High-Throughput Variant Detection Using a Color-Mixing Strategy

Abstract

Many diseases are related to multiple genetic alterations within a single gene. Probing for highly multiple (>10) variants in a single quantitative PCR tube is impossible because of a limited number of fluorescence channels and the limited ability to test one variant per channel, increasing the need for tubes. Herein, a novel color-mixing strategy was experimentally validated that uses fluorescence combinations as digital color codes to probe multiple variants simultaneously. The color-mixing strategy relies on a simple intratube assay that can probe for 15 variants as part of an intertube assay that can probe for an exponentially increased number of variants. This strategy is achieved by using multiplex double-stranded toehold probes modified with fluorophores and quenchers; the probes are designed to be quenched or remain luminous after binding to wild-type or variant templates. The color-mixing strategy was used to probe for 21 pathogenic variants in thalassemia and to distinguish between heterozygous and homozygous variants in six tubes, with a specificity of 99% and a sensitivity of 94%. To support tuberculosis diagnosis, the same strategy was applied to simultaneously probe in Mycobacterium tuberculosis for rifampicin-resistance mutations occurring within one 81-bp region and one 48-bp region in the rpoB gene, plus five isoniazid-resistance mutations in the inhA and katG genes.

Nucleic acid variants are important biomarkers for disease diagnosis. High-throughput detection of variants, including single-nucleotide variants and small insertions and deletions, can improve the sensitivity and specificity of diagnosis. Next-generation sequencing (NGS) is the most commonly used high-throughput variant detection method. However, the overall workflow of NGS is relatively long, and the cost is fairly high.¹ Droplet digital PCR is a highly sensitive detection method with a fast turnaround time that can be used to detect and quantitate variants accurately. However, because of the limitation of the number of fluorescence channels used, droplet digital PCR is usually used for single-plex or duplex variant detection.

Quantitative PCR (qPCR) is the most commonly used method because the instruments are easy to access, and the overall workflow is simple. qPCR can be used for the detection and quantitation of single variants by TaqMan probes and molecular beacon.² However, the detection of highly multiplex variants is challenging in a qPCR because the number of fluorescence channels available in a qPCR instrument is six or less.³ With one channel used for each single-plex variant, simultaneous variant detection using a qPCR instrument is limited to 6-plex. Furthermore, it is hard to distinguish a germline heterozygous variant (50%) from a homozygous variant (100%) using qPCR because the threshold cycle value of a 50% variant sample can be close to that of a 100% variant sample.

Toehold probe is a double-stranded conditionally fluorescent molecular probe that can be used to discriminate single or more base changes, and it is designed on the basis of hybridization. As hybridization proceeds in various buffer conditions, toehold probe can work robustly for a wide range of conditions (ie, salt and temperature).⁴ Toehold probe was previously used for the detection of point mutations in the Escherichia coli rpoB gene.⁵ And a toehold probe set with log-linear response curves enables accurate enzyme-free quantitation in both DNA and RNA.⁶

Herein, a color-mixing strategy was presented that can be used to identify multiplex variants and their heterogeneity. The color-mixing strategy can be implemented through two steps. First, an asymmetric PCR amplification is performed to preferentially enrich one strand within the region of interest over the other; second, end-point fluorescence detection is enabled with multiplex rationally designed toehold probes.^5,7

Based on this strategy, an assay was developed that can detect the 21 most common pathogenic variants in thalassemia and can further distinguish the heterogeneity of those variants.⁸ An assay for Mycobacterium tuberculosis bacteria (MTB) using this strategy was also developed that can theoretically detect any drug-resistant mutation within one 81-bp and one 48-bp region in the rpoB gene, in which 10 variants were experimentally validated, and can also detect another 5 variants in the inhA and katG genes.9, 10, 11, 12

Materials and Methods

Color-Mixing Strategy

Intratube Color-Mixing Design

Four commonly used fluorescence channels with minimum spectral overlap were used for each tube: ROX, CY5, FAM, and HEX. The final output from each fluorescence channel has two states, on (1) and off (0). When making full use of the four fluorescence channels, there are 15 (2⁴ − 1) state combinations indicating 15 variants, excluding the all-off state. Each state combination was given a color code. Note that for this simple design, it is expected that the genetic disorders this color-mixing strategy is targeting will be predominantly caused by a single variant in any individual (ie, there are no coexisting genetic variants within the genome, so the test sample would generally contain one variant). In addition, this strategy would be suitable for infectious diseases in which the genetic variants of pathogens cause drug resistance, assuming that there is usually a single variant within the gene that causes the drug resistance. In the case of multi–drug resistance, multiple genes would contribute to the multi–drug resistance, with each gene likely to have a single drug-resistance mutation.

Multiplex double-stranded toehold probes, modified with four different fluorophores and quenchers, were implemented in one tube. A single on state from probes 1 to 4 can thus detect variants 1 to 4 (Figure 1A). In addition, simultaneous signals from probes with different fluorophores can target other variants: two on states from probe-5-ROX and probe-5-CY5 indicate the presence of variant 5; and three on states from probe-14-CY5, probe-14-FAM, and probe-14-HEX indicate the presence of variant 14.

Figure 1.

Color-mixing concept and experimental workflow. A: Intratube color mixing. ROX, CY5, FAM, and HEX fluorescence channels are used for each tube. Within one tube, each fluorescence channel's signal can be identified separately by the instrument, and the final output from each fluorescence channel has two states: on and off. A combination of four channels yields 15 (2⁴ – 1) state combinations, and each state combination can be denoted as a color code, with 15 color codes representing 15 individual variants. B: Intertube color mixing. When increasing the number of tubes, the number of detected variants can be exponentially increased: a combination of three tubes can be used to identify as many as 3375 (15³) variants. C: Example of experiment workflow. An asymmetric PCR was conducted on a mixture of the DNA template, multiplex asymmetric primers, and variant-specific toehold probes. Fluorescence signals in different channels were measured before and after asymmetric PCR as background signals and raw signals. The signal differences (raw signal – background signal) were used for evaluating detection performance. After asymmetric PCR, variant-specific toehold probes will bind to the variant amplicons but not to the wild-type amplicons. gDNA, genomic DNA.

Intertube Color-Mixing Design

The number of simultaneously identifiable targeted variants is exponentially increased when the number of tubes used in an assay increases; the number of targeted variants is 225 (15²) for two tubes and 3375 (15³) for three tubes. For example, the color codes 1,1,1 can indicate the presence of variant 1, the color codes 1,2,2 can be used to denote the presence of variant 17, and the color codes 15,15,15 can be used to denote the presence of variant 3375 (Figure 1B).

Experimental Workflow

The experimental workflow is shown in Figure 1C. The total volume of each reaction was 30 μL (Supplemental Tables S1 and S2). All multiplex asymmetric primers, a set of rationally designed probes, DNA template, and iTaq Universal Probes Supermix (catalog number 1725134; Bio-Rad, Hercules, CA) were mixed, followed by the execution of a thermal cycling program. The same amount of DNA template was used as the input to each tube. To correct for background noise variation, the fluorescence of the first eight cycles was measured as the background signal before asymmetric PCR. The asymmetric PCR started with 3 minutes at 95°C for polymerase activation, followed by 50 cycles of 30 seconds at 95°C for DNA denaturing, 30 seconds at 64°C for annealing, and 30 seconds or 2 minutes at 72°C for extension (with the extension time dependent on amplicon length). After the asymmetric PCR, the entire system was heated to 95°C for 5 minutes to dissociate all double strands, then cooled to 40°C for probes binding to the target amplicon. Twenty cycles of fluorescence were measured at 40°C as the raw signal (details of the protocol are shown in Supplemental Tables S3 and S4). For further data analysis, the final signals were obtained as raw signals minus background signals for each channel (Supplemental Figures S1 and S2).

Toehold Probe Design

Variant Probe Design

The design principle for a variant probe is shown in Figure 2A. A variant probe comprises a protector strand (P) and a complementary strand (C).^5,7

Figure 2.

Toehold probe design. A: The variant probe comprises two strands, a complementary strand (C strand) and a protector strand (P strand). The variant probe C strand is designed to be perfectly complementary to a variant strand (V strand), with a fluorophore attached at its 5′ end and a C3 spacer attached at its 3′ end; the P strand has a quencher at its 3′ end. Green rectangles outline the predominant species at equilibrium: When the V strand is present, the P strand will be displaced, the C strand will bind to the V strand and fluoresce, and the predominant species will be VC; when the wild-type strand (WT strand) is present, because of the mismatch energy ΔΔG_Mismatch between the WT strand and the C strand, the P strand and C strand will favorably bind to each other, the predominant species will be PC, and the fluorophore will be quenched. Pink-highlighted letters represent the wild-type sequence, and green-highlighted letters represent the variant sequence. B: Wild-type probe design. A wild-type probe also comprises a C strand and a P strand. However, instead of perfectly binding to a specific variant strand, the wild-type probe C strand is designed to perfectly bind to a WT strand. The wild-type probe P strand will be displaced by the WT strand and fluoresce. Any variant within the designed region in a wild-type probe will cause a mismatch energy ΔΔG_Mismatch to the binding of the C strand; therefore, the wild-type probe will not fluoresce when variant amplicons exist. Purple rectangles outline the predominant species of WT toehold probe presented with either WT strand or two different variant strands at equilibrium. Blue-highlighted letters represent the wild-type sequence, and pink-highlighted letters represent the variant sequence. C: The simulated hybridization yield of a variant toehold probe binding to a variant strand or a WT strand. Herein, 5× V or WT strand, 1× P, 1× PC, and [PC]₀ = 20 nmol/L were used for simulation. The final concentration of ${[TC]}_{\infty }$ at equilibrium can be calculated as follows: $\chi =\frac{[TC]\infty }{[TC]\infty +[PC]\infty }=\frac{x}{c}$. The $\Delta G_{rxn\left(variant\right)}^o$ = 0.61 kcal/mol at 40°C and a salinity of 0.18, and binding yield χ = 85.6%. And the $\Delta G_{rxn\left(WT\right)}^o$ = 2.57 kcal/mol at 40°C and a salinity of 0.18, and binding yield χ = 7.0%. Simulated discrimination factor Q of 32 variant toehold probes is shown in Supplemental Figure S6. D: The simulated hybridization yield of a WT toehold probe binding to variant or WT strand, 5× V or WT, 1× P, 1× PC, and [PC]₀ = 20 nmol/L. Two different variant strands with different ΔΔG_Mismatch will cause slight difference in yield. BM, branch migration region; NH, nonhomologous region.

When there is no target, the P and C strands will hybridize to form a double-stranded probe.

When a variant amplicon strand (V) is present, the standard free energy of variant probe binding to the V strand can be calculated as follows:

$$\Delta G_{rxn\left(variant\right)}^o=\Delta G_{Toehold}^o-\Delta G_{NH}^o-\Delta G_{FQ}^o$$ (1)

When a wild-type amplicon strand (WT strand) is present, the standard free energy of variant probe binding to the WT strand can be calculated as follows:

$$\Delta G_{rxn\left(WT\right)}^o=\Delta G_{Toehold}^o-\Delta G_{NH}^o-\Delta G_{FQ}^o+\Delta \Delta G_{Mismatch}^o$$ (2)

When the variant probe reacts with V strand, the $\Delta G_{rxn\left(variant\right)}^o$ is designed to be <0; the P strand will be displaced by the V strand, and the C strand will bind to the V strand and fluoresce. When the probe reacts with the WT strand, the $\Delta G_{rxn(WT)}^o$ will be >0, and thus will be less thermodynamically favorable; the P and C strands will bind to each other and result in fluorescence quenching. The difference between $\Delta G_{rxn\left(variant\right)}^o$ and $\Delta G_{rxn(WT)}^o$ will result in a difference of hybridization yield (Figure 2B and Supplemental Figure S3).

Each toehold probe has three regions: the toehold region (Toehold), the branch migration region, and the nonhomologous region (Figure 2A). Both Toehold and branch migration region can be rationally designed to perfectly bind to variant amplicons. The Toehold and branch migration regions of the C strand are the reverse complement of the V strand, with a fluorophore attached at 5′ end and a C3 spacer attached at 3′ end. The C3 spacer at 3′ end of the C strand prevents polymerase extension and exonuclease activity during the PCR and, therefore, functions as a primer and ensures that the probes' concentrations remain constant after the PCR. The branch migration region of the P strand is identical to the variant sequence and has a quencher modification at its 3′ end. The nonhomologous regions of the P and C strands are random-generated sequences that are designed not to be complementary to the amplicon, ensuring that the overall reaction energy will be <0 kcal mol⁻¹ when binding to variant strand and >0 kcal mol⁻¹ when binding to WT strand. Fluorophores and quenchers are attached in the nonhomologous region. The empirical energy penalty $\Delta G_{FQ}^o$ of different fluorophore and quencher binding is different, as specified in Supplemental Figure S3.

Wild-Type Probe Design

Unlike the C strand of a variant probe that perfectly binds to a V strand, the C strand of a wild-type probe is designed to be perfectly reverse complementary to the WT strand (Figure 2C). When the WT strand is present, the P strand will be displaced by the WT strand, and the C strand will fluoresce. When the V strand is present, with variants at different positions indicated as variant 1 and variant 2, the P strand binds to the C strand and is thus quenched, resulting in a difference in hybridization yield (Figure 2D). The fluorescence signal from a wild-type probe will have an on state when binding to the WT strand, and an off state when binding to the variant 1 or variant 2 strand. Sequences and concentration of probe in each tube were summarized in Table 1, Table 2, Table 3, Table 4, Table 5, and 6.

Table 1.

Primer Sequences Used to Detect Pathogenic Variants in Thalassemia

Name	Sequence
Thal-HBA-FP-0711	5′-CGGGCCTGGGCCGCA-3′
Thal-HBA-RP-0711	5′-GGTCTGAGACAGGTAAACACCTCCAT-3′
Thal-HBB-FP1-0720	5′-CACCCTAGGGTTGGCCAATCTACTC-3′
Thal-HBB-RP1-0720	5′-ACTTATCCCCTTCCTATGACATGAACTTAACC-3′
Thal-HBB-FP2-0720	5′-CTTTGCACCATTCTAAAGAATAACAGTGATAATTTC-3′
Thal-HBB-RP2-0720	5′-CTATTAGCAATATGAAACCTCTTACATCAGTTACA-3′

FP, forward primer; RP, reverse primer; Thal, Thalassemia.

Table 2.

Primer Sequences Used to Detect Pathogenic Variants in Tuberculosis

Name	Sequence
rpoB507 to 576_FP	5′-GGTCGCCGCGATCAAGG-3′
rpoB507 to 576_RP	5′-GAACCCGAACGGGTTGACC-3′
KatG315_FP	5′-GGGTGTTCGTCCATACGACCT-3′
KatG315_RP	5′-CGCTGGAGCAGATGGGCT-3′
inhA194_FP	5′-CGACGATCGCACTCATCGC-3′
inhA194_RP	5′-GGTCAACAGGTTCGTGGCG-3′

FP, forward primer; RP, reverse primer.

Table 3.

Probe Sequences Used in Thalassemia Assay, Tubes 1 to 3

Name in article	Fluorophore sequence	Quencher sequence
Tube 1
V10-T1-FAM + 1	5’-/56-FAM/TAGGACACCTGACTTTCATGCCCAGC/3SPC3/-3′	5′-AAAGTCAGGTGTCCTA/3IABKFQ/-3′
V10-T1-HEX + 1	5’-/5HEX/TTTCCAACCTGACTTTCATGCCCAGC/3SPC3/-3′	5′-AAAGTCAGGTTGGAAA/3IABKFQ/-3′
V11-T1-FAM	5’-/56-FAM/ATCGCCCCATAGAGTCACCCTG/3SPC3/-3′	5′-TCTATGGGGCGAT/3IABKFQ/-3′
V12-T1-FAM	5’-/56-FAM/CCGGTCCATCACTTAAAGGCACC/3SPC3/-3′	5′-AGTGATGGACCGG/3IABKFQ/-3′
V13-T1/T2-FAM	5’-/56-FAM/TCCCAGACCAGCACCTAAGGGTG/3SPC3/-3′	5′-TGCTGGTCTGGGA/3IABKFQ/-3′
V14-T1-ROX	5’-/56-ROXN/TAGGAACATTGCTATTACCTTAACCCA/3SPC3/-3′	5′-AATAGCAATGTTCCTA/3IABRQSP/-3′
V15-T1-FAM	5’-/56-FAM/CTTTATATGTAAAGGACTAAAAGAACCTCT/3SPC3/-3′	5′-AGTCCTTTACATATAAAG/3IABKFQ/-3′
V16-T1-HEX + 2	5’-/5HEX/TTGCTATAATTGACTTTTATTCCCAGCCCTG/3SPC3/-3′	5′-ATAAAAGTCAATTATAGCAA/3IABKFQ/-3′
V17-T1-Cy5	5’-/5CY5/TACCCGGACTCAACCTCTGGGT/3SPC3/-3′	5′-TTGAGTCCGGGTA/3IABRQSP/-3′
V19-T1-HEX + 1	5’/5HEX/ACTCTGGGTGTCTGAGGTTGCTAGT/3SPC3/-3′	5′-CAGACACCCAGAGT/3IABKFQ/-3′
V20-T1/T2-HEX	5’-/5HEX/ATGCGGGACTCAAAAACCTCTGGG/3SPC3/-3′	5′-TTTGAGTCCCGCAT/3IABKFQ/-3′
V3-T1-FAM + 1	5’-/56-FAM/ACTGGCTGATACCAAACTGCCCAGG/3SPC3/-3′	5′-TTGGTATCAGCCAGT/3IABKFQ/-3′
V4-T1-HEX + 1	5’-/5HEX/CTCAACCACCAACCCGCCCAGGG/3SPC3/-3′	5′-GGTTGGTGGTTGAG/3IABKFQ/-3′
Tube 2
V9-T2-HEX + 1	5’-/5HEX/CCCTTTCCCAGGGGCCTCACCA/3SPC3/-3′	5′-CCCTGGGAAAGGG/3IABKFQ/-3′
V11-T2-ROX + 2	5’-/56-ROXN/CCTGAACCCATAGAGTCACCCTGAA/3SPC3/-3′	5′-TCTATGGGTTCAGG/3IABRQSP/-3′
V12-T2-Cy5 + 1	5’-/5CY5/CTGCTCCATCACTTAAAGGCACCG/3SPC3/-3′	5′-AGTGATGGAGCAG/3IABRQSP/-3′
V13-T1/T2-FAM	5’-/56-FAM/TCCCAGACCAGCACCTAAGGGTG/3SPC3/-3′	5′-TGCTGGTCTGGGA/3IABKFQ/-3′
V15-T2-HEX + 1	5’-/5HEX/CTGCATAAAAGGACTAAAAGAACCTCTG/3SPC3/-3′	5′-AGTCCTTTTATGCAG/3IABKFQ/-3′
V16-T2-ROX	5’-/56-ROXN/CCCGGTGACTTTTATTCCCAGCCC/3SPC3/-3′	5′-ATAAAAGTCACCGGG/3IABRQSP/-3′
V18-T2-FAM + 1	5’-/56-FAM/TCAGATGGTTCACCTAGCCCCACAG/3SPC3/-3′	5′-AGGTGAACCATCTGA/3IABKFQ/-3′
V18-T2-HEX + 1	5’-/5HEX/TCTCAATTGTTCACCTAGCCCCACAG/3SPC3/-3′	5′-AGGTGAACAATTGAGA/3IABKFQ/-3′
V19-T2-FAM + 1	5’-/56-FAM/CCGCGGTGTCTGAGGTTGCTAGT/3SPC3/-3′	5′-CAGACACCGCGG/3IABKFQ/-3′
V20-T1/T2-HEX	5’-/5HEX/ATGCGGGACTCAAAAACCTCTGGG/3SPC3/-3′	5′-TTTGAGTCCCGCAT/3IABKFQ/-3′
V5-T2-ROX + 4	5’-/56-ROXN/CCTGATCCCCACCAGGGCAGTAAC/3SPC3/-3′	5′-GTGGGGATCAGG/3IABRQSP/-3′
V6-T2-Cy5	5’-/5CY5/ACTGGCCTGACTTCTATGCCCAG/3SPC3/-3′	5′-AAGTCAGGCCAGT/3IABRQSP/-3′
V8-T2-FAM + 1	5’-/56-FAM/CCCGGACCTTGATAGCAACCTGCC/3SPC3/-3′	5′-TATCAAGGTCCGGG/3IABKFQ/-3′
Tube 3
V1-T3-CY5	5’-/5CY5/CTCCAAGCCTCCCCGGACAAGT/3SPC3/-3′	5′-GGGAGGCTTGGAG/3IABRQSP/-3′
V1-T3-HEX + 2	5’-/5HEX/CCCTCTGCCTCCCCGGACAAGTTC/3SPC3/-3′	5′-GGGAGGCAGAGGG/3IABKFQ/-3′
V21-T3-FAM	5’-/56-FAM/CCCGAACTGGTGGTAAGGCCCTGG/3SPC3/-3′	5′-ACCACCAGTTCGGG/3IABKFQ/-3′
V21-T3-HEX	5’-/5HEX/CCGAATGTGGTGGTAAGGCCCTGG/3SPC3/-3′	5′-ACCACCACATTCGG/3IABKFQ/-3′
V2-T3-Cy5	5’-/5CY5/TACTAGAGAATACCGTCAAGCTGGAG/3SPC3/-3′	5′-ACGGTATTCTCTAGTA/3IABRQSP/-3′
V2-T3-FAM + 1	5’-/56-FAM/CCCAACTAATACCGTCAAGCTGGAGC/3SPC3/-3′	5′-ACGGTATTAGTTGGG/3IABKFQ/-3′
V2-T3-HEX + 1	5’-/5HEX/CCCTTTGAATACCGTCAAGCTGGAGC/3SPC3/-3′	5′-ACGGTATTCAAAGGG/3IABKFQ/-3′
V7-T3-FAM + 1	5’-/56-FAM/CCTTTCCTACCCTTAGACCCAGAGG/3SPC3/-3′	5′-AAGGGTAGGAAAGG/3IABKFQ/-3′

T, tube; V, variant.

Table 4.

Probe Sequences Used in Thalassemia Assay, Tubes 4 to 6

Name in article	Fluorophore sequence	Quencher sequence
Tube 4
T4-ROX	5’-/56-ROXN/TAGGAACGTCCAGGGAGGCGTGCACCG/3SPC3/-3′	5′-GCCTCCCTGGACGTTCCTA/3IABRQSP/-3′
T4-Cy5	5’-/5CY5/TACTAGAAACTCAAAGAACCTCTGGGTCCAAGGGTAGA/3SPC3/-3′	5′-GACCCAGAGGTTCTTTGAGTTTCTAGTA/3IABRQSP/-3′
T4-HEX	5’-/5HEX/ACTACGCACCTTGCCCCACAGGGCAGT/3SPC3/-3′	5′-GGCAAGGTGCGTAGT/3IABKFQ/-3′
T4-FAM	5’-/56-FAM/CTGATGGGGCTCCAGCTTAACGGTATTTGG/3SPC3/-3′	5′-AAGCTGGAGCCCCATCAG/3IABKFQ/-3′
Tube 5
T5-ROX	5’-/56-ROXN/ACCGCCCACCCTTAGGCTGCTGGT/3SPC3/-3′	5′-CCTAAGGGTGGGCGGT/3IABRQSP/-3′
T5-CY5	5’-/5CY5/TCCCGGTGAGGCCCTGGGCAGGTTGGTATCAAGGTT/3SPC3/-3′	5′-ACCAACCTGCCCAGGGCCTCACCGGGA/3IABRQSP/-3′
T5-FAM	5’-/56-FAM/CTATCGCACTTCAGGGTGAGTCTATGGGAC/3SPC3/-3′	5′-TCACCCTGAAGTGCGATAG/3IABKFQ/-3′
Tube 6
T6-ROX	5’-/56-ROXN/CTCGGGTCGGTGCCTTTAGTGATGGCC/3SPC3/-3′	5′-AAGGCACCGACCCGAG/3IABRQSP/-3′
T6-CY5	5’-/5CY5/CAGTTTATATGGCTGGGCATAAAAGTCAGGGC/3SPC3/-3′	5′-TATGCCCAGCCATATAAACTG/3IABRQSP/-3′
T6-FAM	5’-/56-FAM/CCGGGTTCTGGGTTAAGGCAATAGCAATAT/3SPC3/-3′	5′-CCTTAACCCAGAACCCGG/3IABKFQ/-3′
T6-HEX	5’-/5HEX/AATCACCACCTCAAACAGACACCATGGTGCATCT/3SPC3/-3′	5′-TGTCTGTTTGAGGTGGTGATT/3IABKFQ/-3′

T, tube.

Table 5.

Probe Concentrations Used in Thalassemia Assay

Name in article	Fluorophore concentration, nmol/L	Quencher concentration, nmol/L
Tube 1
V10-T1-FAM + 1	10	20
V10-T1-HEX + 1	20	40
V11-T1-FAM	10	20
V12-T1-FAM	20	40
V13-T1/T2-FAM	10	200
V14-T1-ROX	10	20
V15-T1-FAM	10	40
V16-T1-HEX + 2	10	20
V17-T1-Cy5	10	20
V19-T1-HEX + 1	10	20
V20-T1/T2-HEX	10	200
V3-T1-FAM + 1	10	20
V4-T1-HEX + 1	10	20
Tube 2
V9-T2-HEX + 1	10	100
V11-T2-ROX + 2	10	20
V12-T2-Cy5 + 1	20	40
V13-T1/T2-FAM	20	200
V15-T2-HEX + 1	10	20
V16-T2-ROX	10	20
V18-T2-FAM + 1	20	40
V18-T2-HEX + 1	50	100
V19-T2-FAM + 1	10	20
V20-T1/T2-HEX	10	20
V5-T2-ROX + 4	10	20
V6-T2-Cy5	5	10
V8-T2-FAM + 1	10	200
Tube 3
V1-T3-CY5	10	20
V1-T3-HEX + 2	30	60
V21-T3-FAM	10	20
V21-T3-HEX	10	20
V2-T3-Cy5	10	20
V2-T3-FAM + 1	10	20
V2-T3-HEX + 1	30	60
V7-T3-FAM + 1	10	20
Tube 4
T4-ROX	10	20
T4-Cy5	30	60
T4-HEX	50	150
T4-FAM	10	20
Tube 5
T5-ROX	20	40
T5-CY5	50	300
T5-FAM	10	500
Tube 6
T6-ROX	20	200
T6-CY5	10	20
T6-FAM	10	20
T6-HEX	50	100

T, tube; V, variant.

Table 6.

Probe Sequences and Concentrations Used in Tuberculosis Assay

Probe name	Fluorophore concentration, nmol/L	Fluorophore sequence	Quencher concentration, nmol/L	Quencher sequence
Tube 1
rpoB 507 to 533-Cy5-plus	25	5’-/5Cy5/TTCTCAACAGCGCCGACAGTCGGCGCTTGTGGGTCAACCCCGACAGCGGGTTGTTCTGGTCCATGAATTGGCTCAGCTGGCTGGTGCCGA/3SpC3/-3′	600	5′-CCAGCTGAGCCAATTCATGGACCAGAACAACCCGCTGTCGGGGTTGACCCACAAGCGCCGACTGTCGGCGCTGTTGAGAA/3IAbRQSp/-3′
rpoB 561 to 576-HEX-plus	50	5-’/5HEX/TCTCGCGACAGCGAGCCGATCAGACCGATGTTGGGCCCCTCAGGGGTTTCGATCG/3SpC3/-3′	300	5′-CCCTGAGGGGCCCAACATCGGTCTGATCGGCTCGCTGTCGCGAGA/3IABkFQ/-3′
Tube 2
rpoB 507 to 533-Cy5-minus	15	5’-/5Cy5/CCTACGGGCACCAGCCAGCTGAGCCAATTCATGGACCAGAACAACCCGCTGTCGGGGTTGACCCACAAGCGCCGACTGTCGGCGCTGGGGCC/3SpC3/-3′	400	5′-CCGACAGTCGGCGCTTGTGGGTCAACCCCGACAGCGGGTTGTTCTGGTCCATGAATTGGCTCAGCTGGCTGGTGCCCGTAGG/3IAbRQSp/-3′
rpoB 561 to 576-HEX-minus	20	5’-/5HEX/CCCTACATCGAAACCCCTGAGGGGCCCAACATCGGTCTGATCGGCTCGCTGTCGGTGT/3SpC3/-3′	270	5′-CGAGCCGATCAGACCGATGTTGGGCCCCTCAGGGGTTTCGATGTAGGG/3IABkFQ/-3′
Tube 3
KatG S315T1-FAM	20	5’-/56-FAM/CCTCTGGGACGCGATCACCACAGGCATCG/3SpC3/-3′	40	5′-TGGTGATCGCGTCCCAGAGG/3IABkFQ/-3′
KatG S315T1-HEX	20	5’-/5HEX/CCTTGGGACGCGATCACCACAGGCATC/3SpC3/-3′	40	5′-GGTGATCGCGTCCCAAGG/3IABkFQ/-3′
KatG S315T2-FAM	10	5’-/56-FAM/TACCCTGGACGCGATCACCACCGGCATC/3SpC3/-3′	20	5′-TGGTGATCGCGTCCAGGGTA/3IABkFQ/-3′
KatG S315I-FAM	20	5’-/56-FAM/TTCTCAGGGACGCGATCACCATCGGCATCG/3SpC3/-3′	40	5′-GATGGTGATCGCGTCCCTGAGAA/3IABkFQ/-3′
KatG S315I-ROX	25	5’/56-ROXN/CCCTCTGGACGCGATCACCATCGGCATCG/3SpC3/-3′	50	5′-TGGTGATCGCGTCCAGAGGG/3IAbRQSp/-3′
KatG S315R-FAM	20	5’-/56-FAM/CTAGGATCATCACCCGCGGCA/3SpC3/-3′	40	5′-TGATGATCCTAG/3IABkFQ/-3′
KatG S315R-CY5	20	5’-/5Cy5/CTAGAAGTGGACGCGATCACCCGCGGCA/3SpC3/-3′	40	5′-GGTGATCGCGTCCACTTCTAG/3IAbRQSp/-3′
KatG S315R-HEX	20	5’-/5HEX/CTCTAATGGGACGCGATCACCCGCGGCA/3SpC3/-3′	40	5′-GGTGATCGCGTCCCATTAGAG/3IABkFQ/-3′
inhA I194T-HEX	10	5’-/5HEX/CCGACGCCGCAGGCCCTACCCGGACG/3SpC3/-3′	20	5′-AGGGCCTGCGGCGTCGG/3IABkFQ/-3′
inhA I194T-ROX	30	5’-/56-ROXN/CTAGGACGCCGCAGGCCCTACCCGGACG/3SpC3/-3′	60	5′-AGGGCCTGCGGCGTCCTAG/3IAbRQSp/-3′

Oligonucleotides and Repository Samples

All the primers, probes, and synthetic DNA templates were purchased from Integrated DNA Technologies (Coralville, IA). Primers were purchased as standard desalted DNA oligonucleotides and synthetic templates as desalted double-stranded fragments (gBlocks). Solutions of DNA oligonucleotides were stored at 4°C. In thalassemia color-mixing validation, human cell-line genomic DNA (gDNA) sample NA18562 (Coriell Biorepository for Medical Research, Camden, NJ) was stored at −20°C. The gDNA samples were mixed with synthetic DNA templates at a 1:1 ratio to generate samples containing 50% variant sequence. Dilution of gDNA samples and synthetic DNA templates were made in 1 × Tris-EDTA (TE) buffer with 0.1% Tween 20 from Sigma Aldrich (catalog number P1379-100 ML; Burlington, MA). In tuberculosis (TB) color-mixing validation, with the assumption that clinically extracted genomic DNA of M. tuberculosis from sputum specimen is low in amount and contains some human cells or human genomic DNA contamination, the WT sample was the mixture of 2 ng/μL human cell-line gDNA 18572 (Coriell Biorepository for Medical Research) and 1000 copies/μL H37Rv wild-type M. tuberculosis genomic DNA (ATCC, Manassas, VA); each 50% synthetic variant sample was prepared to 2 ng/μL human cell-line gDNA NA18572 (Coriell Biorepository for Medical Research), 500 copies/μL wild-type M. tuberculosis genomic DNA (catalog number 25618D-2; ATCC), and 500 copies/μL synthetic template. All templates were prepared in 1 × TE buffer with 0.1% Tween 20 from Sigma Aldrich.

Results

Identification of Pathogenic Variants in Thalassemia

Thalassemia is a potentially lethal autosomal recessive inherited blood disorder characterized by abnormal synthesis of hemoglobin. There are two types, α-thalassemia and β-thalassemia; α-thalassemia is involved with genetic pathogenic variants in the HBA1 and HBA2 genes that encode hemoglobin subunit α, whereas β-thalassemia is involved with variants in the HBB gene that encodes hemoglobin subunit β.¹³ Individuals with either α-thalassemia or β-thalassemia have shortened life expectancy. Therefore, identification of biallelic pathogenic variants in HBA1, HBA2, and HBB may be informative in genotype screening for expectant parents.

This color-mixing strategy was implemented to detect 21 common pathogenic variants in the HBA2 and HBB genes.¹⁴ The detailed design of different variants with their corresponding reference single-nucleotide polymorphism numbers is shown in Supplemental Table S5. A six-tube assay was designed, with tubes 1 to 3 containing a total of 32 variant probes to identify 21 pathogenic variants and tubes 4 to 6 containing a total of 11 WT probes to further distinguish the samples as homozygous (100%) or heterozygous (50%). Four fluorescence signals were collected from each tube, and a combined 24 fluorescent signals were collected from six tubes.

In each tube, each fluorescence channel has an independent threshold to make a call of an on or off state. To determine the threshold, synthetic DNA with variant strands was used as positive controls and wild-type genomic DNA as negative controls (Supplemental Table S6). Positive controls can be detected by specific variant probes, whereas negative controls do not include any variants that variant probes can detect. Using a receiver operating characteristic curve fit, the threshold was determined as the fluorescence value that could best distinguish the final signal (raw – background) of the positive controls from that of negative controls (Supplemental Figures S4 and S5). Final signals above the threshold are identified as on (1), and final signals below threshold are identified as off (0). A sequence of 12 numbers, all 0 or 1, indicates the state combinations used to represent the 21 pathogenic variants.

As an example, the identification of variant 16 is shown in Figure 3A. In tube 1, the final HEX signal of the variant 16 sample was above the threshold, whereas the final ROX, CY5, and FAM signals of the variant 16 samples were below the threshold. Therefore, a code of 0001 was generated from tube 1. Likewise, codes of 1000 and 0000 were generated in tubes 2 and 3, respectively. (Only 3 channels were implemented in tube 3 for this design; therefore, only CY5, FAM, and HEX signals are shown.) A combination of 0001, 1000, and 0000 from three tubes denotes the identification of variant 16.

Figure 3.

Experimental results of color-mixing detection of pathogenic variants in thalassemia. A: An example of generating color code for positive control variant (Var) 16 in tubes 1 to 3. Triplicate final signals of wild-type (WT) sample are displayed as red lines, and the final signals of the variant sample are displayed as green lines. Thresholds are shown as blue dashed lines. In tube 1, only the HEX signal was in on state (1), whereas in ROX, CY5, and FAM, because there were no probes targeting variant 16, the signals of these three channels were off (0); a color code of 4, representing 0001, was thus generated in tube 1. Likewise, in tube 2, the ROX signal was on, whereas the other channels were off. A color code of 1, representing 1000, was generated in tube 2. In tube 3, because there was no probe targeting variant 16, the color code of this tube is 0. As no probe with ROX fluorophore was included in tube 3, ROX channel's plot is not shown herein. The color codes generated from the three tubes identify variant 16. B: Experimental results of color-mixing design in identifying 21 variants in three variant tubes. Different color codes were used to indicate 21 different variants. Probes were used qualitatively; therefore, the color codes for synthetic 100% and 50% variant-positive controls are the same. Experimental data resulted in a specificity of 99.01% and a sensitivity of 94.12% when compared with the expected results; dark brown rectangular boxed areas marked the experimental results that were not in accordance with expected results. C: Both 50% and 100% variant controls had the same on state and color codes in variant tubes. However, 50% variant controls had the on state in WT tube, and 100% variant control had the off state in WT tube. D: The 0%, 50%, and 100% controls of variant 16 can be separated by WT probes in tube 6. Both 0% and 50% controls were in on state, whereas the 100% control was in the off state. AU, arbitrary unit.

The experimental results of the design for 32 variant probes implemented in tubes 1 to 3 to identify the 21 pathogenic variants in HBA2 gene and HBB gene are shown in Figure 3B. Each variant can be denoted by a unique color code; the 50% and 100% synthetic variant controls have the same color code as a result of this qualitative end-point detection approach, and the WT sample is denoted by 0000, 0000, and 000 in tubes 1 to 3, respectively.

When compared with the expected results (Supplemental Figure S6), the experimental results have a sensitivity of 94.12% and a specificity of 99.01%. The few signals that were misread in the FAM and HEX channels in tube 1 indicated the need for further optimization in probe sequence design or probe concentration to achieve the optimal distinction between variant and WT samples.

The 50% and 100% control samples containing the same variant were represented by the same color codes in tubes 1 to 3; tubes 4 to 6 with wild-type probes were further designed to separate the 50% and 100% control samples (Supplemental Table S7). For example, 50% and 100% variant 16 were already detected by tubes 1 to 3. And in tube 6, a wild-type probe attached with the CY5 fluorophore was used to separate the 50% and 100% variant-16 control samples (Figure 3C). WT (0%) and 50% variant samples were in the on state, whereas only the 100% variant sample was in the off state, in accordance with the design (Supplemental Figure S7).

Together from the six tubes, 21 pathogenic variants related with thalassemia (tubes 1 to 3) could be separated, and individual samples can also be identified to be heterozygous (50%) or homozygous (100%; tubes 4 to 6). In tubes 4 to 6, experimental results had a specificity of 99.46% and a sensitivity of 95.24% comparing with the expected results (Supplemental Figure S8).

Thalassemia Commercial Sample Results

The thalassemia color-mixing design was further validated on four pre-identified commercial samples. Sample identifier, variant identifier in the design, sample name, and the reference single-nucleotide polymorphism number are shown in Figure 4A. Herein, only the results from tubes 1 to 3 are shown because these pre-identified samples were all heterozygous; therefore, the wild-type probes all have a high fluorescence signal in tubes 4 to 6. The expected fluorescence signals of the four samples in tubes 1 to 3 are also shown herein. The four samples were experimentally detected and validated, and are consistent with the expectation (Figure 4B and Supplemental Figure S9).

Figure 4.

Validation of the thalassemia assay on commercial samples. A: This thalassemia assay was tested on four commercial samples, and samples show the sample name, targeted single-nucleotide polymorphism (SNP) identifier (ID), and reference SNP (rs) number for the design. On the right, the expected fluorescence for the color-mixing design is shown. Note that the variants (Vars) for these four samples were pre-identified, and all the samples were heterozygous samples. B: The call of variants in four samples are shown; all samples were successfully identified.

Drug Resistance in Tuberculosis

Furthermore, the color-mixing strategy was implemented in the detection of drug resistance in TB. Tuberculosis, usually caused by MTB, is one of the 10 deadliest contagious diseases worldwide. According to the World Health Organization Global Tuberculosis Report 2019, an estimated 10 million individuals became ill with TB in 2018.¹⁵ Successful pharmacologic treatment began in the 1940s, when streptomycin use was found effective. Soon after that, it became clear that single-drug therapy led to the evolution of drug-resistant MTB strains, and therefore high failure rates in TB treatment.¹⁶

Multi–drug-resistant TB is resistant to the two most potent anti-TB drugs, rifampicin (RIF) and isoniazid (INH). Approximately 5% of newly diagnosed TB cases are rifampicin-resistant TB, and 78% of the 5% rifampicin-resistant TB cases are multi–drug-resistant TB. Thus, rifampicin-resistant TB and multi–drug-resistant TB are of significant concern at the global level, indicating the urgency for development of detection methods for drug-resistance mutations.¹⁵ Prompt diagnosis of multi–drug-resistant TB would tell clinicians to switch to a second-line drug regimen, such as fluoroquinolones.¹⁵

Several mutations in specific genes across the bacterial genome are known to lead to the drug resistance of MTB (Table 717, 18, 19), so it would help to identify drug-resistance–related mutations among the hot spot genetic regions in the MTB genome.

Table 7.

Drug Resistance in Mycobacterium tuberculosis

Mutated codon	Specific mutation	Genetic feature	MIC, μg/mL
None	None	RIF: susceptible	<0.25
D516Y	GAC(Asp) -> TAC(Tyr)	RIF: resistant	64∗
D516V	GAC(Asp) -> GTC(Val)	RIF: resistant	64∗
H526Y	CAC(His) -> TAC(Tyr)	RIF: resistant	128∗
H526D	CAC(His) -> GAC(Asp)	RIF: resistant	128
S531L	TCG(Ser) -> TTG(Leu)	RIF: resistant	128∗
L511M, D516A	CTG(Leu) -> CCG(Pro), GAC(Asp) -> GGC(Gly)	RIF: resistant	64

Drug resistance (ie, the lack of drug susceptibility of Mycobacterium tuberculosis bacteria to the antituberculosis drug) is shown in the difference of MICs between each clinical strain, which is a quantitative parameter describing the concentration of antibiotic to which a bacterium is sensitive. Herein, taking the RIF MIC as an example, the RIF MIC of the susceptible strain (non–RIF-resistant strain) is different from the RIF MICs of mutant strain.⁹

MIC, minimal inhibitory concentration (the lowest concentration of a chemical, usually a drug, that prevents visible growth of a bacterium or bacteria); RIF, rifampicin.

^∗When there were multiple strains with the same mutation under MIC test, the results shown are the median MICs.

Strategies have been applied to detect MTB drug-resistance mutations before. Culture-based phenotypic drug susceptibility test methods are currently the gold standard for drug-resistance detection, but these methods are time-consuming (usually from 2 to 6 weeks) and require sophisticated laboratory infrastructure, qualified staff, and strict quality control.²⁰ Many other methods have been reported before (eg, multiplex PCR,^21,22 melting temperature analysis in the real-time PCR,²³ microarray,²⁴ and real-time PCR using molecular beacon probes,^25,26 the NGS methods,^9,10,27, 28, 29 and the Xpert MTB/RIF).³⁰ However, it usually takes 2 to 6 weeks for drug susceptibility test analysis, up to 2 days from NGS library preparation to interpretation report. Methods like microarray and Xpert MTB/RIF require special sets of instruments and extra training for laboratory researchers and technicians. Certain technologies based on the fluorescent readout, like multiplex PCR, melting temperature analysis, and molecular beacon, are restricted in the total number of targets because of the limited number of available fluorescence channels. Herein, the color-mixing strategy, much less limited in the number of fluorescence channels, was applied to detect drug-resistance mutations in TB, achieving rapid and simultaneous detection of drug-resistance mutations in multiple genes and regions within 3.5 hours.

A three-tube assay was designed to cover the RIF-resistance gene rpoB and the INH-resistance genes katG and inhA (Table 8). RpoB encodes the β subunit of RNA polymerase. An 81-bp region of the rpoB gene has been reported to be the rifampin resistance-determining region (RRDR).³¹ Although around 90% to 96%^31,32 of RIF resistance in MTB is associated with mutations inside the RRDR region, there is another region in rpoB, called the cluster II region,^32,33 that contains the remaining 4% to 10%^31,32 of RIF-resistance mutations in MTB.

Table 8.

Summary of Assays

Tube no.	Target drug	Drug-resistant gene	Covered mutation	Design strategy
1	RIF	rpoB	Plus strand rpoB 507aa to 533aa rpoB 561aa to 576aa	Wild-type probe
2			Minus strand rpoB 507aa to 533aa rpoB 561aa to 576aa
3	INH	inhA, KatG	KatG S315T (AGC -> ACA) KatG S315T (AGC -> ACC) KatG S315I (AGC -> ATC) KatG S315R (AGC -> CGC) inhA I194T (T -> C)	Variant probe

Using wild-type probes, tubes 1 and 2 target the rifampin resistance-determining region and cluster II region of the rpoB gene, whose mutations cause the rifampin resistance of Mycobacterium tuberculosis bacteria. Tube 3 targets multiple mutations in the INH resistance-related katG and inhA genes using variant probes.

Aa, amino acid; INH, isoniazid; RIF, rifampin.

There are difficulties in detecting drug-resistance mutations in the rpoB gene. Previous methods predominantly covered the RRDR region, resulting in lower sensitivity and specificity.31, 32 The 81-bp RRDR and 48-bp cluster II–related regions are relatively long and high in GC content, but also rich in drug-resistance mutations. Most previous methods were limited to detecting only a few hot spot mutations, leading to incomplete mutation profiling and less accurate diagnostic information. To solve these difficulties, an assay was designed to cover both the RRDR and cluster II regions within the rpoB gene in both positive and negative strands (tubes 1 and 2) (Table 8) using wild-type probes to achieve comprehensive detection of rpoB drug-resistance mutations.

INH resistance commonly occurs when mutations occur in the katG gene or the regulatory regions of inhA gene. katG functions to encode catalase-peroxidase, which is an enzyme that can activate INH. Mutations that occur in the katG gene, especially at codon 315, have a high correlation with INH resistance. Another target of active INH is NAD-dependent enoyl-acyl carrier protein reductase, which is encoded by the inhA gene.³⁴ A panel was designed (shown as tube 3) (Table 8) to simultaneously detect multiple hot spot variants within one tube in this work. The targets of tube 3 are mainly single-nucleotide variants and relatively shorter than those in tubes 1 and 2; thus, targeting single variants in only positive or negative strands using variant probes is sufficient for sensitivity and specificity.

Each of tubes 1 and 2 contains two WT probes to identify any drug-resistance mutation occurring within the target region. Only signals from the Cy5 and HEX fluorescence channels were collected from each tube. Examples of the identification of some variants in tube 2 are shown in Figure 5A. Each experiment was performed in triplicate. In Figure 5A, the final signals of variants other than the D516A variant are below the wild-type samples as well as the threshold. That means that all variants except the D516A variant could be confidently detected in tube 2. The threshold is determined by the receiver operating characteristic curve, as described before. The experimental results of the design for tube 1 and tube 2 are shown in Figure 5B. In tubes 1 and 2, if the final signal of one unknown sample is above the threshold, it indicates that the sample does not have any variant in the targeted region; otherwise, the sample should contain drug-resistance mutations in the targeted region. Ideally, every variant regardless of its change pattern or localized strand could be detected in each tube. S531L could not be distinguished with wild type in tube 1, and D516A is indistinguishable from wild type in tube 2, marked as brown frames in Figure 5B. More details are discussed in the Supplemental Table S8 and Supplemental Figures S10 and S11; the failure of D516A is probably from the minimum difference of standard free energy between the binding of the probe to the variant and to the WT, as shown in Supplemental Table S8 and Supplemental Figure S10. The strong hairpin structure formed around S531L probably leads to the indistinguishable region between S531 and the WT in tube 1 (Supplemental Figure S11). The failures in detecting S531L in tube 1 and D516A in tube 2 imply the necessity of targeting both the positive and negative strands. In these cases, combining the results of tubes 1 and 2 would help provide a more accurate and comprehensive result because S531L is easily separated with WT in tube 2 and the final signal of D516A is much lower than that of WT in tube 1.

Figure 5.

Color-mixing strategy in detecting tuberculosis drug-resistance mutations in Mycobacterium tuberculosis. A: In each reaction, 20 ng of human DNA + 1000 Mycobacterium tuberculosis bacteria DNA copies (either 100% genomic DNA or 50% variant template and 50% genomic DNA) were used as input. A final readout was obtained by comparing the final fluorescence with threshold. All wild-type sample final signals are displayed as red lines, all variant sample final signals are displayed as green lines, and thresholds based on the receiver operating characteristic curve for each channel are displayed as dashed lines. For example, variant D516A is called wild type in tube 2 because its final fluorescence is above threshold, and variant P564L is called variant in tube 2 because its final fluorescence is below threshold; summaries of all the tubes are shown in B and C. B: Comparison between expected results and experimental results in wild-type tubes. In tubes 1 and 2, the H37Rv wild-type genomic DNA and 10 synthetic variant templates were tested and summarized. C: Comparison between expected and experimental results in tube 3. In tube 3, the H37Rv wild-type genomic DNA and five synthetic variant templates were tested and summarized. Notes: S315T¹: S315T(AGC -> ACA); S315T²: S315T(AGC -> ACC). AU, arbitrary unit.

Tube 3 contains 10 variant probes to identify five drug-resistance mutations, and four fluorescence signals would be collected. The experimental and expected results for tube 3 are shown in Figure 5C and Supplemental Figure S12. Each variant is denoted by a unique color code, as shown in Supplemental Figure S12. All the covered mutations in tube 3 were tested: one synthetic variant (I194T) in inhA and four synthetic variants [S315T(AGC -> ACA), S315T(AGC -> ACC), S315I, and S315R] in katG (Supplemental Table S9). Each variant could be distinguished from WT and other variants with 100% sensitivity, as shown in Figure 5C.

Given the above data, we conclude that the use of color-mixing strategy successfully detects drug-resistance mutations of MTB simultaneously.

Discussion

In this study, a color-mixing strategy that utilizes end-point fluorescence detection to achieve highly multiplex probing of genomic variants simultaneously was presented. The color-mixing strategy involves two steps: first, enrichment of regions of interest by an asymmetric PCR; and second, detection of genetic variants by a set of rationally designed multiplex toehold probes. A set of toehold probes were either all variant probes or all wild-type probes. The color-mixing strategy was implemented into two applications, the detection of thalassemia-related pathogenic variants and the detection of drug-resistance mutations in tuberculosis. The results demonstrate that a simple color-mixing strategy can be adapted to different applications for different purposes and that it can be used to identify variant and/or wild-type sequences.

Because of the severity of thalassemia and tuberculosis, it is much desired to have affordable, easy-to-use, and comprehensive detection methods with readily accessible instruments for hot spot pathogenic variants or drug-resistance mutations. Compared with traditional gold standard drug susceptibility test analysis for drug-resistance mutations, which usually takes weeks to receive the results, the color-mixing strategy only takes a few hours for the entire workflow. Meanwhile, methods like microarray and NGS usually have a more complex workflow with multiple hands-on steps requiring more operator training and add to the difficulty of applying these methods to low- and middle-income countries with limited resources. The total number of available fluorescence channels has always been a bottleneck for mutation detection. Therefore, the plex scalability of methods based on multiplex qPCR, melting temperature analysis, or molecular beacon is confined. Unlike traditional detection methods, which assign fluorescence channels by targets, the color-mixing strategy offers another solution to exponentially scale up the total number of targets (mutations) by combining fluorescence channels. This way, it is promising to have a detection method with flexible scalability, low protocol requirement, and fast turnaround time using the color-mixing strategy.

A six-tube assay that included three variant tubes was designed to identify 21 pathogenic variants in HBA2 and HBB genes, and another three wild-type tubes that can distinguish the synthetic 50% and 100% variant samples that were designed to mimic the clinical heterozygous and homozygous samples. Note that heterozygous samples are much more clinically common than homozygous samples because the homozygous variant may be lethal in the fetus. This design can be used for prenatal screening for pathogenic variants and can, therefore, provide information for expectant parents.

A three-tube assay was also designed to detect drug-resistance mutations in MTB. This designed assay includes two wild-type tubes that can identify any mutation in both an 81-bp region and a 48-bp region that are related to RIF resistance. It includes another variant tube that can identify five mutations previously reported to be highly correlated with INH resistance. Using the designed multiplex wild-type probes, the identification of 10 synthetic variants within two target regions in the rpoB gene was experimentally validated. Each template used in the experiments was designed to mimic a clinical sputum sample extracted from the lung, with a mixture of bacteria and human cells. The tube 3 (variant tube) was validated by probing for all the variants with 100% sensitivity. Moreover, the compatibility of color-mixing strategy with limited DNA input was demonstrated. Only 1000 genomic copies of MTB were input into each reaction for the assay, which could be achievable by most of the clinical specimen types. To ensure stable reproducibility and good sensitivity, it requires the DNA is not over-fragmented to ensure successful PCR amplification. Other than that, there is no extra requirement for the input sample compared with other detection methods. Some loci or some variants may be difficult to be detected because either the sequences are high in GC content and may form secondary structures or the energy penalty brought by the variant is barely distinguishable by probes. This design can guide clinicians to decide specific treatment plans for each patient promptly.

In both applications, all probes were designed to work best based on the simulation; however, there are parameters that we may not be able to accurately predict (ie, secondary structure of the targeted sequences, interplex toehold probe interactions, energy penalty between different fluorophore and quencher binding, and chemistry purity of each oligo). Therefore, the initial design may not be optimal. On the basis of our previous experimental results, we would like to propose strategies to improve the sensitivity. The first strategy is to modify the probe sequences. It was experimentally found that each toehold probe can be optimized to increase sensitivity by modifying (including adding or removing) the bases at 5′ end of the P strand or the 3′ end of the C strand, resulting in the ΔG change of the toehold region. The second strategy is to modify the probe concentration. Each probe's concentration can be modified individually by changing the ratio of the P strand/the C strand. Increasing the probe's concentration can increase the kinetics of the probe binding to the target and lead to better performance.

Color-mixing strategy can be used to detect discrete variants across multiple regions of interest using variant toehold probes, and to detect dense and continuous hot spot mutations using wild-type toehold probes. For example, in tumor suppressor genes like TP53 and BRCA1/2, continuous hot spot mutations happened within a long genetic region; in the food and agriculture industry, this can be applied to identify dominant species of microorganisms. Furthermore, our method is designed to diagnose genetic disorders that are predominantly caused by a single variant in the individual. Multiple variants within the same reaction may lead to the combination of color codes, and the color code was designed to avoid the misinterpretation of co-existent variants. Therefore, accurate variant detection requires preliminary knowledge of variant information and neighboring sequence, including the population-specific single-nucleotide polymorphism^35,36 or nonpathogenic single-nucleotide polymorphism.^37,38

The color-mixing strategy is compatible with any instrument with fluorescence readout function and enables the exponential scaling up of the total number of detectable variants; therefore, it is not limited to the number of fluorescence channels within an instrument. It is believed that this method is a general method for the detection of multiple variants in other diseases beyond thalassemia and tuberculosis. This method can be used for detection in clinical samples in the future. With limited access to thalassemia and tuberculosis clinical samples, we could not verify the method in actual clinical samples but hope to find collaborators for such verification in the near future. The capability of color mixing to detect variants with relatively high abundance has been revealed in this work. In the future, we want to explore further the feasibility of the color-mixing strategy in detecting variants with low variant allele frequency.

Author Contributions

N.G.X., K.Z., P.S., and D.Y.Z. conceived the project; N.G.X., K.Z., and P.S. performed the experiments and analyzed the data; R.L. and J.L. performed the experiments for thalassemia commercial samples; and N.G.X., K.Z., P.S., and D.Y.Z. wrote the article with input from all authors.

Supplemental Data

Supplemental Figure S1

Example of background signal (BS). There is variation of background fluorescence (Fluor) signal in each well of the quantitative PCR instrument. Before asymmetric PCR, the fluorescence signal would be collected, as BS. a.u., arbitrary unit.

Supplemental Figure S2

Example of background noise subtraction. Left panel: After asymmetric PCR, the fluorescence (Fluor) signal would be collected one more time, as raw signal (RS). Right panel: By subtracting background signal from RS, background correction could be achieved to generate final signal. a.u., arbitrary unit; Min, minutes.

Supplemental Figure S3

Simulation of discrimination factor Q of 32 toehold probes in tubes 1 to 3. Discrimination factor was defined as $Q=\frac{\chi \left(variant\right)}{\chi \left(WT\right)}$; the Q in Figure 2B is 12.3. The summary of simulated discrimination factor Q of 32 toehold probes in tubes 1 to 3 is shown herein. In general, small insertions and deletions have a higher Q than the single-nucleotide variant because of stronger $\Delta \Delta G_{mismatch}^{°}$. The empirical energy penalty ΔG_(FQ) of fluorophore and quencher binding is different among different fluorophores and, to be more specific, for the FAM, the ΔG_(FQ) is from −3.5 to −4 kcal/mol; for the HEX, it is −4 kcal/mol; for the CY5, it is −4 kcal/mol; for the ROX, it is from −3.4 to −4 kcal/mol.

Supplemental Figure S4

Threshold selection in tube 1 ROX channel. As shown in Table S5, we prepared 0%, 50%, and 100% variant controls. Each control can be categorized to be positive and negative for each channel in one tube. A: For example, in tube 1, only an ROX probe was designed to target the variant 14, and control 15 contains 50% variant 14 and control 36 contains 100% variant 14. Therefore, controls 15 and 36 were categorized as positive in ROX channel in tube 1, whereas the rest of the 41 controls were categorized as negative. Positive controls were colored in purple, and negative controls were colored in blue in Supplemental Figure S3A. The x axis is the 43 controls, and the y axis is the background fluorescence-corrected final signal. B: An example receiver operating characteristic (ROC) curve fit is shown to find the optimized threshold in ROX channel in tube 1 in thalassemia color-mixing assay. The threshold was selected from between the minimum and maximum of the final signal and was determined to maximize the area under the curve. When the threshold is set at 3 ∗ 10⁵, all the positive controls can be successfully distinguished from negative controls. ID, identifier.

Supplemental Figure S5

Threshold selection in tube 1 to tube 3 in four fluorescence channels.

Supplemental Figure S6

Expected results of thalassemia pathogenic genetic variant (Var) detection in tube 1 to tube 3.

Supplemental Figure S7

Experimental results of 0%, 50%, and 100% variant 4 in tube 5 Cy5 channel. Subtracted CY5 fluorescence (Fluor) signals of 0%, 50%, and 100% variant controls were plotted in red, blue, and green, respectively. Threshold determined by receiver operating characteristic curve fit was plotted in blue dashed line. According to our design, the wild-type (WT) probe will bind to the WT template and fluoresce, because 0% and 50% variant controls both contained the WT template; therefore, after the asymmetric PCR, 0% and 50% variant controls had above-threshold fluorescence signals, whereas the 100% variant control did not. The 50% variant control had a lower signal than 0% variant control; this is reasonable because the 50% variant control contained fewer WT molecules than the 0% variant control. Nevertheless, a proper threshold was determined to just separate the samples containing WT sequence than those not because our assay was designed to be qualitative but not quantitative. Note that we also observed in other channels that the 50% variant and 0% variant controls gave approximately the same level of fluorescence. A.u., arbitrary unit; min, minutes.

Supplemental Figure S8

Experimental results of thalassemia pathogenic genetic variant (Var) detection in tube 4 to tube 6. Red rectangular boxed areas marked the experimental results that are not in accordance with expected results. Data points specified as N/A are a result of unusable experimental results because of the length limitation of the synthetic gBlocks. All the gBlocks that were ordered had a length of 700 bp; therefore, gBlocks covering region of interest in HBA2 were unable cover regions in HBB genes, and vice versa. For example, variant 1 is on HBA2, and synthetic gBlocks that mimic variant 1 can only cover the 700-bp region in HBA2. However, in reality, a sample should contain both the regions in HBA2 and HBB genes. Also, a sample that contains variant 1 on HBA2 gene can be wild type (WT) in HBB gene. Therefore, a 100% variant 1 sample can be used as a WT for other positions in HBB gene. Nevertheless, our synthetic variant 1 sample did not contain the WT sequence in HBB gene and could not be targeted by the WT probe and fluoresce. Therefore, it could not be used as a reference in specific channels that probes are targeting HBB gene. We specified these data points in 100% variant controls to be N/A to minimize the error in our threshold selection and avoid false-negative results.

Supplemental Figure S9

Results of commercial samples. The assay was validated on four commercial thalassemia samples. The commercial samples are heterozygous samples. The results of tube 1 to tube 3 are shown herein. The x axis is the tube number, and the y axis is the signal/threshold ratio. We analyzed in this way so that if the ratio is >1, it is in on state; and if the ratio is <0, it is in off state. Blue dashed lines show the signal/threshold ratio equals to 1. All four samples can be accurately identified by our assay. Note that some high variation in FAM and HEX channels was observed, which was consistent with our positive and negative control results. This may indicate the following: First, probe sequences and concentration may need to be optimized to reach the best performance. Second, in general, CY5 channel and ROX channel work better than FAM and HEX channel; in the future, one can re-organize the color-mixing design to include more CY5 and ROX probes.

Supplemental Figure S10

Distribution of ΔΔG of each variant used in validation of tube 1 and 2 assay. The ΔΔG values summarized in Supplemental Table S8 were plotted, and the outlier (D516A) in tube 2 is shown by the arrow.

Supplemental Figure S11

Secondary structure folding of amplicon in tube 1 and tube 2 using Nupack. The minimum free energy (MFE) structure is the secondary structure that contributes a minimum of free energy. The red shaded area is what the wild-type (WT) probe with Cy5 modification targets, and the blue shaded area is what the WT probe with HEX modification targets. Because the Mycobacterium tuberculosis genome is GC rich, the possibility of forming a secondary structure is high. We found that the neighboring region of mutation S531L formed a hairpin with a 7-nucleotide stem and 86% GC content. Thus, we hypothesize that the failure of detecting S531L in tube 1 is due to the formation of a high-GC hairpin within the target region. The ΔG_Toehold of Cy5-WT probe in tube 2 is stronger than the ΔG_Toehold of Cy5-WT probe in tube 1; a stronger ΔG_Toehold makes the WT probe more likely to open the secondary structure. In this way, we think this is why tube 2 can successfully detect S531L, but tube 1 does not.

Supplemental Figure S12

Expected results of tuberculosis drug-resistance variant detection in tube 3. S315T¹: S315T(AGC → ACA); S315T²: S315T(AGC → ACC). WT, wild type.