SNP6 array-based estimated genomic coordinates of interstitial copy number deletion/fusion of the CDKN2A gene in cancer cell lines (n=273) with homozygous CDKN2A deletion and estimated genomic coordinates of deep-deleted fragments of CDKN2A, PTEN, RB1, and other frequently deleted genes in cancer tissues were downloaded from the Copy Number Analysis (CONA) datasets in the COSMIC project ( Data Files 1-11 ) (13).

Frozen fresh gastric carcinoma (GC) and paired surgical margin (SM) tissue samples were collected from 156 patients in the WGS study (14). These samples were frozen in liquid nitrogen approximately 30 min after surgical dissection and then stored in a -80°C freezer for 2-5 yrs. Clinicopathological information was also obtained. The 2010 UICC tumor-node-metastasis (TNM) system was used to classify these GCs (15). Genomic DNA was extracted from these samples with a phenol/chloroform method coupled with RNase treatment. Concentrations of these DNA samples were determined with NanoVue Plus (Biochrom LTD, Cambridge, UK). DNA samples with OD_260nm/OD_280nm ratios ranging from 1.7 to 1.9 were used for the detection of gene copy number as described below.

Multiplex primer and probe combinations were designed based on the best multiplex primer probe scores for conserved sequences within the CDR in the CDKN2A (HGNC: 1787) and GAPDH (HGNC: 4141) gene sequences by Bacon Designer 8 software. Multiplex PCR assays were established according to the Applied Biosystems (ABI) TaqMan universal PCR master mix manual. The performance of these assays for the detection of CDKN2A copy numbers was compared with each other. Finally, a multiplex primer and probe combination targeting CDKN2A intron-2 was selected ( Table 1 ), and the concentrations of the components were optimized. Each multiplex PCR assay was carried out in a total volume of 20 μL that included 5-10 ng of input DNA, 10 μM of forward and reverse primers and probe for CDKN2A intron-2, 10 μM forward and reverse primers and probe for GAPDH, and 10 μL of 2x TaqMan Universal Master Mix II with uracil-N-glycosylase (Kit-4440038, ABI, Lithuania). The PCRs were performed in triplicate in a MicroAmp Fast Optical 96-Well Reaction Plate with a barcode (0.1 mL; ABI, China) with an ABI 7500 Fast Real-Time PCR System. The specific conditions of the PCR were as follows: initial incubation for 10 min at 95°C, followed by 40 cycles of 95°C for 20 sec and 58°C for 60 sec. When the Ct value for GAPDH input for a sample was 34 or fewer cycles, this sample was considered CDKN2A SCNV informative. The specificity of the PCR was monitored through running the gel. Distilled water was used as a no-template control for each experiment.

We used genomic DNA from A549 cells containing no CDKN2A allele to dilute genomic DNA from RKO cells containing 2 wild-type CDKN2A alleles, and then we set the standard curve according to the relative copy number of the CDKN2A gene at different dilutions. The ΔCt value and relative copy number for the CDKN2A gene were calculated using the GAPDH gene as the internal reference. When the CDKN2A copy number in the A549-diluted template was consistently lower than that in the RKO control template and the difference was statistically significant (t test, p<0.05), it was judged that the lowest dilution concentration was the detection limit of CDKN2A deletion (the difference in CDKN2A copy number between the 100% RKO template and 80% RKO template spiked with 20% A549 DNA). When the CDKN2A relative copy number in a tissue sample was significantly lower or higher than that of the paired SM sample, the sample was defined as somatic CDKN2A CDR deletion-positive or amplification-positive, respectively. For each experiment, the 100% A549, 100% RKO, and 20% A549 + 80% RKO DNA mix controls were analyzed.

The copy number of CDKN2A exon-1β was detected using the primer and probe set ( Table 1 ) as previously reported (16). When the relative copy number of CDKN2A exon-1β in a tissue sample was significantly lower or higher than that of the paired SM sample, the sample was defined as somatic CDKN2A/P14^ARF deletion-positive or amplification-positive, respectively.

We used Meerkat to predict somatic SVs and their breakpoints in WGS datasets (accession numbers, EGAD00001004811 with 36× of sequencing depth) for gastric adenocarcinoma samples from 168 patients using the suggested parameters (14). This method used soft-clipped and split reads to identify candidate breakpoints, and precise breakpoints were refined by local alignments. CDKN2A deletion information of 157 GC samples was obtained from WGS datasets. We also estimated copy number profiling over 10 kb windows with Patchwork 28 and calculated the ratio of standardized average depth between normal tissue and tumor tissue (log2R ratio). The purity and ploidy of each tumor were calculated using ABSOLUTE software (17).

The CDKN2A allele homozygously deleted cell line A549 (kindly provided by Dr. Zhiqian Zhang of Peking University Cancer Hospital and Institute) was grown in RPMI-1640 medium, and the RKO cell line containing two wild-type CDKN2A alleles was purchased from American Type Culture Collection and grown in DMEM. The medium was supplemented with 10% (v/v) fetal bovine serum (FBS). These cell lines were tested and authenticated by Beijing JianLian Genes Technology Co., Ltd. before they were used in this study. A Goldeneye 20A STR Identifier PCR Amplification kit was used to analyze the STR patterns.

Chi-square or Fisher’s exact tests were used to compare the positive rates of CDKN2A SCND or amplification between different groups of tissue samples. Student’s t test was used to compare the proportion of the CDKN2A gene copy number between genomic DNA samples. All statistical tests were two-sided, and a p value less than 0.05 was considered to be statistically significant.

It has been previously reported that homozygous deletion of approximately 170 kilobase pairs (kb), including the CDKN2A locus, can be detected in human cancers by MSI analyses (18). SCND inactivates the CDKN2A gene in 273 human cancer cell lines according to the COSMIC dataset ( Data File 1 ). We found that an 8 kb estimated CDKN2A CDR could be detected among these cell lines by ordering “start” genomic coordinates of these breaking points ( Figure S2 ). To investigate the prevalence of CDRs within tumor suppressor genes in human cancer tissues with a high deletion frequency (1, 2), we further downloaded the estimated genomic coordinates for deletion fragments that overlapped with these genes. We found that CDRs could be detected not only within the CDKN2A gene ( Figure 1A ; approximately 17 kb) but also within the ATM, FAT1, RB1, PTEN, and miR31HG genes ( Figure 2 ; approximately 1232 kb, 50 kb, 12 kb, 33 kb, and 46 kb, respectively) ( Data Files 2-7 ). No CDR could be observed within CCSER1, FHIT, LRP1B, and WWOX genes according to the SNP-array data ( Data Files 8-11 ).

It was reported that the error in CDKN2A breakpoint estimation based on SNP-array data is approximately 10 kb (19). To characterize the true genomic coordinates of CDKN2A deletion fragments in cancers, we extracted base-resolution sequence information of interstitial CDKN2A deletions from available published articles and our sequencing data ( Data File 12 ) (20–29). We found a 5.1 kb CDR (chr9: 21,970,277 - 21,975,386, hg19) that spanned from the P16^INK4A promoter to intron-2 in 83 (90%) of 92 reported cancer cell lines or tissue samples containing interstitial CDKN2A deletions ( Figure 1B , blue lines). This CDR sequence is the same as the CDKN2A deletion fragment in the HCC193 lung cancer cell line (26). The CDR coordinates were also confirmed in our WGS datasets (average sequencing depth, 36×) of 18 (100%) of 18 GCs (14), in which interstitial CDKN2A deletions/fusions were identified ( Figure 1B , purple lines; Data File 12 ).

It is well known that germline CDKN2A inactivation can lead to a high predisposition for melanoma and pancreatic cancer (30–32). Interestingly, we found that 14 (93.3%) of 15 CDKN2A allelic variants in the Online Mendelian Inheritance in Man (OMIM) database are located within the CDR sequence, especially in CDKN2A exon-2 ( Figure S3 ) (33, 34).

In addition, both P16^INK4A and P14^ARF mRNAs are transcribed from the human CDKN2A gene at chromosome 9p21 but with different transcription start sites; they share the same exon-2 but have different translation reading frames. Because CDKN2A exon-2 located within the true CDR is the essential exon for coding P16^INK4A and P14^ARF proteins, the above findings indicate that P16^INK4A and P14^ARF are coinactivated in 87% (96/110) of human cancer cell lines and tissues containing CDKN2A CDR deletion ( Figure 1B ).

The current clinical method FISH for detecting SCND is composed of a set of biotin-labeled probes that should cover at least 50 kb DNA sequences. Thus, FISH is not a suitable method for detecting the copy number deletion of the 5.1 kb CDKN2A CDR. To provide a convenient method for routine clinical use, we designed and experimentally evaluated a set of multiplex quantitative PCR assays and finally optimized the CDKN2A CDR-specific quantitative multiplex PCR assay called P16-Light for detecting the copy number of a 129-bp amplicon within the CDKN2A intron-2 ( Figure 3A ), which covers 86% (94/110) of known CDKN2A deletion fragments ( Figure 1B , green line).

The copy number of the GAPDH gene was used as the internal reference. Genomic DNA from human A549 cells (with homozygous deletion of CDKN2A alleles) and RKO cells (with 2 wild-type CDKN2A alleles) were used as CDKN2A CDR deletion-positive and deletion-negative controls, respectively. The amplification efficiencies of the two amplicons in GAPDH and CDKN2A were very similar ( Figure 3B ). No template inhibition was observed when the amount of template DNA ranged from 10 to 0.63 ng ( Figure 3C ). The proportions of CDKN2A CDR copy number were linearly correlated with the ratios (0 - 100%) of RKO cell DNA and A549 cell DNA in the input mixtures (10 ng/reaction) when the A549 DNA was spiked in at different proportions for the P16-Light analyses ( Figure 3D ). Furthermore, there was a high reproducibility when DNA with homozygous deletion of CDKN2A was present in ≥20% of the cells verified in ten experimental repeats performed on different days ( Figure 3E ). Thus, when the proportion of CDKN2A copy number was significantly decreased (or increased) in a sample relative to the paired normal control (Student’s t test, p<0.05) in the P16-Light analyses, the sample was defined as CDKN2A SCND-positive (or amplification-positive).

As we described above, information on interstitial copy number deletion/fusion of the CDKN2A gene was extracted from WGS datasets for 156 of 168 GC patients enrolled in a GC genome study (14), and a total of 18 CDKN2A deletion/fusion coordinates at the base resolution were detected in 17 (10.8%) GCs ( Data Files 12, 13 ). To compare the performance of P16-Light with WGS, we analyzed the status of SCNVs, including SCND and amplification, of the CDKN2A gene in 156 of these GCs with enough genomic DNA samples with P16-Light using the paired SM sample as the diploid reference ( Data File 13 ). CDKN2A SCND and amplification were detected in 40 (25.6%) and 34 (21.8%) of these GCs, respectively. The P16-Light analysis was confirmed by the WGS results: the frequency of CDKN2A SCND (or amplification) by P16-Light was significantly higher (or lower) in 17 GCs containing interstitial CDKN2A deletion/fusion than in 139 GCs without interstitial CDKN2A deletion/fusion (chi-square test, p<0.05; Figure 4A ). These results also indicate that there is a significantly higher sensitivity for detecting CDKN2A SCND by the quantitative P16-Light assay than the hemi-quantitative WGS.

Moreover, it is well known that the proportion of cancer cells in tissue samples (i.e., sample purity) may affect the detection values of various genome data. To study whether the cancer cell proportion disturbs the detection of CDKN2A SCNVs, we calculated the cancer cell proportion in the above GC samples using WGS data ( Data File 13 ). We found that the difference in sample purity between GC subgroups with different CDKN2A SCNV statuses was not statistically significant (t test, p=0.075; Figure 4B ), although the proportion was slightly higher in GCs with CDKN2A SCND than in those without CDKN2A SCND. No correlation was observed between the proportion of cancer cells and the relative copy number of the CDKN2A gene among these GCs ( Figure 4C ).

The P14-qPCR assay was previously established for detecting the copy number of CDKN2A/P14^ARF exon-1β (16). Two amplicons in the P16-Light and P14-qPCR assays cover 98% (108/110) of known CDKN2A deletion fragments ( Figure 1B , red and green lines). Therefore, we further compared the performance of P16-Light, P14-qPCR, and their combination using GC and paired SM samples from patients who were recently included in the cross-sectional cohort in our association study (35). GC samples (n=139) with enough genomic DNA were used in P14-qPCR analysis ( Data File 14 ). The SCND-positive rate for P14^ARF was similar to that for the CDKN2A CDR (31.7% vs. 36.7%) ( Table 2 ). CDKN2A SCND was found only in 19 GCs by both assays. While CDKN2A CDR SCND by P16-Light was significantly associated with distant metastasis of GC (odds ratio=4.09, p<0.001), no association was observed between GC metastasis and P14^ARF SCND by P14-qPCR. Using merged CDKN2A SCND data (CDKN2A CDR SCND-positive and/or P14^ARF SCND-positive), only a weaker association was observed. These results suggest that individual P16-Light alone may be good enough for detecting CDKN2A SCND in tissue samples.