All experiments were performed in accordance with relevant guidelines and regulations. The use of hESCs has been reviewed and approved by the Central Ethical Review Board for Stem Cell Research at the Robert Koch Institute, Berlin, Germany (Az.3.04.02/0167). Human PSCs were maintained at 37°C with 5% CO₂ and 21% O₂ in specific media and matrix, as detailed in Supplementary Table S1. Cells were passaged at a 1:6 ratio using the appropriate passage method every 5 to 6 days. A list of hPSCs is provided in Supplementary Table S1.

Cells were treated with 0.04 μg/mL colcemid (Gibco by Thermo Fisher Scientific, Darmstadt, Germany) for 2 h. Single-cell suspensions were produced using Accutase (Sigma-Aldrich by Merck, Darmstadt, Germany) and incubated in a hypotonic solution (1:1 0.075M KCl:0.9% sodium citrate) for 60 min at 37°C. Cells were fixed with a 1:3 acetic acid:methanol solution. G-banding analysis was performed using Stem Genomics (Montpellier, France). At least 20 metaphases were structurally evaluated with a chromosome band resolution between 300 and 500 band levels.

Genomic DNA was extracted using the QIAamp DNA Blood Mini Kit (Qiagen, Hilden, Germany) and processed on the Global Screening Array v3.0 (GSAMD24v3-0, Illumina, Inc. San Diego) by LIFE & BRAIN GmbH (Bonn, Germany). SNP calling was performed using GenomeStudio V2.0.5 with a GenCall threshold of 0.2. CNV analysis was performed using cnvPartition 3.2.0.

There are different values and thresholds which are necessary to be mentioned using GenomeStudio to ensure quality of SNP array data and reliability of CNV detection.

The call rate represents the percentage of SNPs that are successfully assigned to a genotype or copy number state out of the total number of probes on the array. In the literature, a call rate between 95% and 98% is generally considered acceptable (Guo et al., 2014). For quality control of hPSCs in research, we recommend adopting the lower threshold of 95% as the goal is to detect large-scale chromosomal abnormalities rather than to perform high-resolution genotyping like in diagnostic settings. Commonly used CNV analysis tools remain robust at this level, and accepting a 95% threshold ensures a balance between data quality and resource efficiency in routine stem cell research.

The B-allele frequency (BAF) represents the genotype and is determined by the ratio of signal intensity from the B allele, which indicates the relative quantity of the one allele compared to the other. Homozygous SNPs exhibit BAFs close to 0 (AA) or 1 (BB), whereas heterozygous two-copy SNPs have BAFs near 0.5 (AB). Allelic imbalance leads to intermediate values. For instance, in a cell line with triploidy of a certain chromosome, an SNP is represented by three copies, which can have four possible genotypes (AAA, AAB, ABB, or BBB), resulting in BAFs of 0, 0.33, 0.66, or 1, respectively. In the case of a deletion, BAF shows values similar to those of the A/A and B/B genotypes, but no value for the A/B genotype (Attiyeh et al., 2009).

BAF drift quantifies deviations in the BAF from expected values (0, 0.5, or 1) in normal diploid regions. A drift of <0.01 is a benchmark for high-quality SNP data in sensitive analyses (Marenne et al., 2012; Rao et al., 2016; Taghizadeh et al., 2022). It can signal technical issues like poor DNA quality, sample degradation, or hybridization errors, leading to inaccurate genotyping. However, chromosomal aberrations can shift expected BAF positions, and GenomeStudio adjusts for these changes. In such cases, high BAF drift may indicate genomic instability, as observed in cancer cell lines, rather than technical flaws. Notably, GenomeStudio does not display BAF drift, making direct evaluation impossible.

The total fluorescence intensity is used to calculate the log R ratio (LRR). Each SNP probe on the array produces a signal intensity for the DNA sample, representing the observed intensity (R_observed) of both alleles (R_A + R_B). Additionally, a reference intensity R_expected is determined for each SNP, derived from a normalized dataset of many samples with a known, normal diploid genome. The LRR is calculated as follows: L R R = log 2 ⁡   R o b s e r v e d R e x p e c t e d .

The logarithmic transformation in LRR is used to normalize the data, making it easier to interpret copy number changes across a wide range of values. In a log₂ scale, each unit corresponds to a doubling or halving of the copy number. The LRR with a value of 0 represents the presence of two chromosomes [log₂ (1) = 0], representing the normal state. Elevated signal intensity of a region results in an increase in LRR, which represents copy number gain due to amplification or duplication [log₂ (2) = 1, representing doubling]. A decrease in signal intensity is regarded as deletion, resulting in an LRR of approximately −1 [log₂ (0.5) = −1, representing halving] for a hemizygous deletion with only one copy of a region, rather than the normal two copies (Peiffer et al., 2006).

The log R ratio standard deviation (LRR SD) is a key quality control metric in SNP array data analysis, automatically calculated using GenomeStudio. It estimates the overall noise in the data, where a lower LRR SD indicates higher data quality and reliability in detecting chromosomal aberrations. The typical threshold for LRR SD is 0.35. Samples below this value are considered to have good-quality data and are suitable for further analysis. For instance, an LRR SD below 0.2 is often desirable for precise and robust CNV calls. Samples with an LRR SD greater than 0.35 should be excluded from further analysis. Potential reasons for an elevated LRR SD include sample quality issues such as DNA degradation, contamination, or low DNA concentration; experimental factors like hybridization problems, batch effects, or array defects; and biological causes such as mosaicism or high variability in copy number across the genome, as observed in certain cancer cell lines (Bae et al., 2010; Marenne et al., 2012; Staaf et al., 2008).

The size of detected CNVs is an important value to ensure the quality of reported aberrations. Various studies have demonstrated the relationship between CNV size and detection accuracy for identifying genetic alterations. For example, the risk of false positives decreases as the CNV size increases. It is more difficult to distinguish smaller regions from noise or normal variations, whereby the analysis of larger regions is more significant and reliable (Canham et al., 2015). SNP array-based CNV calling algorithms, such as those used in cnvPartition, often miss CNVs smaller than 300 kb. The frequently cited threshold of 350 kb is based on a balance between sensitivity (the ability to detect true CNVs) and specificity (the ability to exclude false positives) (Lavrichenko et al., 2021; Soster et al., 2023; Zhang et al., 2014). CN-LOH is reported for regions of 1 Mb or larger because smaller regions have a higher risk of being false positive.

The CNV confidence value refers to a statistical measure that indicates the reliability or certainty of the detected CNV event and is crucial for distinguishing true CNVs from false positives. For instance, a confidence value above 70 for a called CNV is often considered to indicate a high probability that the CNV is real; scores below 70 may suggest a higher chance of the CNV being a false positive (Lin et al., 2011). We use a CNV confidence threshold of 70 or higher for our analysis. However, according to the literature, a threshold of 40 or higher is also commonly used, depending on the desired stringency of the analysis.

In the detection of CNVs, the minimum probe count defines the lowest number of consecutive array probes that must support a change in signal intensity for a CNV to be reported. This threshold helps distinguish true genomic alterations from background noise or technical variation. By requiring a minimum number of probes to indicate, for example, a duplication or deletion, the analysis software increases the reliability of CNV detection, particularly in regions of the genome where probe density or signal quality may vary. The cnvPartition plug-in uses a default minimum probe count of 3, but this threshold is adjustable and can be increased. In the literature, the minimum probe count for CNV detection is often set between 3 and 7 probes (Mason-Suares et al., 2013; Metzger et al., 2013; Urnikyte et al., 2016; Zhou et al., 2022). However, in hPSC research or in a more stringent context such as diagnostic application, thresholds of 10 or more probes are commonly used, as the risk of false-positive CNV calls increases significantly with lower probe counts (Laurent et al., 2011; Lin et al., 2013; Rajagopalan et al., 2020). In line with this, we applied a minimum probe count of 10 as part of the quality control process for our hPSC lines.

Figure 1 represents all steps necessary for SNP array analysis using GenomeStudio to detect chromosomal aberrations.

The BAF and LRR plots are essential for detecting chromosomal aberrations via the SNP array, showing specific patterns based on the genotype (BAF) and signal intensity (LRR) (Figure 2A). A normal, diploid set of chromosomes shows clear BAF clustering at 0 (AA), 0.5 (AB), and 1 (BB), representing a typical heterozygous SNP pattern. The LRR plot shows a horizontal line of signal clusters around 0, indicating two copies of the chromosome.

For a hemizygous deletion (one copy loss), the BAF plot lacks the heterozygous signal cluster at 0.5 (no heterozygous SNPs) and shows only two signal clusters around 0 and 1, representing homozygous AA or BB genotypes. The LRR plot shows a negative shift in the LRR (around −0.5 to −1), indicating a reduction in the copy number.

In duplications (one extra copy), the BAF plot displays signal clusters between 0 and 0.5 (at 0.33) and between 0.5 and 1 (at 0.66), reflecting the presence of three possible alleles in four genotypes (AAA, AAB, ABB, and BBB), and the LRR plot shows a positive shift (around +0.3 to +0.5), indicating an increase in the copy number.

Finally, CN-LOH, either due to uniparental disomy or chromosomal loss followed by duplication of the remaining copy, shows a complete loss of heterozygosity in the BAF plot, showing only two signal clusters: 0.0 (AA) and 1.0 (BB). There is no middle cluster at 0.5 (AB), which would indicate heterozygosity. The LRR plot remains close to zero because there is no change in the overall copy number. The region is still diploid (two copies of the chromosome), so no gain or loss of genetic material is reflected in LRR.

It is important to use both, BAF and LRR plots, when analyzing chromosomal aberrations because events like a deletion and a CN-LOH can appear identical in the BAF plot but differ in the LRR plot.

We analyzed a total of 40 different hPSC lines for chromosomal integrity using the SNP array (Supplementary Table S1) and detected chromosomal aberrations in nine independent lines (10.9%) (Table 1).

The gain of 20q11.21 is one of the most common aberrations in iPSCs and was detected in three independent iPS cell lines: ZIPi015-K, GM24581, and STBCi101-A. The aberration most likely appeared during cell cultivation (Amps et al., 2011). Amplification of 20q11.21 is visible in the BAF plot as data points clustering around 0.33 and 0.66, along with an increase in signal intensity at this position in the LRR plot, shown for iPSC line GM24581 (Figure 2B, left panel). The same aberration was detected in one clonal line after reprogramming (DSMZi002-C-11), whereas the parental fibroblast culture (M85705) did not show this amplification (Figure 2B, right panel).

Another iPSC line (DSMZi017-A), used in a study for DNA methylation analysis, was derived from a patient with Angelman syndrome, characterized by a large deletion on the maternal chromosome 15q11–q13 (Chamberlain et al., 2010). This deletion was confirmed through SNP array analysis, which revealed a hemizygous deletion in the 15q11–q13 region (Figure 2C, left panel). The BAF plot displayed an absence of signals at the 0.5 value, whereas the LRR plot showed a clear reduction in the signal intensity in the same region. In addition, the SNP array identified a large region of copy-neutral loss of heterozygosity (CN-LOH) on the long arm of chromosome 9 (approximately 75 Mbp). This LOH was evident through a loss of signal in the BAF plot, whereas the LRR plot remained unchanged in this area (Figure 2C, middle panel). The detected CN-LOH probably appeared during cell culture due to the high passage number of the cell line. Additionally, a small gain was detected on chromosome 10q11.22, indicated by clusters of data points at 0.33 and 0.66 in the BAF plot and an increase in signal intensity in the LRR plot (Figure 2C, right panel). As the cell line was reprogrammed several years ago (Chamberlain et al., 2010), no parental material is available for comparison of the detected chromosomal aberration. Numerous gains of 10q11.22 have been reported in the DGV. The CNV gssvG4494, which spans the detected region (chr10:47892511…48055660), is reported with a population frequency of 99.84% (Figure 2D). This strongly suggests that the observed aberration was already present in the parental cell line as a normal genomic variant.

SNP array analysis of the human embryonic stem cell line H9 revealed a gain of 7q11.21, along with a gain in chromosomal region 14q23.3 (Table 1) and a loss of chromosomal region 16q11.2. Although gains of entire chromosomes 7 and 14 are commonly reported in hPSC lines, the specific regions described here have not been explicitly noted in these reports (Baker et al., 2016; Baker et al., 2007). No accordance of the specific positions of the observed gains and loss were found by comparison with the DGV, which could indicate that the aberrations occur during cell cultivation.

The iPSC line RBi001-A exhibited a CN-LOH at chromosome region 4q33–q34.3 (Table 1). Due to the lack of reference material, it remains unclear whether the detected aberration originated during cell culture or was already present in the parental cell line. Notably, CN-LOH at this locus has not been frequently reported in the literature as a common occurrence during hPSC culturing.

In the iPSC line WTSIi021-A, deletions were detected in regions 7q31.1 and 7q31.33 (Table 1). Because deletions affecting parts of the q-arm of chromosome 7 have been frequently described in the literature, we assume the deletion to be an adaptation to cell culture conditions (Andrews et al., 2022).

Cell line ZIPi015-K underwent two rounds of CRISPR/Cas9 gene editing, targeting chromosome 9 (ZIPi015-K-1) and subsequently chromosome 15 (Haake et al., 2024). A clonal cell line of the chromosome 9 editing, with no additional chromosomal aberrations than the 20q11.21 gain, which was already present in the parental cell line, was selected for gene editing, targeting chromosome 15. Among eight clonal cell lines screened post-editing, one exhibited a CN-LOH of the whole q-arm of the targeted chromosome 15 (ZIPi015-K-2-C-3) (Table 1).

Cell line DSMZi017-A was also used for two rounds of CRISPR/Cas9 gene editing at chromosomes 9 (DSMZi017-A-1) and 15, followed by quality control including SNP array analysis and G-banding. Four clonal lines (DSMZi017-A-2-A3/-A8/-C12/-F4) were generated and analyzed via G-banding and SNP array within the quality control of establishing a master cell bank. Although the SNP array data did not show any abnormalities (Figure 3A left panel), G-banding analysis of clone DSMZi017-A-2-A8 exhibited two distinct cell populations: one with a normal karyotype, detected in 11 out of 20 metaphases, and another with an Xp isochromosome, detected in 9 out of 20 analyzed metaphases (Figure 3B). Isochromosomes are classified as structural chromosomal aberrations that are caused by improper segregation during cell division. This missegregation results in an unbalanced chromosomal complement, with the duplication of genetic material from one arm and the loss of material from the other. This results in a chromosome with two identical arms; in the case of an Xp isochromosome, both arms are short arms (p arms) (Chadwick, 2020; McFeely, 1993; Mertens et al., 1994). As cells for G-banding analysis were harvested 13 passages later than for the DNA of the master cell bank samples, we assumed that acquisition of the Xp isochromosome occurred during this extended time of culture. Indeed, it was detected via SNP array in a sample of genomic DNA prepared nine passages after the preparation of the master cell bank (Figure 3A, right panel). In the SNP array, an Xp isochromosome appears as duplication in the BAF plot (signal clusters at 0.33 and 0.66) and a positive shift in the LRR plot, reflecting extra p-arm copies (Figure 3C). In contrast, the region of the plot corresponding to the Xq arm would resemble a deletion in the BAF plot, characterized by the absence of signals at 0.5, and an increase in the LRR plot. For an Xq isochromosome, these patterns are reversed (Figure 3C). In the sample analyzed, the BAF plot of the Xp arm showed the expected pattern for duplication, with clusters around 0.33 and 0.66. However, for the Xq arm, instead of the clean deletion profile observed in a non-mosaic cell population carrying the Xp isochromosome, additional signal intensities around 0.25 and 0.75 were observed. This suggests a mixed cell population, as detected via G-banding, where the presence of both, normal cells and cells with the Xp isochromosome, leads to intermediate BAF values due to partial allelic imbalance (Eberhardt et al., 2023; Glessner et al., 2021; Conlin et al., 2010).