To establish whether melanoma cells exhibit MA channel activity, metastatic WM266-4 melanoma cells (originally isolated from a secondary tumour) were cultured on uncoated pillar arrays made of polydimethylsiloxane (PDMS). Mechanical stimuli were applied directly to cell-substrate contact points by physically deflecting a single pilus subjacent to the cell (Figure 1A,B) and the electrical response of the cell was monitored using whole-cell patch-clamp. Deflection-activated currents were measured in all WM266-4 cells (10/10) and the current amplitude increased with increasing stimulus size (Figure 1C). Variable inactivation kinetics were measured (Figure 1—figure supplement 1, Figure 1—source data 1), as also previously demonstrated for PIEZO1-mediated currents activated by substrate deflection (Poole et al., 2014; Sianati et al., 2019). The reversal potential, as determined from a current-voltage relationship for the peak MA current, was +6.6 mV, indicating that the underlying current was passed by a non-selective cation channel (Figure 1D). Stimulus-response plots were generated by calculating the precise pillar deflection for each applied stimulus (Figure 1E). We additionally tested whether MA currents were activated in WM115 melanoma cells (isolated from the primary tumour from the same patient as WM266-4). Larger deflections were required to activate currents in WM115 cells, compared to WM266-4 (Figure 1—figure supplement 1). These data demonstrate that displacements at the interface between melanoma cells and their substrate evoke MA currents.

To examine whether MA channel activity is correlated to cell migration speeds, we performed pillar-array experiments using LM511, a substrate that supports the migration of metastatic melanoma cells (Oikawa et al., 2011). We first confirmed that LM511 promotes increased migration of WM266-4 cells using transwell assays, compared to other LM isoforms (Figure 1F). We then repeated the analysis of MA channel activity in WM266-4 cells cultured on pillar arrays coated with LM511 and noted that current kinetics were unchanged (Figure 1—source data 1). However, MA currents were more sensitive when evoked in WM266-4 cells attached to LM511 (Figure 1E). Under these conditions, current saturation occurred within the stimulus range, allowing us to use a Boltzmann sigmoidal fit to determine the MA current sensitivity. Half-maximal activation of MA currents was seen with approximately 18 nm of substrate deflection (Effective deflection ED50; standard error = 20.5 nm). These data indicate a correlation between migratory properties and the MA current sensitivity to deflections applied at cell-substrate contact points. The robust MA current activation observed in cells cultured on LM511 also provided an excellent system to investigate the molecules required for this mechanoelectrical transduction.

PIEZO1 is an obvious candidate for mediating this activity, in particular because the biophysical characteristics of the observed deflection-activated currents are consistent with those previously described for this channel (Poole et al., 2014). To directly test if PIEZO1 mediates the MA current in WM266-4 cells, we knocked down PIEZO1 expression and examined whether these currents were still detectable. In PIEZO1 knock-down cells cultured on LM511-coated pillar arrays, MA currents were activated in response to pillar deflection in 10/10 cells measured and the resulting stimulus-response curves were similar to controls (Figure 1—figure supplement 2). Similarly, treatment with Ruthenium Red (RR, a channel blocker that inhibits PIEZO1 and TRP channels) did not inhibit these MA currents (Figure 1—figure supplement 2). From these data we conclude that PIEZO1 is unlikely to be the MA channel responsible for the current measured in WM266-4 cells.

To identify novel MA channels in WM266-4 cells, we undertook a proteomic-based strategy. Given that both intracellular (Poole et al., 2014; Zhang et al., 2017) and extracellular proteins (Chiang et al., 2011) can tune the sensitivity of MA channels, we analysed the proteome of WM266-4 cells rather than taking a comparative proteomic approach (Supplementary file 1). Two known non-selective cation channels were identified, PIEZO1 and TRPV2. However, RR is a channel blocker of both PIEZO1 (Coste et al., 2012) and TRPV2 (Caterina et al., 1999), indicating that neither likely mediates the deflection-evoked currents in WM266-4 cells (Figure 1—figure supplement 2). We then examined the proteomics data for proteins of unknown function with four or more predicted transmembrane (TM) domains. We prioritised the investigation of Elkin1 due to its expression in melanoma cells but not healthy melanocytes, its expression in additional mechanosensitive cells (Alveolar Type II cells) and its upregulation in additional human cancers (Human Protein Atlas [Uhlén et al., 2005] available from www.proteinatlas.org). We generated miRNA constructs targeting Elkin1 and found that knockdown of Elkin1 transcript resulted in a dramatic reduction in MA currents to deflections up to 1000 nm (Figure 2A,B). These data suggested that Elkin1 contributes to MA currents in melanoma cells.

Three human isoforms (representing splice variants) of Elkin1 have been identified: isoforms 1 and 3 (555 and 494 aa respectively), contain six predicted TM domains (Figure 2C). Isoform 2 (181 aa) does not contain any predicted TM domains and was not examined in this study. We cloned hsElkin1-iso1 and hsElkin1-iso3 from WM266-4 cDNA and generated C-terminal GFP fusion constructs. We confirmed the plasma membrane localisation of these two isoforms in transiently transfected HEK-293T cells using cell-surface biotinylation followed by Western blot analysis (Figure 2D, Figure 2—figure supplement 1). Both isoforms were present at the plasma membrane, as well as the intracellular fraction. To further examine the subcellular distribution of hsElkin1, we transiently expressed these plasmids in WM266-4 cells and imaged the cells using total internal reflection fluorescence (TIRF) microscopy, which limits fluorescence excitation to molecules near the cell-substrate interface. hsElkin1-iso1-GFP localised to both the membrane and a dispersed population of dynamic foci, while hsElkin1-iso3-GFP assembled into structures that were co-labelled by Lifeact-mCherry. (Figure 2E–J; Figure 2—videos 1 and 2). Laser-scanning confocal imaging to visualise hsElkin1-iso1-GFP throughout the cell revealed that this protein also localised to the Golgi apparatus (as previously described [Hirata et al., 2015]): in contrast, hsElkin1-iso3-GFP was enriched in actin-based ruffles that were present at the cell periphery (Figure 2—figure supplement 2). The presence of Elkin1 in the plasma membrane, as described here, is consistent with Elkin1 forming an integral component of a mechanoelectrical transduction pathway.

To test the hypothesis that Elkin1 contributes to mechanoelectrical transduction, we examined whether its expression in a heterologous cell system is sufficient to reconstitute MA currents. Un-tagged Elkin1 isoforms were overexpressed in the HEK-293T P1KO cell line (Lukacs et al., 2015), which lacks functional PIEZO1. The control cells exhibited no current activation in response to stimuli within our deflection range of 1–1000 nm (0/8 cells). However, mechanically activated currents were detected in response to similar stimuli in HEK-293T P1KO cells expressing either hsElkin1-iso1 or hsElkin1-iso3 (8/8 and 9/9 respectively) (Figure 3A,B). The observed current kinetics were consistent with direct mechanical activation (Figure 3—source data 1) and the mechanically evoked currents measured in WM266-4 cells (Figure 1—source data 1). We confirmed that we could mechanically evoke currents in a second cell line lacking PIEZO1, N2a ^Piezo1-/- (Moroni et al., 2018), where 6/8 cells expressing hsElkin1-iso1 responded to pillar deflection compared to 4/10 control cells (Figure 3—figure supplement 1).

To investigate whether cells expressing Elkin1 were sensitive to alternative modes of mechanical stimuli, we used cellular indentation and high-speed pressure-clamp (HSPC). The HEK-293T P1KO cell line is unresponsive to indentation, (0/6 cells responding) as previously described (Dubin et al., 2017; Figure 3C,D); in contrast, MA currents were evoked in cells expressing either human Elkin1 isoform (hsElkin1-iso1 6/6 cells, hsElkin1-iso3 5/5 cells responding) (Figure 3C,D). Indentation-evoked currents were only measured in a small fraction of N2a ^Piezo1-/- cells expressing hsElkin1-ios1 or hsElkin1-iso3 (1/10 and 1/12 responding respectively compared to 0/8 control cells, Figure 3—figure supplement 1). When membrane stretch was applied using high-speed pressure-clamp (HSPC), no currents were evoked in negative controls or HEK-293T cells overexpressing human Elkin1 (negative pressure: hsElkin1-iso1 0/11, hsElkin1-iso3 0/13, positive pressure: hsElkin1-iso1 0/5, hsElkin1-iso3 0/5 responding) (Figure 3E). In contrast, both positive and negative pressure evoked currents in all positive controls where PIEZO1 was expressed (Figure 3—figure supplement 2). From these data we conclude that the expression of Elkin1 is sufficient to confer MA channel activity to substrate deflection and cell indentation stimuli in a heterologous system lacking endogenous PIEZO channels, but not to membrane stretch.

The Mus musculus homolog of Elkin1-iso1 shares 93% sequence identity at the protein level with the human variant. However, we did not observe robust MA currents in HEK-293T P1KO cells following expression of mouse mmElkin1-iso1 (5/8 cells responding, Figure 4A,B), indicating a difference in sensitivity between these homologous proteins. These data were confirmed in the N2a ^Piezo1-/- background (Figure 3—figure supplement 1), with 11/26 cells responding. Alignments between the protein sequences (Corpet, 1988) revealed that the divergence in amino acid sequences primarily resides in the N-terminus of the proteins (Figure 4—figure supplement 1). However, expression of N-terminal truncation mutants of hsElkin1 conferred similar MA currents to those of the wild type (WT) protein (Figure 4—figure supplement 1), suggesting that this region is not required for MA currents. To further characterise the differences between the mouse and human proteins, we focused on two non-conservative changes within the region containing the six predicted TM domains. Introducing the two human residues into the mouse polypeptide (mmElkin1-iso1-F271L-N292G), led to robust MA channel activity, in contrast with WT mmElkin1-iso1. Conversely, introducing either of the mouse residues into the human polypeptide (hsElkin1-iso3-L210F or hsElkin1-iso3-G231N) led to a reduction in the amplitude of the MA currents measured (Figure 4A,B), despite the fact that these variants were still present in the plasma membrane (Figure 4—figure supplement 2). These data indicate that two residues can account for the differences in activity between the mouse and human proteins, and that these two residues are required for the activation of robust Elkin1-dependent currents by substrate deflection.

Having demonstrated that human Elkin1 is involved in mechanoelectrical transduction in melanoma cells, we sought to investigate whether the presence of this protein influences cell migration. To address this question, we used CRISPR/Cas9 to gene-edit cells such that no functional Elkin1-iso1/3 were expressed (Figure 5—figure supplement 1). Duplicate clonal populations of wild-type (WT) and Elkin1-knockout (KO) cells were isolated. Using pillar arrays, we observed reduced deflection-activated currents in the Elkin1-KO clones compared to the WT (Figure 5A,B). A residual current at large deflections was still present in some KO cells; we hypothesise that this effect is due to minor compensation from PIEZO1 (Supplementary file 1 and as noted previously [Rocio Servin-Vences et al., 2017]) or the activity of an as-yet-unidentified MA channel. Nevertheless, these data confirm that Elkin1 is required for sensitive MA channel activity in WM266-4 cells.

We first tested whether Elkin1-dependent mechanotransduction regulates cell migration using a transwell assay. While transmigration of the WT clones was not different from control populations, both Elkin1-KO clones exhibited reduced transmigration (Figure 5C). This phenotype could be rescued by the transient overexpression of wild-type hsElkin1-iso3. In contrast, overexpression of hsElkin1-iso3 L210F (Figure 5D), a variant associated with significantly reduced MA currents (Figure 4B), did not rescue the transmigration defect. We additionally generated an Elkin1-KO clone from the A375 melanoma cell line (which has previously been shown to express functional PIEZO1 [Hung et al., 2016]). The A375 Elkin1-KO cells exhibited reduced transmigration, compared to the A375 WT cells (Figure 5E), indicating that the link between Elkin1 activity and migration is not restricted to WM266-4 cells and can be measured in a cell line where PIEZO1 is functionally active. We therefore conclude that disruption of Elkin1 not only impairs mechanosensitivity, but also cell migration.

We expanded our analysis to investigate how Elkin1 regulates different modes of migration. To examine how Elkin1 mediates the unconfined migration of individual cells on a 2D surface, we plated cells on dishes coated with LM511, and observed their movement using phase-contrast and epi-fluorescent microscopy. A representation of the migration tracks shows that the WT clones migrated further over the course of the experiment (Figure 6A). To test whether this effect was exclusive to migration on LM511-coated substrates, we repeated the experiments on dishes coated with poly-L-lysine (PLL) (Figure 6B). The Elkin1-KO clones exhibited reduced track mean speed on both substrates (Figure 6C) and a reduced distance migrated from origin (calculated as the Euclidean distance, Figure 6D).The percent reduction in the mean of the measured track mean speed was similar for experiments conducted on LM511 (Elkin1-KO cells 29% reduction in the mean) to PLL (Elkin1-KO cells 32% reduction). These data demonstrate that Elkin1 modulates cell migration in unconfined environments. To investigate the impact of Elkin1 on confined migration, stripes of LM511 were printed onto glass coverslips and the unprinted regions were passivated to block cell adhesion, thus restricting attachment to the printed regions (maximum width of 5 µm) (Figure 6E). During the experiment cells would occasionally detach and then reattach; our analysis only included periods when cells were attached and elongated on the patterned region. The mean migration speed was lower in the Elkin1-KO clones compared to WT (Figure 6F) (22% reduction in the mean Elkin1-KO track mean speed), indicating that disrupting Elkin1 inhibits the ability of cells to undertake both unconfined and confined migration on hard surfaces.

Given that cell dissociation from the primary tumour is required for tumour metastasis, we investigated the effect of Elkin1 deletion using in vitro organotypic spheroids. We selected representative WT and Elkin1-KO clones and labelled these cells using a viral vector expressing GFP. Spheroids formed from GFP-expressing cells were implanted in 3D collagen gels and imaged at 24, 48 and 72 hr post-implantation. At every time point, more Elkin1-KO cells had dissociated from the spheroid compared to WT (Figure 7A–G) and at 48 and 72 hr the Elkin1-KO cells were further away from the edge of the spheroid (Figure 7H, Figure 7—figure supplement 1). Imaging the first 12 hr after spheroid implantation highlighted the fact that Elkin1-KO cells dissociated more readily from the spheroid mass (Figure 7—videos 1 and 2). We investigated whether isolated Elkin1-KO cells embedded in 3D collagen gels exhibited a change in migratory properties to determine if an increase in migration speed could account for these data. In this 3D environment, the mean track speed was not different between WT and Elkin1-KO clones, however the track straightness was reduced in the Elkin1-KO clones (Figure 7—figure supplement 2). We additionally noted that the Elkin1-KO cells were less spherical as they broke away from the spheroid (Figure 7I, Figure 7—figure supplement 1) as they were in the 3D collagen gels (Figure 7—figure supplement 2)). However, these effects on sphericity and track straightness were moderate, with much of the data set overlapping. These data thus suggest that the increased distance of the Elkin1-KO cells from the spheroid is due to facilitated dissociation, not due to increased migration speed.

One model to account for the increased dissociation of cells from Elkin-1-KO spheroids as well as their altered migration properties is that deletion of Elkin1 changes physical cellular interactions. To test this idea we used atomic force microscopy (AFM) to measure unbinding forces after short-term contact between cells and LM511 or homotypic cell-cell contacts (Hofschröer et al., 2017; Puech et al., 2006). The unbinding of LM511 from Elkin1-KO cells required a larger force than the WT clone (37% increase in the mean of the maximum unbinding force) (Figure 8A,B). In contrast, lower unbinding force was measured after homotypic contact between Elkin1-KO cells, in comparison with WT cells (27% decrease in the mean of the maximum unbinding force) (Figure 8C,D). To determine if deletion of Elkin1 influenced cell-cell organisation over longer time scales, we created chimeric spheroids between Elkin1-KO and WT cells by mixing equal numbers of a GFP-labelled and an unlabelled clone. Regardless of which clone expressed the GFP, the spheroid was organised such that Elkin1-KO cells were found in the outer layer of the spheroid and the WT cells in the core (triplicate experiments, 23 spheroids total) (Figure 8E,F). To test whether the channel activity of Elkin1 is linked to cell partitioning within spheroids, we stably transfected Elkin1-KO cells with either hsElkin1-iso3 or hsElkin1-iso3-L210F. In chimeric spheroids the partitioning of cells was partially rescued in cells expressing hsElkin1-iso3 but not hsElkin1-iso3-L210F, where cells were evenly distributed throughout the resulting spheroids (duplicate experiments, 6 and 9 spheroids, respectively) (Figure 8—figure supplement 1). Taken together these data indicate that deletion of Elkin1 has a differential effect on cell-substrate versus cell-cell binding and regulates cell-cell associations in organotypic spheroids.

All melanoma cell lines (WM266-4, WM115, and A375) were obtained from the ATCC and cultured in complete Minimum Essential Medium Eagle (MEME) or High Glucose DMEM supplemented with 10% FBS and 1% Penicillin/Streptomycin. MEME was additionally supplemented with 1% l-glutamine. HEK-293T and HEK-293T P1KO (Lukacs et al., 2015) (a gift from A. Patapoutian) were cultured in DMEM medium, 10% FBS and 1% Penicillin/Streptomycin. All cultures were maintained at 37°C, 5% CO₂. Fugene HD (Promega) or polyethyleneimine (PEI, Sigma, MW 600–800 kDa) was used to transfect cells at the following ratios (w/w): DNA:Fugene at 1:3 and DNA:PEI at 1:4. WM266-4 cells were transduced with pRRLSIN.cPPT.PGK-GFP.WPRE lentivirus (a gift from Ian Alexander, originally from Didier Trono/Inder Verma, Addgene plasmid #12252), in which a human PGK promoter drives EGFP expression. The virus was produced in HEK-293T cells following PEI transfection of the lentiviral backbone together with packaging plasmid (psPAX2) and the vesicular stomatitis virus (VSV-G) envelope. To create cell lines with stably-integrated DNA for spheroid rescue experiments, plasmids encoding Elkin1 and GFP were linearised using ScaI. Linearised DNA was transfected into WM266-4 3C6 (Elkin1 KO clone) using Fugene (as above). Genetically modified cells (virally transduced or stably expressing Elkin1) were isolated in FACS buffer (2 mM EDTA, 2% FBS, 2% Penicillin/Streptomycin in PBS) using the BD FACS Jazz (low pressure sorting with a 100 μm nozzle at 17 psi, 4°C). For all experiments to be conducted within 6 hr of cell collection, cells were released from culture flasks using enzyme-free cell dissociation buffer (Sigma-Aldrich). Cell lines were authenticated using STR profiling and checked for mycoplasma contamination by CellBank Australia.

To prepare samples for mass spectrometry, peptides were prepared from cultured WM266-4 cells using the Pierce Mass Spec Sample Prep Kit for Cultured Cells as per manufacturer’s instructions (ThermoFisher). LysC and Trypsin were used to generate peptides from 1 mg of total protein. Peptides were separated using isoelectric focussing (IEF) in immobilized pH gradient (IPG) gel strips into six separate fractions, as previously described (Eravci et al., 2014). Desalted peptides of these fractions were then separated on an 8–60% acetonitrile gradient (240 min) with 0.1% formic acid at a flow rate of 200 nL/min using the EASY-nLC II system (Thermo Fisher Scientific) on in-house manufactured silica microcolumns packed with the ReproSil-Pur C18-AQ 3 µm resin. A Q Exactive plus mass spectrometer (Thermo Fisher Scientific) was operated in the data dependent mode with a full scan in the Orbitrap followed by top 10 MS/MS scans using higher-energy collision dissociation (HCD). Analysis of MS and MS/MS spectra was performed using MaxQuant software (version 1.5.1.2) and proteins were identified by searching against the human reference proteome database UP000005640 from Uniprot.

The sequences of all primers used for this study are listed in Supplementary file 2. mRNA was isolated from WM266-4 cells using the RNEasy kit as per manufacturer’s instructions (Qiagen). First strand cDNA synthesis was carried out using 200 ng of isolated mRNA, random primer mix (New England Biolabs) and M-MuLV reverse transcriptase (New England Biolabs). The resulting samples were used for qPCR analysis (see below) and as a template to amplify hsElkin1-iso1/3 cDNA. For electrophysiology experiments, these sequences were cloned into pRK5 with a cistronic eGFP driven by an internal IRES sequence. For the purposes of live cell imaging, hsElkin1-iso1/iso3 sequences were fused to mGFP using the same vector. miRNA knockdown reagents were generated using the BLOCK-iT Pol II miR RNAi Expression Vector, as per manufacturer’s instructions (Invitrogen). Briefly, complementary ssDNA sequences (Eurofins, Belgium) that encode the miRNA of interest were annealed and cloned into the pcDNA 6.2GW-EmGFP vector. Three distinct miRNA sequences were generated for each target gene and assembled in series within a single plasmid to generate the final construct for knock-down experiments. The Golgi network was detected using CellLight Golgi-RFP, BacMam 2.0 (ThermoFisher). Sequences were analysed using the T-Coffee multiple sequence alignment server (coffee.crg.cat) and formatted using Boxshade (https://embnet.vital-it.ch/software/BOX_form.html).

To create plasmid constructs for gene editing, guide RNAs (gRNAs) were designed using the online CRISPR design tool (Zhang Lab, MIT - http://crispr.mit.edu/). Single stranded DNA oligos were obtained from Eurofins (Belgium), annealed and cloned into the pSpCas9n(BB)−2A-GFP plasmid (a gift from Feng Zhang, Addgene plasmid # 48140) (Ran et al., 2013). Four plasmids were generated, each with gRNAs that targeted sequences in intron 7 or exon 9 of the Elkin1 gene. These constructs were transfected into either WM266-4 or A375 cells. Isolation of clonal populations was a two-step process. First, a population of transfected cells was initially isolated based on GFP expression: 24–48 hr post-transfection cells were resuspended in FACS buffer (2 mM EDTA, 2% FBS, 2% Penicillin/Streptomycin in PBS) and cells expressing GFP were collected using the BD FACS Jazz low pressure sorting with a 100 μm nozzle at 17 psi, 4°C. Following expansion of this population, cells were sorted a second time (as above) as single cells in individual wells of a 96 well plate. Single-cell clones were initially cultured in a media comprising of 50% complete media/50% conditioned media for 7–14 days until cells had started to divide and form colonies.

Edited clones were screened for genomic deletions covering the Elkin1 gene using PCR. Primers for screening CRISPR edited clones were designed using Primer-BLAST (NCBI) to span the ~2.5 kb deleted region of the gene. Genomic DNA (gDNA) was extracted from cells using the Illustra Genomic Prep Mini Spin kit (GE Life Sciences), as per manufacturer’s instructions. Genomic PCR was conducted using 10 ng of template DNA and Hot Start Taq DNA polymerase (New England Biolabs).

Primers and probes to analyse Elkin1 and PIEZO1 transcript levels were designed using PrimerQuest Design Tool (IDT, Singapore) and a predesigned assay was used to detect levels of HPRT1 (housekeeping gene) (Integrated DNA Technologies, Singapore). The reaction underwent 40 cycles. Using the difference in cycle threshold (ΔCt), the fold change in expression (2-^ΔΔCt) was calculated and compared to a control sample.

Positive masters and pillar array casting were described previously (Poole et al., 2014). Briefly, positive silicon masters were silanised using vapour phase Trichloro(1H,1H,2H,2H-perfluorooctyl) silane (Sigma-Alrich) for 16 hr. Negative masters were cast from this substrate in polydimethylsiloxane (PDMS) (Sylgard 184, Dow Corning), mixed at a ratio of 1:10 and cured at 110°C for 15 min. Negative masters were silanised as above and used to cast pillar arrays. Arrays were coated with degassed PDMS (1:10) and left for 30 min. A thickness two coverslip activated with oxygen plasma generated using a low pressure Zepto plasma system (Diener, Germany) was placed over the still liquid PDMS. Pillar arrays were cured for 1 hr at 110°C. Pillar arrays were activated using the oxygen plasma system and either cells were directly seeded onto this activated surface or arrays were first functionalised by coating with 10 μg/ml LM-511 (BioLamina, Sweden) for one hour at 37°C. Cells were seeded at a concentration of 2 × 10⁴ cells/mL in complete media and incubated overnight.

Whole-cell patch pipettes were prepared from thick-walled filamented glass (Harvard Apparatus, USA) using a pipette puller fitted with a box filament (P-1000, Sutter Instruments, USA). Pipettes were heat-polished with a homemade micro-forge to give a final resistance of 3 MΩ – 6 MΩ. Pipettes were filled with a solution containing 110 mM KCl, 10 mM NaCl, 1 mM MgCl₂, 1 mM EGTA and 10 mM HEPES (pH 7.3). Extracellular solutions contained 140 mM NaCl, 4 mM KCl, 2 mM CaCl₂, 1 mM MgCl₂, 4 mM glucose and 10 mM HEPES (pH 7.4). Whole-cell patch-clamp data was obtained on either a Zeiss 200 inverted microscope and an EPC-10 amplifier in combination with Patch-master software or a Nikon Ti-E inverted microscope and an Axopatch 200B with pClamp 10 software. Data were analysed using either Fitmaster software (HEKA Electronik GmbH, Germany) or Clampfit software (Molecular Devices, USA). Pipette and membrane capacitance were compensated and to minimise voltage errors, series resistance was compensated by at least 60%. Mechanically-activated currents were recorded at a holding potential of −60 mV.

Mechanical stimuli were applied by serially deflecting an individual pilus using a blunt, heat-polished pipette (tip diameter approx. 2 μm) driven by a MM3A-LS nanomanipulator (Kleindiek Nanotechnik, Germany). Multiple stimuli ranging between 1 nm–1 µm were applied with a delay between each stimulus of at least 10 s. To quantify the stimulus, bright-field images of the deflected pilus were taken before and during stimulation using a 40x/0.6 NA objective. The centre of each pilus was calculated off line by applying a 2D-gaussian fit of intensity values (Igor, Wavemetrics, USA) and the magnitude of pillar deflection calculated from successive images by comparing the difference in the calculated centre point.

Transwell assays were conducted using the HTS FluoroBlok Multiwell Insert System (Corning) with an 8.0 µm pore size. The bottom surface of the membrane was coated with laminin (10 µg/ml) for 3 hr at 37°C. Membranes were then blocked for 1 hr using 2% polyvinylpyrolidone. The bottom well was filled with 500 µl media and 1 × 10⁴ cells were added to the top well. Cells were allowed to transmigrate for 16 hr before the bottom side of the membrane was fixed with 4% PFA. In order to quantify the number of cells that had transmigrated, the fixed sample was permeabilised using 0.1% TX-100 for 5 min at room temperature and nuclei were stained using Hoechst at a concentration of 0.1 µg/ml. Cells were imaged using epifluorescence imaging on an inverted microscope (Zeiss Axiovert 200 or Leica DMIL) fitted with a 10x/0.22 NA objective and standard filters; 4 regions of each filter were imaged. The ‘Analyze Particles’ function of ImageJ was used to count cells within each field. Data from each independent experiment were normalized to the average number cells that were counted on the underside of each filter for triplicate controls.

µ-Slide 8-well dishes (ibidi, Germany) were coated with either 10 µg/ml LM-511 (BioLamina, Sweden), or 0.1% (w/v) poly-L-lysine (PLL) (Sigma-Aldrich, USA). Approximately 2500 cells were added to each well, and after 2 hr, nuclei were labelled by adding Hoechst dye at a concentration of 0.1 µg/ml. Cell migration was monitored for 16 hr using live-cell microscopy (see below). Data were obtained from three experiments. Data were confirmed by repeating this assay with cells expressing cytoplasmic GFP, using the same imaging conditions.

Microcontact-printing (von Philipsborn et al., 2006) was used to pattern substrates for quasi-1D experiments. Microcontact-printing stamps (Chiang et al., 2011) were incubated with 10 µg/ml LM511 and 1 µg/ml of goat anti-mouse IgG antibody labelled with AlexaFluor 594 (to visualize patterns) for 30 min at 37 °C. Stamps were rinsed with ultrapure water, dried with inert gas, then stamped on clean coverslips activated using oxygen plasma. Coverslips were incubated with 1 mg/mL PLL-g-PEG (SuSoS, Switzerland) in PBS, at room temperature for 30 min, to block unprinted regions. Cells (3 × 10⁴ cells/mL) were seeded on each substrate and nuclei were labelled by adding 0.1 µg/ml Hoechst. Cells were monitored using live-cell microscopy for 12 hr and data were obtained from at least 50 cells across three experiments. We only included in our data the periods of movement that correspond to when the cell was attached and elongated on the patterned region. To determine whether cells were confined/unconfined we analysed the cell shape. Cells with an aspect ratio of 2:1 or more were classed as confined and the number of confined vs non-confined cells was determined by sampling still images at five points during each experiment.

To assay individual cells in collagen gels, a cell suspension (1.25 × 10⁶ cells/mL in complete media) was mixed with 1.5 mg/mL rat tail Collagen I (total volume 100 µL) and gelated at 37°C, 5% CO₂ for 15 min. Isolated cells in collagen gels were imaged using a Leica SP8 DLS fitted with a 10x/0.4 NA objective every 10 min for 14 hr. All cultures were maintained at 37°C and 5% CO₂.

Both WT and Elkin1-KO clones were virally transduced with a construct encoding eGFP to enable visualisation of cells. For spheroid formation, 1 × 10³ cells were seeded onto Ultra-low attachment 96-well plates (Corning, USA) in complete MEME media and formed spheroids following 72 hr of incubation. Composite spheroids were created by mixing 5 × 10² of each cell-type to give 1 × 10³ cells total. Collagen gel mixtures were made from 1.5 mg/mL – 2 mg/mL rat tail Collagen I (Corning, USA) in 10 mM NaOH, 1 × PBS and complete media on ice. A base collagen gel was formed in each well of a glass-bottom 96-well plate (Greiner Bio-one, Austria) by incubating 30 µL of the collagen mixture at 37°C, 5% CO₂ for 6 min. The collagen gel mixture containing the spheroid was then applied to the base gel and underwent gelation at 37°C, 5% CO₂ for 15 min. Cells in collagen gels were cultured under 200 µL of complete media. Spheroids and invasive cells were imaged at 24, 48, 72 hr using a Leica SP8 DLS fitted with a 10x/0.4 NA objective.

Imaris 9.1.2 software (Bitplane AG, Zurich, Switzerland) was used to segment cells by creating surfaces with a filter of 200 μm³ to discard cell debris. For organotypic spheroid assay analysis, cell surfaces were filtered by centre of image mass versus z depth, where the z value was determined by orientating the 3D image stacks to the plane of view showing the cells on the glass and cells attached to the glass were excluded from analysis. Cells from the 3D dissociated cell migration assay were tracked using autoregressive motion, applying a threshold of 1500s to filter track duration. Intensity, morphological and tracking data were then exported and further analysed using GraphPad Prism software (La Jolla, CA, USA).

To detect Elkin1 variants at the plasma membrane, the cell surface fraction was biotinylated and subsequently isolated following established protocols (Tarradas et al., 2013). Briefly, HEK-293T P1KO cells attached to PLL-coated culture dishes were transiently transfected with plasmids encoding GFP-tagged Elkin1 variants. After 24 hr the cell surface fraction was labelled, on ice, with freshly prepared 2.5 mg/ml EZ-link Sulfo-NHS-LC-LC-biotin (21338, ThermoFisher Scientific) in DPBS containing Ca⁺⁺. After quenching with 100 mM glycine, cells were lysed as above using RIPA buffer containing protease inhibitor cocktail. A portion of this lysed sample was reserved as the ‘input’ sample. The biotinylated fraction was then isolated using NeutrAvidin Ultralink Resin (53150, ThermoFisher Scientific). After recovery from the NeutrAvidin beads, samples were prepared as for gel electrophoresis by mixing with Bolt LDS sample buffer and Bolt reducing agent (B0007 and B0009 respectively, ThermoFisher Scientific). Samples were separated on a 10% Bolt Bis-Tris Plus gel (NW00102BOX, ThermoFisher Scientific), transferred to a PVDF membrane and subjected to standard antibody detection. GFP-fusion proteins were detected using rabbit polyclonal anti-GFP (SAB4301138, Sigma-Aldrich, 1:1000) and HRP-linked anti-rabbit IgG (7074, Cell Signaling Technologies, 1:1000).

AFM analysis of cell-binding forces was conducted using a JPK Nanowizard, fitted with a CellHesion module (JPK Instruments AG). To measure cell-LM511 interaction forces, cantilevers (MLCT-O10, Bruker) were first activated using oxygen plasma generated using a low-pressure Zepto plasma system (Diener, Germany) and then incubated in a droplet of LM511 (20 µg/ml) for 1 hr. To measure cell-cell interaction forces, cantilevers were coated with wheat germ agglutinin (L4895, Sigma Aldrich). Before collecting data, the sensitivity and spring constant of each cantilever was determined using in-built routines in the JPK software. To measure cell-LM511 binding forces, the following settings were used: speed 2 µm/sec, set point 500 pN, with contact times of 2 s held at constant height. At least five force-distance curves were collected from at least 10 cells per condition (multiple force-distance curves were not collected successively on each cell to minimise adaptive changes). To measure cell-cell binding forces the CellHesion module was used to move the stage through an extended 100 µm pulling range. To attach a cell to the cantilever, cells in suspension were added to a sample dish containing adherent cells, the calibrated cantilever was positioned over an unattached cell and carefully brought into contact. After 10 s attachment time the cantilever was slowly withdrawn from the surface, with a single cell attached. To collect data the cantilever was positioned over an adherent cell and data collected with the following parameters: speed 5 µm/sec, set point 250 pN, contact time 2 s held at a constant height. At least five force-distance curves were collected from at least 10 different cell-cell interaction pairs. Force-distance curves were analysed to determine the maximum unbinding force during cantilever retraction.

All data were analysed using Prism seven or Prism 8 (GraphPad). Normality was determined with D’Agostino-Pearson omnibus normality test. Normally distributed data were analysed using parametric statistical tests. All t-tests were two-tailed and when there were significant differences in the variance, Welch’s correction was used. Data from qPCR experiments and migration assays were analysed using a Kruskal-Wallis one-way ANOVA test with Dunn’s multiple comparisons. Stimulus-response plots were generated by binning data by stimulus size and averaging within bins for each cell, then across cells. Ordinary two-way ANOVA with Sidak’s multiple comparisons was used to analyse stimulus-response plots. The transwell migration assay data were normalised against the parental clone and analysed using a parametric one-way ANOVA test with Tukey’s multiple comparisons. Data from organotypic spheroid experiments were averaged for each spheroid and then compared with one-way ANOVA with Tukey’s multiple comparisons.