Abstract
Circadian rhythms are a key survival mechanism that dictates biological activity according to the day-night cycle1. In animals, cell-autonomous circadian clocks can be found in nearly every cell type and are subjected to multi-layered regulation2–4. Although these peripheral clocks are remotely controlled by the master clock in the brain, they are also sensitive to their immediate physical microenvironment through mechanisms that are still unknown5,6. Here we show that the circadian clock in fibroblasts is regulated mechanically through YAP/TAZ and TEAD. We use high-throughput analysis of single-cell circadian rhythms and apply controlled mechanical, biochemical, and genetic perturbations to study the expression of the core clock gene Rev-erbα7. We observe that Rev-erbα circadian oscillations are disrupted concomitantly with the translocation of YAP/TAZ to the nucleus. By targeted mutations and tuning expression levels of YAP we identify TEAD as the transcriptional effector of this mechanosensitive regulatory pathway. Our findings establish a mechanism that links cell mechanobiology and the circadian clock, which could contribute to explain the circadian impairment observed in cancer and ageing, where the regulation of the mechanical environment and YAP/TAZ is lost8–11.
Main text
Circadian rhythms are defined as an organism’s physiological and behavioural adaptations to the cyclic environment that arises from the Earth’s rotation1. In mammals, these rhythms are intrinsically sustained by the circadian clock, a set of proteins whose expression is self-regulated, at the single-cell level, via a transcription-translation feedback loop with a period close to 24 hours12. Up to 25% of the mammalian proteome is estimated to be influenced, in a cell type-specific manner, by the circadian clock, which deeply conditions dynamics and function of tissues and organs throughout the day13–15.
The central element of the mammalian circadian clock is the transcription factor BMAL1:CLOCK, which activates the transcription of, among many others, Cry and Per. In turn, the sustained expression of CRY and PER generates a transcriptional loop by binding to BMAL1:CLOCK and blocking its activity16. A second stabilizing layer of regulation of the clock is provided by REV-ERB, which, on one hand, is transcriptionally promoted by BMAL1 :CLOCK and, on the other hand, acts as a transcriptional inhibitor of Bmal17.
For every cell residing in a peripheral tissue, the robustness of the circadian clock and its synchrony with the environment are subjected to multi-layered regulation2,3. Decades of work have unveiled the biochemical side of this regulatory network which, mainly through neurocrine and paracrine signalling, drives important changes in the gene expression landscape of the different cell types4,17,18. Besides biochemical regulation, recent evidence suggests that the circadian clock is also influenced by mechanical factors such as actin dynamics, cell-cell adhesions and the stiffness of the extracellular matrix (ECM)5,6,19. However, the mechanosensitive molecular mechanisms that enable the transcriptional regulation of the circadian network are unknown.
By combining confocal microscopy, microfabrication, and a customized computational toolkit, we show that high nuclear levels of the transcriptional regulators YAP/TAZ cause a dramatic change in the expression of Rev-erbα and the subsequent malfunction of the fibroblast circadian clock. This novel molecular mechanism, which we show to be TEAD-dependent, could explain the highly impaired circadian clock in circumstances typically associated with high nuclear YAP/TAZ translocation such as cancer or ageing8–11.
We conducted our work using NIH3T3 fibroblasts, a cellular model extensively reported to express a robust self-sustained circadian molecular clock20,21. To prevent the dominance of specific signalling pathways derived from hormonal shocks22 and reveal underlying mechanobiological signals, cells were not synchronized by exogenous entrainment during the experiment. We generated a stable clonal cell line carrying both a reporter of Rev-erbα transcription (RevVNP)20 and a chimeric histone2B-mCherry that we used as a constitutive nuclear marker (Fig. 1a, left). We also developed a method to systematically track and measure the single cell expression of RevVNP over 3-day time-lapses performed by confocal microscopy. Then, by performing Fourier analysis of the fluorescence emission of each cell and quantifying the fraction of the power-spectrum belonging to the circadian frequency, we obtained a signature of the circadian strength of every single cell (Fig. 1a, Methods).
We first addressed the effect of cell density on RevVNP expression by culturing cells on hydrogels (30 kPa in stiffness) at two distinct densities (~30 cells/mm2 and 1000 cells/mm2, hereafter referred to as low and high density, respectively), which we kept approximately constant over time by arresting the cell cycle with the addition of thymidine, a DNA synthesis inhibitor23. We observed a dramatic cell density-dependent difference in Rev-erbα expression, with cells cultured at low density displaying much higher RevVNP nuclear intensity than confluent cells (Fig. 1b-c). These differences were ratified at the protein level via immunostainings against REV-ERBα (Extended Data Fig. 1). The single cell frequency analysis of the RevVNP signal revealed that the population of cells grown at high density expressed robust oscillations with an average period of 24.7 ± 1.9 h (Fig. 1d-e). By contrast, cells at low density displayed low RevVNP circadian strength, with many cells exhibiting poor circadian oscillations, missing one or several peaks or even experiencing non-periodic fluctuations rather than oscillatory behaviour (Fig. 1d-e).
We next asked whether the detected differences in circadian strength are caused by paracrine signals, as previously reported for intercellular circadian coupling and synchronization24,25. To address this question, we exposed closely packed cells to an abrupt drop in cell density. To do so, two cell monolayers were grown to confluence separated by a PDMS barrier. Upon lifting the barrier (Fig. 2a, left), cells at the monolayer edges begun to spread and migrate towards the freely available gap. These cells experienced a sudden increase in RevVNP intensity that transiently disrupted their circadian oscillations until the gap was closed (Fig. 2a-c; Supplementary Video 1). This disruption contrasted with the behaviour of cells far from the edge, which oscillated robustly throughout the entire experiment as shown in the kymograph depicted in Fig. 2b. We obtained similar results in presence or absence of thymidine (Extended Data Fig. 2), which confirms that the increase seen in Rev-erbα transcription is not a side effect of a cell cycle re-entry at the monolayer edge upon a loss of cell-cell contact. These experiments further demonstrate that cell density regulates the circadian clock and suggest that this phenomenon is not explained by paracrine signals, as high- and low-density cells shared the same extracellular chemical environment. We confirmed that changes in paracrine signalling are not the cause of RevVNP circadian impairment by growing cells at low density with conditioned medium obtained from a high-density culture. We observed no increase in circadian strength compared to the low-density cells grown in fresh medium (Fig. 2d). These results indicate that cell density does not affect the circadian expression of Rev-erbα through paracrine signals, which led us to hypothesize that the observed behaviour had a mechanochemical origin.
To test this hypothesis, we first asked whether the differences in RevVNP circadian oscillations between high and low density depend on cell-cell adhesion. To address this question, we confined individual cells in fibronectin micropatterns of 1200 μm2 (a size comparable to that of fibroblasts within a monolayer) (Fig. 2e). The confined individual cells showed a significant reduction in RevVNP intensity together with an increase in circadian strength when compared to free cells grown at equivalent low density, a behaviour resembling that of high-density cells (Fig. 2e-f; Supplementary Video 2). This result implies that contact-based mechanisms in general, and intercellular adhesion in particular, do not explain the dependence of the Rev-erbα circadian oscillations on cell density.
We thus explored whether the onset of circadian oscillations at high density had a contact-independent mechanical origin. We focussed on two key mechanosensitive transcriptional regulators. The first one is MAL, whose localization is regulated by actin dynamics and has been shown to modulate the transcription of the circadian gene Per226. The second one is YAP, the central component of the Hippo pathway, which translocates in a mechanoresponsive manner between the cytosol and the nucleus27–29. YAP was recently shown to bind to REV-ERBα30 although it has not been linked to the circadian clock. We measured and compared both circadian strength and subcellular localization of MAL and YAP in a series of conditions known to impact their nucleocytoplasmic shuttling. These conditions include the already mentioned high and low cell density and single cell micropatterning; treatment with drugs that alter the actomyosin cytoskeleton such as jasplakinolide, latrunculin A, para-nitro-blebbistatin and cytochalasin D; and the growth of cells on very soft hydrogels (300 Pa)28,31–33. As expected, these conditions had a variety of effects on cell morphology (Fig. 3a, Extended Data Fig. 3a). We found no correlation between nuclear to cytoplasmic ratio of MAL and circadian strength (Extended Data Fig. 3). By contrast, circadian strength showed a progressive decrease with YAP nuclear to cytoplasmic ratio (Fig. 3). This striking anticorrelation reveals that the robustness of the Rev-erbα circadian expression depends on the nucleocytoplasmic transport of YAP and its mechanosensitive regulation.
To test whether this dependence is causal, we induced in high-density cells a sustained overexpression of a non-phosphorylatable YAP mutant, 5SA-YAP, which has abnormally high nuclear retention34. The resulting high levels of 5SA-YAP provoked a severe reduction of RevVNP circadian strength (Fig. 4a) and an increase in REV-ERBα basal levels when compared to high-density control cells (Fig. 4b-d). A similar result was obtained when overexpressing the TAZ mutant 4SA-TAZ, which is also retained in the nucleus due to the mutation of its four Lats inhibitory phosphorylation sites35,36 (Extended Data Fig. 4). Together, these results establish that the circadian clock is controlled through both YAP and TAZ.
The activity of YAP/TAZ as gene expression coactivators requires the interaction with other transcription factors. Since YAP interacts with REV-ERBα through its coiled-coil (CC) domain30, we tested if this interaction gives rise to the deleterious effect of nuclear YAP on the circadian clock. For that, we overexpressed 5SA-YAP but with its CC domain deleted (5SA-CCDEL-YAP), a mutation that had been previously shown to disable the molecular interaction between YAP and REV-ERBα30. We found that the CCDEL mutation only rescued slightly the impaired RevVNP circadian oscillations (Fig. 4a) and the REV-ERBα basal levels (Fig. 4b-d) observed upon 5SA-YAP overexpression. This result shows that the direct molecular binding through the YAP CC domain is not the main regulatory pathway connecting YAP and the circadian clock.
Alternatively, we studied whether this regulation was mediated by the well-known interaction of YAP/TAZ with the TEAD family37. Analysis of published ChIP-seq and microarray databases revealed that Bmal1-but not Rev-erbα- is listed as a YAP and TEAD target38–41, providing a potential direct mechanism for YAP/TAZ to control the circadian clock. To test this possibility, we overexpressed 5SA-S94A-YAP, a mutant version of YAP unable to interact with TEAD40 and observed that the cells recovered, to a large extent, both the RevVNP circadian strength and the REV-ERBα basal levels displayed by the wild-type high-density population (Fig. 4). This result establishes that the circadian clock deregulation in nuclear YAP-enriched cells is caused by the YAP-TEAD transcriptional cascade.
Our discovery that nuclear YAP/TAZ disrupts Rev-erbα oscillations provides an entirely new layer of regulation for the circadian clock. Given the widespread role of YAP/TAZ in physiology and disease, the potential implications of this new mechanism are multiple. For example, this mechanism could contribute to explain the loss of circadian rhythms in ageing and cancer, two conditions in which YAP/TAZ become activated42–45. Moreover, given the role of REV-ERBα in endocrine function, fat storage, thermogenesis, mitochondrial activity, and cholesterol46, we expect this study to contribute new ways to understand the intimate relationship between circadian clocks and metabolism. Considering the role of YAP/TAZ as core mechanosensors, our findings provide a fundamental link between the largely disconnected fields of chronobiology and mechanobiology.
Methods
Cell culture
We cultured NIH3T3 mice fibroblasts in Dulbecco’s Modified Eagle Medium (DMEM) with pyruvate (41966-029, ThermoFisher) supplemented with 10% or 2% FBS (for cell lines maintenance or for experimental purposes, respectively; 10270-106, ThermoFisher), 100 U ml-1 penicillin, 100 μg ml-1 streptomycin and 292 μg ml-1 of L-glutamine (10378016, ThermoFisher). For all the experiments except the gap closure one, thymidine 2 mM (T9250-1G, Sigma) was added 24 hours before fixation or image acquisition. The experiment with conditioned medium was carried out with medium composed by 50% collected medium from a high-density population of cells after two days of culture and 50% fresh medium.
Cell lines and constructs
For the circadian analyses we generated a stable cell line expressing venus fluorescent protein with a nuclear localization signal (NLS) and a destabilization PEST domain under the promoter of Rev-erbα via lipofectamine-based transfection (Lipofectamine 3000, ThermoFisher) of the plasmid RevVNP, courtesy of Prof. Ueli Schibler20. Lentiviral particles carrying plenti6-H2B-mCherry (Addgene plasmid #8976647) were generated by transient transfection of Hek293T cells together with the corresponding envelope and packaging vectors. Several clonal lines were obtained by transduction of the RevVNP cell line with the mentioned particles and flow cytometry sorting (Beckman Coulter). For reproducibility reasons, one of them was selected and used for all the experiments.
The 5SA-YAP, 5SA-CCDEL-YAP, 5SA-S94A-YAP, 4SA-TAZ and control cell lines were obtained by transduction with retroviral particles generated in Hek293T cells and sustained selection with puromycin 2 μg ml-1 (A1113803, ThermoFisher). Specifically, the retroviral FLAG-5SA-YAP and FLAG-5SA-S94A-YAP plasmids were generated via restriction-based cloning from pQCXIH-Myc-YAP-5SA (Addgene plasmid #3309334 and pCMV-Flag-YAP-5SA/S94A (Addgene plasmid #3310340, respectively, and pBABE-PURO (Addgene plasmid #1764 48). The CCDEL deletion was introduced in the latter via overlapping PCR with the primers YAP-DEL-CC-HA-F and YAP-DEL-CC-HA-R as described in 30. The FLAG-5SA-S94A-YAP plasmid corresponds to the Addgene plasmid The FLAG-4SA-TAZ plasmid was a kind gift of Dr. Hyun Woo Park (Yonsei University, Seoul, Korea35). All the cell lines were checked by PCR, DNA sequencing and Western blot.
The cytoskeletal drugs used throughout the study were: jasplakinolide 1 μM (J4580-100UG, Sigma), cytochalasin D 1 μM (C8273-1MG, Sigma), latrunculin A 200 nM (L5163-100UG, Sigma) and para-nitro-blebbistatin 10 μM (DR-N-111, OptoPharma). The control conditions were performed with similar concentrations of DMSO (D8418-50ML, Sigma). All of them were added 24 hours before the initiation of the live imaging or fixation.
Time-lapse imaging of circadian oscillations of NIH3T3 fibroblasts
Time-lapse acquisitions were performed on a scanning confocal microscope (Nikon Ti Eclipse) equipped with thermal, CO2, and humidity regulation. 12-well or 6-well glass-bottomed plates were fixed on the stage and imaging was performed using a 10× objective (NA = 0.30, air) with focusing maintained by the Perfect Focus System (Nikon). The software NIS Elements was used to image every 15 minutes during experiments lasting for 3 to 5 days. Multiple wells, and multiple positions within the different wells, were imaged by phase contrast and in laser scanning mode using a 488nm and a 561nm excitation. For the 488nm acquisition two different sets of settings were used, to better capture the large variations in fluorescence signal seen in the RevVNPchannel: a low-intensity channel with low excitation power, and a high-intensity channel with 5 times higher excitation power. All confocal images were acquired by scanning 1024×1024 pixels on the objective’s field of view, yielding a pixel-size of 1.28um. Pixel dwell times were adjusted independently in each experiment.
Single-cell tracking
Tiff stacks of time-lapse microscopy images were analysed using Fiji49. Fluorescent nuclei visible in the high-intensity channel were tracked in space and time using the FIJI plugin TrackMate50. Within TrackMate, spots were identified using the LoG detector, after which the LAP tracker was used, with a maximum linking distance of 15 pixels and gap closing set to 2 frames. The resulting tracks were then filtered by length, keeping only those that lasted at least 60 hours. In the rare cases of cell-division, only one daughter cell was tracked. The tracking data was then used to additionally obtain tracks for the low-intensity channel. Finally, the position and intensity data for both channels were exported as .csv files.
Frequency Analysis of datasets of single-cell circadian oscillations
All analyses were performed with custom software created using MATLAB R2017a. To calculate a cell’s circadian strength firstly any missing time-points in each single-cell time-intensity track were interpolated. After this, each track was denoised by a low-pass filter, its mean value was subtracted, and the track was divided by its standard deviation. Then the tracks were analysed individually to remove any dying cells: these were defined as intensity traces whose standard deviation in the final 24 hours is less than 5% of the standard deviation during the rest of the experiment. A fast Fourier transform (FFT) was then applied, and the power spectral density (PSD) of each track was generated using the periodogram. All the PSDs were averaged, and the ensemble circadian frequency fc was identified as the coordinate of the peak in the interval [0.7, 1.3] days-1 in the average PSD. The single-cell PSDs were then analysed as follows: the integral of the PSD in the frequency interval W = [fc – 0.2, fc + 0.2] days-1 was divided by the total power of that individual track, giving what we termed the circadian strength of a single-cell oscillation. Note that to set the width of W we used pooled experiments with cells at high-density, identified the ensemble circadian frequency fc and measured the full width at half maximum of the corresponding peak.
Kymography
The denoised and rescaled single-cell tracks were further analysed by pairwise cross-correlation, and the lag values giving maximum cross-correlation were obtained. By virtue of the periodicity of the signals, these lags were restricted to the interval [-1,1] days. Tracks were ordered by decreasing circadian strength and the lags with respect to the median track were used to shift the other tracks and thus synchronize the dataset. These data were plotted as a kymograph of fluorescence intensity by time and track-number.
For the gap closure experiments, the nuclei in time-lapse fluorescence images were segmented using a combination of TrackMate’s spot detection algorithm and intensity thresholding. The kymographs were generated by averaging the intensity of the segmented nuclei in each frame along the direction parallel to the wound edge. These lines were then assembled into a kymograph representing fluorescence intensity as a function of time and distance from the centre of the wound.
Polyacrylamide gels
Polyacrylamide (PAA) gels with Young modulus E= 30 kPa – which matches approximately the stiffness of the majority of abdominal organs and the skin51 - were produced according to a previously published protocol52. A PBS solution containing 12% acrylamide (161-0140, Bio-Rad), 0.15% bis-acrylamide (161-0142, Bio-Rad); plus 0.05% ammonium persulphate (A3678, Sigma) and 0.05% tetramethyl ethylenediamine (T9281-25ML, Sigma) was prepared and allowed to polymerize between a coverslip and a glass-bottomed dish (Mattek). Alternatively, 3% acrylamide and 0.03% bis-acrylamide were used to make 300 Pa gels. The PAA gel surface was then incubated with a solution of 2 mg ml-1 Sulpho-SANPAH (4822589, ThermoFisher) under ultraviolet light for 5 min (wavelength of 365 nm at 5 cm distance). After that, the excess Sulpho-SANPAH was removed by three consecutive 3 min washes with PBS. A solution of fibronectin 0.1 mg ml-1 (F0895, Sigma) was added on top of the gels and left overnight at 4°C.
Gap closure
The gap closure experiments were performed following the protocol described in 53. Magnetic PDMS stencils consisting in two hollow regions separated by a barrier of 600-900 μm were obtained using a customized 3D-printed mould. The cured, autoclaved, passivated, and dried magnetic PDMS stencils were then deposited on a fibronectin-coated PAA gel polymerized in a glass-bottomed dish. After this, the dish was placed on top of a custom-made magnetic holder to keep the gasket in place and avoid medium leakage. The hollow regions were then filled with 40,000 cells contained in 120 μl of medium. After 1 hour, excess of cells was removed by 2 consecutive washes with medium. The attached cells were left overnight at 37°C before the removal of the magnetic PDMS gasket and the filling of the whole glass-bottomed dish with medium.
Immunostainings
Cells were fixed using paraformaldehyde 4% for 15 min and washed with PBS three times. Then, they were permeabilized with 0.1% Triton X-100 for 45 min, incubated with the primary antibodies for 90 min at room temperature, washed and incubated with the secondary antibodies for another 90 min at room temperature. After this step, Hoechst 33342 (H3570, ThermoFisher) was added at a concentration of 1 μg ml-1 for 10 minutes. Finally, cells were mounted in Mowiol reagent (81381, Merck). The buffers used during the whole procedure had fish gelatine 1.6% (v/v) as a blocking agent (G7765, Merck). The primary antibodies used were anti-NR1D1 (ab174309, abcam), anti-MKL1 (ab49311, abcam) and anti-YAP 63.7 (sc-101199, Santa Cruz). The secondary antibodies used were Alexa Fluor-488 anti-mouse (A-11029, ThermoFisher), Alexa Fluor-488 anti-rabbit (A-21206, ThermoFisher), Alexa Fluor-555 anti-mouse (A-21424, ThermoFisher) and Alexa Fluor-555 anti-rabbit (A-21429, ThermoFisher). All the antibodies were incubated in a 1:200 dilution. The acquisition of z-stacks (with a z-axis step of 0.7 μl) was done using a 40x water immersion LD LCI Plan Apo objective (NA = 1.2) in a scanning confocal microscope with Fast AiryScan (Zeiss LSM880) or a 60x oil immersion objective (NA = 1.40) in an inverted microscope (Nikon Eclipse Ti). Maximal projections of the stacks were obtained with Fiji before quantification of nuclear and cytosolic intensity levels or, for the case of YAP and MAL nuclear to cytosolic ratios, they calculated after measuring the mean intensity of two adjacent regions of identical size, inside and outside the nucleus as in 54 using the Hoechst image as a reference for the nuclear position.
Micropatterning
The micropatterning protocol was an adaptation of 55. Single glass-bottomed dishes (35 mm Dish, No. 0 Coverslip, 10 mm Glass Diameter, MatTek) were cleaned with ethanol 96%, dried, and oxidized by exposure to oxygen-plasma (PCD-002-CE, Harrick) for 30 s at 7.2 W. After that, a ring of PDMS with an inner diameter of 7 mm was attached to the coverslip to delimitate the micropatterning area, and that area was immediately filled with 30 μl of PLL-PEG solution (PLL (20 Kda)-g[3,5]-PEG(2), SuSoS) at a concentration of 0.1 mg ml-1 in PBS, which was incubated for 1 h at room temperature. Then, the solution was carefully removed, the surface was washed three times with PBS and 30 μl of photoinitiator PLPP (4-benzoylbenzyl-trimethylammonium chloride; Alvéole) at a concentration of 14.5 mg ml-1 were added. After that, micropatterning was performed in an inverted microscope (Nikon Ti Eclipse), using the PRIMO system controlled by the Leonardo software (Alvéole) and a 20X objective (CFI S Plan Fluor ELWD ADM, Nikon), directing a UV power of 1050 mJ mm-2 following the patterns predesigned in ImageJ. When the patterning was completed, filtered PBS (concentration 1X) was used to clean three times and the PDMS ring was removed. The whole glass surface was then treated for 10 min with a solution of fibronectin (F0895, Sigma-Aldrich) 100 μg ml-1 and Alexa Fluor 647 fibrinogen (F35200, ThermoFisher) 5 μg ml-1, which caused attachment of the coating proteins at both the patterns and the area where the PDMS ring had previously been placed. After three washes, the glass-bottomed dishes were stored, filled with PBS, at 4°C for a maximum of 48 hours before the experiments. 2*105 cells were seeded diluted in DMEM with 0.2% FBS to avoid cell attachment in unwanted areas. After one hour, excess of cells was removed by three intensive washes with the same medium. 24 hours before imaging or fixation the medium was replaced with 2% FBS-containing DMEM.
Statistical analysis
All violin plots were generated using violinplot, an open-source MATLAB script56. Two-sided Wilcoxon rank sum test were used when testing for statistical significance, and p-values smaller than 0.05 were considered as significant, unless specified otherwise.
Author contributions
J.F.A. and X.T. conceived the project. J.F.A., M.M., L.R. and I.A. performed experiments and analysed data. L.R. and J.B. developed software and analysed data. K.K., P.R.-C., S.M., and J.G.-O. contributed technical expertise, materials, and discussion. J.F.A., L.R. and X.T. wrote the manuscript. J.F.A. and X.T. supervised the project. All authors revised the completed manuscript.
Competing interests
The authors declare no competing financial interests
Acknowledgments
We thank S. Aznar-Benitah, G. Solanas, M. Uroz, A. Elósegui-Artola, C. Pérez-González, D. Zalvidea and all the members of the Roca-Cusachs and Trepat laboratories for their discussions and support. We thank A. Menéndez, S. Usieto and N. Castro for daily technical assistance. We also thank N. Montserrat, E. Martínez, U. Schibler, H.W. Park and Y.S. Choi for sharing some of the cell lines and plasmids used in this work. This paper was funded by The Generalitat de Catalunya (AGAUR SGR-2017-01602 to X.T, AGAUR Beatriu de Pinós 2014 BP-B 00105 to J.F.A., the CERCA Programme, and “ICREA Academia” awards to P.R-C. and J.G.-O.); Spanish Ministry for Science and Innovation MICCINN/FEDER (PGC2018-099645-B-I00 to X.T., BFU2016-79916-P and PID2019-110298GB-I00 to P. R.-C, and PGC2018-101251-B-I00 to J.G.-O.); European Research Council (Adv-883739 to X.T.); Fundació la Marató de TV3 (project 201903-30-31-32 to X.T.); European Commission (H2020-FETPROACT-01-2016-731957 to P.R-C. and X.T., H2020 Marie Sklodowska-Curie Actions MECHADIAN - IF/750557 to J.F.A.); La Caixa Foundation (LCF/PR/HR20/52400004 to to P.R-C. and X.T.); IBEC is recipient of a Severo Ochoa Award of Excellence from the MINECO. DCEXS-UPF is recipient of a Maria de Maeztu Award of Excellence from the MINECO.