Background

Copy number variation (CNV) contributes greatly to the genetic variability in human populations. CNVs are a type of structural variation in the genome that vary in number and range in length from one kilobase to several megabases [1]. CNVs have been associated with many complex traits, including neurodevelopmental disorders [2, 3] and obesity [4,5,6].

Several research groups has described a CNV region on chromosome 10q11.22 [2, 5, 7,8,9,10,11] while, one study reports no CNV [12]. The region was initially reported to span almost 194 kb [13] across NPY4R, SYT15 and GPRIN2 genes (Fig. 1). This region is notorious for its complexity due to repeated elements and it has not been fully mapped in the current version of the human genome assembly (GRCh38). Most of the previous reports assumed that the normal copy number of NPY4R is 2 copies per genome [5, 9, 11,12,13] and describe the CNV as either gain or loss, not always specifying the exact copy number. Several studies have demonstrated that a CNV in this region is associated with differences in body weight [2, 5, 11, 13, 14].

Fig. 1
figure 1

Chromosomal positions of NPY4R and RPPH1 genes, and droplet digital PCR assay design. Green and blue arrows indicate primer positions, red lines indicate probe positions. The duplication region on chr10q11.22, as it is annotated in GRCh38 human genome assembly

Full size image

NPY4R encodes Y4, a receptor for pancreatic polypeptide (PP). Due to the fact that PP is involved in appetite regulation, [15, 16] structural or functional changes in NPY4R might be of great importance for energy homeostasis and body weight.

Here we investigate CNV of NPY4R gene in 69 modern human samples from the 1000 Genomes project and of three archaic hominins (two Neanderthals and one Denisovan), using well-established computational methods of CNV detection by read depth analysis [17,18,19,20]. For 19 of the 1000 Genomes samples, we also determined the copy number of NPY4R using the recently developed molecular droplet digital PCR (ddPCR) method [21]. Our results differ from previously published studies of the CNV of this complex genomic region in that we find more copies and more variation with both of these approaches than previously published studies. The differences between the copy number determination methods are discussed.

Methods

The aim of the study was to determine the copy number of the human NPY4R gene in samples from the 1000 Genomes database and to compare the computational read depth analysis methods with the ddPCR method in DNA from a subset of these samples. Finally, we performed read depth analysis of three archaic hominins.

Samples

We have studied 66 samples from the 1000 Genomes Project [22] (see Additional file 1) and three archaic hominins (two Neanderthals and one Denisovan).

The low coverage samples from phase 1 of the 1000 Genomes Project were downloaded from the public repository, (http://www.1000genomes.org, download data 2012-06-01), then the copy number of the CNV region spanning NPY4R was assessed using CNVrd2 read depth analysis method. The phase 3 sequencing data for the same samples from the 1000 Genomes Project (N = 66) was analysed using three different read depth analysis methods, namely CNVrd2 [17] (except for NA18940), Control-FREEC [23] and CNVnator [24]. We have also investigated the high-coverage genomes of two Neanderthals and one Denisovan for their copy number in the NPY4R region using the Control-FREEC method. Finally, ddPCR was performed on 18 samples from the 1000 Genomes Project.

Read depth analysis

CNVrd2

Chromosome 10 was extracted from the alignment for each sample and used for further analysis. The R-package CNVrd2 [17] was used to perform the actual read depth analysis and determine the copy number state of the CNV region spanning NPY4R for each sample. The results are summarized in Table 1.

Table 1 NPY4R CNV determined by read depth analyses
Full size table

Control-FREEC

In order to avoid spurious signals due to either the fragmentation in ancient DNA, differences in coverages or sequencing technologies, ancient samples were analyzed independently of each other and of the 1000 genomes data. BAM files mapped to the human reference GRCh37 were downloaded from the MPI-EVA, Leipzig (Denisovan [25]: http://cdna.eva.mpg.de/neandertal/altai/Denisovan/; Altai Neandertal [26]: http://cdna.eva.mpg.de/neandertal/altai/AltaiNeandertal/bam/; Vindija Neandertal [27], http://cdna.eva.mpg.de/neandertal/Vindija/bam/,). Control-FREEC was employed to call CNVs for the whole genome. First, samtools [28] was used to generate a mpileup file only including reads with mapping quality of at least 30. Then Control-FREEC was run with the following parameters: coefficientOfVariation = 0.05, breakPointThreshold = 0.6, breakPointType = 2. The results for the copy number state for the CNV region spanning NPY4R were extracted from the output. Non-integer values were obtained by multiplying the median ratio of local normalized read depth to global read depth with the ploidy (2).

CNVnator-based read depth

CNVnator [24] v0.3.3 was employed to investigate copy number state of the NPY4R region in 66 samples from phase 3 of the 1000 Genomes Project. We have scanned chromosome 10 (with a window size of 300 bp) for significant duplications overlapping a 200 kb CNV region that spans across the NPY4R. We have used chromosome 10 as a reference for copy number determination.

Droplet digital PCR

Assays were designed according to the guidelines from Bio-Rad Laboratories. The primers and probe for RPPH1 assay were: Forward: 5′-CGCGCGAGGTCAGACT-3′ Reverse: 5′- GGTACCTCACCTCAGCCATT-3′ Probe: 5′-(VIC)CCGGCGGATGCCTCCTT-3′. The primers and probe for NPY4R were: Forward: 5′- TGCATCCATTTGCATCG-3′ Reverse: 5′-CTGCAAGGCTTACTGTGCAC-3′ Probe: 5′- TCAGCTGTTTGTTCCTGGGAGAA(FAM)-3′. For primer and probe location see Fig. 1.

DNA was digested with BstXI restriction enzyme (10 U/μl, ThermoScientific, Cat#: ER1021) in buffer 0 for 1 h at 55 °C, followed by 20 min at 80 °C. A 22 μl mixture of 2 × ddPCR mastermix (Bio-Rad, Cat#: 186–3010), forward and reverse primers for target and reference assay (final concentrations of 900 nM each), probes for both assays (final concentrations of 250 nM each) and 15 ng of digested DNA was emulsified with Bio-Rad Droplet Generator Oil (Bio-Rad, Cat#: 186–3005) in a Bio-Rad QX100™ Droplet Generator (Bio-Rad, Cat#: 186–3001) according to the manufacturer’s instructions. The droplets were manually transferred to a 96-well plate (Eppendorf, Cat#: 951020362) and heat-sealed with Easy Pierce sealing foil sheets (Thermo Fisher Scientific, Cat#: AB-0757). Polymerase chain reaction was performed in a Bio-Rad C1000 thermal cycler (Bio-Rad, Cat#: 185–1197) with the following cycling parameters: 10 min at 95 °C (1 cycle), 30 s denaturation at 94 °C and 1 min annealing and extension at 58 °C (40 cycles), 10 min at 98 °C and a hold at 12 °C. All steps had a ramp rate of 2 °C/s. Droplets were analysed using a Bio-Rad QX100 Droplet Reader (Bio-Rad, Cat#: 186–3001). Fluorescent data from each well were analysed with QuantaSoft software (v1.3.2), where copy number was calculated based on Poisson distribution [29]. All DNA samples were run at least twice.

Data analysis

The degree of correlations between NPY4R copy number data generated by read depth methods and average values of ddPCR were calculated using Spearman correlation (for NPY4R copy number data see Additional file 1). Statistical analysis was performed using SPSS version 22.0.

Results

Using read depth analysis we have confirmed that NPY4R is located in a copy number variable region by analyzing 66 modern human samples from 1000 Genomes Project (for an example of a read depth output see Additional file 2: Figure S1, A). For simplicity, we will refer to the copy number of this CNV region as NPY4R copy number.

The copy number determined by the Control-FREEC and CNVnator methods ranged from three to seven among the 66 individuals with four as the most frequent copy number for the phase 3 data (Table 1). The results of the CNVrd2 method displayed no correlation with either the results of Control-FREEC or CNVnator analyses, nor with the ddPCR. The results of Control-FREEC and CNVnator displayed a statistically significant correlation but were not identical (Spearman’s ρ = 0.822, p < < 0.001) (Fig. 2a).

Fig. 2
figure 2

Results of read depth analyses of the NPY4R in 66 samples from the 1000 Genomes Project. Correlation between Control-FREEC and CNVnator results (all phase 3 data) (a). Results of ddPCR and read depth analyses of the NPY4R in 18 samples from the 1000 Genomes Project (b)

Full size image

In addition to the read depth analysis described above, we have also analyzed the copy number of NPY4R by ddPCR for a subset of 18 samples from 1000 Genomes Project (Table 2) (for an example of a ddPCR output see Additional file 2: Figure S1, B).

Table 2 NPY4R CNV determined by read depth analysis and ddPCR
Full size table

Using the ddPCR method we observed a variation from four to seven copies of NPY4R per genome. The most frequent copy number was four. Both Control-FREEC and CNVnator displayed a correlation with the ddPCR results (Spearman’s ρ = 0.777, p = 1.488*10−4 and Spearman’s ρ = 0.818, p = 3.377*10−5 respectively) (Fig. 2b).

In order to investigate the archaic copy number state, we have analysed the NPY4R copy number in two Neanderthal and one Denisovan genomes. Both Neanderthals and the Denisovan had four copies of NPY4R (Table 1).

Discussion

This study was designed to investigate the previously reported CNV for the NPY4R gene on chromosome 10q11.22 and study the range of the CNV. Earlier studies were contradictory with the majority of studies reporting CNV [2, 5, 7,8,9,10,11, 14] and one study describing no variation in this genomic region [12]. Here, we demonstrate that the NPY4R copy number varied depending on the analysis method and the quality of data. We had initially studied this CNV region using CNVrd2 on phase 1 data from the 1000 Genomes Project, which was the only data available at the time. Subsequently we have analysed the same samples from the 1000 Genomes Project with the CNVrd2 method, using the phase 3 sequencing data. The copy number determined by CNVrd2 in phase 1 and phase 3 data exhibit no correlation. The results of the CNVrd2 displayed no correlation with either the results of Control-FREEC, CNVnator or the ddPCR method, they were therefore considered to be untrustworthy (not shown). Using three independent methods (two types of read depth analysis and ddPCR) we demonstrate the presence of extensive CNV for NPY4R. Two read depth analyses (Control-FREEC and CNVnator) displayed a similar CNV range: three to seven copies, and ddPCR has shown a variation from four to seven copies (Table 1). We discuss below the methods for CNV determination and possible reasons for differences between reports of the CNV of this region.

Differences between CNV determination methods

Read depth analysis was performed using genome re-sequencing data, where reads are mapped back to the reference assembly, and the resulting read coverage in each individual is used to determine the copy number of the region. A limitation of this approach is that any comparison between individuals is relative to the reference assembly, and estimates of copy number will be dependent on the number of copies present in the reference. For regions where the quality of the assembly is low due to e.g. a high frequency of repeats, errors in the reference assembly will affect the possibilities to accurately determine variation relative to the reference genome. When more specifically analyzing genomic short-read re-sequencing data for read depth differences, there is naturally a varying degree of read coverage over the genome that will affect the ability to accurately call copy number differences. While analyzing the results it is also important to keep in mind that the region of interest is small in itself and that we have used a small set of samples. Read depth methods are usually good at determining whether two samples differ in copy number relative to the reference genome. However, the absolute number of copies can be challenging to determine, especially if the copy number included in the reference genome is inaccurate or does not represent the population average. We initially used CNVrd2 to analyse the phase 1 data of the 1000 Genomes Project, where each individual was only sequenced to a moderate coverage, and different sequencing technologies (Illumina, 454 and SOLiD) were used (www.internationalgenome.org/analysis), lowering the accuracy of read depth copy number assignment. We have repeated the same type of read depth analysis on phase 3 data that was generated using the Illumina sequencing platform, had a higher and more even coverage. Due to lack of correlation between CNVrd2 analysis of phase 1 and phase 3 data, and absence of correlation between CNVrd2 and either Control-FREEC, CNVnator (all phase 3) or ddPCR, we concluded that CNVrd2 was unsuitable for analysing such complex regions, unlike, the two commonly used read depth methods that agree with each other and ddPCR.

There was no systematic difference, neither numerical nor ratio-wise, i.e., neither method systematically over- or under-estimates the number of copies relative to the other method.

The ddPCR method has recently emerged as an accurate way for precise quantification of target nucleic acid [21, 29]. It has been used for absolute copy number determination [30,31,32] and was shown to be equal [33] or more reliable than other molecular methods for copy number determination [34, 35] and comparable to other digital PCR methods [36]. The drawbacks of ddPCR are high cost and labour intensity in relation to number of studied CNV regions. In contrast, read depth analysis can determine multiple CNV regions [18] and estimate their copy number [17, 19], provided of course that the whole genome sequence is available (which is far more costly than ddPCR for a specific gene).

The shortcomings of both methods can be overcome by using ddPCR for calibration of read depth CNV estimation based on already available whole genome sequencing (WGS) data. Several studies have evaluated CNV regions based on read depth analysis of WGS data and ddPCR. Although the results were not identical, they demonstrated high concordance [20, 30, 37, 38]. Two of these studies have demonstrated that the difference between ddPCR and read depth analysis becomes grater with higher copy number [37, 38]. Taking into account the precision of ddPCR, we suggest that among the three read depth methods used in this study Control-FREEC and CNVnator are more precise copy number determination tools, and that ddPCR (as an absolute copy number measurement tool) can be used for evaluation and calibration of read depth-based CNV analysis of the WGS data.

Our ddPCR assays generated accurate and replicable results for both 1000 Genomes Project samples and in-house control DNA samples (see Additional file 3: Figure S2, A for showing the low variability of the reference assay between ddPCR runs and Additional file 3: Figure S2, B for comparison between two different reference assays). It is unknown whether, or to what extent, the propagation of immortalised lymphoblastoid cell lines has affected the integrity of their genomes and specifically the copy number state of NPY4R. For the purpose of methodological comparison, it would have been optimal to use DNA extracted from fresh blood samples in order to avoid possible effects of cell culturing. Nevertheless, due to high correlation between the read depth and ddPCR copy number estimates, we think that it is unlikely that transformation and propagation of the cell lines has influenced the NPY4R copy number.

Accuracy of CNV determination is method-dependent

In comparison with previous studies that have reported CNV of the NPY4R gene, our analyses reveal both a generally higher copy number and greater variation. We find that the gene copy number ranges from three to seven (by Control-FREEC and CNVnator) (Table 1), and four to seven (by ddPCR) per genome (Table 2). Previously published studies assumed that the most common copy number of NPY4R is two copies per genome and did not involve accurate calibration to other genes [5, 9, 11,12,13]. That two copies was an incorrect estimate in most genomes was already obvious from SNP (single nucleotide polymorphism) frequency deviations from Hardy-Weinberg equilibrium observed in the 1000 Genomes Project samples (unpublished observations).

We have investigated the genomes of modern humans and archaic hominins in order to study the copy number state and time of the NPY4R gene duplication. As both of the investigated Neanderthal genomes and the Denisovan genome had four copies of NPY4R, which was also the most common copy number in modern humans (Table 1), we conclude that the NPY4R duplication took place before the split of modern humans from the Neanderthals and the Denisovans (400,000–800,00 years ago) [39].

One of the previous CNV studies could not detect any CNV in NPY4R region in young Chinese individuals [12]. We have analysed five individuals of Chinese origin using both of the read depth analysis methods as well as ddPCR, and we found NPY4R copy numbers either being equal to four (Control-FREEC) or ranging between four and five (CNVnator and ddPCR) (Table 2). Our results also agree with the most recent human genome assembly (GRCh38), where two copies of NPY4R are placed on the same chromosome, indicating that four copies per genome is the most likely normal copy number.

We suggest that the differences in NPY4R copy number described here and those reported in previous studies can primarily be explained by methodological differences (although, it is possible that population differences may also contribute). A common issue in CNV analyses is the need to make an assumption about the reference copy number. SNP-arrays, aCGH and RT-PCR-based methods require a reference copy number, which is most commonly set at two copies per genome [5, 9, 12, 13, 40]. We propose that incorrectly chosen reference copy number or inappropriate choice of reference gene in PCR-based copy number determination methods might be sources of error in CNV studies. We addressed this by performing three different kinds of read-depth analysis of chr10q11.22 region, which gave us a copy number relative to the rest of the chromosom 10 or the genome, as well as by using ddPCR, which gave us copy number relative to the reference gene. We observed that the most common copy number for NPY4R is four copies per genome. The haplotype in the current human genome assembly (GRCh38) shows two copies on the same chromosome, thus four copies per diploid genome. The analysis of CNVs in young Chinese individuals [12] was based on the assumption that the most common copy number for genes in the human genome is two and it used the VEGFA gene as a reference for copy number analysis by RT-PCR and presumed that this gene was a suitable reference. However, this gene has previously been shown to display CNV in an Asian cohort [9, 40, 41]. We have used RPPH1 as a reference gene, since it is a commonly used well-known single-copy gene [42, 43].

Most of the previous studies were based on SNP arrays and aCGH, methods [2, 5, 7, 9, 13] that depend heavily on relative fluorescence data quality [44]. Signal intensity fluctuations of SNP arrays may occur as a result of probe length, the GC content of the probes and SNP position in the probe [45, 46]. DNA sequence-specific complications may cause poor probe coverage of certain genome regions, poor reproducibility and elevated risk of false detection. Algorithms for data analysis also differ in their sensitivity to the inherent variation in relative fluorescence between genomic loci on SNP and hybridization arrays [45, 46]. CNV genotyping assays often have a few target-specific probes located far apart, which makes these assays difficult to use with standard statistical methods that rely on the association between closely spaced probes [44]. Current genotyping platforms have limited or no probe coverage for a large number of common CNVs [47] and limited power to detect CNVs in duplication-rich and repeat-rich genome regions, such as chr10q11.22.

Conclusions

In conclusion, our CNV study suggests that NPY4R varies in copy number and that the most common gene copy number is four per genome, not two as previously reported by other investigators. A comparative study would require many more individuals to draw conclusions at the population level and, especially, to investigate copy number differences between populations. Due to the CNV and the role of NPY4R and its ligand pancreatic polypeptide in the regulation of food intake, this gene is a strong candidate for contribution to body weight variation and obesity. However its exact role remains to be investigated, as the CNV in this region has shown both a positive and a negative correlation with BMI [5, 11, 13, 14]. We have demonstrated here that the quality of sequencing data plays a crucial role in read depth analysis and that methods for copy number determination can differ in precision. Based on multiple CNV studies [20, 30, 37, 38, 43, 48,49,50] as well as our own results we suggest that ddPCR is a reliable method for CNV determination that can be used to calibrate read depth analysis.