Amotl1 mediates sequestration of the Hippo effector Yap1 downstream of Fat4 to restrict heart growth

Although in flies the atypical cadherin Fat is an upstream regulator of Hippo signalling, the closest mammalian homologue, Fat4, has been shown to regulate tissue polarity rather than growth. Here we show in the mouse heart that Fat4 modulates Hippo signalling to restrict growth. Fat4 mutant myocardium is thicker, with increased cardiomyocyte size and proliferation, and this is mediated by an upregulation of the transcriptional activity of Yap1, an effector of the Hippo pathway. Fat4 is not required for the canonical activation of Hippo kinases but it sequesters a partner of Yap1, Amotl1, out of the nucleus. The nuclear translocation of Amotl1 is accompanied by Yap1 to promote cardiomyocyte proliferation. We, therefore, identify Amotl1, which is not present in flies, as a mammalian intermediate for non-canonical Hippo signalling, downstream of Fat4. This work uncovers a mechanism for the restriction of heart growth at birth, a process which impedes the regenerative potential of the mammalian heart.

T he growth of the mammalian heart is critical for its contractile function. During development, cell proliferation underlies most of the growth, whereas increase in cell size (hypertrophy) predominates after birth 1 . The growth of the mammalian heart is controlled by the Hippo pathway [2][3][4][5] . Hippo kinases 3 and Hippo effectors 4,5 are required to regulate heart growth during development, and can also be manipulated to reactivate cardiomyocyte division in the postnatal heart, thus improving heart repair after injury 6,7 . However, upstream regulators of the Hippo pathway in this context have remained unknown.
In flies, the atypical cadherin, Fat, is an upstream regulator of Hippo signalling 8 , functioning as a tumour suppressor, and is also a regulator of tissue polarity 9,10 . Fat controls the canonical Hippo pathway by activating the Hippo kinases, which phosphorylate the Yap1 homologue, Yorkie, to sequester it in the cytoplasm, preventing transcriptional activation of target genes involved in cell proliferation and survival 11 . Another level of regulation of Yorkie downstream of Fat is provided by the FERM-domain protein Expanded, which requires Fat for its localization at the membrane 8 and thus sequesters Yorkie by direct binding 12 . Expanded is also localized at the membrane by the cell junction protein Crumbs 13 .
In mammals, Hippo signalling has similarly been shown to be modulated by cell junction proteins, and this is mediated in the early embryo (blastocyst) by the adaptor protein Angiomotin, which interacts with Yap1 (ref. 14). In the liver, Angiomotin can translocate to the nucleus together with Yap1, thereby activating cell proliferation 15 . Mouse mutants for the closest homologue of Drosophila Fat, Fat4, die at birth with polarity defects, exemplified by the kidney or cochlea 16,17 . Fat4 mutations are also associated with abnormal cortical development in Van Maldergem syndrome, due to the misregulation of Yap1 which affects cell differentiation 18 . However, in the cortex, kidney or liver of Fat4 mutants, no activation of the Hippo kinases has been reported 12,[16][17][18][19] . The mouse Fat4 intracellular domain cannot rescue the growth phenotype of fat mutant flies 20 , raising the question of whether and how Fat4 could regulate the Hippo pathway in the mouse.
We show that Fat4 is required to restrict heart growth at birth, by decreasing the activity of Yap1. This regulation does not involve the phosphorylation cascade of Hippo kinases, but rather the scaffold protein Angiomotin-like 1 (Amotl1), a partner of Yap1 not present in flies. When it is not sequestered by Fat4, Amotl1 can translocate to the nucleus together with Yap1, thereby activating cardiomyocyte proliferation.

Results
Fat4 is required to restrict heart growth. In the heart, we assessed the phenotype of Fat4 À / À mutants at birth and observed an abnormal flattened apex (Fig. 1a) and 1.5-fold thicker ventricular myocardium and interventricular septum (Fig. 1b,c), compared with control hearts. This excessive tissue thickness may result from defective orientation or excessive rate of growth. We had shown previously that clonal growth of the heart is oriented 21,22 , and we, therefore, analysed the orientation of cell division in Fat4 À / À mutants. Whereas in the mouse kidney orientation of cell division is disrupted when Fat4 is absent 17 , we found that cell division was still biased in the plane of the heart surface and still coordinated in the left ventricle of Fat4 À / À hearts as in controls ( Supplementary Fig. 1). We then investigated whether the excessive growth of Fat4 mutant hearts was due to increased cell proliferation, and found that the percentage of cardiomyocytes positive for the mitotic marker, phosphorylated histone H3, was significantly higher in the absence of Fat4 (Fig. 2a,b), whereas the effect was milder in non-cardiomyocytes      Fig. 2b). Conditional deletion of Fat4, in cardiac progenitors that express Mesp1, resulted in an increased percentage of cardiomyocytes, whereas the number of nuclei per cell was unchanged (Fig. 2c, Supplementary Fig. 2c). This supports the conclusion of increased cardiomyocyte proliferation. Conditional deletion of Fat4 in non-cardiomyocytes, using Wt1-Cre, did not change the thickness of the ventricular wall or the percentage of mitotic cells ( Supplementary Fig. 2d-f), indicating that Fat4 is required in cardiomyocytes. Analysis of gene expression showed an increase for cell cycle (Ccna2, Cdc20), cytokinesis (Aurkb) and anti-apoptotic (Birc2/5) genes and a decrease of Cdkn1b transcripts implicated in cell cycle exit, in Fat4 À / À compared with control hearts (Fig. 2d). Heterozygotes also show transcript upregulation, although they do not have a detectable heart phenotype, indicating compensation at the level of the proliferation gene network dependent on Fat4 dosage. In primary cultures, knockdown of Fat4 ( Supplementary Fig. 2g-k) significantly enhanced the percentage of proliferating Ki67positive and replicating EdU-positive cardiomyocytes ( Fig. 2e and Supplementary Fig. 2l,m). We next examined whether Fat4 also affects cell size. By measuring the cross sectional area of cardiomyocytes, we found a significant increase of cell size in Fat4 À / À compared with control hearts (Fig. 2f,g). In agreement with a hypertrophic phenotype, reverse transcription quantitative PCR (RT-qPCR) revealed a de-regulation of classical markers of heart hypertrophy 23 , corresponding to the activation, in Fat4 À / À mutant hearts, of genes normally expressed at fetal stages (Acta1), whereas genes normally expressed at adult stages (Myh6) are downregulated (Fig. 2h). The early marker of heart hypertrophy, Nppb 24 , was strikingly increased (11-fold) in Fat4 À / À mutant hearts, whereas the marker of wall stress, Nppa, was not. Our data show that Fat4 is required to restrict heart growth at birth, with an effect on the proliferation and hypertrophy of cardiomyocytes.
Fat4 modulates an effector of Hippo signalling Yap1. Since we observed misexpression of genes ( Fig. 2d) previously shown to be targets of the Hippo pathway in the control of cardiomyocyte proliferation 3,4 , we examined whether Hippo signalling was impaired in Fat4 mutant hearts. The classical targets of Yap1, CTGF and Cyr61, were significantly overexpressed (Fig. 3a), showing that the transcriptional activity of Hippo effectors is increased in Fat4 À / À mutant hearts. Consistent with this, Yap1 was relocalized to the nucleus of cells in which Fat4 was absent in vivo ( Fig. 3b) or downregulated in primary cultures of cardiomyocytes (Fig. 3c,d, Supplementary Fig. 3a,k,l), and  Fig. 3g-j). A phospho-resistant form of Yap1 was relocalized to the nucleus when Fat4 was downregulated (Fig. 4g,h), further supporting the conclusion that the nuclear translocation of Yap1 downstream of Fat4 is not mediated by a change of phosphorylation. The phenotype of Fat4 mutants differs from that resulting from impairment of the canonical Hippo pathway. The onset of excessive myocardial growth in Yap1 gain-of-function mutants 5 or in Hippo kinase-deficient hearts 3 is already seen at embryonic stages (E10.5-E11.5). In contrast, the phenotype of Fat4 À / À hearts is detected much later, from E18.5, although Fat4 is expressed throughout heart development ( Supplementary Fig. 4a-c). The Wnt pathway, which was previously shown to interact with the canonical Hippo pathway, was not found activated in Fat4 À / À mutant hearts ( Supplementary Fig. 4d,e). These observations show that Fat4 is a later modulator of the Hippo pathway and suggest that a mechanism other than that of phosphorylation by the Hippo kinases, regulates the nuclear localization of Yap1, downstream of Fat4.
Amotl 1 mediates Fat4 signalling. Hippo signalling is modulated by cell junction proteins 25 . When we labelled cardiomyocytes with junction markers, we observed that cell junctions  Supplementary Fig. 3b were abnormal in Fat4 À / À hearts. N-cadherin (Cadh2) or Plakophilin2 (Pkp2) staining were broader and less focalized than in control hearts (Fig. 5a). By electron microscopy, the electron dense desmosomal material was more diffuse in cardiomyocytes of Fat4 À / À hearts (Fig. 5b). In agreement with abnormal cell junctions, cardiomyocytes had a rounder shape in Mesp1-Cre conditional Fat4 mutant hearts ( Supplementary Fig. 4f). These observations suggest that a junctional protein may be involved in the effect of the atypical cadherin Fat4 on Hippo effectors. We focussed on the adaptor protein, Angiomotin, which had been shown to interact with Yap1 and to co-translocate to the nucleus, where the complex modulates transcription 15 . This co-translocation Amotl1 c d e f  can occur independently of Yap1 phosphorylation 26 . In the heart, we observed that Angiomotin-like 1 (Amotl1), another member of the Angiomotin family, is predominantly expressed ( Supplementary  Fig. 4g). It is present near the membrane of cardiomyocytes, in clusters where Yap1 is also detected, which is compatible with a localization at cell junctions (Fig. 5c). Strikingly, Amotl1 was relocalized to the nucleus when Fat4 was absent in vivo (Fig. 5d,e) or downregulated in primary cultures of cardiomyocytes ( Supplementary Fig. 3o). When we interfered with Amotl1 ( Supplementary Fig. 2g-k) as well as Fat4 expression, we reversed the increased nuclear localization of Yap1 ( Supplementary Fig. 3k,l) as well as the increased cell proliferation observed when Fat4 expression alone is down-regulated (Fig. 5f). Interference with Amotl1 expression alone shows that it is required for the proliferation of cardiomyocytes, with an effect dependent on the presence of Fat4. On the contrary, forcing Amotl1 to translocate to the nucleus, by addition of a nuclear localization signal, resulted in co-accumulation of Yap1 in the nucleus (Fig. 6a, Supplementary  Fig. 3m-o) and stimulation of cardiomyocyte proliferation (Fig. 6b).
Interference with Yap1 expression shows that the proliferative effect of nuclear Amotl1 is dependent on Yap1. Both Fat4 and Amotl1 are known to interact with the scaffold multi-PDZ domain protein, Mpdz, also known as Mupp1 (refs 27,28). We now show, in transfected HEK293 cells, that Amotl1 interacts physically with Fat4, as well as Yap1 (Fig. 6c), and that the three proteins co-localize in transfected cardiac cells (Fig. 6d). This is consistent with Amotl1 release from cell junctions when Fat4 is absent. These results identify Amotl1 as a modulator of the Hippo pathway in cardiomyocytes, and show that it is prevented from entering the nucleus by sequestration in a Fat4 complex, thus restricting Yap1 mediated tissue growth.

Discussion
Our observations of Yap1 activity in Fat4 mutants, as well as the suppression of Fat4-dependent hyperproliferation by reduced Yap1 or Amotl1 expression, argue that Fat4 is an upstream regulator of Yap1 in the mouse heart, and that it triggers a non-canonical modulation of Hippo signalling. This pathway probably implicates a non-phosphorylated form of Yap1 bound to Amotl1 in the cytoplasm. When Amotl1 is not sequestered at cell junctions with Fat4, we show that it is an intermediate, which bypasses the Hippo kinases, to regulate the nuclear translocation of Yap1. Our results suggest that, depending on the available partners (for example, Fat4), Amotl1 has a role as an inhibitor or facilitator of Yap1. This is in agreement with previous reports showing contradictory roles of Angiomotin in different cell types 15,29,30 . Amotl1 may also directly contribute to the transcriptional activation of target genes, by analogy with Amot in the liver 15 . It remains to be seen whether Tead, a transcription factor that interacts with Yap1, is implicated in this context. The model that we propose is shown in Fig. 7. Amotl1 has no homologue in flies, which explains why the intracellular domain of Fat4 cannot rescue the growth phenotype of fat mutant flies 20 . The function of mouse Amotl1 is similar to that of Drosophila Expanded, a FERM-domain protein which requires Fat for its localization at the membrane 8 and which can directly sequester Yorkie out of the nucleus, independently of canonical Hippo signalling 31 . The mammalian homologue of Expanded, Frmd6, has lost the C-terminal domain of interaction with Hippo effectors, which supports an evolutionary switch in the regulation of Hippo signalling by Fat 19 . Although Fat signalling is implemented differentially between mouse and fly, the function of this cadherin is well conserved, with a dual effect on tissue polarity 17 and also, as we now show, on tissue growth. The effect of Fat4 depends on the cellular context. In the heart, we show that Fat4 regulates tissue growth, rather than polarity. This has also been observed in the cortex 18 , whereas in other organs, such as the kidney or the cochlea, Fat4 is a regulator of tissue polarity 16,17 . Our findings on Fat4 mutants uncover a mechanism that restricts heart growth at birth. Central to this mechanism is the adaptor protein Amotl1, which can shuttle from cell junctions to the nucleus, transporting the transcription co-factor Yap1. Whereas the Hippo pathway was shown to be required at embryonic stages of heart development 3,5 , Fat4 is a later modulator exerting its role at birth. It remains to be established how the Fat4/Amotl1-dependent pathway is activated and what is its relative importance to regulate Yap1, in comparison with canonical Hippo signalling. Canonical Hippo signalling is also modulated by cell junctions in cardiomyocytes, where remodelling of the intercalated discs activates Hippo signalling, with pathological consequences leading to arrhythmogenic cardiomyopathy 32 . Our results showing the importance of Fat4 for the integrity of cell junctions is in agreement with the finding that Fat4 interacts with several junction proteins 12 . Fat4 mutant hearts display hypertrophy, in addition to increased cell proliferation. Although hypertrophy can potentially be induced by Yap1 (refs 2,26), other studies 4,5 would suggest that this is an indirect effect. Due to its positive effect on cardiomyocyte proliferation, Hippo signalling has been shown to be important for prolonging the regenerative potential of the mouse heart 6,7 , which normally ceases during the first week after birth 33 . However, Yap1 is less efficient in promoting cardiomyocyte proliferation at postnatal stages than it is during development, which suggests that other factors block Yap1 activity at later stages. We now identify upstream regulators of Yap1 activity in the heart and anticipate that manipulating the Fat4 pathway will facilitate the reactivation of cardiomyocyte proliferation induced by phospho-resistant Yap1 (ref. 7) or Hippo kinase deficiency 6 . This has major therapeutic implications for the repair of the failing human heart.

Methods
Animal models. The Fat4 mouse mutant line 17 36 and backcrossed in the 129S1 genetic background. Fat4 À / À mutants die at birth, whereas Fat4 flox/ À ; Mesp1 Cre/ þ survive. Animal procedures were approved by the ethical committee of the Institut Pasteur and the French Ministry of Research. For histological analysis, hearts were excised, incubated in cold 250 mM KCl, fixed in 4% paraformaldehyde, embedded in paraffin in an automated vacuum tissue processor and sectioned on a microtome (10 mm). Male and female heart samples were mixed. For immunofluorescence studies, hearts were fixed in 0.5% paraformaldehyde, embedded in gelatine/sucrose, frozen in cold isopentane and sectioned on a cryostat (10 mm). For the quantification of tissue growth, images of sections stained with Haematoxylin Eosin or Hoechst were acquired. A polygonal mask was drawn in order to isolate the two ventricles from the atria. The green channel (with highest contrast) of the resulting image was inverted, thresholded and segmented using Connected Component analysis. The resulting regions were sorted, retaining the myocardial tissue and excluding blood speckles inside the ventricles, to compute the total area of the ventricles. The penetrance of the myocardial excessive growth was 75% (n ¼ 8). Unless otherwise specified, the image analysis was done using the Icy software 37 .
RT-qPCR. Complementary DNAs were reverse transcribed from RNAs extracted in TRIzol from cell cultures and isolated hearts using the Quanti-Tect kit (Qiagen) and Superscript II Reverse Transcriptase (Life Technologies) respectively. Quantitative PCR was carried out on a StepOne System (Life Technologies) using Fast Start SYBR Green Master (Roche). Quantification of gene expression was calculated as R ¼ 2 DCt(control-target) , with Gapdh used as a control. Primers were designed using the NCBI Primer-BLAST software. Primer sequences are listed in Supplementary Table 1.
For flow cytometry analyses, cultures were dissociated to single cell suspensions by trypsin, fixed and permeabilized in eBioscience buffer. Proliferating cardiomyocytes were detected by immunostaining with primary antibodies against Tnnt2 (1/200, ab64623) and Ki67 (1/40, BD 556027) and counted on a BD LSRFortessa Cell Analyser cytometer. Gates were set according to isotype control antibodies (1/100, sc-3887; 1/40 BD556027). At least 900 cells were counted per condition. Alternatively, cardiomyocytes were exposed to EdU during 30 h and counted after immunofluorescence (at least 80 cells per condition). For overexpression experiments, cardiomyocytes were transfected using Lipofectamine 2,000 with Fat4-DECD-Flag (encoding Fat4 depleted for the extracellular domain and for the last C-terminal 297 nucleotides, derived from the constructs in ref. 12 Quantification of cell proliferation. The PH3 channel was thresholded and segmented using Connected Component analysis, filtering objects under a minimum size of 16 mm 3 in order to eliminate non-specific signals. The myocardial volume of the multi-z scan was estimated by manually outlining the myocardial surface in the median Z-slice and computing the area. The total number of cardiomyocyte (a-actinin-positive) nuclei in the scan was estimated by manually counting the number of nuclei in a 200 pixels Â 200 pixels window extending over all the Z-slices, and extrapolating to the total myocardial volume. More than 1,500 nuclei were counted per embryo.
Quantification of cell size. Images of cardiomyocyte transverse sections labelled with Caveolin3 (Cav3) were acquired systematically in the interventricular septum. Cell contours were drawn manually to compute cell area using ImageJ. At least 40 cells were counted per embryo.
Quantification of protein localization. The best in-focus Z-slice of the Hoechst channel was chosen for in vivo cells, whereas in vitro images were Z-projected. The analysis involved three image processing steps. In the first step, the myocardial (Tnnt2-positive) cells were segmented using Connected Component analysis applied after a Z-projection (sum) and thresholding of the Tnnt2 channel; alternatively, in vitro transfected cells were outlined manually. In the second step, the nuclei were segmented by thresholding after application of a Gaussian filter (in vivo), or by the 'Active Contours' plugin (in vitro). In the third step, the average intensity (total intensity divided by the area of the segmented object) of the protein of interest (PI) in the Tnnt2-positive cells (PItot) and in their nuclei (PInucl) was measured by multiplication of the PI channel with the respective binary images obtained in steps (1) and (2). Strong cells were defined as cells in which PInucl is higher than two standard deviations above the mean PInucl of control cells. The nuclear/cytoplasmic ratio was computed as: PInucl / PIcyto ¼ PInucl/(PItot À PInucl).
Quantification of the mosaicism of Cre expression. The nuclei were first segmented by a K-Means classification of the histogram from the Hoechst channel (HK-Means plugin in ICY), extracting objects within a minimum/maximum size range. The average GFP and Tomato signals were then quantified in a box surrounding each nucleus (but excluding the nucleus itself). These two measures were fed into a K-Means clustering algorithm implemented in Matlab (Pattern Recognition and Machine Learning Toolbox), with k ¼ 2, effectively separating the GFP high /Tomato low from the GFP low /Tomato high cells.
Immunoprecipitation and western blots. HEK293 cells (Q-BIOgene AES0503) were transfected with Lipofectamine with the plasmids Amotl1-HA 28 and Flag-Fat4-DECD and cultured for 48 h. Proteins were extracted in a lysis buffer (150 mM NaCl, 5 mM EDTA, 10 mM Tris pH 7.5, 10% glycerol, 1% NP-40) in the presence of protease inhibitors. Immunoprecipitation of protein extracts was performed using a standard protocol based on magnetic beads coupled to bacterial protein G, an immunoglobulin-binding protein. One microgram of anti-HA antibody (Roche 3F10) per 500 mg of protein extracts was used. Proteins were eluted in Laemmli buffer. An isotype antibody (IgG sc38882) was used as a negative control of immunoprecipitation.
Image registration. In order to obtain a full inferior view of the two ventricles at E10.5, confocal scans of the left ventricle, interventricular region and right ventricle were stitched together. The envelopes of the stitched images were computed by Active Mesh segmentation 42 . Ten such envelopes were used to compute an average envelope (referred to as the template), minimizing the deformation distances between the template and the envelopes, plus a residual mismatch cost. The metric distance was built on a group of smooth invertible deformations (that is, diffeomorphisms 43 ). The axial data from each image were then transported through the deformation between the original envelope and the template, as described by the Jacobian matrix of the diffeomorphism (that is, the matrix of partial derivatives of the deformation, a three-dimensional generalization of the gradient). Using the polar part of the Jacobian was required to avoid improvement of the axial correlation.
Quantification of tissue polarity. Whole mount immunostaining was carried out on E10.5 isolated hearts using Scrib and Cadh2 antibodies to detect membranes and Aurkb antibody to detect cytoplasmic bridges. The nuclei and cytoplasmic bridges were segmented with the ICY software, using Filter Toolbox on the nuclear channel and Anisotropic Filter on the membrane channel, followed by subtraction of the two channels to clearly separate nuclei, and three-dimensional Active Mesh to extract the nucleus contour and centroid. On the Aurkb channel, Thresholder, Anisotropic Filter and Adaptive Histogram Equalization were applied. Sister cells were automatically detected using a scoring function of the distance from the bridge extremity and the direction of the bridge 21,44 . The axis of cell division was taken as the axis joining the two centroids of sister cell nuclei. For each genotype, at least three E10.5 embryonic hearts were registered, in order to pool the axial data on a common template. The planar component of each axis of cell division was extracted by projection over the template envelope. The threshold eigenvalue for each region size, E 1(5%) , which was obtained by a bootstrap method 21 , was calculated both before and after the diffeomorphic transport of the axes, and the highest value was retained to compensate for any spurious improvement of the alignment due to the transport. Contour maps of axial coordination were produced by selecting the region, containing at least 50 axes, with the highest eigenvalue E 1 (core region), listing all regions that both included the core region and had an eigenvalue E 1 4E 1(5%) , and drawing these regions on the template, with contour values equal to the ratio E 1/ E 1(5%) . grids covered with formwar and stained with lead citrate. The sections were examined in a JEM 2000FX (Jeol, Japan) electron microscope and recorded using a Gatan DualVision 300W CCD (charge-coupled device) camera (Gatan Inc., USA).
ImageStream. P14 hearts were collected, minced and flash frozen in liquid nitrogen 45 . The defrosted tissue was fixed in 4% paraformaldehyde, digested with 3 mg ml À 1 collagenase type II in HBSS and filtered using a 100 mm cell-strainer. Staining of isolated cells was performed with the BD Cytofix/Cytoperm Fixation/ Permeabilization Kit, using anti-sarcomeric a-actinin (1/600, Sigma) and DRAQ5 nuclear stain. Data acquisition was performed using an ImageStreamX cytometer with INSPIRE software (Amnis). Files were collected with a cell classifier applied to the brightfield channel to capture events larger than 100 mm. At least 23,000 cell events were acquired for each sample and all images were captured with the 40 Â objective. Data analysis was performed with IDEAS software (v6.0, Amnis). Images were compensated using a matrix generated by single-stained samples acquired with identical laser settings in the absence of brightfield illumination. The analysis was restricted to in-focus single cells and to intact cardiomyocytes, selected as actinin and DRAQ5 double positive. An object mask was created on the brightfield channel and the aspect ratio was defined as the ratio between the minor and major cell axis. The number of nuclei per cell was assessed using the DRAQ5 images, in at least 350 cells per heart.
Statistics. Sample size was chosen in order to ensure a power of at least 0.8, with a type I error threshold of 0.05, in view of the minimum effect size that was looked for. The sample size was calculated using the observed variance of the wild-type mice for the phenotype considered. Sample outliers were excluded according to the Thompson Tau test. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment.
Comparisons of centre-values were done on either the average or the geometrical mean when ratios were compared. An analysis of variance was systematically calculated when more than two centre-values were compared, and Tukey-Kramer's two-sided test was used for the assessment of bilateral significance. Otherwise, a Student two-sided test was used. When n410, normality was checked by a Kolmogorov-Smirnov test or by visualization of the distribution. When the test was not positive, a Wilcoxon U-test was used.
Data availability. The authors declare that all data supporting the findings of this study are available within the article and its Supplementary Information files or from the corresponding author upon reasonable request.