Comparison of real-time PCR and nCounter NanoString techniques to validate copy number alterations in oral cancer

Abstract

Copy number alterations (CNAs) are imperative in determining the patient’s prognostic and predictive status. Real-time polymerase chain reaction (PCR) is extensively used to validate the results of global genomic profiling methods. However, nCounter NanoString is yet to be thoroughly assessed and compared for copy number analysis. For the first time, we have comprehensively compared the aforementioned techniques in 119 oral cancer samples to evaluate 24 genes. Spearman’s rank correlation and Cohen’s Kappa score were calculated to identify concurrence. Kaplan–Meier curves with the Log-rank test were used to identify genes related to disease prognosis, including recurrence-free survival (RFS), disease-specific survival (DSS), and overall survival (OS). Spearman’s rank correlation ranged from r = 0.188 to 0.517, and Cohen’s kappa score showed a moderate to substantial agreement. Additionally, in real-time PCR, ISG15 was associated with a better prognosis for RFS, DSS and OS, along with ATM, CASP4, and CYB5A with poor RFS. However, in nCounter NanoString, ISG15 was associated with poor prognosis for RFS, DSS, and OS, while CDK11A showed poor prognosis for RFS. Real-time PCR remains a robust method to validate the genomic biomarkers. However, these observations should be rigorously validated by conducting additional, well-designed, independent studies.

Introduction

Oral cancer accounts for 30% of all cancer cases worldwide, of which one-third are found in India. It is the most common cancer among men in India, the fourth most common cancer in women, and the second most common overall cancer^1,2. Multiple studies have reported the significance of copy number alterations (CNAs) for predicting the prognostic status of patients with various diseases, including cancer^3–5. CNAs play an important role in activating oncogenes and inactivating tumour suppressor genes⁶. Our previous study identified specific CNAs associated with prognosis, which were validated using real-time PCR⁷. Over the years, various techniques have been developed, and constant improvements have been made in resolving and detecting CNAs. Fluorescent in situ hybridization (FISH) is commonly used to validate CNAs^8–10. However, its usage is constrained due to several factors, such as labour-intensive, time-consuming, inability to detect microdeletions and duplications, etc¹¹. Real-time polymerase chain reaction (PCR) is the gold standard for validating results of global genomic profiling and is also commonly used in clinical diagnosis^7,12. In addition to real-time PCR, NanoString’s nCounter analysis system (NanoString Technologies, Seattle, WA 98,109, USA) has been developed, which facilitates customised multiplex analysis of gene targets in multiple samples more rapidly and efficiently by taking advantage of unique colour-coded reporter probes^13,14. Other major advantages over the existing platforms include the direct measurement of target genes without enzymatic reactions, high sensitivity coupled with targeted multiplex capability, less laborious, and a digital readout. In short, the sensitivity of the nCounter NanoString technique is higher than microarrays and comparable to real-time PCR^13,15,16. However, comprehensive validation of CNAs using real-time PCR and nCounter NanoString techniques is lacking. The similarities and differences between the two methods are mentioned in Table 1.

Table 1.

Comparison of real-time PCR and nCounter NanoString.

	Real-time PCR	nCounter NanoString
Technique	Quantitative	Hybridisation
Principle	Counts reaction cycles used to reach amplification slope status in real-time	Senses the hybridized colour-coded probes and captures the signal intensities
Primers/Probes	Fluorescent dyes or taqman probes are used to monitor the accumulation of amplified DNA in real-time	Capture probe and reporter probe are used to count individual the target molecules
Instrument	Thermal cycler: temperature settings to perform the three stages of the PCR: denaturation, annealing, and extension	nCounter prep station and digital counter: The reporter probe carries the signal, while the capture probe allows the complex to be immobilized so that the Digital Analyzer can detect the colour codes
Quantification	The amount of PCR product can be quantitatively measured by fluorescence reporter detection	The Digital Analyzer uses a CCD camera through a microscope objective lens to magnify and image the immobilized Reporters, capturing hundreds of images per sample
Data Analysis	Standard curve method: comparing the Ct values of interest with a control, which are both then normalized to a housekeeping gene	The nSolverTM software performs quality control (QC) and data normalization and analysis
Multiplexing	Yes, relatively fewer genes	Yes, maximum 800 genes
Clinical diagnostics	1. ThyraMIR™: diagnosis of thyroid nodules 2. Oncotype DX: first genomic biomarker assay advising on breast cancer treatment options	1. PROSIGNA: measuring the risk of relapse (ROR), a prognostic factor for relapse-free survival in breast cancer patients

Studies on the comparison of real-time PCR and nCounter NanoString techniques for measuring gene expression, isoform expression and miRNA have been conducted earlier^17–19. However, only one study reported gene expression with clinical outcomes^17–19. Moreover, very few studies have used the aforementioned techniques to detect copy numbers, which are verified at small sample sizes and with fewer genes. Using the appropriate validation technique is important to exclude the probability of random events and false-positive/negative results. In the current study, we aim to extensively compare the existing gold standard method, real-time PCR, with the newly developed platform, nCounter NanoString. We also aim to validate the prognostic biomarkers identified/reported in our study. To the best of our knowledge, for the first time, we have comprehensively validated and compared CNAs in the genes which emerged as predictors of clinical outcomes identified in our previous studies^7,8 using real-time PCR and nCounter NanoString assay. To determine the genomic biomarkers associated with the prognosis, we correlated the CNAs observed in 24 genes evaluated through both platforms in 119 oral squamous cell carcinoma (OSCC) samples. Further, we analysed the genomic biomarkers associated with prognosis, including recurrence-free survival (RFS), disease-specific survival (DSS), and overall survival (OS). Appropriate biomarker validation technology can aid in understanding disease biology, early detection, and clinical management.

Results

Study design

The cross-platform assessment of real-time PCR and nCounter NanoString was performed on N = 119 OSCC patient samples, which is referred to as a subgroup. These cases were derived from treatment-naive 127 OSCC patients, which served as a validation set in our previous report and is referred to as the parent group. Since the amount of DNA was insufficient in eight cases, nCounter NanoString CNA analysis was performed in the remaining 119 samples. The baseline characteristics for parent (n = 127) and subgroup (n = 119) patients are shown in Table 2. Both techniques were assessed using performance parameters, including detection rate, interplatform correlation and association with patient outcomes. The 24 genes which were included in the study are based on our earlier published reports of genomic, transcriptomic and methylomic analysis, wherein the genes and chromosomal loci were associated with clinical outcomes such as nodal metastasis and survival in response to the treatment^7,8.

Table 2.

Baseline characteristics for parental and subgroup cohort.

Characteristics	Parental group n = 127	Subgroup n = 119
Age (Years)
> 60	32	32
≤ 60	95	87
Gender
Male	108	103
Female	19	16
Node
Positive	52	47
Negative	75	72
Clinical stage
Early (I & II)	44	43
Advanced (III & IV)	83	76
Grade
Poor	20	20
Moderate	96	90
Well	10	8
T stage
T1 & T2	69	66
T3 & T4	58	53
Recurrence free survival
Recurrence	57	54
No Recurrence	70	65
Disease specific survival
Alive	30	29
Death	93	87
Overall Survival
Alive	93	87
Death	34	32

The probe sets for both techniques were designed on the basis of the probe sequences present on the array CGH platform, ensuring the coverage of similar gene regions. Female pooled DNA served as a reference for both methods. For nCounter NanoString, we used three probes for genes associated with amplification, and five probes for genes associated with deletion. All the reactions were performed in single as replicates are not required as per the manufacturer’s guidelines. Whereas, for real-time PCR, Taqman assays were used and the reactions were performed in quadruplets as per the MIQE guidelines²⁰. The complete list of primer/probes used for real-time PCR and nCounter NanoString is provided in Supplementary Tables 1 and 2, respectively. The Spearman rank correlation and Cohen’s kappa score were calculated between the CNA obtained by nCounter analyses and real-time PCR analyses for each of the 24 genes, which were further associated with clinical outcomes.

Comparison of CNAs in real-time PCR and nCounter NanoString

The CNAs detected by real-time PCR and nCounter NanoString are presented in Fig. 1. Each box plot represents the minimum and maximum copy number along with median values detected for individual genes. Both techniques displayed extensive variation in the quantification of copy numbers for most genes. Although CNAs were detected through both methods, we observed a lower copy number detection in the nCounter NanoString compared to real-time PCR. The CNAs observed for each gene in 119 patient samples are provided in Supplementary Table 3, and the percentage amplification and deletion of CNAs were calculated and are provided in Supplementary Table 4. In real-time PCR data, the copy number amplification was observed for more than 50% of samples for genes, including ANO1, DVL1, ISG15, MVP, SOX8, and TNFRSF4, compared to nCounter NanoString data.

Fig. 1.

Copy number measurement for 24 genes using real-time PCR (a) and nCounter NanoString (b). Each Box plot represents the minimum and maximum copy number obtained through real-time PCR and nCounter NanoString.

The Spearman rank correlation coefficient was calculated for each gene and is graphically presented as a correlogram Fig. 2. The correlation values obtained for both techniques showed a weak to slightly moderate correlation Supplementary Table 5. We found that of the twenty-four genes analyzed, six genes, including CASP4, CDK11B, CST7, LY75, MLLT11 and MVP, did not correlate. A weak correlation was observed in sixteen genes, including ANO1, ATM, BIRC2, BIRC3, CCND1, CDK11A, CYB5A, DVL1, FADD, FAT1, GHR, ISG15, LRP1B, PDL1, SEPTIN1 and SOX8. A moderate correlation was observed for only two genes, TNFRSF4 (r = 0.513) and YAP1 (r = 0.517), and the lowest was for CDK11A (r = 0.188). In addition, Cohen’s Kappa score was calculated to analyse how well the two techniques agreed on gain or loss of copy number for a particular sample Fig. 3. No agreement was found in nine genes, including CDK11A, CDK11B, DVL1, ISG15, LRP1B, MLLT11, MVP, SOX8, and TNFRSF4. Slight to fair agreement was observed for five genes, including ATM, CASP4, CST7, CYB5A, and SEPTIN, and a moderate to substantial agreement was observed for eight genes, including BIRC2, BIRC3, CCND1, FADD, FAT1, GHR, PDL1 and YAP1 Supplementary Table 6.

Fig. 2.

Correlation between CNAs measured on two platforms. Each box in the correlogram displays concordance between the technique for 24 genes in 119 OSCC samples.

Fig. 3.

Cohen’s Kappa score for agreement between copy number measured using real-time PCR and nCounter NanoString in 119 samples.

Association with clinical outcomes

The real-time PCR and nCounter NanoString data concurrence in determining the prognostic relevance was assessed for each gene Table 3. Real-time PCR analysis revealed that ISG15 was associated with better clinical outcomes, RFS [HR 0.40 (0.20—0.81), p = 0.009], DSS [HR 0.31 (0.13—0.74), p = 0.005] and OS [HR 0.30 (0.13—0.68), p = 0.002]. While genes including CASP4 [HR 3.32 (1.29—8.48), p = 0.008], CYB5A [HR 4.77 (1.85—12.30), p = 0.000], and ATM [HR 2.55 (1.00—6.51), p = 0.041] were associated with poor RFS. The prognostic markers remained identical in the subgroup population upon reanalysis of real-time PCR data, except for the gene ATM. Whereas, in nCounter NanoString, ISG15 was found to be associated with poor RFS [HR: 3.396 (1.52—7.57), p = 0.001], DSS 3.42 [HR: (1.30—8.97), p = 0.008], OS [HR: 3.069 (1.18—7.97), p = 0.015] and CDK11A was found to be linked with poor RFS [HR: 2.542 (1.27—5.08), p = 0.006]. Interestingly, ISG15 was found to be associated with all three survival outcomes (RFS, DSS and OS), with contrasting prognosis obtained for both methods, which we have represented in Fig. 4. Overall, a significant difference in prognostic markers associated with survival was observed for both techniques.

Table 3.

Correlation between nCounter NanoString and Real-time PCR and corresponding Cox proportional hazard regression for each gene.

Sr.No	Gene	Spearman rank correlation		Real-time PCR							nCounter NanoString
				RFS		DSS		OS			RFS		DSS		OS
		Correlation	p value	Hazard ratio (95% CI)	p value	Hazard ratio (95% CI)	p value	Hazard ratio (95% CI)	p value		Hazard ratio (95% CI)	p value	Hazard ratio (95% CI)	p value	Hazard ratio (95% CI)	p value
1	ANO1	.415**	0.000	0.68 (0.30–1.51)	0.344	0.46 (0.17–1.23)	0.116	0.53 (0.20–1.39)	0.193		1.54 (0.83–2.58)	0.162	1.46 (0.64–3.32)	0.356	1.50 (0.69–3.26)	0.290
2	ATM	.352**	0.000	2.55 (1.00–6.51)	0.041	0.67 (0.09–5.00)	0.701	0.61 (0.08–4.50)	0.628		0.82 (0.35–1.92)	0.650	0.84 (0.25–2.80)	0.780	1.01 (0.35–2.89)	0.970
3	BIRC2	.428**	0.000	1.45 (0.65–3.23)	0.360	1.56 (0.53–4.58)	0.407	1.38 (0.47–3.98)	0.551		1.19 (0.47–2.99)	0.710	1.34 (0.40–4.46)	0.620	1.20 (0.36–3.98)	0.750
4	BIRC3	.319**	0.000	1.09 (0.53–2.25)	0.805	0.93 (0.32–2.71)	0.897	0.81 (0.28–2.35)	0.711		0.94 (0.34–2.62)	0.917	0.88 (0.21–3.74)	0.870	0.80 (0.19–3.36)	0.760
5	CASP4	0.145	0.115	3.32 (1.29–8.48)	0.008	2.58 (0.60–11.08)	0.186	3.06 (0.91–10.2)	0.057		1.00 (0.39–2.53)	0.990	1.19 (0.35–3.96)	0.770	1.05 (0.32–3.49)	0.920
6	CCND1	.402**	0.000	0.98 (0.55–1.77)	0.964	0.90 (0.40–2.04)	0.809	0.92 (0.42–2.00)	0.843		1.46 (0.78–2.74)	0.220	1.28 (0.54–3.01)	0.560	1.33 (0.59–2.98)	0.470
7	CDK11A	.188*	0.041	1.32 (0.76–2.29)	0.317	1.16 (0.97–3.47)	0.171	1.40 (0.69–0.68)	0.346		2.54 (1.27–5.08)	0.006	1.39 (0.48–4.02)	0.530	1.26 (0.44–3.60)	0.650
8	CDK11B	0.050	0.586	0.87 (0.39–1.94)	0.743	0.86 (0.30–2.48)	0.788	0.78 (0.27–2.24)	0.650		1.23 (0.49–3.10)	0.650	0.73 (0.17–3.09)	0.670	0.67 (0.16–2.80)	0.580
9	CST7	0.060	0.519	0.04 (0.00–100.9)	0.237	0.04 (0.00–3172.9)	0.416	0.04 (0.00–1407)	0.380		1.70 (0.67–4.30)	0.250	1.67 (0.50–5.55)	0.390	1.50 (0.45–4.95)	0.490
10	CYB5A	.219*	0.017	4.77 (1.85–12.30)	0.000	3.09 (0.92–10.19)	0.054	2.74 (0.83–9.03)	0.084		1.14 (0.60–2.17)	0.680	1.18 (0.50–2.76)	0.700	1.22 (0.55–2.73)	0.610
11	DVL1	.381**	0.000	0.53 (0.16–1.75)	0.300	0.43 (0.10–1.86)	0.250	0.32 (0.09–1.08)	0.055		1.86 (0.90–3.84)	0.086	1.77 (0.67–4.64)	0.240	1.59 (0.61–4.14)	0.330
12	FADD	.273**	0.003	0.89 (0.45–1.78)	0.752	1.03 (0.42–2.53)	0.945	1.09 (0.47–2.52)	0.839		1.28 (0.68–2.39)	0.430	1.18 (0.50–2.78)	0.690	1.23 (0.55–2.76)	0.600
13	FAT1	.292**	0.001	0.88 (0.12–6.40)	0.901	1.59 (0.21–11.86)	0.643	1.14 (0.19–10.45)	0.732		0.81 (0.11–6.00)	0.830	0.04 (0.00–3616)	0.420	0.04 (0.00–2314)	0.400
14	GHR	.420**	0.000	1.47 (0.66–3.27)	0.338	0.68 (0.16–2.89)	0.605	0.61 (0.14–2.56)	0.498		1.68 (0.71–3.93)	0.220	1.57 (0.47–5.22)	0.450	1.91 (0.67–5.48)	0.210
15	ISG15	.230*	0.012	0.40 (0.20–0.81)	0.009	0.31 (0.13–0.74)	0.005	0.30 (0.13–0.68)	0.002		3.39 (1.52–5.57)	0.001	3.42 (1.30–8.97)	0.008	3.06 (1.18–7.97)	0.010
16	LRP1B	.351**	0.000	1.65 (0.70–3.88)	0.241	1.08 (0.25–4.59)	0.908	1.40 (0.42–4.64)	0.572		1.91 (0.46–7.93)	0.360	1.12 (0.15–8.31)	0.900	1.03 (0.14–7.58)	0.970
17	LY75	0.054	0.561	0.63 (0.26–1.47)	0.284	0.58 (0.17–1.92)	0.367	0.51 (0.15–1.70)	0.270		NA		NA		NA
18	MLLT11	0.095	0.561	0.65 (0.23–1.82)	0.413	0.95 (0.28–3.16)	0.937	0.85 (0.25–2.81)	0.795		1.33 (0.63–2.84)	0.440	0.79 (0.24–2.62)	0.700	0.72 (0.22–2.38)	0.590
19	MVP	0.043	0.647	0.76 (0.40–1.43)	0.407	0.91 (0.37–2.25)	0.846	1.02 (0.41–2.49)	0.963		0.67 (0.35–1.29)	0.230	0.59 (0.24–1.46)	0.250	0.53 (0.22–1.30)	0.160
20	PDL1	.484**	0.000	1.09 (0.59–2.01)	0.780	1.26 (0.56–2.87)	0.566	1.48 (0.70–3.14)	0.298		1.39 (0.73–2.65)	0.310	1.84 (0.81–4.18)	0.130	1.87 (0.86–4.06)	0.100
21	SEPTIN1	.193*	0.035	1.07 (0.59–1.93)	0.824	0.93 (0.41–2.12)	0.874	0.84 (0.37–1.88)	0.672		0.95 (0.29–3.07)	0.940	1.05 (0.25–4.44)	0.940	0.96 (0.23–4.06)	0.960
22	SOX8	.258**	0.005	1.03 (0.41–2.63)	0.938	0.87 (0.26–2.88)	0.819	0.73 (0.25–2.11)	0.572		0.90 (0.43–1.86)	0.780	1.18 (0.48–2.94)	0.700	1.07 (0.44–2.63)	0.860
23	TNFRSF4	.513**	0.000	0.83 (0.41–1.66)	0.605	0.78 (0.31–1.92)	0.595	0.73 (0.31–1.68)	0.461		1.47 (0.75–2.87)	0.250	1.41 (0.57–3.50)	0.440	1.27 (0.52–3.11)	0.590
24	YAP1	.517**	0.000	1.08 (0.43–2.57)	0.969	1.19 (0.35–4.00)	0.771	1.05 (0.31–3.25)	0.931		1.05 (0.45–2.47)	0.890	1.01 (0.30–3.36)	0.970	0.92 (0.28–3.02)	0.890

Fig. 4.

Kaplan–Meier estimates of 119 OSCC samples for ISG15. The association of CNAs with clinical outcomes was obtained by real-time PCR for RFS, DSS and OS (a – c), respectively and nCounter NanoString for RFS, DSS, and OS (d – f), respectively. p values originate from log-rank tests.

Discussion

Copy number alterations (CNAs) are complex chromosomal structural variations in the genome, serving as a biomarker in multiple cancers; therefore, an accurate and reliable method to detect CNAs is indispensable to determine the patient’s diagnosis and prognosis. The current study compared CNAs analysed by real-time PCR and the nCounter NanoString assay in OSCC. This is the first report wherein both techniques were compared to investigate the CNAs of 24 genes in 119 OSCC samples with clinical outcomes. We observed a weak to moderate correlation between both techniques: the highest correlation coefficient for YAP1 (r = 0.517) and the lowest for CDK11A (r = 0.188), along with differences in the agreement for gain or loss of individual genes. Moreover, the genomic alterations associated with the prognosis of patients were evaluated, where both techniques showed wide variation in the clinical outcomes, including RFS, DSS, and OS.

Real-time PCR and nCounter NanoString techniques have been extensively used to validate genomic biomarkers; however, only a few reports have thoroughly compared the results of these techniques, with most studies comparing gene expression levels^17–19,21. To date, 13 studies have compared the aforementioned techniques, of which seven studies have found a strong correlation (Supplementary Table 7) and six studies have reported a weak correlation (Supplementary Table 8) between real-time PCR and nCounter NanoString. Schmidt et al. compared nCounter NanoString with real-time PCR for measuring gene expression levels in patients with locally advanced HNSCC for 38 genes, where the median correlation obtained between both techniques was r = 0.84, which shows a strong correlation as compared to our findings¹⁷. On the other hand, Zhang et al. evaluated the isoform and gene expression variations across NanoString, real-time PCR, RNA-Seq, and arrayCGH using 46 cancer cell lines across different cancer types; the agreement on isoform expressions was reported to be lower than gene expressions across the four platforms. Interestingly, the real-time PCR data was more consistent with RNA-seq and Exon-array data than NanoString for isoform quantification²². These observations concur with our previous study, where real-time PCR data demonstrate excellent concordance (r = 0.924) with the gene expression arrays²³. The highest correlation reported for measuring gene expression between real-time PCR and nCounter NanoString techniques is r = 0.84 by Schmidt et al.¹⁷, and the lowest correlation with r =—0.106 by Adam et al¹⁸. Additionally, the association with clinical outcomes was reported by only one study¹⁷.

We have listed studies comparing the results of real-time PCR and nCounter NanoString wherein discrepancies between the correlation or clinical outcomes were observed, which concur with our results (Supplementary Table 8). Adam et al. evaluated 11 target genes to assess the robustness of NanoString in 45 Human FFPE renal allograft tissues, and the results were compared to real-time PCR for gene expression levels. The median correlation coefficient observed was r = 0.487, along with a wide variation, ranging from—0.106 to 0.638 between the two methods, which concur with our results¹⁸. Further, Bergbower et al. assessed the gene expression level of fresh-frozen cardiac explant tissues. They observed lower expression levels in NanoString compared to real-time PCR, along with a high inter-platform variability¹⁹. Similarly, miRNAs in human serum and plasma samples were evaluated using real-time PCR, NanoString and miRNA-Seq platforms. The expression levels of miRNAs in NanoString were lower than those of other platforms, with a weak to moderate correlation with real-time PCR and a non-significant correlation with miRNA-Seq²⁴.

Only four studies have employed nCounter NanoString and real-time PCR to detect CNAs, although they have smaller sample sizes and fewer genes than the current study (Supplementary Table 9). Openshaw et al. identified recurrently amplified genes through NanoString in tumour DNA and validated four genes using real-time PCR, where they observed excellent concordance with r = 0.984, which shows a strong correlation as compared to our findings²⁵. Similarly, Kim et al. found six genes amplified in gastric cancer through NanoString, of which MET correlated well with real-time PCR, but KRAS showed increased copy numbers in real-time PCR for one case²⁶. Further, Lee et al. validated three genes by real-time PCR and, upon comparison with NanoString, KRAS amplification was inconsistent between both techniques²⁷. However, one of the major limitations of these studies is that they do not have adequate sample sizes, and relatively few genes were analysed.

We correlated real-time PCR data from 119 subgroup patients with clinical outcomes and found that gain in the gene ISG15 was a better prognostic marker for RFS, DSS and OS. Deletion of CASP4, CYB5A and ATM genes was associated significantly with RFS; these observations are consistent with the parent cohort of 127 patients⁷. In our earlier report, the ATM gene had a borderline association with RFS. However, a significant association was found in the subgroup cohort of 119 patients, possibly due to a non-random reduction in sample size (Supplementary Table 10). On the contrary, the results of nCounter NanoString revealed that a gain in the gene ISG15 was associated with poor prognosis for RFS, DSS and OS and the CDK11A with poor RFS. In our data, the amplification of ISG15 is 86.6% in real-time PCR and 6.7% in nCounter NanoString, resulting in contrasting clinical outcomes (Fig. 4). Furthermore, we calculated the frequency of CNAs observed in our training set (n = 28) and validation set (n = 127) (Supplementary Table 11). Interestingly, we found an identical percentage for both sets for ISG15, associated with all clinical outcomes. Chong et al. reported amplification of ISG15 at genomic and mRNA levels in patients with OSCC²⁸. Also, overexpression of ISG15 was reported in 80% of OSCC tissues collected from Indian patients²⁹. These data suggest that ISG15 is amplified in OSCC and needs further investigation by conducting additional independent patient cohorts with clinical outcomes, as it is an intricate gene³⁰.

Asnaghi et al. validated the copy number alterations by nCounter NanoString, wherein they could detect gains and losses of very large genomic regions, and alterations in small genomic regions could not be efficiently detected. However, they did not perform a comparative analysis with any other validation technique, including qPCRs³¹. Further, some studies have compared nCounter NanoString with other techniques, including real-time PCR and speculate that target amplification, DNA/RNA isolation techniques, sample type (Fresh frozen or FFPE), and technical mechanism may influence the result obtained from both methods, leading to the discrepancy^19,24,32. nCounter NanoString offers multiplexing with fewer biological samples compared to real-time PCR. However, tumour characteristics such as copy number variations (polysomy), structural alterations in the target gene, and structure and function of proteins predicted to be altered by genetic alteration may be complex to assess³³. On the other hand, real-time PCR has drawbacks such as non-uniform enzymatic responses, pipetting errors, labour-intensive, require a large amount of cDNA to produce limited gene expression analysis, suited for few genes in a limited number of samples, reference gene instability which might affect the precision, repeatability, and specificity of real-time PCR data³⁴.

To the best of our knowledge, our study comprises a comprehensive comparison of CNA of two independent platforms on clinically well-annotated patient DNA samples along with their clinical outcomes. Considering each platform’s strengths and weaknesses and the current study’s observation, we demonstrate that real-time PCR could be a more robust method to validate the genomic biomarkers derived from global genomic profiling. Nevertheless, droplet-digital PCR (DD-PCR), which is a modified version of real-time PCR, can be used to detect small CNAs and overcome the limitations of the technique. DD-PCR is a more stringent, efficient, sensitive method and less susceptible to PCR inhibitors^35,36, with ease of use and increased precision³⁷. While nCounter Nanostring has been reported to be a sensitive technology, our data suggests that researchers should carefully select the validation method. Furthermore, increasing the starting material (DNA) for assays that do not involve target amplification could improve the detection sensitivity³⁸. Here, we have used probes that cover similar gene regions and adequate DNA samples with optimum experimental conditions for both methods, which indicates that the weak correlations obtained may be due to differing techniques.

Employing the appropriate method for validating genomic biomarkers is crucial; our aim in this comparative analysis was to identify a reliable method to enumerate CNAs that can serve as a prognostic or predictive biomarker in clinical settings. As per the guidelines published by manufacturers, for the nCounter CNV assay, replicates are not required (https://nanostring.com/wp-content/uploads/2019_PB_nCounter_CNV_Assay). However, replicate measurements could have further strengthened our observations for nCounter Nanostring, which is one of the limitations of our study. The genes that showed significant associations with clinical outcomes should have also been subjected to protein-based analysis (Immunohistochemistry or Enzyme-Linked Immunosorbent Assay); this would have given interesting insights into the association of CNAs and the expression of proteins, which is another limitation of our study. The primary strength of our study includes the homogenous group of OSCC patient cohorts with their clinical and survival outcomes. In the future, independent studies employing different genetic alterations with prognostic value should be analysed using nCounter NanoString and real-time PCR to strengthen these findings.

Conclusion

Overall, the comparison of copy number analysis by real-time PCR and nCounter NanoString assay demonstrated (i) extensive discrepancy with a weak to moderate correlation in CNAs and poor agreement for most of the genes. (ii) significant difference in the genomic biomarkers associated with disease prognosis. The current study reflects on the importance of selecting the appropriate technology, especially for validating biomarkers associated with prognostic outcomes. Although NanoString offers advantages, including multiplex analysis of targets over real-time PCR, our results demonstrate that detecting/validating the CNAs by real-time PCR remains the gold standard, which needs to be rigorously validated by conducting additional independent studies that are adequately empowered using different sets of genomic biomarkers with prognostic and predictive value. Our study delineates the importance of real-time PCR in detecting copy number alterations, and based on this, we recommend employing it for diagnosis in clinical settings.

Materials and methods

Study design

The current study was conducted on treatment-naive 127 oral cancer (gingivo-buccal) patients, which served as a validation cohort in our previous report⁷. The inclusion criteria were (i) Participants aged above 18 years, no maximum age, with histologically confirmed squamous cell cancers of the oral cavity, (ii) Treatment-naïve, pathologically diagnosed, and surgically resected gingivobuccal tumor (iii) All the patients should be treatment compliant and should have received therapy as per the stage of the disease, (iv) cases with minimum follow-up period of one year. The exclusion criteria were (i) Patients with a previous history of other cancers and those who receive other investigational therapy/agents as part of the treatment, (ii) HPV, HIV-positive, Hepatitis B and C seropositive patients are excluded from this study. The study was approved by the institutional ethics committee of the Tata Memorial Centre (IEC approval no 218 of 2016). The investigation was carried out in compliance with the Declaration of Helsinki principles and recommendations for good clinical practice. The institute’s data safety and monitoring board periodically monitored the study. All participants signed the informed consent. The details of 127 patients are provided in Supplementary Table 12. Due to insufficient DNA, eight cases of the parent group were excluded from the nCounter NanoString analysis. A total of 119 samples were available to compare the results of real-time PCR and nCounter NanoString for 24 genes. All the patients were HPV-negative⁷. Baseline characteristics of the parental (n = 127) and subgroup (n = 119) patients are provided in Table 2.

Real-time PCR

TaqMan probes selected for the assays were from the exonic regions. The genes that emerged as predictors of clinical outcome in our previously published study were selected for the comparison of real-time PCR and nCounter NanoString. The assay IDs of all the target genes analysed by TaqMan real-time PCR are provided in Supplementary Table 1. The real-time PCR reactions were performed as described in our earlier study⁷. The Copy Caller-predicted copy number was used for the analysis; copy number values ranging between 1.5 to 2.5 were considered as no change, more than 2.5 as gain/amplification and less than 1.5 as loss/deletion.

nCounter NanoString

A panel of 24 gene probes was created for CNA detection. The gene list and probe details are provided in Supplementary Table 2. Three probes were designed for genes associated with amplification, and five probes for genes associated with deletion. The nCounter NanoString assay was performed according to NanoString’s MAN-10093–01 standard protocol. Briefly, 100 ng of gDNA was added with 1μL 10X AluI Fragmentation Buffer, 1μL 10X CNV DNA Prep Control, and 1 µl of 5U/µl AluI Fragmentation Enzyme. The contents were mixed thoroughly and incubated in a thermal cycler at 37 °C for 2 h for the fragmentation of the DNA. Next, the fragmented gDNA was denatured at 95 °C for 5 min using a thermal cycler to produce single strands. Further, denatured gDNA was incubated for at least 16–18 h at 65 °C in a hybridization buffer containing the CodeSet, consisting of the reporter and capture probes and the target DNA, forming a tripartite complex. All the additions were performed on ice. Post-hybridization, the complex was loaded in the cartridge, bound by a biotin-labelled capture probe on a streptavidin-coated glass slide. The hybridised DNA-CodeSet complexes were scanned, purified, and the reporters were counted in the nCounter digital analyzer. Quality control assessment for imaging, binding density, positive control linearity and positive control limit of detection was done according to the manufacturer’s recommended default parameters (MAN-C0019-08) with nSolver™ Analysis Software version 4.0 (NanoString Technologies, Seattle, WA, USA). Each sample was mixed with six positive controls in fixed amounts (128—0.125 fm) and eight negative controls. Background subtraction was performed for each sample by subtracting the mean of the negative controls from all data points. Each sample was first normalized to the geometric mean of the positive controls, followed by normalization to the geometric mean of the 10 invariant probes (Probes designed to target regions of the genome that have only two copies in most individuals), which was part of the assay design. Female pooled DNA was used as a reference, and CNAs were determined by averaging over three/five probes used per gene. Copy number values between 1.5 and 2.5 were considered as no change, more than 2.5 as gain/amplification, and less than 1.5 as loss/deletion.

Cross-platform comparison and statistical analysis

Spearman’s rank correlation coefficient was calculated for each gene individually to identify the correlation between real-time PCR and nCounter NanoString assays. Representation of correlation coefficients was done by correlogram using R software version 4.2.0. In addition, Cohen’s Kappa score was calculated to determine the degree of agreement between both techniques for no change and gain or loss for a particular sample. Recurrence-free survival (RFS), Disease-specific survival (DSS), and Overall survival (OS) were defined as described previously⁷. The median follow-up period was 79 months. RFS, DSS, and OS were estimated using the Kaplan–Meier method and compared using log-rank tests. The Cox proportional hazard model was used to estimate hazard ratios (HRs) and compute 95% confidence intervals (CIs). All reported tests were performed in two-tailed. A p-value of < 0.05 was considered statistically significant. All statistical analyses were performed using IBM SPSS version 25.

Abbreviations

OSCC

Oral squamous cell carcinoma

CNA

Copy number alteration

RFS

Recurrence-free survival

DSS

Disease-specific survival

OS

Overall survival

HPV

Human papillomavirus

HR

Hazard ratio

Author contributions

Conceived and designed the experiments: RC, MI, MBM. Performed the experiments: RC, MI. Analysed the data: RC, MI, JG, MBM. prepared figures: RC, JG, MBM.Contributed reagents/materials/analysis tools: MBM. Wrote the paper: RC, MBM. All authors read and approved the final manuscript.

Funding

Open access funding provided by Department of Atomic Energy. This work was supported by a grant from the Terry Fox International Foundation and intramural support of Tata Memorial Centre for Basic and Translational Research in Cancer DAE Grant—No.1/3(7)/2020/TMC/R&D-II/8823 Dt.30.07.2021.

Data availability

The data underlying this article are available in the article itself and in its online supplementary material.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval

The study was approved by the Institute Ethics Committee (IEC) of Tata Memorial Centre, approval number 218 of 2016. Written informed consent was obtained from all the study participants.

Consent for Publication

All authors consent to the publication of the work and data presented in the manuscript.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-08036-9.