Diagnostic accuracy of high-resolution melting curve analysis for discrimination of oncology-associated EGFR mutations: a systematic review and meta-analysis

Abstract

Objective

To investigate the diagnostic value of high-resolution melting (HRM) analysis for oncology-associated epidermal growth factor receptor (EGFR) gene mutations.

Methods

We systematically searched Embase, PubMed, and Web of Science for HRM and EGFR mutation detection studies published through September 2024. True and false positives and negatives were extracted to evaluate the diagnostic accuracy of HRM to detect EGFR mutations. The study was registered at INPLASY (INPLASY202490062).

Results

Twenty-six articles were obtained from 416 references. The overall diagnostic sensitivity and specificity were high at 0.95 [95% confidence interval (CI), 0.94–0.96] and 0.99 (95% CI, 0.99–0.99), respectively. Other indicators, including the positive likelihood ratio, negative likelihood ratio, and diagnostic odds ratio, were 144.91 (95% CI: 69.07–304.04), 0.08 (95% CI: 0.04–0.13), and 2405.21 (95% CI: 1231.87–4696.13), respectively. The Q value of the summary receiver operating characteristic curve was 0.979, and the area under the curve was 0.997.

Conclusion

As a pre-screening method, the high specificity, sensitivity, low cost, rapid turnaround, and simplicity of HRM make it a good alternative for clinical practice, but positive results must still be obtained for diagnostic confirmation. This study provides a transparent overview of relevant studies in design and conduct.

Introduction

The activation of signaling pathways by epidermal growth factor receptor (EGFR) plays an important role in the development of tumor-associated diseases. The EGFR gene, which has tyrosine kinase activity, is a member of the human epidermal growth factor receptor (HER) family composed of HER1 (erbB1, EGFR), HER2 (erbB2, NEU), HER3 (erbB3), and HER4 (erbB4). Over-expression of EGFR is critical for lung, breast, and gastric cancer and squamous cell carcinoma of the head and neck.^1–3 Activation of EGFR launches a series of cellular signaling pathways that promote cancer proliferation, invasion, and metastasis and protects carcinoma cells from apoptosis via an anti-apoptosis pathway.^4,5 Tyrosine kinase inhibitors (TKIs), such as gefitinib and erlotinib, can inhibit this pathway and consequently offer efficacy for patients with an EGFR mutation.^6–8 Therefore, EGFR gene mutational status is the most sensitive target for TKI therapy selection.

EGFR mutations are located on exons 18, 19, 20, and 21 of EGFR, and most include an in-frame deletion of codons 746 to 750 in exon 19 and a missense mutation at codon 858 in exon 21. An activating mutation in EGFR can be found in high incidence in non-smokers, women, those with adenocarcinoma, and individuals of Asian ethnic background.^9,10 Currently, several genotypic methods to screen gene mutations and expand the knowledge of drug-gene relationships have been developed, such as DNA sequencing, ¹¹ single-strand conformation polymorphism analysis, ¹² denaturing high-performance liquid chromatography,^13,14 allele-specific polymerase chain reaction (PCR), ¹⁵ array analysis, ¹⁶ pyrosequencing, ¹⁷ and high-resolution melting (HRM) curve analysis.^18,19 Some of these methodologies require sample separation on a gel or matrix; others require expensive fluorescently labeled probes or special instruments. However, HRM analysis is performed in a closed-tube system that protects the amplified DNA from cross-contamination, which is the main advantage of HRM analysis, and it has been proven as a rapid, cost-effective method that uses few or no probes. ²⁰  As an alternative molecular testing platform for genotyping of polymorphisms, HRM analysis has been applied to various diseases, such as oncological, infectious, and inherited diseases.^21–23

HRM curve analysis is a relatively mature method based on the melting profiles of double-stranded PCR products that is widely used in diagnostic laboratories for identification in disease-associated genotyping, sequence matching, methylation studies, single nucleotide polymorphism analysis, and mutation scanning. ²⁴ It reveals a different melting curve based on DNA duplex melting temperature changes. ²⁵  As the temperature rises, intercalating dye is released, and the fluorescence intensity decreases; then, mutations are distinguished by changes in melt curve shapes compared with a reference profile. ²⁶ Compared with DNA sequencing, HRM analysis requires minimal investment, thus the technology is broadly available, but it also has high sensitivity, rapid turnaround, low cost, and nondestructive and closed tube operations. ²⁷ In selecting a molecular testing platform for genotyping polymorphisms, an important consideration is the rapid delivery of genetic information to meet the need for increasing clinical treatment.

Since its first application for genotyping in 2003, HRM analysis has been extensively used to detect mutations such as KRAS, BRAF, EGFR, and TP53.^{7,18,19,28–30} A recent study suggested that HRM analysis is a promising method to detect EGFR mutations. ³¹ However, its diagnostic accuracy for EGFR identification has not been systematically evaluated. It is essential to investigate the EGFR mutation signature in tumor-associated diseases, which leads to more suitable decision making for treatment by physicians. Therefore, we performed a meta-analysis to evaluate the accuracy of HRM analysis for EGFR mutation identification.

Materials and methods

We performed this meta-analysis according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. ³² Our research was registered at INPLASY, registration number INPLASY202490062 (DOI: 10.37766/inplasy2024.9.0062).

Literature search strategy

Excerpta Medica Database (Embase), Medline (using PubMed as the search engine), and the Web of Science were searched to identify relevant publications in English until 12 September 2024, using the search strategy ‘epidermal growth factor receptor’ or ‘EGFR’ or ‘EGF-R’ or ‘EGF-receptor’ or ‘EGF receptor’ or ‘receptor, epidermal growth factor’ or ‘transforming growth factor alpha receptor’ or ‘ERBB-1 proto-oncogene protein’ or ‘receptor, transforming-growth factor alpha’ or ‘receptor, transforming growth factor alpha’ or ‘C-ERBB-1 protein’ or ‘receptors, epidermal growth factor’ or ‘receptor, EGF’ or ‘urogastrone receptor’ or ‘TGF-alpha receptor’ or ‘epidermal growth factor receptor kinase’ or ‘epidermal growth factor receptor protein–tyrosine kinase’ or ‘epidermal growth factor receptor protein tyrosine kinase’ AND ‘HRMA’ or ‘HRM’ or ‘HRMCA’ or ‘HRMC’ or ‘high resolution melting analysis’ or ‘high resolution melting’ or ‘high resolution melting curve analysis’ or ‘high resolution melting curve’. We also carried out manual searches for additional eligible studies.

Inclusion and exclusion criteria

Studies were included if HRM was used to study EGFR mutations in humans, DNA sequencing (including direct DNA sequencing and pyrosequencing) was used as a reference standard, and true and false positives and negatives (TP, TN, FP, FN) could be calculated from the provided information. The exclusion criteria included studies using only HRM, studies for which the reference standard was not direct DNA sequencing or pyrosequencing, studies for which only positive HRM samples were sequenced, studies examining methylation or epigenetic mechanisms rather than genetic mutations, or studies that were duplicated, a review, a conference abstract, a letter, or a comment.

Data extraction and quality assessment

We extracted the following data: author’s name, publication year, country of origin, specimen sources, mutation prevalence, instruments, disease types, sample number, amplicon length, dye types, and disease-associated mutations. Outcome parameters such as TN, FN, TP, and FP values were calculated based on PCR amplicons, not based on tissue or blood samples. Two authors performed data collection, and disagreements were resolved by discussion or consensus with a third author. We assessed the quality of each study based on the Quality Assessment for Studies of Diagnostic Accuracy (QUADAS-2), ³³ which includes four primary domains to evaluate bias and applicability of included studies by assessing patient selection methods, an index test, reference standards, and patient flow through studies. ³⁴

Statistical analysis

We determined the accuracy of each study with standard methods using Meta-Disc (version 1.4, http://www.hrc.es/investigacion/metadisc_en.htm) and STATA 12.1 software (Stata Corp., College Station, TX, USA). The sensitivity, specificity, positive and negative likelihood ratios (PLRs, NLRs), and diagnostic odds ratio (DOR) from studies and corresponding 95% confidence intervals (CIs) were computed using fixed or random effects models depending on the presence of significant heterogeneity. Degrees of heterogeneity were evaluated with a chi-square test of heterogeneity (Cochran’s Q statistical test) and an inconsistency index (I-square). Alternatively, to quantify the effect of heterogeneity, significant heterogeneity was defined as a Q test with p < 0.10 or I² > 50%. The threshold effect was performed using summary receiver operating characteristic (SROC) curves for each study to ascertain the presence of a “shoulder-arm” pattern, which would suggest a threshold effect. ³⁵ The Spearman correlation coefficient between the logit of sensitivity and logit of 1−specificity for each study was also calculated to assess any threshold effect. A positive correlation (p < 0.05) would suggest a threshold effect. Publication bias was determined using funnel plot analysis with STATA 12.1 software.

Meta-regression analysis and subgroup analysis

Meta-regression analysis was performed to explore heterogeneity sources using Meta-Disc (version 1.4) software. A multivariable regression model was applied, and a backward stepwise algorithm with covariates including disease type, specimen source, instruments, and dye type was used. Variables were retained in the regression model if p < 0.05. Subgroup analysis was performed if reasons for heterogeneity could be found.

Results

Literature search outcome

The results of the literature search and the stage-wise exclusion process are illustrated in Figure 1. All 416 references were found by searching multiple sources and databases. One hundred ninety-seven records were excluded because of duplicates. After reviewing the titles and abstracts, 182 records were eliminated, and 37 articles were deemed potentially relevant for the next detailed screening. Eleven records were excluded for reasons shown in Figure 1. Finally, 26 articles were retrieved in this meta-analysis and divided into 34 subsets for statistical analysis according to specimen sources.

Figure 1.

Flow chart of the literature search and study selection.

Characteristics of the studies

We found 26 eligible studies,^{11,18,20,29,36–57} that reported evaluations of diagnostic accuracy of HRM analysis for human disease-associated mutations, as detailed in Table 1. All samples screened by HRM were followed up with direct sequencing. The flow and timing domain was labeled as “unclear risk”. The patient selection, index test, and reference standard domains were labeled as “low risk” both for risk of bias and applicability concerns.

Table 1.

Characteristics of the 26 studies included in this meta-analysis.

No.	Author	Year	Country	Diease	Number of amplicons	Prevalence	Specimen source	Instrument	Dye	Sequence analyzed	Amplicon size	TP*	FP	FN	TN
1	Willmore-Payne, C. 18	2006	America	Lung adenocarcino-ma	156	5%	FFPE	LightCycler	LCG1	Exons 18–21	186–248	6	0	1	149
2	Willmore-Payne, C. 36	2006	America	Squamous cell carcinoma	120	1.7%	FFPE	LightCycler	LCG1	Exons 18–21	186–259	2	0	0	118
3	Nomoto, K. 37	2006	Japan	NSCLC	74	28%	Cytologic slides	LightCycler	LCG1	Exon-21 L858R mutation/ -19 DEL	51–83	19	0	2	53
4	Takano, T. 38	2007	Japan	NSCLC	132	28%	Methanol-fixed	LightCycler	LCG1	Exon-21 L858R mutation/ -19 DEL	51–83	36	0	1	95
	Takano, T. 38	2007	Japan	NSCLC	56	28%	Cytologic slides	LightCycler	LCG1	Exon-21 L858R mutation/ -19 DEL	51–83	14	0	2	40
	Takano, T. 38	2007	Japan	NSCLC	126	29%	FFPE	LightCycler	LCG1	Exon-21 L858R mutation/ -19 DEL	51–83	34	0	3	89
5	Fukui, T. 20	2008	Japan	NSCLC	70	16%	Cytologic slides	HR-1	LCG1	Exon-21 L858R mutation/ -19 DEL	51–83	11	3	0	56
	Fukui, T. 20	2008	Japan	NSCLC	276	16%	FFPE	HR-1	LCG1	Exon-21 L858R mutation/ -19 DEL	51–83	43	2	3	228
6	Do, H. 39	2008	Australia	NSCLC	800	11%	FFPE	Rotor-Gene 6000	SYTO9	Exons 18–21	121–250	84	70	0	646
7	Fassina, A. 40	2009	Italy	NSCLC	154	1.3%	FFPE	LightCycler 480	LCG1	Exon-21 L858R mutation/ -19 DEL	142, 190	2	0	1	151
8	Jacot, W. 41	2011	France	TNBC	458	1%	Fresh frozen	Rotor-Gene 6000	SYTO9	Exon-21 L858R mutation/ -19 DEL	164, 182	3	0	0	455
9	Sriram, K.B. 42	2011	Austrilia	NSCLC	194	5%	Fresh frozen	Rotor-Gene 6000	SYTO9	Exon-21 L858R mutation/ -19 DEL	250, 210	9	0	0	185
10	Gonzalez-Bosquet, J. 43	2011	America	Breast cancer, endometrial cancer, ovarian cancer	216	2.8%	Fresh frozen	PTC 225 Thermal Cycler	LCG plus	Exon-23	213	6	0	0	210
	Gonzalez-Bosquet, J. 43	2011	America	Breast cancer, endometrial cancer, ovarian cancer	172	4%	FFPE	PTC 225 Thermal Cycler	LCG plus	Exon-23	213	7	19	0	146
11	Hu, C. 45	2012	China	NSCLC	504	15%	FFPE	LightCycler 480	LCG1	Exons 18–21	183–236	74	4	0	426
	Hu, C. 45	2012	China	NSCLC	188	12%	Fresh frozen	LightCycler 480	LCG1	Exons 18–21	183–236	23	2	0	163
12	Stigt, J.A. 46	2013	Netherlands	NSCLC	488	4%	FFPE	LightCycler 480II	LCG1	Exons 18–21	247	15	0	5	468
13	Jing, C.-W. 11	2013	China	NSCLC	480	9%	FFPE	LightCycler 480	LCG1	Exons 18–21	NA	45	3	0	432
	Jing, C.-W. 11	2013	China	NSCLC	480	9%	Plasma	LightCycler 480	LCG1	Exons 18–21	NA	29	2	16	433
14	Lin, J. 47	2014	China	NSCLC	144	8%	Supernatant	LightCycler 480	LCG plus	Exons 18–21	NA	12	6	0	126
	Lin, J. 47	2014	China	NSCLC	144	13%	Cell pellets	LightCycler 480	LCG plus	Exons 18–21	NA	13	0	5	126
	Lin, J. 47	2014	China	NSCLC	144	9%	FFPE	LightCycler 480	LCG plus	Exons 18–21	NA	13	0	0	131
15	Sun, H. 29	2014	China	NSCLC	450	10.2%	FFPE	LightCycler 480II	LCG plus	L858R, T790M, 19DEL	NA	46	21	0	383
16	Clay, T.D. 48	2014	Australia	Lung adenocarcinoma	400	3.5%	FFPE	Rotor-Gene 6000	SYTO 9	Exons 18–21	NA	14	0	0	386
17	Papadopoulou, E. 51	2015	Greece	NSCLC	5888	4%	FFPE	Rotor-Gene 6000	SYTO 9	Exons 18–21	99–156	233	3	0	5652
18	Oyaert, M. 50	2015	Belgium	NSCLC	1320	3%	FFPE	LightCycler 480	NA	Exons 18–21	NA	38	0	0	1282
19	Hinrichs, W.J. 49	2015	Netherlands	NSCLC	100	7%	FFPE	LightCycler 480II	NA	Exons 19–21	NA	6	1	1	92
20	Wayhelova, M. 53	2016	Czech Republic	Squamous cell carcinoma or adenocarcino-ma	96	0	Fresh frozen	Qubit 1.0	NA	Exons 18–21	NA	0	0	0	96
21	Clay, T.D. 52	2016	Australia	Pulmonary adenocarcinoma	712	7%	FFPE	NA	NA	Exons 18–21	NA	53	0	0	659
22	Martínez-Carretero, C. 54	2017	Spain	NSCLC	492	1.8%	FFPE	LightCycler 480	LCG1	Exons 18–21	183–238	7	0	2	483
23	Lu, H.-Y. 55	2018	China	SCLC	2376	0.5%	Blood	LightCycler 480	LCG1	Exons 18–21	NA	12	0	0	2364
24	Frankel, D. 56	2018	France	Lung adenocarcino-ma	28	21%	Cell pellets	LightCycler 480	LCG1	Exons 18–21	NA	5	0	1	22
25	Borràs, E. 44	2011	Spain	Colorectal or lung cancer	108	4.6%	FFPE	LightCycler 480	NA	Exons 19–21	128–250	5	0	0	103
26	Joy, R.A. 57	2020	India	NSCLC	232	24.4%	FFPE	LightCycler 480	NA	Exon 19, exon 21	NA	67	0	8	157

NSCLC: Non-small cell lung carcinoma; SCLC: Small cell lung cancer; TNBC: Triple-negative breast cancer; FFPE: Formalin-fixed and paraffin embedded; TP: True positive; FP: False positive; FN: False negative; TN: True negative; NA: Not available.

^*: Outcome parameters were calculated on the basis of “PCR amplicons”, not on the basis of tissue or blood samples.

Diagnostic accuracy

The diagnostic sensitivity and specificity were 0.95 [95% CI: 0.94–0.96] and 0.99 (95% CI: 0.99–0.99), respectively (Figure 2(a) and 2(b)). As shown in Figure 2(c) and 2(d), a high PLR of 144.91 (95% CI: 69.07–304.04) and a low NLR of 0.08 (95% CI: 0.04–0.13) indicated that HRM analysis had an excellent ability to identify the presence of EGFR mutations. Additionally, the DOR supported that HRM analysis was effective for EGFR mutation screening (Figure 3(a)). Chi-square and I2 tests for heterogeneity confirmed significant heterogeneity for the specificity and sensitivity of the pooled results. The SROC curve is shown in Figure 3(b). The SROC curve from our data showed a Q value of 0.979, while the area under the curve (AUC) was 0.997, further indicating a high overall accuracy of HRM analysis.

Figure 2.

Forest plot estimates of the sensitivity (a), specificity (b), positive likelihood ratio (LR) (c), and negative LR (d) for high-resolution melting with 95% confidence intervals (CIs). Each solid circle represents a subset.

Figure 3.

Forest plot estimates of the diagnostic odds ratio (a) with 95% confidence intervals (CIs) and summary receiver operating characteristic (SROC) curve (b) for high-resolution melting. Each solid circle represents a subset.

Threshold effect and publication bias

Although the Spearman correlation coefficient between the log (sensitivity) and the log (1−specificity) was 0.092, p = 0.605, the typical “shoulder-arm” pattern in the SROC curve suggested a threshold effect. A funnel plot was applied to determine the presence of publication bias, which demonstrated that publication bias was not significant (Figure 4).

Figure 4.

Funnel plot to assess potential publication bias. Each circle is a subset. Publication bias was not significant.

Meta-regression analysis and subgroup analysis

Multivariate meta-regression analysis with covariates, including instruments, dyes, specimen sources, and disease types, was performed to investigate the source of heterogeneity. The regression analysis results showed no statistical significance between studies, and subgroup analysis indicated no source of heterogeneity, but a threshold effect was present (data not shown).

Discussion

EGFR mutations predict TKI sensitivity, thus knowing this status could improve chemotherapy selection and patient outcome. Specific mutations in oncology-associated proteins respond to certain drugs and correlate with increased sensitivity, suggesting personalized therapeutics based on genotype. ²⁴ Takano’s group reported that advanced non-small cell lung cancer patients with EGFR mutations had poorer overall survival, but gefitinib improved the treatment response. ⁵⁸ Because it is part of the HER gene family, EGFR, known as HER-1, is relevant to breast cancer, and there has been ongoing interest in the EGFR gene-associated-oncology status. Therefore, a reliable method of screening EGFR mutations for therapeutic and prognostic triage may be needed to assess the accuracy in a range of tumor samples.

Thirty-four subsets from 26 published studies and 17,778 samples were assayed to evaluate the diagnostic accuracy of HRM analysis to identify EGFR mutations. Although the data show high overall diagnostic accuracy, there was substantial heterogeneity among eligible studies. Exploration of the reasons for heterogeneity rather than computation of a single summary measure has emerged as a main goal of meta-analyses. Thus, it is critical to investigate the sources of heterogeneity to determine whether they alter the appropriateness of statistical pooling of accuracy estimates. The threshold effect is a typical source of heterogeneity that arises when differences in specificities and sensitivities occur because different thresholds are used to define a positive (or negative) result. ⁵⁹ A “shoulder-arm” shape of the points in the ROC curve indicates a threshold effect in our study, which may partially account for the heterogeneity observed. We performed meta-regression and subgroup analyses to explore further heterogeneity sources, including disease type, specimen source, distribution, instruments, and dye type. However, the heterogeneity source was not found. We also tried excluding the studies of Jacot and Gonzalez^41,43 on breast, endometrial, and ovarian cancer to reduce heterogeneity. However, the heterogeneities found before and after exclusion were similar, which may be the main reason for heterogeneities not resulting from the two studies. The DOR and SROC curve are considered when there is substantial heterogeneity, ⁶⁰ as the DOR indicates accuracy when combining sensitivity and specificity data into a single ratio of a positive test result. ⁶¹ DOR values range from 0 to infinity, with higher values indicating better discriminatory test performances. ⁶¹ Our DOR was 2405.21 (95% CI: 1231.87–4696.13). As a global indicator for assessing diagnostic performance, the SROC AUC also indicated a high accuracy of HRM analysis, with a Q value of 0.979 and an AUC close to 1 (0.997). The DOR and AUC data indicated high overall accuracy of HRM analysis for EGFR mutation screening. However, the accuracy of HRM analysis could be affected by sample types, amplicon length, dyes, instruments, PCR specificity, GC content, and melting analysis software.

Although formalin fixation and paraffin-embedding is a commonly used method in EGFR mutation detection, low yields of RNA/DNA are extracted from formalin-fixed paraffin-embedded tissues, and they are often degraded or may contain modifications that inhibit polymerase reactions, which can bias results. Additionally, the detection accuracy is critically dependent on the dye type, instrument resolution, PCR product length, and PCR specificity.^62,63 LCGreen Plus dye detects heterozygotes better than SYTO 9, which is better than EvaGreen. ⁶⁴ Some of the latest real-time thermal cyclers modified to incorporate HRM can yield quality high-resolution data by melting 18 times slower than the HR-1 instrument.^24,65 Melting determination is performed immediately after PCR, and different heterozygotes may produce melting curves so similar that, although they vary from those of homozygous variants, they are not different. ⁶⁶ Therefore, specific amplification of the target of interest is critical, requiring careful choices of primers and optimized temperature cycling.

HRM analysis has been used to discriminate many tumor variants, such as BRAF mutations in colorectal tumors, KIT (the c-kit gene) in gastrointestinal stromal tumors, EGFR, and AKTI in non-small cell lung cancer.^9,17 Driver oncogenes, including EGFR, KRAS, and BRAF, activated by deletion and/or missense/insertion mutations, drive the critical step toward developing non-small cell lung cancer. EGFR, BRAF, and KRAS mutation sensitivities in anti-EGF-receptor therapies are mutually exclusive. Recently published studies reported that HRM analysis is a specific and sensitive method for testing various samples, and a low quantity of DNA is needed for BRAF and KRAS mutation screening.^67,68 We noted that the SROC AUC was accurate for HRM scanning of the EGFR mutation. Therefore, HRM analysis may be a promising method to detect a series of driver oncogene mutations, including EGFR, KRAS, and BRAF mutations, but confirmation by direct sequencing or other methods is necessary, especially in a diagnostic context.

Our study has several limitations, such as substantial heterogeneity across all included studies. Although meta-regression and subgroup analyses were performed, the sources of heterogeneity were undetermined except for a threshold effect. Additionally, inherent discord was observed between HRM and DNA sequencing. There were 136 FPs and 51 FNs. Mutations found by HRM analysis should always be confirmed with DNA sequencing so that FPs are not an issue (they will be wild type afterward). FNs are relatively serious because they cannot be sequenced, and this may cause mutations in patients to be misclassified as wild type by HRM analysis. Therefore, these patients would be denied TKI therapy. However, the proportion of FNs is very low (approximately 0.29%). Thus, HRM analysis offers appropriate diagnostic performance for EGFR mutation screening in oncology-associated diseases and represents a method with high throughput, low labor, low cost, simplicity, and rapid turnaround, but positive results must be sequenced for diagnostic confirmation. Although our meta-analysis focused on the use of HRM analysis to detect EGFR gene exon mutations, we recognize the significant roles that EGFR gene methylation and epigenetic regulation play in tumorigenesis. At present, there are relatively few studies on EGFR methylation status using HRM techniques, and these studies did not meet the criteria for our analysis. Therefore, our analysis does not encompass these areas. Future research may consider applying HRM analysis to detect EGFR methylation status, which could offer new insights into the role of EGFR in tumors.