Dia1 coordinates differentiation and cell sorting in a stratified epithelium

Throughout adulthood, epithelia turn over yet retain distinctive architectures. Such tissues often encompass subdomains distinguished by unique cellular morphology and behavior. The mechanisms, however, by which these domains arise and persist remain a puzzle. Exemplifying the problem, keratinocytes establish stratified structures at the skin surface characterized by layers with distinct morphologies (Luxenburg and Zaidel-Bar, 2019; Simpson et al., 2011). A basal layer of proliferative, columnar cells pack atop a collagenous dermis and give rise to overlying layers with flattened, squamous morphologies. As cells stratify and change shape, a differentiation program ensures that suprabasal cells repress proliferative behavior and express factors required for producing an effective barrier to water loss, pathogens, and mechanical insult (Barrandon and Green, 1985).

Of note, basal cell shape affects the capacity to coordinate stratification with differentiation (Box et al., 2019; Miroshnikova et al., 2017). Reduced cell spread area, concomitant with basal layer packing, also triggers serum-response factor (SRF), Yes-associated protein (YAP) and NOTCH responses which drive differentiation (Connelly et al., 2010; Totaro et al., 2017). Coordination of cell movement with differentiation in these models, however, is contingent upon establishing a densely packed basal layer. The question arises as to how the skin and other stratified epithelia shunt proliferation into packing the basal layer rather than distributing cells homogenously throughout the tissue thickness. To compartmentalize tissues, cells sort out based on differential intercellular adhesion and actomyosin contractility (Fagotto, 2014; Maitre et al., 2012; Manning et al., 2010). The segregation of like cells based on these properties guides prominent developmental processes like gastrulation but is also thought to aid in maintaining compartments at homeostasis (Krens and Heisenberg, 2011). In epidermal tissues, genetic studies indicate that cadherin and integrin-based adhesion ensure proper basal layer formation (Brakebusch et al., 2000; Raghavan et al., 2000; Tinkle et al., 2008; Vasioukhin et al., 2001). Similarly, classic in vitro studies indicate that when mixed in a cell aggregate, undifferentiated keratinocytes separate from differentiated neighbors to form compartments (Watt, 1984). Recent studies highlight the differences in basal and suprabasal layer stiffness (Fiore et al., 2020), pointing to mechanics as a likely determinant of this separation.

How actin regulatory proteins, through their effects on cell–cell adhesion and cortical mechanics, enforce compartmentalization of the basal layer is not well understood. Whereas earlier studies implicate depolymerizing agents like cofilin in epidermal function (Luxenburg et al., 2015), the role of actin nucleators requires further attention. Disruption of the branched actin nucleator, Arp2/3, or its activator, the WAVE complex, perturbs epidermal function with varying effects on the structure (Cohen et al., 2019; van der Kammen et al., 2017; Zhou et al., 2013). Unbranched actin filaments produced by the formin family anchor adhesions and participate in physiological adaptations to mechanical stretch within the epidermis (Aragona et al., 2020; Kobielak et al., 2004). Endogenous miRNA restricts expression of the canonical formin, Dia1, to the basal layer of human epidermis and tongue (Sundaram et al., 2013; Sundaram et al., 2017). Though diabetic ulceration and certain carcinomas yield aberrant expression patterns (Chakraborty et al., 2010; Sundaram et al., 2013; Sundaram et al., 2017; Whitson et al., 2018; Xing and Liu, 2017; Yang et al., 2019), the normal physiological function of Dia1 in the basal layer remains unclear. One hypothesis is that differential Dia1 expression translates into differential adhesion effects amenable to packing and producing a properly shaped basal compartment.

In addressing this hypothesis, we found that Dia1 depletion in a model epidermis disrupted basal layer morphology and led to differentiation defects. Effects on morphology originated from the capacity of Dia1 to accelerate lateral adhesion growth. In monolayers, this feature allowed cells to accommodate crowding by enabling redistribution of the cortex toward building taller lateral adhesions and smaller basal footprints. Dia1 depletion, in contrast, forced cells to evade crowding via delamination. Thus, as proliferation added cells to the Dia1-positive basal layer, the ensuing rearrangement of junctions favored packing the basal layer over movement into the suprabasal compartment. This effect ensured a density amenable to triggering differentiation. In vitro assays also supported the hypothesis that Dia1 expression discrepancies can guide cell compartmentalization. In mature stratified epithelia, we posit that the basally skewed expression pattern of Dia1 establishes a differential adhesion effect which guides redistribution of adhesions during division. Rapid expansion of lateral junctions between Dia1-positive neighbors ensures reintegration of basal daughter cells during division. In contrast, crowding of a Dia1-negative cell by a neighboring division drives delamination and stratification. Our results provide the first indication that formin expression gradients, through effects on cell sorting, enforce the maintenance of distinct morphological zones within complex tissues.

Involvement of Dia1 in immune function and the sensitivity of skin to immunological influence steered us toward the in vitro approach of a well characterized, organotypic culture model (Albanesi et al., 2018; Lin et al., 2011; Peng et al., 2007). To mimic the in vivo juxtaposition of epidermal epithelia to its supporting connective tissue, three-dimensional cultures are produced by plating keratinocytes, like the HaCaT line, upon fibroblast-doped collagen (Schoop et al., 1999). Though they mature slower than primary cell counterparts (Schoop et al., 1999), HaCaT cultures developed at an air–liquid interface construct a stratified tissue with architecture and expression profiles mimicking those of normal tissues (Koehler et al., 2011; Sundaram et al., 2017). Immunostaining of tissue sections produced from 10-day-old, paraffin-embedded control (CTL) cultures indicated that Dia1 expression, as well as the shape and orientation of nuclei, changed with respect to position in the stratified tissues. Dia1 concentrated in the basal layer, approximated as keratinocytes lying within 10 μm of the underlying collagen (dashed line, Fig. 1 A). This localization is consistent with the earlier findings in human epidermis and tongue (Sundaram et al., 2013; Sundaram et al., 2017); we validated the results by Dia1 staining of skin and tongue samples (Fig. S1). Further, we also found Dia1 enriched in the basal layer of the hair follicle outer root sheath (Fig. S1).

Basal layer keratinocytes produced a mean linear density of 13 cells per 100 μm (Fig. 1, B and C). Within the basal layer, nuclei are elongated, with an average aspect ratio of 2.5 (Fig. 1, B and C), and are oriented such that the long axis is nearly perpendicular to the basal surface, with the average angle between these axes being 25° (Fig. 1 E). Both the nuclear shape and orientation change in a stereotypical fashion toward the apical surface. As their relative position increases, the nuclei are less elongated but maintain their orientation (Fig. 1, A–E). In the most suprabasal layers, the aspect ratio and orientation increases such that the long axes are parallel to the basal interface (Fig. 1, A–E). Experimental replicates of Fig. 1, D and E are presented in Fig. S2 and demonstrate similar behavior.

To explore the role of Dia1 on the architecture of stratified tissues, we used CRISPR and shRNA to reduce Dia1 expression (Fig. S3). Both techniques diminished Dia1 expression, with shRNA yielding knockdown efficiencies of 65–80% and CRISPR edits causing near total abrogation of Dia1 protein levels. Next generation sequencing identified two frameshift variants with premature stop codons, accounting for ∼85% of the genomic sequencing signal (Fig. S3). Tissue depleted of Dia1 by CRISPR (Dia1KD) stratified but produced a basal layer that was 20% less dense than CTL cells and lacked the stereotypical, columnar morphology (Fig. 1, A and C). Within the basal layer, the nuclei of Dia1KD cells were less elongated and randomly oriented, at times lying parallel to the underlying collagen (Fig. 1, A, D, and E; and Fig. S2). The nuclear shape remained constant in the suprabasal layers and showed a similar trend toward becoming parallel to the basal interface at high relative positions (Fig. 1 E). Multicellular patches of Dia1KD cells were also observed protruding into the underlying collagen (Fig. S3).

To overcome the limitations of measuring two-dimensional (2D) density in a cross section, we conducted confocal imaging of flat mounted, H2B-labeled cells (CTL-H2B and Dia1KD-H2B) cultured for 6 days (Fig. 1 F). These data revealed similar basal layer density deficits in Dia1KD-H2B cells (Fig. 1 G) and was corroborated by flat-mounted imaging of fixed cultures (Fig. S4). Dia1-depleted tissues successfully stratified but did so without producing a properly packed basal layer, leading us to posit that cells were escaping the basal layer before reaching an appropriate density.

Since morphology dictates the success with which basal cells differentiate (Connelly et al., 2010; Totaro et al., 2017), we next tested whether the aberrant morphology caused by Dia1 depletion impacted epidermal differentiation. To address this possibility, sections of organotypic cultures were immunohistochemically stained for the differentiation marker, keratin-10 (KRT10). As in normal tissue, CTL cells produced a three-dimensional culture in which KRT10 localized specifically to suprabasal layers (Fig. 2, A and B). Dia1KD cells, conversely, produced tissues that stratified but contained lesions of KRT10-negative cells (Fig. 2, A and B). Organotypic cultures derived from shRNA-treated cells (shCTL and shDia1) corroborated these results (Fig. S3).

Spatially segregated proliferation is an important attribute of differentiation in stratified tissues, and its loss is linked to pathologies ranging from barrier disruption to viral oncogenesis (Moody and Laimins, 2010; Rorke et al., 2015). Staining for the proliferation marker, proliferating cell nuclear antigen (PCNA), allowed for visualization of where proliferative cells reside in the stratified layers. Control tissues demonstrated proper restriction of proliferative cells to the basal layer (Fig. 2, C and D). Dia1-depletion led to the extension of PCNA positive cells throughout the tissue thickness (Fig. 2, C and D). Further, the percentage of PCNA positive nuclei was 18% in the Dia1KD cells, reflecting a sixfold overall increase from that of control tissues (2.9%). This is reminiscent of hyperproliferative phenotypes associated with stratified tissue disorders such as psoriasis and esophageal disease (Hugh and Weinberg, 2018; Wang et al., 2005).

Standard 2D culture methods of assessing differentiation corroborated the organotypic model results. A well-known technique for inducing differentiation in vitro entails culturing cells confluently in low calcium (0.03–0.07 mM CaCl₂), then switching the cells to a high calcium (1.2–2.8 mM) media (Deyrieux and Wilson, 2007). Compared to shCTL-treated cells, the mRNA level of differentiation markers involucrin (IVL) and keratin-1 (KRT1) were diminished fivefold in shDia1 cells (Fig. 2 E). To validate the specificity of this result, we transduced Dia1KD cells with a Scarlet-tagged, full-length mouse Dia1, referred to hereafter as Dia1-Rescue cells (Fig. S5). Compared to Scarlet-only controls (Dia1KD-Scarlet), reintroduction of Dia1 facilitated induction of IVL and KRT1 expression in this simplified assay (Fig. 2 F). These results confirmed the specificity of differentiation abnormalities stemming from Dia1 depletion.

To assess the impact on transcriptional changes expected during stratification, we assessed differential gene expression in confluent Dia1KD-Scarlet cells compared to the Dia1-Rescue line via RNAseq. Of the 15,150 genes detected, ∼275 of these were upregulated at least twofold in the rescue line (Fig. 3 A). Strikingly, the top five gene sets enriched by Dia1 rescue, as detected by automated gene ontology, were those of cornification, keratinization, keratinocyte differentiation, and epidermal differentiation/development. Although highly overlapping sets, these strongly implicated Dia1 in epidermal development.

We then performed a manual query of this data with published gene sets for skin differentiation (Joost et al., 2016), skin disease (Lemke et al., 2014), and genes driven by salient pathways like p63 (Truong et al., 2006), YAP (Wang et al., 2018), or NOTCH signaling (Wang et al., 2014). Genes linked to skin disease, differentiation, p63, or NOTCH signaling increased in response to Dia1 rescue (Fig. 3 B). In particular, we observed increased expression of the small proline-rich (SPRR) proteins (Fig. S6), essential for cornification (Carregaro et al., 2013). In contrast, several notable YAP target genes decreased in response to rescued Dia1 expression. Targets of other signaling programs salient to differentiation like SRF (Olson and Nordheim, 2010), SHH (Lewandowski et al., 2015) and Wnt (Watanabe et al., 2014) did not demonstrably trend toward up- or downregulation, but contained individual transcripts displaying sensitivity to Dia1 expression (Fig. 3 B and Fig. S6). Altogether, this data underscored that Dia1 expression, even in in vitro culture models, could drive gene expression signatures pertinent to in vivo epidermal differentiation.

To determine whether aberrant cell area and/or density contributed to the differentiation phenotype of Dia1-deficient cells, we explored whether enforced crowding could rescue transcriptional abnormalities. Inspired by an earlier study (Miroshnikova et al., 2017), keratinocytes were seeded on biaxially pre-stretched (36% by area) PDMS sheets. After incubating cells for 20 h, stretch was released, in effect, increasing cell density (Fig. 3 C). Sampled at the center of the culture area, release from stretch increased Dia1KD-Scarlet density by ∼50% from 2,455 to 3,645 cells per mm² (Fig. 3 D). By comparison, a parallel Dia1-Rescue culture produced a mean cell density of 2,755 cells per mm² (Fig. 3 D). Thus, relaxing prestretch provided a means to enforce crowding of Dia1-deficient cells into densities beyond those achieved by unmanipulated Dia1-Rescue cells. Mitigating the prestretch to 18% and, in turn, dampening the extent of cell density increase, enabled achievement of densities more closely mimicking those of Dia1-Rescue cultures (Fig. S7). Following 5 h of incubation post-release, samples were collected for RNA sequencing (Fig. S5). Dia1KD-Scarlet cells with and without enforced crowding were compared to identify differentially expressed genes. Interestingly, we saw enhanced expression of the 276 Dia1-driven genes identified in Fig. 3 A for both levels of enforced crowding. Similarly, we saw reduced expression of Dia1-inhibited genes. This appeared specific to Dia1-responsive genes, as randomly chosen gene sets demonstrated little response to crowding (Fig. 3 E).

Manual inspection of transcripts from Fig. 3 E that were at least 1.6× enriched or depleted identified a number of differentiation markers as well as components of the pro-differentiation p63 and Notch pathways (Fig. 3 F). GSEA database analysis revealed that both Dia1 expression and enforced crowding of Dia1KD-Scarlet cells suppressed nonepidermal (neural; GO:0007417, GO:0022008) gene expression, specifically among targets of Polycomb Repressive Complex 2 (PRC2)-mediated silencing (Ben-Porath et al., 2008). This set included two YAP-activated genes, TGFB2 and SLIT2, which tracked with connective tissue growth factor (CTGF) in responding negatively to crowding (Cordenonsi et al., 2011; Wang et al., 2018). Consistent with reports of YAP antagonizing differentiation (Totaro et al., 2017; Zhou et al., 2013), both crowding and Dia1 rescue upregulated expression of differentiation markers including DAPL1, DSC1, and CERS3. At this timepoint, crowding does not induce upregulation of the classic epidermal differentiation marker, KRT1. However, by increasing the incubation time to 9 h after releasing prestretch, we were able to detect an upregulation of KRT1 via qPCR within Dia1KD-Scarlet cells (Fig. S8). The crowding-induced upregulation of CERS3, an enzyme crucial to epidermal function (Jennemann et al., 2012), and downregulation of CTGF, a marker of YAP activity (Wang et al., 2018), were, likewise, corroborated by qPCR. When performed on subconfluent cultures, enforced crowding failed to induce KRT1 and CERS3, suggesting that induction did not arise solely from a transient drop in cellular tension but required dense culture conditions (Fig. S8). Together, the data indicated that Dia1-dependent effects on basal layer crowding contributed to keratinocyte differentiation.

To explore the formation of stratified tissue, we used live cell imaging to monitor the onset of suprabasal layer production in cultures of GFP-H2B marked cells plated atop fluorescently labeled collagen (Fig. 4 A and Fig. S9). CTL-H2B and Dia1KD-H2B cells initially formed monolayers (Fig. 4 A). At 24 h, basal layer density in CTL-H2B cultures increased by twofold, accompanied by the onset of a suprabasal layer (Fig. 4, A and B). Over the next 75 h, the basal layer density remained constant, whereas the number of suprabasal cells continued to increase (Fig. 4, A and B). In comparison, Dia1KD-H2B cells populated the culture at a reduced rate, possibly due to slower rates of proliferation (Fig. 4 B and Fig. S10). The basal layer achieved a steady-state density around 24 h post-plating, albeit 20% lower than that of CTL-H2B cells (Fig. 4 B). On the surface, this reduced proliferation is in apparent contrast with the hyperproliferative phenotype shown in (Fig. 2 C). However, the data in Fig. 2 C were acquired several days after forming 3D organotypic cultures, and we suspect arises from a failure in Dia1KD tissues to repress proliferation, consistent with a differentiation defect.

Though cells were observed in the Dia1KD-H2B suprabasal layer, the basal layer density was significantly reduced at all times. To measure the basal layer density corresponding to the onset of suprabasal layer formation, we measured the linear cell density in regions proximal to instances with 1–2 suprabasal cells per 100 µm. Consistent with our earlier results, the linear density of the basal cell layer was 20% reduced in Dia1KD cells compared to the control cells (Fig. 4 C and Fig. S9). Thus, Dia1-deficient cultures stratified but did so later and with lower basal cell densities.

Dia1 expression is sharply attenuated in the suprabasal cells of control tissues. We wished to explore whether this segregation would spontaneously occur in mixed cultures of Dia1KD and CTL cells. We produced mixed cultures containing even portions of Dia1KD cells expressing the fluorescent protein Scarlet, marking the cytoplasm, with CTL-H2B cells. After 4 days of growth on collagen gels, confocal imaging was used to assess the spatial relationship between Dia1KD-Scarlet and CTL-H2B cells (Fig. 4 D). The two cell types did not readily mix; CTL-H2B cells formed strands of connected islands weaving in between Dia1-negative regions (Fig. 4 D). We assessed the lateral positioning of the cells by imaging orthogonal sections of three independent trials (Fig. 4 E). In control cultures at similar time points, we would expect all cells to have suprabasal partners at this timepoint (Fig. 4, A and B). However, we found that 25–35% of control cells lacked suprabasal partners (Fig. 4 E). For those that did have a suprabasal partner, we found that 30–41% of CTL-H2B cells had acquired Dia1KD-Scarlet cells as suprabasal neighbors, whereas 11–17% of CTL-H2B cells had assumed a suprabasal position overlying a Dia1KD-Scarlet cell (Fig. 4 E). In 13–25% of cases, CTL-H2B cells wedged between two layers of Dia1KD-Scarlet cells (Fig. 4 E). These data indicated that Dia1KD-Scarlet cells preferentially localize to the suprabasal compartment when challenged by CTL-H2B cells.

This point becomes strikingly clear when Dia1KD-Scarlet cells are more heavily diluted among CTL-H2B cells (1:50; Fig. S11). Here, the cell density is largely controlled by the CTL-H2B cells, and we found that Dia1KD-Scarlet cells are found in suprabasal layers (Fig. S11). This suggests that Dia1-negative cells were intolerant of densities produced by Dia1-positive cells, leading to basal layer egress. We noted here that inhibiting ROCK, an upstream activator of myosin that cooperates with Dia1 to produce contractile actomyosin networks, blunted preferences for the suprabasal compartments and yielded monolayered cultures. In monocultures, 24% of CTL-H2B cell divisions orient perpendicular to the underlying collagen compared to 6% of Dia1KD-H2B divisions (Fig. S12). This is consistent with the lower cell density achieved by Dia1KD-H2B cells and with published work (Box et al., 2019) indicating that cell density controls division orientation. This argued against the idea that Dia1KD cells populate a suprabasal compartment by shifting to asymmetric divisions at an abnormally low density. Thus, we turned our attention to stratification by delamination as a suitable context within which to examine the apparent sorting of Dia1KD cells into the suprabasal compartments.

To understand how this sorting occurred, we then considered how Dia1 expression impacted the cross-sectional shape of cells, as a means of inferring the likelihood of apical protrusion. A previous study noted that Diaphanous depletion in Drosophila led to enhanced apical cross sections, consistent with increased apical protrusion (Homem and Peifer, 2008). To facilitate this assay, we formed monolayers and exploited the cytoplasm outlines produced by Scarlet and Scarlet-mDia1 signals in Dia1KD-Scarlet and Dia1-rescue cells, respectively. Immunofluorescence of fixed cells revealed enhanced spreading and protrusion of Dia1KD-Scarlet cells both atop and below neighbors (Fig. 4 F). The wedge-shape associated here with Dia1-deficiency mimicked morphology marking the delamination and protrusion of keratinocytes into the suprabasal layer during stratification (Miroshnikova et al., 2017). Together, the results supported a hypothesis that heterotypic interactions between Dia1-positive and negative cells guided the protrusive activity of Dia1-negative cells toward the suprabasal compartments. The resulting arrangement mimicked the spatial relationship observed in vivo.

The observed changes in cell morphology implicated Dia1 in modifying cell adhesion and/or cortical mechanics. To query this, we used immunofluorescence to visualize the actin and collagen in the vertical cross section of newly forming adhesions between CTL or Dia1KD cells 4 and 20 h post-plating (Fig. 5 A). We observed, at 4 h post-plating, that nascent cell–cell contacts between CTL cells induced distortion of the underlying collagen; this was not observed at 20 h (Fig. 5 B). This matrix distortion is accompanied by the production of phospho-myosin decorated, perijunctional actin filaments (Fig. S13), consistent with a role of formin-dependent actin assembly supporting cortical tension (Acharya et al., 2017; Bovellan et al., 2014; Harris et al., 2014). Indeed, nascent cell–cell contacts formed by Dia1KD cells did not distort the underlying collagen at either 4 or 20 h (Fig. 5, A and B), failed to produce the perijunctional actin pool and possessed modestly diminished overall levels of phospho-myosin (Fig. S13). This did not correspond to an overall reduction in F-actin intensity, but rather represented changes in the organization of F-actin. However, at 4 h, Dia1KD cells failed to extend their junctions vertically to match the plane of the cell apex, particularly at sites of tricellular junctions (Fig. 5 C and Fig. S13). Thus, whereas CTL cultures produced monolayers with flat apical surfaces, those of Dia1KD cultures slightly undulate. Measurement of contact angles at the apical and basal planes of cell junctions revealed larger, flatter angles at the apical junctions of CTL cultures when compared to those obtained from the basal surface (Fig. 5, D and E). Dia1KD cells produced an opposite effect. This indicated that cortical tension along the apical surface exceeded that along the basal surface in CTL cells, characteristic of a mature epithelium. Dia1KD failed to do so, demonstrating a delayed capacity to mechanically integrate apical cell surfaces during epithelial maturation.

These data suggested that, in CTL cells, force produced along the apical surface was transmitted between cells at the apical junction as well as down the lateral membrane into the collagen. Indeed, examination of E-cadherin staining revealed a discontinuous staining pattern at the apical surface in Dia1KD cells when compared to CTL cultures, arising from failure of cells to extend junctions vertically (Fig. S14). However, when limiting analysis to the most apical patch of nascent junctions, though discontinuous, mean E-cadherin intensity was not notably diminished in Dia1KD cells. Consistent with the idea that junctions may exist under tension at this timepoint in CTL cells, there was a modest recruitment of vinculin to E-cadherin positive plaques in these cultures. Dia1KD cells failed to execute this recruitment.

Though suggestive, the contact angle measurements and immunostaining experiments described above cannot clearly describe the distribution of force along the cortex at cell–cell contacts and contact free sites. To isolate these effects, we measured contact angles formed between cell doublets in suspension, to remove contributions made by adhesion to collagen. Doublets formed between CTL-H2B and Dia1-rescue cells produced similar contact angles to those produced by Dia1KD-H2B/Dia1KD-Scarlet doublets (Fig. 5, F–H). This suggested that, to some extent, adhesion to collagen exacerbated the inability of Dia1KD cells to increase apical contact angles. Heterotypic mixtures of CTL-H2B/Dia1KD-Scarlet or Dia1KD-H2B/Dia1-rescue, intriguingly, produced significantly lower contact angles when compared to homotypic interactions. Thus, in contexts where both Dia1-positive and Dia1-negative cells are present, heterotypic junctions fail to develop as robustly as homotypic interactions. This observation, and prior literature (Canty et al., 2017), led us to hypothesize that heterotypic interactions might elicit separation of Dia1-positive and Dia1-negative cells into distinct compartments.

In matrix-free cell sorting assays, differential tension and adhesion can drive the separation of distinct cell populations (Foty and Steinberg, 2005; Maitre et al., 2012). We sought to explore whether sorting of CTL and Dia1KD cells would also occur in a hanging drop assay (Foty and Steinberg, 2005). To form aggregates, CTL cells were mixed 1:1 with Dia1KD cells and suspended from petri dish lids in drops of media. Mixtures totaling 1,000 or 20,000 cells were tested. To distinguish populations, cell lines carrying the GFP, nuclear label (CTL-H2B or Dia1KD-H2B) were mixed with those carrying the Scarlet, cytoplasmic labels (CTL-Scarlet or Dia1KD-Scarlet). After 20 h, confocal microscopy enabled detection of local maxima in z-slices and networks of nearest neighbors were drawn for each signal. CTL-H2B and Dia1KD-Scarlet cells separated within such aggregates, yielding networks that intersected seldom compared to those constructed from randomized positions (Fig. 6 A). In contrast, there was a high degree of overlap between networks derived from aggregates containing mixtures of CTL-H2B and CTL-Scarlet cells (Fig. 6 B). To quantify separation, the number of intersections determined per z-slice from randomized networks was divided by the number of actual, observed intersections. Thus, randomly distributed mixtures would yield a separation index of 1 with higher values representing a greater degree of separation. Analyzing 10 z-slices per aggregate indicated that CTL/Dia1KD mixtures (mean index = 1.40 ± 0.13 SE, n = 21) displayed significantly (P = 0.04) higher levels of separation than CTL/CTL mixtures (mean index = 1.00 ± 0.05 SE, n = 11).

Though Dia1 coordinates with ROCK-dependent myosin activity to generate tension along intercellular contacts (Acharya et al., 2017), treatment of CTL/Dia1KD mixtures with the Y-27632 ROCK-inhibitor had little impact on the separation index (mean index = 1.41 ± 0.17 SE, n = 12). However, inspection of 3D structures revealed a key distinction between untreated and inhibitor-treated mixtures. In large aggregates (20,000 cells), Dia1-positive and negative cells separated into stratified layers, remarkably, mimicking the arrangement of cells in actual stratified tissue (Fig. 6 C). Cells in the presence of ROCK-inhibitor continued to sort but did so horizontally within monolayered aggregates. This could represent local clonal expansion. Alternatively, differential Dia1 expression may drive cell sorting independent of ROCK but rely on ROCK-dependent myosin activity to establish a stratified arrangement. Indeed, ROCK inhibition blunted the suprabasal movement of Dia1KD-Scarlet cells when diluted 50-fold into CTL-H2B cells and allowed to adhere to collagen gels (Fig. S11). We concluded that when presented with Dia1-negative counterparts, Dia1-positive cells preferentially adhere to each other and that addition of ROCK activity drives this separation into a multilayered structure.

Division in the basal layer figures prominently among events which affect cell position in the epidermis (Lechler and Fuchs, 2005; Miroshnikova et al., 2017). In light of the cell sorting data, a reasonable scenario would apply the given role of Dia1 in building tension-bearing actin networks at adherens junctions (Acharya et al., 2017) toward remodeling events associated with division (Higashi et al., 2016). Static analyses also indicate that Dia1 supports lateral adhesion height (Carramusa et al., 2007), suggesting that it may aid in rebuilding lateral adhesions following cytokinesis. To test whether Dia1 expression affected the rate of cell–cell contact expansion, the behavior of nascent cell–cell contacts was observed. CTL or Dia1KD cells, labeled with siR-actin, were imaged immediately after plating (t = 0 min) to capture the formation of cell–cell interfaces along the apical–basal direction. In control cells, the adhesive interface of the lateral membrane domain expanded over the first 4 h, increasing to 7 μm in height and increasing the angle formed between cell apices at the top of the contact (Fig. 7, A and B; and Fig. S10). By contrast, in Dia1KD, the adhesive interface failed to expand over the first 4 h (Fig. 7, A and B). This result represented one of the first attempts to visualize a Dia1-dependent effect on lateral membrane growth kinetics. Reintroducing mDia1 to Dia1KD cells (Dia1-rescue) also accelerated the accumulation of actin at nascent junctions visualized in the XY plane as early as 1 h post-plating (Fig. S10). As cells must reallocate adhesive surfaces following division, these data argued in favor of Dia1 expression possibly guiding division-associated rearrangements by favoring an actin-driven expansion of interfaces between Dia1-positive siblings.

The extent to which adhesion kinetics affected division-associated rearrangement was explored by live imaging of mixed cultures. Juxtaposition of H2B-marked cells with neighbors carrying Scarlet-marked cytoplasm enabled tracking of cell–cell and cell–matrix interfaces during cell division. In all mixtures tested, neighboring cells responded to division by entering the space formerly occupied by the dividing cell, often at the cytokinetic furrow (Fig. 7 C). We predicted that, when presented with the option of reintegrating with a Dia1-positive sister or a Dia1-negative neighbor, differential adhesion would favor rapid reintegration of CTL-H2B daughter cells and recession of the Dia1KD-Scarlet neighbor. Conversely, in the absence of differential Dia1 expression, intrusions made by CTL-Scarlet neighbors could become stable, possibly leading to neighbor exchange or stratification. Measuring the frequency with which intruding cells occupied a position between daughter cells as a function of time post-anaphase supported these predictions (Fig. 7 D). Measurement of intrusions made within the first hour after anaphase and tracked up to 160 min post-anaphase revealed that Dia1 expression favored production of long-lasting intrusions (Fig. 7 E). Roughly 60% of intrusions produced by CTL-Scarlet cells lasted >80 min. Intrusions produced by Dia1KD-Scarlet cells, in contrast, only exceeded 80 min in 25% of attacks made upon Dia1KD-H2B neighbors and ∼35% of those made upon CTL-H2B divisions. Thus, juxtaposition of Dia1-positive and negative cells enhanced the probability of Dia1-positive daughter cells reintegrating after division. In an in vivo context, these behaviors would favor channeling proliferative output into packing the basal layer.

Early in development, when keratinocytes exist as a monolayer, cells often divide in an out-of-plane, oblique manner to begin building stratified structures (Box et al., 2019). In our in vitro division analysis, we noted that intrusions aimed initially at the cytokinetic furrow would often expand across one of the daughter nuclei when observed via maximum projections (see Fig. 7 C, CTL/CTL panels). This suggested that intrusions were either expanding underneath or spreading atop the daughter cell. To clarify the position of intrusions, orthogonal images were created from confocal z-slices sectioned perpendicular to the division plane (Fig. 7 F). At 2 h post-anaphase, in CTL-H2B/CTL-Scarlet mixtures, 15% of all divisions analyzed resulted in a Scarlet-labeled neighbor expanding over an H2B-marked daughter cell. A roughly equivalent number of divisions (20%) resulted in the Scarlet-labeled cell positioned beneath the daughter, in effect, an oblique division. However, when the neighbor was a Dia1KD-Scarlet cell, 32% of CTL-H2B divisions resulted in one daughter cell packing underneath a Scarlet-marked neighbor. Only 12% assumed the opposite arrangement. In Dia1KD-H2B/Dia1KD-Scarlet mixtures, no divisions resulted in daughters being packed beneath a Scarlet-labeled neighbor, whereas 16% placed a daughter on top of a neighbor. These behaviors likely account for the static observations made in Fig. 4 E concerning the position of CTL cells with respect to Dia1KD neighbors. The timing of Dia1 downregulation with respect to stratification is not yet known. However, these data indicate that downregulation of Dia1 by a basal cell would result in delamination and stratification when crowded by dividing, Dia1-positive neighbors.

We show that Dia1 is required to construct a crowded basal layer which, in turn, supports differentiation of stratified epidermal tissue. In its absence, enforced crowding enables differentiation, indicating a primary role for Dia1 in determining basal layer mechanics. Epidermal tissue relegates YAP activity to undifferentiated basal cells and its inhibition by crowding elicits differentiation (Elbediwy et al., 2016). Accordingly, the present study suggests that Dia1-dependent crowding impinges on this pathway en route to supporting differentiation (Fig. 8). We speculate that Dia1 supports crowding by controlling remodeling of cell–cell adhesions required in response to cells added by proliferation. Cells in a monolayer can accommodate an increasingly crowded environment by building taller lateral junctions and occupying a smaller basal footprint or they can avert crowding by delaminating and leaving the monolayer entirely. Our data shows that Dia1 accelerates the growth of lateral cell–cell adhesions, which we surmise favors the option of packing cells into a crowded configuration. Accordingly, Dia1 depletion led to keratinocytes leaving the basal layer before reaching an appropriate density. This ultimately results in a stratified tissue marked by aberrant basal layer morphology and a diminished capacity for differentiation.

This work also has implications for the role of Dia1 suppression in the initiation of tissue stratification. Two dominant theories address the coupling of differentiation to cell position in stratified epithelia. Reoriented division and delamination among basal cells both contribute to populating the suprabasal compartment. However, recent intravital imaging reveals that dynamic factors pertaining to adhesion and geometry actively ensure proper execution of these processes (Box et al., 2019; Lough et al., 2019; Mesa et al., 2018; Miroshnikova et al., 2017). Though the present study did not aim to compare delamination with asymmetric division, we posit that preferential adhesions established by Dia1 expression gradients would act in both cases to channel proliferative output into basal layer packing (Fig. 8). Our data show that Dia1-negative cells interact transiently with dividing Dia1-positive cells, but these interactions fail to stabilize. In vivo, this instability would favor reintegration of dividing, Dia1-expressing cells with the basal layer, whilst driving the Dia1-suppressed cells to delaminate and form a suprabasal layer. A more resolved temporal analysis is required to determine at what timepoint inhibition of Dia1 expression, presumably by miR-198, occurs. However, this study suggests that loss of Dia1 expression among basal cells and subsequent delamination, coupled with oriented cell division, could represent a step in the stratification process. More generally, differential formin expression as a mechanism to drive differential adhesion and sorting may widely impact any epithelia in which proliferation must be shunted into populating specific compartments. Though we focused on intercellular adhesion, it should be noted that given the complex roots of cell morphology, there are likely effects stemming from cell–matrix interactions that will require further examination. Recent work, indeed, has demonstrated that localized formin expression can alter the shaping of underlying stroma by compartments within single-layered epithelia (Sumigray et al., 2018).

Since suprabasal cells lack Dia1 but robustly resist disintegration by exogenous force, we cannot claim that Dia1 is formally required for strong intercellular adhesion in all cases. In the present work, this pointed to adhesion kinetics and cortical mechanics as the behaviors capable of driving cell segregation. Specifically, Dia1 expression accelerates expansion of a lateral contact under tension, capable of distorting the underlying collagen. We postulate that this function favors the creation of a columnar shape by expanding the lateral “wall” between two Dia1+ cells at the expense of area dedicated to the apical “roof” or basal “floor.” Thus, Dia1 may specifically populate basal cells because it uniquely provides the polymerization activity necessary to expand an adhesive surface, the lateral membrane, while under tension. Previous work has implicated cortical tension in restricting basal layer exit (Miroshnikova et al., 2017). Dia1 is a force-activated nucleator (Zimmermann and Kovar, 2019) and is implicated in cortical mechanics (Bovellan et al., 2014); these suggest particular advantages in controlling cell shape and junction remodeling during division. In the context of rearrangements elicited by basal layer divisions, expansion of lateral, Dia1-positive, homotypic junctions would outpace expansion of the apical heterotypic junction, favoring retention of cells in the basal layer. An effect on cell shape occurs both at the monolayer stage when cells lack suprabasal neighbors as well as in the presence of Dia1-negative suprabasal cells. Together, one interpretation of these results would suggest that the Dia1 expression gradient functions to mechanically isolate suprabasal cells in terms of guiding basal cell shape while maintaining adequate adhesion between the two compartments.

In summary, we have demonstrated that differential formin expression aids in defining the morphology and packing of basal layer epidermal keratinocytes with downstream impacts on differentiation. The study broaches new ground in offering evidence that specific formin expression profiles can guide cell positioning through modulating cell–cell adhesion kinetics. In addition to the implications for other formins and tissues that must channel stem cell proliferation into defined architectures, the present study offers a unique perspective regarding the basic process of cell sorting. One implication of our work is that the adhesion kinetics associated with a formin become particularly meaningful in tissues populated by cells with heterogeneous formin expression profiles. Viewing formin expression gradients as determinants of tissue, domain architecture represents a novel perspective. Extending this possibility to the entire formin family opens a potentially fruitful avenue for exploration of morphology maintenance in all regenerative epithelia. Toward expanding upon the present work, future studies will begin tackling the potential contribution of Dia1-deficiency to in vivo wound healing assays, potentially in the context of skin lesions associated with myelodysplastic diseases (Keerthivasan et al., 2014; Lepelletier et al., 2019).

The HaCaT line of immortalized keratinocytes was provided by Yu-Ying He (University of Chicago) and was originally developed by the laboratory of Norbert Fusenig (Schoop et al., 1999). HaCaTs were maintained in low calcium DMEM prepared from calcium-free DMEM powder (#09800; US Biological), spiked with 40 μM CaCl₂, penicillin-streptomycin, L-glutamine, and 10% fetal bovine serum depleted of calcium by Chelex-100 (Sigma-Aldrich) treatment. For calcium-induced differentiation experiments, calcium concentration was raised to 2.8 mM for the indicated time span. ROCK-inhibitor (Y-27632; Cayman Chemical) was used at 10 μM.

The following antibodies were used in this study. Mouse anti-Keratin-10 (MS-611-P0; Thermo Fisher Scientific) and mouse anti-PCNA (AB29; Abcam) were used at 1:100 for immunohistochemistry. Rabbit anti-Dia1 (20624-1-AP; ProteinTech and ab129167; Abcam) was used at 1:1,000 for Western blotting and 1:100 for immunohistochemistry. Rabbit anti-GAPDH (G9545; Sigma-Aldrich) was used at 1:2,000 for Western blotting. Rabbit phospho-myosin light chain (S19; 3671; Cell Signaling) was used at 1:100 for immunostaining and 1:1,000 for Western blotting. Mouse anti-E-cadherin (ab1416; Abcam), and rabbit anti-vinculin (v4139; Sigma-Aldrich) were used at 1:100 for immunostaining.

To generate tagged mDia1 constructs, SnapGene software (from GSL Biotech; available at snapgene.com) was used to develop InFusion cloning (Takara) reactions for disabling the miRNA cassette and replacing the GFP in pGIPZ with mScarlet. The mScarlet insert was purchased from IDT as a gene block. Full-length mDia1 (NM_007858), was cloned into pGIPZ-scarlet via PCR amplification (Forward primer: 5′-GGA​CTC​AGA​TCT​CGA​ATG​GAG​CCG​TCC​GGC-3′ and reverse primer: 5′-TCT​AGG​ATC​CCT​CGA​TTA​GCT​TGC​ACG​GCC​AAC​C-3′) and the InFusion cloning system (Takara).

For experiments conducted on collagen gels, 75 μl of 3.6 mg/ml rat tail collagen I (#354236; Corning) was spread upon activated 25 mm diameter coverslips. Activated coverslips were prepared by sequential incubation with 3-aminopropyltrimethoxysilane and glutaraldehyde as described in Aratyn-Schaus et al. (2010). Gel polymerization was induced by exposure to ammonium hydroxide vapor for 4 min, followed by neutralization in PBS (pH 7.4). An 8 mm (inner diameter) glass cloning ring was placed upon the center of the gel which was, subsequently, incubated overnight at 37°C in culture media prior to receiving cells. Labeling of collagen was accomplished by spiking the mixture, 1:50, with AlexaFluor-647-conjugated collagen prepared with a protein labeling kit (A20186; Thermo Fisher Scientific). Unless otherwise noted, 1.75 × 10⁵ cells were plated per cloning ring.

Production of 3D organotypic cultures was based on protocols described by Simpson et al. (2009) with the following modifications. Collagen I solutions (ALX-522-435-0100; Enzo) of 1.5 ml (final concentration 4 mg/ml) containing J2 fibroblasts at 4.0 × 10⁵ cells/ml were cast in 3 μm pore filter inserts (657630; Greiner) set in deep 6-well plates (#355467; BD). Fibroblasts were allowed to condition the collagen plug overnight prior to 1.0 × 10⁶ keratinocytes being added per collagen plug. After 2 days of submerged culture, keratinocytes were brought into contact with air by removing media from the upper chamber and allowed to stratify for the indicated duration.

Both organotypic and keratinocyte cultures grown on coverslip-supported collagen gels were carried out in E-media, based on a 3:1 DMEM:F12 media mixture supplemented with following reagents from Sigma-Aldrich: 180 μM adenine, 5 μg/ml insulin, 5 μg/ml transferrin, 5 μg/ml T3, 0.4 ng/ml hydrocortisone, 10 ng/ml cholera toxin, 4 mM L-glutamine (Sigma-Aldrich), and 10 μg/ml gentamicin/0.25 μg/ml amphotericin B described by Simpson et al. (2009), containing 5% fetal bovine serum. For the 2 days prior to raising to an air–liquid interface, E-media is spiked with 5 ng/ml EGF (Chemicon).

A pGIPZ construct carrying shRNA targeting nucleotides 1528–1546 (GGA​TTA​ATT​GAT​CAA​ATG​A) of the human DIAPH1 mRNA (NM_005219.4) was purchased from Dharmacon (Cat# V2LHS_43611). The Dia1 shRNA had been previously described by Bovellan et al. (2014). pGIPZ constructs were co-transfected with pHR1-8.2-deltaR and a VSV-G pseudotyping plasmid (gifts from M. Rosner) into 293T cells for lentiviral particle production via a second-generation system as described by Zufferey et al. (1997). After concentrating virus using Amicon Ultra-15 30 kD centrifugal filters, HaCaTs were incubated 6 h with concentrated virus in the presence of 4 µg/ml polybrene. shRNA expressing cells were then selected for by incubation with puromycin. pGIPZ lentiviral transduction was also utilized for introducing Scarlet-tagged protein. GFP-H2B lentivirus were produced, similarly, but with a pLKO.1-based vector (gift from Elaine Fuchs; Addgene plasmid # 25999; http://n2t.net/addgene:25999; RRID: Addgene_25999). GFP-H2B transduced cells were sorted by the University of Chicago Cytometry and Antibody Technology Core.

Before conducting CRISPR editing, parental HaCaT cells were subcloned and tested to ensure that the chosen clone produced differentiated, 3D cultures. Alt-R CRISPR-Cas9 crRNA targeting DIAPH1 exon1 (5′-UCU​UCU​UGU​CCC​GGG​UCC​CGG​UUU​UAG​AGC​UAU​GCU-3′) and Alt-R CRISPR-Cas9 tracrRNA ATTO-550 acquired from IDT were complexed with Cas9 enzyme (GeneArt) and transfected into a suitable HaCaT clone using Lipofectamine CRISPRMAX (Thermo Fisher Scientific) as per IDT instructions. Through flow cytometry, successfully transfected cells were identified and deposited as single cells into separate wells and allowed to expand. Successful disruption of the DIAPH1 locus in individual clones was determined by Western blot and editing at the genomic level was assessed by the CRISPR amplicon sequencing service provided by the Massachusetts General Hospital CCIB DNA Core. Specifically, a qPCR amplified portion of the DIAPH1 exon1 genomic sequence (Forward primer: 5′-CAG​CGT​GAA​CCG​GGA​CAT​GGA-3′ and reverse primer: 5′-GAC​CGA​GCG​AAA​CGA​GGA​AGG-3′) was sent for sequencing.

Cell lysates were collected in Laemmli buffer and sheared with 22- and 25-gauge needles before carrying out amido black protein assays. Western blots were conducted by running equivalent amounts of protein on 4–20% Tris-Glycine SDS/polyacrylamide gradient gels (Lonza) after reduction with 2-mercaptoethanol and heating to 95°C for 5 min. Protein was transferred to nitrocellulose, overnight at 4°C in standard Tris-Glycine buffer containing 20% methanol without SDS at 30 V. Blots were blocked for 1 h in a 5% milk/0.05% Tween-20/TBS (pH 7.8) solution. Primary antibody incubation occurred overnight at 4°C in blocking buffer followed by washing with TBS/Tween and a 1 h incubation of HRP-conjugated secondary antibody at room temperature. After a final series of TBS/Tween washes, blots were developed using chemiluminescence reagent (Pierce) and film (CL-Xposure; Thermo Fisher Scientific) development. Blots for assessing shRNA-based knockdowns were done similarly; however, a LiCor Odyssey system was used to image fluorescent secondary antibodies and LiCor proprietary blocking buffers and washing instructions were used.

For qPCR, total RNA was collected using NucleoSpin kits (#740955; Macherey-Nagel). First-strand synthesis was carried out with the SuperScript III system (18080-051; Invitrogen) using an oligo dT primer and 200 ng of total RNA as input. First-strand reactions were diluted threefold, and 2 μl was used as template in 20 μl reactions containing PrimeTime master mix (1055772; IDT) and PrimeTime predesigned qPCR primer/probe mixtures from IDT (IVL: Hs.PT.58.39460547; KRT1: Hs.PT.58.24741966; GAPDH: Hs.PT.39a.22214836). qPCR reactions were run on a StepOnePlus system (Applied Biosystems). Relative mRNA levels were calculated using the 2^−ΔΔCt method with GAPDH serving as a reference gene.

Three-dimensional organotypic cultures were fixed in 10% formaldehyde for 3 days then transferred to 70% ethanol. The University of Chicago Human Tissue Resource Center paraffin-embedded, sectioned and provided hematoxylin/eosin stained slides of the skin cultures shown in Fig. 1 A, Fig. S3 D, and Fig. S5 A. This center also provided banked, paraffin-embedded sections of human skin (Fig. S1 A) and tongue samples (Fig. S1 B). For immunohistochemistry, 5 μm sections were de-paraffinized by baking at 55°C for 1 h followed by sequential bathing in xylene and ethanol. Antigen-retrieval was conducted by holding the sections in a pH 6.0 citrate solution at 95°C for 10 min. Sections were then blocked in a 2% BSA/TBS solution and incubated with primary antibodies overnight followed by washing and incubation with secondary antibodies for 1 h. After a final series of TBS washes, sections were mounted in gelvatol.

Plucked hair follicles were isolated from the arm and processed via whole-mount immuno-staining (Fig. S1, C and D) as described by Cheng et al. (2018) and Fujiwara et al. (2011) modified so as to use TBS, rather than PBS as the buffer, and 0.5% Triton X-100, rather than Tween-20 as the detergent. Briefly, plucked hair follicles were placed in 4% paraformaldehyde/PBS for 1 h at room temperature, followed by a 1 h blocking step in 2% BSA/TBS + 0.5% Triton X-100. Primary antibody incubation occurred overnight at 4°C in blocking buffer followed by a 4-h wash in blocking buffer at room temperature. The final counterstain for nuclei and secondary antibody incubation was conducted for 3 h at room temperature in blocking buffer and utilized SPY650-DNA (Spirochrome), rather than DAPI. Following another wash, samples were mounted and imaged via confocal microscopy. The same technique, without the primary antibody incubation step, was utilized for staining nuclei of organotypic skin cultures in Fig. S4, A–C.

For staining cultured cells on coverslips, a PBS-buffered, 4% paraformaldehyde solution was applied for 10 min (20 min for cells on collagen gels). Blocking, permeablization and quenching of paraformaldehyde was carried out by placing coverslips in a 2% BSA/TBS solution containing 0.5% Triton X-100 for 30 min. To stain nuclei and the actin cytoskeleton, coverslips were incubated with 5 μg/ml Hoechst 33342 (H3570; Thermo Fisher Scientific) and 13 nM AlexaFluor-labeled phalloidin (A12379, A12380, A22287; Thermo Fisher Scientific) in 2% BSA/TBS for 30 min at 37°C. After washing in TBS, coverslips were mounted in a Tris-buffered glycerol mountant containing N-propyl gallate.

To conduct crowding experiments, 1 mm thick sheets of Sylgard184 (Thermo Fisher Scientific) with a 10:1 polymer to crosslinker ratio were cured at 70°C overnight. The molds created 12 mm diameter, 0.17 mm deep wells on the sheet surface. The sheets, as well as, steel washers with an inner diameter and thickness of 25 mm and 3 mm, respectively, were subjected to UV-ozone cleaning for 2 min. Washers were adhered to PDMS sheets, centered around the wells, using PDMS of a 5:1 polymer to crosslinker ratio and baked at 70°C. The 12 mm wells were biaxially stretched by clamping the washer down over a 6 mm post to increase the well area by 36% or a 4.5 mm post to acquire a 18% increase. Stretched and unstretched wells were subjected to UV-ozone cleaning followed by an overnight incubation at 4°C with 200 μg/ml collagen I in 0.017 N acetic acid. In unstretched wells, 2.13 × 10⁵ cells were plated in low calcium media. In wells stretched to increase area by 18 and 36%, 2.50 × 10⁵ or 2.90 × 10⁵ cells were plated, respectively, such that all conditions received a starting cell number of ∼1,900 cells per mm². Nonadherent cells were rinsed off at 1 h, and cells were incubated a further 16 h. At that point, stretched wells were released from stretch. Parallel samples were fixed immediately to check cell densities, whereas samples destined for RNA-sequencing were allowed to incubate a further 5 h prior to lysis for total RNA purification. Release of 36% prestretched PDMS caused a roughly 48% increase in cell density at the center of the well, whereas the 18% prestretched membrane resulted in a 14% increase. For induction of KRT1, cultures were allowed to incubate 9 h after stretch was released.

Total RNA samples were produced as described above and submitted to the University of Chicago Genomics Facility. mRNA libraries were prepared and subjected to Illumina Hi-Seq 4000 (SR 50 bp) sequencing. MultiQC analysis indicated that samples produced between 70 and 110 million reads with 84–87% being uniquely mapped for each sample. The Illumina BaseSpace software suite was utilized to convert raw reads into transcripts per million (TPM) values for each gene. For forced crowding experiments, two independent experiments were conducted, each containing uncrowded and crowded conditions for Dia1-deficient and Dia1-rescue samples. After converting to fold-change over Dia1-deficient values, average uncrowded, Dia1-rescue values exceeding twofold (and exceeding at least 1.5-fold in each replicate) were fed into the PANTHER ontology program to produce unbiased functional annotation of Dia1-induced genes. Average fold change values of Dia1-rescue samples were also used in assessing the response of published gene sets pertaining to various signaling pathways. The reference sets used were as follows: suprabasal/differentiated epidermal genes (Joost et al., 2016), skin cornification diseases (Lemke et al., 2014), genes supported by the expression or activity of p63 (Truong et al., 2006), Wnt/β-catenin (Watanabe et al., 2014), Gli/Hh (Lewandowski et al., 2015), Notch (Wang et al., 2014), SRF (Olson and Nordheim, 2010), and YAP/TAZ (Wang et al., 2018). Regarding Gli/Hh, only genes reported by Lewandowski et al. (2015) as reduced twofold by SHH knockdown with FDR < 0.05 were queried. Since a different level of crowding was reached in each experiment, the crowded/uncrowded ratios are reported separately for each experiment, rather than as averages. The enrichment of neuronal and polycomb targets among genes both reduced greater than twofold by re-expressing Dia1 on average and 1.25-fold upon 36% crowding was identified using GSEA software (Broad Institute).

These experiments aimed to identify early indications of differentiation and, in turn, low abundance genes were not filtered out of datasets. As such, trends within gene sets (e.g., differentiation genes, Dia1-responsive genes, etc.) were deemed more meaningful than changes to individual genes. Specific genes listed in Fig. 3 are provided as examples but require more rigorous testing to achieve statistical significance.

Cells marked with GFP-H2B and cytoplasmic mScarlet were mixed 1:1 following trypsinization, then, suspended in 20 μl droplets of high calcium (2.8 mM) media on petri dish lids overhanging a pool of PBS. The aggregates, containing between 1.0 to 20 × 10³ cells, were cultured overnight at 37°C in a cell culture incubator. Aggregates were then fixed with paraformaldehyde and transferred to a coverslip with PDMS applied to edges as a spacer. After rinsing with PBS, gelvatol was used to mount aggregates prior to imaging aggregates in a series of 0.25 μm Z-sections.

To produce cell doublets, cell line pairs were cocultured together at subconfluent density for 1 day, then treated with trypsin until detachment from cell culture dishes. Cells were stained with CellMask Deep Red Plasma membrane Stain (Thermo Fisher Scientific) for 10 min, then resuspended in culture medium. Around 15,000 cells in 200 µl of medium were transferred to a PDMS well, which was pretreated with 1% Pluronic f127 for 1 h then washed three times with 1X PBS. Cells were incubated in the wells at 37°C and 5% CO₂ for 4 h, then imaged at room temperature for less than 30 min. Cells were imaged by mounting PDMS wells on coverglass and using imaging methods described below. The contact angle was manually measured between isolated heterotypic pairs of cells using ImageJ.

Confocal microscopy of fixed and live samples was conducted on an inverted microscope (Ti-E; Nikon) with a confocal scan head (CSUX; Yokogawa Electric Corporation) and laser-merge module housing 491-, 561-, and 642-nm laser lines (Spectral Applied Research), a stage controller (Prior), and a cooled CCD (HQ2-Roper Scientific) or CMOS (Zyla-Andor) camera. Images were acquired using 20 × 0.75 NA Plan Fluor multi-immersion, 40 × 1.15 NA Plan Apochromat water immersion extra-long working distance and 60 × 1.2 NA Plan Apochromat water immersion objectives (all from Nikon). This set up enabled capture of GFP, mScarlet, AlexaFluor 488/561/642, and SPY650 fluorophores throughout the study. Metamorph software was used for image acquisition. Live-cell imaging utilized a stage incubator for temperature, humidity, and CO₂ control (Chamlide TC and FC-5N; Quorum Technologies). The stage adapter, stage cover, and objective were maintained at 37°C, whereas humidified 5% CO₂ air was maintained at 50°C at its source to prevent condensation within its tubing. Unless otherwise noted, serial z-sections were done at 0.25-μm steps for fixed samples and 0.5-μm steps for live samples. For live, lateral membrane growth studies, cells were preincubated with 50 nM siR-actin (Cytoskeleton) for 1 h prior to trypsinization and immediate transfer to an imaging chamber. A 50 nM siR-actin concentration was maintained in the imaging media.

ImageJ and MatLab were used for image analysis. To make basal nuclei more visible in images of living H2B-labeled 3D cultures the following filtering technique was applied to average projections of the planes spanning 10 μm up from the lowest in focus nuclear signal. The original image was modified by applying a sigma = 40 pixel median filter. The original image was then divided by the transformed image to remove uneven background signal. For studies concerning positional relationships, H2B-labeled cells were considered to have a suprabasal neighbor if Scarlet signal from a neighbor occluded the nuclear signal by >50%. For studies utilizing only H2B-labeled cells, cells were considered suprabasal if their H2B signal occluded >50% of an underlying H2B signal or could be seen to overlie 2 or more H2B-labeled nuclei. Collagen deformation below forming junctions was taken to be the distance between the line connecting the lowest observed collagen positions between two adjacent cells and the highest position marked at forming junctions.

Measurement of cell sorting was conducted by taking serial z-slices through cell aggregates, then, reconstructing 10 orthogonal, radially configured slices per aggregate using ImageJ. Positions of local maxima for Scarlet (noise = 300 px) and GFP-H2B (noise = 60 px) signals were ascertained. MatLab was utilized to determine and connect via a line each maximum to the nearest neighbor bearing the same signal. For each cell aggregate, the number of intersections and maxima detected among the 10 slices was summed and used to calculate the average number of intersections per maxima. Matlab was used to rearrange actual maxima positions to randomized positions within the spatial constraints corresponding to each slice. Dividing the intersections per maxima value for randomized maxima by that acquired from actual observations produced, for each aggregate, a cell separation index score. Higher scores indicated less interaction between networks and a higher degree of cell sorting.

To quantify the extent to which neighboring Scarlet labeled cells invaded the space of dividing, H2B-labeled cells, the Scarlet signal was turned into a binary mask. At each timepoint after anaphase, the presence of this mask within a 5-μm circle centered between the chromosomes of what would become two daughter cells was counted as an intrusion. The circular region of interest was centered upon premitotic and metaphase nuclei to establish a baseline level of intrusion.

Where indicated, independent two-sample Student’s t-tests were utilized to calculate statistical significance with P values <0.05 indicative of a significant difference being detected between means. Data distribution was assumed to be normal, but this was not formally tested. In cases where the sample size was small (n ≤ 6), a nonparametric two-tailed Mann–Whitney U test was performed.

Fig. S1 demonstrates Dia1 localization in human tissue samples. Fig. S2 contains extra analysis of nuclear measurements taken from organotypic cultures. Fig. S3 characterizes efficiency of Dia1 disruption techniques and additional histological images of CRISPR-edited organotypic cultures. Fig. S4 contains imaging of flat-mounted organotypic cultures. Fig. S5 demonstrates shRNA-mediated phenotypes in organotypic culture, presents evidence of ectopically expressed constructs, and provides additional RNAseq data. Fig. S6 shows the response of SPPR genes to Dia1-rescue. Fig. S7 indicates the response of cell density to forced crowding upon 18% prestretched substrates. Fig. S8 has qPCR corroboration of crowding induced changes suggested by RNA-seq. Fig. S9 is a duplicate experiment of the stratification time course. Fig. S10 has additional data regarding proliferation, junctional actin and intercellular contact angles. Fig. S11 demonstrates cell positioning effects of cultures mixed at 50:1 ratios. Fig. S12 presents observations regarding division angles. Fig. S13 has immunostaining images of pMLC and actin during junction development. Fig. S14 demonstrates E-cadherin and vinculin staining during junction development. Tables S1, S2, S3, S4, and S5 provide separate files of the tables displayed in Figs. 3, 4, 7, S10, and S12.

Thank you to Yvonne Beckham for feedback on manuscript preparation and Tracy Chmiel for scripts used in quantifying vertical cell position in Fig. S11.

The above work was supported by the following funding entities: M.L. Gardel acknowledges funding from NIH RO1 GM104032. RMH acknowledges funding via an NIH National Research Service Award (NIGMS:1F32GM117928-01). Christy Schmehl from University of Chicago Human Tissue Resource Center conducted histological embedding and sectioning.

The authors declare no competing financial interests.

Author contributions: R.M. Harmon conceptualized and wrote the manuscript, in addition, to devising, performing, and analyzing experiments. J. Devany aided in quantitative analysis, concept development, and manuscript preparation. M.L. Gardel aided in manuscript preparation, conceptualization, experiment planning, and data interpretation.