Introduction
Wiring the central nervous system demands precise formation and maintenance of neuronal connections. Functional connections between neurons are established by means of micron-sized interfaces, called synapses [
19,
44]. Synaptic activity and the adjoined opening of gated calcium channels generate calcium transients that drive morphological changes such as dendritic growth and arborization, but which can also influence synapse strength [
2]. This dynamic remodeling of both neurites and synapses fosters improved communication between neurons allowing synchronous functional activity, thereby reinforcing the overall connectivity of the neuronal network. All long-lasting adaptations of the brain, including learning, memory, addiction and chronic pain sensation, rely on the continuous remodeling of neuronal network connectivity [
2]. And, disruption of this process is a hallmark of numerous neurological diseases, including schizophrenia, major depressive disorder and Alzheimer’s disease (AD) [
40]. For example, the cognitive impairments witnessed in AD patients correlate with synapse and dendritic loss, as well as with a reduction of the brain activity, indicating an overall decrease in neuronal connectivity [
16,
35,
60]. Therefore, therapeutic developments for neurodegenerative disorders are directing their focus to identifying regulators that promote neuronal connectivity or prevent the loss thereof.
The dense, three-dimensional organization and spatial heterogeneity of the brain makes studying neuronal connectivity in vivo a daunting task, which is not amenable to upscaling. Therefore, systematic screening efforts most often rely on simplified models such as neuronal cell cultures. Although some immortalized tumor cells (e.g.
, SH-SY5Y human neuroblastoma cells, NT2 human teratocarcinoma cells, PC12 rat pheochromocytoma cells) can be differentiated to take on a neuron-like phenotype (as evidenced by e.g.
, neurite outgrowth, expression of synaptic markers, induced or subtle spontaneous calcium activity), none fully recapitulate the full feature set of a physiologically connected neuronal network [
1,
11,
18,
23,
27,
29]. Neurons derived from human induced pluripotent stem cell have a much higher translational value in comparison with other cell lines and primary cultures prepared from rodents, but the differentiation process is very labor and time intensive [
15,
25,
36]. That is why, as yet, primary neuronal cultures represent the model of choice for genetic and pharmacological high-throughput screens [
3,
6,
8,
9,
28,
31,
43,
53,
55]. However, most of these screens tend to focus on one or two specific readouts such as neuron number [
6,
31], neurite outgrowth [
3,
6,
8,
31] or synapse density [
53,
55]. This reductionist approach can be misleading since it neglects multifactorial effects. For instance, measured changes in synapse density can be a direct reflection of real alterations in the number of synapses [
28,
43], but could also be influenced by changes in the neurite density [
62]. Furthermore, solely gauging morphological correlates may mask potential changes in functional connectivity. Indeed, it has already been shown that spontaneous activity in primary cultures does not scale linearly with synaptic density [
5], and that morphological aberrations of primary neuronal networks do not always result in functional impairments [
62]. In order to accurately quantify the overall neuronal connectivity in primary cultures, information gained from several readouts should thus be combined. Hence, instead of using one or a few descriptors, here we comprehensively assess the major morphological and functional correlates of neuronal network connectivity, and we integrate them to accurately map changes between subsequent maturation stages in vitro. In a targeted assay, we identified dual leucine zipper kinase (DLK) inhibition as a positive modulator of neuronal connectivity in unperturbed cultures and as a neuroprotector in cultures grown under suboptimal or compromised conditions.
Discussion
The continuous remodeling of the neuronal network is crucial for learning, memory and behavior, but is disrupted in several psychiatric and neurodegenerative disorders [
40,
49]. Identification of novel therapeutic targets requires a method that is able to quantify neuronal network connectivity over time with high accuracy and throughput. To our knowledge, the high-throughput analysis proposed in this paper is the first to comprehensively gauge neuronal connectivity in primary cultures by including descriptor sets reporting on the dendrite network, synapse markers, nuclei and calcium bursting activity. Other high-throughput studies aimed at finding regulators of neuronal connectivity [
3,
6,
9,
28,
43], have mainly focused at one or few readouts such as neuron number [
6] and neurite outgrowth [
3,
6], synapse density [
53,
55] or calcium responses [
65]. The strength of an integrative approach lies in the fact that it can account for a number of potential sources of bias. For instance, when only considering synapse density, observed changes could be the result of true variations in the synapse number in cultures with comparable dendrite density [
28,
43], but they could equally arise from an altered dendrite density with preserved synapse count [
62], an increased density of only one or both synaptic partners (pre/post) [
43], an increased spot size of one or both synaptic partners, or an altered clustering of neuronal somata and fasciculation of dendrite bundles. Furthermore, the inclusion of calcium data as a proxy for spontaneous electrical network activity allows determining whether alterations in the network morphology also have a functional impact and vice versa. For example, previously published data by our group shows that overexpression of human tau in hippocampal cultures decreased the neurite density but increased the synaptophysin density [
62]. At a functional level, the percentage of active neurons and frequency of synchronous bursts decreased while the synchronicity was preserved. This underscores the argument that conclusions may vary depending on the descriptors used and that an integrative approach is desirable.
In line with other studies, we found an increase in dendrite density and synapse density with culture age, and - although not linear – a strong increase in the synchronous bursting behavior of the network at later time points (> 10 DIV). The time scales used in literature range from DIV6 to DIV21, which corresponds with the time range we identified in our initial experiment to be the time span in which dendrite density, synapse density and functional activity increased drastically. The decrease in neuronal cell count with culture age confirmed previous findings [
28,
33], but we now also showed that certain treatments affect cell culture composition, in particular the ratio of nuclear and glial cells (e.g.
, AraC). To our knowledge, no other connectivity studies have made such a distinction and it shows the value of including nuclear descriptors.
When applying PCA to the unfiltered morphological descriptor set, we unveiled distinct temporal trajectories [
14], suggesting that the culture age could serve as an indicator of connectivity. This was confirmed by the fact that a RFC could predict culture age with an accuracy of 95%. To include functional parameters in the analysis, we defined a comprehensive connectivity score. Using this score as readout, we confirmed that inhibition of mTOR activity (using rapamycin) impairs the development of the dendritic network [
34,
58]. Nevertheless, rapamycin has been used to slow down or block neurodegeneration in mouse models of Alzheimer’s, Parkinson’s and Huntington’s disease, through the induction of autophagy that cleared accumulated autophagosomes and/or aggregated proteins [
4,
37,
52]. Therefore, an extension of this work could focus on assessing the effect of rapamycin on neuronal connectivity under conditions of induced toxic protein aggregation. Antagonists of NMDA-R, such as memantine and MK801, were also tested. These antagonists have the potential to block the excessive NMDA-R activity in many neurodegenerative diseases, hereby reducing the increased calcium influx in the neurons [
41,
51]. NMDA-R antagonists can however also block the normal function of these receptors, which was suggested by the overall negative effect on the connectivity score for MK801 (in which the functional descriptors had a major contribution). The score indicated that memantine treatment had only a slight negative impact on neuronal connectivity in comparison with MK801, which may be the result of memantine having a shorter dwell time on the NMDA-R then MK801 [
41]. Tubastatin, a HDAC inhibitor that improved cognitive deficits in mouse models of Alzheimer's Disease by improving microtubule stability [
72], did not have a negative, nor a neurotrophic effect on the overall neuronal connectivity in control conditions, and may therefore be further explored in compromised conditions. Treatment with another HDAC inhibitor, SAHA, resulted in a decreased neuronal connectivity at later time points. This may be due to the fact that SAHA targets multiple HDAC classes, whereas tubastatin only targets HDAC6 [
69].
The sole compound that showed a clear neurotrophic effect in unperturbed primary cultures was an inhibitor of DLK, which is why we chose to pursue the efficiency of this compound in compromised conditions. DLK, also known as MAP3K12, is an upstream regulator of the JNK stress response pathway, which becomes activated in both acute and chronic neurodegenerative conditions [
54]. Its activation induces a broad transcriptional injury response (via c-Jun and ATF4) [
66]. The fact that it is upstream in this highly conserved pathway makes that DLK inhibition has a broad action range, at least in preclinical models. Pharmacological or genetic DLK inhibition protects against excitotoxicity [
50], growth factor deprivation [
22,
42,
68], amyloid and tau pathology [
32,
38], nerve crush and traumatic brain injury [
66,
71] retinal ganglion cell degeneration [
67,
68] and SOD1-mediated neurodegeneration [
38]. At this moment, one inhibitor is in Phase I clinical trial for Amyotrophic Lateral Sclerosis (ALS) by Genentech (Roche) [
30]. Our results now show that also in unperturbed primary cortical cultures, DLK inhibition results in enhanced morphofunctional connectivity. Yet, it should be noted that the inhibitor treatment started 4 h after plating. It is possible that the cells already experienced stress during the dissociation procedure, which could have been attenuated by DLK inhibition.
We found that DLK inhibition could prevent and even rescue neurodegeneration induced by hTau.P301L overexpression. This aligns with literature data where double mutant mice (Tau
P301L;DLK
cKO) show attenuated cell loss in the subiculum compared to single mutant (Tau
P301L) mice without having altered tau pathology [
38]. Our in vitro approach now also shows that hTau-P301L-induced neuron loss is accompanied by impaired dendrite formation, synapse density and functional activity and that this can be prevented and rescued by pharmacological DLK inhibition. DLK inhibition reduced the increased levels of unphosphorylated and phosphorylated (ser63) c-Jun that were detected in cultures overexpressing hTau.P301L.
The omission of anti-oxidants from the culture medium of primary cortical neurons resulted in gradual impairment of morphofunctional connectivity, which could be prevented by DLK inhibition as well. Neurons are particularly sensitive to reactive oxygen species (ROS) and most neurodegenerative disorders are associated with increased oxidative stress [
61], making our in vitro model attractive for drug screening. In line with our results, it was described before that the SOD1
G93A mouse model for ALS exhibits neuronal cell death due to an impaired oxidative stress defense and that this is accompanied by aberrant JNK pathway activation [
38]. Double mutant mice (SOD1
G93A;DLK
cKO) show enhanced neuronal survival and myelinization as well as reduced neuroinflammation compared to single mutants (SOD1
G93A), resulting in enhanced grip strength and longer life span of the former model. In AO-depleted cell culture, we did not find evidence for the activation of the JNK pathway, but we did detect a downregulation of the pathway upon DLK inhibition.
Finally, we also verified whether DLK inhibition could sustain morphofunctional connectivity in aging cultures, as it is known that JNK signaling is elevated in the aging brain [
70,
73]. However, the effect was rather limited. Increased variability across aged cultures and reduced sensitivity of the morphological readout due to the very high density of dendrites covering most of the culture plate could have masked a potentially crisper effect of DLK inhibition on ageing. Across experiments, we found differential efficacy of both DLK inhibitors in various growth conditions. These discrepancies are likely the result of different treatment duration and starting times, as well as different target selectivity of both inhibitors [
46,
54]. Literature reports on the selectivity of both compounds show that GNE-3511 had a K
i(DLK) below 0.5 nM and a pJNK IC
50 of 30 nM, which was considerably lower than the IC
50 for other kinases (MKK4 > 5.000, MKK7 > 5.000, JNK1 = 129, JNK2 = 514, JNK3 = 364, MLK2 = 767, MLK3 = 602 nM), except for MLK1 (IC
50 = 67.8) [
46]. For GNE-8505, the K
i(DLK) and pJNK IC
50 values were higher than for GNE-3511 (4 nM and 144 nM, respectively), yet the selectivity for DLK over other kinases was markedly better (MKK4 > 5.000, MKK7 > 5.000, JNK1 > 10.000, JNK2 > 10.000, JNK3 > 10.000, MLK1 = 3.500, MLK2 = 5.150, MLK3 > 10.000 nM). This illustrates that the DLK inhibitors used in this study can target other kinases, especially in the higher dose range, which might explain the differential effects of both compounds.
To increase the sensitivity of the current approach, it could be advantageous to include more synapse markers so as to better map the full landscape of synapse types (e.g.
, inhibitory vs excitatory). To bypass spectral limitations, one could resort to the use of narrow-emission band labels, such as quantum dots [
20], or divert to cyclic immunofluorescence staining protocols [
26,
39]. The latter would however lower the throughput and put a higher demand on downstream analyses (e.g.
, image registration). A particular caveat of the current approach was the high variability of the functional readout. To strengthen the sensitivity of this assay one could resort to imaging larger cell populations by increasing the field of view. Previous studies have revealed the co-existence of separate networks of high connectivity within a large neuronal network that may not always fire in sync [
48,
57] Identification of stratified connectivity patterns may therefore expose more subtle modifications of the functional connectivity. Selective addition of chemical stimuli (e.g.
, glutamate) could further unveil cell type specificity as well as differences in spontaneous and induced functional activity [
10].