Three-dimensional neuroimmune co-culture system for modeling Parkinson’s disease microenvironments in vitro

Parkinson’s disease (PD) is a complex and multifaceted neurodegenerative disorder that results from multiple environmental factors and multicellular interactions. Although several PD neuropathologies have been identified and described, the thorough understanding of PD pathophysiology and research has been largely limited by the absence of reliable in vitro models that truly recapitulate PD microenvironments. Here, we propose a neuroimmune co-culture system that models PD neuropathologies by combining relevant multicellular interactions with environments that mimic the brain. This system is composed of: (i) 3D bioprinted cultures of mature human dopaminergic (DA) neurons grown on extracellular matrix (ECM)-derived scaffolds doped with electroconductive nanostructures, and (ii) a direct co-culture of human astrocytes and differentiated monocytes that models neuroinflammatory responses. When co-cultured in a transwell format, these two compartments recreate relevant multicellular environments that model PD pathologies after exposure to the neurotoxin A53T α-synuclein. With immunofluorescent staining and gene expression analyses, we show that functional and mature DA 3D networks are generated within our ECM-derived scaffolds with superior performance to standard 2D cultures. Moreover, by analyzing cytokine secretion, cell surface markers, and gene expression, we define a human monocyte differentiation scheme that allows the appearance of both monocyte-derived macrophages and dendritic cell phenotypes, as well as their optimal co-culture ratios with human astrocytes to recreate synergistic neuroinflammatory responses. We show that the combined response of both compartments to A53T α-synuclein stimulates the formation of intracellular α-synuclein aggregates, induces progressive mitochondrial dysfunction and reactive oxygen species production, downregulates the expression of synaptic, DA, and mitophagy-related genes, and promotes the initiation of apoptotic processes within the DA networks. Most importantly, these intracellular pathologies were comparable or superior to those generated with a rotenone-stimulated 2D control that represents the current standard for in vitro PD models and showed increased resilience towards these neurotoxic insults, allowing the study of disease progression over longer time periods than current models. Taken together, these results position the proposed model as a superior alternative to current 2D models for generating PD-related pathologies in vitro.


Introduction
Parkinson's disease (PD) is the second most common neurodegenerative disorder worldwide, affecting 1.5%-2% of adults over 60 years of age [1]. It is known for the death of dopaminergic (DA) neurons in the Substantia Nigra pars Compacta (SNpC), a midbrain region responsible for dopamine synthesis and clearance, followed by dysfunctional dopamine circuits in the striatum [2]. This progressive and irreversible loss of DA neurons affects how neurons talk to each other and causes both cognitive (such as depression, anxiety, and dementia) and motor symptoms (such as bradykinesia, tremors, and rigidity) [2,3], though these usually show up decades after the cellular neuropathology started [3]. Physiological alterations that result in neuronal DNA and an increase in oxidative stress and chronic neuroinflammation play a pivotal role in disease progression and have been closely associated with key PD neuropathological hallmarks. Namely, dysregulations in the production and folding of the presynaptic protein α-synuclein (α-syn) induce the formation of aberrant soluble oligomeric conformations, termed protofibrils, that exert neurotoxicity by multiple mechanisms [4]. Besides accumulating in neurite projections and disrupting synaptic function, α-syn aggregates also accumulate throughout the soma within cytoplasmic inclusions, termed Lewy bodies, and interfere with several intracellular pathways [5]. This results in nuclear and mitochondrial dysfunction, disruption of intracellular trafficking, and obstruction of autophagy/lysosomal routes [6], each of which also promote severe oxidative damage and may ultimately lead to neuronal death [7].
However, although α-syn dysregulation is one of the most studied and recognized propagators of PD, there is a body of evidence suggesting that, in certain cases, the α-syn-related pathology may be an epiphenomenon and not causally related to neuronal dysfunction and death [5]. This is supported by the fact that the toxic effects elicited by α-syn protofibrils also favor further α-syn aggregation in a feedforward manner [8]. For instance, mitochondrial dysfunction leads to the overproduction of reactive oxygen species (ROS), which create an oxidative environment where α-syn aggregates are prone to form. Similarly, neuroinflammatory environments inherently favor excessive oxidative stress and thus the accumulation and secretion of α-syn aggregates, which further trigger pro-inflammatory pathways upon their uptake by resident and peripheral immune cells [9]. In fact, these microenvironmental factors and intracellular pathologies are thought to be equally important to unravel a neuropathological outcome, beyond their causal relation to α-syn aggregation [9].
Accordingly, the multifactorial nature of PD has considerably challenged our understanding of the underlying mechanisms that dictate its pathogenesis. Specifically, more than 90% of PD cases fail to exhibit a common genetic background and, thus, their origin remains unknown [10,11]. This can be highly correlated with the current inability to reliably study the disease progression in a high-throughput manner. In this regard, most in vitro studies focus on DA neurons alone and under non-physiological adherent conditions (i.e. 2D monolayers), ignoring microenvironmental factors that influence disease progression [12][13][14][15][16]. Animal disease models, despite presenting the most comparable pathologies to PD in humans [17], are highly limited by their low throughput, as well as by species-specific physiological differences. Also, studying patient-derived xenografts in immunocompromised animals makes it much harder to study how the immune system interacts with the brain as PD gets worse [18]. Additionally, human clinical studies rely solely on non-invasive analysis techniques that provide limited molecular insights, or post-mortem tissue analyses, where most of the relevant cells have already died. Reliable in vitro models would, therefore, be an extremely valuable tool for the thorough study of PD pathophysiology at the molecular level, as well as for the high throughput screening of potential therapeutic candidates with higher confidence.
Organoids, which are self-organizing 3D coculture systems, have become the gold standard in the field of in vitro tissue models because they can closely mimic the different types of cells that make up tissues in 3D cell aggregates [19,20]. Most of the time, these aggregates are made by directing induced pluripotent stem cells (iPSCs) in a way that mimics embryonic development toward certain lineages. Accordingly, midbrain organoid models have become especially promising for PD research, since they allow the joint study of multicellular responses to PD-related stimuli within tissue-relevant in vitro structures [21,22]. However, organoid modeling is costly, extremely laborious, highly time consuming (given the long differentiation periods required), and show highly variable cellular compositions and structures due to their uncontrollable self-assembly [23].
In an attempt to circumvent these limitations, we propose the use of extrusion-based 3D bioprinting for the biofabrication of midbrain DA constructs that model cellular and microenvironmental dependencies relevant to the development of PD pathologies. In particular, we differentiate midbrain neural progenitor cells (NPCs) into DA neurons within extracellular matrix (ECM)-derived hydrogels doped with electroconductive nanostructures (i.e. graphene oxide (GO)). This approach allowed us to obtain dense DA networks with improved scalability and reproducibility, given the ease for tuning relevant biological parameters such as cellular composition and extracellular stiffness [24,25]. Moreover, maturation periods can be considerably reduced with respect to organoid-based approaches due to the closer lineage relationship between NPCs and DA neurons. In addition, we incorporate neuroimmune interactions to the described model by co-culturing neural cores with astrocytes and monocyte-derived macrophages, thereby obtaining a multicellular system involving both resident and peripheral immune inputs which are key for the development of PD-like pathologies. Accordingly, we propose the induction of these pathologies upon exposure to a mutant isoform of α-syn (i.e. A53T α-syn), whose genetic link to PD and higher aggregation potential in solution [26] favors neurotoxic responses within the DA core and elicits neuroinflammatory responses from co-cultured immune cells that propagate pathologies related to PD. With this method, we present a comprehensive model that lets us study multicellular responses to a complex PD-mimicking environment that mimics the main pathological hallmarks of this disease. This brings us one step closer to reliable PD research in a lab setting.

SISMA-GO hydrogel preparation
Porcine small intestines were decellularized as previously reported [27,28]. Briefly, the jejunum was thoroughly washed with tap water and the submucosa layer was isolated by mechanically removing the tunica mucosa and serosa muscularis layers. This layer was then decellularized with a solution of hydrogen peroxide, sodium hypochlorite and autoclaved type II water for 15 min under constant agitation, followed by successive washes with sterile 1X phosphate buffered saline (PBS) and autoclaved type II water to remove cellular and chemical remnants. Finally, washed membranes were dried overnight under laminar flow and subsequently pulverized with cryogenic grinding in a freezer mill (6875 Freezer/Mill, SPEX SamplePrep, Metuchen, NJ, USA).
ECM-derived hydrogels were prepared as described previously [27]. Briefly, dry small intestinal submucosa (SIS) powder was solubilized in 0.5 M acetic acid with 0.1% (w/v) pepsin at a 0.4% (w/v) concentration for 48 h. The obtained SIS pre-gels were then functionalized for 24 h with MA, using EDC and NHS as zero-length crosslinkers, and subsequently dialyzed for 48 h against 0.25 M acetic acid with medium changes every 12 h. Functionalized pre-gels (methacryloyl-modified decellularized small intestinal submucosa (SISMA)) were then lyophilized for 48 h at 0.14 mbar and sterilized by ethylene oxide treatment. Finally, sterilized SISMA was resuspended in 0.02 M acetic acid and mixed in a 1:1 (v/v) proportion with a working solution of Advanced DMEM:F12 supplemented with 1X N2 supplement, 2 mM L-glutamine, 100 mM Tris-HCl and 3 mg ml −1 LAP. For the electrosensitive SISMA-GO hydrogel, GO nanoflakes (synthesized as previously described in [27]) were pre-treated for 1 h by submersion in a 10 µg ml −1 solution of fibronectin [29] and subsequently added at a 0.5 mg ml −1 concentration in the working solution. The resulting hydrogels were transferred to 3 ml sterile syringes and stored under refrigeration at 4 • C and protected from light.

Mechanical testing
The mechanical stiffness of the SISMA and SISMA-GO hydrogels before and after blue light irradiation was assessed via time sweep experiments performed on a Discovery Hybrid Rheometer-1 (TA Instruments, New Castle, DE, USA). The concentrations evaluated ranged between 10 and 25 mg ml −1 for both hydrogels, and all assays were performed under oscillatory mode with 1% constant strain and 10 rad s −1 frequency. Blue light irradiation was performed using the 405 nm photocuring module included in the Bio X bioprinter (Cellink AB, Gothenburg, Sweden) at an irradiance of 34.7 mW cm −2 for 5 min.

Cell culture
Lund human mesencephalic (LUHMES) NPCs were purchased from Applied Biological Materials Inc. (T0284, ABM, Richmond, BC, Canada). Prior to cell seeding, all culture vessels were pre-coated with 50 µg ml −1 poly-l-ornithine and 1 µg ml −1 fibronectin in sterile type I water overnight. Cells were maintained in complete growth medium consisting of Advanced DMEM: F12 with 1X N2 supplement, 2 mM L-glutamine and 40 ng ml −1 bFGF under a humidified atmosphere at 37 • C and 5% CO 2 . Media was changed every other day. Cells were passaged before 80% confluency was reached.
The 2D differentiation of LUHMES NPCs was carried out following an enhanced two-step protocol proposed by Harischandra et al (figure S1) [30]. Briefly, cells were seeded at a 50 000 cells cm −2 density in pre-coated flasks and maintained in complete growth medium for 24 h. Then, media was changed to a freshly prepared differentiation medium composed of Advanced DMEM: F12 with 1X N2 supplement, 2 mM L-glutamine, 1 mM dibutyryl cAMP, 1 µg ml −1 tetracycline, 20 ng ml −1 glial derived neurotrophic factor (GDNF), 20 ng ml −1 brain derived neurotrophic factor (BDNF), 20 ng ml −1 transforming growth factor βIII (TGF-βIII), 10 ng ml −1 leukemia inhibitory factor and 0.2 mM ascorbic acid. After 48 h, cells were dissociated with 0.025% trypsin/EDTA and replated on pre-coated plates at a cell density of 1.5 × 10 5 cells cm −2 in differentiation medium. Differentiation media was changed every other day from then on. Differentiation markers were assessed on day 7 [30], when mature DA phenotypes are already obtained.
THP-1 cells (TIB-202™, American Type Culture Collection, Manassas, VA, USA) were maintained in RPMI-1640 medium supplemented with 10% (v/v) FBS and 0.05 mM b-mercaptoethanol under a humidified atmosphere at 37 • C and 5% CO 2 . Macrophage differentiation was induced upon incubation with 50 nM or 200 nM of PMA in RPMI-1640 medium supplemented with 10% (v/v) FBS for 48 h, followed by a resting period of 48 h with regular culture medium.
NHA cells were purchased from Lonza (Cat. # CC-3565, Basel, Switzerland) and maintained in the complete growth medium recommended by the manufacturer (AGM Astrocyte Growth Medium BulletKit, Cat. # CC-3186, Lonza, Basel, Switzerland) under a humidified atmosphere at 37 • C and 5% CO 2 .

3D bioprinting and differentiation of LUHMES NPCs
Prior to bioprinting experiments, the Bio X bioprinter (Cellink AB, Gothenburg, Sweden) was thoroughly wiped with 70% ethanol and placed inside a biosafety cabinet, where all subsequent steps were performed. Next, the clean chamber protocol was activated, and the temperature-controlled printhead was set to 13 • C, while the printbed was set to 37 • C. This preconditioning was required to accelerate the bioprinting process once the bioinks had been prepared.
Bioinks at a final concentration of 1 × 10 7 cells ml −1 were prepared by mixing a concentrated suspension of LUHMES NPCs and the SISMA or SISMA-GO hydrogels at a 1:10 proportion with the aid of 3 ml sterile syringes and a female-tofemale Leuer lock. To prevent unwanted temperatureinduced gelation, bioinks were subsequently transferred to UV-shielding cartridges and mounted on the temperature-controlled printhead of the bioprinter. After 5 min of temperature conditioning, cylindrical constructs (D = 6 mm, H = 2 mm) were bioprinted on 24-well plate polycarbonate inserts (Cat. # CLS3399, Corning®, Corning, NY, USA), using 20 G conical nozzles and an extrusion pressure of 8 kPa, and subsequently irradiated for 1 min as described in section 2.3. Immediately after crosslinking, differentiation medium was added to the apical and basal chambers of the well plate inserts to initiate the 3D differentiation scheme (figure S1). Bioprinted constructs were incubated under a humidified atmosphere at 37 • C and 5% CO 2 and the culture medium was changed on days 2, 5 and 7.

Metabolic activity of bioprinted constructs
For evaluating the metabolic activity of the bioprinted DA neuron constructs as a direct indicator of cell survival, the Alamar Blue cell viability assay was performed at days 0, 3 and 7 after bioprinting. Briefly, the Alamar Blue reagent was added to the culture medium on each well at a final concentration of 10% (v/v) and left for incubation at 37 • C and 5% CO 2 for 4 h. After incubation, media was collected from each well and 100 µl were transferred to an opaque 96well plate. Culture medium with Alamar Blue reagent in the absence of cells was also incubated and used as background control. Fresh differentiation media was added to assayed constructs to continue their maturation, and metabolic activity readings were performed on the same constructs throughout the assayed time period. The fluorescence of collected media was measured using an excitation of 570 nm and an emission at 585 nm with a FluoroMax spectrofluorometer (Horiba, Kyoto, Japan). To calculate the fold change in metabolic activity, background fluorescence was subtracted and subsequently normalized to medium fluorescence of the respective construct at day 0. This assay was performed in triplicate.

A53T α-synuclein-induced inflammation
Both undifferentiated and differentiated THP-1 monocytes were exposed to the A53T α-syn mutant to assess their inflammatory response. For the former, THP-1 monocytes were seeded at a 2 × 10 5 cells ml −1 density in 12-well plates and stimulated with 10 µg ml −1 of A53T α-syn for 24 h. For the latter, THP-1 monocytes were differentiated at the same density as described above and, at day 4, were stimulated with either 5 or 10 µg ml −1 of A53T α-syn. High cell viability after 24 or 72 h of stimulation was confirmed with MTT and LDH-leakage assays to validate the used A53T α-syn concentration range. Both assays were performed on 96-well plates. For MTT assays, cells were incubated with 0.5 mg ml −1 of the MTT reagent for 2 h and subsequently lysed with DMSO. Absorbance was measured at 595 nm. For LDH-leakage assays, 50 µl of culture media were collected and incubated with the LDH reagent for 30 min at room temperature. Absorbance was measured at 490 nm. Unstimulated cells were used as negative control (C − ) and 10% (v/v) Triton X-100-lysed cells were used as positive control (C + ). The percentage of metabolic activity and LDH leakage was determined according to equations (1) and (2), respectively. A s , A C − and A C + are the absorbances of the sample, negative control and positive control Due to optimal viability results, these stimulation schemes were used to evaluate cell surface marker expression, mRNA expression and cytokine secretion, which were measured at 0 and 24 h after stimulation.
For co-culture studies, THP-1 monocytes were seeded at 30 000 or 22 500 cells cm −2 in 24-well plates and differentiated with 200 nM PMA as described above. On day 4, confluent NHA cultures were detached with 0.05% Trypsin/EDTA and seeded over differentiated monocytes at either 15 000 or 22 500 cells cm −2 , yielding co-culture ratios of 2:1 and 1:1with the same total number of cells, respectively. Co-cultures were maintained in complete AGM, since differentiated THP-1 cells showed insignificant changes in viability when cultured with this medium, as determined by the MTT assay ( figure  S8). After 24 h, co-cultures were treated with the predetermined A53T α-syn concentration from inflammation experiments in complete AGM. At 0 and 24 h after stimulation, the expression of surface markers and the release of cytokines were looked at for both co-culture ratios. For cytokine secretion assessments, monoculture controls of each cell type at the same seeding densities as in each co-culture scenario were also stimulated and assayed at the same time points.

Quantification of cytokine secretion
Culture supernatants were collected, centrifuged at 5000 rpm for 5 min to precipitate cell debris and assayed immediately. The concentration of 13 cytokines (e.g. IL-1β, IFN-α2, IFN-γ, TNF-α, MCP-1 (CCL2), IL-6, IL-8 (CXCL8), IL-10, IL-12p70, IL-17A, IL-18, IL-23 and IL-33) was determined in culture supernatants with the Human Inflammation LEGENDPlex™ Panel (Biolegend, San Diego, CA, USA), a bead-based assay that follows the same basic principle of sandwich immunoassays. Briefly, antibody-conjugated beads were incubated with samples and, after analyte binding, were marked with biotinylated detection antibodies. These beadanalyte-detection antibody sandwiches were then fluorescently labeled with streptavidin-phycoerythrin (PE) to quantify the bound analyte. Analyte-specific beads were differentiated by size and internal fluorescence intensity with the aid of flow cytometry (BD FACSCanto II, San Jose, CA, USA). The obtained results were analyzed with the LEGENDPlex™ Software. Principal component analysis was conducted on the cytokine secretion profiles of each stimulation scenario by evaluating the spatial distribution of the first two principal components with the Python Sklearn library.

PD-mimicking neuro-immune co-culture generation
Differentiated THP-1 monocytes and astrocyte cocultures at 2:1 ratio were first exposed to 10 µg ml −1 A53T α-syn in AGM without serum. After 4 h, transwells with DA neuron constructs differentiated for seven days prior were transferred over immune co-culture wells. Apical medium was changed to complete LUHMES differentiation medium supplemented with 10 µg ml −1 A53T α-syn and basal medium was left unchanged. On day 3, apical and basal media were changed to fresh LUHMES differentiation medium and AGM without serum, respectively, both with 5 µg ml −1 of A53T α-syn. DA constructs were collected on days 1, 2, and 5 and analyzed for apoptosis, ROS production, mitochondrial membrane potential (MMP), and intracellular-syn immunostaining. To increase RNA yield for gene expression analyses, DA constructs and immune co-culture maturation were scaled up to a 6-well plate format. A53T α-syn stimulated DA construct controls (without neuroinflammatory inputs), were exposed to A53T α-syn in LUHMES differentiation medium at both apical and basal chambers.

Rotenone 2D controls
As described in section 2.5, LUHMES NPCs were differentiated in 6-well pre-coated plates. At day 7 of differentiation, cells were stimulated with 0.5 µM rotenone in complete differentiation medium, as recommended by previous rotenone studies with these cells [31]. DMSO levels were less than 0.001% (v/v) at final concentration to avoid unwanted DMSO effects over cell viability. For apoptosis, ROS production, MMP and intracellular α-syn immunostaining cells were directly differentiated and exposed in precoated confocal glass slides to facilitate visualization. Analyses were only performed after one and two days of stimulation, since cell viability was negligible for longer time points.

Immunofluorescent staining
To facilitate visualization with confocal laser scanning microscopy (CLSM), immunofluorescent staining of 2D adherent cultures was performed directly on pre-coated glass bottom dishes for LUHMES. The cells were fixed with 10% (v/v) buffered formalin for 10 min, then washed twice with 1X PBS and permeabilized with 0.1% (v/v) Triton-X 100 for 10 min. Cells were blocked for 30 min at room temperature with 10% (v/v) goat serum in 1X PBS with 0.1% Tween 20 (PBS-T) following two washes with 1X PBS. Samples were then incubated with primary antibodies (e.g. rabbit anti-CCR7, chicken anti-βIIItubulin, mouse anti-α-synuclein or mouse anti-GFAP) at the recommended manufacturer's dilutions in PBS-T with 1% (v/v) goat serum for 1 h. Secondary antibodies (e.g. anti-rabbit IgG Dylight 680, anti-chicken IgY Northern Lights 637, or antimouse IgG Dylight 488) were incubated with cells in PBS-T containing 1% (v/v) goat serum for 1 h. When desired, nuclei were stained in this step with the addition of Hoechst 33342 at a 1:10 000 dilution ratio. If using fluorophore-conjugated primary antibodies (e.g. anti-tyrosine hydroxylase-Dylight 532), this step was skipped. The cells were then washed twice and observed with an Olympus FV1000 confocal microscope (Tokyo, Japan) using laser excitations at the recommended wavelengths for secondary antibodies and 365 nm for Hoechst 33342. To ensure adequate comparisons, five images per sample were acquired with constant settings, and each sample was performed in triplicate. Unless otherwise stated, ImageJ was used to calculate the mean fluorescence intensity (MFI) and Pearson correlation coefficients for a total of 200 manually segmented cells per sample. For immunofluorescent staining of 3D constructs, the same process was performed with slight modifications. Namely, incubations with primary and secondary antibodies were performed for 4 and 2 h, respectively, and washes were performed three times for 5 min each in a rocking platform. Moreover, zstacked images were acquired from 25 adjacent planes with 1 µm distance between them.
The immunofluorescent staining of CD68 for flow cytometry analyses was performed using the same fixation and permeabilization method described above, followed by 20 min of incubation in the dark with a PE-labeled anti-CD68 antibody. Unstained cells subjected to the same fixation and permeabilization procedure were used as control. For CD14 staining of live cells, cells were washed in cold 1X PBS with 0.1% sodium azide and 2% FBS and resuspended in the same solution containing an allophycocyanin (APC)-labeled anti-CD14 antibody for 45 min at 4 • C. Cells were then washed twice and fixed with 2.5% formalin for 10 min. Unstained cells were also used as control. Stained CD68 and CD14 were quantified with flow cytometry at 565 and 652 nm excitation, by analyzing 10 000 cells per treatment. The MFI was determined by subtracting the MFI of the control samples from the MFI of the treated samples.

RNA isolation and Quantitative reverse transcription PCR
RNA was extracted according to the manufacturer's instructions using the Monarch Total RNA Miniprep Kit (New England Biolabs, Ipswich, MA, USA). To prevent trypsin-induced expression changes, adherent cultures were lysed with the lysis buffer directly in culture plates, whereas suspension cells were lysed after centrifugation. RNA was extracted from 3D constructs by manually dissociating the matrices, submerging them in 1 ml of TRIzol reagent, and snapfreezing them with liquid nitrogen for a minimum of 24 h. After samples were thawed, they were homogenized by passing them through 25G needles until complete dissociation. Then, 0.2 ml of chloroform per 1 ml of TRIzol were added and vortexed thoroughly, followed by centrifugation at 10 000 rpm and 4 • C for 15 min. Following the instructions of the aforementioned kit, the aqueous phase was collected, combined with an equal volume of 96% ethanol, and purified. RNA was resuspended in nuclease-free water, and its concentration and quality were determined spectroscopically by ensuring that the ratios 260/230 and 260/280 were greater than 2.0. RNA samples were stored at −80 • C until further use, while avoiding multiple freeze-thaw cycles. RT-qPCR was performed with the Luna Universal One-Step RT-qPCR Kit (New England Biolabs, Ipswich, MA, USA) following the manufacturer's instructions. Processing and signal acquisition were carried out with a Rotor-Gene Q real time PCR cycler (Qiagen, Hilden, Germany). Primer pairs were designed to have melting points around 60 • C and are shown in table S1. PCR runs included an initial 10 min period at 55 • C for reverse transcription, followed by 40 amplification cycles. Fold changes in gene expression were calculated with the 2 −∆∆CT method and are reported as log 2 (fold change). The GAPDH reference gene was used to normalize expression levels.

Apoptosis assay
Apoptotic cells in the assessed PD models were determined with acridine orange (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA) and ethidium homodimer (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA) (AO/EH) staining. Briefly, 2D cultures or 3D constructs were incubated for 10 min at 37 • C with 20 µg ml −1 AO and 4 µM EH in regular culture medium, followed by immediate CLSM visualization with laser excitations at 488 and 620 nm. Apoptotic phenotypes were determined according to the following criteria. Non-apoptotic: homogeneous AO staining without nuclear condensation; early apoptotic: AO staining with clear nuclear condensation and fragmentation, but without EH staining; late apoptotic: dual AO and EH staining of condensed nuclear regions; dead: EH staining with minimal AO staining. Per sample, 200 cells were examined, and each condition was replicated three times.

MMP estimation
MMP of DA neurons in the assessed PD models was determined by incubating 2D cultures or 3D constructs with 2 µM JC-1 dye (Santa Cruz Biotechnology, Dallas, TX, USA) in regular culture medium for 30 min at 37 • C prior to CLSM visualization. Fluorescent images were acquired with laser excitations of 488 and 550 nm. This dye emits green fluorescence in its monomeric form and red fluorescence in its aggregated form, but only aggregates in the presence of energized mitochondria. Accordingly, red and green fluorescence were quantified with ImageJ and red-to-green (R/G) fluorescence ratios were calculated as indicators of mitochondrial polarization. At least 200 cells were analyzed per sample and each condition was performed in triplicate.

Intracellular ROS quantification
The production of intracellular ROS by DA neurons in the assessed PD models was determined with the DCFDA/H2DCFDA-Cellular ROS Assay Kit (Abcam, Cambridge, UK). Briefly, 2D cultures and 3D constructs were washed once with 1X buffer and incubated with 20 µM DCFDA reagent in 1X buffer for 45 min at 37 • C. Due to the high sensitivity of DCFDA to light-induced oxidation, cells were washed with 1X buffer and immediately visualized with CSLM (Ex/Em: 485 nm/55 nm), with special care taken to avoid light overexposure. ImageJ was used to quantify intracellular fluorescence after the manual segmentation of 200 cells per sample, with each condition being performed in triplicate.

Intracellular α-synuclein distribution assessment
The mitochondrial bodies of PD models were stained with 0.5 M Mitotracker Deep Red FM (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA) diluted in regular culture media for 45 min at 37 • C. Then, samples were fixed with 10% (v/v) formalin for 10 min and immunofluorescently stained for α-syn as described in section 2.12, with special care taken to shield each step from light. Nuclei were also stained with Hoechst 334 (1:10 000) at the last incubation step. After manually segmenting each cell, stained samples were visualized with CLSM, and α-synmitochondria and α-syn-nuclei colocalizations were performed in ImageJ. A total of 200 cells were analyzed per sample, and each condition was performed in triplicate.

Results
Due to their high susceptibility to degenerative environments, DA neurons have become the main focus of PD research. However, most pathological hallmarks observed in DA neurons result from the synchronous action of a variety of PD-related microenvironmental factors that go beyond the intracellular level. For instance, the high interconnectivity and functional interdependence of neuronal networks plays an important role in spreading and exacerbating PD pathologies, especially since axonal transport mediates neuron-to-neuron transmission of PD mediators such as α-syn aggregates and dysfunctional mitochondria [32]. Moreover, the PD-induced transition of glial cells (e.g. astrocytes, microglia) from neuroprotective to neurotoxic phenotypes stimulates persistent neuroinflammatory responses that perpetuate characteristic PD neuropathologies such as oxidative stress, α-syn aggregation and mitochondrial dysfunction [33,34]. These responses are further propagated by the infiltration of peripheral monocytes at PD lesion sites in response to early glial reactivity, especially since pro-inflammatory macrophage and dendritic cell phenotypes derived from in site monocytic differentiation are known to be critical orchestrators of inflammatory environments [35][36][37][38]. In fact, recent studies have shown that blocking monocyte entry reduces inflammatory responses and stops neurodegeneration in PD mouse models [39], highlighting the importance of these cells in disease progression. All of these microenvironmental factors must be taken into account when modeling PD microenvironments in vitro, because their contribution, along with the pathophysiology of DA neurons, is important for accurately simulating the progression of the disease.
To address this challenge, we propose a neuroimmune co-culture approach that allows the assessment of relevant multicellular responses to PD-related stimuli (figures 1(A) and (B)). This brain-mimicking co-culture facilitates the paracrine interactions between two main compartments: (i) a bioprinted neuronal core composed of interconnected networks of DA neurons three-dimensionally grown on a biomimetic ECM-derived matrix and (ii) a 2D immune compartment comprising the direct co-culture of brain resident (i.e. astrocytes) and peripheral (i.e. differentiated monocytes) immune cells. With this setup, we intend to recapitulate the 3D DA circuits most susceptible to PD and evaluate their simultaneous response to PD-related neurotoxins (i.e. A53T α-syn) and the secreted pro-inflammatory signals from the immune compartment.

3D bioprinted neuronal cores model functional DA networks in vitro
To recreate functional DA cores in vitro, we employed hydrogel-based 3D culture systems that compositionally and physiochemically resemble brain environments. In particular, we optimized the 3D bioprinting and in situ differentiation of immortalized LUHMES NPCs within our previously developed biomimetic bioinks composed of SISMA with homogenously dispersed GO nanosheets [27], henceforth termed SISMA-GO ( figure 1(A)). SIS is one of the most studied ECM sources for tissue engineering applications due to the low cellular content and rich ECM content of native SIS [40][41][42]. It is primarily composed of type I collagen, but other structural proteins such as elastin, fibronectin, laminin and type III, IV and VI collagens have also been identified [43,44]. In addition, SIS contains a wide variety of glycosaminoglycans and proteoglycans that provide further cell attachment sites within the hydrogel and serve as repositories of growth factors secreted by native cells [44]. Figure 1(C) shows that SISMA hydrogels preserve a variety of these growth factors, many of which have important roles in the central nervous system. For instance, abundantly conserved growth factors of the insulin-like growth factor (IGF) system, such as IGF-I and IGF binding proteins (IGFBPs) 2, 3, and 4, play crucial roles in neural growth, development and function [45], and other growth factors of the fibroblast and vascular endothelial growth factor (FGF and VEGF) families are known to stimulate neurogenesis [46][47][48]. This suggests that these hydrogels pose a favorable environment for the 3D culture of NPCs. Also, the abundance of bone morphogenetic proteins (BMPs) 5 and 7 may be helpful for implementing DA differentiation profiles, since these have been shown to be involved in the differentiation of neural stem cells into midbrain DA neurons [49].
Further, methacryloyl moieties in SISMA hydrogels facilitate photocrosslinking reactions with cytocompatible blue-light irradiation dosages [27], rendering them with improved mechanical stability and tunable stiffness as a function of SISMA concentration and crosslinking degree. This biochemical modification has shown to overcome the low biomechanical stiffness and poor long-term stability of ECM-derived hydrogels [50] by permitting the formation of dense and stable hydrogel networks [27,51]. Moreover, we previously showed that exposure to cytocompatible doses of ascorbic acid induces the in situ reduction of dispersed GO and significantly improves its electroconductivity, which is particularly appealing for the proposed application given the inherent electrical properties of neural tissue. Therefore, we ensure a homogenous dispersion of GO nanosheets within SISMA bioinks by coating them with fibronectin, a strategy that we previously showed to stabilize these nanostructures in electrophilic environments [27] and that is expected to facilitate its interaction with cellular substrates given the important role of this protein in cell-matrix interactions of neural lineages [52,53].
We induced the differentiation of LUHMES NPCs in bioprinted SISMA-GO constructs by incubating them for seven days with an ascorbic aciddoped differentiation cocktail, previously described for 2D cultures (figure S1) [30]. Moreover, we modulated hydrogel stiffness to optimize this process in terms of both metabolic activity and the expression of differentiation markers. Figure 2(A) shows that both SISMA concentration and GO addition readily modulate hydrogel stiffness in bioprinted constructs, and these mechanical differences affect the metabolic activity of maturating NPCs ( figure 2(B)).
Specifically, matrices with storage moduli above 1 kPa appear to hamper the sustained increase in metabolic activity observed in 2D controls during their seven-day differentiation period, which is in line with previous reports that have shown that neural lineages are optimally grown in soft matrices [54]. We observed a similar effect for matrices with storage moduli below 0.5 kPa, suggesting too soft matrices might not be suitable either for supporting cellular growth. This positions 0.5-1 kPa as the optimal stiffness range for SISMA-based neural constructs. After seven days of differentiation, GO-tethered SISMA matrices within this range exhibit strikingly increased metabolic activity, with fold change values nearly twice as large as their GO-deficient counterparts and slightly higher than 2D controls. Moreover, although 15 mg ml −1 SISMA-GO and 20 mg ml −1 SISMA constructs present comparable stiffness values, we only observed a marked increase in metabolic activity from days 3 to 7 in the former.
Previous research has demonstrated that neuronal differentiation induces metabolic reprogramming from aerobic glycolysis to oxidative phosphorylation [55]. Considering that the used metabolic assay (i.e. Alamar blue) is more sensitive to the latter [56], GO addition may also be promoting neuronal differentiation. This is further supported by the fact that, by day 5, tetracycline supplementation within the differentiation medium has already induced the transition of most NPCs to post-mitotic states [30,57], which suggests cellular proliferation dynamics should not be largely contributing to the observed metabolic increase. This hypothesis was confirmed by evaluating the expression of neuronal and DA markers in SISMA and SISMA-GO constructs at both concentrations. After the seven-day differentiation period, the neuron-specific cytoskeleton marker βIII-tubulin (Tuj1) was distinctly upregulated in all constructs (figure 2(C)), confirming the appearance of neuronal phenotypes within embedded cells. Most importantly, the expression of tyrosine hydroxylase (TH), a rate-limiting enzyme in dopamine synthesis [58], was also upregulated in most cells, thus suggesting the effective appearance of DA neuronal populations. We assessed the extent of these DA phenotypes in each formulation by comparing the change in TH gene expression during this differentiation period (figure 2(D)), as well as that of two other markers of mature DA neurons: the potassium inwardly rectifying channel subfamily 6 (KCNJ6) and the dopamine active transporter (DAT) (figures 2(E) and (F)) [59]. In line with the previously observed metabolic differences, matured cells in 15 mg ml −1 SISMA-GO express higher levels of TH, KCNJ6 and DAT when compared to those in GO-deficient analog constructs and 20 mg ml −1 SISMA constructs of comparable stiffness. Notably, KCNJ6 and DAT were downregulated in the latter constructs, indicating that mature DA phenotypes were not fully obtained. In contrast, only matured 15 mg ml −1 SISMA-GO constructs consistently demonstrated comparable (e.g. KCNJ6, DAT) or increased (e.g. TH) levels of these DA markers when compared to the 2D differentiated control, confirming the superior generation of mature DA neurons within these 3D systems.
Remarkably, synaptic activity was also enhanced in matured 15 mg ml −1 SISMA-GO constructs, as evidenced by the higher expression levels of the presynaptic protein synapsin 1 (Syn1) (figure 2(G)) and their markedly superior neurite outgrowth and network formation (figure 2(C)). Their superior synaptic activity with respect to 20 mg ml −1 SISMA emphasizes the important role of these electroconductive GO nanostructures in the formation of neuronal networks, beyond the possible effects of hydrogel stiffness. The microenvironment of 15 mg ml −1 SISMA-GO constructs is also effectively preventing the non-specific differentiation of LUHMES NPCs into oligodendrocytic and astrocytic phenotypes, as evidenced by the marked downregulation of the cell type-specific oligodendrocyte transcription factor (OLIG2) and GFAP in these constructs (figures 2(H) and (I)). This is especially advantageous in light of the significant presence of oligodendrocytic-like populations in 2D differentiated cells utilizing the same scheme (figure 2(H) and [60]). Taken together, these results suggest that both the precise modulation of hydrogel stiffness and the incorporation of GO nanosheets within our ECM-derived matrices are essential for the optimal development of functional and interconnected DA neuronal networks.

Co-cultured resident and peripheral immune cells exposed to A53T α-synuclein model neuroinflammatory environments in vitro
To recreate representative paracrine signals of PD neuroinflammatory environments, we model an inflammatory scenario where the joint response of brain resident and infiltrating peripheral immune cells to aberrant A53T α-syn orchestrates the feedforward secretion of pro-inflammatory mediators. Specifically, we used human astrocytes (NHA) to represent brain resident immune inputs and human THP-1 monocytes, pre-differentiated into macrophage-like and dendritic cell-like phenotypes, to represent peripheral inputs.
We first characterized the individual inflammatory responses in differentiated THP-1 monocytes to identify optimal stimulation parameters that produce inflammatory phenotypes most similar to those described for these cells in PD. In this regard, monocytic differentiation towards both monocyte-derived macrophage (MDM) and dendritic cell (moDC) subpopulations is desired, as each of these cells plays crucial yet distinct roles in the modulation of innate immune responses [61][62][63]. Accordingly, we treated THP1 monocytes with varying doses of PMA, a widely reported monocyte differentiating agent [64,65], and jointly optimized these pre-differentiation schemes with posterior A53T α-syn stimulation. This is due to the fact that PMA exposure enhances the response of differentiated cells to posterior proinflammatory stimuli [66].
We exposed THP-1 monocytes to low and moderate PMA concentrations (i.e. 50 and 200 nM) and characterized their posterior response to two A53T α-syn concentrations (i.e. 5 and 10 µg ml −1 ) in terms of acquisition of MDM and moDC phenotypes while preventing immunotoxic side effects derived from overstimulation. We determined that these concentration ranges do not induce cytotoxic responses by measuring metabolic activity and LDH leakage in each stimulation scenario, which reached values above 80% and below 20%, respectively (figure S2). We also validated that treating THP-1 monocytes with PMA is essential for achieving differentiated phenotypes as A53T α-syn exposure alone does not induce cell adherence ( figure  S3), which is a hallmark of monocyte differentiation [67]. In contrast, PMA-treated cells acquired rounded adherent morphologies and transitioned towards characteristic 'fried-egg' and dendritic morphologies of M1-polarized macrophages and mature DCs upon A53T α-syn stimulation (figure S3) [68,69]. Moreover, the expression of CD68, a classic macrophage marker, was significantly upregulated by PMA exposure in a concentration-dependent manner. Most importantly, only PMA pre-treated cells exhibited marked increases in CD68 expression upon A53T α-syn stimulation, a previously reported behavior in macrophages exposed to inflammatory stimuli [70] (figures 3(A) and S4). This further confirms the absence of macrophage phenotypes without PMA. We observed a similar behavior with the expression of CD14, a surface marker involved in immune recognition and activation of toll-like receptor 4-mediated inflammatory pathways in myeloid cells [71], where PMA pre-treatment strengthened CD14 upregulation in A53T α-syn stimulated cells (figures 3(B) and S5). However, pre-treated cells with the highest PMA concentration express lower levels of CD14 in all stimulation scenarios, despite them being more sensible to inflammatory stimuli [72]. Considering moDCs have been reported to express lower CD14 levels than MDMs [73,74], this could suggest the presence of larger subpopulations of moDC in these inflammatory scenarios. We confirmed this hypothesis by assessing the expression of the DC-specific marker DC-STAMP [75], which was most prominent in cells pre-treated with the highest PMA concentration ( figure 3(C)). These findings suggest that heterogeneous populations containing both MDMs and moDCs can be obtained most efficiently by pre-treating cells with the highest concentration of PMA evaluated.
We characterized the A53T α-syn-induced inflammatory responses in these pre-differentiated populations by comparing the expression of the pro-inflammatory and anti-inflammatory surface markers CCR7 and CD206, respectively, each of which is readily expressed by both MDMs and moDCs [76][77][78]. As expected, A53T α-syn exposure induces concentration-dependent increases in CCR7 (figures 3(D) and S6) and decreases in CD206 (figure 3(E)), and this effect is irrespective of the intensity of PMA pre-treatment. However, we observed marked differences in cytokine secretion profiles as a function of both PMA pre-treatment and A53T α-syn concentration, as determined by the secreted levels of 13 cytokines closely implicated in inflammatory responses (figures 3(F)-(K) and S7). For instance, A53T α-syn stimulation elicited dose-dependent increases in the secretion of interleukin 1β (IL-1β), tumor necrosis factor-α (TNFα), interleukin 6 (IL-6) and interferon γ (IFN-γ) (figures 3(G)-(J)), four pivotal cytokines in neuroinflammation that, beyond their role in promoting innate immune responses, induce direct DA toxicity in PD environments [79][80][81][82]. Moreover, pre-treated cells with the highest PMA concentration showed increased sensitivity to A53T α-syn stimulation and achieved maximal secretion levels of these four proinflammatory cytokines, especially at higher A53T α-syn doses. Monocyte chemoattractant protein-1 (MCP-1) secretion also increased in response to A53T α-syn stimulation, reaching over 1500-fold values with highest doses ( figure 3(K)). This is a crucial effect for immune cell recruitment in PD [83] and it was only observed in PMA pre-treated cells. In fact, PMA pre-treatment significantly upregulated basal secretion levels of these and other pro-inflammatory cytokines such as IFN-α2, IL-8, IL-12p70 and IL-18 ( figure S7) and thus facilitated their further upregulation in response to A53T α-syn. Again, this was most prominent in pre-treated cells with the highest PMA dosages. Taken together, these results suggest that THP-1 monocytes pre-treated with 200 nM PMA and further stimulated with 10 µg ml −1 A53T αsyn most closely resemble peripheral inputs in PD neuroinflammatory environments in terms of both phenotypical profiles and inflammatory responses.
Consequently, we co-cultured PMA-pretreated THP-1 monocytes with human astrocytes (NHA) and evaluated their joint inflammatory responses to A53T α-syn ( figure 4(A)). We evaluated two THP-1: NHA co-culture ratios (i.e. 1:1, 2:1) that did not significantly affect the viability of either cell type (figures 4(B) and S9) in order to define optimal conditions for maximizing their interaction. Figures 4(C)-(E) show that when these two cell lines are grown together, pro-inflammatory pathways are activated even before exogenous stimulation. This is shown by the increased expression of the proinflammatory markers CCR7 and GFAP in differentiated THP-1 monocytes and NHA, respectively [77,84,85]. Moreover, A53T α-syn exposure further increased the expression of these markers at both co-culture ratios, reaching values significantly higher than in individual monocultures. This correlates well with the drastic increases in pro-inflammatory cytokine secretion within co-culture scenarios in response to A53T α-syn stimulation, which were consistently greater than the sum of individual secretions in monoculture scenarios (figures 4(F)-(O)). In this regard, the secretion of pro-inflammatory cytokines such as TNF-α, IFN-α2, MCP1, IL-12p70, IL-8, IL-18 and IL-23 was at least double in co-culture scenarios, with TNF-α and IL-12p70 reaching values over 10 and 500 times higher. This synergistic effect suggests that, beyond the individual effect of A53T αsyn over each cell type, direct and/or paracrine interactions between the diverse immune phenotypes of these co-culture environments are also largely contributing to the inflammatory outcome. Moreover, co-culture conditions alone induced maximal secretions of TNF-α, IL-6 and IL-1β (figures 4(F)-(H)), the three pivotal cytokines in PD [79,81,86], and these were only marginally increased further with A53T α-syn stimulation. In contrast, the secretion of other cytokines such as IFN-γ, IFN-α2, MCP-1, IL-12p70, IL-8, IL-18 and IL-23 was significantly heightened by the presence of this neurotoxin (figures 4(I)-(O)), all of which have been shown to contribute to PD progression [83,[87][88][89][90][91][92]. This, in turn, suggests that co-culture settings set the basic disease-mimicking environmental conditions and A53T α-syn enhances the complexity of immune responses. Further, the fact that individual monocultures of both NHA and differentiated THP1 exhibited differential pro-inflammatory cytokine secretion levels in response to A53T α-syn may imply that the inflammatory input of each cell type in A53T α-syn stimulated co-culture conditions is unique, and thus resembles the multifaceted nature of neuroinflammatory environments. Consistently, the secretion of anti-inflammatory cytokines such as IL-10 [93] and IL-33 [94] remained comparable to unstimulated scenarios, and we observed no synergistic changes in co-culture conditions (figure S10), which corroborates that the elicited responses in these environments are primarily neurotoxic.
Although both co-culture ratios exhibited similar inflammation profiles, cytokine secretion in THP1-NHA co-cultures at a 2:1 ratio was consistently higher, both before and after A53T α-syn stimulation. Accordingly, we selected this optimal ratio to recapitulate neuroinflammatory responses in our proposed neuro-immune PD model.

A53T α-synuclein stimulated neuro-immune co-culture systems recapitulate PD-specific pathologies
We elicited PD-mimicking pathologies within our 3D DA core by simultaneously exposing it to soluble A53T α-syn oligomers and A53T α-synstimulated immune co-cultures of NHA, MDM and moDC subpopulations ( figure 5(A)). Specifically, we exposed our 3D bioprinted and matured DA constructs (section 2.1) to our 2D immune co-culture (section 2.2) in a transwell format, with soluble A53T α-syn in the culture media, for up to five days. We looked at how well this neuro-immune co-culture system could recreate the microenvironments of PD by observing the development of key pathologies of this disease and comparing it to the current gold standard for in vitro PD research. We compare our model to 2D DA neuron cultures that were exposed to rotenone, a mitochondrial complex I inhibitor that is known to cause PD-related intracellular pathologies [31,95]. We also compared our model with A53T α-syn stimulated DA cores in the absence of neuroinflammatory inputs to accurately grasp their contribution in the developed pathologies. Consistent with previous reports [31], over 80% of DA neurons in the rotenone-stimulated control were either early apoptotic, late apoptotic or dead during the first 48 h of stimulation, and all cells were dead by the 5th day ( figure 5(B)). Conversely, only 30% of stimulated DA neurons in our neuroimmune α-syn and immune co-cultures for one, two and five days, as a marker for cellular viability and apoptotic outcomes. While rotenone is lethal to DA neurons within this stimulation period, over 70% in A53T α-syn stimulated neuroimmune co-cultures remain alive. Live cells: homogenous AO stain (green) without nuclear condensation, early apoptotic cells: AO stain with clear nuclear condensation and fragmentation, late apoptotic cells: overlapping AO and EH (red) nuclear staining, dead cells: EH staining with low AO staining. Percentage of live, early apoptotic, late apoptotic and dead cells is shown on the left panel. (C) Representative CLSM images of JC1 aggregation in A53T α-syn-stimulated neuroimmune co-cultures after zero, one, two and five days. JC1 aggregates appear in energized mitochondria (red) and monomeric forms remain in depolarized mitochondria (green), indicating mitochondrial dysfunction. Scale bar: 20 µm. (D) Estimation of mitochondrial membrane potential (MMP) in DA neurons exposed to the three stimulation scenarios mentioned above for up to five days, as determined by the red-to-green (R/G) intracellular fluorescence ratio of JC1 emission. (E) Intracellular ROS production during these stimulation scenarios as determined by ROS-sensitive DCFDA fluorescence. DA neurons in A53T α-syn stimulated neuroimmune co-cultures exhibit the highest mitochondrial dysfunction and ROS production. ( * ) and (#) denote statistical significance against unstimulated and rotenone controls, respectively. Statistical analyses were performed with Two-way ANOVAs coupled to Tukey's multiple comparison tests. co-cultures underwent apoptotic processes throughout the whole five-day stimulation period. Apoptotic outcomes were even more reduced in A53T α-synstimulated constructs without neuroinflammatory inputs, where late apoptotic phenotypes were not even observed within this period. Despite the reduced lethality in our model, we observed marked mitochondrial dysfunction as early as 24 h after the start of stimulation ( figure 5(C)). This is evidenced by the notable accumulation of monomeric JC1 in mitochondrial membranes throughout the whole stimulation period, which can only occur in low MMP conditions. Strikingly, the drop in MMP was significantly larger in our A53T α-syn stimulated neuroimmune co-culture model than in 2D cultures stimulated with rotenone (figures 5(C) and S11), whose mechanism of action is known to directly disrupt mitochondrial function [31,95]. Mitochondrial dysfunction was also significantly lower in the absence of immune responses (figures 5(C) and S11), suggesting the secreted neuroinflammatory mediators are also contributing to the observed pathologies. These results were further corroborated with the observed changes in intracellular ROS production, which peaked in our A53T α-syn stimulated neuro-immune co-culture model after only 24 h ( figure 5(D)). These findings were supported by changes in intracellular ROS production, which peaked after only 24 h in our A53T α-syn stimulated neuro-immune co-culture model ( figure 5(D)). This was expected considering that neuroinflammatory inputs directly influence ROS production through the activation of stress signaling pathways [96], which, together with the effects of mitochondrial dysfunction, ignite the development of oxidative environments. This oxidative response remained significantly higher than unstimulated and A53T α-syn stimulated controls, and comparable to the rotenone control through the following days, confirming the persistence of oxidative stress in our model. Intracellular α-syn distribution was also considerably shifted in DA neurons of our neuroimmune co-culture systems. Normal α-syn nuclear localization [97] is drastically reduced in both non-apoptotic and apoptotic cells (figures 6(A) and (B)) and rather shifted to cytoplasmic regions. Mitochondrial colocalization increases significantly with time, reaching correlation values over three-fold higher after 48 h (figures 6(A) and (C)). This is consistent with the described interaction of α-syn with mitochondrial inner membranes in PD [98] and the enhanced mitochondrial dysfunction observed above ( figure 5(C)). Notably, distinct cytoplasmic α-syn aggregates are detectable as early as 24 h after the onset of stimulation, which resemble the α-syn protofibril bodies observed in PD. The fact that A53T αsyn exhibits a high aggregation potential in solution [26] suggests that exogenous A53T α-syn is likely transitioning into its insoluble fibrillary form and accumulating within intracellular compartments. As previously reported [99], the appearance of fibrillary A53T α-syn aggregates may trigger the conversion of endogenous monomeric α-syn into these aggregated structures. Moreover, given that α-syn cytoplasmic distribution remained homogeneous in the absence of neuroinflammatory inputs, the recreated neuroinflammatory microenvironments may contribute significantly to the dynamics of α-syn fibrilization. This correlates well with the stimulatory effect of neuroinflammatory inputs on oxidative stress ( figure 5(D)), as evidence suggests oxidative environments exacerbate α-syn fibrilization via oxidative modifications to the protein backbone [100,101]. In addition, rotenone-stimulated control cells expressed significantly lower levels of α-syn in both the nucleus and the cytoplasm. Although few distinct cytoplasmic aggregated bodies are discernible in some cells after 48 h, there was no observable increase in mitochondrial colocalization during this period. This further implies that our neuro-immune co-culture system is superior to rotenone-based PD models in terms of the recapitulation of α-synucleinopathies.
During the first 48 h of stimulation, DA neurons in our stimulated neuroimmune co-culture models also exhibited significant functional alterations, as evidenced by marked downregulations in critical DA (e.g. TH, KCNJ6 and DAT) and synaptic connectivity (e.g. Syn1) genes. TH expression was markedly reduced by almost 64-fold (figure 6(D)), suggesting a clear decrease in dopamine synthesis [102], coupled to over 2000-fold decreases in KCNJ6 and DAT (figures 6(E) and (F)), two transmembrane proteins dictating synaptic release and recycling of dopamine [59,103]. These results, taken together with the over 1000-fold decrease in Syn1 expression (figure 6(G)), suggest that these disease mimicking microenvironments may be significantly interfering with the synaptic activity and interconnectivity of embedded DA neuronal networks. Notably, we observed attenuated downregulations of DA genes but no significant change in Syn1 expression in A53T α-syn-stimulated constructs lacking neuroinflammatory inputs, indicating that the effect of these environments also directly interferes with synaptic function. Again, the extent of downregulation of these proteins in the rotenone-stimulated control was significantly less marked than in neuro-immune coculture systems. In addition, we observed marked downregulations of two PD-related genes involved in mitophagy pathways within neuroimmune cocultures, namely, PINK1 and Parkin (figures 6(H) and (I)). These two proteins are critical regulators of the efficient elimination of damaged mitochondria, and their downregulation or deletion is closely associated with increased mitochondrial pathologies and sensitivity towards aberrant α-syn [104]. In contrast, rotenone-stimulated controls exhibited unaltered or upregulated expression levels of these proteins 48 h after exposure. This suggests that, while rotenone insults do not completely impair mitophagy, this process is significantly impaired in A53T α-synstimulated neuroimmune co-cultures, making DA neurons more prone to mitochondrial damage.
Taken together, these results suggest that our proposed model is effectively recapitulating key pathological hallmarks of PD, while demonstrating improved resilience to neurotoxic insults. This is a clear advantage over current in vitro rotenone standards, given that rapid rotenone-induced cell death largely limits the precise study of disease progression [31]. The possibility of utilizing this approach to model PD for longer time periods is particularly exciting, as it would allow the implementation of more extensive studies on disease initiation and progression, as well as a more reliable evaluation of the potential of therapeutic pharmacological candidates for this disease.

Discussion
The multifactorial nature of neurodegenerative diseases has significantly hampered the development of reliable PD research in vitro, as the variety of difficulties in replicating the complexity of brain structures and multicellular responses has resulted in oversimplified models. The neuroimmune co-culture proposed here is intended to increase the robustness of in vitro responses by incorporating key microenvironmental factors that drive the progression of PD.
We showed that mature and interconnected DA neuron networks can be generated within our 3D ECM-derived matrix with improved functionality and differentiation specificity, which was largely driven by its readily-tuned stiffness and the presence of electroconductive nanostructures. Moreover, we showed that immune responses greatly reminiscent to neuroinflammatory environments can be adequately recreated with our optimized immune cocultures of astrocytes and differentiated monocytes, whose response to A53T α-syn recapitulates complex inflammatory responses with both resident and peripheral immune cell inputs. Coupled with direct neurotoxic responses of A53T α-syn over DA neurons, these environments induced a spectrum of PD-mimicking pathologies within DA networks, including synaptic and mitochondrial dysfunction, α-syn-based aggregated bodies, and oxidative environments, with comparable or superior effects to rotenone-stimulated 2D DA neuron cultures that are standardly used to model PD. Most importantly, the improved resilience of DA neurons to these neurotoxic insults is a promising advantage over these rotenone-based models, whose rapid cell death considerably limits the study of disease progression. In fact, this effect is also superior to other comparable PD models in 3D spheroid or organoid modalities, which are only exposed to mitochondrial stressors (i.e. rotenone, MPTP, 6-OHDA) for 48 h at most [105][106][107] and whose cytotoxic responses are comparable to respective 2D cultures [105].
From a broader perspective, our model poses additional advantages over previous 3D models of PD. For instance, our tetracycline-inducible NPC differentiation regime allows the obtention of DA neuronal networks after only 7 days, which contrasts widely with organoid protocols derived from iPSCs [108,109] or neuroepithelial stem cells [110] that need over 21 days for the appearance of DA phenotypes. Moreover, the closer lineage relationship between DA neurons and NPCs in our approach makes the final cellular composition much more reproducible, especially in comparison to iPSCderived approaches [23]. Thus, the reduced maturation time and high reproducibility of our model could facilitate its use in higher throughput applications such as pharmacological or genetic screening. In addition, the in situ differentiation of DA neurons within our 3D matrices represents a less invasive approach on neuronal network formation than previous 3D bioprinted PD models, which rely on embedding already differentiated DA neurons [105,111]. These approaches require longer acclimation periods for the regeneration of axonal networks, at best, and may lead to suboptimal phenotypic baselines, at worst. Further, our dual stimulation approach combining PD neurotoxins and pro-inflammatory mediators demonstrated to be instrumental for the generation of complex PD-related microenvironments and goes beyond the one-dimensional stimulation approaches employed previously using either mitochondrial stressors [105][106][107] or single genetic mutations of PD risk genes [23,108,[111][112][113]. To the best of our knowledge, this is also the first time that neuroimmune co-cultures are used to promote PDrelated pathologies in vitro.
However, despite the increased complexity of the PD-mimicking responses generated in our in vitro model, these are still simplified versions of those generated in vivo. Including other brain resident cell types (e.g. oligodendrocytes, endothelial cells, microglia) would certainly increase the robustness of the generated responses as they all contribute to neuronal degeneration in PD. Affected oligodendrocytes and endothelial cells, for example, promote axonal demyelination [114] and vascular leakage and regression [115], whereas microglia play important roles in propagating neuroinflammation [114]. Although some studies have already started to incorporate the first two in 3D midbrain in vitro models to aid neuronal network formation [105,106], their precise effects over degenerative responses in vitro are yet to be studied in detail. Microglia, in particular, would be an interesting addition to our neuroinflammatory co-cultures to not only exacerbate but increase the complexity of the generated paracrine signals by adding another layer of crosstalk between immune cell types. However, for the purpose of inducing paracrine pro-inflammatory signaling, the cytokineproducing capacity of astrocytes and their ability to secrete pro-inflammatory mediators [116] seem to be sufficient to induce relevant responses in synergy with peripheral immune cells.
Finally, it should be noted that α-syn mutations are not the only way by which this disease can appear and progress in patients. In fact, mutations in several other proteins (e.g. LRRK2, PRKN, DJ1, ATP13A2, PINK1) have been involved in triggering PD [117] and over 90% of patients present idiopathic disease conditions, making the origin of PD elusive in such cases [10,11]. Therefore, given that we use standard cell lines and A53T α-syn for stimulation, our model holds limited potential for personalized medicine approaches, especially when compared to models using PD patient-derived iPSCs as the starting population [23,108,112]. Nevertheless, our model holds the potential to address a wide variety of unknowns within α-synucleinopathies and their important role within PD progression, independent of its origin.

Conclusion
The translational gap between in vitro and in vivo PD models has been largely breached by the lack of reliable in vitro models that accurately mimic PD neurodegeneration. With our proposed neuroimmune co-cultures, we demonstrate the development of PD-related phenotypes across multiple areas of neurodegenerative pathologies, including protein aggregation, oxidative stress, and mitochondrial and synaptic dysfunction. With its high reproducibility, short maturation period and improved resilience to neurotoxic insults, our models holds much potential across many areas of basic α-synucleinopathy research and drug discovery, ranging from early target identification to toxicology screens. The encouraging results presented here move us one step closer to the development of dependable in vitro models that will facilitate PD research.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).