt Lake City, UT; 2 current address: Laboratory Center, the Second Hospital of Dalian Medical University, Dalian, China; 3 Department of Mathematics, Utah Valley University, Orem, UT. * Address correspondence to this author at: Department of Pathology, University of Utah Medical School, 50 N Medical Drive, Salt Lake City, UT 84132. Fax [phone]; e-mail [e-mail]. Received December 6, 2013; accepted March 25, 2014. Previously published online at DOI: 10.1373/clinchem.2013.220160 4 Nonstandard abbreviations: SNV, single nucleotide variant; Tm, melting temperature; LS96, LightScanner 96; Rotor36/72, RotorGene Q (Rotor36, RotorGene Q with a 36-tube rotor; Rotor72, RotorGene Q with a 72-tube rotor); LC480, LightCycler 480; Piko, PikoReal 96. Clinical Chemistry 60:6 864-872 (2014) Molecular Diagnostics and Genetics 864 of stability effects that go beyond the nearest neighbor theory (13 ). The genotyping accuracy of high-resolution melt-ing instruments has not been directly assessed. In prior studies, the precision of early instruments was mea-sured by the SD of the Tm (TmSD) of multiple identical samples (14-16). At that time, there were large differ-ences among melting instruments, including sample format, melting rates, data density, and acquisition mode. Today, many differences remain, new instru-ments have been introduced, and it is not clear which of the many factors determine genotyping accuracy. Our goal was to establish the genotyping accuracy of instruments that claim to perform high-resolution DNAmelting by use of a range of PCR product lengths. Anticipating excellent results for heterozygotes on all instruments, we focused on differentiating homozy-gotes by using a difficult class 4 SNV from human genomic DNA with a predicted Tm of 0.0 °C. We also sought to find the instrument and assay characteristics that contribute to genotyping accuracy. Materials and Methods GENERATION OF PCR PRODUCTS Four PCR products of different lengths (51, 100, 272, and 547 bp) covering the same SNV in human genomic DNA [rs3213784, c.3405-29AT in CPS1 (carbamoyl-phosphate synthase 1, mitochondrial)] were amplified by nested PCR. To create the outer PCR products of 691 bp, human DNA of each genotype (A/A, A/T, and T/T) was amplified from CPS1 with primers GAAATCAGGTTCTGGGCTGA and TTC CTCTTTTTCCACCAACC. Amplification was per-formed in 10-L volumes with 50 ng genomic DNA, 200 mol/L each deoxynucleotide triphosphate, 0.4 U KlenTaq™ (AB Peptides), 88 ng TaqStart antibody (Clontech), 2 mmol/L MgCl2, 50 mmol/L Tris (pH 8.3), 500 g/mL BSA, 0.5 mol/L primers, and 1 LCGreen Plus dye (BioFire Diagnostics) with a 25-L mineral oil (Sigma-Aldrich) overlay. We used a CFX96™ (Bio-Rad) instrument with an initial denaturation at 95 °C for 2 min, followed by 25 cycles of 95 °C for 20 s, 65 °C for 20 s, and 72 °C for 30 s and cooling to 15 °C. The 3 genotypes (A/A, T/T, and A/T) were confirmed by sequencing. Each PCR product length and genotype was am-plified in 96-well plates in 50-L volumes with a 40-L overlay on either PTC200 (Bio-Rad) or CFX96 instru-ments. The reactions included 2 L of a 10 000-fold dilution of the 691-bp PCR product. Other PCR com-ponents were the same as those used to amplify the 691-bp template. Primers for 51-, 100-, 272-, and 547-bp inner products were: AGTCAAGTCTAG TATTAGCATAAACCT and AAGGAAGGGGAAA AAAAGCAG; ATAGGTTGTCTGGAACTGTTCTG and TCATAGCAGACCCACTGGAA; TTGGTTGAT TGTCCTGGTGA and CAGTCACTACAAAGAAAT TGGACA; and CAGAAAGGGCAAACTTTGGA and GGAGACTAGAGGGTAGAAGAGGAAA, respectively. PCR was performed with an initial denaturation at 95 °C for 2 min, followed by 25 cycles of 95 °C for 15 s, annealing at 65 °C (15 s for 51 and 100 bp; 20 s for 272 and 547 bp), and extension at 72 °C (10 s for 51 bp; 15 s for 100 bp; 20 s for 272 bp; 30 s for 547 bp), then 95 °C with a 20-s hold, and cooled to 15 °C. We pooled PCR products by genotype for each product length. The min-eral oil was discarded after centrifugation (1500g for 5 min). AllPCRproducts were scanned on a LightScanner 96 (BioFire Diagnostics) (LS96) to confirm the genotypes. We used nested PCR products as template instead of genomic DNA because the intent was to isolate and test the melting function of each instrument. INSTRUMENTS AND TEMPERATURE VERIFICATION We used 9 high-resolution melting instruments with excitation and emission wavelengths compatible with LCGreen Plus for comparison:HR-1™(BioFire Diagnos-tics), LightScanner 32 (BioFire Diagnostics) (LS32), LS96, Rotor-Gene Q (36 and 72 sample carousels, Qia-gen) (Rotor36/72), LightCycler 480 (Roche) (LC480), CFX96, StepOnePlus™ (Thermo Fisher Scientific), Eco™ (Illumina), and PikoReal™ 96 (Thermo Fisher Scientific) (Piko). The instrument characteristics are listed in Sup-plemental Table 1, which accompanies the online version of this article at http://www.clinchem.org/content/ vol60/issue6. The sample temperatures of all instruments except for the Rotor36/72 were monitored with a J-type mi-crothermocouple (5SRTC; Omega) and conditioned (USB-TC01; National Instruments) to calculate the ramp rate during melting. The sample temperatures of Ro-tor36/ 72 were measured with the temperature-sensitive sulforhodamine B dye, and the ramp rate was calculated by correlating fluorescence to temperature (17). MELTING ACQUISITION We randomized 32-96 samples of 3 genotypes into capillaries, tubes, or plates, ranging from 10 to 20 L, depending on the batch size and sample volumes rec-ommended by the manufacturers. Four runs, 1 for each product size, were melted in each instrument along with 3 known genotype standards. We used both the 36 and 72 rotors on the RotorGene, for 10 instru-ment configurations. The 4 PCR products were melted from 65 to 92 °C in all instruments under high-resolution melting conditions as recommended by the manufacturers. Melting was also performed at a common rate (0.1 °C/s) with a 10-L oil overlay (17 ) on the LC480, High-Resolution DNA Melting Accuracy Clinical Chemistry 60:6 (2014) 865 CFX96, Rotor36/72, and Piko. Because ramp rates can-not be programmed directly in these instruments, we adjusted instrument settings (acquisitions/°C, step, ramp, and/or hold) empirically to obtain 0.1 °C/s with a thermocouple or fluorescence. HIGH-RESOLUTION MELTING ANALYSIS The data obtained with the manufacturers' recom-mended melting conditions were analyzed on custom software written in LabView (1, 18, 19 ).We used anal-ysis with common software to focus on differences be-tween instrument platforms, not software. The custom software sequentially processes melting data in 3 steps as previously detailed (1 ). First, data is normalized be-tween 0% and 100% fluorescence, including exponen-tial background subtraction (19 ), so that the pre- and post- melting regions are horizontal, relate to DNA he-licity, and can be compared to predicted curves gener-ated by uMelt (20 ) (https://www.dna.utah.edu/umelt/ umelt.html). Second, and at user discretion, curves are overlaid at low fluorescence (high temperature) by shifting curves along the temperature axis until they are superimposed to focus on curve shape, not Tm. Finally, curves are displayed as difference plots that display all curves as the difference from wild-type curves. The custom software is available for noncommercial use by request from 1 of the authors (R.A. Palais). The custom software best detects heterozygotes by shape changes of normalized and overlaid melting curves displayed on difference plots. Homozygotes are detected on difference plots with or without overlay and the use of A/A and T/T standards. Overlay of melt-ing curves (also known as temperature shifting) is used when it helps to distinguish genotypes. In this study, overlay was particularly useful in analyzing homozy-gotes on some instruments with full-plate imaging. Re-gardless, all samples were analyzed both with and without curve overlay. All data were genotyped to de-termine: (a) the accuracy of heterozygous calls, (b) the accuracy of calling A/A vs T/T homozygotes, and (c) the analytical sensitivities of A/A and T/T calls. The Tm was taken as the peak temperature on negative deriva-tive plots. For the 2-domain, 547-bp product, the Tm of the larger (higher-temperature) transition was used. In some studies, we compared the instrument's own commercial software to the custom, common software. Commercial software results were obtained by manual inspection, autoclustering, or instrument genotyping as available from the different manufacturers. Linear regression, 2, and t-test analysis were per-formed with SPSS 21 (IBM). Results Many parameters can affect the accuracy of genotyping by high-resolution melting. For example, different manufacturers recommend widely different melting rates for fluorescent high-resolution DNA melting. The instruments we studied here vary over a 60-fold range from 0.005 to 0.3 °C/s, resulting in turnaround times of 1.5 to 95 min (Table 1). Instead of the direct and simple °C/s as the unit of melting rate, various manufacturers have adopted creative but confusing units, such as, "X acquisitions/°C," "X°C step with a Y s hold," "X°C ramp with a Y s hold," or "X% ramp." Empirical measurement is necessary to convert these units to °C/s. Additional differences between instru-ments include container format (capillaries, plates, and tubes), sample volumes, number of tubes or wells, ex-citation and emission wavelengths, detection mode (full-plate imaging vs single sample interrogation), de-tector type, continuous or step acquisition, and analy- Table 1. Recommended and measured melting rates on high-resolution melting Instruments. Instrument Recommended instrument setting Measured ramp rate (°C/s)a Turnaround time (min) Batch size Throughput (samples/h) HR-1 0.3 °C/s 0.3 1.5 1 40 LS32 0.3 °C/s 0.3 2.5 32 24 LS96 0.1 °C/s 0.1 5 96 1152 Eco 0.1 °C/s 0.08 7 48 411 LC480 25 acquisitions/°C 0.04 9 96 640 Piko step (0.04 °C)/hold (1 s) 0.01 40 96 144 Rotor36/72 ramp (0.1 °C)/hold (2 s) 0.01 40 36/72 54/108 CFX96 step (0.2 °C)/hold (10 s) 0.01 50 96 115 StepOnePlus 0.3% ramp 0.005 95 96 61 a Measured with a microthermocouple or by temperature-sensitive fluorescence [Sanford and Wittwer (17 )]. 866 Clinical Chemistry 60:6 (2014) sis software (see online Supplemental Table 1). Com-bining these instrument differences (that may all affect melting curve precision) with assay differences (Tm, PCR product size, SNV class, PCR master mix, and dyes) makes determining the most relevant parameters a daunting task. Therefore, we limited our study to comparing instruments and PCR product sizes sur-rounding a single SNV using a single master mix and dye. Certain parameters, such as melting rate and dye comparisons, will best be done on single instruments and are left for future studies. Representative melting curves of the 3 different genotypes (A/A, A/T, T/T) within 4 different PCR product lengths are shown in Fig. 1. PCR products 300 bp appear to melt in a single domain, whereas the larger 547-bp product melts in 2 domains. Heterozy-gotes are usually easy to distinguish from homozy-gotes, although the area between the curves becomes smaller as the product size increases. The overall analytical sensitivity of heterozygotes was 99.7% (2135/2141) including all 10 instrument configurations, 2 melting conditions (manufacturer recommended and 0.1 °C/s with oil), and 2 software analyses (original manufacturer and custom software). The analytical specificity was 100%. False negatives oc-curred only in the 547-bp product. With manufacturer-recommended melting conditions and software, 5 heterozygous variants were misjudged as homozy-gotes, including 1 sample classified as A/A on the LC480 (97%) and 4 as T/T on StepOnePlus (87%). With a ramp rate of 0.1 °C/s, oil, and custom analysis, 1 heterozygote was misjudged as A/A on the Rotor72 (96%). Under manufacturers' melting conditions and custom software, 100% heterozygote accuracy was ob-tained on all instruments and product sizes. In contrast to heterozygote detection, the 2 differ-ent homozygotes were very hard to distinguish from each other, as expected by their predicted Tm of 0.0 °C (2, 13 ). Genotyping accuracy varied with the in-strument and with PCR product size. Overall, only 3121 of 4441 homozygotes (70.3%) were correctly genotyped. The effect of curve overlay on genotyping ho-mozygotes on instruments with full-plate imaging is shown in Fig. 2. On some instruments, curve overlay clearly enabled homozygous genotyping, whereas without overlay, temperature gradients across the plate could mask genotype shape differences. However, this advantage of curve overlay was instrument dependent; on another 96-well instrument with full-plate imaging, overlay had little effect, i.e., overlay could reveal geno-type patterns only if the underlying precision of the curve shape was high enough. With all relevant data considered, instrument fac-tors that were highly significant included the melting Fig. 1. Normalized melting curves of the 3 SNV ge-notypes within 4 PCR products of different sizes [(A), 51 bp; (B), 100 bp; (C), 272 bp; and (D), 547 bp]. The black, light gray, and dark gray curves indicate the wild-type, homozygous, and heterozygous genotypes, re-spectively, without overlay. In all cases, the heterozygotes are easy to distinguish from the homozygotes, but the wild-type and homozygous genotypes are difficult to dif-ferentiate. Single melting domains are present in the first 3 PCR products, but 2 are present in the longest product. Each panel shows 32 samples of random genotype ob-tained on the HR-1 at 0.3 °C/s. High-Resolution DNA Melting Accuracy Clinical Chemistry 60:6 (2014) 867 protocol, detection mode, and software (Table 2). Melting protocols were more accurate when per-formed as described by the manufacturer, rather than at 0.1 °C with an oil overlay (P 0.0005). Single sam-ple detection was more accurate than full-plate imag-ing (P 0.0005). Those instruments that interrogate samples individually did better as a group than those with full-plate illumination and acquisition. Finally, the custom software genotyped more accurately than the various commercial software packages provided with the instruments (P0.002). Although no mono-tonic trend was apparent, the 51-, 100-, 272-, and 547-bp products had accuracies of 72.3%, 83.1%, 59.8%, and 65.9%, respectively (P 0.0005). Instrument and PCR product size effects are indi-vidually evaluated in Fig. 3, Fig. 4, online Supplemental Fig. 1, and online Supplemental Table 2. Online Sup-plemental Table 2 compares the homozygous genotyp-ing accuracy for specific instruments and PCR product sizes, analyzed by both commercial and custom soft-ware. Custom software improved accuracies on the LS32, LS96, and LC480 for multiple size products. In only 1 instance (100-bp product on the StepOnePlus) was the commercial software significantly better than the custom software. Further instrument and size com-parisons used only the custom software. The analytical sensitivities of correctly detecting A/A and T/T homozygotes are shown in Fig. 3, separat-ing each genotype along 1 dimension. All 10 different instrument configurations are plotted, and 1 panel is provided for each PCR product length. Any biases for incorrect A/A or T/T calls are reflected by the instru-ment position in each graph (upper left vs lower right). Overall homozygote accuracies by instrument and PCR product size are shown in Fig. 4 and online Sup-plemental Table 2. In Fig. 4, the instruments are or-dered according to decreasing recommended melting rates. Contrary to expectation, accuracy did not appear to improve with slower melting rates. Indeed, there was a trend toward better accuracy with faster rates. When the log of the melting rate was plotted against accuracy similar to Fig. 4, the correlation of rate to accuracy was slightly positive (linear regression r 0.3, P 0.057, data not shown). Of the homozygotes, the 100-bp product was the easiest to genotype, with 100% accuracy in 7 of the 10 instrument configurations. The 272-bp product was the most difficult to genotype, and no instrument achieved 100% accuracy. Some instruments were con-sistently good across all product sizes (HR-1 and CFX96). Others were excellent with short products, but not with long products (Rotor36 and 72). Variation in the genotyping accuracy of the different rotor sizes of the same instrument may result from (a) different rec-ommended sample volumes for each carousel, (b) total Fig. 2. Effect of curve overlay on homozygote differentiation. Wild-type and homozygous genotypes are not separated on some instruments with full-plate imaging unless curve overlay is applied. The inset between panels (A) and (B) shows the heterozygote wells in white with the homozygotes as different shades of gray. The gray level correlates to different rows of a 96-well plate, not the homozygous genotypes. (A), Without curve overlay, the spread observed on difference plots corre-lates to plate position, not genotype. That is, a preponderance of lighter lines lower in the graph indicates that the difference plot is dominated by temperature differences across plate columns. (B), After curve overlay, the difference plots split into 2 clear genotype clusters that both include all different shades of gray (plate positions), indicating that the overlay has re-moved the temperature gradient effect to allow definitive genotyping. (C), The plate wells are now shaded by genotype (black AA, gray TT), showing the clear separation between homozygous genotypes on the difference plot. (D), On some full-plate imaging instruments, the homozygous genotypes are not separated, even after overlay. Wild-type and homozy-gous variant data shown in (A) through (C) were obtained on a LS96 at 0.1 °C/s, and (D) was from a Piko at 0.01 °C/s. All panels show difference plots of the 100-bp PCR product. 868 Clinical Chemistry 60:6 (2014) volume load per carousel, and (c) different fluid (air) dynamics between 36- and 72-tube carousels. Assay characteristics also contributed to homozy-gote genotyping success. Genotyping accuracy was bet-ter in smaller products than in larger products (P 0.0005), although the relationship was not monotonic. The Tm between homozygotes predicts better geno-typing success in instruments with single sample detec-tion (P 0.0005), but not full-plate imaging (P 0.12). Furthermore, the TmSD is lower in instruments with single sample detection than those with full-plate imaging (P0.01). The smaller the TmSD of an instru-ment, the greater the expected accuracy of genotyping, as previously described (16 ). For any given SNV, the probability of distinguishing homozygotes increases as Tm increases. Attempts to combineTm and TmSD as a predictor of accuracy were not very successful, giving an R2 of only 0.256, P 0.001 (see online Supplemen-tal Fig. 1). Discussion High-resolution DNA melting was first introduced in 2003 (1, 18 ). It was rapidly adapted into most real-time Table 2. Characteristics that affect the ability to distinguish homozygotes. Characteristic and classification n Accuracy (%) Pa Instrument factors Melting protocolb <0.0005 Recommended by manufacturer 1004 72.7 Ramp rate 0.1°C/s, oil overlay 1038 62.9 Detection mode <0.0005 Single sample detection 2268 78.1 Full-plate imaging 2173 61.8 Acquisition mode 0.114 Continuous 1631 68.9 Step 2810 71.1 Softwarec 0.002 Custom 1709 74.4 Commercial 1618 69.5 Assay factors Amplicon size <0.0005 51 and 100 bp 2222 77.7 272 and 547 bp 2219 62.8 Tm d Single sample detection <0.0005 P 0.05 611 88.4 P 0.05 297 64.0 Full-plate imaging 0.12 P 0.05 222 65.8 P 0.05 655 71.3 TmSD (°C) 0.010 Single sample detection 24 0.058 (0.067)e Full-plate imaging 16 0.112 (0.056)e a All characteristics evaluated by 2 analysis (except t-test for TmSD. P values 0.05 are shown in bold. b Two melting protocols were compared on 5 instruments for each product size. c Commercial software provided by each instrument was compared to custom software [Gundry et al. (18 ), Palais and Wittwer (19 )] modified to accept input from all instruments. All 10 instrument configurations and all 4 product sizes were analyzed by both manufacturer and custom software, except for the HR-1 (instrument and custom software are the same) and samples melted at 0.1 °C/s (custom software only). d Tm difference between A/A and T/T obtained from the temperature peaks on negative derivative curves using custom software. e Values [mean (SD)] are of TmSD, not accuracy. High-Resolution DNA Melting Accuracy Clinical Chemistry 60:6 (2014) 869 PCR instruments, often by slowing the melting rate down to obtain adequate resolution. The term "HRM" was trademarked by 1 instrument manufacturer, but there is no definition or standardization of what con-stitutes high-resolution melting, and recommended protocols vary widely among manufacturers. No guidelines such as Minimum Information for Publica-tion of Quantitative Real-Time PCR Experiments (21 ) for quantitative PCR are available for melting. Prior studies to establish instrument precision at-tempted to use constant melting rates of 0.1 °C/s, but in more than half the instruments, slower rates had to be used (16 ). In this study, we focused on measuring the genotype accuracy of 10 different instrument configu-rations under the manufacturers' recommended con-ditions (ramp rates, sample volumes, number of sam-ples, etc.), holding as many assay parameters constant as possible. We found that detection mode, analysis software, and PCR product size were highly signifi-cant factors that affected the genotyping accuracy of homozygotes. The detection mode of an instrument refers to whether an instrument detects samples 1 at a time or by imaging multiple samples on a plate. Single sample de-tection can arise from only measuring 1 sample, or more commonly by scanning the optics over samples 1 at a time. Although imaging is convenient, single sam-ple detection appears superior for differentiating ho- Fig. 3. Analytical sensitivity of homozygote detection across 10 instrument configurations and 4 PCR product sizes. By plotting the sensitivity of detecting the T/T genotype against the A/A genotype, each panel correlates genotype sensitivities in 2 dimensions. Instruments with 100% accuracy appear at the top right corner. All samples were melted under manufacturers' recommended conditions and analyzed with the same custom software. Fig. 4. Accuracy of homozygous genotyping on 10 in-strument configurations with 4 PCR product lengths. The instrument melting rates decrease from left to right as given in Table 1. The manufacturers' recommended melting conditions were used with custom analysis software. 870 Clinical Chemistry 60:6 (2014) mozygotes. Both fluorescence and temperature homo-geneity can be compromised by spatial imaging, which adversely affects the precision of melting curves. Use of a common software improved homozygote genotyping over manufacturers' software. This may re-sult from the ability of the overlay function to focus on curve shape rather than position (Tm). This is surpris-ing because homozygotes are usually differentiated by Tm, and curve overlay is not applied for fear of elim-inating the major difference. However, our evidence suggests that curve overlay and shape differences can be used, at least in some cases, to better genotype homozygotes. Early studies on PCR product length showed that the best accuracy for heterozygote detection occurred when PCR products were400 bp (10 ). Homozygous variant accuracy was assumed to depend on Tm val-ues that are greater with shorter PCR products (18 ). Indeed, genotyping by high-resolution melting is often called "small amplicon genotyping." Our current data indicate that homozygote accuracy does depend on PCR product length, but the dependence is not a sim-ple monotonic function. As the product length in-creases, the accuracy first goes up, then down, and fi-nally up again. Although a general correlation with PCR product size and accuracy may exist, this appears to be highly sequence dependent. Better homozygote accuracy with multiple domains (as seen in the 547-bp product) has been suggested before (5 ) and may ex-plain its improved accuracy over the 272-bp product. The TmSD has been used before as an important metric to predict temperature precision in melting in-struments (14 ). Our data correlate TmSD with the de-tection mode. The TmSD in instruments with full-plate imaging is about twice that observed in instruments with single sample detection. In addition to instrument factors, assay factors are important for genotyping ac-curacy, specifically theTmbetween the 2 homozygous genotypes. However, Tm and TmSD do not explain a majority of the variance in genotyping accuracy (see online Supplemental Fig. 1). Limitations of this study include both instrument and assay limitations. Only certain instruments were available for study, and only a single instrument of each type was assessed. Some instruments were excluded from study because their optics did not match the dye used, and selected instruments may vary in the degree of optical matching. To focus on the instruments, the assay chemistries were held as constant as possible. Only 1 dye and 1 buffer composition were evaluated. Only 1 SNV and 4 PCR product lengths were geno-typed. Because of these limitations, generalizations to other conditions (e.g., instruments and dyes) are tenuous. The template was obtained by nested PCR to elim-inate any differences in concentration or contaminants and pooled before melting. This removed variation at-tributable to PCR and focused only on melting. As such, our protocol could be considered a best-case sce-nario, not to be expected in real-world performance. However, we chose the worst-case scenario in the SNV, where homozygotes are predicted to have identical melting curves. Our purpose was to focus on melting and obtain different accuracies on different instru-ments to determine characteristics that affect genotyp-ing accuracy. Detection of gene variants by high-resolution melting depends on subtle melting curve differences, including shape changes and melting temperature shifts (22-24 ). Alternate methods can been used to in-crease the detection sensitivity of melting assays, such as decreasing the PCR product length (10 ), calibration with an internal oligonucleotide duplex (13 ), quanti-tative heteroduplex analysis after mixing with a known genotype (25 ), or the use of unlabeled probes (14, 15 ) or snapback primers (26, 27 ). Nevertheless, melting of PCR products has an inherent simplicity that is very attractive. We hope to continue improving the accu-racy of this method by identifying and implementing critical parameters. High-resolution melting is not a static achievement, but a continuum. As instruments and methods get better, so will melting accuracy and performance. Author Contributions: All authors confirmed they have contributed to the intellectual content of this paper and have met the following 3 re-quirements: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; and (c) final approval of the published article. Authors' Disclosures or Potential Conflicts of Interest: Upon man-uscript submission, all authors completed the author disclosure form. Disclosures and/or potential conflicts of interest: Employment or Leadership: C.T. Wittwer, Clinical Chemistry, AACC. Consultant or Advisory Role: None declared. Stock Ownership: None declared. Honoraria: None declared. Research Funding: M. Li, the China Scholarship Fund; C.T. Wit-twer, BioFire Diagnostics, Canon US Life Sciences. Expert Testimony: None declared. Patents: R.A. Palais, US 8068992 B2, US 8296074 B2; C.T. Wittwer, US 6174670. Other Remuneration: L. Zhou, University of Xiamen. Role of Sponsor: The funding organizations played no role in the design of study, choice of enrolled patients, review and interpretation of data, or preparation or approval of manuscript. Acknowledgments: We thank Ying Wang for technical assistance and Jesse Montgomery for reviewing the manuscript. High-Resolution DNA Melting Accuracy Clinical Chemistry 60:6 (2014) 871 References 1. Wittwer CT, Reed GH, Gundry CN, Vandersteen JG, Pryor RJ. High-resolution genotyping by am-plicon melting analysis using LCGreen. Clin Chem 2003;49:853- 60. 2. Liew M, Pryor R, Palais R, Meadows C, Erali M, Lyon E, Wittwer C. Genotyping of single-nucleotide polymorphisms by high-resolution melting of small amplicons. Clin Chem 2004;50: 1156-64. 3. Wittwer CT. High-resolution DNA melting analysis: advancements and limitations. Hum Mu-tat 2009;30:857-9. 4. Vossen RH, Aten E, Roos A, den Dunnen JT. High-resolution melting analysis (HRMA): more than just sequence variant screening. Hum Mutat 2009;30:860-6. 5. Montgomery JL, Sanford LN, Wittwer CT. High-resolution DNA melting analysis in clinical re-search and diagnostics. Expert Rev Mol Diagn 2010;10:219-40. 6. Zhou L, Palais RA, Ye F, Chen J, Montgomery JL, Wittwer CT. Symmetric snapback primers for scanning and genotyping of the cystic fibrosis transmembrane conductance regulator gene. Clin Chem 2013;59:1052- 61. 7. Li M, Liu L, Liu Z, Yue S, Zhou L, Zhang Q, Cheng S, Li RW, Smith PN, Lu S. The status of KRAS mutations in patients with non-small cell lung cancers from mainland China. Oncol Rep 2009; 22:1013-20. 8. Li M, Zhang Q, Liu L, Liu Z, Zhou L, Wang Z, Yue S, Xiong H, Feng L, Lu S. The different clinical significance of EGFR mutations in exon 19 and 21 in non-small cell lung cancer patients of China. Neoplasma 2011;58:74-81. 9. Dobrowolski SF, Wittwer CT. High resolution melt profiling. In: Rapley R, Harbron S, eds. Molecular analysis and genome discovery. Hoboken, NJ: Wiley 2011:81-113. 10. Reed GH, Wittwer CT. Sensitivity and specificity of single-nucleotide polymorphism scanning by high-resolution melting analysis. Clin Chem 2004; 50:1748 -54. 11. Santa Lucia J Jr, Hicks D. The thermodynamics of DNA structural motifs. Annu Rev Biophys Biomol Struct 2004;33:415- 40. 12. Montgomery J, Wittwer CT, Kent JO, Zhou L. Scanning the cystic fibrosis transmembrane con-ductance regulator gene using high-resolution DNA melting analysis. Clin Chem 2007;53: 1891-8. 13. Gundry CN, Dobrowolski SF, Martin YR, Robbins TC, Nay LM, Boyd N, Coyne T, Wall MD, Wittwer CT, Teng DH. Base-pair neutral homozygotes can be discriminated by calibrated high-resolution melting of small amplicons. Nucleic Acids Res 2008;36:3401- 8. 14. Herrmann MG, Durtschi JD, Bromley LK, Wittwer CT, Voelkerding KV. Amplicon DNA melting anal-ysis for mutation scanning and genotyping: cross-platform comparison of instruments and dyes. Clin Chem 2006;52:494 -503. 15. Herrmann MG, Durtschi JD, Bromley LK, Wittwer CT, Voelkerding KV. Instrument comparison for heterozygote scanning of single and double heterozygotes: a correction and extension of Herrmann et al., Clin Chem 2006;52:494-503. Clin Chem 2007;53:150 -2. 16. Herrmann MG, Durtschi JD, Wittwer CT, Voelk-erding KV. Expanded instrument comparison of amplicon DNA melting analysis for mutation scanning and genotyping. Clin Chem 2007;53: 1544-8. 17. Sanford LN, Wittwer CT. Monitoring temperature with fluorescence during real-time PCR and melt-ing analysis. Anal Biochem 2013;434:26 -33. 18. Gundry CN, Vandersteen JG, Reed GH, Pryor RJ, Chen J, Wittwer CT. Amplicon melting analysis with labeled primers: a closed-tube method for differentiating homozygotes and heterozygotes. Clin Chem 2003;49:396-406. 19. Palais R, Wittwer CT. Mathematical algorithms for high-resolution DNA melting analysis. Meth-ods Enzymol 2009;454:323- 43. 20. Dwight Z, Palais R, Wittwer CT. uMELT: predic-tion of high-resolution melting curves and dy-namic melting profiles of PCR products in a rich web application. Bioinformatics 2011;27:1019- 20. 21. Bustin SA, Benes V, Garson JA, Hellemans J, Huggett J, Kubista M, et al. The MIQE guidelines: minimum information for publication of quanti-tative real-time PCR experiments. Clin Chem 2009;55:611-22. 22. Vandersteen JG, Bayrak-Toydemir P, Palais RA, Wittwer CT. Identifying common genetic variants by high-resolution melting. Clin Chem 2007;53: 1191-8. 23. Reed GH, Kent JO, Wittwer CT. High-resolution DNA melting analysis for simple and efficient molecular diagnostics. Pharmacogenomics 2007; 8:597- 608. 24. Erali M, Voelkerding KV, Wittwer CT. High reso-lution melting applications for clinical laboratory medicine. Exp Mol Pathol 2008;85:50-8. 25. Palais RA, Liew MA, Wittwer CT. Quantitative heteroduplex analysis for single nucleotide poly-morphism genotyping. Anal Biochem 2005;346: 167-75. 26. Zhou L, Errigo RJ, Lu H, Poritz MA, Seipp MT, Wittwer CT. Snapback primer genotyping with saturating DNA dye and melting analysis. Clin Chem 2008;54:1648 -56. 27. Farrar JS, Palais RA, Wittwer CT. Snapback primer genotyping of the Gilbert syndrome UGT1A1 (TA)(n) promoter polymorphism by high-resolution melting. Clin Chem 2011;57:1303-10. 872 Clinical Chemistry 60:6 (2014)