Fig 1.
Temperature-dependent quenching of fluorescence of free and dsDNA-bound fluorophore and its correction.
(A) Plot of first-order polynomial curve fit of raw RFU vs. temperature in no template controls (NTC). The equation shown in the plot is the mean ± SD of six different sample slopes and constants. (B) The unprocessed high-resolution melting profile (blue trace) and the extrapolation from first-order polynomial curve fitting of the post-melt curve region (red dashed line) from an amplicon of an unedited target site. (C) High-resolution melting profile of background subtracted RFU (BcRFU, blue trace) and that of ‘unquenched’ or fluorescence-compensated BcRFU (FcRFU, green trace) from an unedited target site. The red dashed line shows extrapolation of pre-melt region from first-order polynomial curve fitting of BcRFU and depicts the predicted BcRFU in the absence of DNA melting. D) Comparison of first-order polynomial curve fitting of post-melt and pre-melt portions of melting curves. Normalized data were used to enable plotting of the two sets of data.
Fig 2.
GD of first derivative of high-resolution melt curves of amplicons from gDNA of unmodified target sites.
gDNA from mock-transfected HEK293T cells (Mocks) were PCR amplified using primer pairs targeting F8-S2 or CCR5 loci to obtain high resolution melt curve data as described in Materials and Methods. The normalized and fluorescence corrected melt curve data (nFcRFU) from F8-S2 (A and C) and CCR5 (B and D) target sites were numerically differentiated as described in Materials and Methods (Eq 7). 1-GD (A and B) and 2-GD curve fitting of derivative melt curves were done using CurveExpert Professional using Eq 9 and Eq 10, respectively. The first derivative (y-axis: -d(nFcRFU)/dT) was plotted against temperature (x-axis) and is shown as blue dots. The 1-GD curve fit to the first derivative data is shown as a red trace in A and B. The individual Gaussians of 2-GD curve fit are shown as brown (g2(x)) or green dashed lines (g3(x)) and their sum (g2(x) + g3(x)) is depicted as a solid red line in C and D. Table E shows the Gaussian parameters determined from 1-GD curve fitting of A and B, while Table F shows the parameters identified by 2-GD curve fitting of C and D using the CurveExpert Professional software.
Table 1.
2-GD model shows better fit than 1-GD for derivative melt curve data of mocks.
Fig 3.
3-GD of first derivative of high-resolution melt curves for estimation of mutant percentage in genome-edited samples.
gDNA was isolated from HEK293T cells transfected with F8-S2 targeting RGENs or CCR5 targeting TALENs and PCR amplified using corresponding primer pairs to obtain high resolution melt curve data (Materials and Methods). 3-GD curve fitting was done on first derivative melt curves using CurveExpert Professional and Eq 12 as described in Materials and Methods. The individual Gaussians-g1(x) (purple dashed line), g2(x) (brown dashed line) and g3(x) (green dashed line) and their sum- g1(x)+ g2(x) + g3(x) (red solid line) were overlaid over the first derivative melt curve (blue dots). GD of F8-S2 is shown in A and of CCR5 in B. Table C shows the parameters (weights, centers and SDs) of 3-GD. The parameters that were fixed from GD of mocks and those that were set free during 3-GD of edited samples are shown in the Comments column. The g1 weight (w1) represents the mutation frequencies in the amplicons of genome-edited F8-S2 and CCR5 target sites, respectively.
Fig 4.
Comparison of mutant percentage estimation by 2- and 3-GD.
First derivatives of high-resolution melt curves from genome-edited samples were curve fitted using 2- or 3-GD models as described in Materials and Methods (Eq 11 and Eq 12, respectively). The mutant percentages estimated from curve fitting are shown along the y-axis for F8-S2 (A) and CCR5 (B). Two molecular clones (10 and 11) of dgRNAs targeting F8-S2 site and two pairs of TALENs (L1R1 and L2R2) targeting CCR5 site were tested. The mutant percentages were compared using Student’s t-test (two-tailed). The p-values of the pair-wise comparisons of 2-GD and 3-GD are shown above the bars.
Table 2.
3-GD model achieves better fit than 2-GD for derivative melt curve data of genome-edited samples*.
Fig 5.
Validation of GD method using predefined amplicon mixes.
High-resolution melt curves of samples containing different proportions of F8-S3 amplicon (S3Wt) or F8-S2 mutant (S2Mt) in F8-S2 amplicon (S2Wt) were analyzed by GD as detailed in Materials and Methods. (A) Derivative melt curve data (blue dots) of indicated S3Wt-S2Wt mixes were fitted using 3-GD (red traces). The nominal percentage of S3Wt in the mix is shown below (indicated by S3Wt%) and the GD-estimated amount in the top left corner of each plot. (B) Derivative melt curve data (blue dots) of indicated S2Mt-S2Wt mixes were fitted using 3-GD (red traces). The nominal percentage of S2Mt in the mix is shown below (indicated by S2Mt%) and the GD-estimated amount in the top left corner of each plot. (C) Scatter plot of nominal F8-S3Wt% in mix (X-axis) vs 3-GD estimated F8-S3Wt% (Y-axis). (D) Scatter plot of nominal F8-S2Mt% in mix (X-axis) vs 3-GD estimated F8-S2Mt% (Y-axis). The equations of linear regression analysis of both types of dose-response curves and the correlation coefficients (R2) are shown. The samples were tested in duplicate (replicates ‘a’ and ‘b’).
Fig 6.
Mutant frequency determination by 3-GD and comparison to difference curve areas (DCAs) and next generation sequencing (NGS) data.
HEK293T cells were transfected with F8-S2 targeting dgDNA clone 10 (F8-S2 Cl.10) or clone 11 (F8-S2 Cl.11) together with a dCas9-FokI construct. The cells were also cotransfected with either pBackbone or pDonor-F8 targeting plasmids (Materials and Methods). Following transfection, gDNAs were isolated from unselected cells or cells selected with puromycin and used for amplification by PCR using appropriate primer pairs targeting F8-S2 loci to obtain high-resolution melt curve data. (A) Mutant percentage estimations by 3-GD for the four different categories of samples from unedited and edited F8-S2 site are identified on the left. The derivative melt curves are shown as blue dots and the fitted curves from GD as red traces. Four PCR replicates were analyzed for each clone with one exception (F8-S2 clone 10, pBackbone/Unselected) for which only three replicates were tested. The mutant frequency (percentage) estimated from the area of the mutant peak (w1 parameter from g1(x)), of 3-GD) for each replicate is shown within the plot. (B-D) The average mutant frequency determined by GD for the different categories in A were compared to mutant frequencies determined by difference curve areas (DCA) (C) and to mutant frequency determination from next generation sequencing (NGS). NGS was only done on unselected samples. (E) Mutant frequency estimation from GD of high resolution melt curve data from gDNA of HEK293T cells transfected with TALENs (two independent pairs of molecular clones L1R1, L2R2) targeting CCR5 locus. CCR5 edited samples were also analyzed by NGS. Error bar = 1 SD.
Fig 7.
Size of PCR product does not affect determination of mutant percentage by GD.
The CCR5 target site in gDNA of unmodified or genome-edited cells were amplified using two pairs of primers designed to produce two distinct sizes of product (107 bp and 140 bp, respectively). The amplicons were subjected to high-resolution melting and then processed to correct for temperature-dependent quenching of fluorescence of free and dsDNA-bound fluorophore. The resulting melt curves of genome-edited (for clone pair L1R1) and unmodified controls (Mock) are shown (A & C). Corresponding first-derivatives of processed melt curves are shown in B and D. Replicates G1 and G2, A1 and A2 refer to gDNA samples amplified using primers that produce 107 bp amplicon, whereas G5 and G6, and A5 and A6 refer to gDNA samples amplified using primers that produce 140 bp amplicon. The derivative melt curves were decomposed using the 3-GD model to estimate the mutant frequency. The estimated mutant frequencies for both sizes of amplicons are shown in (E). Error bar = 1 SD.
Table 3.
Parameters determined by 3-GD of two different size amplicons from the CCR5-edited target site.
Fig 8.
Flow diagram of steps involved in processing high-resolution melt curve data for GD.
(A-D) The steps needed to process the raw melt curve data to correct for background from free fluorophore (A-B) and to correct for temperature dependent quenching of bound-fluorophore (C-D) by dividing background-corrected RFU (BcRFU) by the efficiency (C). The normalized fluorescence corrected RFU (nFCRFU) of mocks is then differentiated (E) before curve fitting using 2-GD (F) to determine the parameters values for use as constants in 3-GD of genome-edited samples. (G) Processed melt curves of genome-edited samples using the same steps outlined in A-E are then curve-fitted by 3-GD as described in the text. Equation numbers refer to those in Materials and Methods.
Fig 9.
Comparison of different methods of processing melt curve data for background and fluorescence quenching correction.
Melt curve data from amplicons of unmodified or control samples from F8-S2 (A) or CCR5 target loci (B) were either unprocessed (-dF/dT, blue trace) or corrected using exponential background subtraction method of Palais and Wittwer (24) (-dF/dT-dB/dT, red dashes) or the method described in this study (-d(nFcRFU)/dT, green trace).
Fig 10.
Analysis of F8-S2 and CCR5 target sequence features and melting properties in silico.
Sliding window analysis of percentage of AT (%AT) in F8-S2 (A) or CCR5 (B) sequences of target sites amplified by PCR. The percentage of As and Ts were determined in a sliding overlapping window of 10-mers. The shift was by 1 bp. These are shown as green dashes. The data was smoothed using running averages with a period of 5 (solid green line). The sum of free energies (∆Gs) in a sliding window of 10-mers and a shift of 1 bp is shown along the left y-axis in kJ/mol (blue dots). The running averages were calculated as for %AT traces and are shown as blue traces. Putative AT-rich domains are marked I-IV. (C- H) The F8-S2 and CCR5 target sequences were used as input in the UMelt web analysis tool (29). UMelt predicted derivative melt curve (C and D), "Dynamic Profile” of melting (E and F) using a sliding temperature control that was situated close to the predicted Tm for each sequence to identify portions of the target sequences (nucleotide position indicated on the x-axis) that may have melted earlier than the rest. The web tool also provided a "Melting Profile" analysis that shows potential regions that might show greater tendency to melt earlier (G and H).