Nanostructures to Engineer 3D Neural‐Interfaces: Directing Axonal Navigation toward Successful Bridging of Spinal Segments

Neural interfaces are the core of prosthetic devices, such as implantable stimulating electrodes or brain–machine interfaces, and are increasingly designed for assisting rehabilitation and for promoting neural plasticity. Thus, beyond the classical neuroprosthetic concept of stimulating and/or recording devices, modern technology is pursuing toward ideal bio/electrode interfaces with improved adaptability to the brain tissue. Advances in material research are crucial in these efforts and new developments are drawing from engineering and neural interface technologies. Here, a microporous, self‐standing, 3D interface made of polydimethylsiloxane (PDMS) implemented at the interfacing surfaces with novel conductive nanotopographies (carbon nanotubes) is exploited. The scaffold porosity is characterized by 3D X‐ray microtomography. These structures are used to interface axons regenerated from cultured spinal explants and it is shown that engineering PDMS 3D interfaces with carbon nanotubes effectively changes the efficacy of regenerating fibers to target and reconnect segregated explant pairs. An improved electrophysiological performance is shown when the spinal tissue is interfaced to PDMS enriched by carbon nanotubes that may favor the use of our substrates as regenerative interfaces. The materials are implanted in the rat brain and a limited tissue reaction surrounding the implants at 2, 4, and 8 weeks from surgery is reported.


Introduction
Neural prostheses are artificial, implantable devices designed to restore functions that are lost in injured or diseased cen tral nervous system (CNS). In manufacturing CNS implants, the tissue-electrode interface represents a crucial component potentially aid in cell maturation and in the formation of com plex, functional neuronal networks. [10,11] A polymer material incorporated with carbon nanotubes is, therefore, likely much more advantageous when attempting to build an interactive interface to improve axonal growth and guide synapse forma tion and function.
Organotypic spinal slice cultures represent a complex in vitro model where both sensorymotor cytoarchitecture and electrical properties are retained in a 3Dfashion. [10,14,15] We recently used such a model to test the ability of pure carbon nanofiber 3D interfaces in promoting successful reconnection of separated spinal explants in vitro. [16] Here, we exploit cul tured slices used as pairs to test 3D PDMS interfaces with or without carbon nanotubes. Previous studies have shown that spinal slices separated at distances more than 300 µm fail to reconnect under basal conditions, [7,[15][16][17][18] despite the huge out growth of nerve fibers, [10,14,16] allowing for the investigation of potential reconnection in the presence of efficacious cues. In this study, we investigate by electrophysiology, immunofluores cence, and confocal microscopy, organotypic slices interfaced to 3D scaffolds made of polymeralone or polymer and nanotubes. In particular, we address the navigation within the artificial structures of neuronal fibers exiting the slices and we correlate neurite distribution at the interface with functional slicetoslice reconnection.
We also investigate the biocompatibility of the PDMS con structs in vivo upon implantation into adult rat visual cortex [16] observed at 2nd, 4th, and 8th week postsurgery timepoints. We use immunofluorescence labeling to quantify the density of inflammatory responsemediating cells surrounding the implant material and of neuronal markers to assess the level of cellular infiltration into the scaffold in situ.
Our data indicate that the incorporation of carbon nano tubes into a 3D polymer construct confers increased neuronal activity in the spinal networks developed at the interface and guides regrowing axons toward functional reconnection of sep arated spinal explants. We also suggest that PDMS with carbon nanotube materials elicit minimal immune response following implantation into the CNS.

Scaffolds Used to Interface Organotypic Slice Cultures: A Microtomography Analysis
Two different, microporous, freestanding, elastomeric scaf folds, able to sustain the development of organotypic slices, were developed following the procedure described in our previous work. [7] PDMS scaffolds are the negative replica of a generating sugar framework; this allows for fabrication of 3D spongelike structures characterized by irregular porosity, with random channels connecting the pores. Within the same fabrication procedure, we generate micrometric cavi ties by a sugar mold previously mixed to multiwalled carbon nanotubes [7] (MWCNTs, Figure 1). This permits the formation of 3D sponges with the pore surfaces layered by a MWCNT carpet, stably trapped in the PDMS matrix. [7] Scanning electron microscopy (SEM) analysis of the PDMS scaffolds and of the PDMS decorated by the MWCNTs (PDMS+CNT) reveals similar "spongelike" morphologies ( Figure 1a,b), with the PDMS+CNT showing the presence of a layer of MWCNTs lining the inner facets of the pores (Figure 1b, inset).
3D Xray microtomography (µCT) reconstructions allow a comparison of the inner pore and channel morphologies of the two materials ( Figure 1c). These measures show that both scaffolds display similar hollow features. Regarding the general porosity, PDMS scaffolds (n = 3) contain 50% ± 11% empty volume and similarly, in PDMS+CNT scaffolds (n = 3), the empty volume is 54% ± 6% (Figure 1d). Interestingly, the isotropic index is 0.89 ± 0.03 and 0.91 ± 0.02 for PDMS and PDMS+CNT samples, respectively. This reveals that pore shape and channel directions are both isotropic traits of the manu factured scaffolds (Figure 1d). The diameter of the inscribed sphere (mean ± SD) in PDMS samples is 29 ± 8 µm, with the interconnecting channel mean size of 16 ± 1 µm, while in PDMS+CNT scaffolds the mean inscribed sphere diameter is 23 ± 2 µm, with 14 ± 1 µm channel size (Figure 1e). These values were not significantly different (P = 0.25 and P = 0.14 for pores and throats, respectively; Figure 1e).
After this characterization, PDMS and PDMS+CNT sub strates were used over extended periods to interface the devel opment of neural networks in longterm, cocultured pairs of mouse organotypic spinal cord explants. [10,16]

Spinal Explants Interfaced to the PDMS and PDMS+CNT Constructs: Axonal Colonization of the Interfaces in the Third Dimension
Spinal organotypic slices after >10 d of in vitro development exhibit a characteristic outgrowth of nerve fibers [10,16] and a rich neurite network observed in PDMS and PDMS+CNT condi tions (Figure 2a,b). Pairs of slices, when interfaced to PDMS or to PDMS+CNT, sprout neuronal processes that rapidly invade the 3D structure of the interfaces. In this set of experiments, we explore the infiltration of fibers within the sponges [10,16] and we investigate their patterns of growth by confocal microscopy and comparing the regions of interest (ROI: 640 × 640 µm; 20 and 28 ROI, n = 5 and n = 9 cultures, PDMS and PDMS+CNT, respectively) sampled from the area surrounding the slice explants. [16] To determine if there were any differences between the PDMS and PDMS+CNT substrates' ability to guide neuronal process distribution, we acquired zstacks to reconstruct the sampled ROIs (Figure 2a,b, right panels) and measured the thickness through which the βtubulin IIIpositive neurites were detectable within the interface materials.
Image stacks demonstrate that neurites are distributed in complex network formations across the surface of the sub strates, reaching varying depths into the manufactured inter faces (Figure 2a,b, right panels). Interestingly, we found that the depth through which neurites are traveling in the PDMS substrates (see Figure 2a,b, bottom insets to right panels) is sig nificantly greater than the neurite distribution in PDMS+CNT substrates (105.5 ± 8.6 and 52.7 ± 3.9 µm, respectively, P < 0.001; Figure 2c), despite the absence of any significant dif ference in substrate pore/channel size (Figure 1e), excluding that a mere variability (in the micrometer scale) of the PDMS and PDMS+CNTinterfaced skeleton influenced the permeation of growing neurites.
To further explore the differences in fiber outgrowth and morphology within the 3D interfaces, we also quantified the overall neurite density and the degree of dispersion. [16] In fact, depending on the substrate that supports the elongation of neu ronal projections exiting the interfaced spinal explant, neural fibers can be organized in thick bundles of aligned processes [10] or in complex webs of randomly oriented fibers. [16] First, we estimated neuronal fiber density. [16] By this pro cedure, we detect no differences in the amount of βtubulin IIIpositive area between PDMS (28.05 ± 2.89% positive area, n = 29 ROIs acquired from n = 10 cultures) and PDMS+CNT (26.77 ± 3.53% positive area, n = 20 ROIs acquired from n = 6 cultures) samples, suggesting that the amount of βtubulin III positive fibers outgrowing from the slices is similar (Figure 2d), although differently scattered in the third dimension (Figure 2c).
In a recent work, [16] we showed that pure carbon nanotube 3D scaffolds, when interfaced to cultured spinal slices, shape the neuronal processes extending from organotypic slices into a meshlike structure, compared to bundled fasciculations that occur when slices are cultured interfaced to fibrin glue layered on 2D glass coverslips. [16] We perform a similar analysis here to further investigate if this reduction in fiber orientation is tied more closely to the porous nature of the substrate or directly to the presence of carbon nanotubes. When fiber directionality is assessed in terms of degree of angular dispersion, [16] we find no differences in the mean dispersion angle values of the two samples (average degree of dispersion: 17° ± 2° for PDMS, n = 28 ROIs; 16° ± 3° for PDMS+CNT, n = 20 ROIs; Figure 2e). The relatively high degree of dispersion (>15°) in both sam ples confirms our previous experiments [16] by indicating that, once interfaced to 3D structures, the majority of fibers are not oriented in "bundlelike" assemblies but, instead, they tend to form a random network. [16] To summarize, both constructs support and attract neuronal process growth, as shown by the comparable βtubulin III positive area measured, but the presence of MWCNTs on the porous surface seems to limit, maybe by adhesion processes, the growth of neural fibers deep within the third dimension of the structure. Both porous constructs also favor the formation of intricate networks of variably oriented fibers, regardless the presence of MWCNTs, characterized by fewer thick, aligned bundles when compared to traditional 2D interfaces. [10,16]

Engineering PDMS Interfaces by Carbon Nanotubes Affects the Functional Rewiring of Segregated Spinal Explants
After characterizing the scaffolds' structure and the mor phology of the neurite network infiltrating the structures upon interfacing spinal slices for >1 week, our primary interest is to investigate the in vitro effects of PDMS and PDMS+CNT inter faces on spinal network activity and their ability to promote functional reconnections between slice pairs. Cultured spinal explants display prominent spontaneous elec trical activity and wellcharacterized motor (ventral) outputs. [14,16] We monitored extracellular potentials (local field potentials, LFPs) by placing electrodes in the ventral regions of each slice (within 20-100 µm of the ventral fissure; sketched in Figure 3a).
To examine the functional impact of interfacing slices with the two different materials, we use the glycine/GABA A receptor antagonists strychnine (1 × 10 −6 m) and bicuculline (20 × 10 −6 m) to weaken synaptic inhibition in PDMS and PDMS+CNT slice pairs (n = 13 and n = 16 cultures, respectively). These antago nists of inhibitory synapses are known to shift electrical motor outputs from random bursting to synchrony, as previously reported for the entire spinal cord [19] and for organotypic spinal slices, [20,21] leading, in all cultures tested, to the emergence of a slowpace bursting (see sample tracings in Figure 3b).
A characterizing feature of spinal disinhibited rhythm is the frequency of bursting episodes that we quantified by measuring the interevent intervals (IEI). A direct comparison of IEI meas ured in the PDMS or PDMS+CNT constructs, reveals that the frequency distribution of IEI values is significantly shifted to the left in spinal slices cultured in the presence of carbon nano tubes (P < 0.001; Figure 3c), the average IEI values being 7.0 ± 0.6 s in PDMS+CNT (n = 28 slices) and 11.2 ± 1.1 s in PDMS (n = 23 slices). Such an increase in rhythmic burst activity is reminiscent of the improved excitability usually detected when interfacing spinal slices to MWCNT carpets. [10] We can specu late that the MWCNTs exposed on the PDMS surfaces alter synaptic connectivity and/or firing activity in cultured slices, [10] possibly at the layer of neurons in direct contact to MWCNTs, affecting the excitability of the spinal networks. [10] However, we cannot exclude that the increased bursting activity is also reflecting an improved functional coupling among pairs of slices in the presence of carbon nanotubes. That is, when interfaced to PDMS+CNT, neuronal processes have more prob ability to interweave within the construct and to synapse to the cocultured spinal tissue, compared to those interfaced to PDMS alone. To clarify this issue, we further studied the functional connectivity between the two slices assessing, by crosscorrela tion analysis of the disinhibited bursting, the presence of syn chronous ventral outputs. [15,16] Ventral LFP recordings of disinhibited bursts were taken simultaneously from cocultured slices in PDMS or in PDMS+CNT (see Figure 3a for the experimental setup and Figure 4a for an example of the recordings). We use a MATLAB algorithm [16] to calculate the Pearson crosscorrelation coeffi cient (CCF) between the two cocultured slices and to determine   )). Scale bars: 50 µm. c) Analysis of the depth through which β-tubulin III-positive processes penetrate the scaffolds. d) No differences were observed in β-tubulin III processes when evaluated as a percent of image area. e) The degree of directionality dispersion, a measure of neuronal fiber orientation, indicated no differences between processes developed on PDMS or PDMS+CNT supports.
whether the CCF values obtained are significantly larger than that expected by chance, thus supporting the presence of functional contacts between the explants. An example of this analysis, performed for the recordings shown in Figure 4a, is shown by the plots of Figure 4b, where the distribution of CCFs obtained by randomly sampling the two time series in each experiment [16] is compared to the average CCF computed sampling consecutive time windows (the red line in the plots of Figure 4b). In PDMS+CNT, 81% of the cocultured explants are correlated, displaying CCF values significantly larger than that expected by chance; on the contrary, in PDMS, correlated slices are detected only in 37.5% of samples (a significant dif ference; P < 0.05, Figure 4c). These results suggest that there are increased functional connections between the spinal slices interfaced with CNTcontaining substrates, compared with those interfaced by PDMS alone. The low probability of recon nection in the PDMS interfaces is not surprising and is in line with our preliminary findings: [16] slices cultured sidebyside in pairs at distances known to inhibit functional reconnection in basal conditions [15,16] reconnect with a low probability even in the presence of 3D PDMS interfaces. [16] In our experiments, the average measured distance between slices is 1.9 ± 0.1 mm for PDMS (n = 7) and 1.6 ± 0.1 mm for PDMS+CNT (n = 15). The emergence of functional coupling is unrelated to the interslice distance ( Figure 4d; note that filled symbols indicate functional correlated pairs measured as described above and in Figure 4b; r PDMS+CNT = 0.30 ± 0.30, n = 15; r PDMS = 0.22 ± 0-29, n = 7), as well as to the age in culture (sampled from 10 to 23 d; Figure 4e r PDMS+CNT = 0.20 ± 0.29, n = 16; r PDMS = −0.08 ± 0.29, n = 8).
Engineering PDMS 3D interfaces with MWCNTs effec tively changes the efficacy of regenerating fibers to target and reconnect segregated explants. We also show an improved electrophysiological performance that may favor the use of our substrates as regenerative interfaces.
In our previous study, we reported that when interfacing pairs of spinal explants to scaffolds, the improved directionality dispersion and the third dimensional outgrowth of neurites are needed to improve the fraction of reconnected pairs, regardless the amount of βtubulin IIIpositive area. [16] However, 3D con structs, such as the PDMS interface used here, were not able to replicate the ability of pure 3D MWCNTs sponges [16] in recon necting separated spinal slices by dispersed webs of axons, leaving unresolved what properties of the latter were making the difference: the tubular morphology and elastic properties of freestanding MWCNTs skeleton [16] or the physical and chem ical properties of the MWCNT material in the third dimen sion? [12,16] Here we compare, for the first time, two PDMS based interfaces exhibiting highly similar scaffold structures but for the presence of a layer of MWCNTs. We used PDMS elastomer since this material is a standard in microfabrication or microfluidics developments in biological applications, due to the PDMS simple manufacturing together with its gas per meability, optical transparency, flexibility. [22][23][24] In addition, this elastomer has been recently engineered in 2D devices showing excellent mechanical flexibility and leading to soft implants integrated in the subdural space. [25] PDMS has also been filled with MWCNTs using microcontact printing and casting mold techniques [26] to provide conductive PDMS patterns, with potential applications in biology. Here, by entrapment in the PDMS surface of nanofeatures such as carbon nanotubes, we have engineered an interfacing grid with the ability to direct the formation of reconnecting webs of neurites. The inter actions among the geometry, chemistry, and the numerous physical and biological factors translating the exposure to  MWCNTs into axonal regrowth and synapse formation are unknown. Interestingly, we found that neurites penetrated sig nificantly greater depths into PDMS substrates compared with the PDMS+CNT substrates. We speculate that the presence of MWCNTs improves neurites mechanical tight associations with the interface, whereas in the case of PDMS, neurites tend to drop down into the pores following gravity, without associ ating with the material. That the neuriteMWCNT association is the rationale for the neuronal processes remaining within the first layers of the scaffold is supported by previous works which show that MWCNTs form extremely tight contacts with biological membranes. [8,10] We exclude large differences in the scaffolds' mechanical features, since in both PDMS and PDMS+CNT elasticity is due to the porosity of the structure. [7] In our previous study, we reported for similar porous scaffolds a Young's modulus within the range of that estimated in rodent and human brains. [7] Our in vitro results show that both scaf folds are of a mechanical stiffness conducive to neural tissue growth and sustainment.
We suggest the use of 3D PDMS scaffolds incorporated with MWCNTs as a potential candidate for manufacturing regener ating interfaces. The increasing application of nanomaterial related technologies for engineering brain interfaces may thus ameliorate some intrinsic features of interfacing devices (such as the stability or the charge transfer to and from neurons), [27] as well as their ability to facilitate recovery of function via guiding regenerating fibers.

Biocompatibility of PDMS+CNT Implanted In Vivo
Toward further biomedical applications, any new device needs to be challenged with the biological milieu in vivo. Given the beneficial interactions in vitro of PDMS+CNT materials com pared with PDMSalone, in the last set of experiments we assess the tissue response to PDMS+CNT implantation in the cortex of adult rats.
The use of PDMS as inert polymer for generating implant able scaffolds is traditionally accepted [28] nonetheless advan tages and limitations of any suitable material need to be evalu ated when microfabrication technologies are designed for bio logical applications. [29] Unfortunately, in our early observations, we detected a com plete absence of interaction of the brain tissue with implanted PDMS scaffolds when not incorporating MWCNTs (n = 2 ani mals; data not shown). In these animals, after 2 and 8 weeks (18 sections) postsurgery, the lack of integration of the PDMS scaffold with the surrounding cortical tissue is to an extent that the material no longer remains within the tissue fol lowing sectioning and subsequent histological processing for  Red arrowheads indicate the onset of each burst in each slice. b) Histograms represent the null distribution [12] of individual CCF values obtained from time windows independently sampled for the left and right recordings of both conditions; the red line indicates the average value of the actual CCF of sampled traces. Plots refer to the two explicative traces shown in (a), and point to a significant synchronicity in PDMS+CNT traces (**P < 0.01). c) The fraction of correlated pairs was significantly greater in slice pairs cultured on PDMS+CNT; *P < 0.05. d) Distance between slices was not correlated to CCF, indicating that distance had no effect on synchronous activity. e) Similar to distance, CCF did not correlate with the number of days in vitro of slice cultures.
immunofluorescence labeling. This makes it virtually impos sible to quantify the response in the surrounding tissue at the immediate interface with the material; i.e., we cannot predict, for example, how much of the glial scar, is removed along with the implanted material during the tissue processing. These results were not entirely unexpected, as other studies have dem onstrated a lack of PDMS integration with surrounding tissues when the PDMS material is not modified (such as via etching or the incorporation of adhesion molecules). [30,31] Because the in vitro results suggest that MWCNTs have a significant func tional benefit over PDMSalone substrates, we thus focused our biocompatibility study on the PDMS+CNT material.
We measured the surrounding microenvironment responses to PDMS+CNT implants, in particular we focus on the distribu tion patterns and infiltration of microglia together with astro cyte aggregation at the interface. [16,[32][33][34] Electrodes are common neuralimplanted devices; extensive reporting in the literature illustrates that many, if not most, implanted electrodes have reduced functionality over time due to the tissue response and glial scarring, an indicator of inade quate biocompatibility. [35][36][37][38] Glial scar formation typically occurs during the first 2-4 weeks following material insertion and is characterized by a dense reactive astrocyte and microglia region that can extend beyond 100 µm from the implant. [35,36,[38][39][40] This region can also undergo to relative neurodegeneration with a decrease in the number of local neurons. [35,36] We implanted PDMS+CNT scaffolds into the adult rat visual cortex, [16] this allowed accessing large cortical surfaces while avoiding areas containing too many blood vessels. In addition, the choice of this region allowed effective comparisons of out comes with our recent published work. [16] Rats were sacrificed at 2 weeks (n = 3), 4 weeks (n = 3), and 8 weeks (n = 3) postim plantation. Brain tissue sections were immunolabeled for astro cytes and microglia (Figure 5a,b) using antibodies against glial fibrillary acidic protein (GFAP) and Iba1, respectively, to deter mine the level of tissue reactivity surrounding the implanted . c) Increased GFAP immunoreactivity was observed primarily in the 50 µm immediately proximal to the implant. d) Iba1-positive tissue immunoreactivity up to 500 µm from the edge of the implant was found to significantly decrease over time; *P < 0.05, **P < 0.01, ***P < 0.001. material and any presence of a glial scar. Tissue sections were also immunolabeled with βtubulin III and NeuN, to determine if any neurons were able to infiltrate the material. [16] Overall, the immunoreactivity of astrocytes and microglia surrounding the implanted material suggests a generally low immune response. In fact, as shown in Figure 5a, the increased GFAP immunoreactivity surrounding the implant is substantially lim ited to the first 50-60 µm from the implant edge at all time points (quantified in Figure 5c). Further analysis reveals that, while the mean intensity up to 50 µm from the implant is not different between the three time points investigated (P = 0.119), the mean intensity up to 150 µm (namely, to larger distances from the implant edge), of the GFAP fluorescence is signifi cantly (P ≤ 0.001) lower in 4 and 8 weeks animals when com pared to 2 weeks (2 weeks = 1.73 ± 0.02 a.u., 4 weeks = 1.55 ± 0.02 a.u., 8 weeks = 1.57 ± 0.03 a.u.; Figure 5c). This suggests that, generally, the glial response in the surrounding tissue is likely decreasing when moving past the acute, inflammatory phase of the tissue reactivity. Regarding microglia cells, Iba1 immunoreactivity surrounding the implant (up to 500 µm) significantly decreases over time (Figure 5b,d). When meas ured, the average intensity is 0.22 ± 0.02 a.u. at 2 weeks, 0.14 ± 0.02 a.u. at 4 weeks; P = 0.004 and 0.08 ± 0.01 a.u. at 8 weeks (P ≤ 0.001 vs 2 weeks and P = 0.02 vs 4 weeks). Iba1positive microglia are observed to have infiltrated the implant material as early as 2 weeks postimplantation and continue to reside within the material at 8 weeks postimplantation.
Compared with other reports regarding the tissue response to cortical implants, we observe a general decrease in GFAP positive astrocyte labeling from 2 to 8 weeks after implantation and, by 8 weeks postimplantation, limited to the area immedi ately surrounding the implanted material ( Figure 5a). [36,37,40,41] In accordance with a reduction in inflammatory phase, there is a significant decrease in the intensity of Iba1positive micro glia labeling surrounding the implant over time. This is in con trast to the persistent upregulation of microglia associated with compromised biocompatibility and foreign body rejection often seen with implanted electrodes. [35][36][37][40][41][42] Interestingly, we observe Iba1 immunoreactivity inside the implanted material at all timepoints ( Figure 5b). It is interesting to note that an increased Iba1 immunoreactivity within MWCNTs, both when used as substrates [11] and as 3D scaffolds [16] has been observed before, without any indications that this increase is associated with an immune response. In vitro, [11] we observed an increase in Iba1positive cell density in dissociated neural cells with an insubstantial immune response, while in vivo [16] the coloniza tion of the material with native cells, as described here with the PDMS+CNT material, may sustain biocompatibility with neural tissue. Further research will be necessary to establish a thorough response profile and the true functionality of this material in vivo.
Additionally, in order to define whether the PDMS+CNT scaffold was fully integrated within the neural tissue, we inves tigated the presence of neurons within the material. [16] We observe βtubulin IIIpositive neuronal processes (Figure 6a) permeating the 3D scaffold at the three time points studied (27%: 6 of 22 horizontal slices, 57%: 16 of 28 horizontal slices, and 44%: 8 of 18 horizontal slices at 2, 4, and 8 weeks post implantation, respectively; Figure 6c). Therefore, βtubulin IIIpositive cells invade the implanted material as early as 2 weeks postimplantation and their presence increases at later stages, though without reaching a statistical difference. In a subset of slices (2 animals, 2 weeks, n = 8), we provide further evidence of infiltrated neuronal cells by exploiting the specific neuronal nuclear marker, NeuN. [16] In these experiments, we detect NeuNpositive neurons within the construct, as shown in an exemplifying image in Figure 6b. The presence of neu ronal cells, perhaps migrating neuroblasts, within the implant is surprising, however is in agreement with our recent report describing NeuNpositive cells localized in implanted MWCNT based scaffolds. [16] These findings, together with our previous work demonstrating very similar GFAP and Iba1 immunore activity in MWCNTalone scaffolds, [16] confirm the MWCNTs based materials as biocompatible 3D scaffolds and support their application as neural interfaces.

Conclusion
Controlled engineering of polymeric interfaces into 3D porous constructs favors the formation of intricate networks of variably oriented axons regrowing from the interfaced spinal tissues. The further incorporation of MWCNTs promotes the manufac turing of more efficient regenerating interfaces and guiding of regenerating axons to targets. Generally, the possibility to deco rate elastomeric structures with nanomaterials implements the adhesion of axons to the interfacing devices and can be used to fabricate conductive pathways within a 3D construct. In this work, a novel 3D structure was interfaced to spinal explants in vitro and its biocompatibility was assessed in vivo. The  presented results hold the potential to exploit the use of 3D hybrids and carbon nanotubes in the area of (nano)engineering regenerative interfaces.

Experimental Section
PDMS 3D Interface Fabrication: 3D PDMS scaffolds are free-standing, porous structures made by PDMS (Sylgard 184, Down Corning Co.), an elastomeric material commonly used in biomedical applications. [22][23][24]43] The fabrication procedures involve water dissolution of a sugar scaffold, resulting in a cast PDMS framework consisting of interconnected cavities as previous described. [7] In brief, 500 mg of food-approved sugar was passed through a No. 60 mesh sieve (Sigma-Aldrich) and mixed with 20 µL of deionized water, placed in a silicon mold, and subsequently dried at 65 °C for 30 min. PDMS was vacuum-forced to percolate inside the resulting sugar framework until all spaces were filled, and cured in an oven at 85 °C for 1 h. After cooling, the sugar was dissolved by submersing the sample in distilled water overnight. The water-driven dissolution of sugar grains produces a network of interconnected micropores inside the PDMS scaffold.
Prior to use, PDMS and PDMS+CNT substrates were treated with lowpressure air plasma for 5 min at high power using a PDC-32G Plasma Cleaner (Harrick Plasma) and sterilized under UV light for 20 min. PDMS is characterized by pronounced hydrophobicity-the major drawback for use of the material for in vitro and in vivo applicationswhich was addressed by an oxygen plasma treatment. This treatment turns the PDMS surface hydrophilic by introducing hydroxyl groups en masse to the surface. [22] Synchrotron Radiation (SR) Computed Microtomography and Scanning Electron Microscopy (SEM): To characterize 3D PDMS and PDMS+CNT scaffolds, computed microtomography (µCT) studies were performed. X-ray µCT experiments were carried out at the SYRMEP beamline of the ELETTRA synchrotron light source (Trieste, Italy). In order to enhance the image contrast, a single distance phase retrieval preprocessing algorithm [44] was applied to the CT projections, prior to the reconstruction procedure based on the standard Filtered Back Projection approach. The experiment was done in the white beam configuration mode with a polychromatic spectrum. Filters were applied for low energies (1 mm of aluminum) resulting in a hardened beam with average beam energy of 22 keV. The detecting system consisted in a sCMOS chip-based camera, placed at 10 cm away from the sample, and coupled with a high numerical aperture optic so the resulting pixel size was set to 0.9 µm. Each dataset consists in 1800 projections covering a total angle range of 180° with an acquisition time of 0.25 s per projection. The application of the tomographic method provides the benefit of reconstructing, in a nondestructive manner, the complex 3D organization of pores and channels inside the material and to provide a map of the void spaces. 3D volumes were obtained reconstructing the series of 2D projections using the STP (SYRMEP Tomo Project) software [45] and then analysis was performed with Pore3D software library. [46] The porous matrix was evaluated in terms of number and dimension of pores and throats, connectivity density, and anisotropy index. [47] Pore and throat numbers and size distributions were obtained following a skeleton analysis [48] and using the concept of maximal inscribed sphere, while connectivity density, a simple global measure of connectivity which gives higher values for better-connected structures and lower values for poorly connected structures, was evaluated finding the Euler characteristic. [47] PDMS and PDMS+CNT scaffolds morphologies were assessed via SEM. Images were acquired collecting secondary electrons on a Gemini SUPRA 40 SEM (Carl Zeiss NTS GmbH, Oberkochen, Germany). Bare scaffolds were mounted on conductive double side carbon tape (Ted Pella, Inc., USA) and imaged at 5 keV. Prior to SEM characterization samples were metalized with a thin layer of gold (<10 nm) using a metal sputter coater (Polaron SC7620).
Organotypic Spinal Cord Slice Preparation and Culture: Organotypic slice cultures were obtained as previously described. [10] Briefly, embryos at embryonic day 12 (E12), were isolated from timed-pregnant mice (C57Bl) euthanized by CO 2 overdose and decapitation. Spinal columns were isolated and the low thoracic to high lumbar regions were dissected away from the surrounding tissue. This tissue was sliced by tissuechopper transversely (275 µm) and further isolated the spinal cord and dorsal root ganglia tissue. Spinal cord slices were placed in cold GBSS and refrigerated for 1 h prior to placement on the substrates. Pairs of spinal cord slices were embedded into a thick matrix of chicken plasma (Rockland) and thrombin (Sigma) clot and placed upon the PDMS and PDMS+CNT substrates the distance between the slice pairs being within the range of 1.0-2.5 mm (interslice distance measured from center of the two slices); slice pairs not falling within this distances were discarded. The specified distance was necessary to ensure homogeneity in experimental conditions between the two populations (PDMS and PDMS+CNT) for electrophysiological and immunofluorescence experiments. Slice pairs were cultured for 10-22 d in vitro (DIV) in plastic Nunc tubes with 1.5 mL medium containing 67% Dulbecco's modified Eagle medium (DMEM) (Invitrogen), 8% sterile water for tissue culture, 25% fetal bovine serum (Invitrogen), and 25 ng mL −1 nerve growth factor (Alomone Laboratories); osmolarity, 300 mOsm; pH 7.35. The tubes were kept in a roller drum rotating 120 times per hour at 37 °C in humidified conditions with 5% CO 2 .
Electrophysiological Recordings and Analysis: Organotypic slice cultures at >10 DIV were mounted in a Perspex recording chamber, installed on an upright microscope (Leica DM LFS) with a continuous superfusion of control physiological saline solution containing (in mm): 152 NaCl, 4 KCl, 1 MgCl 2 , 2 CaCl 2 , 10 HEPES, and 10 Glucose; pH adjusted to 7.4 with NaOH. Simultaneous extracellular recordings were obtained at room temperature (RT; 20-22 °C) using low resistance glass micropipettes (4-6 MΩ) filled with 20 × 10 −3 m KCl saline solution. The electrodes were advanced manually using micromanipulators (Luigs and Neumann SM1, Germany) into the ventral region in both slices. Voltage LFP, as a consequence of synaptic activity and action potential firing, were recorded in regions 20-100 µm from the ventral fissure. [23,43] The recorded signals were amplified with the offset neutralized manually through the amplifier by current injection (Axopatch 1D; Axon Instruments). Signal recordings were acquired using the Clampex 8.2 software (pClamp suite, Axon Instruments) and digitized at 10 kHz (Digidata 1322A, Axon Instruments), filtered lowpass at 2 kHz. Strychnine and bicuculline (1 × 10 −6 and 20 × 10 −6 m, respectively) were administered into the bath to induce disinhibited rhythmic bursting. [49] Disinhibited bursts were evaluated by mean interevent interval analysis (the time between the onset of each burst; ≥20 bursts were used for each measure).
Disinhibited burst LFPs obtained from each spinal slice were lowpass filtered at 3 Hz and the data reduced and imported to MATLAB. [16] The synchrony was further assessed between the bursting activities of the two explants in each pair by computing the Pearson correlation coefficient between the two voltage time series. The statistical significance of the correlation coefficient was determined by performing a permutation test, [16] that allowed to measure the distribution of correlation coefficients that one would expect to observe if the voltage signals recorded from a pair of explants happened to correlate purely by chance. By measuring how likely it was for the values of this null distribution to be larger or equal than the real correlation coefficient, it was possible to understand whether the correlation between the pair of time series was significantly larger than expected by chance. This procedure allowed for determining what fraction of cocultured slices exhibited a significantly synchronous bursting activity, for all the tested conditions. A full amelioration of the cross-correlation analysis can be found in Usmani et al. [16] The fraction of significantly correlated pairs (P < 0.05) was compared between PDMS and PDMS+CNT groups by performing a chi-squared test for homogeneity.
Spinal Culture Imaging and Analysis: Images of immunolabeled organotypic spinal slice pairs were acquired using a Leica DM6000 epifluorescence microscope using a 2.5× dry objective. This low magnification allowed for visualizing both slices in the same field in order to identify slice location and to measure interslice distance. Further, a Nikon C2 confocal microscope was used to acquire higher quality images of the neurites spreading 3D into the substrate (40× oil objective, N.A. 1.3, oil mounting medium R.I. 1.515). The total stack thickness was set in order to obtain all β-tubulin III-positive processes from the sample with a z-step of 1 µm. Analysis and 3D reconstruction of the image stack was accomplished using NIS-Elements AR software (Nikon), Volocity (PerkinElmer), and the open source image processing package, FIJI (http://fiji.sc/Fiji).
Assessment of any differences in the 3D distribution of neuronal processes on PDMS and PDMS+CNT substrates was evaluated by considering image stacks taken from regions central (area between slices) and peripheral (at the substrate edge) to the slices. Utilizing the FIJI processing software, the thickness through which the β-tubulin III signal was present (indicating neural process extension) was measured and compared between PDMS and PDMS+CNT samples.
To measure the density of outgrowing β-tubulin III-positive neuronal fibers on PDMS and PDMS+CNT scaffolds, similar image stacks, as described above, with a z-step of 2 µm, were used. The FIJI software was used to measure the intensity of β-tubulin III-positive labeling of neuronal processes surrounding the spinal slice. The background intensity for each image was defined automatically by the software. The signal-positive area above the background threshold intensity was recorded to define the area of the image (depicted as a percentage of the total image area) positive for neuronal fibers (28 randomly sampled images from n = 10 PDMS samples; 20 images from n = 6 PDMS+CNT samples).
The alignment of fibers emerging from the spinal slices was quantified in terms of relative orientation by means of fiber directionality analysis [see ref. [16] for detailed methods]. This was carried out using the Directionality plugin of Fiji software. Fourier component analysis is executed to identify orientation of "structures;" in our case β-tubulin III immunolabeled neuronal processes. Confocal images measuring 640 × 640 µm 2 were used to compute mean fiber direction and the degree of dispersion (29 randomly sampled images from n = 10 PDMS samples; 20 images from n = 6 PDMS+CNT samples).
In Vitro Data Statistical Analysis: For all in vitro data, results are depicted as mean ± SE of the mean, unless otherwise stated; statistically significant differences between datasets were assessed using Statistica software (StatSoft, Inc.). Student's t test (after validation of variance homogeneity by Levene's test) was used for parametric data, Mann-Whitney test for nonparametric data; results were further confirmed using MATLAB (http://www.mathworks.com). Differences in the relative cumulative frequency distribution were obtained using the paired Kolmogorov-Smirnov (K-S) test. Chi-squared method analysis was performed for estimating significantly synchronous slices in the two groups (PDMS and PDMS+CNT) based on results obtained by the use of MATLAB. Statistical significance was established at P < 0.05.
In Vivo Implantation of PDMS+CNT Scaffold: All surgical procedures were performed on Wistar male rats (Harlan Laboratories; n = 9, three animals per time point), 3-6 months of age and weighing 450-550 g. Anesthesia was induced with isofluorane (2% in 100% O 2 , Sigma-Aldrich) administered via a nose cone and maintained throughout the procedure. Anesthetic depth was monitored by checking the absence of tail and paw reflexes. The anesthetized animal was placed in a stereotaxic apparatus (Narishige SR-5R) and body temperature was maintained at 37 °C with a thermostatically controlled heating pad (temperature control unit HB 101/2; Panlab/Harvard Apparatus) to avoid anesthesia-induced hypothermia. Heart rate and oxygen saturation level were monitored via a pulse oxymeter (Pulsesense LS1P-10R, Nonin Medical Inc., Medair AB).
The implant consisted of a sharp-pointed cylinder (≈2 mm in length and 0.5 ± 0.2 mm in diameter) carved out from a larger PDMS+CNT scaffold using a razor-blade and mounted on an electrode holder (ZIF-Clip, Tucker-Davis Technologies). [16] Implant stiffness was increased via polyethylene glycol (PEG-8000, Sigma-Aldrich) scaffold permeation after the carving procedure to facilitate subsequent insertion of the material into the brain tissue.
A square, 2 × 2 mm 2 craniotomy was performed over the left hemisphere using a microdrill and the dura mater was removed, exposing the brain. Tissue adhesive (B. Braun Aeusculap, Germany) was applied on the inner perimeter of the craniotomy to prevent brain dimpling. The coordinates for implantation of the PEG-stiffened PDMS-CNT implant were A-P −4.5 mm from Bregma and M-L −3.5 mm and relate to the visual cortex (specifically the cortical junction between V2ML and V1) [50] in the adult rat. The scaffold was inserted via a microdrive at a speed of about 0.5 mm s −1 to a depth of ≈D-V 2 mm until the material was completely within the cortical tissue. The high speed of insertion is necessary to prevent premature melting of the PEG. The craniotomy was filled with silicone (Kwik-Cast&Kwik-sil, World Precision Instrument) in order to isolate the exposed brain tissue and the skull surface covered by bone cement (Super Bond, C&B).
Throughout the procedure, the animal's eyes and cortex were periodically irrigated using an ophthalmic solution (Epigel, CevaVetem) and a 0.9% NaCl saline solution, respectively. Atropine (1.5 mg kg −1 ) and lactated Ringer's solution (1 mL) were administered subcutaneously every hour to reduce secretions and maintain homeostasis. Analgesic solution (Rymadyl; 5 mg kg −1 , intraperitoneal) and antibiotic (Baytril; 5 mg kg −1 , intramuscular) were administered to the rat via injection immediately before the surgery and 1 h prior to the end of the surgical procedure. Both solutions were administered following surgery via the animals' drinking water for 3 d postsurgery. Animals were monitored for recovery immediately after the surgery and at least three times per day for the first 72 h. A general distress-scoring sheet (Institutional Standards for the Care and Use of Animals in Research and after consulting with a veterinarian) was used to assess pain and distress in animals using physiological (appearance) and behavioral (natural and provoked behavior) parameters. For each parameter, a numerical score starting from 0 = normal, and reaching 3 = grossly abnormal, was determined. An extra point was added for each 3 given. The provoked behavior was tested after the first 72 h postimplantation. This method allows an objective monitoring of animals general conditions, to evaluate the need of analgesics or euthanization. All the animals used in the study did not show any sign of pain or distress (score = 0). Following this 72 h period, animals were monitored once per day until sacrifice. Animals were sacrificed at 2, 4, and 8 weeks postimplantation.
Brain Tissue Preparation and Immunohistochemistry: At the selected time points (2, 4, and 8 weeks postimplantation) animals were anesthetized with 5% chloral hydrate (7 mL kg −1 ) and perfused transcardially with 0.1 m PBS followed by 4% paraformaldehyde in PBS. Brains were removed, postfixed for 24 h at 4 °C in 4% PFA, and cryoprotected in 15% sucrose in PBS at 4 °C for 24 h and subsequently in 30% sucrose in PBS at 4 °C for at least 24 h. Brains were embedded in optimal cutting temperature compound (Tissue-Tek), frozen at −20 °C, and sectioned at 25 µm horizontally onto SuperFrost-Plus slides (Thermo Scientific) using a cryostat.
Brain Tissue Image Acquisition and Analysis: Immunoreactivity of the brain to the implanted material was measured by labeling tissue sections for reactive astrocytes and microglia using antibodies against GFAP and Iba1, respectively. Antibodies against β-tubulin III and NeuN were used to visualize neurons that may have infiltrated the scaffold. Fluorescence images were acquired using a Leica DM6000 upright microscope. In order to quantify GFAP and Iba1 labeling, a 10× dry objective was used to take images surrounding the entirety of the implanted material. Higher magnifications were used to take detailed images for observation of cellular infiltration. Image analysis was performed using FIJI software.
For GFAP intensity measurements, 24 intensity profile lines were drawn at random intervals around the implanted material, starting from the edge of the implant and extending 500 µm into the surrounding tissue, as previously shown. [16] These lines produced a fluorescence intensity profile as a function of distance from the implant edge. To account for variations in labeling, an image of the contralateral hemisphere in the same anatomical region as the implanted material was taken at 10× and used to define the background intensity for each section (the fluorescence intensity was measured with 9 randomly placed intensity profile lines). In each tissue section, GFAP intensity values at each micrometer were calculated for 0-150 µm from the implant edge, normalized to the average intensity for the contralateral hemisphere, and finally averaged across all sections (n ≥ 7) for each animal to create a mean intensity profile. Because differences in GFAP intensity were observed to be within 50 µm from the implant, for each animal the data from 0 to 150 µm were further normalized to the average intensity measured at 51-150 µm from the edge of the implant; these data were used for statistical analyses. [16] For Iba1 intensity measurements, n = 8 ROIs (100 × 500 µm 2 ) were selected starting from the edge of the implant and extending 500 µm into the surrounding tissue. [16] The background intensity threshold was defined for each section using the Iba1 labeling intensity measured in the contralateral hemisphere in the same anatomical region as the implanted material. The area within each ROI with intensity above the background threshold was calculated and used for statistics. The ROI for all sections were averaged for each animal and are depicted as the mean ± standard error of the mean (SEM).
Statistical analysis was performed using Statistica software (StatSoft, Inc.). One-way analysis of variance (one-way ANOVA) was used to determine statistical significance at P < 0.05, with posthoc analysis performed using the Fisher's least significant difference (Fisher's LSD) method to compare datasets.
Ethics Statement: This study was carried out in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and the appropriate international and institutional standards for the care and use of animals in research (Italian Ministry of Health, in agreement with the EU Recommendation 2007/526/CE). The protocols and all performed experiments are approved by the local veterinary service and the institutional (SISSA) ethical committee, in accordance with the EU guidelines (2010/63/UE) and Italian law (decree 26/14). All efforts were undertaken to minimize pain and suffering of the animals and to reduce the number of animals used.