Intermediate filament reorganization dynamically influences cancer cell alignment and migration

The interactions between a cancer cell and its extracellular matrix (ECM) have been the focus of an increasing amount of investigation. The role of the intermediate filament keratin in cancer has also been coming into focus of late, but more research is needed to understand how this piece fits in the puzzle of cytoskeleton-mediated invasion and metastasis. In Panc-1 invasive pancreatic cancer cells, keratin phosphorylation in conjunction with actin inhibition was found to be sufficient to reduce cell area below either treatment alone. We then analyzed intersecting keratin and actin fibers in the cytoskeleton of cyclically stretched cells and found no directional correlation. The role of keratin organization in Panc-1 cellular morphological adaptation and directed migration was then analyzed by culturing cells on cyclically stretched polydimethylsiloxane (PDMS) substrates, nanoscale grates, and rigid pillars. In general, the reorganization of the keratin cytoskeleton allows the cell to become more ‘mobile’- exhibiting faster and more directed migration and orientation in response to external stimuli. By combining keratin network perturbation with a variety of physical ECM signals, we demonstrate the interconnected nature of the architecture inside the cell and the scaffolding outside of it, and highlight the key elements facilitating cancer cell-ECM interactions.


Results and Discussion
Keratin phosphorylation does not affect Panc-1 cell area but does affect actin inhibition responses. The effects of cytoskeletal elements on cell morphology were analyzed by treating Panc-1 cells on flat PDMS substrates with chemical inhibitors of microtubules (nocodazole), and actin (cytochalasin D) in conjunction with keratin phosphorylation (SPC). Phase contrast images were captured in 6.5 minute intervals, starting 45 minutes prior to treatment and finishing once equilibrium area was reached. Cell area at each time point was calculated from cell outlines, with the area at each time point normalized to the area of the cell immediately prior to treatment (Fig. 1A).
In untreated cells, the keratin network spans the cell from the perinuclear region to the cell membrane ( Fig. 1D) 33 , while the SPC treatment concentrates the keratin almost completely around the nucleus (Fig. 1E), a pattern that is also seen in the intermediate filament vimentin (Fig. 1F,G). The absolute and normalized areas of untreated and SPC-treated cells were found to be equivalent over the course of the treatment (Fig. 1B). As with SPC treatment alone, nocodazole either alone or in parallel with SPC did not have a significant impact on cell area up to 100 minutes after treatment (Fig. 1B), allowing for the conclusion that microtubules do not play a significant role in the stabilization of cell area. Cytochalasin D treatment has previously been reported to be sufficient to drastically reduce cell area 55 , and in Panc-1 cells a 46% reduction was observed (Fig. 1C). Interestingly, this effect appears to be enhanced by SPC treatment, with parallel application resulting in a cell area decrease of as much as 59% (Fig. 1C), significantly larger (p < 0.01) than the decrease in area from Cytochalasin D alone. Based on these results, it appears that keratin and actin dynamics are interconnected-when keratin reorganizes, leaving a void in the previously enmeshed cortical region of the cell, the actin cytoskeleton may be more free to reorganize without the steric hindrance that a well defined keratin network provides. This makes keratin an attractive element for perturbation-on its own, SPC treatment will not affect simple cellular metrics like spread area, but it can free up cytoplasmic space through which the more responsive actin network can move.

Keratin dynamics are independent of actin orientation in stretched Panc-1 cells. Structural
interactions between the actin and keratin networks were analyzed by comparing their spatial orientation in relation to each other. To do this, cells on flat control and cyclically stretched (2 Hz, 8%) substrates were stretched for eight hours and stained for both actin and keratin, and a custom ImageJ macro was used to correlate cytoskeletal alignment ( Fig. 2A-D). Electron microscopy performed after solubilizing non-keratin cytoskeletal elements 33 was used to compensate for the fact that immunofluorescence was not sufficient to visualize the keratin network at the periphery of the cell (Fig. 2E-H).
For cells on unstretched control PDMS substrates, the orientation of the actin fibers was randomly distributed, while cells exposed to cyclic strain exhibited fibers aligned perpendicularly to the direction of force. These alignment patterns were unaffected by SPC-mediated keratin network phosphorylation (Fig. 2K). Cells plated on unstretched substrates showed no preferential alignment of the perinuclear or peripheral keratin network, but did exhibit keratin alignment perpendicular to the direction of strain in both the periphery and the perinuclear regions, although at a lower magnitude than the actin fiber alignment (Fig. 2I,J). SPC treatment was sufficient to suppress this mechanoresponse in the perinuclear region of the cell, but in the periphery, alignment was only reduced by 33% (Fig. 2J). Higher levels of keratin alignment in the periphery of the cell during cyclic strain could be a result of the increasing role of mechanosensitive focal adhesion activity and turnover 56 .
Correlation coefficients were calculated using orientation values for any actin and keratin fibers found to intersect. From these measurements, an average correlation coefficient was calculated for each experimental condition (Fig. 2L). Although the orientation of the keratin in the stretch experiments had a slight alignment bias perpendicular to the direction of pull, no significant correlation between actin orientation and keratin orientation in fiber intersection points was found, as all values were statistically equivalent to 0. This result is surprising given the fact that both systems realign in the same direction, but it reinforces the hypothesis that the two networks behave independently and likely do not bind to each other during the realignment process. In special consideration of the fact that keratin alignment in response to cyclic strain has never been reported, only SPC-mediated keratin phosphorylation was analyzed in conjunction with substrate topography-based migration and morphology. Panc-1 cell orientation dynamics as a function of cyclic strain and keratin phosphorylation. Untreated and SPC-treated Panc-1 cells were found to be randomly polarized on flat PDMS substrates (S control = S SPC = 0.00 ± 0.02) (Fig. S2A), but after nearly 8 hours of uniaxial cyclic substrate strain (2 Hz, 8%), a large percentage of the cells had realigned in a direction perpendicular to the strain, mirroring the previous alignment of keratin fibers (Fig. 3A).
To investigate the effect of SPC treatment on cell morphology restructuring, the frequency of cyclic substrate strain was varied between 0.05 and 2 Hz, both with and without SPC treatment, and cell orientation was analyzed after 9 hours. In general, as the frequency increased above a threshold of 0.2 Hz, SPC-treated cells were able to realign perpendicular to the direction of strain more efficiently than the untreated cells (Fig. 3B). Thus, keratin network phosphorylation and subsequent restructuring plays a role in force-induced cell orientation by increasing the degree of morphological adaptation in response to periodic substrate strain.
To understand if cellular keratin restructuring affects the dynamics of this reorientation, the four frequencies of cyclic strain found to induce significantly different orientation parameters in combination with SPC treatment (0.2, 0.5, 1.0, and 2.0 Hz) were applied to untreated and SPC-treated cells, and orientation parameters were calculated at 22 time points over five hours, at which time each cell population had reached an orientation equilibrium. The time constant τ (time to reach 63% of the equilibrium value) was then calculated based on these curves using equation 2 (Fig. 3C). Regardless of SPC treatment, τ decreases as the frequency increases, indicating that higher frequencies stimulate faster cell realignment. SPC treatment is capable of augmenting this effect, but only up to a point; at 2.0 Hz the time constant is equivalent for untreated and SPC-treated populations. We can infer from this that keratin reorganization provides a degree of responsiveness to the cells that can be saturated at high stretch frequencies-it is possible that this level of strain may partially induce defects in the keratin network and thus show no further effect 57 . The general observation that an intact keratin network in the cell periphery slows down actin-driven cell reorganization provides evidence that keratin may function as a type of cytoskeletal dashpot, providing viscoelastic steric hindrance to the more reactive actin network that has been shown to drive cell reorganization 58 .
Cyclic strain influences Panc-1 cell migration. In addition to cell orientation, the migratory properties of the Panc-1 cells could be calculated during the stretching experiments. This complements the orientation analysis, which focused on cell spanning cytoskeletal networks, by considering the cyclic strain effect on the emergence of newly polymerizing lamellipodia-like networks. Cells seeded on flat control substrates without any strain moved randomly with no preferred direction (Fig. S2A), while on flat control substrates being stretched in the X-direction, cells exhibited biased migration along the Y-axis, perpendicular to the direction of pull (Fig. 3D).
To characterize this behavior, the orientation parameter S M was calculated using the angle between the cell's migration vector and the direction of strain (the X-axis) (Fig. S1B). S M was then calculated for each strain frequency f, both with and without SPC treatment. A decrease in S M can be observed with an increase in frequency f, which corresponds to an increase in the number of cells migrating perpendicular to the direction of strain. This effect is enhanced in cells treated with SPC, but only for frequencies of 0.5 Hz and greater. A decrease in S M at high frequencies as a result of SPC treatment was observed, with the frequency required for the cells to reach the arbitrary migratory orientation value S M = − 0.37 halving from 2 Hz to 1 Hz (Fig. 3E).
Additionally, the average velocity of Panc-1 cells exposed to different strain frequencies was determined. A large distribution of cellular velocities was observed, with the average speeds of the fastest cells as much as sixfold higher than the slowest cells. In many cases, the average speed of cells on stretched substrates was found to not be significantly different from those on unstretched surfaces. Thus, the change in migration vector orientation was likely not a function of any changes in migration speed (Fig. 3F), as no systemic effect of the stretching on speed was observed. On cells treated with SPC, average cell velocity was significantly higher for most stretching conditions, including the unstretched control. This is consistent with our hypothesis that the restructuring of the keratin network provides a less hindered environment for the cell migration-driving actin network to function more efficiently, with the output in this case being an increase in cytoskeleton-generated velocity.

Panc-1 cell migration on grooved substrates is dependent on keratin phosphorylation and groove depth.
To examine the relationship between contact guidance and keratin network organization, untreated and SPC-treated Panc-1 cells were grown on substrates with and without grooves. On grooved substrates, immunofluorescence staining revealed that the actin network tended to show groove direction-dependent patterns, while the keratin network did not respond as noticeably (Fig. S3). In general, cells plated on grooved substrates tended to align and migrate along the grooves (S > 0), as opposed to the random alignment and movement on flat control substrates (S ≈ 0) (Fig. 4A,B). The angular distributions of untreated and SPC-treated cell migration were plotted in the first quadrant (Fig. S2B), illustrating this bias in migration direction. Furthermore, a decrease in x-axis directed migration is seen in SPC-treated cells on grooved substrates. Thus, the microstructured surface can affect the migration direction of Panc-1 cells, and this effect is enhanced by destabilization of the keratin network. This could potentially be due to an increased degree of cell body deformation into and around the physical microenvironment as a result of the reduction in the stiff keratin network in the periphery of the cell.
By altering the depth or width of the grooves, several important observations could be made. First, the Panc-1 cells aligned along the grooves for all substrates, regardless of the groove dimensions or SPC treatment (Figs 4A,B, S2B). SPC treatment did result in an increased alignment along the grooves for all tested dimensions. Increasing the depth of the grooves from 200 nm to 350 nm and increasing the width of the grooves from 2 μ m to 4 μ m were both sufficient to increase alignment, with the biggest effect coming from increasing depth. SPC treated cells were also more aligned on deeper grooves, but the groove width and spacing had no effect on SPC treated cells (Figs 4B, S2B).
On control substrates, the ratio of Y-axis movement to X-axis movement is equivalent to 1, indicating unbiased migration. For untreated cells, this ratio increases with increasing groove width and depth, indicating migration parallel to the groove structures. SPC treatment results in further increases in M y /M x , but again, among SPC-treated populations, the only physical parameter that resulted in different migration characteristics was groove depth (Fig. 4C). Increases in groove depth and width also spurred an increase in directional persistence, indicating smoother, more biased migration, while SPC-treated populations were only affected by increases in groove depth and unaffected by groove width dimensions, although migration persistence was higher across the board in treated cells compared to untreated ones (Fig. 4D). Interestingly, groove dimensions played no role in cell velocity, as all conditions resulted in untreated cell velocities of approximately 0.5 μ m/minute. The same was true for the SPC-treated populations, which are generally faster than untreated cells on control substrates. Their velocity was independent of groove dimensions at approximately 0.75 μ m/minute (Fig. 4E). It is of note that these observed velocities were statistically similar to values obtained on flat control PDMS surfaces, indicating that grooved substrates do not alter the inherent cell velocity of Panc-1 cells.
Altogether, it is clear that the grooved substrates are capable of biasing untreated cell alignment, migration direction, and directional persistence, and that this effect is amplified by both deeper and wider grooves, as opposed to SPC-treated populations that are only affected by groove depth. Despite the fact that SPC treatment did not play a role in the mechanotransduction of wider grooves, the response to groove depth was generally stronger than in untreated cells, suggesting improvements in contact guidance. For the treated cells, this contact guidance enhancement can be explained by noting that the cells are mechanically softer as a result of the loss of an intact keratin network, allowing them to physically adapt better the grooves or other microscale features. To see if other microscale features also effect differences in contact guidance, cells were next analyzed on PDMS micropillars.
Scientific RepoRts | 7:45152 | DOI: 10.1038/srep45152 Migration on pillar substrates. As an alternative to the grooved structures, pillar substrates were used in order to present a micro-structured surface to the cells with a different type of anisotropy. In contrast to the flat control PDMS and the grooved substrates, the fibronectin was only added to the tops of the pillars. As a result, cells could only adhere to the pillar tops via focal adhesions (any cells moving between the pillars were excluded from analysis). This results in a cell morphology in which the periphery is supported along the tops of the columns (Fig. 5A).
The average velocity of cells on the pillars was calculated to be 0.45 ± 0.02 μ m/minute, not significantly different from velocity on control substrates. SPC treatment increased the speed of the cells to 0.58 ± 0.02 μ m/minute, which was a lower increase than that observed on control and grooved substrates.
Trajectories of untreated and SPC-treated cells migrating on pillars with 4 μ m diameters, 8.7 μ m heights, and 6 μ m spacings were calculated (Fig. 5B,C). The difference between untreated and SPC-treated cells is smaller than the differences observed on grooved substrates, but close inspection reveals that SPC-treated cells tend to prefer movement in either the X-or Y-axis. To investigate this, frequency histograms of the migration trajectories of untreated and SPC-treated cells were constructed (Fig. S2C) 59 . The untreated plot shows the characteristic flat distribution expected for randomly migrating cells, but the SPC-treated plot shows distinct peaks in both the vertical Y-axis direction and the horizontal X-axis direction, indicating a clear orthogonal bias in migration. This corresponds to the axes of symmetry in the pillared substrates leading to the conclusion that keratin phosphorylation is capable of enhancing the mechanosensitivity of migrating cancer cells, as on the pillars the X-and Y-axis directions correspond to the closest physical signal available to the cells.  For both the untreated and SPC-treated cell populations, the average orientation parameter S was 0.00 ± 0.04, indicating random cell alignment. This result appears to be at odds with the orthogonally aligned results from SPC-treated cells shown in Fig. S2C, but this is likely due to preferred alignment in the X or Y direction being cancelled out. This is similar to the calculation of the M y /M x parameter, which was found to be 1.01 ± 0.04 for untreated cells and 1.02 ± 0.06 for SPC-treated cells. Overall, treatment of Panc-1 cells with SPC and the resultant keratin reorganization can influence the migration dynamics of Panc-1 cells on pillars in a similar manner to that found on grooved substrates.

Conclusions
In general, cells with less keratin expression have been shown to be softer and more susceptible to deformation 16 . Furthermore, invasive cancer cells have been found to be much softer than both healthy cells and non-invasive cancer cells 60,61 . In many cases, this loss of stiffness has been analyzed in the context of three-dimensional cell migration and invasion 16,60 . While the migratory mechanisms underlying three-dimensional invasion are in many ways distinct from those guiding two-dimensional migration, there is evidence that nanoscale topography of two dimensional substrates is correlated with alterations in cell stiffness 62 . As the interactions between nanoscale substrate conditions and cancer cells has been shown to play a role in cancer treatment in vivo 63 , an elucidation of this link in vitro is important. We have shown that in Panc-1 pancreatic cancer cells, the phosphorylation of keratin leads to perinuclear reorganization, in agreement with previous work 64 . Our observation that Panc-1 cells align perpendicularly to cyclic substrate stretch is in line with previous observations in other cell types 57 , but in contrast to a study on tumor capillary endothelial cells, which showed reduced perpendicular rearrangement after exposure to a 1 Hz, 10% uniaxial strain 65 . This may be due to differences in tissue origin; it is intriguing to wonder whether and how SPC treatment might play a role in stretch-responsive orientation in other cancer cell lines similar to the effect seen in Panc-1 cells. Ultimately, this effect may be due to the effects of steric hindrance. As the keratin network (and likely the vimentin network, based on recently published studies 32 and the data in Fig. 1G) becomes destabilized and reorganizes around the nucleus, the cell volume in the periphery becomes less entangled. This reduction in steric hindrance or entanglement may then be enough to generally unencumber the actin cytoskeleton, which has been shown in many studies to be a key player if not the main effector of a wide variety of cytoskeleton-based characteristics, including cell orientation, migration, and mechanotransduction [66][67][68] .

Materials and Methods
Cell Culture. All chemicals were purchased from Life Technologies (Darmstadt, Germany). Human pancreatic cancer cells (Panc-1) (ECACC #87092802) were used for all cell experiments due to their previous keratin cytoskeleton characterization 33  Sample Preparation. PDMS substrates were cleaned in 70% ethanol followed by PBS for 5 minutes each.
The substrate surface was then coated with 50 μ g/mL human fibronectin (Calbiochem #341635, Germany) for two hours, followed by two PBS rinses. Panc-1 cells were then seeded at a concentration of 80 cells/mm 2 and stored overnight in an incubator. One hour prior to the start of an experiment, the substrate was mounted, leveled, and allowed to equilibrate in experimental media (5% FBS) in an incubating chamber and the cells were also placed in experimental media. For inhibitor experiments, the experimental media was dosed to a concentration of 10 μ Μ . Additional images were taken with a scanning electron microscope (SEM) (Ultra-55 Field Emission SEM, Gemini, Zeiss, Germany) and a white light interferometer (WLI) (Zygo NV5000, Middlefield, CT).
Cell Substrate Stretching. Cell substrate stretching was performed using a stretching device developed and elaborated on in previous works 57 , consisting of a DC servo motor (Faulhaber, Schöneich, Germany) which can apply a periodic stretch to clamped PDMS substrates along a single axis. The stretching device and microscope were controlled by an integrated, custom-built Visual Basic program (Microsoft, USA) in AxioVision software. The stretch frequencies used were 0.05, 0.1, 0.2, 0.5, 1.0, and 2.0 Hz with a constant amplitude of 8%. Three independent experiments were performed for each frequency setting, each with a duration of 8 hours.
Morphology Analysis. Time-lapse phase contrast images were used for cell morphology analysis in ImageJ (NIH). Metrics calculated were cell orientation, a time constant representing the response to strain over time, and cell area. For each metric, multiple cells from three independent experiments were averaged. The orientation parameter S was calculated from outlines of each cell manually drawn in ImageJ and corresponding fitted ellipses. The angle between the major axis of the ellipse and the direction of substrate stretching was then calculated. In control experiments, the X-axis was used as a reference direction. The average of the orientation of cells in response to the cyclic strain of the substrates was quantified by the non-polar orientation parameter S in equation 1 57,59 . The angular distribution function f(ϕ) and the orientation angle ϕ are given by the Fokker-Planck equation 71 from which a order parameter can be calculated.   Analogous to the metric of chemotactic index 45 , a parameter measuring the cell migration projected onto two perpendicular axes was calculated. For this study, the X-axis corresponded to the direction of pull in strain experiments, while the Y-axis corresponded to the direction of the grooves in the grooved substrates. In control experiments, the X-axis corresponded to the arbitrary axis. The movements of a given cell along these axes were determined using equation 4 for every migration step, allowing for the ratio (M y /M x ) i to be calculated. A mean value for each cell could be calculated by averaging the ratio from each time point, shown here as = Photolithography. Photolithography with S1818 positive photoresist was utilized to produce the groove and pillar structures. A 3-inch p-type silicon wafer (100 orientation, Siegert Consulting, Aachen, Germany) was placed in Piranha solution (3:1 sulfuric acid to 30% hydrogen peroxide) for 2 hours. Next, the wafer was rinsed in acetone and isopropanol for three minutes each, followed by a rinse in ultrapure water and heating for 10 minutes on a 200 °C hotplate. The wafer was then centered on a spin coater (WS-400B-6NPP-Lite, Laurell Technologies Corp., USA) and blown dry with a nitrogen gun. 3 mL of S1818 positive photoresist were added to the center of the wafer, and the resist was spread in a two step spinning program (500 rpm for 5 seconds, followed by 3000 rpm for 30 seconds) optimized to obtain a desired photoresist thickness of approximately 1.8 μ m. Excess solvent was then evaporated by heating at 115 °C for one minute, and the wafer stored at room temperature for five hours. Alignment with a micropatterned chrome photomask (ML&C, Jena, Germany) displaying either 1 cm 2 fields of circles of diameter 4 μ m and center spacing 10 μ m in a square arrangement or 1 cm 2 fields were filled with long rectangles of 2 or 4 μ m width and 2 or 4 μ m center spacing was performed in an MJB4 manual mask aligner (Sweet and Microtech, Germany) with an LIF350 filter. Exposure was carried out using a 350 W mercury vapor lamp with a wavelength of 365 nm for two seconds with hard contact wedge error correction (WEC) at 1.7 bar. Areas of photoresist exposed to UV were dissolved and washed away in 30-40 seconds using MF-319 Microposit developer (MicroChem, USA) followed by rinsing with ultrapure water and drying with a nitrogen gun.

Reactive Ion Etching (RIE). RIE with the ion etcher Plasmalab 80 Plus (Oxford Instruments, England)
was used in conjunction with photolithography to achieve the high aspect ratios associated with micropillars. After the photolithography-patterned mask was placed inside the etcher, the chamber was cooled to − 20 °C to increase etching selectivity. The etching process consisted of 4 sub-steps constituting one cycle. After each etching and passivation step, the chamber was pumped with a vacuum to eliminate waste products. Table 1 contains the complete steps and parameters for the entire etch process. A mixture of SF 6 and HCF 3 at a ratio of 5:6 was used as the etching gas, and during passivation steps only HCF 3 was introduced into the chamber. The photoresist-covered Si wafers were subjected to 150 cycles, providing an etch depth of approximately 8.7 μ m. A polyurethane polymer consisting of 66% EBECRYL284 diacrylate prepolymer (Cytec Surface Specialties, Germany), 30% M3160 (MIWON, Korea), 1.5% Irgacure184 and 1.5% Darocur1173 (Ciba Chemicals Specialty, Switzerland) and 1% Rad2200N (Tego Chemie Service, Germany) was then cast to make the final mold. The sample was stored for 30 minutes at room temperature to remove air pockets, then placed in an ultraviolet radiation chamber (UVACube 100, Honle UV Technology, Germany) and irradiated for one minute with UVA light, followed by baking for one hour at 65 °C. Following this, the polyurethane mold was mounted in a metal frame and used as a master mold for the fabrication of PDMS substrates. Thermal Chromium Deposition. Following the fabrication of grooved surfaces on silicon wafers with photolithography, thermal deposition with a PVD system (ZWE Thin Film Laborator, MPI, Stuttgart, Germany) was used to coat the wafers with either 200 nm or 250 nm of chromium, dictating the height of the grooves. The chromium coated wafers were then heated for 2 hours at 65 °C to drive voltages in the chromium layer, then placed in a glass dish and covered with S1818G2 developer (MicroChem, USA) and sonicated for one minute, yielding a Si-chromium master mold for use with PDMS.
Substrate fabrication and characterization. The silicone-based elastomer polydimethylsiloxane (PDMS) (Sylgard 184, Dow Corning, USA) was used as a bulk material at a 10:1 base to crosslinker ratio for the fabrication of the substrates, yielding substrates with a Young's modulus of approximately 2 MPa. Degassed PDMS was poured into the appropriate molds, with control substrates fabricated against a flat mold, and allowed to polymerize overnight at 65 °C. PDMS substrates containing no features (control), grooves, and pillars were prepared and characterized by white light interferometry (Zygo NewView 5000, Middlefield, CT, USA) as shown (Fig. S4A). The grooves were found to have a width B of 4.01 μ m ± 30 nm, a center spacing A of 6.02 μ m ± 40 nm, and a depth H of 351 nm ± 5 nm (Fig. S4B). The average height of the pillars was 8.7 μ m ± 50 nm, the average Step t(s) p(mTorr) SF 6 (sccm) CHF 3 (sccm) P RF (W) P ICP (W) width B was 4.02 μ m ± 40 nm and the distance between the center positions of the pillars A was 10.03 μ m ± 50 nm (Fig. S4C), confirming that the dimensions of both the column substrates and the groove substrates match the dimensions of the patterned masks used during the photolithography process.

Statistics.
All experiments were performed in triplicate unless otherwise indicated. Error bars shown are the standard error of the mean (s.e.m.), except in Figs 3F and 4E, which display 95% confidence intervals. Significance for single comparisons was assessed by t-test at a significance threshold of p < 0.05 or lower as indicated. Significance for multiple comparisons was assessed with ANOVA and a post-hoc Tukey test at a significance threshold of p < 0.05 or lower as indicated. Values less than 0.1 were noted. For instances where data is not significantly different, N.S. is stated.