CXCL12-induced rescue of cortical dendritic spines and cognitive flexibility

Synaptodendritic pruning is a common cause of cognitive decline in neurological disorders, including HIV-associated neurocognitive disorders (HAND). HAND persists in treated patients as a result of chronic inflammation and low-level expression of viral proteins, though the mechanisms involved in synaptic damage are unclear. Here, we report that the chemokine CXCL12 recoups both cognitive performance and synaptodendritic health in a rodent model of HAND, which recapitulates the neuroinflammatory state of virally controlled individuals and the associated structural/functional deficiencies. CXCL12 preferentially regulates plastic thin spines on layer II/III pyramidal neurons of the medial prefrontal cortex via CXCR4-dependent stimulation of the Rac1/PAK actin polymerization pathway, leading to increased spine density and improved flexible behavior. Our studies unveil a critical role of CXCL12/CXCR4 signaling in spine dynamics and cognitive flexibility, suggesting that HAND - or other diseases driven by spine loss - may be reversible and upturned by targeting Rac1-dependent processes in cortical neurons.


Introduction
The neurological complications of human immunodeficiency virus 1 (HIV-1) infection, collectively 13 known as HIV-associated neurocognitive disorders (HAND), remain an important and unmet clinical need 14 (Heaton et al., 2010). While the introduction of combination antiretroviral therapy (ART) has significantly 15 reduced the severity of neurological impairments in HIV+ individuals, approximately 30-50% of infected 16 patients will develop some form of neurocognitive dysfunction (Saylor et al., 2016). Additionally, these 17 impairments continue to be important determinants of quality of life, as well as disease progression 18 (Simioni et al., 2010;Tozzi et al., 2007). 19 The neuropathology of HAND is complex and has shifted dramatically since the implementation of animals, as in patients, viral proteins are expressed at low levels in the brain -including in the prefrontal 36 cortex, an area involved in higher cognitive functions and neuroHIV (Ann et al., 2016). 37 The chemokine CXCL12 and its main signaling receptor CXCR4 regulate several critical steps of 38 CNS development, including neuronal survival and neuronal-glial communication (Bhattacharyya et al.,39 2008; Guyon and Nahon, 2007;Khan et al., 2004;Nicolai et al., 2010). Our previous findings in 40 differentiated central neurons suggested that the chemokine could also modulate spine morphology, an 41 important factor for input-specific alterations in synaptic strength (Yuste, 2011), particularly in the mature 42 brain. Dendritic protrusions are often grouped into four major classes: mushroom, thin, and stubby spines 43 and filopodia (Peters and Kaiserman-Abramof, 1970). Mushroom and thin spines are classically considered 44 mature structures, while stubby spines and filopodia are thought of as immature or spine precursors, 45 respectively (Bourne and Harris, 2007). 46 Here we report that intracebroventricular (ICV) administration of CXCL12 completely rescues 47 dendritic spine loss and cognitive dysfunction in the HIV-Tg rat. The chemokine specifically increased the 48 number of thin spines, which are associated with learning and plasticity, in layer II/III pyramidal neurons of 49 the medial prefrontal cortex (mPFC). In the same animals, upregulation of spine density by CXCL12 50 correlates with an improvement in cognitive flexibility, a task mediated by the mPFC. Our mechanistic 51 studies demonstrate the essential role of the Rac1/PAK pathway in mediating the effects of CXCL12 on 52 dendritic spines and cognition. Overall, this study provides evidence of complete restoration of structural 53 and functional deficits in a model of HAND, reveals a novel fundamental role of CXCL12 in the mature 54 brain, and identifies the molecular pathway responsible for CXCL12-mediated restoration of dendritic 55 spines and cognitive performance. that exogenous administration of CXCL12 could alleviate structural and cognitive deficits in Tg rodents. Tg 90 rats were implanted with a unilateral cannula targeted to the lateral ventricle and allowed to recover for 91 seven days prior to initiation of once-a-day infusions with vehicle (0.1% BSA in PBS, 5L total volume) or 92 CXCL12 (5ng/L, 5L total volume) throughout the duration of the three-phase behavioral task ( Fig. 3-93 figure supplement 1). 94 Both groups performed similarly across the first two phases (position discrimination and position 95 reversal) of the behavioral task, with animals reaching criterion in approximately one day (Fig. 3-figure  96 supplement 2). In the shift to cue task, CXCL12-treated rats took significantly fewer trials to reach criterion 97 than vehicle-treated counterparts (for animals who reached criterion: 701.3  97.26 TTC for CXCL12 98 compared to 1191  114.5 TTC for vehicle). Additionally, a greater number of rats given CXCL12 (n=7 out 99 of 10) completed the task compared to vehicle-treated animals (n=2 out of 9; p=0.03; Fig. 3A). A linear 100 regression analysis on the cumulative proportion of animals reaching criterion over the ten sessions 101 demonstrated that the CXCL12-treated rats completed the task at a significantly faster rate, F (1,16) =44.3781, 102 p<0.001 (Fig. 3B), further emphasizing that CXCL12 rescues cognitive flexibility in this animal model, even 103 after deficits are already present. Parallel studies in male WT rats (Fig. 3-figure supplement 3) showed that 104 vehicle-and CXCL12-treated WT rodents performed at a similar level across all three phases of the task, 105 indicating that long term exposure to CXCL12 does not result in any deleterious effects on cognition as 106 measured by the attentional set-shifting task. 107 CXCR4 inhibitor (AMD3100; 100ng/mL, 20 minutes) or PTX (100ng/mL, 18 hours) completely blocks the 134 ability of CXCL12 to activate Rac1 in cortical neurons (Figs. 4B and C). These results, as observed with 135 dendritic spine density (Pitcher et al., 2014), support the involvement of CXCR4 and its G-protein 136 dependent signaling for CXCL12-induced Rac1 activation in cortical neurons. 137

CXCL12 phosphorylates downstream mediators of Rac1 and results in a shift in the F/G actin ratio. 138
Activation of Rac1 facilitates dendritic spine stability and alterations in actin polymerization through 139 phosphorylation of downstream proteins, including PAK1, LIMK1, and cofilin. In CXCL12-treated neurons, 140 we observed a time-dependent increase in phosphorylation in PAK1 (Thr423), LIMK1 (Tyr507/508), and 141 cofilin (Ser3) (Fig. 5A). In line with our observations regarding Rac1 activation, inhibition of CXCR4 and Gαi 142 signaling completely blocked CXCL12-induced phosphorylation of PAK1, LIMK1, and cofilin (Figs. 5B and 143 C). Thus, CXCL12-mediated activation of Rac1 results in the phosphorylation of proteins critical for 144 dendritic spine stability through CXCR4 and PTX-sensitive Gi proteins. 145 The protein phosphatase Slingshot homolog 1 (SSH1) is known to regulate the phosphorylation 146 status of both LIMK1 and cofilin (Sparrow et al., 2012). Dephosphorylated/active SSH1 exhibits dual 147 activity in that it dephosphorylates and activates cofilin and, in parallel, dephosphorylates and inactivates 148 LIMK1, resulting in the release of the brake imposed by LIMK1 on cofilin (Niwa et al., 2002). However, 149 phosphorylation of SSH1 at Ser978 causes inactivation such that SSH1 can no longer dephosphorylate 150 and inactivate LIMK1/activate cofilin (Mizuno, 2013). CXCL12 increased phosphorylation of SSH1 after one 151 hour, demonstrating that not only does CXCL12 regulate the phosphorylation status of LIMK1 and cofilin 152 through activation of Rac1 and PAK1, but also through inactivation of SSH1 (Fig. 5D). 153 Increases in phosphorylated cofilin (Ser3) by either LIMK1 activation or SSH1 inactivation culminate 154 in F-actin polymerization and a reduction in actin filament turnover. Therefore, we would expect a shift in 155 the F/G-actin ratio in favor of F-actin in response to activation of this pathway by CXCL12. Cortical neurons 156 treated with CXCL12 for one or two hours exhibited a four-fold increase in the F/G-actin ratio relative to 157 vehicle-treated controls, indicative of a shift towards increased actin polymerization (Fig. 5E). This 158 functional change in the actin ratio is consistent with the time-dependent increase in dendritic spines 159 induced by CXCL12 (Pitcher et al., 2014). These findings suggest that CXCL12-induced activation of the 160 Rac1/PAK pathway in cortical neurons through CXCR4-, Gαi-dependent mechanisms, results in 161 measurable, functional changes in actin polymerization, consistent with those needed for dendritic spine 162 stability. 163 CXCL12 specifically increases thin spine density via activation of Rac1. We observed activation of 164 Rac1 and its downstream mediators by CXCL12 in cortical neurons, resulting in functional alterations in the 165 F/G-actin ratio. Since these changes are associated with dendritic spine stability, we sought to investigate 166 whether CXCL12 could upregulate a specific spine type. Thin spines appear to be likely targets for this 167 signaling cascade due to the rapid F-actin polymerization and depolymerization that leads to their quick 168 formation and elimination, often over the course of minutes. Rat primary cortical neurons (21 DIV) were 169 exposed to CXCL12 (20 nM) for three hours; this treatment has previously been shown to elicit a peak 170 effect of CXCL12 in cortical neurons (Pitcher et al., 2014). Consistent with our signaling data, CXCL12 171 specifically increased thin spine density on these neurons (Fig. 6A). On the same segment of dendrite 172 analyzed for thin spine density, we observed a subsequent reduction in stubby spine density. This 173 suggests, together with our data regarding the shift in the F/G-actin ration, that spine stabilization is 174 increased following CXCL12 treatment. 175 To determine whether CXCL12 depends on the activation of Rac1 to modulate dendritic spine 176 density and morphology we used both pharmacologic and molecular approaches. First, we determined that 177 pretreatment of neuronal cultures with the specific Rac1 inhibitor NSC23766 (Gao et al., 2004) (100µg/mL; 178 15 mins) completely blocked CXCL12-induced activation of Rac1 (Fig. 6B) and phosphorylation of its 179 downstream mediators (Fig. 6C). Next, we assessed alterations in dendritic spines by CXCL12 following 180 either NSC23766 pretreatment or knockdown of Rac1 via shRNA (neurons infected at DIV 18 and treated 181 on DIV 21). As shown in figure 6D, the ability of CXCL12 to affect overall spine density and thin spine 182 density was inhibited by blockade of Rac1 activation; furthermore, NSC23766 did not exhibit any effects on 183 spine density or morphology on its own, demonstrating that acute inhibition of Rac1 does not negatively 184 affect dendritic spines. Similarly, CXCL12 was unable to modulate spine density or morphology in Rac1 185 shRNA neurons (Fig. 6E); in line with other studies (Tashiro and Yuste, 2004), prolonged inhibition of Rac1 186 expression by shRNA reduced total spine density (in either vehicle-and CXCL12-treated neurons). Taken 187 together, these results point to the essential role of Rac1 activation in CXCL12-mediated dendritic spine 188 alterations and provide the first mechanistic insight into this novel function of CXCL12. 189 We further investigated the role of Rac1 activation in regulating CXCL12-mediated dendritic spine 210 dynamics and cognition in vivo. Prior to behavioral studies, the efficiency of in vivo Rac1 inhibition was 211 assessed. Four-month old male rats were implanted with a unilateral cannula as previously described and 212

In vivo inhibition of Rac1 stimulation prevents CXCL12-induced alterations of dendritic spines and
following a seven-day recovery period, once daily infusions of either vehicle (diH 2 O) or NSC23766 (1g/L 213 or 2g/L; 5L total volume) were initiated for eight days. At the end of the eight-day period, animals were 214 sacrificed and one hemisphere was processed for Rac1 pulldown, while the other was utilized for 215 immunohistochemical analysis of pPAK1. Immunoblotting revealed that both doses of NSC23766 216 significantly attenuated activation of Rac1 in frontal cortex lysates in a dose-dependent manner ( Fig. 8-217 figure supplement 1). Additionally, there was a dose-dependent decrease in pPAK1 in NeuN+ cells in the 218 contralateral hemisphere (Fig. 8-figure supplement 1). Thus, prolonged treatment with NSC23766 via ICV 219 administration was well tolerated, and successfully downregulated activation of Rac1 and subsequent 220 phosphorylation of its downstream mediators.  Based on the findings in WT rats, we extended our studies on Rac1 inhibition to HIV-Tg rats in 236 order to determine the role of Rac1 activation on the ability of CXCL12 to positively modulate cognitive 237 flexibility and dendritic spines in a neuroinflammatory environment. Co-treatment with NSC23766 238 completely abrogated the ability of CXCL12 to enhance attentional set-shifting in HIV-Tg rats, with fewer 239 animals reaching criterion (55% for Veh+CXCL12 compared to 25% for NSC+CXCL12) and taking 240 significantly longer to reach criterion (F (3,32) =7.503, p=0.006, Fig. 9B). As previously observed, inhibition of 241 administration (data not shown) suggesting that injury in the mPFC is not layer-specific. Thus, the increase 295 in stubby spine density on these neurons may result in aberrant synaptic transmission and hyper-296 excitability, leading to inappropriate activation of Ca 2+ -dependent enzymes and structural damage to the 297 neuron, resulting in considerable circuit alterations and behavioral dysfunction. 298 Notably, treatment with exogenous CXCL12 completely restored the reduction in dendritic spine 299 density observed in the mPFC of HIV-Tg rats, as well as the associated dysfunction in set-shifting. To our 300 knowledge, this is the first study reporting recovery of both structural and behavioral deficits in the HIV-Tg 301 rat model. Importantly, studies were conducted at an age where animals were already impaired, 302 demonstrating that dendritic spine injury and cognitive dysfunction are completely reversible. This has 303 important implications for HIV+ patients on ART where viral replication has been suppressed, but viral 304 proteins and inflammatory mediators may still be produced, especially in the CNS, and continue to induce 305 damage (Saylor et al., 2016). It also supports the notion that targeting homeostatic mediators of neuronal 306 function and synaptic transmission that are disrupted during HIV infection may be a valid therapeutic 307 strategy for HAND patients. Our data also suggests that restoring highly plastic thin spines is a key driver in 308 regulating cognitive function mediated by the mPFC. This is the first time this phenomenon has been 309 reported in a rodent model of HAND, suggesting that therapeutics to target this spine type are an 310 appropriate strategy to combat the deficits seen in these patients. Interestingly, the extended treatment 311 with CXCL12 did not reveal any side effects in both WT and Tg animals, indicating that potential effects of 312 the chemokine on other cellular targets (Guyon, 2014) do not interfere with its neuroprotective function -at 313 least within the time frame of this study. Most importantly, our work dissecting out the molecular 314 mechanisms of CXCL12-mediated spine alterations points to the key involvement of the Rac1/PAK 315 pathway. Therefore, therapies targeting this pathway specifically in neurons would be beneficial in HAND 316 and other types of cognitive dysfunction, such as aging (Hao et al., 2006). 317 318 Role of Rac1/PAK pathway in CXCL12-enhanced cognitive performance: Though the chemokine 319 CXCL12 is well known for its homeostatic action in the developing brain (Lysko et al., 2014), its relevance 320 in the mature brain has been mainly associated with inflammatory responses. Here, we demonstrated that 321 CXCL12 activates the small GTPase Rac1 and its downstream mediators to modulate dendritic spine 322 density in mature, excitatory cortical neurons. Activation of this pathway is associated with actin 323 polymerization and this resulted in a specific increase in thin spine density on cortical neurons. The 324 involvement of the Rac1/PAK pathway in mediating the effects of CXCL12 on actin polymerization has not 325 been described in cortical neurons. Our data show that Rac1 activation is the exclusive mediator of 326 CXCL12's alterations on spine density and morphology, suggesting that targeted therapeutics to selectively 327 activate this pathway through CXCR4 could recover dendritic spine loss. CXCL12 not only activated the 328 with studies in non-neuronal cells, the mechanisms leading to SSH1 phosphorylation might also be 339 conserved. However, this needs to be investigated. 340 Inactivation of Rac1 via a small molecule inhibitor completely blocked the beneficial effects of 341 CXCL12 on set shifting as measured by trials to criterion and the rate to reach criterion. Furthermore, 342 changes in spine density and morphology induced by daily CXCL12 treatment were completely dependent 343 on its ability to activate Rac1 in the mPFC. Interestingly, blockade of Rac1 activation had no perceived 344 effects on the first two phases of the behavioral task, suggesting specificity of the small GTPase in 345 functions mediated by the mPFC but this warrants further investigation. Nevertheless, to our knowledge, 346 this is the first study to connect Rac1 activation with behavioral flexibility, further underscoring the critical 347 role of the small GTPase in regulating synaptic function. Additionally, our data supports the notion that 348 utilizing therapeutics to modulate Rac1/PAK signaling may be beneficial in reversing structural and 349 behavioral deficits in the mPFC. It is important to note that the relationship between CXCL2 and CXCR4 is not exclusive, as 373 CXCL12 can also interact with another receptor, atypical chemokine receptor 3 (ACKR3), also known as 374   returned to the University of Maryland, who originally developed this transgenic line (Reid et al., 2001). 420

Rats were singly housed in isolation in our Association for Assessment and Accreditation of Laboratory 421
Animal Care-accredited barrier facilities in accordance with the National Institutes of Health guidelines and 422 institutional approval by the Institutional Animal Care and Use Committee. Animals were food-restricted 423 one week prior to the initiation of lever pressing training. 424

Cannula implantation and intracerebroventricular administration. Rats were implanted stereotaxically 425
under isoflurane anesthesia and ketamine/xylazine (50 mg/kg; 10 mg/kg, PennVet), with 26-gauge 426 stainless-steel guide cannulas (Plastics One) placed 0.96mm posterior and 2.00mm lateral to bregma, 427 3.5mm below the surface of the cranium. Four stainless-steel screws (#0-80) were placed around the 428 cannula and acrylic dental cement was used to anchor them. Following surgery, animals were allowed 429 seven days to recover before the initiation of infusions and behavioral testing. Animals received once daily 430 infusions of either vehicle (0.1% BSA in PBS, 5 L total volume), CXCL12 (5 ng/L in 0.1% BSA in PBS, 5 431 L total volume), or NSC2377 (2g/L in diH 2 O, 5L total volume) via a micropump set for an infusion rate 432 of 0.5L/minute throughout the duration of behavioral testing. 433 Attentional set-shifting task. Reversal learning and strategy shifting were assessed using an automated 434 operant-based approach whereby the behavioral tasks occurred within operant chambers controlled by 435 custom software programs (Brady and Floresco, 2015). Each chamber was equipped with a house light, 436 fan, two retractable levers, a tone generator above each lever, a pellet dispenser, a stainless-steel pellet 437 trough located between the levers, and a panel light just above the pellet trough. Food-restricted rats were 438 initially trained to respond by lever pressing under a fixed-ratio (FR1) schedule of reward presentation (45 439 mg sucrose pellet, Bio-Serv, Flemington, NJ). Side biases were avoided by presenting each lever an equal 440 number of times in a pseudorandom order. Following training, the 3-phase task was conducted, consisting 441 of position discrimination, position reversal, and rule shifting. During position discrimination, one lever (left 442 or right) was designated as the "correct" choice. Both levers were presented but only responses on the 443 "correct" lever were rewarded while responses on the other had no programmed consequences. An inter-444 trial interval of 20 seconds followed each response, during which time the levers were retracted. Rats 445 underwent position discrimination until they attained the criterion of 10 consecutive correct trials with a 446 maximum of 150 trials per day. One day following completion of the discrimination phase, the rats 447 performed a reversal learning task in which the opposite lever of the "correct" choice from the 448 discrimination phase was now rewarded. Twenty reminder trials of the previous rule were conducted prior 449 to the reversal trials. As with position discrimination, rats underwent reversal learning until they attained 10 450 consecutive correct trials. After successful completion of the reversal task, the strategy shift task was 451 initiated on the following day. In this phase both levers were presented in each trial. In addition, an audible 452 tone was presented above one of the two levers. The "correct" lever, which resulted in reward presentation, 453 was that which was associated with the tone, regardless of position (left or right) within the chamber. Tone, 454 rather than a cue light, was used in this phase to account for any inherent visual impairment due to cataract 455 development in Tg rats. Rats conducted the strategy shift phase until they attained the criterion of 10 456 consecutive correct trials, with a maximum of 150 trials on a single day. If animals failed to reach criterion 457 by day 10 of the shift to cue phase, the task was terminated. Cognitive performance was determined by 458 measuring the number of trials needed to reach criterion during each phase of the task, as well as the rate 459 at which they reached criterion. For tissue homogenates, brain cortices were rapidly removed, and a 5 mm portion of the frontal cortex was 489 separated using a brain matrix. Tissue was dissociated by pipetting and incubated for 1 hour at 4C with 490 frequent vortexing. Lysates were then centrifuged at 20,800 x g for 10 minutes at 4C and protein 491 concentration was assessed via BCA assay. Equal amounts of protein (30-40 g/lane) were used for SDS-492 PAGE followed by immunoblotting. 493 The following antibodies were used: anti-pPAK1 (Thr423, Cell Signaling Technology, 1:1000,  Rac1 activation assay. The Rac1 activation assay was performed using Cell Biolabs Rac1 Activity Assay 501 Kit (#STA-401-1). Briefly, cells or tissue were lysed as described above and the active form of Rac1 (GTP-502 Rac1) was selectively pulled down from the lysate with p21-binding domain (PBD) of PAK agarose beads. 503 Subsequently, the precipitated GTP-Rac1 was detected by Western blot analysis as described above. 504 Each assay also consisted of lysates that were loaded with GTPS or GDP as positive and negative 505 controls respectively. A separate Western blot was run to evaluate total levels of Rac1 in each lysate. 506 F/G-actin ratio. F/G-actin ratio was assessed as previously described (Pyronneau et al., 2017). Briefly, 507 cells were lysed in cold lysis buffer [10 mM K 2 PO 4 , 100 mM NaF, 50 mM KCl, 2 mM MgCl 2 , 1 mM EGTA, 508 0.2 mM DTT, 0.5% Triton-X 100, 1 mM sucrose (pH 7.0)] and centrifuged at 15,000 x g for 30 minutes. 509 Separation of F-actin and G-actin was achieved in that F-actin is insoluble (pellet) in this buffer, whereas G-510 actin is soluble (supernatant). The G-actin supernatant was transferred to a fresh tube and the F-actin 511 pellet was resuspended in lysis buffer plus an equal volume of a second buffer [1.5 mM guanidine 512 hydrochloride, 1 mM sodium acetate, 1 mM CaCl 2 , 1 mM ATP, 20 mM tris-HCl (pH 7.5)] and then 513 incubated on ice for one hour with gentle mixing every 15 minutes to convert F-actin into soluble G-actin. 514 Samples were centrifuged at 15,000 x g for 30 minutes and the supernatant (containing F-actin which was 515 converted to G-actin) was transferred to a fresh tube. F-actin and G-actin samples were loaded with equal 516 volumes and analyzed via Western blot. Latrunculin A (5 M, 2 hours), a potent actin polymerization 517 inhibitor, and jaspakinolide (5 M, 2 hours), an inducer of actin polymerization, were used as internal 518 controls for the assay. 519 Immunocytochemistry. Immunocytochemistry was performed as previously described (Pitcher et al., 520 2014). Cells were washed in PBS, fixed in 2% paraformaldehyde (PFA) for 10 minutes at room 521 temperature and 4% PFA at 4C for 20 minutes, and permeabilized with 0.1% Triton-X 100 for five minutes. 522 Blocking was performed with 5% normal goat serum for 30 minutes. The following primary and secondary 523 antibodies were used: anti-MAP2 (Millipore, 1:1000, RRID:AB_91939) and goat anti-rabbit Alexa Fluor 568 524 (Invitrogen, 1:250, RRID:AB_143157). Cells were counterstained with both Hoechst (Invitrogen, 1:10,000) 525 and phalloidin Alexa Fluor 488 (Invitrogen, 1:400). After staining, coverslips were rinsed in H 2 O and 526 mounted using ProLong Gold Antifade mounting media (Invitrogen). 527 DiOlistic labeling of brain slices. Rats were sacrificed following the end of behavioral studies. Brains 528 were rapidly removed, rinsed in H 2 O, fixed in 4% PFA for 1 hour, and washed 3 times in 1X PBS. After 529 fixation, serial coronal slices were sectioned via vibratome at a thickness of 150 m and placed in tissue 530 culture plates until further processing. DiOlistic labeling was performed according to published techniques 531 (Seabold et al., 2010). 300 mg of tungsten beads (Bio-Rad, Hercules, CA) were suspended in 99.5% pure 532 methylene chloride (Fisher Scientific) and sonicated in a water bath for 1 hour. Crystalized DiI (13.5 mg; 533 Invitrogen) was dissolved in methylene chloride and protected from light. Following sonication, 100 L of 534 the tungsten bead solution was placed on a glass slide and 100 L of DiI solution titrated on top, which was 535 slowly mixed using a pipette tip and allowed to dry. A razor blade was used to collect the dry bead/dye 536 mixture onto weigh paper, placed into a 15 mL conical tube with 3 mL ddH 2 O, and sonicated in a water 537 bath for 20 minutes. The bead/dye mixture was drawn into Tezfel tubing coated with polyvinylpyrrolidone 538 (Sigma-Aldrich, St. Louis, MO) and dried using nitrogen gas for 1 hour. Once dry, tubing was cut into 13 539 mm cartridges and loaded into the Helios Gen Gun (Bio-Rad). Helium gas flow was adjusted to 120 PSI 540 and bullets were delivered to slices through 3 m pore filter paper. Slices were quickly washed 3 times with 541 1X PBS and stored overnight at 4C to allow diffusion of the dye. The following day, slices were mounted 542 using ProLong Gold Antifade (Invitrogen), coverslipped, and stored at 4C until imaging. 543 Dendritic spine analysis. For in vitro experiments, neurons were cultured for 21 days, then fixed and 544 stained for MAP2 (Alexa 568) and counterstained with phalloidin (Alexa 488) or infected with GFP-tagged 545 lentiviral particles, as described above. Images were acquired with the Olympus FLUOVIEW FV3000 546 confocal microscope equipped with a 100x silicone oil-immersion objective at 2x electronic zoom and taken 547 at 0.5 m Z steps. Neurolucida 360 software (MBF Bioscience) was utilized to automatically quantify 548 dendritic spines, as well as classify them into their respective morphologies based on established 549 parameters (Rodriguez et al., 2008). For each experiment, four dendrites, at least 100-150 m in length, 550 from each coverslip were analyzed and a total of three coverslips were imaged for each condition. Each 551 coverslip was averaged as a single data point and the experiment was repeated across three separate 552 neuronal dissections. 553 For in vivo studies, dendrites in layer II/III pyramidal neurons from the prelimbic (PrL) region of the mPFC 554 were imaged using the Olympus FLUOVIEW FV3000 using a 100x silicone oil-immersion objective with at 555 2x electronic zoom and taken at 0.15 m Z-steps. Neurolucida 360 software was used as mentioned 556 above. Eight dendrites, at least 100-150 m in length, were analyzed from eight separate neurons and 557 averaged together as a single data point per animal. 558 Tyramide stain and amplification. The hemisphere contralateral to the one used for DiOlistic staining 559 was fixed in 4% PFA for 24 hours, moved to 70% ethanol, and then paraffin embedded and sectioned at 560 5m for immunohistochemical analysis. Tissue was sequential dual-stained with pPAK1 (Thr423, 561 Invitrogen, 1:25, RRID:AB_2554427) and the neuronal marker NeuN (Cell Signaling, 1:400, 562 RRID:AB_2651140). After rehydration, antigen retrieval was performed in a citrate buffer (Life 563 Technologies) at 95C for 20 minutes followed by quenching of endogenous peroxidase activity with 564 hydrogen peroxide and methanol for 30 minutes at room temperature. Tissue was blocked with 10% 565 normal goat serum before incubation with primary antibody overnight at 4C in a humidity chamber. and in vivo experiments. The number of trials to criterion for each behavioral task was analyzed using a 590 two-tailed Student's t-test or one-way ANOVA followed by Tukey post hoc. Dendritic spine density and 591 morphology were analyzed using a two-tailed Student's t-test or one-way ANOVA followed by Dunnett post 592 hoc if comparing experimental groups to a control group or Tukey post-hoc if comparing all groups against 593 each other. The correlation between spine density and trials to criterion was calculated as Pearson's r. The 594 rate for reaching criterion was assessed by linear regression analysis to determine if the slopes of each line 595 were equal. For multispectral imaging, a two-tailed Student's t-test was used. For the remaining in vitro 596 studies, statistical significance was determined by two-tailed Student's t-test or one-way ANOVA followed 597 by Dunnett post hoc. A p<0.05 was considered statistically significant. P and n are reported in the figure  598 legends; for correlation analysis, r values are also reported in the text. Additional information regarding 599 statistical analyses and raw data is provided in the source data files for each figure. All statistical analysis 600 was performed with GraphPad Prism, version 8.0 (GraphPad Software, RRID:SCR_015382). 601 602

Acknowledgements 603
The authors thank Renato Brandimarti for assistance with viral particle production, David Sulzer (Columbia 604 University) and members of the Meucci laboratory for discussion and critical reading of the manuscript, 605 particularly Bradley Nash for support with editing and statistics. 606 607

Competing Interests 608
The authors declare no competing interests.          E. The ability of CXCL12 to decrease stubby spine density was blocked by NSC23766. N=12 963 animals/group, 8 dendrites measured for each animal and averaged into single data point, **p<0.01. 964 F. As previously observed, overall dendritic spine density was negatively associated with trials to criterion 965 on the set-shifting phase of the behavioral task. N=23 animals, Pearson's r=-0.7862, p=0.0127. 966 G. This relationship became even stronger when only thin spine density was considered. N=23 animals, 967 Pearson's r=-0.8350, p=0.0051. 968    during the first two phases of the behavioral task. N=12 animals/group. 1034