Nanodiamonds-induced effects on neuronal firing of mouse hippocampal microcircuits

Fluorescent nanodiamonds (FND) are carbon-based nanomaterials that can efficiently incorporate optically active photoluminescent centers such as the nitrogen-vacancy complex, thus making them promising candidates as optical biolabels and drug-delivery agents. FNDs exhibit bright fluorescence without photobleaching combined with high uptake rate and low cytotoxicity. Focusing on FNDs interference with neuronal function, here we examined their effect on cultured hippocampal neurons, monitoring the whole network development as well as the electrophysiological properties of single neurons. We observed that FNDs drastically decreased the frequency of inhibitory (from 1.81 Hz to 0.86 Hz) and excitatory (from 1.61 to 0.68 Hz) miniature postsynaptic currents, and consistently reduced action potential (AP) firing frequency (by 36%), as measured by microelectrode arrays. On the contrary, bursts synchronization was preserved, as well as the amplitude of spontaneous inhibitory and excitatory events. Current-clamp recordings revealed that the ratio of neurons responding with AP trains of high-frequency (fast-spiking) versus neurons responding with trains of low-frequency (slow-spiking) was unaltered, suggesting that FNDs exerted a comparable action on neuronal subpopulations. At the single cell level, rapid onset of the somatic AP (“kink”) was drastically reduced in FND-treated neurons, suggesting a reduced contribution of axonal and dendritic components while preserving neuronal excitability.

ND coatings also represent excellent substrates for neuronal cultures, preserving intrinsic neuronal excitability as well as spontaneous synaptic transmission, even though nanoparticle size and their surface features may affect neuronal adhesion 20,21 . However, if the bulk material is non-toxic and inert, when dealing with diamond at the nanoscale, quite heterogeneous results emerge. FNDs have been proven to be suitable for long-term in vivo imaging in Caenorhabditis elegans, without causing toxicity 22,23 ; in cancer and stem cells, division and differentiation are not affected by 100 nm NDs 24 . In a different experimental model, airway epithelial cells, functionalization may reduce ND inflammogenicity 25 , while a remarkable dose-dependent decrease of neurite length of central and peripheral cultured neurons after FNDs seeding has been demonstrated 26 .
Focusing on the interaction of FNDs with hippocampal neurons, in this work we combined patch-clamp and microelectrode array (MEA) recordings to evaluate their functional implications on hippocampal neurons functionality. On mature neurons, FNDs application drastically impairs AP spontaneous firing, as well as the occurrence of spontaneous miniature synaptic currents. On the contrary, the network synchronism, the balance between excitatory and inhibitory synapses and the proportion of fast-versus low-spiking neurons are not affected by FNDs. At the single-cell level, the AP shape analysis suggests that FNDs alter the cell morphology rather than modifying the passive membrane properties and the cell responsiveness to applied electrical stimuli. Hippocampal neurons functionality is also preserved following ODMR test measurements on cell-internalized FNDs.

Results
FND characterization. The size of FNDs was estimated by means of SEM imaging, after finalizing their preparation. Figure 1a and b show typical SEM micrographs, where most of the nano-particles present the expected dimensions, namely below 100 nm. The powders are classified as Ib type, with a nominal substitutional N concentration of 10 ÷ 100 ppm, and contain a low amount of native NV − centers. Ion beam irradiation (Fig. 1c) and a subsequent thermal annealing were performed to increase the number of NV − centers in each nanocrystal (see Methods).
Room temperature photoluminescence (PL) spectroscopy was performed with the purpose of assessing the spectral features of the ND emission. A confocal microscope was employed to this scope, with 532 nm laser excitation and a single-photon-sensitive avalanche Silicon detector. The typical spectrum of NV − centers in NDs is reported in Fig. 1d: consistently with what reported in literature 27,28 , the 638 nm zero phonon line and relative phonon bands are observable. No features involving organic contaminants or graphite are observable, thus confirming the effectiveness of the cleaning procedure (see Methods).

FND internalization. FNDs internalization mechanisms and their related motional dynamics inside cells
have been previously investigated in various experimental models (HeLa cells, embryo hippocampal neurons, as well as lung cancer cells and embryonic fibroblasts) 23,24,29,30 . Here we assessed FNDs internalization by means of confocal microscopy. Embryo hippocampal neurons were incubated with 40 µg/ml FNDs 26 , being this concentration far below the threshold of cytotoxicity (i.e. 250 µg/ml) 12 . After 24 hours, the cytoplasmic labelling dye (CellTracker ™ Green CMFDA, ThermoFisher,) was added to the medium. This allowed to identify the cell boundaries together with the internalized FNDs, characterized by a red fluorescent emission. Representative images are shown in Fig. 2a. It is worth noting that since FNDs aggregation may occur in physiological culture media 18 , detection of single FNDs may be hindered using our acquisition system.
To exclude the possibility that the FNDs concentration in neurons declines over time, we performed FNDs seeding at two different stages of culture development: at 7 DIV (young cultures) and at 14 DIV (mature cultures) and measured the number of internalized FNDs at 18 DIV, that was respectively after 11 and 4 days of incubation (Fig. 2b). We found that longer periods of incubation (11 vs. 4 days) increase the number of internalized FNDs from 14.6 ± 3.6 (4 days) to 29.4 ± 2.9 (11 days) (p < 0.01) (Fig. 2c).
Firing frequency of the hippocampal network is differently affected by FND administration at early and late stages of development. Hippocampal neurons, when cultured on microelectrode arrays (MEAs), create a network that exhibit spontaneous firing activity which is characterized by different patterns along with culture maturation: asynchronous firing is typical of younger neurons (7-13 days in vitro, DIV), while in elder cultures activity becomes highly synchronized and organized into bursts (>14 DIV) 31 . Here, by means of microelectrode arrays (MEAs), we compared the firing frequency of hippocampal neurons under control conditions and after exposure to FNDs. FNDs seeding was performed at two different stages of culture development, respectively at 7 DIV (young cultures) and at 14 DIV (mature cultures). Unless otherwise specified, firing activity was measured at 18 DIV.
Representative recordings are shown in Fig. 3 respectively for control neurons (a), for neurons incubated with FNDs at 7 DIV (b) and neurons incubated with FNDs at 14 DIV (c).
Regardless of the exposure stage of FNDs, we found no changes in the number of spontaneously active neurons, as demonstrated by the unaltered percentage of electrodes exhibiting firing activity: 49 ± 7% in control (n = 26 MEAs), 41 ± 7% when FNDs were administered at 7 DIV (n = 17 MEAs) and 47 ± 8% when FNDs were administered at 14 DIV (n = 16 MEAs).
On the contrary, the firing frequency was differently affected depending on the developmental stage of incubation with FNDs (7 versus 14 DIV). When FNDs were applied at 14 DIV ( Fig. 3c and d), they drastically reduced the mean frequency on 18 DIV neurons with respect to controls (36%), from 1.03 ± 0.07 Hz to 0.66 ± 0.05 Hz (p < 0.001). On the contrary, early incubation at 7 DIV preserved the firing frequency during culture maturation, as confirmed by measurements at 11 DIV (0.23 ± 0.03 Hz for controls and 0.29 ± 0.02 Hz with FND, p > 0.05) (Fig. 3e), at 14 DIV(0.8 ± 0.2 Hz for controls and 0.60 ± 0.04 Hz with FND, p > 0.05) (Fig. 3f) and at 18 DIV, with a mean firing frequency of 1.03 ± 0.07 Hz in control and 1.06 ± 0.07 Hz in the presence of FNDs (Fig. 3d). Similarly, the mean number of bursts, evaluated over 120 s, decreased by 44%, (i.e. from 16 ± 1 to 9 ± 1, p < 0.001), when FNDs were applied on mature networks, but remained unaltered when FNDs incubation occurred earlier (Fig. 3g). Burst duration was not significantly different among controls and FNDs (Fig. 3h).
Overall, these data suggest that spontaneous firing is significantly impaired only if the exposure to FNDs occurs at later stages of network maturation. Since early incubation (7 DIV) did not induce any significant effect, all other experiments were carried out by adding the FNDs at 14 DIV and recording the firing activity at 18 DIV.
Despite this reduction of spontaneous frequency, we could not find any significant difference on the degree of hippocampal network synchronization 31 , independently of FNDs application at 7 or 14 DIV. This is clearly visible in Fig. 4, which shows the raster plots and the related cross-correlogram plotted for the different experimental conditions. In more details, the cross correlation probability in control neurons was 0.29 ± 0.01, and remained 0.28 ± 0.01 (7 DIV incubation) and 0.31 ± 0.02 (14 DIV incubations, p > 0.1, Fig. 4b and c). Due to the role of GABAergic stimuli in triggering network synchronization 32,33 , we next checked the effect of FNDs exposure on inhibitory and excitatory postsynaptic currents.
The frequency of excitatory and inhibitory miniature postsynaptic currents is equally reduced by FNDs. Analysis of miniature inhibitory and excitatory postsynaptic currents (mIPSCs and mEPSCs) revealed that exposure to FNDs at 14 DIV drastically reduced the mean frequency of spontaneous GABAergic (from 1.81 ± 0.18 Hz to 0.86 ± 0.11 Hz, p < 0.01, n = 5) and glutamatergic transmission (from 1.61 ± 0.17 Hz to 0.68 ± 0.08 Hz, p < 0.01, n = 5) 34 . Representative traces are shown in Fig. 5a and b, respectively for inhibitory and excitatory spontaneous events. Experiments were performed at 18 DIV. The mean frequency reduction was comparable for inhibitory and excitatory synapses (52% and 57% respectively, Fig. 5c). Unitary current amplitude peaked at 14.5 ± 0.8 pA and 12.4 ± 0.3 pA, respectively for inhibitory and excitatory spontaneous events, and these mean values were preserved after FNDs exposure, i.e. 14.2 ± 0.6 pA and 12.8 ± 0.7 pA, respectively (Fig. 5c). Similarly, other parameters such as the time-to-peak and the half-width of mPSCs remained unaltered upon FNDs application (p > 0.1), as summarized in Fig. 5c. Overall, these findings suggest that, despite the marked frequency changes, GABAergic and glutamatergic transmission was equally affected during exposure to FNDs and thus likely preserving their balance.
Resting membrane properties are preserved upon FNDs exposure. Membrane input resistance (R in ) was measured to evaluate whether cell membrane integrity could be impaired by FNDs exposure. The protocol consisted in applying hyperpolarizing current pulses (from −100 pA to −120 pA in 10 mV steps), after holding the membrane resting potential at −70 mV and measuring the membrane voltage (V m ) deflection at the steady state (Fig. 6a). Representative traces for control neurons and those treated with FNDs are shown in Fig. 6a; R in was evaluated from the slope of the corresponding voltage-current relationship (Fig. 6b). Mean values of R in (inset) were in good agreement with previous values obtained from dissociated hippocampal neurons 35     Balance of fast-spiking versus slow-spiking neurons in cultured hippocampal neurons is mantained following exposure to FNDs. Within the hippocampus, different populations of GABAergic interneurons tune the activity of pyramidal cells. These distinct phenotypes have been characterized in hippocampal slices on the basis of their electrophysiological properties and post hoc morphology 37,38 . Here we investigated this etherogeneity in dissociated hippocampal neurons. To this purpose, we measured in current-clamp configuration the firing properties of hippocampal neurons (18 DIV) by applying consecutive current pulses of increasing amplitude (from 100 pA to 300 pA, 10 pA steps, as described in Methods). As shown in Fig. 7a, the firing frequency increased with higher current injections; nevertheless, for some neurons the maximum value did not exceed 10 Hz (7.5 ± 1.4 pA for controls and 8.0 ± 1.3 pA for FNDs), whereas in other cases, it exceeded 20 Hz (23.5 ± 1.7 pA for controls and 25 ± 2 pA for FNDs). Thus, as reported for hippocampal slices 39 , we could identify two different groups of neurons on the basis of their firing frequency, named as "slow-spiking" (SS) and "fast-spiking" (FS) neurons. Representative traces of (SS) and (FS) neurons are shown in Fig. 7b, either without FNDs (control, CTRL) or after FNDs treatment. AP firing patterns clearly differ among (SS) and (FS) neurons, rather than among controls versus neurons exposed to FNDs.
Slow and fast spiking neurons also differ for other aspects. First af all, they are characterized by different cell density, being the (SS) neurons more frequently recorded than the (FS) neurons. These latter indeed represent a minority of cells, namely 26% for controls and 38% for FNDs-treated cells. This may suggest that under our experimental conditions (dissociated neurons), FNDs do not significantly alter the balance among neurons subpopulations or do not specifically affect the activity of either neuron subpopulation.
(SS) and (FS) neurons also exhibit distinct R in , in good agreement with the values reported for similar hippocampal neuronal phenotypes 37 . SS neurons had lower R in than (FS) neurons (respectively 0.25 GΩ versus 0.38 GΩ, p < 0.05 for controls) and the same difference was mantained in the presence of FNDs (0.26 GΩ versus 0.32 GΩ, p < 0.05, Fig. 7d). It is worth noting that the mean R in value, obtained by pooling all neurons together (as reported above in Fig. 6a), is effectively balanced among the two subpopulations.
Considering the action potential parameters, we found that in (SS) neurons the AP half width was higher than in (FS) neurons, i.e. 2.5 ms vs. 1.9 ms (p < 0.05) and this difference was mantained after FNDs exposure: 2.3 ms vs. 1.9 ms (p < 0.05, Fig. 7d). On the contrary, the threshold of activation (AP threshold, Fig. 7d), as well as the AP amplitude (not shown), were the same in all conditions (i.e. with or without FND, both for (SS) and (FS) neurons).
A further difference among (SS) and (FS) neurons is represented by the firing adaptation, evaluated from the ratio of the istantaneous frequency measured at the beginning versus the one measured at the end of the AP train.
The onset frequency (f o ) was estimated between the first and second AP, whereas the steady-state frequency (f ss ) corresponds to the firing frequency measured between the last two APs. Thus, from our data we can argue that fast spiking neurons were strongly adapting, being the decrease in firing frequency during the 2 s pulse equal to 40 ± 5% for FNDs and 33 ± 7% for controls (p < 0.05), while adaptation was not significant in slow spiking neurons (p > 0.1). Neuronal adaptation is correlated with the ability of neurons to synchronize with the surrounding network 40,41 . Again, these adaptation properties remained unaltered after FNDs seeding, in good agreement with the comparable cross correlation probability values measured by means of MEAs (Fig. 4).
Finally, another relevant difference among (SS) and (FS) neurons was observed using the phase-plot analysis (Fig. 8), which represents the time derivative of AP voltage (dV/dt) plotted versus voltage 42 . In this analysis we found that the maximum dV/dt value, associated with the density of voltage-gated Na + channels Na v 36,42 , was significantly lower in (SS) with respect to (FS) neurons (115 versus 150 mV/ms, p < 0.05), and, interestingly, the same difference was mantained after FNDs exposure (107 versus 145 mV/ms, p < 0.05, Fig. 7d). From these data we can argue that the contribution of Na v currents to slow and fast-spiking neurons is preserved upon FNDs exposure.
Action potential shape is altered by FNDs. Concerning the action of FNDs on the single action potential shape, we found thatFNDs altered the AP shape by drastically reducing the percentage of cells exhibiting the "kink", visualized as an abrupt voltage transition seen at the spike onset recorded from the soma 43 . The kink reflects the initiation of the spike in the initial segment of the neuron and is associated to the size and geometry of the axon initial segment and dendritic arborization. The kink, monitored in 40% of controls but only in 15% of FNDs-treated neurons, was proportionally distributed among SS and FS neurons (Fig. 8C). Using phase-plane plots, the kink can be visualized as the appearance of two distinct components reflecting the AP initiation in the axon initial segment and the subsequent somato-dendritic compartment 42 . Representative phase-plot graphs (with and without kink) are respectively shown in Fig. 8A and B, for both control and FNDs-treated neurons. The rapid AP onset was found in most of the controls, but was less evident in FNDs-treated neurons. Since the shape of the AP onset at the axon initial segment is accelerated in neurons with larger dendritic surface area 44 , this suggests that a reduced geometry of the dendritic compartment, as in the case of FNDs exposure 26 , or an impaired neurite extension, can be responsible for the slower spike onset and attenuation of the kink in FNDs-treated neurons. In addition, also an altered geometry, as the one induced by FNDs, can cause a reduced resistivity of the critical coupling between the axonal initial segment and soma 45 .
It is worth noting that the FNDs-induced reduction of neurons showing with a fast onset of the AP could also be due to a different contribution of voltage gated Na + or Ca 2+ channels, caused by altered resting membrane potentials. However, we could not observe any significant difference among the resting potential of neurons with and without kink, in the presence or absence of FNDs (p > 0.1 following ANOVA Bonferroni test).

ODMR towards bio-sensing applications.
With the purpose of performing a preliminary feasibility test, we applied the Optically Detected Magnetic Resonance (ODMR) 46 technique to detect signals from FNDs incorporating NV centers that were internalized into hippocampal neurons. Negatively charged nitrogen-vacancy (NV − ) color centers in diamond have been recently proposed as an effective system for nanoscale magnetic field sensing in different contexts, including biological measurements in vitro 5,8,10,18 , on the basis of their excellent magnetic field sensitivity and nano-scale size allowing for high spatial resolution. In particular, each NVis can be regarded as an atomic-sized magnetic field sensor that can be optically read-out optically using ODMR.
Following the protocol described in Methods (Fig. 9a), we successfully performed a demonstrative ODMR measurement in a sample of hippocampal neurons hosting NVcenters in nanodiamonds. During an ODMR test, the sample is continuously illuminated and irradiated by microwaves at a frequency not far from the spin resonance. The frequency is slowly scanned while the photoluminescence (PL) is monitored. When approaching the resonance frequency, the photoluminescence will be diminished and a dip corresponding to the transition frequency (2.87 GHz) between the two spin states can be observed (Fig. 9b). Under these experimental conditions, spontaneous firing was preserved, as shown in Fig. 9c for two representative cells, respectively without (CTRL) and after FND treatment, suggesting that cell functionality was maintained after the implementation of the ODMR measurement protocol (ND hosting, mW-regime laser exposure, microwave excitation, as detailed in Methods).

Discussion
Combining patch-clamp electrophysiology to detect AP and mPSC together with microelectrode arrays recordings, to monitor extracellular AP from developing neuronal networks, we investigated the functional effects of fluorescent diamond nanocrystals on primary cultures of hippocampal neurons. It is worth noting that, despite the outstanding optical properties and lack of cytotoxicity of FNDs have been largely explored 1,2 , little is known regarding the effects of FNDs seeding on neuronal activity.
Here we show that FNDs induce a marked reduction of bursts firing frequency on reconstituted hippocampal networks and reduces the percentage of neurons displaying an abrupt AP onset phase (kink). Concerning the effect on hippocampal networks, we demonstrate that the stage of FNDs seeding (7 versus 14 DIV) is critical to modify the spontaneous firing frequency, causing a drastic reduction only if nanoparticles are applied at later stages of neuronal development. This functional impairment is in good agreement with previous findings reporting that, although FNDs do not exhibit cytotoxicity, they drastically alter the network morphology 26 . More specifically, the smaller the nanodiamond size, the longer the neurites extend 20 .
Reduction of spontaneous firing frequency can be ascribed to a variety of pre-and post-synaptic factors: (1) the reduction of network excitability associated with a selective Na + or Ca 2+ channels inhibition or K + channels up-regulation 47 , (2) the alteration of glutamatergic versus GABAergic transmission and the balance of fast-spiking versus slow-firing neurons 48 , and (3) the modification of neuron morphology at the neuritic level 44,45 . To address these alternative possibilities, basic synaptic properties were measured by patch-clamp recordings. The frequency of both mIPSCs and mEPSCs events are equally reduced by FNDs, while their unitary amplitude and time course are preserved, suggesting an altered presynaptic activity induced by FND 49 , although we cannot exclude a postsynaptic effect. This is in agreement with an impaired neurite growth observed in central and peripheral primary neurons without inducing cytotoxicity 26 . However, since the depressive effect on glutamatergic and GABAergic inputs are comparable, it is reasonable to hypothesize an "indirect" aspecific damage to the network rather than an effect of FNDs on specific cellular targets. As a direct consequence of reduced GABAergic and glutamatergic activity, the network burst firing frequency is halved. Despite the reduced excitability, it appears that FNDs do not compromise the network synchronization, in good agreement with the preservation of the inhibitory-excitatory balance and adaptation properties 50,51 .
Regarding the AP generation in isolated neurons, it is relevant to underline how, even in the presence of FNDs and their aggregates, hippocampal neurons are still able to fire APs with unaltered input resistance, rheobase, firing frequency and AP waveform (Figs 7 and 8), suggesting little or no effects to the gating and expression density of the ion channels sustaining the regular AP firing. Nevertheless, our data reveal an altered dynamics of the AP initiation induced by FNDs internalization. In particular, the abrupt onset of APs, which appears as a sharp kink detected in the majority of untreated neurons, is lacking in most of FND-treated neurons. Sharpening of the spike onset was originally attributed to a cooperation of Na + channels at the axonal initial segment 52 , and later to the backpropagation of the AP from the axonal initial segment toward the soma 53 . A different interpretation, based on the "compartmentalization" hypothesis 43 , attributes the kink to the distal initiation and the current sink caused by the different size of the soma and axon. In our work the first hypothesis cannot be confirmed by experimental findings, since the Na + channels density at the soma is preserved by FNDs (Fig. 8). Apart from that, the presence of the kink in the isolated hippocampal neurons used in our study reflects a good morphologic structure of the axonal initial segment 45 and the existence of a sufficiently well-extended dendritic surface area 44 that may be comparable to the AP recording in CA3 neurons of hippocampal slices with well-preserved morphology 54 . On the contrary, dendritic disruption (possibly associated with impaired neurite outgrowth) indeed alters the AP shape, even if it does not inhibit the firing itself. Thus, a reasonable hypothesis could be that FNDs and their aggregates alter the geometrical shape of hippocampal neurons in vitro 26 by limiting the extension of the neuronal branches and, consequently, reducing the network firing frequency. This is a critical issue worth to be considered in the perspective of applying FNDs for in vivo measurements.
Finally, we successfully implemented a test ODMR detection scheme to NDs internalized in hippocampal neurons demonstrating that these cells are not affected by the implementation of the measurement protocol, which (apart from the NDs internalization) is based on the application of both MW fields and laser illumination. Since each NV − centers can be regarded as atomic-sized magnetic field sensors that can be read-out optically using ODMR, in perspective their excellent magnetic field sensitivity can be applied to detect action potential propagation in 2D along the entire neuron with high-time resolution 5 .
In conclusion, our findings give strong support to the use of FNDs for in vivo imaging and targetable drug delivery. The nanoparticles aggregation problem remains, in our opinion, the main cause of the observed alteration both at the single-cell level and in neuronal networks. Once solved this issue, FNDs could be more widely employed to "view" neuronal excitability using optical tools. It would be also useful to perform in vivo studies of FNDs delivery over longer periods of time than those used here, in order to exclude any possible long-term cytotoxicity due to the non-biodegradable nature of NDs. Council on Animal Care and approved by the local Animal Care Committee of Turin University. Hippocampal neurons were obtained from black-six mouse 18-day embryos. Hippocampus was rapidly dissected, kept in cold HBSS (4 °C) with high glucose, and digested with papain (0.5 mg/ml) dissolved in HBSS plus DNAse (0.1 mg/ ml) 31,49,55 . Isolated cells were then plated at the final density of 1800 cells/mm 2 onto the MEA and 400 cells/mm 2 on 35 mm dishes (previously coated with poly-DL-lysine and laminine). Cells were incubated with 1% penicillin/ streptomycin, 1% glutamax, 2.5% fetal bovine serum, 2% B-27 supplemented neurobasal medium in a humidified 5% CO 2 atmosphere at 37 °C. For MEA recordings, each MEA dish was covered with a fluorinated ethylene-propylene membrane (ALA scientific, Westbury, NY, USA) to reduce medium evaporation and maintain sterility, thus allowing repeated recordings from the same chip.
FND preparation. The sample under exam is produced by ElementSix TM (UK) and consists of a synthetic diamond powder produced by disaggregation of High Pressure High Temperature (HPHT) single-crystals. The size of the diamond nanoparticles nominally ranges between ∼10 nm and ∼250 nm. The crystals are classified as type Ib having a nominal concentration of single substitutional nitrogen comprised between 10 ppm and 100 ppm.
Crystals also present impurities (superficial graphitic layers) and contaminants adsorbed on the surface, which are by-products of the fragmentation process. A preliminary cleaning process was therefore necessary to remove these impurities from crystals surface.
NDs were dispersed in an acid solution containing H 2 SO 4 and HNO 3 (volumes ratio 9:1) for 72 hours at 75 °C. After filtration, nano-crystals still exhibited an acid behavior, and were therefore neutralized by means of a NaOH bath (2 hours, 90 °C). Processed powders were deposited over a suitable substrate for ion irradiation (1 × 1 cm 2 squares) creating an uniform layer with a thickness of ~20 µm (see Fig. 1b).
Since, due to the low amount of vacancies in the pristine powder, only a few percent of the substitutional nitrogen present in the diamond is involved in the formation of NV centers, ion-induced damaging is necessary to increase the vacancy density and therefore the total amount of luminescent NV centers upon subsequent thermal activation.
The samples were implanted with a 2 MeV H + ion broad beam at the AN2000 accelerator facility of the INFN Legnaro National Laboratories (INFN-LNL). An implantation fluence of 5 × 10 15 cm −2 was delivered, guaranteeing the introduction of a vacancy density of 5 × 10 18 cm −3 that is sufficiently far from the graphitization threshold of implanted diamond 56 . As resulting from the SRIM 57 , Monte Carlo code simulation reported in Fig. 1a, the penetration depth of 2 MeV proton is sufficient to cross through the thin (∼20 µm) film of nanodiamond powder, guaranteeing a nearly-uniform vacancy creation over the totality of the nano-particles.
Thermal treatment (800 °C for 1 h in nitrogen environment) was performed after ion implantation to promote the formation of NV − centers. Assuming a NV formation efficiency of ∼10% 27 , the described processing protocol guarantees that even the smaller nano-particles, i.e. the ones that can be easily uptaken by cells, contain several tens of NV color centers.

Microelectrode arrays recordings.
Multisite extracellular recordings were carried out with the MultiChannel System MCS, (Reutlingen Germany). Data acquisition was controlled through the MC_Rack software, by setting the threshold for spike detection at −30 μV and sampling at 10 kHz. Recording duration was 2 minutes. Burst analysis was performed using Neuroexplorer software (Nex Technologies, Littleton MA USA) after spike sorting. A burst is defined as a group of spikes with decreasing amplitude 42 , thus we set a threshold of at least 3 spikes and a minimum of 10 ms duration. We set interval algorithm specifications such as maximum interval to start burst (0.17 s) and maximum interval to end burst (0.3 s) recorded in 0.02 s bins. Burst analysis was conducted considering two main parameters: mean frequency and number of bursts. In order to examine synchronicity, cross-correlation probability vs. time diagrams were constructed by means of Neuroexplorer software, using ±0.5 s and 5 ms bin size. Raster plots were created by using the specific function on Neuroexplorer software.

Cell incubation with FNDs and confocal imaging.
Hippocampal neurons incubation with FNDs was performed at 7 DIV or 14 DIV, as specified in the Results. For both MEAs and plastic dishes, half of the culture medium (i.e. approximately 1 ml) was replaced with FND-medium to reach a final concentration of 40 µg/ml. Incubation with the cytoplasmic membrane dye was performed 24 hours later. For confocal imaging, hippocampal neurons were plated on 35 mm dishes (ibidi GmbH, Planegg/Martinsried, Germany).
Imaging was performed using a Leica TCS SP5 confocal system (Leica Microsystems) equipped with an argon ion and a 561 nm DPSS laser. Cells were imaged using a HCX PL APO 63×/1.4 NA oil immersion objective at a pixel resolution of 0.08 × 0.08 × 0.3 μm.
The luminescent emission from the FNDs was excited by 561 nm laser, while the emission was collected in the 650-750 nm spectral range. The same excitation wavelength was uneffective in untreated neurons. Green fluorescence for intracellular staining was obtained by 488 nm wavelength. Image analysis was performed using ImageJ software.
Patch-clamp experiments. Patch-clamp experiments were performed using pclamp software (Axon Instruments, Foster City, CA) and a 12-bit A/D Tecmar Laboratory Master board (125 kHz). All the experiments were performed at room temperature (22-24 °C Voltage-clamp experiments. Miniature postsynaptic currents (mPSCs) were recorded under voltage-clamp conditions from mature hippocampal neurons (18 DIV) as detailed in 47,49,55,58 . For both controls and FND-incubated cells, holding membrane potential (V h ) was maintained at −70 mV and data acquired for two minutes.
Miniature postsynaptic currents were acquired with sample frequency ranging between 10 kHz and filtered at half the acquisition rate with an 8-pole low-pass Bessel filter. Recordings with leak currents ≥100 pA or series resistance ≥20 MΩ were discarded. Miniature inhibitory and excitatory currents were recorded by superfusing the whole-cell clamped postsynaptic neuron with Tyrode solution containing (in mM): 2 CaCl 2 , 150 NaCl, 1 MgCl 2 , 10 Hepes, 10 glucose, 4 KCl (pH 7.4). Tetrodotoxin (0.3 µM) (Tocris Cookson Ltd, Bristol, UK) was added to block voltage dependent Na + channels and spontaneous action potentials propagation. GABAergic and glutamatergic currents were respectively isolated by D-AP5 (50 μM) and Picrotoxin (100 μM). The standard internal solution contained (in mM): 90 CsCl, 20 TEACl, 10 EGTA, 10 glucose, 1 MgCl 2 , 4 ATP, 0.5 GTP and 15 phosphocreatine (pH 7.4). For unitary mini analysis, all double or multiple events were discarded 49,55 . The rheobase was evaluated as the minimum amount of injected current necessary to induce AP firing, when holding neurons at V h = −70 mV (see ref. 36  These trials were performed in physiological Tyrode saline solution enriched with kynurenic acid (1 mM) and picrotoxin (100 μM). Action potential firing was evoked by applying 2 second current pulses injection (from 100 pA to 300 pA, 10 pA steps). Analysis of AP shape, phase-plane plot and data interpretation on AP trains were performed as previously detailed 36 . ODMR. The ODMR technique is based on the dependence of the photoluminescence (PL) emission intensity from NVdefects in diamond on the spin state of the defect, thus enabling different types of electro-magnetic detection schemes 5,8,18,46 . In our setup (Fig. 9a) a non-resonant 532 nm continuous laser excitation was employed to initialize the NVcenters in the bright m s = 0 spin state. Then, a microwave field of −20 dBm power was generated at variable frequency in the 2.80-2.96 GHz range, by positioning a micro-strip antenna in close proximity of the investigated sample. This enabled the observation of a local minimum in the PL emission intensity corresponding to the resonance frequency (2.87 GHz) of the ground-state zero-field splitting between the |0> and |±1> states 46 . In this way we have exposed the cells to a measurement protocol that in perspective would enable detecting in a non-destructive way the weak magnetic field associated to the action potential firing 5 .
Statistical analysis. Data are given as the mean ± SEM for the number (n) of cells. Statistical significance was estimated with paired Student's t tests in case two groups of measurements had to be compared and with a one-way ANOVA followed by post hoc Bonferroni analysis in case more than two groups had to be compared with one another.
Data were found statistically significant when p < 0.05 (*). Data analysis was performed with pClamp and Origin software (OriginLab Corporation, Northampton, MA, USA).
Data availability statement. The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.