The glutathione S-transferase mu 1 (GSTM1) gene is located in the 1p13.3 band and codes for the cytosolic enzyme GST-μ that plays a role in carcinogen detoxification. Many structural variations have been described that overlap this gene in the Database of Genomic Variation 1 (Figure 1). A common copy number variant (CNV) has been well characterized and reported on the basis of the frequency of homozygous deletions of the entire coding region (known as the GSTM1-null genotype), which varies between 29% and 51% across ethnic groups 234.

Due to the established role of GSTM1 in detoxification and the high frequency of homozygous deletions in the population, GSTM1 has been extensively investigated in association with many chronic diseases, in particular, with asthma 5 and different types of cancers 46, including bladder cancer 7. In a large case–control study, we reported that 63% of bladder cancer cases and 47% of controls harbored the GSTM1-null genotype, leading to an OR = 1.7 (95%CI 1.4 – 2.0) 8. This is a well-established association as it was replicated in several independent studies and confirmed in a meta-analysis 8.

Reliable and accurate technologies such as qPCR or Multiplex Ligation-dependent Probe Amplification (MLPA) are available to genotype CNVs in targeted regions of the genome. High-throughput SNP-array platforms now offer the possibility to explore CNVs at a genome-wide scale. For instance, Illumina Infinium 1 M provides intensity data of both alleles at each SNP allowing the detection of CNV breakpoints and the estimation of the associated number of copies. Monomorphic probes were also included in genomic regions known to harbor CNVs but that were not well covered by SNPs. Overall, this platform contains 1,071,820 probes, among them 206,665 are located in reported CNV regions and 17,202 are monomorphic probes specially designed for CNV purpose.

Assessing CNV data across the genome-wide level using SNP-arrays is a daunting problem and requires stringent quality control (QC) measures to minimize the noise related to an analytical scheme that relies on sliding windows across SNPs 91011. The assessment also requires that CNVs is called based on the raw intensity data (Log-R ratio, LRR) from each probe. Several calling algorithms, such as PennCNV, have been developed 9101213 but these algorithms have a very low sensitivity when applied to large data sets 141516. In turn, the probability of false negatives remains a major challenge.

To limit false negative callings, a strategy that by-passes the calling step and directly performs the association test using LRR measures has been proposed 171819. Alternatively, methods have been developed allowing the simultaneous calling and association test estimation. These methods account better for the calling uncertainties but are yet to be validated in sufficiently large studies 2021.

The objective of this study was to compare the assessment of a CNV at the GSTM1 locus and its association with bladder cancer by applying LRR and PenCNV to data derived from the Illumina Infinium 1 M SNP-array platform with that derived from TaqMan (qPCR) and MLPA genotyping in subjects included in the Spanish Bladder Cancer/EPICURO (SBC/EPICURO) Study.

The concordance rate between TaqMan and MLPA for GSTM1-null identification was high but not complete: 96.2% over the entire sample (cases and controls) and 95.8% when considering only the controls. TaqMan and MLPA detected 402 (52.96%) and 401 (52.83%) controls carrying a homozygous deletion, respectively, and 289 controls (38.1%, both platforms) carrying a heterozygous deletion (Additional file 1 Table S2). These values were consistent with the reported rate of deletion in the European population. In contrast, when we conducted an analysis with PennCNV on the Illumina 1M SNP-array data, no deletion was detected among the 759 controls and 773 cases.

As shown in Figure 2, the average LRR values for the 5 Illumina probes located at the GSTM1 locus was higher than expected based on the number of copies reported by TaqMan or MLPA. Indeed, for the individuals with two copies according to both TaqMan and MLPA, the LRR for the five probes was 0.17 on average, higher than 0, the value expected when 2 copies of the gene are present. Similarly, although negative LRR values are expected when a deletion is present, we observed positive values (0.08) in individuals with one copy and only slightly negative values (−0.10) in individuals with a homozygous deletion according to TaqMan and MLPA callings. These differences between the observed pattern of LRR at the GSTM1 locus and the one expected by the PennCNV algorithm could explain that no detection was detected by PennCNV.

However, there was a strong correlation between LRR values and the estimated number of copies using TaqMan or MLPA (analysis of variance - p < 10^-15), suggesting that the LRR values might be useful as an exploratory approach (Figure 2). Using LRR values from each of the five probes as explanatory variables in the logistic regression models, we detected the association with bladder cancer risk for three of them: (rs4147567: p = 8.00 × 10^-4; rs2239892: p = 0.01 and rs12068997: p = 0.02) (Table 1). Estimated OR at the 5 SNPs (Table 1) was <1.0 indicating that an increase of the LRR was protective for bladder cancer, which was concordant with the evidence in the literature. Indeed, applying the callings provided by TaqMan and MLPA in this sample, we obtained a significant association (p = 3.40 × 10^-4, OR = 0.74 [0.62-0.87] and p = 1.15 × 10^-4, OR = 0.72 [0.61-0.85], respectively) between the trend on the number of copies at GSTM1 and bladder cancer risk.

The association was performed by logistic regression models adjusted for age, gender, region and smoking status.

When using CNVtools to test for association, individuals were clustered into two categories based on their copy numbers. These categories fitted well with their actual status of having a homozygous deletion or not as detected by MLPA or TaqMan. We observed a high concordance rate, 93.7% and 93.0%, between CNVtools clustering and TaqMan and MLPA callings across the overall sample set. An association signal was detected (OR = 0.66, p = 1.74 × 10^-3) using this method adjusting for age, gender, region and tobacco use, replicating the known association between GSTM1 and bladder cancer.

Genome structural variants and, particularly CNV, are thought to play an important role in phenotypic variation and in the development of many complex diseases. In the last few years, several calling algorithms have been developed to identify CNVs at the whole genome scale using the same SNP-chips used to perform GWAS. However, studies that have evaluated the available tools have concluded that they lack sensitivity leading to a large number of false negative callings 141516. While PennCNV algorithm was found to be the one performing the best in previous comparisons, here we demonstrate the lack of sensitivity of PennCNV in a particular scenario. In the well-characterized region of GSTM1, we found that PennCNV did not detect any deletion in a large sample of cases with bladder cancer and controls where homozygous deletion was known to have a frequency of 50% using Taqman and MLPA technologies. Because PennCNV was designed to identify unknown CNV regions, we also applied the cnvHap algorithm that was designed to genotype known CNV regions 24. As expected, cnvHap did not detect any deletion in the GSTM1 region in our sample, either. It is noteworthy the fact that, using CNV calls derived from Illumina 1 M platform, GSTM1 would have never been associated with bladder cancer. However, when individual probe LRR values are compared between cases and controls, the association can be detected and provide results similar to those obtained when using Taqman or MLPA. This observation clearly shows that PennCNV lacks sensitivity to detect CNV in the GSTM1 region.

A possible explanation for the lack of sensitivity of PennCNV (and cnvHap) is the high frequency of the GSTM1 deletion in the studied population. Indeed, CNV calling is done using the LRR that depends both on the observed (R_obs) and the expected (R_exp) R values. The R_exp is determined based on the clusters of genotypes. In the case of GSTM1 where the homozygous deletion is very frequent, these clusters include a high number of subjects with a homozygous deletion (GSTM1-null genotype). Thus, R_obs and R_exp are expected to be similar in a GSTM1-null individual and, accordingly, the LRR value is around 0. The normalization process could also play a role as it aims at finding three clusters and this is not possible for GSTM1 locus since the BAF of homozygous deleted sample is uniformly distributed between 0 and 1, thus normalization is affecting the intensity values, too.

The fact that the association between GSTM1 CNV and bladder cancer can be detected with LRR values without applying a calling CNV confirms the utility of this measure as a complementary screening strategy to test for association at the genome-wide level, as already suggested 171819. Indeed, the LRR is a continuous measure that approximates and correlates well with the actual discrete number of copies. Nonetheless, it is affected by the noise contained in the intensity measurement of both alleles obtained through the hybridization experiments. Thus, using LRR in the association test may decrease the power of some probes in detecting the association in comparison of using an accurate calling of the discrete number of copies. This loss of power would explain that two of the five probes located in the GSTM1 locus failed to show association with bladder cancer risk in our study, and that the three significant probes only showed a moderate significant p-values (between 8x10^-4 and 0.019, Table 1). Nevertheless, even if the significance of these probes was moderate, we observed an excess of significant p-values in comparison to what we could expect under the null hypothesis of no association in that region. Thus, methods working at the genome-wide level and searching for regions with an excess of significant probes could have identified the GSTM1 region in our study. Alternatively, CNVtools, performing a joint calling and association testing, might also be considered, though it is more difficult to apply than that based on LRR and takes longer to run. The main caveat with CNVtools and equivalent methods is the definition of regions of interest.

The GSTM1 deletion is located in a region of high sequence homology neighbored by a segmental duplication and this might explain that its breakpoints may slightly vary and, thus, the difficulties of calling. However, the locus is defined since the deletion in GSTM1 was already known and approaches based in probes are able to identify it. Nevertheless, there might be other still unknown CNVs in the genome showing similar characteristics that might thus not be easy to call 25. To increase sensitivity in CNV identification at the whole genome scale, we propose performing a genome-wide screen for association using LRR values at each probe and then applying CNVtools for a fine-tuning in the most promising regions.

In conclusion, our study provides insights into the limitations of CNV-calling algorithms applied to SNP-array platforms in regions harboring common CNVs, especially those with full gene deletions. Though our results focused on a previously characterized CNV, they raise the possibility that there could be a substantial problem across unknown regions of the genome with common CNVs. On the other hand, we showed that by comparing LRR between cases and controls we were able to identify hot genomic areas associated with the trait of interest, supporting the use of this exploratory association assessment at the whole-genome level, which should be pursued with promising efficient calling algorithms.

We declare we have no competing interest.

GM, SJC, NM and EG participated in the design of the study and drafted the manuscript. GM performed the statistical analysis. FXR, NR, BR, LP, MK, MG and DTS are involved in the SBCS study and help to draft the manuscript. All authors read and approved the final manuscript.

Co-last authors: Núria Malats, Emmanuelle Génin and Stephen J Chanock