Mitochondrial DNA integrity and metabolome profile are preserved in the human induced pluripotent stem cell reference line KOLF2.1J

Summary

Quality control of human induced pluripotent stem cells (iPSCs) is critical to ensure reproducibility of research. Recently, KOLF2.1J was characterized and published as a male iPSC reference line to study neurological disorders. Emerging evidence suggests potential negative effects of mtDNA mutations, but its integrity was not analyzed in the original publication. To assess mtDNA integrity, we conducted a targeted mtDNA analysis followed by untargeted metabolomics analysis. We found that KOLF2.1J mtDNA integrity was intact at the time of publication and is still preserved in the commercially distributed cell line. In addition, the basal KOLF2.1J metabolome profile was similar to that of the two commercially available iPSC lines IMR90 and iPSC12, but clearly distinct from an in-house-generated ERCC6^R683X/R683X iPSC line modeling Cockayne syndrome. Conclusively, we validate KOLF2.1J as a reference iPSC line, and encourage scientists to conduct mtDNA analysis and unbiased metabolomics whenever feasible.

Highlights

• •The mtDNA of the KOLF2.1J reference line is intact
• •The metabolomic signature of the KOLF2.1J reference line is inconspicuous
• •We recommend assessing mtDNA integrity of iPSC lines
• •ERCC6^R683X/R683X Cockayne syndrome model line shows an altered metabolomic signature

Rossi and colleagues analyze the mtDNA and metabolomic signature of the KOLF2.1J reference line using various sequencing techniques and untargeted metabolomics. The KOLF2.1J line displays intact mtDNA and an inconspicuous metabolomic profile. In contrast, a genetically modified Cockayne syndrome iPSC model line shows an altered metabolomic signature.

Introduction

Human induced pluripotent stem cells (iPSCs) have revolutionized basic research and regenerative medicine (Hirschi et al., 2014). Analysis of their genomic integrity is important to guarantee reproducibility (Rossi et al., 2022; Volpato and Webber, 2020). The male-derived iPSC line KOLF2.1J was recently suggested as a reference for Alzheimer disease modeling (Pantazis et al., 2022). The authors performed deep genotyping, ruling out the presence of detrimental nuclear genome variants (Pantazis et al., 2022). Although several studies reported associations of mtDNA mutations of human iPSCs that could affect stability, functionality, and differentiation potential (Prigione et al., 2011; Wei et al., 2021), mtDNA integrity was originally not assessed.

Mitochondria play key roles in the maintenance of stemness and differentiation (Lisowski et al., 2018). Specific mtDNA single nucleotide variants (SNVs) are disease associated (Chinnery and Gomez-Duran, 2018). SNVs affecting all mtDNA copies (homoplasmy) or only a fraction (heteroplasmy) can cause diseases in ∼0.02% of all individuals. Different mtDNA profiles also likely contribute to the observed heterogeneity among iPSC clones (Deuse et al., 2019; Hämäläinen, 2016; Perales-Clemente et al., 2016; Prigione et al., 2011; Wei et al., 2021). Given this impact, mtDNA analysis is recommended to guarantee the overall integrity of cell lines (Rossi et al., 2022).

Consequently, we decided to assess KOLF2.1J mtDNA integrity at the time of publication and in the distributed commercially available clones. We further performed untargeted metabolomics to rule out aberrant metabolism of the globally distributed cell line.

Results

mtDNA integrity analysis

First, we analyzed whole-genome sequencing (WGS) data previously generated (Pantazis et al., 2022) (Figure 1, left). Second, we used targeted (Figure 1, right) and WGS Oxford Nanopore Technologies (ONT) long-read sequencing of KOLF2.1J cultured in our laboratory.

Figure 1.

Overview of the 2 workflows used to analyze mtDNA integrity of the KOLF2.1J reference cell line

(Left) WGS data were obtained from a publicly available repository. After rigorous data preprocessing, reads corresponding to mtDNA were extracted and realigned to a canonical and a shifted mtDNA reference genome. Finally, SNV and indel mutation analysis was performed using Mutect2 and Mutserve. (Right) For analysis using long-reads sequencing, cells were lysed and DNA extracted. Afterward, mtDNA was enriched by performing 9 overlapping PCR reactions, and a sequencing library was prepared. Long-read sequencing data were generated using a minION M1kC with a Flongle adaptor. Raw data were basecalled and analyzed using Mutserve.

SNVs and insertion/deletion mutations (indels) of short-read WGS data were analyzed using Mutserve (Weissensteiner et al., 2021) and Mutect2 (Laricchia et al., 2022). This is challenging because low mutation frequencies require clear distinction from contamination, sequencing errors, and erratic alignment to nuclear sequences of mitochondrial origin (Albayrak et al., 2016; Li et al., 2012; Maude et al., 2019; Wei et al., 2022). The circular mtDNA further complicates the alignment at the artificial breakage point to the revised Cambridge Reference Sequence (rCRS, GenBank: NC_012920). To account for these features, we used rigorous sequencing data processing based on the Mutect2 mitochondria workflow (Laricchia et al., 2022). Briefly, we mapped the WGS data to the GRCh38 reference genome using the Burrows-Wheeler alignment (Li and Durbin, 2009), filtered duplicates, extracted mtDNA reads, and mapped a reversed FASTQ file derived from this processing to the rCRS while accounting for the artificial reference breakage point. To ensure the absence of residual genomic DNA, we used haplocheck (Weissensteiner et al., 2021). Indel calling was performed using Mutect2 and variant calling using Mutect2 and Mutserve at a variant calling threshold of 0.01. Variants were checked for entries in the Mitomap disease variant database (www.mitomap.org). Two homoplasmic indels were residing in short tandem repeats (m310T>TC, m513G>GCACA) and thus classified as irrelevant. Other regions prone for sequencing errors (Laricchia et al., 2022) were also neglected for analysis.

Overall, 38 (Mutect2) and 39 (Mutserve) variants were detected (Figure 2A; Table S1). The vast majority of variants (Mutect2: 33; Mutserve: 34) were annotated as known polymorphisms. Of the remaining 5, 2 were annotated as synonymous and 2 as benign variants. One unknown variant was located at position m3650 in the coding sequence of the MT-ND1 gene. This variant was found to be synonymous with no predicted impact on translation.

Figure 2.

Analysis of KOLF2.1J mtDNA integrity

(A) SNV detection of the mtDNA of the iPSC line KOLF2.1J using Mutserve. The x axis represents the mtDNA nucleotide position. The y axis represents the heteroplasmy level. Of the identified 39 SNVs, 34 were known common polymorphisms (green dots). Of the 5 additional SNVs (pink dots), 4 were described as not disease associated. These were located in MT-ATP6, MT-ND3, MT-ND4, and MT-ND5. The last SNV was unknown and was located in MT-ND1. Manual inspection revealed the variant to be synonymous. This analysis was performed on WGS data obtained from a public repository.

(B) SNV detection of the mtDNA of the iPSC line KOLF2.1J using Mutserve. Of the identified 42 SNVs, 37 were known common polymorphisms (green dots). Of the 5 additional SNVs (pink dots), 4 were described as not disease associated. These 4 SNVs were located in MT-ATP6, MT-ND3, MT-ND4, and MT-ND5. The last SNV was located in an artifact-prone poly C stretch and subsequently identified as a false hit. This analysis was performed on sequencing data obtained by in-house long-read ONT sequencing based on a custom pipeline.

Next, we purchased the KOLF2.1J cell line and performed mtDNA analysis using both ONT WGS and targeted mtDNA sequencing to determine whether mtDNA integrity remains intact in distributed cell lines. SNVs were detected using Mutserve at a variant calling threshold of 0.05 (targeted) or 0.01 (WGS). This approach identified 37 SNVs for the ONT WGS data (Figure S1; Table S2) and 42 SNVs for the targeted approach (Figure 2B; Table S3) in line with the original WGS data. Of these 42, 37 were annotated as known polymorphisms. Of the remaining 5, 4 were already detected in the original WGS dataset. The remaining 1 resided in an artifact-prone poly C stretch and was a false positive hit. Read phasing of long ONT reads on the variants called with high-quality WGS data revealed the presence of only 1 clear haplotype because the vast majority of detected variants were homozygous (data not shown).

Metabolome profile analysis

To corroborate these findings, we assessed the basal metabolome profile of KOLF2.1J and compared it to established iPSCs (IMR90, iPSC12) and an in-house-generated Cockayne syndrome disease model (ERCC6^R683X/R683X) iPSC line (Sanchez-Roman et al., 2018). Notably, principal-component analysis (PCA) analysis revealed a cluster of iPSC12, IMR90, and KOLF2.1J that was clearly distinguishable from that of the ERCC6^R683X/R683X line (Figure 3A). This was also reflected by the high number of significantly differently abundant metabolites detected for ERCC6^R683X/R683X compared to the other lines (Figure 3B), whereas comparison of KOLF2.1J with iPSC12 and IMR90 revealed only a low number of differently abundant metabolites (Figure S2). The majority of identified metabolites accounting for the differences of ERCC6^R683X/R683X compared to the other iPSCs was of the superclass organic acids (Figure 3C), including amino acids (Figure 3D), in line with altered cellular metabolism in Cockayne syndrome patients (Sanchez-Roman et al., 2018). Accordingly, we found the amino acids ornithine, proline, and glycine (Figure 3E) to be less abundant in the ERCC6^R683X/R683X disease line compared to the isogenic iPSC12, the IMR90, and the KOLF2.1J reference line. Although the definition of a “healthy” basal iPSC metabolome is beyond the scope of the present study, these results may serve as an indication.

Figure 3.

Analysis of untargeted metabolomics data

(A) PCA indicates similarity between analyzed cell lines. Although all of the commercially available human iPSC lines (KOLF2.1J, iPSC12, IMR90) cluster together, the disease iPSC line ERCC6^R683X/R683X is found in a different cluster strongly separated on the first principal component.

(B) Volcano plot showing differentially abundant metabolite features in the basal metabolome of the 3 commercially available iPSC lines (KOLF2.1J, iPSC12, IMR90) compared to the disease cell line ERCC6^R683X/R683X.

(C) Organic acids make up the majority of the detected metabolites.

(D) Amino acids are differentially abundant in ERCC6^R683X/R683X compared to healthy iPSC lines.

(E) Detailed view on selected amino acids with lower abundance in ERCC6^R683X/R683X compared to healthy iPSC lines. Comparison of differentially abundant metabolites in (B) indicates high similarity of metabolomes between the 3 commercially available established iPSC lines compared to a disease model iPSC line.

Discussion

By combining short- and long-read WGS and targeted ONT sequencing data analysis, we performed an in-depth characterization of the KOLF2.1J mtDNA integrity at the time of publication and in the cell line in use in our laboratory. We found no detrimental disease-associated mtDNA SNVs or indels. Although very low yet potentially detrimental heteroplasmy may exist under the detection limit, the high (short-read) WGS coverage depth of ∼300× for most mtDNA positions gives us high confidence in our analysis. Further confidence is provided by our targeted ONT sequencing analysis. Both approaches showed equal performance, supporting the notion that expensive WGS data generation is not strictly necessary when focusing on mtDNA in specific use cases. Furthermore, untargeted metabolomic analysis of KOLF2.1J revealed that the abundance of detected metabolites was in line with regularly used iPSC lines and distinct from that of a disease iPSC line with altered metabolism. Although our experimental setup is not suited to define a “healthy” iPSC profile, the presented data can serve as a starting point to raise awareness of the value of metabolomics data in the stem cell field (Table S4). In summary, these data support the usability of KOLF2.1J as a male-derived reference human iPSC line with respect to mtDNA integrity and metabolism.

These findings highlight that to ensure the absence of disease-related mtDNA mutations, researchers should monitor the mtDNA integrity of iPSC lines whenever feasible (Rossi et al., 2022). Since WGS data generation yields data on mtDNA as a by-product, no additional major effort is necessary to assess mtDNA integrity in such cases. Nevertheless, monitoring mtDNA remains a nontrivial task demanding careful evaluation. The emphasis for future endeavors should be on developing user-friendly software that allows for a cost-effective evaluation of mtDNA integrity. This will facilitate the inclusion of mtDNA analysis as part of the standard quality control measures for newly produced or engineered iPSCs. To some extent, we tackled this issue by conducting a cost-effective experiment using ONT sequencing to examine the mtDNA. It must be noted, however, that variant calling using sequencing data from long-read ONT sequencing needs to be performed carefully. Here, when using FASTQ files directly from the sequencing device (high accuracy mode), Mutserve detected a m625G>A transition located in the tRNA MT-TF at a heteroplasmy level of >0.9. Manual inspection of the binary alignment map (BAM) file actually indicated a 0.5/0.5 G/A alignment. One contributing factor to this problem is the suboptimal performance of Mutserve for noisy sequencing data. In light of the documented detrimental effects of this mutation on cellular fitness (Sudo et al., 2011), we conducted an analysis of m625 using Sanger sequencing (Figure 4A). In contrast to the sequence derived from ONT sequencing, we found no alteration at the position of interest (Figure 4B). This highlights the necessity for a pipeline facilitating precise variant detection via cost-effective long-read ONT sequencing specifically designed for users who possess minimal knowledge of sequence analysis.

Figure 4.

Analysis of m625 by Sanger sequencing

(A) To analyze the suspicious position m625, the respective part of the mtDNA was amplified by PCR. Afterward, the PCR product was purified and sent for Sanger sequencing using 2 internal primers.

(B) Sanger sequencing chromatograms using forward (top) and reverse (bottom) primers of KOLF2.1J surrounding the potentially detrimental mutation. Inspection of the chromatograms revealed that the sequence is not altered and the variant calling failed at this position. This warrants careful analysis of mitochondrial DNA mutations. Red arrowheads point to the positions of interest.

Collectively, we extend an invitation to scientists to explore the possibility of mtDNA integrity analysis to assess iPSC quality, particularly when WGS data are accessible, or by using cost-effective ONT sequencing devices. Based on our experience, the targeted ONT sequencing described in this paper yields reliable results down to a heteroplasmy level of 0.05. To correctly represent such a population within a 90% confidence interval, we recommend >6,000 reads for a given position. For homoplasmic variants, as little as 40 reads are sufficient to correctly represent their frequency. This number is likely to further decrease with new releases of the ONT sequencing chemistry. Although read phasing did not yield additional value in the case of the KOLF2.1J line, in general, we recommend performing such an analysis to monitor cosegregation of potentially harmful variants.

In conclusion, the KOLF2.1J iPSC line displayed intact mtDNA integrity at the time of publication, still maintains mtDNA integrity and inconspicuous metabolism, and can thus indeed be used as a reference iPSC line.

Experimental procedures

Resource availability

Further requests should be directed to the corresponding authors jochen.dobner@iuf-duesseldorf.de and andrea.rossi@iuf-duesseldorf.de.

Materials availability

The ERCC6^R683X/R683X iPSC line can be obtained via material transfer agreement.

Data and code availability

Details on the code to reproduce the results can be found at Mendeley Data (https://doi.org/10.17632/n4mccs2zfg.2).

Original short-read WGS data analysis

WGS data were downloaded from the Alzheimer’s Disease Workbench (https://www.alzheimersdata.org/ad-workbench) and preprocessed according to the guidelines of the genome analysis toolkit (GATK; Van der Auwera et al., 2013). Reads from FASTQ files were aligned to the GRCh38 reference genome using bwa mem -K 100000000 -p -v 3 -t 24. Duplicates were marked (gatk MarkDuplicates), and reads mapped to the mtDNA extracted (gatk PrintReads) using the read filters MateOnSameContigOrNoMappedMateReadFilter, MateUnmappedAndUnmappedReadFilter, and NotDuplicateReadFilter. The aligned BAM file was reverted to FASTQ via generation of an unaligned BAM file (gatk RevertSam, gatk SamToFastq), and realigned to either the reference mtDNA genome or a mtDNA reference genome shifted to move the breakage point to the middle of the sequence (bwa mem -K 100000000 -p -v 3 -t 24). Afterward, aligned BAM files were merged with unaligned BAM files (gatk MergeBamAlignment), duplicates filtered (gatk MarkDuplicates and PrintReads) with NotDuplicateReadFilter, and variants called with Mutect2 (gatk Mutect2) on either the control region spanning the breakage point (8,025–9,144 in the shifted reference mtDNA) or the remaining sequence (576–16,024 in the original reference mtDNA). Variant calls from the shifted reference were then lifted over to the correct position numbering using R (R Foundation for Statistical Computing) and the results were merged (gatk MergeVcfs). Finally, calls were filtered (gatk FilterMutectCalls) using -mitochondria mode true -max-alt-allele-count 2 -min-allele-fraction 0.01 and variants left aligned and trimmed to improve visualization (gatk LeftAlignAndTrimVariants). For Mutserve, the bwa-aligned BAM file was used after filtering out duplicates and extracting mtDNA-aligned reads. SNV detection was carried out using default settings and a detection threshold of 0.01. Afterward, the data were processed and visualized using a custom R script.

Targeted long-read ONT sequencing analysis

After total DNA isolation, we enriched for mtDNA by PCR to generate nine overlapping amplicons (Ramos et al., 2009; Zascavage et al., 2019) and sequenced using an ONT minION M1kC device with a Flongle adaptor, as previously described (Nguyen et al., 2022) (ONT). Briefly, pooled PCR products were used to prepare the sequencing library using the Sequencing by Ligation Kit (SQK-LSK109 and NBD104 native barcoding kit) according to the manufacturer’s instructions (ONT). The final library was loaded at a total of 15 fmol on a Flongle flow cell and sequenced for 24 h. Postsequencing, FAST5 files were basecalled with Guppy basecaller (ONT) and the following settings: --min_qscore 10 --require_barcodes_both_ends --enable_trim_barcodes --barcode_kits "EXP-NBD104" -c dna_r9.4.1_450bps_sup.cfg. FASTQ files were concatenated into single FASTQ files and aligned to the rCRS using minimap2 (Li, 2018). Variant calling was performed with Mutserve at a threshold level of 0.05. Afterward, data were processed using a custom R script and visualized using ggplot2 (Wickham, 2016). Sanger sequencing was performed on a PCR amplicon (forward primer: 5′-CTGTATCCGACATCTGGTTCCT-3′, reverse primer: 5′-GTTTAGCTCAGAGCGGTCAAGT-3′) by Microsynth using internal custom forward and reverse primers (forward primer: 5′-CCCTAACACCAGCCTAACCA-3′, reverse primer: 5′-AGGGTGAACTCACTGGAACG-3′).

Untargeted metabolomics

Untargeted metabolomics was conducted using small adaptations of an established protocol for the analysis of regulatory myeloid immune cells by ultra-high performance liquid chromatography-time-of-flight-mass spectrometry (Baumann et al., 2020). Details regarding this specialized experiment can be found in the supplemental information.

Published: February 22, 2024

Supplemental information

Table S1. Identified and annotated SNVs of short-read WGS, related to Figure 2

SNV information on the KOLF2.1J reference line derived from short-read WGS. Detected with Mutect2 mitochondria mode and Mutserve.

Table S2. Identified and annotated SNVs of long-read nanopore WGS, related to the mtDNA analysis Results section

SNV information on the KOLF2.1J reference line derived from long-read WGS. Detected with Mutserve.

Table S3. Identified and annotated SNVs of targeted long-read nanopore amplicon sequencing, related to the mtDNA analysis Results section

SNV information on the KOLF2.1J reference line derived from targeted long-read mtDNA sequencing. Detected with Mutserve.