Mitochondrial DNA 3243A>G heteroplasmy is associated with changes in cytoskeletal protein expression and cell mechanics

Abstract

Mitochondrial and mechanical alterations in cells have both been shown to be hallmarks of human disease. However, little research has endeavoured to establish connections between these two essential features of cells in both functional and dysfunctional situations. In this work, we hypothesized that a specific genetic alteration in mitochondrial function known to cause human disease would trigger changes in cell mechanics. Using a previously characterized set of mitochondrial cybrid cell lines, we examined the relationship between heteroplasmy for the mitochondrial DNA (mtDNA) 3243A>G mutation, the cell cytoskeleton, and resulting cellular mechanical properties. We found that cells with increasing mitochondrial dysfunction markedly differed from one another in gene expression and protein production of various co-regulated cytoskeletal elements. The intracellular positioning and organization of actin also differed across cell lines. To explore the relationship between these changes and cell mechanics, we then measured cellular mechanical properties using atomic force microscopy and found that cell stiffness correlated with gene expression data for known determinants of cell mechanics, γ-actin, α-actinin and filamin A. This work points towards a mechanism linking mitochondrial genetics to single-cell mechanical properties. The transcriptional and structural regulation of cytoskeletal components by mitochondrial function may explain why energetic and mechanical alterations often coexist in clinical conditions.

1. Introduction

Clinical diseases are often studied at the cellular level in order to elicit greater knowledge about their underlying mechanisms. Current trends in cell-based research tend to fall into one of two groups. Biologists tend to consider that metabolic and mitochondrial dysfunction are primary aspects of cellular dysfunction, while bioengineers are largely focused on how pathology manifests as alterations in cell mechanics.

A wealth of literature now suggests that cellular mitochondrial and mechanical dysfunction are in fact related. Structural studies have mostly focused on the association between mitochondria and microtubules [1], but other work suggests that mitochondria are structurally connected to actin filaments [2] and that filament breakdown impairs mitochondrial motility [3]. Previous studies further suggest that mitochondrial subcellular localization contributes to cell mechanics. For example, mitochondrial subcellular localization is regulated during cell migration [4,5], and mitochondria at the cell periphery have a higher membrane potential in some cell types [6], where traction forces are usually highest [7]. These observations collectively suggest a functional connection between mitochondria and cellular mechanical activity.

Additionally, both acute and chronic cellular pathology are often characterized by concurrent alterations in cell mechanics and mitochondrial function. Anoikis, or cell death caused by detachment from the substrate, activates pro-apoptotic mitochondrial proteins [8], whereas engagement of substrate-attaching integrins promotes expression of anti-apoptotic ones [9]. Integrin inhibition also triggers the release of mitochondrial reactive oxygen species [10], which may further contribute to cell death. In addition to acute cellular toxicity, clinical pathologies often demonstrate parallel mitochondrial and mechanical dysfunctions at the cellular level. Cancer cells exhibit the well-known glycolytic Warburg effect [11], are softer [12] and generate weaker traction forces [7] than normal cells. Metabolic alterations are also pervasive in endothelial and cardiomyocyte cells from patients with cardiovascular disease [13], and these cells tend to be less flexible than cells from healthy patients [14]. Thus, mitochondrial and mechanical alterations frequently occur in parallel under pathological conditions. However, the directionality and causality of this relationship remain unclear.

In this work, we studied a unique cellular model of localized mitochondrial dysfunction, caused by increasing levels of a mitochondrial DNA (mtDNA) point mutation in the cell cytoplasm. We focused on the mtDNA 3243A>G mutation, the most common pathogenic mtDNA defect that ultimately results in electron transport chain deficiency [15]. Because the cytoplasm contains hundreds of mtDNA copies, normal (3243A) and mutant (3243G) mtDNAs can coexist within a single cell in a state termed heteroplasmy. MtDNA 3243A>G heteroplasmy profoundly affects clinical presentation, with low levels (<30%) associated with diabetes and deafness [16], higher levels (50–90%) causing mitochondrial encephalomyopathy, lactic acidosis and stroke-like episodes (MELAS syndrome) [17] and extremely high levels causing Leigh syndrome and early death in infancy [18]. We previously showed that selectively increasing mtDNA 3243A>G heteroplasmy perturbed mitochondrial gene expression and resulting bioenergetics and additionally had a profound, nonmonotonic influence on a variety of nuclear encoded genes [19]. Since these cell lines are distinguished from one another exclusively on the basis of mtDNA heteroplasmy level, this experimental paradigm is ideal for studying the effects of an isolated mitochondrial defect on cytoskeletal mechanics. Thus, we conducted further experiments on the cybrid cell lines to test the hypothesis that mtDNA mutations will breed alterations in the cellular cytoskeletal gene expression, structure and ultimately cell mechanics.

2. Material and methods

2.1. Generation of transmitochondrial cybrids

Cytoplasmic hybrid cell lines (cybrids) having 0, 20, 30, 50, 60, 90 and 100% mtDNA 3243A>G heteroplasmy were generated as described previously [19]; see electronic supplementary material for more information.

2.2. Measurement of mitochondrial DNA heteroplasmy

Heteroplasmy levels were monitored by two methods: (1) polymerase chain reaction amplification of the 264 nt fragment, digestion with HaeIII (New England Biolabs #R0108) and separation of the fragments initially by agarose gel electrophoresis and densitometric analysis and subsequently by capillary electrophoresis on an Genetic Analyzer 3500 automated sequencer (Applied Biosciences) [20]; and (2) by sequencing the entire mtDNA using next generation sequencing on an Ion Personal Genome Machine (PGM) using Ion PGM Sequencing 200 v. 2 kit (Life Technologies, Grand Island, NY). Data were analysed using the Ion Torrent Suite v. 4.0.1 and NextGENe software v. 2.3.3 and A/G heteroplasmy levels at nucleotide position 3243 were verified. Results were in the expected range and are given in greater detail in our earlier work [19].

2.3. Gene expression measurements

Transcript levels were determined by RNA sequencing (RNA-Seq) as previously described [19]. Total RNA was extracted with Trizol and depleted from ribosomal units (RiboMinus, Life Technologies #10837-08). RNA was quality checked and quantified on a Bioanalyzer 2100, RNA 6000 Nano kit (Eukaryote Total RNA Nano, Agilent Technologies) and Qubit 2.0 fluorometer RNA assay kit (Molecular Probes #Q32852). Five hundred nanograms of RNA was used for cDNA library preparation using Ambion, Total RNA-Seq kit #4445374.

RNA from all seven transmitochondrial cybrid cell lines plus the ρ₀ parental cell line were sequenced on the ABI SOLiD 5500 platform. Barcoded triplicates were sequenced on the same slide in different lanes. All sequencing was performed using paired-end chemistry of 50 (forward) × 35 (reverse) base pairs. Sequenced reads were mapped and processed as previously described [21].

Data were further analysed using Partek Genomics Suite software, v. 6.12.0109. Out of a total of 22 449 uniquely annotated genes for which transcripts were detected, 15 652 were differentially expressed (ANOVA model, p < 0.0001). Data were normalized and expressed as reads per kilobase per million (RPKM) reads. Normalizing data using R package DESeq yielded similar results. Reads for each gene were extracted, averaged among triplicates and expressed relative to 0% heteroplasmy. While gene expression was measured in triplicate, error bars were extremely small for all data points, so they are omitted from both graphs in figure 1.

Figure 1.

Cytoskeletal gene expression varies in the 3243A>G cybrid cell lines. (a) RNAseq data showing absolute cumulative gene expression of various cytoskeletal proteins in each of the eight cell lines. (b) Data normalized to the 0% cell line. Gene expression was measured in triplicate, but error bars are omitted in the graphs since they are too small to be easily resolved. ρ₀ cells lack any mitochondrial DNA, but contain the same nuclear genome as the other cell lines.

2.4. Cell culture

Cells were cultured in T25 and T75 flasks containing Dulbecco's modified Eagle medium high glucose buffer (Life Technologies) supplemented with 10% fetal bovine serum (Sigma, St Louis, MO), 1% non-essential amino acids (Life Technologies) and 50 µg ml⁻¹ uridine (Sigma). Cells were generally passaged at approximately 80% confluency using 0.25% trypsin (Life Technologies). All experiments were performed within several passages, when the cells had been growing in culture for approximately 48 h, except gene expression which was measured previously on earlier passage numbers [19].

2.5. Protein isolation and western blotting

Western blotting experiments were performed in duplicate. Cells were grown in T75 flasks and detached using 0.25% trypsin and centrifuged twice in culture media with the supernatant discarded. The cell pellet was lysed in RIPA buffer (Bio-World, Dublin, OH) supplemented with protease inhibitors (Roche, Manheim, Germany) and then centrifuged at 14 000×g to remove the insoluble pellet. The protein concentration of each cell line was determined using a Pierce BCA Protein Assay Kit (Thermo Scientific, Waltham, MA) as per the manufacturer's instructions. Standards and samples were appropriately diluted and pipetted into a Costar 96-well plate (Corning) in triplicate. Absorbance at 562 nm was measured using a Biotek Synergy H1 microplate reader and Gen5 software (Biotek, Winooski, VT), and sample protein concentrations were extracted from the resulting standard curve.

For western blotting, 10 µg total protein was separated on 4–15% SDS-PAGE gels (Biorad, Hercules, CA) and transferred to polyvinylidene difluoride membranes (Millipore, Billerica, MA). Membranes were blocked with 2.5% bovine serum albumin (Sigma) in TBS with Tween-20 solution (TBST) for 1 h at room temperature and incubated with primary antibodies (1 : 1000) overnight on a rocker at 4°C. After washing three times with TBST, membranes were incubated with secondary antibody (1 : 5000) and washed three times with TBST. Protein–antibody complexes were detected using the Amersham ECL Western Blotting Analysis System (GE Healthcare, Piscataway, NJ) according to manufacturer's instructions and exposed using a Kodak Image Station 4000 MM Pro Digital Imaging System and Carestream MI software (Carestream, Inc., Rochester, NY). When two proteins of similar molecular weights were of interest in the same blot, the membrane was stripped using Restore western blot stripping buffer (Life Technologies). Protein isolation followed by western blotting was performed in duplicate for each protein of interest.

Primary antibodies used for blotting and immunofluorescence included anti-α-actinin 1 isoform (Abcam, Cambridge, MA, #18061), anti-pan-actin (Cytoskeleton, Inc., Denver, CO, #AAN01) and anti-TATA binding protein (TBP, Abcam #51841) as a housekeeping protein. For blotting, secondary antibodies included anti-mouse IgG labelled with horseradish peroxidase (Thermo Scientific) and anti-rabbit IgG labelled with horseradish peroxidase (GE Healthcare).

Quantification of band density was done using ImageJ. Each band was highlighted using an identically sized rectangular selection, and its background-corrected density was normalized to the corresponding TBP-band background-corrected density to yield a final normalized protein level. To determine whether the observed pattern of actin protein abundance was linearly correlated with the gene expression data, we first tested whether the data followed a mixed effects model, with experiment number serving as a random effect. The experiment number was shown to have no effect on the protein expression data, so data from both experiments were pooled when testing for correlation. Each normalized band density was then plotted against the corresponding average normalized value for actin gene expression, shown in figure 2c.

Figure 2.

Actin protein production varies in the 3243A>G cybrid cell lines and correlates with gene expression levels. (a) Sample actin western blot showing actin bands and corresponding TBP bands used as a housekeeping protein. (b) Resulting protein expression obtained by normalizing the actin band density to the TBP band density. This experiment was repeated twice, with each experiment represented by a different colour. (c) Linear fits between normalized actin protein expression and actin gene expression (from figure 1a). Each data point was considered individually in the analysis after a mixed effects model showed that experiment number did not serve as a random effect on the protein level. Linear fits were performed for gene expression of β-actin (blue squares), γ-actin (purple triangles) and combined β- and γ-actin (black circles). Significant correlations were found for γ-actin (adjusted p = 0.00573, R² = 0.508) and combined β- and γ-actin (adjusted p = 0.01361, R² = 0.395), but not for β-actin (adjusted p = 0.407, R² = 0.050).

2.6. Immunofluorescence and fluorescence microscopy

For imaging, cells were plated at a concentration of 25 000 cells/dish on 30 mm glass-bottom dishes (Mattek, Ashland, MA) coated with 10 µg ml⁻¹ fibronectin (Corning, Tewksbury, MA) for 30–40 min. Microtubules, actin and nuclei were stained using commercially available fluorescent dyes, while actinin staining was performed using immunofluorescence. For more information on staining, fluorescence microscopy and image processing and analysis, see electronic supplementary material, Methods.

2.7. Atomic force microscopy

Atomic force microscopy (AFM) experiments used approximately 100 000 cells cultured on 22 × 40 mm coverslips (Fisher Scientific, Pittsburgh, PA) coated with 10 µg ml⁻¹ fibronectin for 30–40 min and placed in 60 mm Falcon dishes (Corning). This work used tipless cantilevers with nominal resonant frequencies of 10–20 kHz and nominal spring constants of 0.03–0.09 N m⁻¹ (NanoAndMore, Lady's Island, SC, #CSC38/tipless/No Al). For more information on AFM tip fabrication, see electronic supplementary material, Methods.

An Asylum MFP-3D AFM conjugated to a Nikon total internal reflectance fluorescence (TIRF) microscope was used in force spectroscopy mode to probe cellular mechanical properties. Colloid-attached cantilevers were calibrated in deionized water prior to each experimental session using the thermal-noise method [22]. The measured spring constants used for force spectroscopy ranged 0.06–0.21 N m⁻¹. Deflection sensitivity of the cantilevers was measured from the slope of the force–distance (FD) curves obtained on a clean glass slide in recording Hanks balanced salt solution (HBSS) before the start of the experiment. FD curves on cells were acquired at the perinuclear region of the cell with an approach rate of 500 nm s⁻¹. The perinuclear region of the cell was chosen so as to avoid both substrate effects, which can be substantial at the cell periphery [23], and mechanical contributions from the nucleus, which may differ from the cytoplasm in its mechanical properties [24,25]. Repeated cell indentation was performed every few seconds, which is likely infrequent enough to allow for relaxation in between indentations [26].

Cells were kept in an incubator with 5% CO₂ at 37°C until experiments, which were performed at room temperature in HBSS. FD curves were taken from 6 to 10 cells per dish, and dishes were discarded 30–60 min after experimentation began. Cells were initially identified using optical microscopy, with the probe then lowered into contact with the cell to ensure contact in the perinuclear region. A representative FD curve measured on a cell is shown in figure 4a. Asylum Research software AR 12 version was used to fit the Hertz model to the approach curves. The Hertz model, rather than a more complex model accounting for the pericellular coat [27], was used because electron micrographs of the 3243A>G cybrid cell lines did not reveal the presence of a significant pericellular coat [19]. For further information on the Hertz model and how model fitting was performed in this work, see electronic supplementary material, Methods. Since E_cell measurements often widely varied during initial measurements before converging on a narrow range (electronic supplementary material, figure S2), force curves were repeatedly obtained for a given cell until four to six consecutive curves gave consistent measures of E_cell (with the extreme measures not differing by more than approximately 15%). The resulting averaged E_cell was used for data analysis. If E_cell measurements did not appear to converge for a given cell, measurements from that cell were discarded in the final analysis. Each cell line had data pooled from 34 to 57 cells total, measured over two separate days. For linear regression, the input stiffness value for each cell was the mean elastic modulus, and the input gene expression value was the mean gene expression level for that particular gene in that cell line. Regressions were also performed by plotting the geometric means of elastic modulus by cell line against the mean expression level of a given gene for that cell line.

Figure 4.

Stiffness measurements highlight mechanical differences among the 3243A>G cybrids. (a) Sample force curve resulting from AFM indentation of a cybrid cell. The Hertz model, shown as a solid black line, is fitted to the approach curve, ultimately giving a cell elastic modulus of 535 Pa. Inset: AFM tip indenting a perinuclear region of a cell. Dotted white lines outline the cell periphery and nucleus. (b) Linear fits between the elastic modulus geometric means and the corresponding normalized gene expression for each of the eight cell lines. Geometric means were used because the data appear to trend more towards lognormal rather than normal distributions. Data are shown for (i) γ-actin, (ii) γ + β actin, (iii) α-actinin and (iv) filamin A gene expression. See electronic supplementary material, figure S5, for stiffness measurements of individual cells across the entire dataset.

2.8. Statistics

Matlab and R were both used for statistical analysis. Linear regressions were performed using the corrcoef command in Matlab or the lm command in R. R was also used for adjusting p-values for multiple comparisons (p.adjust) using the Benjamini–Hochberg method and linear mixed effects modelling (lm) with ANOVA to test for effects of experiment number. Analysis of variance (ANOVA) was performed using the anova1 command in Matlab. The p-values are generally accepted as significant when p < 0.05.

Correlations between cellular elastic moduli and gene expression data were performed on both the levels of individual data points and geometric means of each cell line. Using the former approach, each data point represents the elastic modulus measurement for a cell plotted against the mean gene expression level for the corresponding cell line. The latter approach had a total of eight data points for each correlation, where each data point represents the geometric mean of elastic moduli measured for a given cell line plotted against the corresponding gene expression level for that cell line. This approach results in higher R² values, but lower p-values in general due to the drastically reduced size of the dataset and resulting reduced power.

3. Results

3.1. Gene expression

Transcript levels for various cytoskeletal genes in the eight mtDNA 3243A>G cybrid cell lines were measured by RNA sequencing. We analysed genes encoding α-actinin, filamin A, vimentin and expressed isoforms of actin and tubulin. Figure 1a illustrates the cumulative gene expression for these genes in each of the cell lines. The graph indicates, for example, that the γ-actin isoform is expressed at higher levels than the β-actin isoform in all of the cell lines. In comparison, α-actinin expression levels are substantially lower than other genes examined. Figure 1b shows the same data normalized to the 0% cell line (no mtDNA mutation), enabling comparison of relative changes as mitochondrial function increases. This illustrates, for example, that actinin and γ-actin gene expression vary by approximately threefold across the cell lines, while β-actin and vimentin exhibit variations of lower magnitude.

Consistent with our previous studies [19], cytoskeletal gene expression did not monotonically correlate with mtDNA 3243A>G heteroplasmy levels in the different cybrid cell lines. Gene expression appeared to follow similar patterns across functionally related cytoskeletal genes (e.g. tubulin α1a and tubulin α1b, α-actinin and filamin A), consistent with co-regulation of functionally related cytoskeletal elements. To test whether associations between different genes were statistically significant, we applied linear regression analysis to the dataset and adjusted the resulting p-values for multiple comparisons (table 1). The strongest correlation was between α-actinin and filamin A, with an adjusted p-value of 2.48 × 10⁻¹¹. In addition, α-actinin expression was also positively correlated with both isoforms of actin studied (p < 0.001 for γ-actin and p < 0.0001 for β-actin), and filamin A was positively correlated with both isoforms of actin (p < 0.01 for γ-actin and p < 0.001 for β-actin). Transcript levels of other cytoskeletal elements, including various tubulin isoforms, were also significantly correlated with one another.

Table 1.

Coefficients of determination and adjusted p-values showing significance of linear relationships between expression of different cytoskeletal genes in 3243A>G cybrids. The values of gene expression for each gene in each of the cybrid cell lines were tested pairwise for linear regression. N = 3 per data cell line, total 24 total data points per regression. P-values were adjusted using the Benjamini–Hochberg method for the 28 total comparisons. Values with p < 0.05 are shown in italics.

		γ-actin	β-actin	α-actinin	tubulin α1a	tubulin α1b	tubulin β	filamin A
β-actin	R2	0.407	X	X	X	X	X	X
	adjusted p-value	2.05 × 10−3	X	X	X	X	X	X
α-actinin	R2	0.564	0.598	X	X	X	X	X
	adjusted p-value	1.31 × 10−4	6.63 × 10−5	X	X	X	X	X
tubulin α1a	R2	0.109	0.127	0.00206	X	X	X	X
	adjusted p-value	0.170	0.145	0.833	X	X	X	X
tubulin α1b	R2	0.0712	0.225	0.0459	0.610	X	X	X
	adjusted p-value	0.264	0.0337	0.383	6.17 × 10−5	X	X	X
tubulin β	R2	0.114	0.682	0.552	0.389	0.546	X	X
	adjusted p-value	0.166	9.58 × 10−6	1.49 × 10−4	2.63 × 10−3	1.49 × 10−4	X	X
filamin A	R2	0.368	0.470	0.906	0.00731	0.0913	0.484	X
	adjusted p-value	3.59 × 10−3	6.79 × 10−4	2.48 × 10−11	0.717	0.212	5.55 × 10−4	X
vimentin	R2	0.315	0.00787	0.0280	0.454	0.243	0.0821	0.0271
	adjusted p-value	8.69 × 10−3	0.717	0.496	8.58 × 10−4	0.0267	0.233	0.496

3.2. Protein abundance

We focused our subsequent analyses on actin due to its well-known role in cell mechanics [28]. To assess whether gene expression variations were reflected in protein levels, we measured the relative abundance of actin and actinin in these cell lines by western blotting. We avoided traditional glycolytic (i.e. GAPDH) and cytoskeletal elements as housekeeping genes and instead selected the nuclear TATA-binding protein (TBP) as a housekeeping protein for normalizing actin levels. Figure 2a shows a representative western blot of actin across cybrid cell lines, as well as the corresponding TBP bands. Quantified band densities from two experiments are given in figure 2b.

Measured total actin protein significantly correlated with γ-actin transcripts level (figure 2c). No correlation was found between actin protein and β-actin gene expression. However, combining total mRNA transcripts of both actin isoforms normalized to the 0% cell line yielded a significant correlation with total actin protein abundance.

3.3. Actin and actinin imaging

We next used fluorescence microscopy to visualize relative amounts of tubulin, actin and α-actinin in the cybrid lines at the single cell level. For tubulin, we observed no major differences in staining (intensity and sub-cellular distribution) between the different cell lines (electronic supplementary material, figure S3). On the other hand, the actin cytoskeleton filaments showed varying structures among the different cell lines. We focused on the adherent layer of the cell in order to maximize visualization of cytoskeletal proteins. Figure 3a shows representative images from several of the cell lines, while electronic supplementary material, figure S4, displays multiple cells from each of the cell lines. Consistent with the oncogenic nature of cybrid cell lines, actin staining largely presented as disordered aggregates and fewer organized stress fibres than are typically present in normal cells.

Figure 3.

Sample fluorescent images demonstrate cytoarchitectural heterogeneity among several of the 3243A>G cell lines. (a) Images are identically contrast-adjusted for better visualization. Scale bar is 20 µm. Also shown are masks corresponding to each sample image, and thresholded actin images, which are both described in detail in §2.7. (b) Proportions of cell cytoplasm occupied by actin-containing pixels for each heteroplasmy level, ANOVA, p = 0.089. The percentages of actin in the individual sample cells shown in (a) measured 21.92% (20%), 32.1% (90%) and 20.3% (ρ₀). (c) Boxplot showing circularity measurements by cell line. One outlier is omitted from the plot for better data visualization. The overall ANOVA for the analysis gave p = 0.0407. The circularity measurements of the sample cells shown in (a) were 0.3108 (20%), 0.0685 (90%) and 0.0927 (ρ₀). For both (b) and (c), N = 15 cells per group. Box borders show 25th (lowest line), 50th (centre line) and 75th (highest line) percentiles for each cell line; whiskers show 10th and 90th percentiles. Outliers are shown as +signs.

In general, we observed substantial cell-to-cell heterogeneity in actin staining as might be expected (electronic supplementary material, figure S4). However, when looking at populations of >10 cells, certain trends were apparent. In particular, the 90% heteroplasmic cell line, where mitochondria are highly dysfunctional and energy production is minimal, consistently showed a denser actin network with more pronounced stress fibres than any other cell line (figure 3a and electronic supplementary material, S4). Most other cell lines showed actin structural networks which appeared overall similar to one another, with the 30% and 60% cell lines displaying slightly denser stress fibre networks. The relative percentage of surface area occupied by actin was quantified in each cell line (figure 3b). Although the one-way ANOVA comparing the different cell lines did not achieve significance (p = 0.0892), the data confirmed that the 90% cell line harboured the highest actin density (median of 27.4%), followed by the 60% (median 26.6%) and 30% (median 26.0%) cell lines. Of these cell lines, the 90% had the smallest spread, indicating a more consistently dense actin cytostructure.

The ρ₀ cell line, which lacks mtDNA and therefore functional mitochondria, surprisingly lacks dense bundles of stress fibres which might be expected from the gene expression (figure 1) and protein abundance (figure 2) data. However, ρ₀ cells typically contained relatively more actin-rich [29] filopodia protruding from the cytoplasm than in most other cell lines. We quantified this observation by considering the circularity of a mask of each cell (figure 3a). The resulting data (figure 3c) showed the lowest circularity measurements for the 90% and ρ₀ cells (ANOVA, p = 0.0407), suggesting superior spreading and dynamic actin turnover in these cell lines.

Actinin, which acts as a cross-linker of actin filaments [30], tended to colocalize with actin. Higher magnification confocal images (figure 3, insets) illustrate actinin positioning along actin stress fibres and at the cell periphery. Colocalization analysis showed that across all cell lines, 56.4 ± 6.0% of actinin-containing pixels were also occupied by actin, and this measurement rose to 80.0 ± 5.8% when actin pixels within a 2-pixel (264 nm) radius of each actinin pixel were considered for colocalization. Since on average, approximately 25% of a given cell contained actin according to our thresholding analysis, and approximately 20% contained actinin, random colocalization would occur about 5% of the time. Our observation showing over ten times that percentage supports the association between actin and actinin and is consistent with the general correlation between actin and actinin gene expression (table 1).

3.4. Cell stiffness measurements

Finally, we addressed whether the alterations in cytoskeletal gene and protein expression translated into quantitative changes in cell mechanical properties as a function of mitochondrial heteroplasmy. Figure 4a shows a typical force–distance curve from an AFM cybrid cell measurement, as well as the fitted Hertz model and resulting elastic modulus. The figure inset shows an image of an AFM tip with an attached colloid in contact with the perinuclear region of a cybrid cell.

Data were collected from 34 to 57 cells per line. The cybrid cell lines in general were relatively soft, with elastic moduli of the order of approximately 500 Pa. This is expected based on the cybrids' osteosarcoma origin and is consistent with mechanical measurements of other cancer cell lines [12]. Similar to our imaging results, significant heterogeneity existed in elastic modulus measurements between different cells of each cell line (electronic supplementary material, figure S5), as is typically seen in biological systems [31]. Despite this cell-to-cell variability, the overall stiffness pattern mirrored that of gene expression shown in figure 1, with the 90% mutant and ρ₀ cell lines appearing to be the stiffest in parallel with highest expression of α-actin and actinin.

Our statistical analysis focused on the correlation between cellular elastic modulus and gene expression for the corresponding cell type, done on both the levels of individual cells and geometric means of elastic moduli measured for all cells in a given cell line (table 2). For effective data visualization, we plotted geometric means of elastic moduli for each cell line against the corresponding mean gene expression for γ-actin, combined γ and β actin, α-actinin and filamin A (figure 4b). Comparisons using the overall data showed that stiffness measurements significantly correlated with gene expression of γ-actin, combined β- and γ-actin, α-actinin and filamin A (adjusted p < 0.05 for all). The p-values correlating the geometric mean of the elastic modulus with γ-actin or combined γ and β actin gene expression are also significant. Notably, no correlations existed between stiffness data and vimentin or the different tubulin genes measured.

Table 2.

Coefficients of determination and adjusted p-values between stiffness measurements of mtDNA 3243A>G cybrids and different cytoskeletal genes. Both parameters are evaluated by correlating all individual elastic moduli data points with the mean gene expression for each gene, and separately by correlating geometric means of elastic moduli for each of the cell lines with the mean gene expression for each gene. The p-values are adjusted for multiple comparisons using the Benjamini–Hochberg method. Values with p < 0.05 are shown in italics.

		γ-actin	β-actin	β + γ-actin	α-actinin	α1a tubulin	α1b tubulin	β-tubulin	filamin A	vimentin
individual data points	R2 for gene expression versus E	0.0289	0.00841	0.0245	0.0241	0.00237	0.00337	0.00374	0.0188	0.00185
	adjusted p-value	0.0144	0.174	0.0144	0.0144	0.426	0.378	0.378	0.0286	0.437
geometric means of E	R2 for gene expression versus E	0.764	0.330	0.734	0.600	0.0507	0.111	0.0928	0.441	0.102
	adjusted p-value	0.0295	0.246	0.0295	0.0722	0.592	0.521	0.521	0.163	0.521

4. Discussion

4.1. Overall findings

The most salient finding in this work is that cells with the same nuclear genome differing only by their degree of mtDNA 3243A>G mutation differ in expression levels of various cytoskeletal genes and cellular mechanical properties. Consistent with a model whereby mitochondrial dysfunction directly alters various cellular functions by signalling transcriptional and post-translational processes, our findings suggest that the relationship between mitochondrial dysfunction caused by this mutation and cell mechanics may be mediated by specific changes in the transcriptional state and protein levels of cytoskeletal elements related to cortical stiffness and cytoplasmic cytoskeletal organization. This hypothesis is conceptually illustrated in figure 5. In accordance with this proposed model, our analyses focus on the relationship between gene expression of cytoskeletal elements and cytoskeletal protein content, cytoskeletal structure and cellular mechanical properties. Our observation of a conserved relationship between these measures, rather than their direct correlation with mtDNA heteroplasmy, supports the notion that mitochondrial dysfunction has a significant but complex effect on the cytoskeleton and cell mechanics.

Figure 5.

mtDNA 3243A>G heteroplasmy affects cell mechanics. (a) The different cytoskeletal effects of mtDNA heteroplasmy are sequentially related to one another. First, nuclear expression of cytoskeletal genes changes in response to mtDNA mutation. This in turn affects expression of cytoskeletal proteins, particularly actin and α-actinin, which then affect cellular cytostructure and ultimately cause relative changes in cell stiffness. (b) Graphic representation of how alterations in mtDNA affect cell mechanics in a sequential manner.

The mechanisms underlying the complex relationship between mtDNA 3243A>G heteroplasmy level and cytoskeletal gene expression remain to be studied. Our earlier work demonstrated that increasing levels of mtDNA 3243A>G heteroplasmy induced abrupt transcriptional changes in both mitochondrial and nuclear genes [19]. The current study highlights that these massive cellular effects extend to transcription of cytoskeletal genes, which in turn are correlated with variations in cytoarchitectural framework and cell mechanical properties. We anticipate that future studies will address the mechanistic pathways underlying this pattern of effect.

While many types of cellular dysfunction are characterized by both mitochondrial and mechanical dysfunction, we believe this is the first experimental demonstration of a mtDNA mutation directly affecting cellular mechanical properties. Recently published findings showed that accumulated mtDNA mutations caused by a defect in the nuclear gene encoding the mtDNA polymerase-γ led to arterial stiffening in mice [32]. In this model, the underlying genetic defect is nuclear and leads to generalized mitochondrial mutations which are not identified by type or quantity. Additionally, cell stiffness is measured on the tissue level. While this work is highly relevant to the study of interactions between mitochondrial defects and cell mechanics, our work differs in that it focuses on mechanical effects on the single cell level which result from graded effects of a particular mtDNA mutation.

Our work additionally shows that actin gene expression does not correlate 1 : 1 with cell stiffness, with large changes in gene expression translating to relatively small alterations in cellular elastic modulus. Changes in actin gene expression translate into variations in protein abundance of roughly the same magnitude. The cell phenotype appears to change drastically with alterations in actin gene and protein levels, as illustrated by the denser stress fibre network shown in 90% heteroplasmic cells or the abundance of actin-rich filopodia shown by the ρ₀ cells. Despite these pronounced alterations in actin gene expression and ultimately cytostructure associated with different heteroplasmy levels, cell stiffness exhibits relatively low variation (approx. 45% maximal change in median elastic modulus between cell lines).

In general, gene expression may be regarded as the ‘attempt’ of the cell to respond to a certain set of environmental conditions. Cellular mechanical properties and cytoarchitectural characteristics represent what is actually achieved by the cell, which may or may not match gene expression. Our work identifies an unexpected but conserved relationship that extends from this attempt to its actualization. Future work should build upon these findings to elucidate the molecular mechanisms underlying this relationship.

4.2. Roles of different cytoskeletal proteins in cell stiffness

Although actin has mostly been identified for its central role in cell mechanics [28], several studies have identified roles for α-actinin in cell stiffness preservation [33] and focal adhesion maturation [34]. Similarly, filamin has been shown to work cooperatively with α-actinin to maintain more solid-like behaviour of cross-linked actin gels [35], and filamin A in particular has been highlighted for its role in cell mechanosensing in certain environments [36]. It is thus reasonable to assume a combined role for actin, α-actinin and filamin A in maintaining cellular elasticity, as actinin and filamin primarily function to cross-link actin filaments and anchor them to the cell membrane [30,37]. This is consistent with the predominant presence of actinin at the cell periphery (figure 3). The cooperative relationship between the three proteins supports the correlations between actin, actinin and filamin A gene expression observed in the 3243A>G cybrid cell lines (table 1). Furthermore, the proven role of all three proteins in maintaining mechanical homeostasis supports our observation that cell stiffness is correlated to transcript expression of each of their corresponding genes.

We did not observe any correlation between elastic moduli of the cybrid cells and either tubulin or vimentin gene expression. Unlike actin, the role of microtubules in cell mechanics is a matter of debate, with some studies showing no direct relationship between the two [38,39] and others finding that microtubule depolymerization results in lowered elastic moduli in cells [40,41]. Our current work supports the former view, demonstrating an absence of correlation between tubulin gene expression and measured 3243A>G cybrid cell stiffness. The same is true for vimentin gene expression, which may seem more surprising since vimentin integrity has been shown to be important for preserving cytoplasmic cell stiffness in several cell types [42,43]. However, it was recently shown that vimentin does not contribute to stiffness of the cell cortex [44], which is the sub-cellular region probed by AFM experiments in the current study. The lack of correlation between vimentin gene expression and cell stiffness is thus consistent with previous results. Still, because our measurements are limited to gene expression without assessment of protein polymerization status of the corresponding proteins, we cannot rule out the possibility that mitochondrial bioenergetics impact cell stiffness via post-translational mechanisms regulating polymerization of vimentin and/or various tubulin isoforms. Future work should investigate the possible effect of mitochondrial (dys-)function on cytoskeletal dynamics, including those of vimentin and microtubules, and cell stiffness.

While we address the roles of multiple cytoskeletal components in determining cell stiffness, other proteins previously related to cytoskeletal structure were not assessed. One particular such protein, myosin II, has been shown to be critical to maintaining cell stiffness due to its role in actin cross-linking [45]. We hope that future work will expand these results in order to provide a more comprehensive picture of the different cytoskeletal effects of mtDNA 3243A>G heteroplasmy.

4.3. Methodological considerations

It is possible that substrate effects may have influenced our elastic modulus measurements in a small number of cases. This could be true for the softest cells, for which a greater indentation was required to achieve the predefined tip deflection during experimentation. As we did not measure the cell thickness, it is possible that indentations for these cells (up to approximately 500 nm) exceeded 10% of the cell thickness, which may in turn have led the mechanical properties of the substrate to influence the resulting mechanical measurements. However, substrate effects would make these cells appear stiffer, lessening our concern. Additionally, our imaging technique for probe positioning provided assurance that we were making AFM measurements far from the leading edge of the cell, such that the indentation depth has less influence on the results [46]. Moreover, as these greater indentation depths were only reached for softer cells, any effects of the substrate (fibronectin-coated glass) would have ultimately led to an understatement of the mechanical variation we observed among the 3243A>G cybrid lines.

In addition to avoiding substrate effects, the use of the Hertz model to measure the cellular elastic modulus with a colloidal probe relies in part on establishing an input force that is directed normal to the cell surface. Unless cells have a very flat, horizontal region of membrane that is being probed, local membrane curvature can lead to probe–membrane orientation effects which contribute to modelling error. To minimize these effects, we used optical microscopy to locate the perinuclear area of the cell for AFM measurements, avoiding both the thin cell periphery (i.e. substrate effects) and potential contribution of the nucleus to the local elastic modulus. Additionally, by acquiring data from numerous cells for each cell line, it is highly likely that the randomness of the different spatial orientations sampled has kept any systematic error introduced by the membrane curvature/probe orientation to an experimentally reasonable level.

Another potential concern relates to the elastic modulus measurements deriving from repeated indentations of a given cell. While our experimental platform can be justified by the need for data reproducibility, repeatedly probing the cell may result in cytoskeletal rearrangement and thereby change the cell modulus. If such rearrangement occurs, the degree of alteration of local actin structure is likely dependent on multiple interrelated factors, including cell bioenergetic state and the local availability of actin to polymerize or depolymerize. These factors may vary between the different cell lines studied, suggesting an interesting subject for additional study. In any case, we did not see a consistent response of the cell to repeated indentations. Electronic supplementary material, figure S2, shows a cell that ultimately softens after multiple indentations, but our dataset was rich with examples of cellular moduli increasing or fluctuating after repeated indentations.

In our western blotting experiments, we found a linear relationship between actin protein levels and gene expression of γ-actin, the most highly expressed actin isoform in the 3243A>G cybrids. Measured protein levels did not correlate with genetic expression of β-actin, but did with combined β- and γ-actin. There are several possible reasons for these findings. First, most traditional housekeeper proteins for western blot normalization concern either glucose metabolism (e.g. GAPDH) or the cytoskeleton, both of which were not possible in our case. We chose TBP as an alternative, but it is possible that TBP expression is not consistent across the 3243A>G cell lines, since 3243A>G heteroplasmy strikingly affects many nuclear genes [19]. Quantification of protein abundance by western blotting in general is subject to greater technical variability [47], so we place more weight on gene expression and cell stiffness measurements.

In a similar vein, while it may seem more logical in principle to correlate cell stiffness with measurements of actin density resulting from fluorescent images, we chose instead to compare cell stiffness with gene expression data. While our quantitative measurements of the fluorescent images did support our qualitative impressions, the actin density measurements in particular can be very sensitive to normalization and threshold values. Circularity measurements are less dependent on input parameters, but relating cell stiffness to circularity is not as theoretically straightforward as comparing stiffness with actin density. Additionally, we found that the trend shown by actin density (figure 3b) does not precisely recapitulate differences observed in gene expression or the stiffness measurements, as the ρ₀ and 90% lines had similar gene expression and AFM measurements but drastically different actin cytostructures. These differences would undermine the potential correlation between imaging data and stiffness data. Since actin quantification and circularity measurements overall indicated a persistence in the original gene expression pattern of actin and other related genes, we ultimately chose to correlate gene expression with cell stiffness based on the objectively quantitative nature of both these experimental systems.

A potential methodological limitation of this work is that while gene and protein expression studies were performed on homogenates of confluent cell monolayers, immunofluorescence and AFM experiments used individual isolated cells. One obvious consequence of this is our inability to directly correlate gene or protein expression in a given cell with the cell's elastic modulus. Instead, we assumed that each cell expressed a given gene at a level equivalent to the population average. While cell-to-cell structural and mechanical heterogeneity suggested that this assumption is probably incorrect in a strict sense, collecting data from many cells per cell type showed that the population-wide expression patterns of particular genes were indeed quantitatively correlated with mechanical properties within this smaller population. Another possible problem with pooling experimental data from confluent and single cells is that degree of confluency may affect cell mechanical properties [48]. We expect, though, that this change would remain similar across the different 3243A>G lines and that the relative differences in cytoskeletal gene expression, protein production and cell stiffness would persist whether cells were taken from a confluent population or analysed at the individual level. Although we did not directly confirm this, our experimental results support this hypothesis.

5. Conclusion

This study demonstrates a connection between mtDNA 3243A>G heteroplasmy and single cell mechanics. The particular mechanisms linking this mutation with nuclear cytoskeletal genes remain decidedly nonlinear, but future work might address relevant underlying mechanisms. The connections between mitochondrial and mechanical function may also be better explored in cells exposed to different environmental conditions, including situations that promote acute mitochondrial respiratory chain dysfunction (e.g. sepsis), where energy supply is limited, or where the cytoskeleton is directly disrupted. Overall, this work suggests potential mechanisms that may link mitochondrial and mechanical dysfunction.

Data accessibility

Gene expression data are available in the Gene Expression Omnibus (GEO) database (www.ncbi.nlm.nih.gov/geo, accession no. GSE56158). Other data are available on figshare at the following links: western blotting (https://figshare.com/s/545b0b4e51eddc8fd6e1), confocal images (https://figshare.com/s/d9bc2907e0acde74836f) and cell stiffness data (https://figshare.com/s/d07ee7cd18ca4e2db959). Matlab analysis codes have been provided as part of the electronic supplementary material for this article.

Authors' contributions

J.K., M.P., and D.M.E. conceived of the study. J.K. and M.P. carried out experiments. J.K. analysed the data and wrote the paper with critical contributions from M.P. and D.M.E. D.C.W. provided important overall guidance.

Competing interests

The authors declare no competing interests.

Funding

This work was funded by Office of Naval Research grant N000141612100, National Institutes of Health grant no. T32HL007954, NS21328 and DK73691, the Horatio C. Wood Endowment at the University of Pennsylvania, Simons Foundation grant 205844 and the Canadian Institute of Health Research. The TIRF-AFM instrumentation is additionally funded through the Penn Nano/Bio Interface Center via NSF Major Research Instrumentation grant no. DBI0721913 and NSF Nanoscale Science and Engineering Center grant no. DMR0425780.