Control of Axonal Growth and Regeneration of Sensory Neurons by the p110δ PI 3-Kinase

The expression and function of the 8 distinct catalytic isoforms of PI 3-kinase (PI3K) in the nervous system are unknown. Whereas most PI3Ks have a broad tissue distribution, the tyrosine kinase-linked p110δ isoform has previously been shown to be enriched in leukocytes. Here we report that p110δ is also highly expressed in the nervous system. Inactivation of p110δ in mice did not affect gross neuronal development but led to an increased vulnerability of dorsal root ganglia neurons to exhibit growth cone collapse and decreases in axonal extension. Loss of p110δ activity also dampened axonal regeneration following peripheral nerve injury in adult mice and impaired functional recovery of locomotion. p110δ inactivation resulted in reduced neuronal signaling through the Akt protein kinase, and increased activity of the small GTPase RhoA. Pharmacological inhibition of ROCK, a downstream effector of RhoA, restored axonal extension defects in neurons with inactive p110δ, suggesting a key role of RhoA in p110δ signaling in neurons. Our data identify p110δ as an important signaling component for efficient axonal elongation in the developing and regenerating nervous system.


INTRODUCTION
Phosphoinositide 3-kinases (PI3Ks) are a family of lipid kinases which regulate a wide variety of biological responses in different cell types [1]. In the nervous system, PI3K activity contributes to the establishment of appropriate connectivity by regulating various cellular processes, including neuronal differentiation, survival, migration, extension and guidance [2][3][4][5]. The 8 isoforms of mammalian PI3Ks have been grouped into three classes (I, II, and III) [1]. The class IA subset of PI3Ks signal downstream of Tyr kinases and Ras, and are heterodimers composed of one of three p110 catalytic subunits-p110a, p110b or p110d-in complex with one of the three regulatory subunit (collectively called 'p85s'). Detailed information on the tissue distribution of the p110a and p110b isoforms is not available, although evidence for a broad expression of both isoforms has been presented [6][7][8][9]. On the other hand, p110d is known to be highly enriched in leukocytes [7,10,11]. Gene-targeting studies in the mouse have uncovered non-redundant roles of specific p110 PI3K isoforms in immunity, metabolism and cardiac function [12,13]. In contrast, the expression and function of the distinct PI3K isoforms in the nervous system have not been investigated.
Here, we report that expression of p110d PI3K is highly enriched in the embryonic nervous system in the mouse at stages concomitant with the extension and guidance of neuronal processes. Genetic or pharmacological inactivation of p110d in sensory neurons led to a reduction in PI3K signaling, increased sensitivity to growth cone collapse and deficient axonal elongation under limiting growth conditions. In addition, mice with inactive p110d show impaired axonal regeneration and functional recovery following a sciatic nerve crush injury. These results identify p110dmediated PI3K signaling as a crucial component for efficient axonal elongation.

RESULTS AND DISCUSSION p110d expression is highly enriched in the nervous system
To date, no detailed information on the distribution of the distinct PI3K isoforms in neuronal tissue has been reported. In order to analyze the expression of p110a and p110d, reporter mice were generated in which a b-Gal/LacZ reporter gene was inserted into the endogenous p110a or p110d gene locus by homologous recombination [14,15]. An internal ribosome entry site (IRES)-LacZ sequence was targeted into the last exon of the p110 gene, immediately after the stop codon (schematically shown for p110d in Figure 1A). This allowed independent production of the p110 protein and the b-Gal enzyme from the same bicistronic mRNA, encoded by the p110 gene promoter(s). These mice are hereafter referred to as p110a lz and p110d lz mice.
As expected, the p110a isoform of PI3K was found to be widely expressed ( Figure 1B), and high expression of p110d in the embryonic liver, the principal location of hemopoiesis, supports previous reports on the expression of p110d in adult mice leukocytes [7,11]. Unexpectedly strong signals of p110d were also observed in the nervous system, especially in the spinal cord, dorsal root ganglia (DRG), cranial sensory ganglia and peripheral nerves ( Figure 1C). Stainings at various time points showed that this pattern of LacZ expression generally reflects the appearance of differentiated neurons in both the central and peripheral nervous system. At E12.5, for example, the p110d/LacZ signal followed the wave of retinal ganglion cell differentiation in the central retina ( Figure 1D), and was not detected in the neuroblast layer at any developmental stage analyzed (E12-E18; Figure 1D and data not shown). In addition, p110d/LacZ expression was enriched during neuronal migration, for example at E12.5 in the facial motor nuclei within the hindbrain during movement to the final rhombomere location ( Figure 1E). Axonal processes and cell bodies within the vomeronasal epithelium also expressed p110d/ LacZ, shown in Figure 1F. Cross sections through the spinal cord at lumbar levels revealed highly enriched p110d/LacZ staining in the DRG, interneurons of the spinal cord and the spinal motor neuron pool ( Figure 1G).
In adult mice, high p110d/LacZ expression was also present in neurons, for example in specific brain regions, including the hippocampus, cortex and thalamus (Figure S1A-C). Immunoblotting of brain extracts confirmed the enrichment of p110d protein in distinct brain regions ( Figure S1D). This contrasts with the uniform distribution of p110a and p110b, as well as various forms of the p85 regulatory subunits ( Figure S1D).

Inactivation of p110d increases the vulnerability of sensory neurons to growth cone collapse and decreases axonal extension
To determine the contribution of p110d to neuronal development and function, we analyzed mice in which p110d was inactivated as a result of the introduction of a germline point mutation which renders the kinase inactive (p110d D910A ; [14]. Homozygous p110d D910A/D910A mice, hereafter referred to as p110d kinaseinactive (KI) mice, are viable and fertile [14]. In these mice, expression of the mutated p110d protein and the other PI3K subunits was equivalent to that of the wild-type (WT) proteins in brain homogenate ( Figure S2), demonstrating the absence of compensatory PI3K expression. Gross morphology of the nervous system (data not shown) and hippocampus-dependent learning behavior ( Figure S3) were unaffected in p110d KI mice, suggesting that the establishment and functioning of the neuronal circuitry required for complex behavioral tasks does not depend on p110d activity.
We next assessed the responsiveness of neurons to PI3K inhibition using the pan-PI3K inhibitor LY294002 [1], which induces growth cone collapse in sensory neurons [16,17]. Our analyses showed that this response was significantly greater in p110d KI than in WT DRG neurons (Figure 2A). This indicates that DRG neurons with inactive p110d are more sensitive to global PI3K inhibition, and also provides evidence that the remaining PI3K isoforms could not compensate for the loss of p110d activity. IC87114, a p110d-selective small molecule inhibitor [18], also induced growth cone collapse in WT DRG neurons, but had little effect on p110d KI DRG neurons (Figure 2A), indirectly confirming the selectivity of this compound. Responsiveness of DRG neurons to physiological stimuli that utilize PI3K/Akt signaling was also assessed. Growth cone collapse induced by the axon guidance molecule Sema3A, known to decrease PI3K signaling [17,19], was 50% higher in p110d KI DRG neurons at low concentration of Sema3A ( Figure 2B, C). Integrin activation through the substrate laminin is known to activate PI3K signaling in the growth cone [20]. Lamininmediated axonal elongation in p110d KI DRG neurons was reduced by almost 30% at lower (10 mg/ml) but not at higher (20 mg/ml) concentrations of laminin ( Figure 3). Taken together, these observations uncover a p110d PI3K signaling pathway, important for the maintenance of optimal axonal outgrowth in an inhibitory environment and under lower substratum availability. Genetic or pharmacological inhibition of p110d in cultured DRG neurons did not alter apoptosis (as measured by apoptotic nuclei; data not shown), indicating that this pathway is independent of PI3K-mediated survival responses.

Reduced axonal regeneration in the injured sciatic nerve of mice with inactive p110d
Given that the absence of p110d activity limits axonal outgrowth in embryonic neurons, we assessed the axonal growth potential of adult p110d KI neurons. Adult peripheral nerve axons are capable of functional regeneration in mammals, with a crush injury of the sciatic nerve being a well-established injury model to study axonal growth [21]. On day 3, extension of regenerating neuronal fibers 2 mm distal from the injury site was reduced in p110d KI mice ( Figure 4A,B), without any differences in the presence of  p110d/PI3K Function cytoskeletal breakdown products 4 mm distally from the site of injury (data not shown), indicating that loss of p110d activity does not alter normal axonal degeneration. p110d activity has been shown to be essential for CSF-1-driven in vitro chemotaxis of macrophages [22,23], which might affect axonal regeneration due to impaired inflammatory response at the injury site in p110d KI mice. No changes in recruitment of macrophages (stained by F4/80) into the injured nerve were detected ( Figure 4B). Next, we compared the capacity of DRG soma to upregulate growth-associated SPRR1A (small proline-rich repeat protein 1A) during regeneration. SPRR1A has been shown to peak 1-2 weeks following sciatic nerve injury in adult mice and its depletion reduces axonal outgrowth in vitro [24]. We co-stained for ATF3 (Activating Transcription Factor 3), which is produced de novo in sensory neurons following sciatic nerve injury and is widely used as a marker for nerve injury [25]. No significant difference was observed in the percent of ATF3-positive neurons in injured DRGs obtained from WT and p110d KI mice in L4 DRGs 7 days after nerve injury, demonstrating that the sciatic nerve injury was equivalent in the two groups (data not shown). However, the regeneration marker SPRR1A was significantly reduced in L4 DRG soma of p110d KI mice ( Figure 4C), indicating that loss of p110d function impairs regenerative capacity of adult sensory neurons.

Inhibition of axonal regeneration correlates with impaired functional recovery
We next assessed whether the anatomical evidence for p110ddependency for optimal axonal regeneration correlated with functional recovery using an automated quantitative gait analysis system, the CatWalk, to assess recovery of locomotion following nerve injury on days 1, 3, 7, 10, 14 and 21 [26]. Prior to injury, gait analysis assessed by the CatWalk on a number of locomotor parameters revealed no differences between WT and p110d KI mice ( Figure 5A, B, S4). In contrast, following unilateral injury to the sciatic nerve, p110d KI mice showed a significant decrease in the recovery of the ability to bear weight on the injured paw ( Figure 5A, B). Whilst both WT and p110d KI mice display functional recovery during the first 10 days post-lesion, the relative  paw pressure intensity in p110d KI mice was significantly lower in comparison to WT mice ( Figure 5B; p,0,001, 2-way repeated measured ANOVA). These differences were most apparent at later time-points (day 10 and day 14 post-lesion; Figure 5B; p,0.05, Tukey test). These results indicate that following sciatic nerve injury, the lack of functional p110d led to a decreased ability of the axons to undergo regenerative growth, which in turn led to decreased functional recovery.

Reduced Akt and increased RhoA/PTEN signaling in neurons with inactive p110d
We next investigated the effect of p110d inactivation on signaling in neurons. PI3K signaling drives many aspects of neuronal morphology through the coordinated phosphorylation of proteins that regulate cytoskeletal dynamics, protein synthesis, and transcriptional activity. Akt is an important effector through which PI3K controls axon elongation and morphological responses induced by neurotrophins [27,28]. A substantial decrease in activatory Akt phosphorylation was observed in p110d KI DRG neurons cultured in the presence of NGF ( Figure 6A). In contrast, phosphorylation of GSK-3b, an effector of the PI3K/Akt pathway in many cell types [29] and a crucial determinant of axonal growth and guidance [27,30,31], was not affected ( Figure 6A). This lack of inactivation of GSK-3b may allow the observed normal neurite outgrowth under non-limiting conditions in p110d KI mice ( Figure 3) and suggests the presence of alternative, p110d-independent regulatory pathways for GSK- p110d/PI3K Function 3b. Another member of the signaling pathway downstream of PI3K activation is mTor and its effector p70S6K, which controls the initiation of protein synthesis [1]. DRG neurons from p110d KI mice showed a substantial decrease in p70S6K phosphorylation, whilst levels of MAPK phosphorylation and Bcl-xl protein were not affected ( Figure 6A). These results indicate that p110d PI3K activity exerts control over the Akt/p70S6K pathway, which is not compensated for by p110a or p110b.
p110d has recently been demonstrated to inhibit the activity of the tumor suppressor PTEN through a pathway involving RhoA [23]. Similary, in p110d KI DRGs and brain homogenates, PTEN lipid phosphatase activity was constitutively elevated ( Figure 6B). In addition, GTP-loading of RhoA, but not Rac, was significantly increased in p110d KI brain extracts ( Figure 6C). These observation is in line with the idea that p110d can suppress cellular RhoA but not Rac activity under basal conditions [23]. RhoA is a critical mediator of the inhibitory effect of several axon guidance molecules and myelin associated inhibitors [32,33]. Inhibition of RhoA, or its downstream effector ROCK, is able to restore outgrowth in an inhibitory environment provided by myelin or myelin associated inhibitors [32,33]. In order to test the significance of increased RhoA function for p110d signaling during axonal elongation, we inhibited ROCK using the small molecule inhibitor Y27632 [34][35][36]. As expected, axons extending from p110d KI DRG explants were significantly shorter than WT axons ( Figure 6D). However, treatment with Y27632 restored axonal length in p110d KI neurons to the lengths seen in WT neurons ( Figure 6D). These data are consistent with a model whereby inactivation of p110d leads to increases in ROCK activity as a consequence of higher levels of active RhoA. Previous work has indicated RhoA/ROCK signaling in the control of PTEN [37], thus raising the possibility that PTEN activity in neurons functions downstream of RhoA. Such deregulation of signaling by p110d is consistent with the phenotypes observed in this study [17].

Conclusion
The major finding of this study is the identification of a function of the p110d PI3K in controlling effective axonal elongation under less favorable conditions and during insult to the nervous system. This is consistent with the idea that although p110d is similar to p110a and p110b in terms of structure and substrate specificity, it is restricted in its expression and function ( Figure 1; [7,11,38]). It remains to be determinded if the p110a and p110b isoforms of PI3K play similar roles in neurons. Homozygous inactivation of p110a or p110b leads to embryonic lethality [8,9,15], precluding investigation in this area until conditional p110a/p110b mutant mice will become available. p110d is considered to be an interesting therapeutic target in inflammation and auto-immunity [12,14,39], and the development of small molecule inhibitors against p110d is in progress [18]. The data reported here suggest that p110d inhibitors may not have adverse effects on the steadystate functioning or the development of the nervous system. Nonetheless, caution should be exercised given that these compounds may have undesirable effects under conditions of nerve injury or ongoing neurological degeneration.

Mice
Gene targeting to create the p110a lz mice and p110d lz mice has been described elsewhere [14,15]. All mice were littermates and backcrossed for 10 generations onto the C57B16/J strain. Animals were maintained in individually-ventilated cages on a 12 h lightdark cycle, with free access to food and water. All experiments were undertaken in accordance with the UK Animals (Scientific Procedures) Act 1986.

Sciatic nerve crush
For in vivo regeneration studies, 6 to 8 week-old male mice were used. The mice were assessed before surgery to establish baseline-walking patterns. Animals were anesthetized with a mixture of medetomidine (0.5 mg/kg) and ketamine (75 mg/kg), and the left sciatic nerve was exposed. A crush injury was performed 10 mm distal to the obturator tendon using forceps compression (10 sec) and the crush site labeled with lamp black. The muscle and skin layers were sutured and animals were allowed to recover in the cage post-operatively.

The CatWalk
The CatWalk gait analysis system was used to assess functional recovery of locomotion following sciatic nerve crush injury [26]. The animal traversed a meter long walkway with a glass floor and 2 perspex walls spaced 8 cm apart, housed in a darkened room. Light from 2 encased white fluorescent tubes entered the glass floor through the distal edge of the glass, and was totally internally reflected. Light scatters only where a paw contacts the glass, illuminating the area of paw contact. This reflected light was captured using videocamera (Sentech 705, 8.5 mm, f = 1.4, variable focus and variable iris) equipped with a wide-angle objective and a frame grabber (Matrix Vision SG-board) connected to a PC running the CatWalk 500 software for capture and analysis [26]. Each mouse ran across the CatWalk one day before surgery to establish baseline locomotor parameters. Following surgery, animals ran the Catwalk on days 1, 3, 7, 10, 14 and 21. The program was set to capture the paw prints from the middle section of the run. At least 2 runs per animal were performed on each day. Data was analyzed by labeling all areas containing one or more pixels above a certain analysis threshold. In a second interactive pass, these areas were assigned to each of the paws (left and right fore and hind paws: LF, RF, LH, RH). Data generated from the program was exported to Excel, yielding several parameters including average area and intensity for each paw, the regularity index, and duration of swing and stance phase. Statistical significance was evaluated using twoway repeated measures ANOVA and Tukey post-hoc comparisons (Sigma Stat 3.0.1, SPSS Inc.).

Morris water maze
Mice for behavioral testing were housed in groups of 2 to 4 in individually-ventilated cages, with food and water ad libitum and maintained on a 12 h light-dark cycle. Mice tested were males p110d/PI3K Function between the ages of 2 and 6 months. The Morris water maze protocol has been described previously [40]. Briefly, the mice were trained in a 1.5 m pool with a 10 cm platform. The water was maintained at 24-27uC, and made opaque with non-toxic white paint. The animals received 12 training trials per day, in blocks of 4, with 1 h in between each block, and a 90 sec maximum swim time. On days 3 and 5, a probe trial was given, in which the platform was removed and the animal was allowed to swim for 90 sec before being removed from the pool. The movement of the animals whilst in the pool was videotaped and recorded by a computer tracking system (HVS Image, Hampton, UK). The behavioral data were analyzed with the 'HVS water program' and Sigmastat (SYSTAT Software, SSPS Science Inc).

Neuronal cultures
Age-matched WT and p110d KI embryos were isolated at the appropriate age, and transferred into ice-cold DMEM. DRGs were dissected from E13.5 mouse embryos and extramesenchymal tissue was removed using a sharpened 0.2 mm tungsten wire. DRGs were then plated on glass coverslips previously coated with poly-L-lysine and laminin (both at 20 mg/ml; Sigma). For primary neuronal cultures, DRGs were incubated in trypsin (1 mg/ml, diluted in HBSS) for 10 min at 37uC, and dissociated using a firepolished Pasteur pipette. Neurons were either cultured at low density (50 cells/mm 2 ) for neurite outgrowth assays on glass coverslips coated as described above, or in laminin-coated 6-well dishes (150 cells/mm 2 ) for biochemical analysis. Explants and primary DRG neurons were incubated at 37uC/5% CO 2 for 24 h in DMEM/10% FCS/Pen/Strep supplemented with 20 ng/ml NGF (Promega). Pharmacological inhibitors were used as described; LY294002 was purchased from Calbiochem. The collapse assay using purified Sema3A-Fc or PI3K inhibitors was performed as previously described [17].

Immunocytochemistry
Neuronal cultures were treated as indicated and paraformaldehyde-fixed (4% paraformaldehyde/PBS/10% sucrose) for 30 min before permeabilization for 5 min with PBS/1% Triton 6100. Neurons were then labeled with Phalloidin-Alexa488 (1:50 in PBST) and anti-bIII-tubulin antibody (1:800; Covance). For each experiment, the collapsed growth cones were counted and represented as a % ratio. Each experiment was performed a minimum of 4 times, and the average % collapsed was determined. Standard errors of the mean were determined as the (standard deviation/ square root (number of experiments)). For determination of neurite length, neurons were labeled with the anti-bIII-tubulin antibody, and the KS300 program (Zeiss) was used to measure neurite length in each treatment.

GTPase and PTEN lipid phosphatase activity assays
The Rac or RhoA activation assays were performed using GST-PBD (p21-binding domain of PAK) or GST-RBD (Rho binding domain of Rhotekin), respectively, as described [23]. In brief, brain tissue was lysed in Mg 2+ lysis buffer (Upstate) and mixed with GST-PBD or with GST-RBD bound to glutathione-agarose and incubated for 1 h at 4uC. Bound protein was washed, and suspended in sample buffer. Proteins were then separated by SDS-PAGE, transferred to PVDF membranes and blotted with the indicated antibodies. PTEN lipid phosphatase activity was measured as previously described using malachite green reagent for the detection of phosphate release [23]. Similar results were also obtained using a PTEN activity ELISA kit (Echelon). Figure S1 Expression of p110d and other class IA PI3K isoforms in the brain. Coronal sections of the brain of (A) p110d lz and (B) p110d/PI3K Function WT adult mice reveal restricted expression of p110d/LacZ in several brain regions, including the cortex (Cx), hippocampus (H) and thalamus (Th). Sections were counterstained with nuclear fast red. Scale bar, 1 mm. (C) p110d expression in different brain areas as assessed by X-gal staining of adult lacZ (b-Gal) reporter mice. (D) Expression of PI3K isoforms and the CD45 pan-leukocyte marker in lysates of different brain regions and thymus of adult WT mice. CD45 was found to be expressed in thymus and not in the brain, indicating that X-gal signals do not derive from resident leukocytes in the brain. Found at: doi:10.1371/journal.pone.0000869.s001 (1.37 MB TIF) Figure S2 Expression of class IA PI3K proteins in the hippocampus of p110d KI mice. Tissue extracts from the hippocampus from adult WT and p110d KI mice were immunoblotted with PI3K isoform-specific antibodies as indicated. Anti-b-actin staining was used as internal control for equal protein loading. Found at: doi:10.1371/journal.pone.0000869.s002 (0.15 MB TIF) Figure S3 Normal spatial memory development of p110d KI mice in the Morris water maze. (A) WT and p110d KI mice were trained with 12 trials per day in blocks of 4 trials. The time to reach the hidden platform is shown; there was no difference between the genotypes. (B) After training, day 3 and 5 probe trials were performed to assess selective searching in the quadrant where the platform used to be (TQ). Both genotypes searched selectively indicating normal spatial memory in p110d KI mice. (C) The 'platform crossings' during the probe trials showed the same accuracy in WT and p110d KI mice. Each data point represents the mean+SEM (n = 11 mice/group). During the probe trials the swim speeds did not differ between the genotypes (data not shown). Found at: doi:10.1371/journal.pone.0000869.s003 (1.70 MB TIF) Figure S4 p110d KI mice display normal locomotor parameters prior to injury. WT and p110d KI mice were assessed for 6 locomotor parameters using the CatWalk quantitative gait analysis system to obtain baseline values. (A) The Regularity Index, an index that quantifies the % of steps assigned to one of 6 normal step sequences [26], is equivalent between WT and p110d KI mice. (B) The base of support (measured in arbitrary units) represents the width between the two hind paws and indication of the stability of posture during locomotion. The base of support does not differ between WT and p110d KI mice. (C, D) For each hind paw, the average area of contact and the average intensity of light reflected at each point of contact (which is indicative of the pressure applied by the paw upon the glass surface) are equivalent between WT and p110d KI mice. Both average area and average intensity are measured in arbitrary units. (E, F) The stance phase is timed while the paw is placed upon the glass and the swing phase is timed between paw placements. The duration of swing and stance phases (in sec) between the two groups do not differ. In each evaluation, every data point represents the mean+SEM (n = 6 mice/group). p.0.1 in all parameters. Found at: doi:10.1371/journal.pone.0000869.s004 (0.47 MB TIF)