Patterning human neuronal networks on photolithographically engineered silicon dioxide substrates functionalized with glial analogues

Interfacing neurons with silicon semiconductors is a challenge being tackled through various bioengineering approaches. Such constructs inform our understanding of neuronal coding and learning and ultimately guide us toward creating intelligent neuroprostheses. A fundamental prerequisite is to dictate the spatial organization of neuronal cells. We sought to pattern neurons using photolithographically defined arrays of polymer parylene-C, activated with fetal calf serum. We used a purified human neuronal cell line [Lund human mesencephalic (LUHMES)] to establish whether neurons remain viable when isolated on-chip or whether they require a supporting cell substrate. When cultured in isolation, LUHMES neurons failed to pattern and did not show any morphological signs of differentiation. We therefore sought a cell type with which to prepattern parylene regions, hypothesizing that this cellular template would enable secondary neuronal adhesion and network formation. From a range of cell lines tested, human embryonal kidney (HEK) 293 cells patterned with highest accuracy. LUHMES neurons adhered to pre-established HEK 293 cell clusters and this coculture environment promoted morphological differentiation of neurons. Neurites extended between islands of adherent cell somata, creating an orthogonally arranged neuronal network. HEK 293 cells appear to fulfill a role analogous to glia, dictating cell adhesion, and generating an environment conducive to neuronal survival. We next replaced HEK 293 cells with slower growing glioma-derived precursors. These primary human cells patterned accurately on parylene and provided a similarly effective scaffold for neuronal adhesion. These findings advance the use of this microfabrication-compatible platform for neuronal patterning. © 2013 The Authors. Journal ofBiomedicalMaterials Research Part APublished byWiley Periodicals, Inc.Wiley Periodicals, Inc. J Biomed Mater Res Part A: 102A: 1350–1360, 2014.


INTRODUCTION
Engineering and interacting with bespoke in vitro neuronal networks have the potential to enhance understanding of information processing in real neuronal networks 1 and may form a useful platform for pharmacological screening in diseases, such as epilepsy and stroke. 2 As bidirectional interaction with such networks becomes possible, this also offers a promising approach to developing neuroprosthetic devices. However, creating such networks requires precise control of cell body adhesion and also neurite outgrowth and connectivity. To interact with a defined network, methods that enable stimulation and recording from patterned cells must also be amenable to incorporation. These collective demands motivate the approach of merging silicon semiconductor microelectronics with neuronal cell patterning.
The concept of building bespoke neuronal networks on silicon is not new. [3][4][5] Contemporary work 6 has furthered the idea to take advantage of various cellular lithographic techniques, has explored the impact of glia in patterned networks, and has utilized multielectrode arrays to record cellular activity. Our group focuses on the use of parylene-C as a neuronal patterning substrate. Parylene-C is a biocompatible polymer used commercially to coat printed circuit boards.
Photolithographic patterning of parylene-C on silicon dioxide, followed by activation with serum, has enabled patterning of primary murine hippocampal cells, 7-10 a human teratocarcinoma cell line, 11,12 and the human embryonal kidney (HEK) 293 cell line. 13 This straightforward and reliable technique is significantly simpler than some multistage protocols used for neuronal patterning. Specifically, patterned parylene substrates are biologically stable and can be stored until needed (whereupon they are activated).
Whilst parylene-C has been used previously in the context of cell patterning and cell trapping, 14 its use for neuronal patterning after serum activation is in its infancy. Exploration of the mechanisms underlying cell patterning suggests that both adhesive and repulsive components in serum interact to imbue each substrate with contrasting cytoadhesive or cytorepulsive characteristics, although these components are not yet characterized. 13 Although patterning primary murine hippocampal cells (which contain both neurons and glia) is effective, it remains unclear whether neurons in isolation are capable of patterning or whether glia adhere and (by close association) enable neurons to respect the underlying parylene geometry. The presence of glia amongst patterned neurons, though better reflecting the in vivo environment, may complicate downstream efforts to record from and stimulate individual neurons. We therefore sought to pattern neurons in isolation, questioning whether neurons themselves will pattern or whether they are dependent on the presence of glial (or other) cell types.
The lund human mesencephalic (LUHMES) cell line manifests well-described functional neuronal characteristics. 15 These conditionally immortalized cells can be induced to differentiate by shutting down the myc transgene. Inactivation of the oncogene by tetracycline-mediated gene expression allows neuronal differentiation to proceed, resulting in a pure source of postmitotic neurons in 5 days. Important phenotypic characteristics include formation of one to two neurites (>500 mm long), dynamic growth cone behavior, and timely generation of spontaneous electrical activity.
We initially attempted to pattern isolated LUHMES in both their undifferentiated (UD) and their differentiated state. Subsequently, coculture environments were tested. These sought to assess neuronal behavior (with respect to cell adhesion and morphological differentiation) in the presence of a different prepatterned cell type. Toward this end, patterning behavior was assessed in a range of different cell lines and was quantified by measuring adhesive and repulsive indices on parylene and SiO 2 , respectively. Cell lines tested were UD N2a (Neuro 2A): a mouse neuroblastoma-derived neuronal cell line. HEK 293: previously considered a derivative of mouse embryonic fibroblastic or endothelial renal cells, 16 current research suggests an early neuronal lineage (evidenced by presence of messenger RNA and gene products typically found in neurons neurofilament-M, neurofilament-L, and a-internexin) and the endogenous expression of several voltage-gated ion currents. 17,18 N9 microglia: a mouse-derived cell line with phenotypic characteristics similar to primary microglia. 19 UD 3T3 L1 preadipocytes: derived from 3T3 cells (themselves derived from primary mouse embryonic fibroblasts), these cells have fibroblast-like characteristics. 20 HEK 293 cells illustrated the most robust patterning characteristics and were therefore used for initial coculture experiments with LUHMES neurons.
We subsequently assessed patterning characteristics of three different human glioma-derived stem-like cultures (GSC); our hypothesis being that these glial analogues would also pattern accurately on parylene. Finally, LUHMES neurons were sequentially cocultured with one of the human GSC lines. Our aim was to create a network of human neurons with a configuration defined by underlying human glial-like cells, with glial adhesion itself demarcated by parylene-patterned silicon dioxide.

MATERIALS AND PROTOCOLS
Fabrication of parylene patterns on SiO 2 : Process flow 1. Silicon wafers (Siltronix, Archamps, France) were oxidized in an atmospheric horizontal furnace (H 2 1.88 SLM and O 2 1.25 SLM) at 1100 C for 40 min to produce a 500 nm SiO 2 layer (measured using a Nanometrics Nano-Spec 3000 reflectometer). 2. Oxidized wafers were primed with Merck Silane A174 adhesion promoter, followed by deposition of 100 nm coating of parylene-C (22 C at a rate of 1.298 nm/mg of dimer using a SCS Labcoter 2 deposition Unit, Model PDS2010). 3. Hexamethyldisilazane adhesion promoter was deposited on parylene-coated wafers in a SVG 3 inch photoresist track followed by application of 1 mm thick film of Rohm & Hass SPR350-1.2 positive photoresist, by spinning at a speed of 4000 rpm for 30 s. 4. Wafers then soft baked for 60 s at 90 C. 5. Wafers and premanufactured photo mask (Compugraphics International, Glenrothes, Scotland) were placed in Suss Microtech MA/BA8 mask aligner and ultraviolet (UV) exposed. The primary parylene design consisted of three iterations of circular parylene nodes with a centered "cross-hair" (node diameters 250 mm, 100 mm, and 50 mm, cross hairs 450 mm in length for largest node size, 300 mm for the two smaller nodes) on chips 7.7 mm 3 5.9 mm in dimension. All node/cross hair complexes were separated from one another by a distance of 100 mm. Nodes were arranged orthogonally with the chip designed such that there were three regions for each node size. A second chip design was used to further explore differences in on-chip behavior of GSC lines and LUHMES neurons. Here, 50 mm diameter nodes were arranged in a grid configuration separated from one another by 400 mm horizontally and vertically. Two different variations were created; one in which a 2-mm wide parylene track extended diagonally from node to node, and another in which only very short (30 mm long and 2 mm wide) parylene tracks partially extended from nodes.

Maintenance of cell lines: Protocol 1. HEK 293 cells (human embryonic kidney cells; American
Type Culture Collection, VA) maintained at 37 C and 5% CO 2 in Dulbecco's modified Eagle's medium (DMEM, Gibco Invitrogen) supplemented with 10% FBS. In monoculture experiments, cells applied to chips as a suspension of 5 3 10 4 cells/mL in growth media and imaged live after 3 days in vitro.

LUHMES cells (American Type Culture Collection) main-
tained at 37 C and 5% CO 2 in precoated plastic culture flasks (coated with 50 mg/mL poly-L-ornithine and 1 mg/ mL fibronectin in H 2 O for 3 h). Proliferation media consisted of Advanced DMEM/F12 (Gibco Invitrogen), 13 N2 supplement, 2 mM L-glutamine and 40 ng/mL recombinant basic fibroblast growth factor (FGF; Gibco Invitrogen). Differentiation media consisted of Advanced DMEM/F12 (Gibco Invitrogen), 13 N2 supplement, 2 mM L-glutamine and 1 mg/mL tetracycline. To differentiate into postmitotic neurons, media changed from proliferation to differentiation 24 h after passage. After 2 further days in differentiation media, cells trypsinized and plated. For monoculture experiments, UD LUHMES plated as a suspension of 5 3 10 4 cells/mL in proliferation media, predifferentiated LUHMES plated as a suspension of 30 3 10 4 cells/mL in differentiation media and both imaged alive after 3 days in vitro.

3T3 L1 (a gift from Dr Luke Chamberlain, Strathclyde
Institute of Pharmacy and Biomedical Sciences, University of Strathclyde) maintained at 37 C and 10% CO 2 in DMEM (Gibco Invitrogen) supplemented with 10% fetal calf serum and 1% Pen-Strep. Cells applied to chips as a suspension of 5 3 10 4 cells/mL in growth media and imaged alive after 3 days in vitro.

N2a cells (American Type Culture Collection) maintained
at 37 C and 5% CO 2 in DMEM (Gibco Invitrogen) supplemented with 10% FBS. Plated on-chip at a density of 5 3 10 4 cells/mL and imaged alive after 3 days in vitro.

N9 cells (a gift from Prof Alun Williams and Dr Clive
Bate; The Royal Veterinary College, University of London) maintained at 37 C and 5% CO 2 in Iscove's modified Dulbecco's medium (Gibco Invitrogen) with 5% FCS, 100 IU/mL penicillin, and 100 lg/mL streptomycin. Plated on-chip at a density of 15 3 10 4 cells/mL and imaged alive after 3 days in vitro. 6. Human glioma-derived primary cell cultures were obtained from fresh human glioma tissue removed intraoperatively during surgery. All patients gave informed signed consent.  Although LUHMES fail to adhere in isolation, they are capable of patterning effectively if a chip is prepatterned with HEK 293 cells. Figure 2 illustrates LUHMES in coculture with HEK 293 cells. Prepatterned HEK 293 cells enable secondary adhesion of predifferentiated LUHMES. These neurons also show morphological signs of differentiation with neurites extending from parylene nodes to explore the surrounding SiO 2 environment. The geometric arrangement of underlying parylene informs neuronal configuration, promoting formation of linear neurite connections between patterned nodes; as in Figure 2(B,C). Note, however, the continued rapid growth of underlying HEK 293 cells in When HEK 293 cells were grown on un-patterned polystyrene in media containing citrinin (to retard overgrowth), their doubling time was increased from 24.8 h to 63.4 h. However, in the context of patterned parylene substrates and LUHMES coculture, this reduction in proliferation was insufficient to prevent problematic node overgrowth.
Considering the possibility that patterned but nonviable HEK 293 cells (or cell membrane fragments) might still enable LUHMES adhesion, hypotonic cell stress was applied to HEK 293-patterned chips, before LUHMES application. HEK 293 cells lifted off extensively in response to all hypotonic stress time periods. Secondary application of LUHMES was globally unsuccessful with no adhesion to areas previously occupied by HEK 293 cells. Directionality of neurite growth is shown in radial plots in Figure 3 E)]. In contrast, when LUHMES are cocultured on GSC-A, they illustrate the capacity to extend neurites between nodes [compare left and right panels of Fig. 6(A,B)]. Figure 6(C) shows neuron-specific bIII tubulin-stained cocultures, confirming the presence of neurites projecting between nodes.

LUHMES in isolation
We set out to establish whether isolated LUHMES neurons were capable of selective adhesion and network formation on activated parylene-patterned SiO 2 substrates. Isolated LUHMES do not pattern, either in an UD or differentiated state, nor do they manifest morphological changes suggesting differentiation (Fig. 1). For LUHMES the chip constitutes a globally cell repulsive environment, reflected by the very low PAI and very high SRI. This behavior contrasts starkly with the accurate patterning seen with primary murine hippocampal cells, [7][8][9] suggesting that the presence of glia in these preparations may be key to enabling neuronal patterning.
Heterogeneous patterning behavior across cell lines We next explored patterning behavior across a range of cell lines, confirming the heterogeneity of responses toward the parylene/SiO 2 chip construct (Fig. 1). The interface between a cell and adjacent foreign substrate is complex, dynamic, and bidirectional. 22 Having illustrated this variation in cell adhesion behavior, a future strategy may involve identifying differences in the expression of key cell surface adhesion molecules that contribute to a cell's capacity to pattern. For example, interrogation of DNA microarray databases from a variety of cell types with contrasting patterning behavior should allow correlation between gene expression profiles and a cell's tendency to adhere to or be repulsed by the two contrasting substrates. This may open the way to targeted cell manipulation so as to empower a cell to pattern. Moreover, this approach may reciprocally help to elucidate important substrate-specific aspects of patterning.

Coculture of LUHMES and HEK 293 cells
Given the failure of isolated LUHMES to pattern we questioned whether neurons would pattern in the context of a supporting cell substrate, reminiscent of the glia-neuron patterning seen with primary rodent cocultures. [7][8][9] Of the array of four cell lines tested, HEK 293 cells manifest the best patterning profile and were used for initial coculture experiments. HEK 293 cells enable LUHMES to adhere selectively to the chip, by binding with patterned HEK cell clusters on parylene. As such, HEK 293 cells appear to perform a role analogous to glia. Moreover, LUHMES neurons that attach to HEK 293 cell clusters differentiate morphologically, evidenced by neurite formation. HEK 293 cells facilitate neuronal patterning by providing a physical point of attachment on-chip. In addition, this cell:cell interaction may enable LUHMES differentiation to proceed, where this was not possible in isolation.

Control of neurite behavior
For the creation of neuronal networks capable of meaningful interrogation, neurite growth direction and connectivity needs to be controlled. We illustrated the ability to grow orthogonally arranged networks, as defined by the geometric configuration of parylene nodes. Each of the three node sizes/configurations used promotes some orthogonality but accuracy improves from largest (250 mm diameter) to smallest (50 mm diameter) node.
Neurite growth cones, in contrast with LUHMES and HEK 293 cell somata, appear capable of adhering and traversing unpatterned SiO 2 . It is unclear whether on-chip neurite growth is dominated by haptotactic or chemotactic (or both) mechanisms. Neurites appear to explore surrounding areas [ Fig. 2(A)] and, after encountering another node, straighten up under tension. That growth cones of developing neurons generate tensile forces is well described 23 and this selforganizing behavior has been observed before in the context of carbon-nanotube patterned substrates. 1 Recent work using primary hippocampal cells has also noted how the geometry of patterned adhesive regions can influence axonal outgrowth, 24 in this case using different polygons of microcontacted printed poly-L-lysine and laminin.
Although this approach allows a degree of network organization, it is hindered by the rapid proliferation of HEK 293 cells. As unrestricted HEK 293 proliferation continues, in contrast with arrested growth of postmitotic LUHMES neurons, patterning is overwhelmed. The combination of HEK 293 overgrowth, and the internode tension applied by linking neurites, tends to cause cell lift off and network obliteration over time . This is an insufficient time window for the establishment of spontaneous electrical activity of neurons in the network. In addition, our ability to reliably fix and stain LUHMES/HEK 293 cocultures was compromised due to network fragility and cell lift-off. To ameliorate, methods for retarding or arresting HEK cell growth were explored. Citrinin (a nephrotoxic mycotoxin which disrupts microtubule function) has been used to induce cell cycle arrest in HEK 293 cells. 25 Citrinin successfully reduced HEK 293 doubling time on unpatterned polystyrene substrates but its effect was insufficient in the context of patterned coculture. However, such network lift-off [seen best in Fig. 3(D)] is potentially useful for neuroregeneration purposes. For example, it opens the possibility of creating a neuronal network on-chip that could then be detached and implanted with the intention of in vivo nervous system repair.
Questioning the mechanics by which HEK 293 cells facilitate LUHMES adhesion, we used hypotonic cell stress before LUHMES application. Hypotonic cell stress using dH 2 O for intervals ranging from 30 s to 60 min resulted in almost total HEK 293 cell lift off. However, secondary application of LUHMES was universally unsuccessful, suggesting residual cell fragments/membranes are insufficient to enable LUHMES adhesion.

Coculture of LUHMES and glial analogues
Given the success of HEK 293 cells as a glial analogue, a source of human glial cells was sought to improve the patterning model and potentially circumvent the issue of cell overgrowth. Our group has previously established primary cultures of human glial stem-like cells from a range of human gliomas. Cells from these cultures have been shown to have predominantly astrocytic characteristics, but also have tumor-associated mutations. Of the three primary cultures GSC-C and GSC-E have a much less disordered genome than GSC-A, because GSC-A is derived from a higher grade tumor. Despite their genetic aberrations, and evident ineligibility for downstream uses including neuroprosthetics, these cultures are more readily available than primary human astrocytes and easier to work with at this proof-of-concept stage. They also grow more slowly than the serially passaged HEK 293 cells, facilitating better observation and intervention of the glial-neuronal interaction.
Given previously published primary hippocampal cell work, we first hypothesized that these glial cell lines would pattern effectively. This is confirmed (in Fig. 4) with excellent patterning indices, comparable to those of HEK 293 cells. Using GSC-A to prepattern the parylene template, we again illustrated successful coculture with LUHMES and the generation of reticular orthogonally arranged networks (see Fig. 4). GSC-A cells are slower growing than HEK 293 cells, providing a greater opportunity for long-term establishment of stable neuronal networks. Glioma-derived cell lines and the HEK 293 cell line robustly respect the underlying parylene design; being repulsed from bare SiO 2 and adhering to parylene. Only by creating a design with a parylene track between nodes can GSC-A be induced to connect with adjacent nodes. This contrasts with neurite behavior; with neurites being capable of bridging the nonparylene gap. This is important for downstream techniques to interrogate electrical and synaptic activity in the patterned network. By focusing on the parylene-free internode gap [ Fig. 6 (C), region demarcated by arrow head] one can be confident that neurites alone are being interrogated.
Future work will seek to better establish the rules governing neurite growth and organization. Axon guidance molecules or other topographical features on-chip may allow another level of control. Work is ongoing to assess electrophyiological and other functional behaviors of patterned neurons, and to confirm formation of functional synapses in the network. In addition, improved understanding of the underlying mechanisms governing cell adhesion to activated parylene will further enhance the utility of this platform.

CONCLUSIONS
LUHMES neurons require an intermediate cell type to adhere to serum-activated parylene-patterned SiO 2 . HEK 293 cells fulfill this role and, in doing so, perform a function analogous to glia. These engineered cocultures organize into orthogonally configured reticular networks, informed by the spatial geometry of underlying parylene nodes. However, HEK 293 overgrowth, neurite tensile forces, and resultant cell lift-off compromise long-term in vitro network viability.
Human tumor-derived glial precursor cells pattern accurately and directly on serum-activated parylene-patterned SiO 2 . These prepatterned glial analogues enable defined secondary adhesion of LUHMES neurons, and in so doing promote formation of orthogonally arranged neurite connections. This work illustrates the potential utility of activated parylene as another tool for generation of bespoke neuronal networks on silicon.