The geometry of photopolymerized topography influences neurite pathfinding by directing growth cone morphology and migration

Topographically micropatterned substrates were generated as previously described using photopolymerization (figure 1(a)) [19]. First, a monomer solution was formulated consisting of 40 wt% HMA (Aldrich), 59 wt% 1,6-hexanediol dimethacrylate (HDDMA, Aldrich), and 1 wt% of 2,2-dimethoxy-2-phenylacetophenone (DMPA, BASF). Then, to create the patterned substrates, this solution was evenly dispersed on a silane-coupled piece of cover glass by placing glass-chrome custom photomask (Applied Image Inc., figure 1(b)) on top. These samples were then exposed to 365 nm light at an intensity of 16 mW cm⁻² using a high-pressure mercury vapor arc lamp (Omnicure S1500, Lumen Dynamics, Ontario, Canada) to polymerize the monomer solution (figure 1(a)). The opaque portions of the mask direct exposure of the UV radiation thereby modulating the rate of the polymerization locally to generate features on the surface. This process creates raised features or ridges underneath transparent bands where UV light intensity and the polymerization rate are highest. These substrates direct the growth of various cells and neurons, including SGNs and DRGNs [43]. Additionally, various geometries of topographical substrates can be generated by varying the photomask [34] or time of UV light exposure to change feature amplitude [19]. The duration of UV light exposure was changed in this work to create three microfeature amplitudes/depths of 2 μm, 4 μm, and 8 μm with the custom photomask.

Zoom In Zoom Out Reset image size

To generate multiangle features and micropatterns, a novel photomask was used to generate the micropatterns utilized here (figure 1). This custom photomask consists of 6 different microfeatures, i.e. channels with gradually sloped edges, that are 20 μm in width arranged in a repeating hexagonal pattern. In addition to the control (straight) microchannel, the other five microfeatures consist of turns with varied angles (30°, 45°, 60°, 90°, and 120°) in which a neurite would need to deviate 20 μm perpendicular to the trajectory of the microfeature to navigate the angled turn (labeled in blue on supplementary figure 1(a)). For the microfeature nomenclature, we defined the microfeature in reference to the interior angle with 30° being the sharpest turn, 120° the most gradual, and 45°, 60°, and 90° representing intermediate turn challenges.

Prior work suggests that neurites can bypass the angle turn challenges in situations with zero or small overlapping distance (distance perpendicular to the direction of the microchannel in blue on supplementary figure 1(a)) in the zigzags [34]. Therefore, we chose to design each angle turn challenge to have an equivalent distance (20 μm) that the neurite would need to deviate perpendicular to the direction of the feature to navigate the turn. As such, the length of the straight portion of each microfeature's zigzag is varied in order to standardize this design constraint (supplementary figure 1(c)).

The dimensions and characteristics of these micropatterned substrates were assessed using three methods. First, the topography of every substrate used in the following experiments was measured using white light interferometry (Dektak Wyko 1100 Optical Profiling System, Veeco; supplementary figure 2(a)). Feature amplitude was characterized in five regions of the glass coverslip and neurons were only cultured on areas within ±10% of the target microfeature amplitude. Vision software (Bruker) was used to create 3D images of substrates using this measurement technique.

A second method to characterize the micropatterned surface was SEM (Hitachi S-3400N; supplementary figure 2(b)). Conductive tape was used to mount sample substrates to SEM stubs and each sample was sputter coated with carbon prior to microscopy examination. The specimen was imaged top–down to assess the polymer surface.

The third method to assess the micropattern feature dimensions was confocal microscopy (STELLARIS 8, Leica; figure 1(c) and supplementary figure 2(c))). This approach was possible since the polymer interacts with ultraviolet light, therefore, enabling it to be imaged via confocal microscopy using a 405 nm laser. By creating a z-stack, the topographical dimensions as well as the character of the substrate surface can be determined.

All procedures involving animals were conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals and were approved by the University of Iowa Institutional Animal Care and Use Committee. All mice were maintained on a C57BL/6 (Envigo) background, housed in groups on a standard 12:12 h light: dark cycle with food and water provided ad libitum, and used at p3-5 of age when SGN peripheral processes already contact hair cells in the organ of Corti. For live cell imaging, neurons were harvested from Pirt-GCaMP3 transgenic mice [44].

One day before culturing, substrates were sterilized by soaking in 70% ethanol for 2 min and then exposing them to UV light in a cell culture hood for 15 min. Cloning cylinders were placed on the patterned coverslip and then filled with poly-L-ornithine solution (Sigma-Aldrich) for 1 h at RT. After washing the substrate with sterile Milli-Q^® water, a laminin solution was added (20 μg ml⁻¹, Sigma-Aldrich) and incubated overnight at 4 °C. SGN cultures were prepared from mice between post-natal day 3 and 5 (p3-p5). Mice were decapitated, and their cochleae were isolated from their temporal bones in ice-cold PBS. The spiral ganglia were then isolated and placed into ice-cold HBSS(−/−) as previously described [45]. Enzymatic dissociation was then performed in calcium- and magnesium-free HBSS with 0.1% collagenase and 0.125% trypsin at 37 °C for 25 min. 100 μl of FBS was added to stop dissociation. Ganglia were washed with Neurobasal media before being placed in supplemented Neurobasal Medium (Thermo Fisher) containing: 5% FBS, 2% N2 Supplement (Thermo Fisher), 10 μg ml⁻¹ Insulin, 50 ng ml⁻¹ BDNF (R&D Systems), and 50 ng ml⁻¹ NT-3 (Sigma-Aldrich). Once in supplemented media, ganglia were triturated first with a 1000 μl pipette tip followed by a 200 μl tip. Cultures were plated onto micropatterned substrates and maintained in a humidified incubator with 6% CO₂ for 48 h.

DRGNs were isolated as previously described [46], however, using neonatal (p3-p5) mice. First, a 24-well polystyrene plate was coated with poly-L-ornithine solution (Sigma-Aldrich) for 1 h at RT. The surface was washed with sterile Milli-Q^® water three times prior to a laminin solution (20 μg ml⁻¹, Sigma-Aldrich) being added and incubated overnight at 4 °C. After warming the coated well plate for 1 h at 37 °C, the freshly dissected DRGNs were cultured on this plate for 72 h before being replated. The replating procedure has been described previously [47], but after a 1 min incubation with TrypLE Express (Thermo Fisher), warm media was used to gently triturate the culture surface and lift the adhered neurons. The resulting replated DRGNs (rDRGNs) were then cultured on the micropatterned or unpatterned HDDMA/HMA substrates in a humidified incubator with 6% CO₂ for 24 h.

After culture, media was removed and the culture surface was washed three times with phosphate-buffered saline without calcium and magnesium (PBS(−/−)). 4% paraformaldehyde in PBS(−/−) (Fisher Scientific) was added to fix the neurons. After 20 min at RT, cells were washed with PBS(−/−) three times, and then blocking buffer was added (5% normal goat serum [ThermoFisher], 0.2% Triton™ X-100 [Fisher Scientific], and 1% BSA [Research Products International] in PBS[−/−]). After incubating for 30 min at RT, different approaches were taken for the antibodies. For SGNs, mouse monoclonal anti-NF200 antibody (RRID: AB_260781, Millipore Sigma) was added (1:400 in blocking buffer) for 2 h at 37 °C. Following this incubation, the culture surface was washed 3 times with PBS (−/−) and then Goat anti-mouse Alexa Fluor^®546 (RRID: AB_2534089, ThermoFisher) in blocking buffer (1:800) was added for 1 h at RT in the dark.

For the rDRGNs, chicken polyclonal antibody to NF200 (RRID: AB_2313552, Aves) was added (1:800 in blocking buffer) for 1 h at RT. Following this incubation, the culture surface was washed three times with PBS (−/−) and then Goat anti-Chicken Alexa Fluor^®546 (RRID: AB_2534097, ThermoFisher) in blocking buffer (1:1000) was added for 1 h at RT in the dark. Following the secondary antibody incubation, both systems were treated the same; cultures were again washed with PBS (−/−) three times. Lastly, the polymer-coated cover glass was mounted onto a glass slide using Fluoromount-G^® (SouthernBiotech) and kept in the dark for 24 h before imaging.

Every SGN neurite that encountered a micropatterned, topographical microfeature was assessed in this analysis, in which neurite length in the microfeature was measured using NeuronJ [48]. Neurite length in microfeature was measured from where the neurite entered the microfeature to where it either exited the microfeature or the endpoint of the neurite was reached. These length data were then grouped both by turn angle and microfeature amplitude. For this approach, the unpatterned control condition was generated by overlaying the shape of the 60° microfeature over a flat portion of the substrate, and length data was measured in the same manner as neurites that encountered this pseudo-pattern.

Another method for assessing neurite pathfinding behavior was qualitatively scoring every neurite encounter with a microfeature. Neurites were scored as either turning or not turning. In particular, two conditions for which they were scored as not turning were: (1) did not follow the microfeature; neurite growth is not oriented in the direction of the pattern and the neurite crosses out of the microfeature nearly perpendicular to the microfeature's direction, and (2) the neurite follows the straight portion of the microfeature but fails to complete the turn challenge and leaves the microfeature. Two categories of neurites were deemed to have successfully turned in response to the microfeatures: (1) the neurite follows the microfeature and navigates one turn; the neurite enters the microfeature more than 30 μm before the turn challenge and remains positioned in the microfeature 30 μm past the vertex of the turning challenge, or (2) the neurite navigates multiple turns and/or the shaft is reoriented where it is aligned across multiple of the zigzagging microfeatures. Like the previous metric, these data were grouped both by turn angle and microfeature amplitude.

Neurons growing in response to these turns were imaged live. To do this, rDRGNs were cultured from mice expressing Pirt-GCaMP3, a genetically encoded calcium indicator to allow imaging with a standard GFP/FITC/488 filter. These neurons were cultured on substrates with 4 μm and 8 μm amplitude microfeatures for 18 h. At that timepoint, neurites that were currently growing in a microfeature and actively engaging with the ridge of the feature were selected randomly and imaged every 15 s for 1.5 h to capture the growth dynamics. These videos were assessed by tracking the growth cones that were in the microfeature at the beginning of the recording and determining if they remained in the microfeature throughout the 1.5 h of imaging.

To study the morphology and behavior of neuron growth cones in response to topographical growth cues, replated DRGNs (rDRGNs) were cultured on a different topographically patterned substrate consisting of repeating rows of ridges and grooves (10 μm periodicity and 3 μm amplitude). These micropatterns offered a consistent pattern and thus a systematic methodology for this analysis. The substrates were made in the same manner as described previously [19].

For this assessment, rDRGNs were cultured on these micropatterns and on unpatterned substrates, then immunofluorescently labeled as above; however, the secondary antibody step was modified to assess growth cone morphology. For the modified secondary antibody step, Goat anti-chicken Alexa Fluor^®488 (RRID: AB_2534096, ThermoFisher) (1:800) and Alexa Fluor™ 546 Phalloidin (ThermoFisher) (1:200) in blocking buffer was added for 1 h at RT. Fixed growth cones were imaged via confocal microscopy (STELLARIS 8, Leica). Images of the growth cones were analyzed by two approaches using Imaris software. First, the prolate ellipticity of the shape of the growth cone was computed (supplementary equation (1)). For this analysis, the software approximates the shape of the growth cone as an ellipsoid and derives the prolate ellipticity, which is a measure for how elongated the growth cone is. This shape was assessed since others have shown that the shape of the growth cone is influenced by substrate cues, thus the micropatterned substrate here may cause the shape to become elongated [49, 50]. Second, the 2D shape of the growth cone was approximated as an ellipse and its major axis was derived. The angle made between this major axis and the neurite shaft was measured (figure 5).

Since the neurons were encountering turn challenges of various microfeature amplitudes, the tendency of these neurons to keep their shaft in the microfeature around a turn was assessed. All the neurites that were deemed to have turned in the analysis above (section 2.8) were further studied and the proportion of these neurites that remained in the microfeature around the turn was determined.

For each microfeature turn angle, a mathematical definition of what characterizes a neurite holding its position around a turn was derived. For this definition, the greatest angle path a neurite could take from halfway down each straight segment and remain in the microfeatures was measured (supplementary figure 1(b)). This was found to be 134°, 138°, 142°, 151°, and 160° for the 30°, 45°, 60°, 90°, and 120° microfeature turn angles, respectively. ImageJ was used to measure the angle of the neurite's shaft from the point on the shaft at the microfeature turn's vertex to points on the shaft 30 μm on each side of the vertex (supplementary figure 1(d)). Neurites with angle of tension lower than that of the theoretical path of least tension for its given microfeature angle were characterized as holding tension, while those with angles greater were categorized as not holding tension. Neurite shafts that did not remain in the microfeature at the point of the vertex of the microfeature turn were also categorized as not holding tension. These data were grouped by microfeature amplitude.

In order to better visualize the neurites that held tension and/or reoriented, confocal microscopy was used to confirm the observations made by epifluorescence image analysis. The images were color coded by z-position to determine what plane the neurite shafts were positioned in these angled microfeatures.

Neurite branching was also assessed as a function of the angle at which the neurite encountered the edge of the microfeature. First, the angle at which the neurite encountered the ridge was measured using the angle tool in ImageJ (supplementary figures 3(a) and (b)). Then each neurite was scored as either follow, exit, or branch (if it branches at the microfeature boundary wherein one branch follows and the other exits). For this approach, only the distal-most neurite encounter with the micropattern for each neurite was assessed. Additionally, as part of this approach, the proportion of neurites that successfully turned was also measured as a function of the angle they approached the microfeature ridge and microfeature amplitude.

Using a customized photomask designed with varied angles (30°–120°), photopolymerization techniques were used to generate methacrylate surfaces with the specified micropatterning consisting of grooves in masked regions (figure 1). By altering the duration of light exposure, patterns were engineered with microfeatures of varied amplitude, or depth (2 μm, 4 μm, and 8 μm). Thus, the photomask dictates pattern geometry while light exposure determines microfeature amplitude. The micropatterned substrates were characterized using multiple methods. The first method is via white light interferometry (supplementary figure 2(a)). This method enabled measurement of microfeature amplitude quickly and accurately. The substrates were also imaged with SEM to confirm that the microfeature ridges consist of sloping transitions rather than the typical, abrupt on/off features characteristic of other engineered micropatterns from lithography, etching, etc. (supplementary figure 2(b)). Lastly, z-stacks created with confocal microscopy also allowed for the measurement of the amplitude of the microfeatures, visualization of the smooth nature of the surface, and additionally the creation of depth color-coded images to aid in the quantitative and qualitative assessment of the substrate (figure 1(c) and supplementary figure 2(c)).

The topographical multi-angled micropatterns enable the evaluation of neurite pathfinding in response to various turn challenges of precisely controlled geometry. In particular, SGN neurite pathfinding in response to six different microfeature turns (30°, 45°, 60°, 90°, 120°, and straight) was assessed across 3 different microfeature amplitudes (2 μm, 4 μm, and 8 μm). Importantly, these are modest and biologically relevant growth cues consisting of gradually sloping transitions (figure 1(c) and supplementary figure 2). To provide an overall assessment of how strongly a neurite follows these various micropatterns, we measured the distance that each neurite remained in the microfeature channel and compared the data with two-way ANOVA. The analysis of neurite behavior on these substrates shows that the length of a neurite that remains in the microfeature increases both for more gradual turn challenges (straighter or larger angle microfeatures) as well as deeper microfeatures (figure 2). In particular, neurite length in a microfeature increases by an average of 20% from the 2 μm (42.8 μm) to 8 μm amplitude (51.6 μm). Thus, these results suggest that modest increases in channel depth enable neurites to be guided by the cues more efficiently. Additionally, the length also increases by an average of 23% from the sharpest turn to most gradual (42.7 μm for 30° to 52.4 μm for 120°). This implies that the neurites also exhibit greater fidelity in navigating more gradual turns.

Zoom In Zoom Out Reset image size

This conclusion is further evidenced by similar findings when the neurite behavior at each type of turn challenge is scored as noted in section 2.8. The proportion of neurites that successfully navigate the turns increases with microfeature amplitude increasing when assessed with multinomial logistic regression and chi-square (figure 3). On average, the percent of neurites that navigated the turns almost doubles from 18% making the turns on average in the 2 μm amplitude microfeatures up to 31% in the 8 μm amplitude microfeatures, a 72% increase in fidelity. In terms of microfeature angle, we see neurites are best able to turn in response to the moderate angle microfeatures (45°, 60°, 90°) (figure 3 and supplementary table 1). This second analysis does not perfectly isolate the variable of turn angle since the straight portion of the microfeature varies in length for each turn (supplementary figure 1(c)). Thus, the conclusions of the regression for this assessment are inconclusive and instead these data suggest that it is more challenging for a neurite to navigate a longer straight portion and more gradual turn (120° microfeature) than a shorter straight portion despite a sharper turn (45°, 60°, and 90° microfeatures). Lastly, we also see that regenerating neurites from another sensory neuron, rDRGNs, also successfully navigate these microfeatures in a geometry-dependent manner (supplementary figure 4).

Zoom In Zoom Out Reset image size

To better understand how neurites grow in response to these cues, replated DRGNs with a genetically encoded fluorescent indicator, Pirt-GCaMP3, were imaged live as they extended their neurites in the patterned angled microfeatures. In the videos, the growth cones were highly dynamic and strongly responded to the microfeature ridges in the 8 μm amplitude condition (supplementary video 1). Based on the high fidelity with which the growth cones appeared to navigate the 8 μm amplitude microfeatures, we compared the proportion of the growth cones that remained in the microfeature throughout the 1.5 h videos in the 4 μm and 8 μm amplitude microfeatures. Due to the low throughput of this analysis and the requirement that the growth cone of interest be actively engaging with a microfeature ridge at the beginning of recording, the growth cones analyzed were drawn from a random sample of turn angles and the growth cone positions within the microfeature (supplementary table 2). The proportion of growth cones that remained in the microfeature was much greater in the 8 μm amplitude microfeatures (95% ± 5%) compared to the 4 μm (55% ± 15%) (figure 4). In this analysis, we see the fidelity increase from about 50% guidance in the 4 μm amplitude to nearly perfect guidance in the 8 μm amplitude system. While the observed effect is robust, it is important to note that the duration of imaging (1.5 h) is shorter than the culture duration of the other studies (24 or 48 h).

Zoom In Zoom Out Reset image size

Given the stark difference in rDRGN growth cone pathfinding fidelity in different microfeature amplitudes in figure 4, we further assessed growth cone behavior on micropatterned substrates by growing rDRGNs on a topographical substrate consisting of narrow repeating rows of ridges and grooves (10 μm periodicity and 3 μm amplitude, figure 5(a)). We sought to determine if and how the growth cone morphology and orientation are altered when grown on topographically patterned substrates to inform how the strong differences in pathfinding in figure 4 manifests. When cultured on this micropatterned substrate, the ellipse shape of the rDRGN growth cones become more elongated and prolate than on unpatterned substrates, with prolate ellipticity values of 0.431 (±0.037) and 0.274 (±0.043), respectively (figure 5(d) and supplementary equation (1)). Additionally, the angle difference between the major axis of the growth cone and the neurite shaft was smaller in the neurons grown on the micropatterned substrate, a median difference of 5° compared to 20° on the unpatterned substrate (figure 5(e)). In combination with figure 4, these observations suggest both the importance of the growth cone in transducing the biophysical cues for proper pathfinding and that the growth cone can function as an assessment marker of pathfinding.

Zoom In Zoom Out Reset image size

After appreciating the significant role of the growth cone in figures 4 and 5, we next studied the behavior of the neurite shaft when growth cones are pathfinding in response to angled turn challenges. We observed that a significant subset of neurites aligned across, as opposed to with, the zig-zagged microfeatures in the fixed sample studies (figure 3(a), supplementary figure 5). To dynamically assess the behavior leading to this observation, live imaging of neurite shafts was conducted. The live imaging demonstrates that the neurite shafts are constantly reorienting their position as the neurite is elongating. The neurite shafts appear to be mobile and drift as the growth cone pathfinds (supplementary video 2). In particular, the neurites tend to reposition their shaft such that it becomes aligned in the direction of elongation. This movement thereby minimizes the number of neurites that hold their position in a zigzag or hold their tension around a turn. This behavior occurs since the neurite shaft moves to reposition itself outside of the microfeature and becomes more aligned in the direction of neurite elongation (supplementary video 2). Additionally, SEM images of these rDRGNs display neurites tightly following the microfeature ridge or just hopping over the vertex of the angle challenge of the microfeature (supplementary figure 5).

The neurons that successfully turned in figure 3 were again assessed to determine if the proportion of neurites reorienting around the turns was related to microfeature geometry (supplementary table 3). In this analysis, there is no significant effect of feature amplitude on neurite shafts remaining in the microfeatures when making turns. Though a higher proportion of neurites remain in the microfeature when turning in the 8 μm amplitude condition, this difference is not statistically significant via Chi square test since we are comparing a proportion in small subsets of the data and the magnitude of the difference is not large (31% vs. 46%, respectively) (figure 6). Additionally, no significant relationship was found between the angle of the turn and the ability of a neurite to hold tension when making turns. Despite the considerable reorientation exhibited by the neurites, this behavior appears to be a fundamental property of the neuron that is not affected by the modest changes to the microfeature amplitudes studied here.

Zoom In Zoom Out Reset image size

During the live imaging of the neurite sharfs, the growth cones were observed to often branch when the growth cone encountered a ridge or turn challenges on the angled microfeatures. Additionally, the growth cones were observed to have a branched, arborizing appearance as they navigated the microfeatures (supplementary video 2). To quantitatively assess neurite branching, we scored neurite behavior (follow, exit, or branch) as a function of microfeature amplitude and the angle at which the neurite encountered a topographical ridge from within the angled microfeatures (supplementary figures 3(a) and (b)). This modified approach was used since the exact angle at which a growth cone encounters the feature ridge varies within each microfeature condition. Firstly, the results of this modified quantification echo the findings of figures 2 and 3 showing that the proportion of neurites successfully turning increases with greater microfeature amplitude and with more obtuse angles (supplementary figure 3(c)). The neurites are observed to better follow the microfeatures by a magnitude of at least 3.5-fold when they approach the microfeature ridge at a less severe angle. Additionally, neurites are seen to better follow as microfeature amplitude increases from 2 μm to 8 μm by an average of 2.8-times (supplementary figure 3(c)).

When studying the relationship between neurite turning and branching using this modified assessment, we see neurite branching occurs most frequently when the probability of neurite turning approximated 50% (figure 7). Branching is observed to be less likely to occur when the neurite turning outcome becomes more certain, i.e. when approaching no chance of turning or all neurites making the turn. The microfeature angle where branching was maximized varied across microfeature amplitudes. For the 2 μm microfeatures, maximal branching occurs in gradual turns (15°–30°), 4 μm in moderate turns (30°–45°), while sharper turn challenges maximized branching in the 8 μm amplitude microfeature (75°–90°).

Zoom In Zoom Out Reset image size

Herein, we have demonstrated a novel system to assess the ability of neurites to turn in response to topographical cues of varied angles and amplitudes (depths). Using engineered micropatterned substrates, we see that the geometry of the angle turn challenge, i.e. microfeature amplitude and angle of turn, governs the ability of the sensory neurites to navigate the turn and follow the micropattern (figures 2 and 3). This conclusion appears straightforward; however, we see a wide variation in how neurites respond to the same geometric cue (figure 2). Furthermore, neurites can follow or not follow any microfeature studied implying, that in these patterned microfeatures, there is no distinct upper or lower limit of a neurite's ability to navigate a turn (figure 3). Therefore, while the geometry of the angled features influences how well sensory neurites track the microfeature guidance cues (figure 2) and navigate the turns (figure 3), the process by which these neurites navigate these turns is not simply causal. These data suggest that as SGN neurites navigate complex turn challenges, their behavior reflects a balance of deterministic cues guiding their growth and their innate stochastic outgrowth patterns. Therefore, the strength of the guidance cue does not perfectly establish growth cone behavior, but rather shifts the probability distribution of growth cone turning behavior.

This random nature of growth cone motility has been described previously [56–58]. Neurite pathfinding results from an interplay between the strength of the turn cue [17, 19], stochastic actin polymerization [58], and signaling from the support cells from the primary culture influencing guidance cues [54]. Within the experimental system used here, these factors result in non-uniform pathfinding observations within each geometric condition, e.g. neurites do not navigate the gradual turns in the 8 μm amplitude with perfect fidelity. It is important to note, that sufficiently strong cues can overpower the stochastic nature of the neurite pathfinding to create nearly deterministic systems [19, 59]. While the biophysical cues utilized here do not overpower the innate randomness, we show that these topographical cues, of biologically relevant dimensions, can strongly shift the probability distribution of neurite turning behavior. Thus, this is an appealing model to study turning and guidance since it reflects the balance of both directed and random behavior inherent to neurite guidance in vivo [60, 61]. Such a model of neurite turning in response to topography can be applied to study fundamental signaling mechanisms that contribute to neurite guidance and translation to in vivo applications.

To better understand the neurite pathfinding process, live imaging of rDRGN neurites navigating the microfeatures was conducted (supplementary video 1 and 2). In these videos, the growth cones showed a strong tendency to closely track the microfeature ridge for the 8 μm amplitude microfeatures (95%), while in the shallower 4 μm microfeature, the fidelity to the microfeature was less consistent (55%) (figure 4). These data suggest two main findings. First, since the growth cones consistently remain in deep microfeature (8 μm amplitude), strong, biologically relevant topographical cues can be engineered that precisely direct neurite guidance and turning. Second, based on the degree of the difference observed in figure 4, it appears that growth cone behavior accounts for the differences in pathfinding that is observed in the prior analyses, figures 2, 3 and supplementary figure 4.

After observing differential rDRGN growth cone behavior to the various angled microfeature geometries, we further assessed how growth cone morphology and behavior change using a topographical substrate with narrow repeating rows of ridges and grooves. This system offered a systematic methodology and further support for growth cone behavior being involved in differential pathfinding to topographical cues. Importantly the growth cone is more prolate when grown in the microfeatures (figure 5(d)) as well as the growth cone appears elongated to become strongly oriented in the direction of the microfeatures and neurite shaft (figure 5(e)). The alignment of the growth cone and shaft suggests that changes in growth cone morphology and behavior underlie the pathfinding observations downstream. In particular, by integrating these findings with previous work, it appears that the topographically patterned substrate and growth cone interact to cause the growth cone to become elongated in the direction of the microfeatures. This prolate shape drives growth in the direction of the major axis of the ellipsoid growth cone which is in line with the ridges and grooves. The neurite shaft then orients in the same trajectory since the growth cone is uniformly driving growth in that direction.

It is evident that topographical features studied here change rDRGN growth cone behavior and that the growth cones have differential responses across topographical feature amplitudes, consistent with other studies [62, 63]. However, it is important to note that when imaged live the neurite shafts appear tremendously dynamic and reorient around the turns as the growth cone traverses the environment (supplementary video 2). The neurite shafts reorienting their position may obscure the inferences of our work since how a shaft is oriented at a given timepoint may not be representative of the path a growth cone took to create that neurite. Importantly, the conclusion that microfeature geometry determines neurite turning remains sound since even with the chaotic reorientation occurring, we still see clear and significant differences in our comparisons of interest (figures 2, 3 and supplementary figure 3). That being said, the reorientation of the neurite shafts is an important finding that motivates a reinterpretation of prior studies using fixed neurons to study pathfinding in response to various cues [17, 34]. Our data suggest that the growth cone may be consistently navigating a microfeature, but the shaft may reorient bringing it out of the microfeature. Therefore, tracing fixed neurites may not represent the true trajectory that the growth cone traversed in its elongation.

Beyond influencing neurite pathfinding, we also evaluated the relationship between the tendency of a neurite to branch and the ability of neurites to navigate a turn on these micropatterned surfaces [31, 64, 65]. We observed that branching occurred most often when the probability of neurites successfully making a given turn was essentially a tossup, near 50% (figure 7). When interpreted in combination with the live imaging data, these data suggest that the dynamic growth cones initiate branching at decision points on the micropatterned substrate. When the two paths offer equal resistance, i.e. making a turn or exiting out of the microfeature, both neurite branches are more likely to persist [66]. By contrast, on micropatterns with stronger guidance cues and greater certainty of growth behavior, the neurite either never initiates a branch, e.g. the growth cone remains in the microfeature for a gradual turn in a high amplitude feature, or ends up retracting the branch in the less favorable path, e.g. growth cone easily hops out a shallow microfeature at a sharp turn.

Though our work does not directly address the causes of branching, it complements other work that informs the process. First, others have indicated that mechanical force plays a significant role in neurite branching, in that neurons retract a growth cone branch if sufficient mechanical stress is applied [67]. In this context, mechanical stress on the growth cone from either the topographical microfeature ridge or turn likely initiates growth cone branching or dictates which arbor retracts. Additionally, topography has been implicated in branching frequency where topographical cues were presented to promote neurite branching [68]. The ability of specific cues to promote branching is consistent with the findings here that branching does not occur randomly, but can be stimulated to an extent by precise cues presenting a challenge with two equally likely directions of growth simultaneously. Overall, this angled microfeature system, or similarly constructed cues, offer a useful model to study neurite branching decisions.

This work is one of the first to use a precisely engineered topographically patterned substrate to systematically study neurite turning to angled and tunable microfeatures (figure 1), thus there are a few limitations of this initial work. First, neurons used in this study are derived from primary cell cultures, and cells on the micropatterned substrate are a heterogeneous mixture of the various cell types present in the ganglia. Therefore, the topographical growth cues inherent to the substrate are not the only guidance cues the neurites encounter in this system, as the glial cells may affect neurite guidance as well [54]. This tradeoff is necessary in order to study the pathfinding behavior of primary neurons with their full expression of receptors and channels. Second, all data are from neurons derived from neonatal animals. Neonatal SGNs are more dynamic compared to adult SGNs, the target cells of patients receiving CIs. This shortcoming is necessary as the yield and growth of SGNs from adult mice is limited. Likewise, growth cone size and dynamics are enhanced in neonatal DRGNs. Thus, using neonatal DRGNs for the live imaging and growth cone studies is necessary. That being said, adult SGNs and DRGNs have been shown to respond to topographical growth cues as well so the results are expected to translate [43]. A third potential shortcoming of the experimental system is the proximity of the angled turn cues to one another. While our system does not have simple, repeating cues arranged directly next to each other as in many other studies [30, 34], the cues are near each other and could impact behavior of the neurites. Importantly, our findings suggest that this impact is likely minimal. This is due to two observations. First, we use a variety of different observations to draw our conclusions, so it is unlikely that the neighboring channels influence the separate observations in the same manner across all experiments. In particular, these include observations of real time, local interactions of the growth cone with the channel ridge (figure 4), which should be unaffected by adjacent channels. Another limitation of this work was the duration of real time studies. While the real time observations were only for 1.5 h, they were of adequate duration for the stated goals. In particular, we sought to observe the behavior differences in the growth cones in real time, so neurons were imaged continuously and robust differences between moderate feature geometries were observed. Future work would explore methods to increase the duration of imaging including more resilient model neuronal culture systems, which approximate primary cells [69, 70] or non-continuous imaging to minimize high intensity light exposure. Lastly, this work did not assess neuronal function after the neurites navigated these biophysically patterned turn challenges. Follow up work using this system ought to further assess neuron functionality by performing similar work as others have done with in vitro studies by guiding neurite growth toward an electrode array and assessing the excitability and functionality of the neurite after being guided around various angled turns [71, 72].

To conclude, it is important to discuss how these findings inform the next steps for the field such as translating these findings to in vivo applications as well as expanding the understanding of fundamental mechanisms underlying neurite pathfinding. The findings from this work demonstrate fundamental principles that will be essential for translational endeavors in vivo, as the microfeatures in our work have sloped edges, so when growth cones grow within a microfeature, growth mechanics will mimic closed tubes with similar curvature [19]. A first finding that will inform in vivo application is that based on the decreased ability of neurites to pathfind to sharper angle cues, turns should be engineered to be gradual and to minimize zig-zagging as neurite shafts appear to resist this reorientation. For example, this may favor a lateral wall placed CI electrode where the neurite would need a gradual turn from the organ of Corti. Second, since the growth cones change their morphology and migration behavior to drive pathfinding in response to these microfeatures, translational biomaterials may be better suited to have aligned pores to guide growth. This paradigm has been proposed to be advantageous for neural conduits when compared to a biomaterial with a random meshwork [36, 37]. Lastly, minimizing neurite branching at the electrode interface will be necessary to optimize signal resolution in next generation CI. Our work suggests that neurite branching occurs as a function of neurite growth uncertainty when presented with multiple paths of growth. Therefore, in future material designs, instructive growth guides should be of sufficient strength to minimize exuberant branching. This may be challenging to accomplish since branching may occur at positions with unclear cues, such as when entering and exiting the designed biomaterial. Thus, as with pathfinding, neurite branching must also be considered, and minimized, for optimal recapitulation of the architecture and function of native neural system where growth is being induced.