Within the Drosophila pupal notum tissue cells divide according to their intephasic cell shape long-axis (a,b), thereby following the 130-year-old Hertwig rule. However, upon entry into mitosis cells round up (the cell shown in a, -15 to -2 min and ).
(a) Time-lapse images of Dlg:GFP in a dividing cell (out of 249 cells quantified in b) in the pupal notum tissue illustrating cell rounding during mitosis (The same cell is shown as inset in .). Prior to mitosis (-30 min) the cell (marked by asterisk) is clearly elongated and divides according to its interphasic cell shape (5 min). Upon entry into mitosis (-15 min) the cell rounds up and reaches a minimal anisotropy just prior to anaphase (-2 min, see also ).
(b) Rose plot of the difference between the experimental (θ_division) and predicted division orientations by the average (60-30 min prior to mitosis) interphase cell long-axis (θ_shape). The data are duplicated relative to 0° line (light green). Number of cells (n) analysed is indicated.
(c,d,e) Gαi localization in fixed epithelial dorsal thorax tissue (c), Pins:YFP localization in pins mutant tissue (d) and GFP:Mud localization (e) showing cells in G2 interphase (left) and mitosis (right). Gαi is hardly detected at the cell cortex in G2 phase and Gαi is mostly homogenously distributed around the cortex during mitosis. Pins:YFP is homogenously distributed around the cell cortex in both interphase and mitotic cells. In mitosis Pins:YFP also weakly localizes at the mitotic spindle. GFP:Mud localize at TCJs during interphase and mitosis (see also f). n=24 cells (c, left); n=19 cells (c, right); n=80 cells (d, left); n=12 cells (d, right); n=111 cells (e, left) and 54 cells (e, right).
(f) GFP:Mud time-lapse images from G2 interphase to telophase (n=21 cells). White arrows: GFP:Mud at TCJ (numbered at t=-22min). Red and yellow arrowheads: GFP:Mud on the spindle and its poles, respectively. The same panels -22min to 4min are shown in . See also .
(g) Apical-basal (AB) sections of the cell in (f) at t=-22min (top) and t=-1min (bottom). White arrows: GFP:Mud at TCJ. n=21 cells.
(h) GFP:Mud kymograph along the cortex (x axis) from t=-22 to t=0 min of the cell in (f). TCJ numbered as in (f). The kymograph shows that during mitotic rounding GFP:Mud spread only modestly along the cortex of the dividing cell. n=21 cells.
(i) AB sections of GFP:Mud, adherens junction marker E-Cad and septate junction marker Dlg (top, n=16 cells) or septate TCJ marker Gli (bottom, n=30 cells).
(j-m) Localizations of GFP:Mud (white in j-m and green in j”-m”) and Gli (white in j’-m’ and red in j”-m”) in fixed pupal wing (j-k) and larval wing disc (l-m) tissues. GFP:Mud colocalizes with Gli at TCJ in G2 interphase and mitotic cells in both the pupal wing and larval wing disc epithelium. Asterisks mark Mud punctate structures present on the nuclear envelope of early G1 cells. Yellow arrows indicate GFP:Mud on the spindle poles. n=20 cells (j-j”); n=5 cells (k-k”); n=63 cells (l-l”) and n=12 cells (m-m”).
(n-o) Localizations of Mud (white in n, o and green in n”, o”) and Gli (white in n’, o’ and red in n”, o”) detected by antibody staining in G2 interphase and mitotic cells in the pupal dorsal thorax tissue. As observed for GFP:Mud ( and ), the endogenous Mud is enriched at TCJ where it co-localizes with Gli in G2 interphase and mitotic cells. Yellow arrows indicate Mud on the spindle poles. n=37 cells (n-n”) and n=21 cells (o-o”).
Scale bars: 1μm (a, c, d, e, f, g, i, j, k, l, m, n, o). 3 min (h).

(a) Scheme depicting the accumulation of the Drosophila FUCCI reporters during the cell cycle. ECFP:E2F1 accumulates during G1 phase, G2 phase and mitosis, whereas mRFP1:CycB accumulates during S phase, G2 phase and mitosis.
(b) Localization of GFP:Mud (green left column and white in the second column panels), mRFP1:CycB (red in the left column and white in the third column panels) and ECFP:E2F1 (blue in the left column and white in the right column panels) in epithelial cells of the pupal notum tissue. Confocal sections at the level of septate junctions are shown. Cells in G1 (n=21), S (n=6), G2 (n=35) phases and mitosis (n=6) are indicated in the left panels. During both G1 and S phase (upper two row panels), GFP:Mud is weakly localized at the nuclear envelope membrane, weakly localized at the cortex and at the apically localized centrioles (not shown). During G2 phase GFP:Mud becomes prominently localized at the TCJ (one cell in the 1^st row panels and 2 cells in the 3^rd row panels). Arrows indicate examples of TCJ GFP:Mud accumulation. During mitosis GFP:Mud remains localized at the TCJ and accumulates on the spindle and the spindle pole (bottom row panels). Similar results were obtained on fixed tissue for which the cell cycle phases were determined using the PCNA S-phase marker and the nucleus size to distinguish cells in G1 or G2 interphases (not shown).
(c) GFP:Mud (green arrows) and ChFP:Mud (red arrows) in adjacent tissue patches in G2 (n=31) and mitotic (n=8) cells. The FLP/FRT system was used to generate adjacent groups of cells labelled with either GFP:Mud or ChFP:Mud. By analysing the distribution of GFP:Mud in dividing cells adjacent to ChFP:Mud interphasic cells, we found that GFP:Mud was localized at the TCJs of the dividing cell from G2 through mitosis.
Scale bars: 1μm (b, c).

(a,b) Localizations of Gli (white in a and green in a’, n=2 clones) and Dlg (white in b and green in b’, n=3 clones) in fixed notum tissues harbouring mud clones (identified by loss of mRFP:nls, red in a’-b’). The loss of Mud function does not modify the Gli and Dlg localizations at septate junctions.
(c) Localization of Gli (white in c and green in c’) in fixed notum tissue harbouring a clone of dlg (identified by the loss of mRFP:nls, red in c’, n=13 clones). The loss of Dlg function results in a loss of Gli localization at TCJ.
(d) Localization of Dlg:GFP (white in d and green in d’) in live epithelial dorsal thorax tissue harbouring a Gli clone (identified by expression of PH:ChFP, red in d’, n=5 clones). The loss of Gli function does not affect the distribution of Dlg:GFP at the septate junctions.
Scale bars: 5μm (a, b, c, d).

(a,a’) Time-lapse images of ChFP:Mud (green in a and top panels a’, white in bottom panel a’) and of Jupiter:GFP (red in a and top panels a’, white in middle panel a’) in dividing cells (n=11) in the Drosophila pupal notum tissue. The panels in a’ are magnifications of the boxed region in a. Yellow arrow points at an astral MT that contact ChFP:Mud at the cortex and shortens concomitant to the spindle pole movement towards the TCJ and spindle rotation. The dashed line corresponds to the initial spindle orientation and the solid lines correspond to its orientation at the final time point (see ). Similar results were obtained in cells expressing GFP:Mud and Tub:RFP to label the MTs (data not shown).
(b) Schematic of the laser-ablation assay used to estimate the origin and magnitude of forces on astral MT required for spindle orientation in the Drosophila pupal dorsal thorax epithelium. Upon ablation (red lines, top), pulling forces (green arrows, left column) or pushing forces (green arrows, right column) yield recoil away (grey arrow, left column) or toward the ablation site (grey arrow, right column), respectively.
Scale bars: 1μm (a).

Loss of Mud activity is known to induce defects in mitotic spindle orientation relative to the apical-basal axis (AB) of the cell,,. Nonetheless, in mud, dlg, dynein (gl^DN) mutant cells around 50% of the epithelial cell divide with an AB angle (α_AB) in the range of the wt tissue (a-e). Since a large proportion of the spindles remain within the plane of the tissue, all analyses reported in the manuscript were performed on cell divisions that occurred within the plane of the tissue. Furthermore, in a central region of the tissue (box in f and f’), 88% of the divisions in mud tissue occur with an α_AB in the range of the wt tissue (g). This region was analysed to compare TCJ bipolarity and cell shape based predictions of division orientation in wt and mud tissue ().
(a) AB views of a dividing epithelial cell in a wt (left panel, out of 257 cells quantified in b) or mud (right panels, out of 176 cells quantified in b) tissue. The spindle is labelled using Jupiter:GFP (green) and the centrosomes using Sas-4:RFP (red). α_AB varies from 0° (spindle parallel to the plane of the tissue) to 90° (spindle perpendicular to the plane of the tissue).
(b) Quantification of α_AB in wt, mud and in mud tissue expressing GFP:Mud (mud, GFP:Mud). In wt tissue, α_AB varies between 0 and 22° (blue dashed line). In mud tissue, 56% of cells divide with a α_AB angle lower than 22° (dashed red lines). The expression of GFP:Mud in mud tissue rescues the spindle AB orientation phenotype caused by Mud loss of function. Numbers of cells (n) for each genotype are indicated. The distribution of angles in mud tissue is significantly different from wt (p<10^-4), and is restored in mud, GFP:mud (p<10^-4). p values, Kolmogorov-Smirnov test.
(c) Quantification of α_AB in wt , Gαi and pins. The loss of either Gαi or Pins function does not affect the orientation of the spindle relative to the plane of tissue (p>0.3) in agreement with our findings that Mud localization at TCJ is independent of Pins and Gαi. The analysis in pins tissue confirmed previously published findings. Number of cells (n) are indicated. p value, Kolmogorov-Smirnov test.
(d, e) Quantification of α_AB in wt, Gli and dlg tissues at 25°C (d) and in wt and gl^DN tissues at 29°C (e). Gli loss of function does not affect α_AB orientation, whereas 46% of the dlg cells (p<10^-4) and 59% of the gl^DN cells (p<10^-4) divide with α_AB lower than 22° and 24°, respectively. Numbers of cells (n) are indicated. p values, Kolmogorov-Smirnov test.
(f-f’) Identification of a region of the notum where AB orientation of the spindle is not affected in mud mutant tissue. Defects in AB orientation of the mitotic spindle result in size asymmetry of the two daughter cells. Therefore daughter cell size was initially used as a proxy for the magnitude of spindle misorientation along the AB axis in mud tissue. The maps of daughter cell size asymmetry in wt (f) and mud (f’) tissues (green, no size asymmetry; purple strong size asymmetry) revealed that a region (highlighted by the black box, f, f’) in the mud notum tissue exhibits almost no defects in daughter cell size asymmetry. Accordingly the quantification of spindle AB orientation within the region in wt and mud tissue revealed that 88% of the cells of the region divide within the range of the wt cells (see g).
Anterior is to the right and the dashed back line indicates the midline. Colour coding; purple: daughter cells with strong size asymmetry, green: daughter cells with normal size symmetry, cyan: cells for which no division was detected, grey: cells which left the field of view and were not analysed, yellow: macrocheatae, white: sensory organ precursors (SOPs).
(g) Quantification of α_AB in wt and mud tissue in the boxed regions in f-f’ was performed as in b-e. Numbers of cells (n) for each genotype are indicated.
Scale bars:1 μm (a), 100 μm (f-f’).

(a) Mitotic cell in the Drosophila pupal notum labelled with Jupiter:GFP to label MTs (n=23 cells). White arrows point at astral MTs. Yellow arrowheads indicate spindle poles. Scale bar: 1μm.
(b) Representation of the different parameters that were varied for the predictions based on the GFP:Mud cortical intensity and shape model to estimate their contribution. L: length of the mitotic spindle, N: number of astral MTs, Φ: the angle covered by the astral MTs and α: the GFP:Mud intensity scaling factor. See also .
(c-f) Cumulative plots of the differences between the theoretical spindle orientation (θ_theory) and the experimental spindle orientation (θ_division) angles in GFP:Mud expressing cells (same cells as in ) for different spindle lengths (c), MT number (d), angular extension of astral MTs (e) and different scaling factor between the GFP:Mud intensity and mechanical pulling force (f). The GFP:Mud model predictions are mostly independent of spindle length, the number of astral MTs, the angle covered by the astral MTs or the scaling factor between GFP:Mud intensity and MT pulling force.
(g) Dependence of model prediction on shape or GFP:Mud effective potential depth (±s.e.m.). The y-axis quantitates the difference between the theoretical angle (θ_t) and experimental angle (θ_d) (1: aligned, -1: perpendicular). A larger potential depth corresponds to more deformed cells for the shape model, and to a sharp and anisotropic GFP:Mud distribution for the cortical model. Model predictions improve with potential depth, suggesting the model can capture the effect of GFP:Mud distributions in a dose-dependent manner. n=140 cells
(h) Definitions of the angles used in the analytical calculation of the contribution of different harmonics to the potential (U(θ). The spindle (heavy black line) makes an angle θ with the positive x axis. An astral MT (thin black line indicated by the black arrow) projects to the cortex (circle) at an angle φ with respect to the spindle. The same MT contacts the cortex an angle β = ψ + θ above the positive x axis.
(i) Normalized magnitudes |ũ_n|/|ũ₂| of the Fourier coefficients of the kernel ũ(Ψ) for n even. The magnitudes |ũ_n| drop off substantially with increasing n, indicating that for many purposes it should be sufficient to approximate the function U by its lowest, n = 2 mode. To calculate numerical values for the Fourier coefficients, we took the average of the normalized spindle length ϵ = L/2R for the 140 cells (n) analysed in this paper, obtaining ϵ ≈ 0.76 ± 0.03; because it is difficult to precisely estimate Φ from the available data, coefficients are shown for Φ = 180° and 270° in agreement with the astral MT distribution observed in a.
(j) Schematic illustrating the difference between cell shape and cell TCJ bipolarity measurements. An elongated cell and a rounded cell are overlaid (left panels) and shown side-by-side (middle and right panels). In this example, although the two cells have distinct shapes, they have the same TCJ bipolarity. The upper panels illustrate the measurement of cell shape, which uses all the pixels making up the cell (blue bars). The lower panels illustrate the measurement of TCJ bipolarity (red bars), which is solely based on the angular distribution of the TCJs (red dots), only using the unit vectors pointing from the cell center (black dot) to each cell TCJ. The TCJ bipolarity therefore characterizes TCJ distribution independently of cell shape, and a correlation observed between the two quantities is not due to a shape bias in the TCJ bipolarity measurement.

(a) Rose plots of the difference between the theoretically predicted (θ_theory) and the experimental division (θ_division) orientation of the mitotic spindle in pins tissue (orange left rose plot) and wt tissue (green right) based on the GFP:Mud intensity. To facilitate the comparison between the left and the right rose plots, the data are duplicated relative to 0° line (light orange and light green).Number of cells (n) analysed is indicated. p value, Kolmogorov-Smirnov test.
(b) Quantifications of the co-localization of GFP:Mud with Gli in pins in metaphase cells (±s.e.m.). Number of cells (n) analysed is indicated. ns: not significant (Student t-test).

(a) Diagram of the domains of the Mud protein: putative actin binding calponin homology domain (CH, aa 1-246, blue), coiled-coil domain (CC, aa 246-1868, grey), conserved Numa/Lin-5/Mud domain (NLM, aa 1968-1998, red), putative transmembrane domain (TM, aa 2456-2499, yellow), 9x repeat domain (aa 1137-1515), MT binding domain (MT, aa 1850-2039) and Pins binding domain (aa 1928-1982)–. GFP or ChFP tagged deletion constructs and the MudΔCH constructs were generated by BAC recombineering (see for details). The MudΔC allele was generated at the mud locus using a CRISPR/Cas9 approach (see for details). For each mutant allele, its localization at the TJCs and its localization at the spindle pole are indicated.
(b) Localization of the GFP:Mud, MudΔCH, GFP:MudΔCC, GFP:MudΔPINS, GFP:MudΔTM and in G2 interphase and mitotic mud epithelial cells. GFP:Mud, GFP:MudΔCC, GFP:MudΔPINS, GFP:MudΔTM proteins were imaged in living tissue, whereas MudΔCH was localised on fixed tissue using Mud antibodies. GFP:Mud (n=56), MudΔCH (n=33), GFP:MudΔCC (n=165), GFP:MudΔPINS (n=42) and GFP:MudΔTM (n=67) interphase cells. GFP:Mud (n=15), MudΔCH (n=4), GFP:MudΔCC (n=67), GFP:MudΔPINS (n=18) and GFP:MudΔTM (n=11) mitotic cells.
(c) Localization of the Mud^ΔC protein (white in the left panels, green in the right panels), Gli (white in the panels in the middle and red in panels at the right) and Cora (magenta in the right panels) in fixed G2 interphase (n=71) and mitotic (n=6) cells. The Mud^ΔC protein is not enriched at TJCs and its localization at the spindle pole is strongly reduced.
(d) Rose plots of the difference between the theoretically predicted (θ_theory) and experimental (θ_division) spindle orientation angles in wt (left rose plot, green) and mud (right rose plot, orange) tissues based on the distribution of GFP:MudΔCC. The right rose plot is identical to the one shown in . To facilitate the comparison between the left and the right rose plots, the data are duplicated relative to 0° line (light green and light orange). Number of cells (n) analysed is indicated. p value, Kolmogorov-Smirnov test.
Scale bars: 1 μm (b, c).

(a-a”) Rose plots of the magnitude of the difference between experimental (θ_division) and predicted division orientations by the average (60-30 min prior to mitosis) interphase TCJ bipolarity (θ_TCJ) or cell long-axis (θ_shape) in cells for the indicated |θ_TCJ – θ_shape| intervals. To facilitate the comparison between the left and the right rose plots, the data are duplicated relative to 0° line (light blue and light red). Kolmogorov-Smirnov test (p values), percentage of total cells (n=29388). Panels b and b” are identical to panels e and e’ in .
(b-b”) Rose plots of the magnitude of the difference between experimental (θ_division) and predicted division orientations by the average (60-30min prior to mitosis) interphase TCJ bipolarity (θ_TCJ) or cell long-axis (θ_shape) for the indicated η_shape intervals. To facilitate the comparison between the left and the right rose plots, the data are duplicated relative to 0° line (light blue and light red). Kolmogorov-Smirnov test (p values), percentage of total cells (n=29388). Panels c and c” are identical to panels f and f’ in .
(c) Plot of the spindle orientation prediction improvements (color-coded from dark blue to red) based on TCJ bipolarity over those based on cell shape versus the magnitude of their angular difference (|θ_TCJ – θ_shape|) and the cell shape anisotropy (η_shape). The plot height is the normalized cell number in each domain of the plot (29883 cell were analysed in total). As |θ_TCJ – θ_shape| increases, the TCJ bipolarity predictions improve over cell shape prediction for both rounded (low η_shape) and elongated cells (high η_shape). Whereas the rounded cells are characterized by an even distribution along the |θ_TCJ – θ_shape| axis, the elongated cells are mainly characterized by a strongly skewed distribution towards low |θ_TCJ – θ_shape|.

(a,b) Images of the scutellum tissue before and after ablation (ablated region in yellow) in early and late pupa characterized by small isotropic stress (a) and high anisotropic stress (b). Tissue stress was estimated by determining the initial recoiled velocity upon circular ablation in the x and y directions. First and last images of two time-lapse movies out of the 18 quantified in c are shown. Scale bars: 10μm (a, b).
(c) Plot of the difference between the orientation of TCJ bipolarity (θ_TCJ) and principal strain axis (θ_strain) as a function of normal stress differences (σ_yy – σ_xx, note that σ_xy = 0) as estimated up to a prefactor by circular laser ablation. Number of ablations (n) analysed is indicated. The same plot is shown in .
(d) Plot of the difference between the orientation of TCJ bipolarity (θ_TCJ) and the orientation of strain (θ_strain) as a function of the percentage of cell elongation applied to a simulated cell lattice. When cell elongation increases TCJ bipolarity orientation becomes aligned with the direction of cell elongation. Number of simulations (n) analysed is indicated.

(a) GFP:Mud from interphase to telophase (t=0 min, anaphase). GFP:Mud at TCJ (arrows), spindle poles (arrowheads). n=21 cells.
(b) GFP:Mud and Gli co-localization in interphase (top, n=54 cells) and metaphase (bottom, n=8 cells).
(c) GFP:Mud localization in mud (n=15), pins (n=22), Gαi (n=5) cells and GFP:MudΔPINS in mud cells (n=18).
(d,e) GFP:Mud distribution (d, images representative of quantifications shown in e) and mean±s.e.m. TCJ intensities (e) in wt, Gli, dlg and pins cells. Fas3, cell contours. t-test (ns: not significant, ***:p<0.0005). n: cell numbers.
Scale bars: 1μm (a, b, c, d).

(a) Ablation of astral MTs (red line), n=21 cells quantified in b.
(b) Mean centrosome velocity relative to MT ablation site (left), mean velocity amplitude after ablation (mean±s.e.m, right) in wt, mud, dlg and Gli cells at 25°C and in wt and gl^DN cells at 29°C. t-test (*:p<0.05). Orientations are different in mud, dlg and gl^DN (Watson’s U² test, p<0.01).
(c,d) Cell shape (c) and Mud intensity (d) models: pulling forces scale with MT length (blue arrows) or Mud cortical intensity (red arrows) to exert a torque (T, arrows).
(e-g) Experimental spindle orientation (green cross) and predictions based on cell shape (blue circles, f) or GFP:Mud intensity (red circles, g) potentials at t=-1min for cell in e (n=121 cells).
(h) Difference between theoretically predicted (θ_theory) (blue: shape, red: GFP:Mud intensity) and experimental (θ_division) spindle orientation. Data are duplicated in a lighter colour relative to 0° line in this and subsequent plots. Kolmogorov-Smirnov test (p value).
(i) Localizations of GFP:Mud in wt (n=54) and mud (n=15) cells as well as of GFP:MudΔCC in wt (n=18) and mud (n=67) cells.
(j) Quantifications (mean±s.e.m) of GFP:Mud or GFP:MudΔCC co-localization with Gli in wt and mud cells. t-test. ns: not significant.
(k) Mean centrosome velocity relative to MT ablation (left), mean velocity amplitude after ablation (mean ± s.e.m, right) in wt and in mud tissues expressing GFP:Mud or GFP:MudΔCC. t-test (**:p<0.005). Orientation is different in mud, GFP:MudΔCC (Watson’s U² test, p<0.001).
(l) Difference between θ_theory (from cortical GFP intensity) and θ_division in mud cells expressing GFP:MudΔCC or GFP:Mud. GFP:Mud in mud and wt (h, red) tissue are similar (p=0.12). Kolmogorov-Smirnov test (p values).
Scale bars: 1μm (a, e, i), n: cell numbers (b, h, j, k, l).

(a) TCJ (red dots) bipolarity and cell shape (blue) anisotropies η and orientations θ represented by the length and orientation of red and blue bars.
(b) η_TCJ and η_shape from interphase to anaphase (mean±s.e.m). Insets: time-lapse images of a cell from interphase to mitotic rounding (n=249 cells). t-test (*:p<0.05, ***:p<0.0005).
(c-c’) Difference between experimental (θ_division) and predicted division orientations by the average (-20 to -10min interphase, -4 to -3min metaphase) cell long-axis (θ_shape) (c) or TCJ bipolarity (θ_TCJ) (c’). Kolmogorov-Smirnov test (p values).
(d) Top: θ_shape and θ_TCJ align with θ_division. Bottom: only θ_TCJ aligns with θ_division. Time-lapse images of 2 cells out of the 29388 cells analysed.
(e-f’) Difference between experimental (θ_division) and predicted division orientations based on interphase TCJ bipolarity (θ_TCJ) or cell long-axis (θ_shape) for |θ_TCJ – θ_shape| intervals (e-e’) and indicated η_shape intervals (f-f’). Kolmogorov-Smirnov test (p values), percentage of 29388 cells.
(g) TCJ bipolarity (θ_TCJ) prediction improvement over cell long-axis (θ_shape) versus |θ_TCJ – θ_shape|. mean±s.e.m of three movies for a total of n=29388 cells analyzed.
(h) Differences (mean±s.d.) between θ_shape and θ_TCJ versus η_shape. Correlation coefficient A_experiment = 0.88.
(i-i’) Difference between experimental (θ_division) and predicted division orientations based on interphase cell long-axis (θ_shape) (left) or the TCJ bipolarity (θ_TCJ) (right) in wt and mud cells. Kolmogorov-Smirnov test (p values).
Scale bars: 1μm (a, b), n: cell numbers (b, g, h).

(a-b) Regular hexagonal cells (a) and simulated cell lattice (b) before and after elongation.
(c) Differences (mean±s.d.) between θ_shape and θ_TCJ versus η_shape for experimental () and simulated cells. A_stimulation = 0.94. n: cell numbers.
(d) Difference between θ_TCJ and principal strain axis (θ_strain) versus normal stress differences. n: ablation number.
(e) Upon mitotic rounding, Mud interphase localization is maintained at TCJs orienting the spindle along the interphase cell long-axis.