Feeder-free culture of human pluripotent stem cells drives MDM4-mediated gain of chromosome 1q

Abstract

Culture-acquired variants in human pluripotent stem cells (hPSCs) hinder their applications in research and clinic. However, the mechanisms that underpin selection of variants remain unclear. Here, through analysis of comprehensive karyotyping datasets from over 23,000 hPSC cultures of more than 1,500 lines, we explored how culture conditions shape variant selection. Strikingly, we identified an association of chromosome 1q gains with feeder-free cultures and noted a rise in its prevalence in recent years, coinciding with increased usage of feeder-free regimens. Competition experiments of multiple isogenic lines with and without a chromosome 1q gain confirmed that 1q variants have an advantage in feeder-free (E8/vitronectin), but not feeder-based, culture. Mechanistically, we show that overexpression of MDM4, located on chromosome 1q, drives variants' advantage in E8/vitronectin by alleviating genome damage-induced apoptosis, which is lower in feeder-based conditions. Our study explains condition-dependent patterns of hPSC aberrations and offers insights into the mechanisms of variant selection.

Keywords: MDM4; culture conditions; genetic changes; genome damage; human pluripotent stem cells.

Graphical abstract

Figure 1

Retrospective analysis of aberrant karyotypes and conditions for culturing hPSCs (A) Karyotyping data from two independent centers (WiCell and Centre for Stem Cell Biology [CSCB]) were annotated for different parameters and analyzed to ascertain a possible association of chromosomal aberrations with culture conditions. (B) Percentage of hPSC cultures containing cells with abnormal karyotypes is similar between WiCell and CSCB datasets. ns, non-significant; Fisher’s exact test. (C) The breakdown of abnormal karyotypes according to the type of abnormality. ns, non-significant; ^∗∗∗∗p < 0.0001; Fisher’s exact test. (D) The majority of aberrations involving both gains and losses of chromosomal material are isochromosomes. ns, non-significant; Fisher’s Exact test. (E) The frequency of gains/losses of each cytoband across all abnormal karyotypes in the WiCell dataset. (F) The frequency of gains/losses of each cytoband across all abnormal karyotypes in the CSCB dataset. See also Tables S1, S2, and Figures S1–S5.

Figure 2

Changes in patterns of recurrent aberrations over time (A) The relative frequency of the six most common abnormalities in the WiCell dataset over time. (B) The relative frequency of the four most common abnormalities in the CSCB dataset over time. (C) The proportion of abnormal karyotypes containing one of the six most common abnormalities in the WiCell dataset sampled in the year 2009 versus the year 2021. ns, non-significant; ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001, ^∗∗∗∗p < 0.0001; Fisher’s exact test. (D) The proportion of abnormal karyotypes containing one of the four most common abnormalities in the CSCB dataset sampled in 2002–2007 versus 2017–2019 (several years are pooled together due to relatively low n numbers in this dataset). ns, non-significant; ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗∗p < 0.0001; Fisher’s exact test.

Figure 3

Gains in chromosome 1q are associated with KOSR-free conditions (A) Timeline of the introduction of popular hPSC culture conditions over the past 25 years beginning with predominantly KOSR-based systems and transitioning into KOSR-free culture regimens. CSCB started using predominantly KOSR-free systems in 2015. (B) The relative frequency of the most used media in the WiCell dataset over time. The use of KOSR-based media had decreased over time. (C) The percentage of abnormal karyotypes from the cultures using the most represented media across the WiCell dataset. ^∗p = 0.0253; Fisher’s exact test. (D) The percentage of abnormal karyotypes from the cultures using the most represented matrices across the WiCell dataset. ns, non-significant; Fisher’s exact test. (E) The percentage of the most common abnormalities in the WiCell dataset across (i) all media (5,650 abnormalities), (ii) KOSR-based media (470 abnormalities), (iii) E8 (1,279 abnormalities), (iv) mTESR (2,295 abnormalities), (v) NutriStem (256 abnormalities), and (vi) StemFlex (391 abnormalities). Less frequent but recurrent aberrations (above 1% of abnormal karyotypes) are indicated in shades of gray. In comparison to KOSR-based medium, the frequency of 1q gain is elevated in E8 (^∗∗∗∗p = 7.6 × 10⁻¹⁵), mTESR (^∗∗∗∗p = 4.5 × 10⁻⁶), NutriStem (^∗∗∗∗p = 8.9 × 10⁻¹⁷), and StemFlex (^∗∗p = 0.0014); Fisher’s exact test.

Figure 4

v1q have selective advantage in KOSR-free but not KOSR-based conditions (A) A panel of wild-type and v1q sublines across four genetic backgrounds (H7, H9, MIFF3, and WLS-1C) used in this study. (B) Selective advantage was tested by mixing ∼10% v1q with their wild-type counterparts, with either of the lines being fluorescently labeled. Mixed cells were plated into either KOSR/MEF or E8/VTN, and the ratio of variants was monitored over subsequent passages. (C) v1q overtake wild-type cells rapidly in E8/VTN but not in KOSR/MEF. Data shown are the mean ± SD of three independent experiments. ns, non-significant; ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001; ^∗∗∗∗p < 0.0001; two-way ANOVA. (D) v1q have a significantly higher growth rate than wild-type cells in E8/VTN but not KOSR/MEF. Data shown are the mean ± SD of three independent experiments. ns, non-significant; ^∗∗p < 0.01, ^∗∗∗∗p < 0.0001; two-way ANOVA. (E) v1q have a significantly higher cloning efficiency than wild-type cells in E8/VTN but not KOSR/MEF. Data shown are the mean ± SD of three independent experiments. ns, non-significant; ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001, ^∗∗∗∗p < 0.0001. two-way ANOVA, Fisher’s least significant difference. (F) Wild-type and v1q hPSCs display similar levels of expression of a marker of undifferentiated state, SSEA3, in E8/VTN and in KOSR/MEF conditions. (G) Lineage trees tracked from time-lapse images of wild-type cells (upper two panels) and v1q (lower two panels). Red crosses indicate cell death. Gray-shaded area indicates the first 24 h post-plating when the cells were grown in the presence of Y-27632, required for single-cell passaging. (H) v1q show a trend toward a faster cell-cycle time compared to wild-type counterparts. Data points indicate 114 and 286 divisions for wild-type and v1q cells, respectively, from two independent experiments. (I) Percentage of cell fate outcomes of daughter cells following cell division, with SS denoting survival of both daughter cells, SD survival of one and death of the other daughter cell, and DD death of both daughter cells. (J) v1q have decreased levels of cleaved caspase-3 marker of apoptosis. Data shown are the mean ± SD of three independent experiments. ^∗∗∗p < 0.001; two-way ANOVA followed by Holm-Sidak’s multiple comparison test. See also Video S1 and Figure S6.

Figure 5

MDM4 overexpression provides selective advantage to v1q (A) The minimal region on chromosome 1q32.1 identified from the karyotyping datasets in this study. (B) The minimal region on chromosome 1q32.1 identified from overlaying the karyotyping datasets in this study, SNP array data from MIFF3 v1q used in this study, and data published by (Merkle et al., 2022). The minimal amplicon contains 13 genes expressed in hPSCs (blue boxes), which were compared based on their essentiality scores (indicated in italics underneath the genes; for comparison, an essentiality score for POU5F1 is −1.85). MDM4 (red) is a candidate of interest, based on its known role in p53 signaling and cancer. (C) RNA-seq analysis of H7 and H9 v1q versus wild-type sublines revealed differential expression of the p53 pathway. (D) MDM4 expression is increased in v1q. Western blot analysis of H7, H9, and MIFF3 wild-type and v1q cells. β-actin was used as a loading control. (E) Knockdown of MDM4 with small interfering RNA (siRNA) in v1q was confirmed by quantitative PCR. siRNA for Renilla Luciferase (siRNA LUC) was used as a negative control. Data shown are the mean ± SD of three independent experiments. ^∗∗∗p < 0.001; Student’s t test. (F) MDM4 knockdown suppresses cloning efficiency of v1q. Data shown are the mean ± SD of three independent experiments. ns, non-significant, ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001; One-way ANOVA followed by Holm-Sidak’s multiple comparison test. (G) Overexpression of MDM4 in wild-type hPSCs. Western blot analysis of MIFF3, MIFF3 v1q, and MIFF3 wild-type cells overexpressing MDM4 (wild-type-MDM4). β-actin was used as a loading control. (H) MDM4 overexpression provides selective advantage to wild-type hPSCs. Mixed cultures were analyzed at seeding and after three passages. Data shown are the mean ± SD of three independent experiments. ns, non-significant, ^∗∗∗∗p < 0.001; One-way ANOVA followed by Holm-Sidak’s multiple comparison test. (I) MDM4 overexpressing and v1q cells have lower percentage of cleaved caspase-3 marker of apoptosis compared to wild-type counterparts. Data shown are the mean ± SD of three independent experiments. ns, non-significant, ^∗∗p < 0.01; one-way ANOVA followed by Holm-Sidak’s multiple comparison test.

Figure 6

KOSR-based medium, rather than MEFs, diminishes advantage of v1q cells in KOSR/MEF compared to E8/VTN (A) Schematic representation of experiments aimed at revealing the impact of matrix (VTN or MEFs) versus medium (E8 or KOSR) on the selective advantage of v1q. (B) Similar numbers of MIFF3 wild-type and v1q hPSCs attach regardless of the medium (E8 and KOSR) or the matrix (VTN and MEFs) at 2 h post-plating. Data shown are the mean ± SD of three independent experiments. ns, non-significant; Unpaired t test. (C) Cell area at 2 h post-plating is similar between MIFF3 wild-type and v1q hPSCs in each of the conditions tested, indicating similar ability of cell to attach post-plating. Circles represent the cell area of individual cells measured at 2 h post-plating. For each test condition, 30–50 cells were measured across three independent experiments. The line represents the mean of all measurements. ns, non-significant, Mann-Whitney test. (D) Quantification of the number of phosphorylated focal adhesion kinase (pFAK)-positive focal adhesions. Circles represent the cell area of individual cells measured at 2 h post-plating. For each test condition, 21–32 cells were measured across three independent experiments. The line represents the mean of all measurements. ns, non-significant, Mann-Whitney test. (E) Representative images of pFAK (green) in MIFF3 wild-type and v1q hPSCs plated in different media (E8 and KOSR) or matrices (VTN and MEFs) at 2 h post-plating. Nuclei are counterstained with Hoechst 33342. Scale bar: 25 μm. (F) Numbers of MIFF3 v1q hPSCs are higher than wild-type cells in E8/VTN but are not significantly different from wild-type cells in other test conditions at 48 h post-plating. ns, non-significant, ^∗p < 0.05; Unpaired t test. See also Figures S7A and S7B.

Figure 7

Variant 1q and MDM4-overexpressing cells are less sensitive to DNA double-strand breaks, which are more abundant in hPSCs grown in E8/VTN compared to KOSR/MEF (A) Western blot analysis of MDM4 in MIFF3 and MIFF3 v1q cells under different conditions. MDM4 abundance is reduced in KOSR/MEF. Also, the addition of KOSR to E8/VTN reduces MDM4 expression. β-actin was used as a loading control. (B) The addition of KOSR to E8/VTN reduces growth rates of v1q to levels like those of wild-type cells. Data shown are the mean ± SD of three independent experiments. ns, non-significant; ^∗∗p < 0.01; two-way ANOVA with Holm-Sidak’s multiple comparison test. (C) MDM4 has a more nuclear and punctate localization in hPSCs grown in E8/VTN compared to KOSR/MEF. Representative images of MIFF3 hPSCs grown in E8/VTN (upper panels) or KOSR/MEF (lower panels) stained with antibodies against MDM4 and OCT4 (POU5F1). The nuclei are counterstained with Hoechst 33342. Scale bar: 50 μm. (D) Quantification of the MDM4 nuclear expression in E8/VTN versus KOSR/MEF condition in MIFF3 line. Data shown are the mean ± SD of three independent experiments. ^∗∗p < 0.01; Unpaired t test. (E) Genome damage is heightened in E8/VTN compared to KOSR/MEF condition. Quantification of γH2AX marker of double-strand breaks in hPSCs grown in E8/VTN compared to KOSR/MEF. Data shown are the mean ± SD of three independent experiments. ^∗p < 0.05; paired t test. (F) Cells with a gain of chromosome 1q or overexpressing MDM4 are more resistant to genome damage-induced apoptosis. Quantification of cell numbers of wild-type, v1q and wild-type-MDM4 cells treated with 10 μM camptothecin (CPT) for 2 h. Data shown are normalized to untreated control and represent the mean ± SD of three independent experiments. ^∗∗∗p < 0.001; One-way ANOVA. (G) A model summarizing a differential competitive advantage of v1q in E8/VTN versus KOSR/MEF conditions. E8/VTN confers high levels of genome damage in hPSCs. Amplification of MDM4 through the gain of chromosome 1q bestows v1q cells with the resistance to genome damage-induced cell death. Consequently, v1q outcompete wild-type hPSCs in E8/VTN. The KOSR/MEF condition does not generate the same selective pressure as the levels of genome damage are reduced compared to E8/VTN. The shift from feeder-based to feeder-free conditions over the last two decades has contributed to an increase in frequency of chromosome 1q gains detected in hPSC cultures. See also Figures S7C–S7E.