Cancer-mediated axonal guidance of sensory neurons in a microelectrode-based innervation MPS

Despite recent advances in the field of microphysiological systems (MPSs), availability of models capable of mimicking the interactions between the nervous system and innervated tissues is still limited. This represents a significant challenge in identifying the underlying processes of various pathological conditions, including neuropathic, cardiovascular and metabolic disorders. In this novel study, we introduce a compartmentalized three-dimensional (3D) coculture system that enables physiologically relevant tissue innervation while recording neuronal excitability. By integrating custom microelectrode arrays into tailored glass chips microfabricated via selective laser-etching, we developed an entirely novel class of innervation MPSs (INV-MPS). This INV-MPS allows for manipulation, visualization, and electrophysiological analysis of individual axons innervating complex 3D tissues. Here, we focused on sensory innervation of 3D tumor tissue as a model case study since cancer-induced pain represents a major unmet medical need. The system was compared with existing nociception models and successfully replicated axonal chemoattraction mediated by nerve growth factor (NGF). Remarkably, in the absence of NGF, 3D cancer spheroids cocultured in the adjacent compartment induced sensory neurons to consistently cross the separating barrier and establish fine innervation. Moreover, we observed that crossing sensory fibers could be chemically excited by distal application of known pain-inducing agonists only when cocultured with cancer cells. To our knowledge, this is the first system showcasing morphological and electrophysiological analysis of 3D-innervated tumor tissue in vitro, paving the way for a plethora of studies into innervation-related diseases and improving our understanding of underlying pathophysiology.


Introduction
Microphysiological systems (MPSs) technology has made significant strides in enabling the investigation of complex biological interactions in accurately represented organs and tissues.However, the importance of innervation for proper tissue function has been largely overlooked [1].Peripheral innervation plays an essential role in mediating the response of skin, bone, muscles and visceral organs to environmental changes and maintaining tissue homeostasis [2][3][4].Additionally, in diseases such as cancer, aberrant innervation patterns can have significant effects on pain, highlighting the critical role of innervation in pathological conditions.Therefore, incorporating nerves in three-dimensional (3D) in vitro models is urgently required.Although previous approaches have used micropatterned structures to manipulate axons by injury and/or growth factor cocktails, these are limited to two-dimensional (2D) space and not appropriate for innervation of complex 3D structures such as organoids and spheroids [5][6][7][8][9].To address these shortcomings, we developed a compartmentalized 9-well INV-MPS for coculture of peripheral nervous system (PNS) neurons with 3D tissues, all embedded in hydrogels.In addition, we have integrated microelectrode arrays (MEA) to allow direct and parallel recording of electrophysiological activity in 3D neurons with high spatio-temporal resolution, which is highly relevant to the study of sensory function, more physiological than calcium imaging and more efficient than patch-clamp recording [10].This represents a substantial innovation over currently available MEA-based microfluidics, which are limited to 2D applications.Various MEA configurations have been commonly combined with poly-dimethylsiloxane (PDMS)-based microfluidics to study, for example, the sensitizing (painful) effects of exposing the PNS to the secretome of mature osteoclasts in bone disease [11].Researchers also exploited MEA technology to create flexible polyimide-based devices that interface with unmyelinated C-fibers using protruding 3D gold pillars to record elusive low-amplitude neural signals in the nerve plexus [12].Lastly, relevant work includes combining stereolithographic printing of polycarbonate devices and membranes with PDMS for placement inside multiwell MEA plates for functional analysis of iPSC-derived sensory neurons innervating spinal cord organoids [13] and skin [14].As a proof-of-concept for the potential use of the newly developed INV-MPS for innervation studies, we specifically investigate sensory innervation of cancer spheroids.This is also motivated by the significant unmet medical needs associated with cancer-induced pain (CIP), as well as the extensive amount of literature available on nociceptive fibers, both in vivo and in vitro, and growing ethical concerns regarding in vivo animal research in this field.
The INV-MPS is fabricated by bonding a selectively laser-etched glass chip to a custom MEA, creating two individual compartments connected by micro-tunnels.Suspended in hydrogel, murine dorsal root ganglion (DRG) neurons and colorectal or pancreatic cancer cells are deposited into the respective compartments.Once axons extend through the barrier, interaction of neurites and cancer spheroids can be monitored via time-resolved live imaging and electrophysiological recording.Current literature, mainly based on in vivo studies, reveals an increase in nerve density/nerve alteration and pain states, also triggered by tumor-mediated release of factors such as NGF and prostaglandins (PGE 2 ) [15][16][17][18][19][20][21].By exposing compartmentalized nerve terminals to pain and inflammatory mediators after applying a NGF gradient we confirmed the potent effect of NGF on inducing a nociceptive phenotype.Remarkably, in the absence of NGF, colorectal cancer spheroids were also able to attract sensory neurites and become innervated, accompanied by the induction of excitability through application of pain agonists.These findings suggest that cancer cells alone can drive a nociceptive phenotype, similar to NGF.Finally, live-imaging of Ca 2+ transients in cancer-innervating fibers confirmed that the neurites are responsive to noxious stimuli while interacting closely with cancer cells.Functional innervation of cancer spheroids by sensory fibers as shown here serves as a case study for the INV-MPS, an innovative 3D in vitro model amenable for exploring tissue-induced axonal growth, nerve sprouting and associated changes in neuronal electrophysiology.However, the platform has the potential to benefit a much broader community of microphysiological scientists who aim to recapitulate and explore the complexity of innervated tissues.Thus, the INV-MPS represents a significant advancement towards novel human-relevant drug discovery models, particularly in pathologies where tissue innervation plays a critical role.

A novel MPS with integrated electrodes for 3D axon manipulation and recording from sensory neuron cultures
Although MPSs have advanced with rapid strides recently, the development of new in vitro PNS models has been hampered by challenges in integrating effective readouts into complex 3D cultures.However, improving existing models, particularly with respect to tissue innervation, is critical for studying pathophysiological processes and the development of new therapeutic strategies.Therefore, we developed a novel MPS that integrates a non-invasive electrophysiological readout with glass microfluidics to produce a valuable 3D neurite outgrowth model that separates soma and axons.Specifically, we grew peripheral neurons in a 3D scaffold and facilitated their projection through a physical barrier, enabling them to innervate axonally residing tissue (figure 1(A) and movie S1).To achieve this, we produced a fused silica MPS patterned with µm-scale features and attached it to a MEA substrate, as previously described [10].The MPS includes pairs of adjacent two-layer tissue wells (somatic and axonal compartment) featuring defined hydrogel layers (GL; approx.8.0 µl) below media layers (ML; approx.110.0 µl).
The two adjacent tissue wells are separated by a 300.0 µm thin barrier featuring 12 rows of 110 microtunnels (10.0 × 10.0 µm cross-section) each, over a surface area of approx.4.0 × 1.0 mm (figure 1(B) and SI appendix, figure S1(A)).This 3D-arrangement of µ-tunnels is possible due to the unique capabilities of selective laser etching (SLE) that enables the selective removal of glass from a monolithic piece in 3D.The microtunnels (figure 1(C)) allow only DRG neurites to pass into the GL of the axonal compartment and constrain DRG cell bodies to the somatic compartment.To record from gel-embedded 3D neurons inside the INV-MPS, we utilized capped microelectrodes (CMEs) positioned at the base of the somatic compartment that effectively trap axons (figure 1(D)).Neurons occupying CME-microtunnels also extended towards the barrier and crossed into the axonal compartment in 3D.To confirm the spatial separation of soma and nerve terminals, we used live-cell confocal fluorescence microscopy of DRG neurons transduced with adenoviral vectors encoding enhanced green fluorescent protein (eGFP) (figure 1(E) and movies S2 and S3).Additionally, we confirmed that neurites grew inside the confined spaces of the CMEs (5.0 × 3.0 µm), allowing us to detect electrophysiological signals (up to ∼100 µV) generated upon firing of action potentials (SI appendix, figure S1(B)).Each micropatterned glass chip was bonded onto a 49.0 × 49.0 mm glass substrate containing 117 embedded recording electrodes.Thirteen recording-CMEs and a reference electrode were positioned within each of the nine somatic compartments.CMEs were connected to edge-pads, which secured the connection to pins of the MEA-recording system (figures 1(F) and (G)).

3D axonal guidance and nerve terminal excitability modulated by NGF inside the MPS
Our study examined the responsiveness of 3D gelembedded DRG neurons to neurotrophic gradients, which may have implications for tissue innervation under various pathophysiological conditions in vivo.
We focused on NGF, a well-known neurotrophic factor that promotes the development of neuronal growth cones and can enhance neuronal survival in vitro [22,23].Here, experiments were performed using an INV-MPS without integrated electrodes (SLE INV-MPS; SI appendix, figure S2(C)), which features a single row of tunnels at the bottom of the barrier, allowing for accurate comparison of neurite outgrowth and microtunnel diffusion kinetics.To analyze the impact of NGF on 3D sensory fiber`extension into the axonal compartment, we supplemented NGF solely axonally (NGF −/+), thereby creating a stable gradient over the barrier for 1 week.We found a significant increase in fibers crossing the barrier (figures 2(A) and (B) and SI appendix, figure S2(A)) with limited variability in survival across different conditions (SI appendix, figure S2(B)).When NGF was supplied to both compartments (NGF +/+), few axons extended through the barrier and neurites predominantly grew within the somatic compartment, likely due to the lack of a chemoattracting gradient.To mimic a NGF gradient across adjacent wells, we monitored the concentration of fluorescently labelled dextran moieties (applied to the axonal compartment) for 12 h and observed a ∼10x lower concentration of both 4 kDa and 20 kDa tracers in the somatic compartment.It is important to note that the small amount of dye that diffuses through the barrier is progressively diluted into a much larger volume in the somatic compartment, resulting in a significantly lower bulk concentration.Having optimized the conditions for neurite outgrowth in the SLE INV-MPS, we then transitioned to using the INV-MPS with integrated MEA for electrophysiological characterization of 3D sensory terminals in the NGF −/+ condition (figure 2(D)).Application of TRPV-1 agonist capsaicin [24] to the axonal compartment of the INV-MPS resulted in measurable action potentials in the majority of CMEs located in the somatic compartment (figure 2(E)).This provides evidence that compartmentalized 3D neuronal culture in the INV-MPS can successfully replicate peripheral activation of both receptors and ion channels involved in generating and transmitting nociceptive signals from nerve terminals to the soma.Recorded signals appeared within a timeframe of 3-5 s from administration of capsaicin, indicating the system's ability to rapidly detect and measure terminallyevoked changes in neuronal activity.Given the noninvasive nature of MEA electrophysiological recordings, our system enables investigation into the longterm effects of substances within an inflammatory milieu.This was demonstrated by recording changes in neuronal activity induced by an inflammatory mediator cocktail (bradykinin, prostaglandin, serotonin and histamine), where action potentials could be continuously measured for up to 30 min by the majority of electrodes (figure 2(F)).Our findings indicate that the INV-MPS can continuously monitor activity generated in 3D nerve terminals, providing a valuable tool for studying the mechanisms involved in peripheral inflammatory pain.Examining electrode response rates following capsaicin application, we found that a response rate of over 80% can be achieved within 6 d from plating.This is longer compared to typical MEA recordings of DRG cultures in 2D (2-3 d), most likely due to the additional time required for neurites to extend into the axonal compartment (figures 2(G) and (H)).Application of different doses of capsaicin (0.1-1.0 µM) on day 6 revealed a direct relationship between the concentration of the agonist and the number of observed responses and their potency (figures 2(I) and (J)), suggesting that the INV-MPS could facilitate applications in drug discovery.In summary, the INV-MPS could be utilized to study the engagement of specific ligand-gated receptors or voltage-gated ion channels that may play a role in modulating pain states, as well as to monitor the potential of various factors to induce neuronal outgrowth or nerve degeneration in toxicity studies.

Induction of sensory neurite outgrowth by colorectal cancer spheroids in the absence of NGF supplementation
We then examined whether the INV-MPS could be utilized as a tool for evaluating changes in nerve density and innervation induced by cancer in a neuroncancer coculture.This investigation could contribute to the comprehension of the underlying neuronal mechanisms involved in pathological pain in vivo [25][26][27][28].We seeded DRG neurons in the somatic compartments of the SLE INV-MPS, while colorectal (HT29) or pancreatic (PANC1) cancer spheroids were seeded in the axonal compartment.HT29 spheroids compacted efficiently and exhibited a round morphology (between 50-75 µm in diameter), whereas PANC1 spheroids tended to have a less homogeneous morphology and were larger (between 100-150 µm), as shown in figures 3(A) and (C).We monitored neurites as they traversed the barrier and innervated axonal targets following application of NGF to the distal compartment (NGF −/+).No cancer-mediated effects on neuronal outgrowth were observed compared to the NGF −/+ control condition (figures 3(A) and (B)).This suggests that NGF in the media of the axonal compartment may saturate the outgrowth potential of nerve fibers and override guidance clues.Therefore, we opted to omit NGF from the cocultures (NGF −/−) and were then able to demonstrate a cancer-mediated increase in nerve fiber density in the axonal compartment of HT29-DRG cocultures (figures 3(C) and (D)).Remarkably, we were able to simultaneously demonstrate innervation and complete envelopment of individual cancer spheroids, as observed in vivo [16,19].3D representations of nerve fibers extending into the axonal compartments among resident cancer spheroids further confirm that the described cell-cell interactions occur in all dimensions (figures 3(E) and (F) and movies S4, S5, S6 and S7).Furthermore, for optimal comparison of neuronal outgrowth conditions and standardization, cancer cells were cultivated in neuronal media instead of cancer media without clear negative effects on viability of the cancer spheroids (SI appendix, figures S3(A) and (B)).Interestingly, while HT29-DRG cocultures exhibited a significant enhancement of neuronal outgrowth compared to control conditions (DRG alone, NGF −/−), this effect was not observed in PANC1-DRG cocultures (figures 3(C) and (D)).This observation suggests that innervation can be highly cell-type-specific, possibly due to the presence of different molecular or paracrine cues associated to the microenvironment of each cancer type.

Sensory neurons innervating 3D colorectal cancer spheroids can be activated by pain mediators
The combination of electrical and image-based techniques is crucial for investigating neuronal functions.However, current methods for monitoring activity in 3D tissues, such as calcium imaging, are typically limited to qualitative analysis.In this study, we demonstrate the potential of the MEA-integrated INV-MPS for screening purposes in pain research.As shown in figure 3, cancer spheroids occupying the axonal compartment of the SLE INV-MPS can stimulate nerve growth, raising the question of whether nerve terminals also exhibit excitability (Movie S8).To address this, we applied either capsaicin or bradykinin to the axonal compartment of INV-MPSs containing DRG (NGF−/−).with or without cancer spheroids.Prior to chemical excitation, we confirmed that HT29 did not cause any rise in spontaneous activity (SI appendix, figure S4(A)).In the presence of cancer spheroids it was possible to measure characteristic responses to both compounds in axonally extended nerve fibers, which were absent in the control group (figures 4(A) and (B)).Although the responses were less pronounced than those evoked in DRG monocultures (NGF−/+; figures 2(E)-(J)), our data suggests that the inclusion of HT29 cancer spheroids not only drives neurite outgrowth, but it is sufficient to induce nerve terminal excitation in response to chemical stimuli.The response to both depolarizing stimuli was quantified and compared between DRG and HT29 coculture vs. DRG control (figures 4(C) and (D) and SI appendix, figure S4(B)).The results show that capsaicin application to HT29 coculture induced responses characterized by a less pronounced initial cumulative peak and lower responsiveness compared to NGF−/+ conditions (figures 2(I) and (J)), which is consistent with the difference in neurite outgrowth (<50%).Deeper analysis, including time-course and number of responses elicited by capsaicin or bradykinin applied to DRG (NGF−/−) alone or cocultured with HT29 cancer spheroids, revealed that colorectal cells have the capacity to enhance both outgrowth of 3D nerve terminals and transduction of noxious stimuli into action potentials.To explore a more direct association between nerve terminal excitability and cancer innervation, we employed AAV-GCaMPf6 to directly visualize calcium transients induced by application of chemical agonists in DRG neurons innervating cancer spheroids (figure 4(E) and movies S9 and S10).Although Ca 2+ imaging can indeed provide an optical measure of nerve terminal activation, monitoring responses in a multiplexed 3D setting remains technically challenging and time consuming, which emphasizes the advantages of the INV-MPS platform for quantifying electrical excitability in 3D microtissues.

Discussion
The MPS-field aims to replicate human organ functions not properly mimicked by conventional cell cultures or animal models [29].However, MPSs for the study of organ innervation are currently not available due to the added complexity of incorporating nerve fibers, despite the potential benefits for therapeutic domains such as visceral pain, metabolic and cardiovascular disease and neurological disorders [30].Novel designs and platforms are required to integrate peripheral neurons and target tissues, whilst providing a robust electrophysiological readout that is consistent with neuronal activity.By accurately monitoring and controlling electrical activity of innervating fibers, these will aid the understanding of underlying mechanisms that drive peripheral neuron-tissue interaction leading to pathological phenotypes.Leveraging expertise in combining custom MEA, high-resolution laser glass etching and 3D neuronal cultures [10], we developed an innervation platform that meets the requirements of the MPS-field towards creating intricate tissue models and incorporates a reliable electrophysiological readout.Unlike conventional 2D axonal outgrowth models [31], the INV-MPS is used to closely mimic 3D in vivo-like nerve-tissue interplay and simultaneously overcomes the limitations of conventional MEA systems solely designed for recording from 2D cultures.This is achieved by integrating CMEs that allow electrophysiological recording from 3D peripheral nerve fibers innervating hydrogel-embedded target tissues.Distinct from alternative methods such as patch-clamp, our approach presents several unique features, including non-invasive readout, enhanced throughput, simple handling, and seamless integration across platforms.Notably, the INV-MPS avoids the utilization of patterned polymers such as PDMS, as it is susceptible to small molecule absorption and requires a time-consuming molding procedure, among other considerations.Instead, we designed an easy-to-assemble, reusable device utilizing a glass microfluidic chip and MEA to compartmentalize and record from cultured sensory neurons, effectively ensuring the required separation of soma and terminals through barrier-etched microtunnels [5].Axonal growth through the barrier was shown to be strongly dependent on a NGF gradient across the 3D hydrogel, closely mimicking physiologically relevant nerve guidance [32].This was further demonstrated in the NGF +/+ condition, where the externally administered large neurotrophin pool above the DRG soma serves as a primary cue for the growth of neurites in this compartment.This overrides the growth toward the axonal compartment, despite the presence also here of NGF, as neurites would need to extend through the barrier to reach this compartment.We demonstrated that our platform enables non-invasive measurement of the electrical activity of 3D sensory terminals and that the response profiles to pain mediators are similar to those observed in vivo and ex vivo [33,34].For instance PGE 2 , a component of the IS, exerts a crucial influence on the sensitization of nociceptive terminals in the NGF −/+ condition, impacting their excitability.Notably, both in vivo and in vitro studies consistently demonstrate that PGE 2 enhances the sensitivity of nociceptive neurons [35].Overall, these results suggest that the INV-MPS can provide a physiological relevant measure of the engagement of receptors and ion channels localized at distinct locations (terminal vs. somatic) potentially mediating nociception.Simultaneously, it allows for the monitoring of neurotrophic factors/inhibitors' effects on neuronal outgrowth and arborization, providing valuable insights into these processes.Next, we aimed to apply INV-MPS' unique readout capabilities to the study of disease-related phenotypes by exploring CIP.Our findings indicate that when cancer spheroids and DRG neurons are cocultured in the presence of NGF, neural outgrowth is primarily driven by the exogenously applied neurotrophin in the axonal compartment.Remarkably, we observed that when NGF is omitted, the presence of colorectal cancer spheroids is sufficient to stimulate neural outgrowth and ultimately the innervation of individual cancer structures, as observed in vivo [16,19].This emphasizes the complex interplay between cancer and the PNS, highlighting the need to consider the impact of exogenously applied NGF or other growth factors when studying nerves and nerve-tissue crosstalk in vitro.Understanding the role of released factors is crucial to elucidate underlying mechanisms of chronic pain and identify potential therapeutic targets for the management of CIP.An increase in nerve fiber density could be demonstrated for colorectal cancer spheroids (HT29) but not for pancreatic cancer spheroids (PANC1).However, it would be premature to conclude that pancreatic cancer cannot induce outgrowth within the INV-MPS setting, as induction of axonal growth can have varying degrees of dependency on tumor-associated stromal cells and inflammation [16,[36][37][38][39][40], only partially recapitulated here.Neuronal outgrowth was clearly shown to depend on exogenously applied NGF, suggesting a potential correlation between the observed cancermediated outgrowth and NGF presence.However, given that both HT29 and PANC1 release comparable amounts of NGF in vitro [41,42], the observed disparity in innervation between these two cancer types poses that NGF release alone may not be sufficient to drive outgrowth in this specific in vitro context.Expanding on this by investigating putative paracrine mediators and exploring alterations in gene expression and regulation of relevant protein levels is beyond the scope of our current study.However, the observed difference in 3D innervation between colorectal and pancreatic cancer spheroids emphasizes the ability of the INV-MPS to accurately recapitulate and detect differences in cancer-specific environments.This was further assessed by measuring the electrophysiological response to prototypic noxious and inflammatory stimuli in innervating fibers of HT29-DRG neuron cocultures.Here, nerve terminals extending into the axonal compartment and enveloping cancer spheroids were found to respond to both capsaicin and bradykinin.Measured responses were remarkably greater (duration and frequency) than those observed in cultures without cancer and exogenously applied NGF.This was complemented by live-imaging of evoked calcium transients occurring in 3D cancer-innervating fibers, providing a direct qualitative evaluation of tumor innervation and neuronal activity.Having the ability to additionally image nerve terminals and cancer cells in direct contact suggests that the INV-MPS holds potential as a valuable multimodal platform for investigating the reciprocal influences between cancer and closely associated innervating fibers.While the primary objective of this study was to validate the INV-MPS by examining the impact of cancer microtissues on sensory innervation and excitability, emerging evidence indicates that peripheral neurons may additionally influence cancer homeostasis and dissemination [43], making it a promising avenue for future exploration using this in vitro model.The potential applications of the INV-MPS extend beyond CIP and cancer, as this model offers the opportunity to investigate a diverse array of diseases by faithfully reproducing 3D peripheral innervation and measuring excitability in various tissues.In conclusion, the INV-MPS is purposefully designed to facilitate the exploration of complex in vitro biology, presenting great potential for researchers in the MPS field who seek to understand the intricate interactions between the PNS and innervated organs.

Animals
Wild-type Swiss mice obtained from Janvier Labs (France) were housed in the animal facility of the institute.Mice were kept under standard conditions with controlled temperature and a 12-hour light-dark cycle, with ad libitum access to food and water.All experimental procedures described below were performed in compliance with the regulations outlined in the European Union (EU) legislation for the care and use of laboratory animals (Directive 2010/63/EU of the European Parliament on the protection of animals used for scientific purposes) and the German TierSchG (Tierschutzgesetz) with the latest revision in 2019.

Microfabrication
INV-MPS devices were prepared following the protocol described previously [10], with modifications.Briefly, compartmentalized microfluidic chips (see SI appendix, figure S1(A)) were fabricated by SLE using fused silica.MEAs with integrated CME were produced by structuring two layers of permanent epoxy photoresist on gold microelectrodes.SU-8 2002 (MicroChem, USA) was spin-coated to achieve a thickness of approximately 3.0 µm (10 s at 500 rpm, then 30 s at 1000 rpm) (see SI appendix, figure S1(B)).ADEX TDFS A20 (Micro Resist Technology GmbH, Germany) was then laminated onto the SU-8 layer using a pouch laminator (GMP Photonex@325, EF02015) at a temperature of 75 • C. Microfluidic chips were bonded to the MEA using EPO-TEK 301-2FL (Epoxy Technology) after alignment with a Fineplacer® lambda (Finetech GmbH, Germany).The bonded devices were cured at 80 • C for 3 h.
Prior to use, the devices were rinsed with bi-distilled water and incubated in a 1% Tergazyme solution for 3 h.Subsequently, they were washed overnight with bi-distilled water.This cleaning process was repeated between experiments if the devices were deemed suitable for reuse.Alternatively, the devices were detached from the MEA substrate by immersion in concentrated sulfuric acid (H2SO4 ROTIPURAN® 98%-Carl Roth, # H290-H314) overnight and then reattached to a new MEA substrate.SLE INV-MPS devices do not acquire additional assembly as they are created from a monolithic fused silica substrate and lack microelectrodes.

Fused silica microfluidic chip production
Fabrication of fused silica microfluidic chips is based on a subtractive, direct-write microfabrication process.Exploiting a femtosecond laser and etching creates integrated 3D micro-systems with micrometer precision [44].The process exploits highly localized material modifications, triggered by non-linear absorption during the laser exposure step.The size of the modified region is in the order of 2.0 µm in XYaxis and from 10.0 µm to 100.0 µm in Z-axis, depending on the optics used to focus the laser.These modifications cause a local increase of 'etching rate' in the exposed material.Since modification happens only within the focal spot (voxel) e.g., buried structures or free-form surfaces can be produced.The exposed material is later etched in KOH based solution to create 3D devices.

Profilometry
Barrier integrated microtunnels were analyzed using a 3D optical profiler (Contour GTK-A, Bruker UK Ltd) equipped white light illumination and a 10x magnification objective.The instrument provides lateral resolution of 1.0 µm and Z resolution below 10.0 nm.Stylus profilometry (DektakXT, Bruker UK Ltd) was performed to characterize CME height.Both SU8 (no ADEX cap) and SU8-ADEX (with cap) were measured by using a 2.0 µm tip.Both sets of measurements were processed and analyzed using dedicated Vision64 software (Bruker UK Ltd).

3D DRG neuron culture
DRG neurons were isolated and dissociated by modification of a previously published protocol [45].
Postnatal mice (P4-6) were sacrificed by decapitation and a narrow incision through the skin and muscle of the back exposed the spine, after which the ganglia could be removed for preservation in Hibernate™-A Medium (Thermo Fisher, #A1247501).Cleaned ganglia were then incubated in enzymatic solution(s) containing collagenase (IV) (Worthington, #LS004186) and dispase (Worthington, #LS02109), followed by manual dissociation in DNAse solution (Worthington, #LS002139

Research animals ethics statement
All animal research procedures were conducted in accordance with the EU legislation for the care and use of laboratory animals (Directive 2010/63/ EU) and the German Animal Welfare Act ('Tierschutzgesetz' , 2019).

Figure 1 .
Figure 1.A novel MPS with integrated electrodes for 3D axon manipulation and recording from sensory neuron cultures.(A) Recapitulating the innervation of soft tissues by the peripheral nervous system.3D-surface rendering of DRG terminals extending into 3D space in the axonal compartment.Dimensions 600 × 450 × 300 µm.(Scale bar: 100 µm).(B) Both somatic and axonal compartments are filled with a soft 3D scaffold inside the gel layer (GL).Supporting media is added on top in both media layers (ML) (width: 300.0 µm).(C) High magnification color map of micro tunneled barrier profilometry data.(D) DRG neurons grow neurites in 3D through the barrier while simultaneously projecting neurites into recording CMEs residing in the somatic compartment.(E) Live-cell confocal images (max projection) of GFP-transduced DRG homogeneously distributed across the somatic compartment with neurites traversing the barrier and entering CMEs (scale bar: 75 µm).(F) INV-MPS holds 9 identical cocultures with 13 CMEs each and assembles by bonding on top of the dedicated 120-electrode MEA.(G) Technical drawings of MEA-CME production photomasks.

Figure 2 .
Figure 2. 3D axonal guidance and nerve terminal excitability modulated by NGF inside the MPS.(A) Live-cell confocal images of GFP-transduced DRG neurons protruding from the barrier in 3D after 5 d in culture.Neurobasal-A culture medium either contains no NGF (−/−), NGF in both somatic and axonal media compartment (+/+) or NGF only in the axonal media compartment (−/+) (scale bar: 75 µm).(B) Quantification of neurite outgrowth after 5 d in culture.Total number of neurites was measured at 100 µm, 300 µm, and 500 µm from the barrier.Bar graphs represent mean ± SEM (n = 4 wells) (C) micrograph shows diffusion of Dextran-FITC (MW: 4 kDa) or Dextran-Rhodamine B (MW: 20 kDa) through the micro-tunnel barrier from axonal (right) to somatic (left).Fluorescent intensity profiles generated from ROIs taken 5 µm from the micro-tunnel barrier, plotted against time as percentages (normalized to intensity at T = 12 h).Data is represented as mean ± SEM (n = 3 wells).(D) Agonists are applied to the media layer atop the 3D nerve terminals of a NGF −/+ culture in the axonal compartment.3D-surface rendering of DRG neurons and their soma.Dimensions 1450 × 750 × 300 µm.(Scale bar: 300 µm).(E) Trace plots representing the activity recorded by 13 CMEs after application of 1 µM Capsaicin (black asterisk) at different time scales.(F) Trace plots comparing control to activity recorded 5 min after application of inflammatory soup (IS containing: 3.0 µM Bradykinin, 3.0.µM Prostaglandin, 3.0.µM Serotonin and 3.0 µM Histamine).(G) Temporal dynamics of Capsaicin (1 µM) responses in −) rcultures of DRG neurons after 4 (blue), 5 (grey) and 6 (red) days.Dots represent the mean spikes per second per well over a 60-second span post-application (n = 3-4 wells).(H) Bar graphs comparing the development of nociceptive phenotype (percentage of responding electrodes) and total spikes recorded per well in the 60-second span post application at different time points (time in culture).(I) Dose-response plots comparing temporal dynamics following application of 0.1 (blue), 0.3 (grey) and 1.0 (red) µM capsaicin in −/+ cultures of DRG neurons (n = 3-4 wells).(J) Bar graphs comparing CMEs excitability probability (percentage of responding electrodes) and total spikes recorded per well in the 60-second span post application following application of different capsaicin's concentrations.In B, H, and J, significant differences between mean values of groups are displayed as asterisks: * p < 0.05; * * p < 0.01; * * * p < 0.001; * * * * p < 0.0001; Brown-Forsythe and Welch's ANOVA test.

Figure 4 .
Figure 4. Sensory neurons innervating 3D colorectal cancer spheroids can be activated by pain mediators.(A) Trace plots representing individual CMEs (1-13) and cumulative raster after application of 1.0 µM Capsaicin (black asterisk) to a DRG-only culture (NGF −/−) after 6 d. (B) Trace plots and cumulative raster after application of 1.0 µM Capsaicin or 1.0 µM Bradykinin to DRG neurons innervating HT29 cancer spheroids (NGF −/−) after 6 d. (C) Temporal dynamics of capsaicin responses in DRG neurons innervating HT29 cancer spheroids (NGF −/−).Dots represent the mean spikes per second per well over a 60-second span post-application (n = 10-11 wells).Bar graphs comparing excitability of DRG neurons responding to axonal stimuli (percentage of responding electrodes) and total spikes recorded per well in the 60-second span post application.(D) Temporal dynamics of bradykinin responses in NGF −/− cultures of DRG neurons innervating HT29 cancer spheroids.Dots represent the mean spikes per second per well over a 120-second span post-application (n = 3 wells).Bar graphs comparing excitability of DRG neurons responding to axonal stimuli (percentage of responding electrodes) and total spikes recorded per well in the 120-second span post application.(E) Live-cell confocal images of GCaMP6f-transduced DRG neurons innervating HT29 cancer spheroids after 6 d in a NGF −/− culture.Application of 1.0 µM Capsaicin and 4.0 µM Bradykinin induces Ca 2+ transients (also seen in movies S9 and S10).In C and D, significant differences between mean values of groups are displayed as asterisks: * p < 0.05; * * p < 0.01; * * * p < 0.001; * * * * p < 0.0001; Brown-Forsythe and Welch's ANOVA test.
Ca 2+ indicator was imaged before and immediately after application of inflammatory/excitatory compounds using a spinning disk Zeiss Cell Observer® System (20x/63x objective) and processed in FIJI.At 5-6 d in culture, cocultures were placed inside the 384 well microplate adaptor and left to acclimate.ROI (350 µm × 350 µm) was selected by qualifying the presence of innervated cancer spheroids at various heights within the gel layer before stimulation with either 4 µM Bradykinin (Tocris/Biotechne, #3004) or 1 µM Capsaicin (Tocris/Biotechne, #0462).Acquisition time was set at ∼2 frames per second.Electrophysiology of 3D innervating fibers was recorded using a USB-MEA 120-System (Multi Channel Systems MCS GmbH, Germany) at 4-6 d in culture for DRG monocultures and at 6 d in culture for cancer cocultures.During experiments, a customized MEA incubation chamber (Okolab Srl, Italy) was used to continuously monitor temperature, CO2 and humidity.The temperature was set to 36 • C, humidity at 85% and CO2 at 5%.MEAs were allowed to acclimate inside the MEA chamber before applying agonists and inflammatory mediators to the axonal compartment.Agonists were applied 30 s into a recording to identify spontaneously active electrodes.Inflammatory mediators were applied 5 min before recording.Recordings were acquired using MC_Rack v. 4.6.2(Multichannel Systems MCS GmbH, Germany) and filtered using a fourth-order bandpass filter (60-6000 Hz), before processing with NeuroExplorer (version 5.300) for spike detection and further analysis.Statistical analysis was performed on the collected data using GraphPad Prism 8 developed by GraphPad Software Inc. (Boston, USA).Degree of normality of the data distribution was assessed both visually and quantitatively by employing quantilequantile plots, frequency histograms, and Shapiro-Wilks tests.Within the neurite outgrowth assay domain, conditions are represented as bar graphs with mean ± SEM.Differences were assessed by Brown-Forsythe and Welch's ANOVA test with posthoc test Dunnett T3 (n < 50) for comparison of means.Electrophysiological data were represented as either bar graphs comparing conditions or connected-points for temporal dynamics with mean ± SEM.Differences were assessed similarly by either Brown-Forsythe and Welch's ANOVA with post hoc test Dunnett T3 (n < 50) or Welch's unpaired t-tests for comparison of means.Asterisks denote statistical significance as follows: * p < 0.05; * * p < 0.01; * * * p < 0.001; * * * * p < 0.0001.2020 Framework Programme for Research an Innovation under the Marie Sklodowska-Curie Grant Agreement No. 814244 (BONEPAIN II).P C acknowledges funding from Eureka's Eurostars Programme under grant Agreement 115217 (NEUROCHIP).FE acknowledges funding from the Ministry of National Education of Turkey.