Overview of a mechanomics approach for de novo identification of genes involved in cell mechanics regulation. (A) Data curation. Datasets originating from different biological systems encompassing cell states with distinct mechanical phenotypes, as characterized by real-time deformability cytometry (RT-DC), and associated transcriptomics profiles are collected. (B) Target prediction. A subset of collected datasets is used to perform machine learning-based network analysis on transcriptomic data and identify conserved module of genes associated with cell mechanics changes. PC – principal component. (C) In silico validation. The classification performance of individual genes from module identified in (B) is evaluated in silico on remaining datasets. TPR – true positive rate, FPR – false positive rate, AUC – area under the curve. (D) Experimental validation. Targets with highest classification performance in silico are verified experimentally in perturbation experiments.

Mechano-transcriptomic datasets used in this study. Pred – prediction, Val – validation, PI/II– positive hypothesis I/II, N – negative hypothesis, CCLE – cancer cell line encyclopedia, HT Seq – high-throughput RNA sequencing, CAGE – cap analysis gene expression, AFM – atomic force microscopy, adeno – adenocarcinoma, wt – wild type, PP – proliferating progenitors, NNs – newborn neurons.

Mechanical properties of divergent cell states in five biological systems. Schematic overviews of the systems used in our study, alongside with the cell stiffness of individual cell states parametrized by Young’s moduli E. (A) Human patient-derived glioblastoma cells with three distinct signaling states maintained by indicated culture conditions. (B) Human small-cell carcinoma and adenocarcinoma cell lines originating from intestine, lung, and stomach. (C) Human breast epithelium MCF10A cell line bearing single-allele H1047R mutation in the PIK3CA with parental wild type (wt) as a control. (D) Murine F- and C-class iPSCs cultured in the presence or absence of doxycycline (dox) activating ectopic expression of OSKM factors (Oct4, Sox2, Klf4, and cMyc). (E) Proliferating progenitors (PPs) and newborn neurons (NNs) isolated from brains of mouse embryos. Horizontal lines delineate medians with mean absolute deviation (MAD) as error, datapoints represent medians of the individual replicates, the number of independent biological replicates is indicated below each box. Statistical analysis was performed using generalized linear mixed effects model. The symbol shapes represent cell lines derived from three different patients (A), matched experimental replicates (C), two different reprogramming series (D), and different cell isolations (E). Data presented in (A) and (D) were previously published in ref (29) and (30), respectively.

Identification of putative targets involved in cell mechanics regulation. (A) Glioblastoma and iPSC transcriptomes used for the target prediction intersect at 9,452 genes. (B, C) PCA separation along two first principal components of the mechanically distinct cell states in the glioblastoma (B) and iPSC (C) datasets. The analysis was performed using the gene expression data from the intersection presented in (A). The symbol shapes in (B) represent cell lines derived from three different patients. (D) Schematic representation of PC-corr analysis and the combination of the PC-corr results for two systems. (E to G) Gene networks based on filtering gene pairs by the combined PC-corr score. The presented networks were obtained by setting the cut-off value to 0.75, when using the mean PC-corr approach (E), and to 0.70 (F) and 0.75 (G), when using the minimum value approach. In (E to G) edge thicknesses represent the |PC-corrcomb| (Table S1 and S2) and the colors of the nodes represent the average processed PC loadings (Table S3).

List of identified target genes comprising the conserved module.

Expression of identified target genes in the prediction and validation datasets. Panels show unsupervised clustering heat maps of expression data from transcriptomic datasets corresponding to the following systems: (A) glioblastoma, (B) induced pluripotent stem cells (iPSCs), (C) carcinoma, cell lines from intestine, lung and stomach used for positive hypothesis I (see Table 3) are highlighted in pink, green, and orange, respectively; *mechanically tested cell lines (here the FANTOM5 dataset is presented as it contains all the cell lines that were tested mechanically in our study, for the remaining carcinoma datasets see Figs. S4-S6), (D) non-tumorigenic breast epithelia MCF10A, and (E) developing neurons (dev. neurons). Comb – combinatorial marker, wt – wild type, PPs – proliferating progenitors, NNs – newborn neurons. Clustering was performed using clustergram function in MATLAB (R2020a, MathWorks) on log-normalized expression data.

Overview of the hypotheses and datasets used for validating universality and specificity of obtained markers. Hypotheses are listed in the column headings. Under every hypothesis, sample groups used for the hypothesis testing are listed. Numbers of samples used in every group are indicated in brackets.

Validation of identified target genes and the combinatorial marker. Minimum AUC-ROC (min AUC-ROC) and JVT p values are reporter for the two positive hypotheses and one negative hypothesis for each target genes and the combinatorial marker (comb). The specific datasets and comparisons used for testing of each hypothesis are listed in Table 3 above. The results presented in this table can be reproduced using the code and data available under the GitHub link reported in the methods section.

Perturbing levels of CAV1 affects the mechanical phenotype of intestine carcinoma cells. (A) CAV1 levels in small-cell (ECC4) and adenocarcinoma (TGBC) cell lines from intestine. (B to D) Mechanical phenotype of ECC4 and TGBC cells measured with RT-DC (B, as in Fig. 2B), AFM indentation (C), and AFM microrheology (D). (E) Verification of CAV1 knock-down in TGBC cells using two knock-down system: three esiRNA constructs (esiCAV1-1. esiCAV1-1, and esiCAV1-3 with rLuc as a control), and pooled siRNA mixture (CAV1-pool with non-targeting mixture (nonT) as a control). (F to H) Mechanical phenotype change of TGBC cells upon CAV1 knock-down as measured by RT-DC (F), AFM indentation (G), and AFM microrheology (H). (I) Verification of transient CAV1 overexpression in ECC4 and TGBC cells. (J) Mechanical phenotype change of ECC4 and TGBC cells upon CAV1 overexpression as measured by RT-FDC. Gating for fluorescence positive and negative cells based on dTomato expression in ECC4 (top) and TGBC (bottom) cells (left-hand side). Fluorescence positive cells correspond to cells expressing CAV1-IRES-dTomato (CAV1iT). For comparison, mock transfection sample is shown in the background (mock). Apparent Young’s modulus changes of ECC4 and TGBC cells upon CAV1 overexpression (right-hand side). CAV1iT- and CAV1T+ are dTomato negative and positive cells, respectively. For protein quantification in (A), (E), and (I), representative Western blots (top) as well as quantification of specified replicate numbers N (bottom) are shown. In (B), (F) and (J), horizontal lines delineate medians with mean absolute deviation (MAD) as error, datapoints represent medians of N experiment replicates, statistical analysis was performed using generalized linear mixed effects model. In (C) and (G), box plots spread from 25th to 75th percentiles with a line at the median, whiskers span 1.5 × interquartile range (IQR), individual datapoints correspond to values obtained for n individual cells, statistical analysis was performed using two sample two-sided Wilcoxon rank sum test. In (D) and (H), datapoints correspond to means ± standard deviation of all measurements at given oscillation frequencies for n cells. Lines connecting datapoints serve as guides for the eye. E – apparent Young’s modulus, G* – complex shear modulus, ΔE – apparent Young’s modulus change relative to respective control measurements. In (E), (F), (I), and (J), the symbol shapes represent matching experiment replicates.

Perturbations of CAV1 levels in MCF10A-ER-Src cells result in cell stiffness changes. (A) Inducing transformation of MCF10A-ER-Src cells by tamoxifen (TAM) treatment, as opposed to vehicle control (ethanol, EtOH), causes a decrease of CAV1 expression over time, as captured by microarray analysis (GEO accession number: GSE17941, data previously published in (75)). Datapoints with error bars represent means ± standard deviation (N = 2, unless indicated otherwise). (B) Western blot analysis shows the decrease of CAV1 at protein level 72 h post induction. (C) MCF10A-ER-Src cells show decreased apparent Young’s moduli 72 h post TAM induction. (D) CAV1 knock-down in uninduced MCF10A-ER-Src cells results in lowering of the apparent Young’s modulus. (E) Overexpression of CAV1 in TAM-induced MCF10A-ER-Src cells causes increase in the apparent Young’s modulus and effectively reverts the softening caused by TAM induction (compare to panel c). Box plots in (C to E) spread from 25th to 75th percentiles with a line at the median, whiskers span 1.5 × interquartile range (IQR), individual datapoints correspond to values obtained for individual cells, the number of measured cells per conditions, pooled from N = 3 independent experiments, is indicated below each box. Statistical analysis was performed using a two-sided Wilcoxon rank sum test. In (B), (D), and (E), the symbol shapes represent experiment replicates.