enlargement of axo-somatic contacts formed by gad-immunoreactive axon terminals onto layer v...

OR I G INA L ART I C L E

Enlargement of Axo-Somatic Contacts Formed byGAD-Immunoreactive Axon Terminals onto Layer VPyramidal Neurons in the Medial Prefrontal Cortex ofAdolescent FemaleMice Is Associated with Suppressionof Food Restriction-Evoked Hyperactivity and Resilienceto Activity-Based AnorexiaYi-Wen Chen, Gauri Satish Wable, Tara Gunkali Chowdhury, and Chiye Aoki

Center for Neural Science, New York University, New York, NY 10003, USA

Address correspondence to Chiye Aoki, Center for Neural Science, New York University, 4 Washington Place, New York, NY 10003, USA. Email: [email protected]

AbstractMany, but not all, adolescent female mice that are exposed to a running wheel while food restricted (FR) become excessivewheel runners, choosing to run even during the hours of food availability, to the point of death. This phenomenon is calledactivity-based anorexia (ABA).Weused electronmicroscopic immunocytochemistry to askwhether individual differences inABAresilience may correlate with the lengths of axo-somatic contacts made by GABAergic axon terminals onto layer 5 pyramidalneurons (L5P) in theprefrontal cortex.Contact lengthswere, onaverage, 40%greater for theABA-inducedmice, relative to controls.Correspondingly, the proportion of L5P perikaryal plasma membrane contacted by GABAergic terminals was 45% greater for theABAmice. Contact lengths in the anterior cingulate cortex correlated negatively and strongly with the overall wheel activity afterFR (R =−0.87, P < 0.01), whereas those in the prelimbic cortex correlated negatively with wheel running specifically during thehours of food availability of the FR days (R =−0.84, P < 0.05). These negative correlations support the idea that increases in theglutamic acid decarboxylase (GAD) terminal contact lengths onto L5P contribute toward ABA resilience through suppression ofwheel running, a behavior that is intrinsically rewarding and helpful for foraging but maladaptive within a cage.

Key words: anxiety, cingulate cortex, electron microscopic immunocytochemistry, exercise, prelimbic cortex

IntroductionAdolescence is a period of peak physical health and mentalcapacity, though brain maturation is still incomplete. Brain mat-uration during adolescence is susceptible to environmental in-fluences (Andersen et al. 2000; Chowdhury et al. 2014) andmany mental illnesses emerge for the first time during adoles-cence (Spear 2000; Romeo 2010). One mental illness that isparticularly prevalent among adolescent females is anorexia

nervosa (AN), features of which include refusal to maintain nor-mal body weight, fear of weight gain, and dysmorphia (AmericanPsychiatric Association 2013). Once succumbed to the conditionof AN, relapse is prevalent (30–50%) (Birmingham et al. 2005)and mortality is extremely high (10–20%) (Sullivan 1995; Zipfelet al. 2000; Birmingham et al. 2005; Bulik et al. 2007), evensurpassing depression. Yet, there is at present no accepted phar-macological treatment for this condition. Nearly all adolescent

© The Author 2015. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: [email protected]

Cerebral Cortex, 2015, 1–16

doi: 10.1093/cercor/bhv087Original Article

1

Cerebral Cortex Advance Access published May 15, 2015 at N

ew Y

ork University on M

ay 15, 2015http://cercor.oxfordjournals.org/

Dow

nloaded from

http://www.oxfordjournals.org

http://cercor.oxfordjournals.org/

females experiment with dieting, while only 1% of them succumbto the condition of AN (Hudson et al. 2007). What is the basis forthe individual differences in vulnerability to AN? This was thequestion of our study strived to answer through the use of amouse model of AN, namely activity-based anorexia (ABA).

ABA is a condition whereby severe voluntary hyperactivityand weight loss is induced simply by restricting food accesswhile providing ad libitum access to a running wheel (Gutierrez2013). ABA induction captures 3 hallmarks of AN (AmericanPsychiatric Association 2013): Severe weight loss (Gutierrez2013), over-exercise (Epling et al. 1983; Davis et al. 1997, 1999;Hebebrand and Bulik 2011; Zunker et al. 2011; Gutierrez 2013),and anxiety (Kaye et al. 2004; Perdereau et al. 2008; Dellavaet al. 2010; Thornton et al. 2011; Wable et al. 2015). Importantly,although ABA is induced by the initial imposition of food restric-tion (FR), once animals have begun to lose body weight, their vol-untary wheel activity becomes so excessive that animals choosewheel running over eating, even during the limited hours of foodaccess. In this way, ABA-induced animals exhibit one additionalstriking hallmark of the human condition of AN—namely, volun-tary FR. When applied to adolescent C57BL6 female mice, thisanimalmodel also reveals individual differences in vulnerability,with approximately half of the animals exhibiting severe hyper-activity and weight loss that would be lethal unless removedfrom the ABA-inducing environment, while approximately aquarter of the population exhibits resilience, due to minimalhyperactivity and weight loss (Chowdhury et al. 2013). Havingidentified a cohort with variable degrees of vulnerability to ABAinduction, our next goal has been to identify the cellular basesfor the individual differences in ABA vulnerability.

It has recently been demonstrated convincingly that wheelrunning is not an abnormal behavior induced by rearing in cap-tivity but is highly rewarding, even to mice in wilderness (Meijerand Robbers 2014). A number of ideas have been proposed to ex-plain why restricted food access evokes heightened activity onthewheel. One idea is that wheel running is in response to hypo-thermia, since FR-evoked hyperactivity can be reversed throughhigh ambient temperature (Gutierrez et al. 2006). Another isthat wheel running is anxiolytic (Hare et al. 2012), thus being per-petuated by individuals that are experiencing anxiety due to thestress of FR. Wheel running may also be an attempt at food for-aging [discussed by Gutierrez (2013)]. However, when occurringwithin a closed environment, wheel running is detrimental, be-cause it exacerbates weight loss (Chowdhury et al. 2013) withoutany opportunity of increasing food access. Individual differencesin ABAvulnerability could arise, at least in part, from differencesin the animals’ neural circuitry underlying the decision to sup-press wheel running, after undergoing an experience wherewheel running is not suitable for survival. This idea is based onour observation that the resilient animals are those that reducewheel running, therebyminimizing bodyweight loss and extend-ing survival, while those animals that exhibit the greatest ABAvulnerability, based on the severity of weight loss, are the onesthat continue to respond to FR with excessive running on thewheel (Chowdhury et al. 2013). In search for differences in neuralcircuitry underlying the decision to suppress wheel running, wehave turned our attention to the prefrontal cortex (PFC).

Our rationale for analyzing the PFC stems from human data(Kaye et al. 2009). Among individuals with AN, presentations ofvisual stimuli depicting physical activity result in an exaggeratedresponse of the PFC: The authors suggest that visual stimuli inthe form of physical activitymay bemore rewarding, thereby pla-cing increased demand on the inhibitory control system of ANpatients when challenged with a Go/NoGo task (Kullmann et al.

2014). Another study reported that AN patients’ performance ofa PFC cognitive task, theWisconsin Card Sorting Test, was signifi-cantly poorer than that of healthy controls (Sato et al. 2013).There is also an intriguing case study, indicating that symptomsof AN were relieved in a male after glioma in the frontal lobe wasremoved surgically (Goddard et al. 2013). Not only is activation ofthe PFCof individuals with AN consistently altered, but also in theratmodel of ABA, the PFC has been shown to be hypometabolic bymicro-positron emission tomography (PET; van Kuyck et al. 2007).These studies suggest that altered activity of the PFCmay contrib-ute to the behavioral phenotype of AN amonghumans and of ABAamong rodents, therebyultimatelyaffecting individuals’decisionsto exercise or to eat.

Excitability of the PFC is determined, in part, by the intracor-tical circuitry comprised of excitatory pyramidal neurons and in-hibitory interneurons. Of them, the layer V pyramidal neuronscontribute strongly to the excitatory outflow from the PFC to des-cending cortical and subcortical regions mediating motor output(Berendse et al. 1992; Gabbott et al. 2005; Riga et al. 2014). Further-more, while excitability of the layer V pyramidal neurons inthe PFC is controlled by the balance of their excitatory and inhibi-tory inputs, individual gamma-aminobutyric acid (GABA)ergicaxo-somatic inhibitory synapses are more influential than theGABAergic axo-dendritic synapses or the numerous but moredistally located excitatory axo-spinous synapses. Moreover, par-valbumin (PV) neurons in the PFC, which exert the axo-somaticinhibition, have been shown to fire as animals make decisions toleave a reward (Kvitsiani et al. 2013). This indicates that GABAergicaxo-somatic inhibition of pyramidal neuronswithin the PFC couldplay a particularly central role in an animal’s choice-related beha-viors, such as to eat food or to run on the wheel. Thus, amongthe many types of synapses that exist within the PFC, we choseto assess the extent of axo-somatic inhibitory contacts formedby GABAergic axon terminals onto layer V pyramidal neurons.

We chose to take an electron microscopic approach for 2 rea-sons: First, we wished to be able to study a well-defined singleclass of neurons targeted by GABAergic innervation, in case en-vironmentally and experientially evoked changes in GABAergiccontacts differed depending on the postsynaptic neuronal type(pyramidal vs. interneuronal; layer II/III vs. layer V vs. layer VIpyramidal cells). Electron microscopy (EM) allows for differenti-ation of cell bodies belonging to pyramidal cells versus interneur-ons (Peters and Jones 1984; White and Keller 1989). Moreover, wewished to be able to use the resolving capability of EM for differ-entiating those contacts that are direct versus others that are in-direct, due to the interposition of thin (<30 nm, below the limit ofresolution of light microscopes) astrocytic processes in betweenGABAergic axon terminals and perikaryal plasma membranesthat can reduce the activation of GABAergic receptors [review inZhou and Danbolt (2013)].

The source of PFC tissue for this analysis was derived, in part,from an earlier published study that examined GABAergic in-nervation of pyramidal cells in the hippocampus of adolescentfemale C57BL6 mice exhibiting variable degrees of hyperactivityfollowing 2 ABA inductions during adolescence, spaced 1 weekapart (Chowdhury et al. 2013).

Materials and MethodsAnimals and ABA Induction

Fourteen female mice, 7 ABA animals, and 7 controls, from 4 lit-ters, were used in this study. All animals used in this study werebred in New York University’s animal facility. All but one of the

2 | Cerebral Cortex

at New

York U

niversity on May 15, 2015

http://cercor.oxfordjournals.org/D

ownloaded from


PFC tissue used for this study came from brains used for anotheranatomical study on the hippocampus (Chowdhury et al. 2013).Two breeding pairs were wild-type C57BL/6J mice obtainedfrom Jackson Laboratories. Two other breeding pairs were alsoobtained from Jackson Laboratories (Bar Harbor, ME, USA) on aC57BL6/J background but crossed with BALB/c strains: CB6-Tg(Gad1-EGFP)G42Zjh/J (G42), which express green fluorescent pro-tein (GFP) in PV-containing interneurons (Chattopadhyaya et al.2004). GFP label was not utilized as a marker for the electronmicroscopic study described here. Details of these animals withGFP expression are described in a previous publication from thislaboratory: in short, thesemice exhibited behavioral and glutam-ic acid decarboxylase (GAD) expression phenotypes that were nodifferent from wild types (Chowdhury et al. 2013).

After weaning at postnatal day 25 (P25), the animals weregroup-housed 2–4 per cage with their littermates, separatedfrom males. Starting on P36, the animals were assigned into 2groups: ABA animals and controls. Control (CON) animals weresingly housed in standard home cages with ad libitum accessto food and no running wheel access for the duration of thisstudy. ABA animals were housed individually with unrestrictedfood and water access. Starting on P36, ABA animals were given24-h/day access to a running wheel (Med Associates Low-ProfileWireless Running Wheel for Mouse Product: ENV-044), andtheir wheel-running activity was recorded continuously on aPC, using Med Associates’ “Wheel Manager” software (Product:SOF-860). In addition, wheel running during the hours 7 PM to9 PM and the 24-h running record (from 7 PM to 7 PM) wasmanu-ally recorded, together with the animals’weight and food intake.The cageswere kept undera 12-h light/dark cycle (lights on at 7AM;lights off at 7 PM). The daily measurements of animals’ bodyweight, food consumption, and running wheel activity for 3 daysserved as baseline data under the singly housed condition, to becompared against the additional impact of FR that began on P40.On P40, food was removed at 12 PM. ABA mice were only allowedfree access to food for the first 2 h of the dark cycle (7 PM–9 PM)for 3 days (first ABA). After 3 days of exposure to ABA, starting at12 PM, ad libitum food access was restored and the runningwheel was removed from the cage. The animals’ body weights re-turned tobaseline levelswithin 48 h.After 7daysof 24-h/dayaccessto food, and without wheel access, animals were placed in a cagewith a running wheel, so as to reassess the running wheel activityin the absence of FR. After reacclimation to the running wheelfor 4 days, ABA animals were returned to the ABA environment(second ABA), whereby they were given restricted food access for2 h, at the beginning of the dark cycle, in the presence of a runningwheel. The second ABA continued for 4 days.

The animals’ estrous cycles were not monitored, becausecycles are already known to become disrupted by FR (Nelson1985; Dixon et al. 2003) and because pubertal animals are too im-mature to be cycling regularly (Hodes and Shors 2005). We alsoreasoned that vaginal smears, needed for assessing the estrouscycle, is a stressful procedure that could exacerbate or mask theeffect of stress associated with FR.

All procedures relating to the use of animalswere according tothe NIH Guide for the Care and Use of Laboratory Animals andalso approved by the Institutional Animal Care and Use Commit-tee of New York University (A3317-01).

Preparation of Brains

ABA animals were euthanized at the end of the second ABA, atages P59 or P60, by anesthetizing them with urethane (i.p. 34%;0.15 mL/20 g body weight) during the hours of 12 PM to 4 PM,

then transcardially perfusing them with a buffer containing 4%paraformaldehyde. Glutaraldehyde fixation was withheld untilafter immunocytochemistry, so as to optimize antigen retention.Brains were also collected from CON mice at ages P59 or P60.

The GAD Immunocytochemistry Procedure

Details of theGAD immunocytochemical procedure are similar tothe steps described in our previous publication (Chowdhury et al.2013). Tissues from all animals were processed for immunocyto-chemistry and EM synchronously, so as to minimize variabilitiesarising from differences in reagent qualities, incubation time,ambient temperature, etc. The rabbit anti-GAD antibody usedin this study (Lot 2069924 of Millipore, Bellerica, MA, USA; catalognumber AB11; hereafter referred to as anti-GAD) recognizes boththe GAD65 and GAD67 isoforms. Multiple vibratome sectionsfrom each animal’s brain, prepared in the coronal plane at athickness of 40 μm and containing 3 regions [cingulate cortex,area 1 (Cg1), prelimbic cortex (PrL), and medial orbital cortex(MO)] of the PFC (Paxinos and Franklin 2001; Fig. 1), were incu-bated overnight, at room temperature, under constant agitation,in a buffer consisting of 0.01 M phosphate buffer/0.9% sodiumchloride (PBS), adjusted to pH 7.4, together with 1% bovineserum albumin (BSA, w/v, from Sigma Chem., St Louis, MO,USA), 0.05% sodium azide (w/v, from Sigma Chem.), and theGAD antibody at a dilution of 1 : 400. At the end of the incubationperiod, sections were rinsed in PBS over a 30-min period, andthen incubated for 1 h at room temperature under constant agita-tion in the PBS/BSA/azide buffer containing biotinylated goatanti-rabbit IgG (Vector, Inc., Burlingam, CA, USA) at a dilution of1 : 200. At the end of the incubation period, the sections wererinsed in PBS over a 30-min period and then incubated in PBS con-taining the avidin-biotinylated horseradish peroxidase complex(HRP) (ABC, from Vector’s ABC Elite kit) for 30 min at room tem-perature, under constant agitation. The sectionswere then rinsedin PBS and reacted with the HRP substrate, consisting of 3,3′-diaminobenzidine hydrochloride (DAB, 10 mg tablets fromSigma Chem.) per 44 mL of PBS buffer and hydrogen peroxide(4 μL of 30% hydrogen peroxide per 44 mL of DAB solution). TheHRP reaction was begun by adding hydrogen peroxidase, and ter-minated at the end of 11 min by rinsing repeatedly in PBS. Vibra-tome sections underwent post-fixation using 2% glutaraldehydefor 16 min (EM Sciences, Martifield, PA, USA) in PBS.

Tissue Processing for EMFollowing the immunocytochemical procedure, the vibratomesections were processed by a conventional electron microscopicprocedure, consisting of post-fixation by immersing in 1%osmiumtetroxide/0.1 Mphosphate buffer for 1 h, thendehydratedusing graded concentrations of alcohol up to 70%, and then post-fixed overnight using 1%uranyl acetate (Terzakis 1968; Lozsa 1974)dissolved in 70% ethanol. On the following day, the sections werefurther dehydrated using 90% ethanol, followed by 100% ethanol,then were rinsed in acetone, and infiltrated in EPON 812 (EMSciences), which was cured by heating the tissue at 60 °C, whilesandwiched between 2 sheets of Aclar plastic, with lead weightsplaced on top of the Aclar sheets, so as to ensure flatness of theEPON-embedded sections. These flat-embedded vibratome sec-tions were re-embedded in BEEM® capsules (EM Sciences) filledwith EPON 812, then ultrathin-sectioned at a plane tangential tothe vibratome sections. The ultrathin sections were collectedonto formvar-coated, 400 mesh thin-bar nickel grids (EMSciences). Ultrathin sections were counterstained with Reynolds’lead citrate before viewing.

Enlarged GABA Synapses in the Medial Prefrontal Cortex Correlates with ABA Resilience Chen et al. | 3

at New

York U



ownloaded from


The Sampling ProcedureTwo steps were taken to maintain random, unbiased sampling.One, the electron microscopist sampling the ultrathin sectionwas blinded to the animal’s antemortem behavioral characteris-tics and treatments. Two, for each cortical area of each animal,somatic profiles were analyzed, strictly in the order that theywere encountered from the surface-most portions of vibratomesections spanning layer V. Cell bodies of pyramidal neuronswere identified as being positioned in layer V, if they were locatedmidway between pial surface and the white matter. Care wastaken to never re-sample an area, so as to avoid analysis of over-lapping sectors of perikaryal plasma membrane.

The 3 PFC regions of interest that were sampled and analyzedseparately were Cg1, PrL, and MO. The anatomical landmarks

of white matter, sulci, overall shape of the coronal section, celldensity, and all other features shown in Figure 1A were used todetermine the AP levels and boundaries of the subregions ofthe PFC. We sampled the neuropil surrounding cell bodies ofpyramidal cells located in layer V of the PFC.

Statistical analysis did not reveal any difference in the propor-tion of perikaryal plasma membrane contacted by GAD terminalsor of the individual lengths of contact formed by GAD terminalsonto cell bodies of layer V pyramidal neurons across the AP levels.There is alsonobehavioral literaturepointing todifferences in func-tion across the AP levels. Therefore, data from all AP levels werecombined for each subregion of the PFC (Fig. 1A) of each animal.

Sampling of the axon terminals contacting cell bodies oflayer V pyramidal neurons in prefrontal areas was designed to

Figure 1. Description of the sampling procedure for quantification of GAD terminal contact sizes onto layer V pyramidal neurons of the PFC. EM was used to verify direct

contact between GAD axon terminals and pyramidal cell bodies. (A) Mouse brain atlas highlighting the 3 regions of interest in the PFC, consisting of the cingulate cortex

area 1 (Cg1), prelimbic cortex (PrL), andmedial orbital cortex (MO). Anterior–posterior position (frombregma) of a schematic of amouse brain’s coronal section is indicated

below the atlas, but the actual sampling ranged from 2.34 to 1.98 mm. Each trapezoid represents the position of the ultrathin section prepared for electron microscopic

analysis. The atlas was modified from figures of Paxinos and Franklin’s atlas (Paxinos and Franklin 2001). Calibration bars on the left = 1 mm. (B) An illustration of the

procedure we followed to quantify the lengths of direct contacts made by GAD-immunoreactive axons along the perikaryal plasma membrane. The blue lines

represent the total membrane length of a profile identified to be belonging to a pyramidal neuron. The purple lines represent the contact lengths of GAD axon

terminals. The red polygons represent the cross-sectional area of GAD axon terminals forming contacts. (C) This example shows 6 GAD axon terminals forming direct

contacts onto a cell body, taken from MO of the PFC of a CON animal. The smooth contour of the nuclear membrane and absence of asymmetric excitatory synapses

indicates that this is a cell body of a pyramidal neuron. The white asterisks (to the right of contact 6; below contact 6; and below contact 2) are GAD axon profiles that

are near but not directly contacting the cell body plasma membrane. The 2 gray arrows below contact 2 are examples of axon terminals with levels of HRP-DAB

reactivity that are indistinguishable from background labeling. To maintain consistency, only those terminals with HRP-DAB labeling that were detectably more

electron dense than those of mitochondria were considered immunolabeled, and included in the quantitative analysis. Calibration bar = 2 μm. (D) Contact 3 is shown

at a higher magnification of 60 000× to reveal synaptic specialization (synaptic junctions, between the 2 arrows), indicated by the clustering of vesicles and parallel

alignment of plasma membranes of the axon terminal and of the cell body. The arrowheads indicate the lateral extents of the contact, which includes the flanking

regions that lack clearly parallel alignment of the 2 plasma membranes. Ast, astrocytic profile in (D), which often flanks synaptic junctions. Calibration bar = 500 nm.

4 | Cerebral Cortex

at New

York U



ownloaded from


optimize immunodetection. To achieve this, for each region ofinterest, multiple ultrathin sections revealing the surface-mostportions of vibratome sections for that region were chosen. Fiveto 10 somata were sampled from each subregion of the PFC ofeach animal. Since there were 3 subregions of the PFC analyzedfor each animal, the total number of somata sampled per animalwas 15–30. The number of ultrathin sections collected from anyone animal’s brain section depended on the natural curvaturethat the vibratome section took on, even while cured underlead weights. Usually, 5–10 ultrathin sections needed to be col-lected from each vibratome section, so as to reach the target ofsampling from the 3 subregions of the PFC.

Quantification of Axo-Somatic Contact Sizes Made by GAD AxonTerminals and the Proportion of Perikaryal Plasma MembranesContacted by GAD Axon TerminalsSomatic profileswere identified using standardmorphological cri-teria (Peters et al. 1991). Somatic profiles were further identified tobe of pyramidal cells and not GABAergic, based on the absence ofdeep indentations along the nuclear membranes, homogeneousdistribution of chromatin within the nucleoplasm, and the ab-sence of GAD-immunoreactivity in the cytoplasm (Peters andJones 1984; White and Keller 1989). Contacts formed by GAD-immunoreactive axon terminals (heretofore referred to as “GADterminal contacts,” for short) were considered to be synapticwhen they exhibited >5 vesicles within the profile, a parallel align-ment between the plasmamembrane of the axon terminal and ofthe soma and thin or complete absence of postsynaptic density.Essentially, these are the features associated with symmetricsynapses (Peters and Jones 1984). Typically, symmetric synapticjunctions were flanked by plasma membranes lacking the pre-to-postsynaptic parallel alignment but with the 2 plasma mem-branes still juxtaposed directly against one another. These non-junctional flanking regions were included in the quantificationof GAD terminal contact lengths (see below for further details).

Digital electronmicroscopic imageswere captured fromultra-thin sections by using a 1.2-megapixel Hamamatsu CCD camerafrom AMT (Boston, MA, USA) from the JEOL 1200 XL electronmicroscope (JEOL Ltd, Tokyo, Japan). The extent of contactformed by GAD terminals upon layer V pyramidal cells of theCg1, PrL, and MO regions of the medial PFC (Fig. 1A) was quanti-fied using the segmented line tool of NIH’s software, Image J(version 1.46r), to measure plasma membrane lengths of theneuronal profiles within digital images of the ultrathin sections.After determining the total perikaryal plasma membrane lengthof a profile identified to be belonging to a pyramidal neuron, theproportion (in units of percent) of the plasma membranecontacted by GAD terminals was determined by retracing thoseportions of the plasma membrane directly opposite to the HRP-DAB-labeled axon terminals. Portions of GAD-immunoreactiveaxon terminals lacking the parallel alignment to the plasmamembrane of the cell bodies but still directly juxtaposed were in-cluded in ourmeasurements of GAD terminal contacts (Figs 1–3).So as to optimize consistency in discriminating immunolabeledfrom unlabeled or background labeling, axon terminals contain-ing faint HRP-DAB-like density in the cytoplasm that was lessthan the electron density of mitochondria were categorized asbelow threshold for definitively positive immunoreactivity andnot included in the quantification (Fig. 1C). Conversely, some ofthe profiles contacting cell bodies were so densely filled withHRP reaction products that vesicles could not be visualized easily.However, inspection of the same profiles recurring in adjacent ul-trathin sections more removed from the surface of vibratomesection surfaces verified that these did contain vesicles (Fig. 3),

thereby indicating that these intensely darkened profiles wereGAD-immunoreactive axon terminals that could be included inthe quantification.

The size of GAD terminals forming contacts with layer V pyr-amidal neurons’ cell bodies was assessed in 2ways: The length ofcontact, using the measurement described above, and cross-sec-tional areas of GAD terminal cytoplasm at the site forming directcontacts with the cell bodies (Fig. 1B). GAD terminal area wasmeasured using Image J’s polygon tool.

The groupmean values of the proportion of layer V pyramidalcells’ perikaryal plasma membrane contacted by GAD terminals,plasma membrane lengths contacted by GAD terminals, and thecytoplasmic areas of the GAD terminals at the site of axo-somaticcontact were compared across the ABA versus CON groups. Thegroup mean values were compared for each of the 3 PFC subre-gions separately and after pooling across the areas.

Statistical Analyses

Statistical analysis was done by SPSS statistical software 22.0(IBM Corp., Armonk, NY, USA) and GraphPad Prism version 6.0(GraphPad Software, San Diego, CA, USA). Data were verified tobe of normal distribution by using the Kolmogorov–Smirnov,the Lilliefors, and Shapiro–Wilk’s W tests. All variables werefound to be normally distributed. For comparison of 2 groups,Student’s t-test was used to determine significance of the differ-ence. For comparison of 3 groups, an one-way repeated-mea-sures ANOVA was conducted, followed by Bonferroni’s post hocanalysis. Pearson correlation was conducted between behavioralactivities and GAD-immunoreactive measurements. P-values of<0.05 were considered statistically significant.

ResultsVariability in Wheel Activity during the Firstand Second ABA

Seven female C57BL6Jmice, derived from 3 litters, were subjectedto 2 sessions of ABA inductions. The ABA inductions were sepa-rated by a week of recovery, consisting of ad libitum food accessfor 24 h and no wheel access. Both ABA inductions occurred dur-ing adolescence (P36–P60).

All 7 mice exhibited hyperactivity within 24 h after theonset of FR, relative to the pre-FR baseline activity, and all butone were hyperactive within the first 6 h of FR. The average ofthe 7 animals’ wheel activity (in km/day) during the 3 days ofFR of the first ABA induction increased by 68%, relative to thebaseline activity that preceded FR and this increase was signifi-cant (18.7 ± 1.68 km/day during first ABA; 11.14 ± 1.55 km/dayduring baseline; F2,12 = 5.08, P < 0.05, by one-way repeated-mea-sures ANOVA; P < 0.05, Bonferroni’s post hoc test, comparingfirst ABA with baseline).

Notably, although all animals exhibited FR-evoked hyper-activity, their degrees of hyperactivity were variable. Two of the7 animals exhibited only a minimal increase in activity followingthe first ABA (<3.5 km/day increase, Animals 3 and 4), whereasthe remaining animals exhibited an increase in activity by atleast 7 km/day (Animals 5–8 and 11; Fig. 4A). The time that ani-mals dwelled on the wheel correlated strongly with the distancethat animals ran on thewheel (R = 0.9926, P < 0.0001; Fig. 4B), indi-cating that individual differences in wheel-running distancearose mostly from the individual differences in the “preference”of using the running wheel, rather than the speed with whichthey ran on the wheel.


at New

York U



ownloaded from


Animals restored their body weight within the first 48 h of therecovery period, as was reported earlier (Chowdhury et al. 2013).Greater variability inwheel activitywas observed following expos-ure to the secondABA induction.One of the animals (Animal 3) ex-hibited strong resistance to the second ABA exposure, in that itsactivity level after FR was less than the level measured duringthe initial acclimation to the wheel. Two animals exhibited onlya minimal increase in activity following the second ABA (≤5 km/day increase, relative to the running during the baseline; Animals4, 6, and 7). The remaining animals exhibited an increase in activ-ity by at least 7 km/day (Animals 5, 8, and 11; Fig. 4A). These indi-vidual differences were apparent within the first 12 h of thesecond ABA. Owing to this individual variability in the FR-evokedhyperactivity, there was no statistically significant difference in

the group mean average of wheel running during the second ex-posure to ABA, relative to the pre-FR baseline activity (17.81 ± 2.49km/day during second ABA; 11.14 ± 1.55 km/day during baseline;P = 0.296).

Thus, wheel analysis confirmed that all 7 adolescent femalemice exhibited FR-evoked hyperactivity but that the extent ofthis hyperactivity was variable, especially during the secondABA induction. Each animal’s average hyperactivity evoked dur-ing the first and second ABA inductions relative to the baselineactivity was ranked and shown in Figure 4A. Using this measureto rank ABA resilience, Animals 3, 4, and 6were themost resilientamong the 7 mice.

As was reported earlier by others (Gelegen et al. 2008; Gutier-rez 2013) and us (Chowdhury et al. 2013), amore detailed analysis

Figure 2. GAD axon terminals forming direct contacts onto a cell body of the PrL of the PFC of an ABA animal are shown. Same criteria used for the CON cell bodies were

applied to verify that this cell body belongs to a pyramidal neuron (E). Five white asterisks (near contact 3, in between contacts 4 and 5, to the right of contact 1, and 2 in

between contacts 7 and 8) are GAD axons that are near but not directly contacting the perikaryal plasmamembrane. The 8 direct GAD terminal contacts onto the cell body,

indicated by arrows, are shown in higher magnification in (A–D) and (F–I). Arrowheads indicate the lateral borders used tomeasure GAD terminal contact lengths. Arrows

(A and D) indicate the lateral borders identifiable as synaptic junctions, based on the parallel alignment of the plasma membranes of the axon terminal and of the cell

body. The contacts showing only one or the other of the salient features of synaptic junctions could be verified to be synaptic by viewing adjacent ultrathin sections,

examples of which are shown in Figure 3. Sometimes, an endoplasmic reticulum saccule abutted the intracellular surface of the plasma membrane of the contact site

(C). ER, endoplasmic reticulum; Ast, astrocytic profile. Calibration bar = 2 μm in (E); 500 nm in (I), and applies to all other panels as well, except for E.

6 | Cerebral Cortex

at New

York U



ownloaded from


of the animals’wheel running revealed that once food access be-comes limited to certain hours of the day, animals begin to ex-hibit particularly high wheel running during the 6-h bin thatimmediately precedes feeding, referred to as the food anticipa-tory activity (FAA). The remaining 18 h also contribute signifi-cantly, albeit less, to the overall increase in wheel running thatis evoked by FR (Chowdhury et al. 2013). Analysis of this cohort’swheel running indicated that these animals also exhibited FAAduring the 6-h preceding feeding (Fig. 4C). The 4 animals exhibit-ing the greatest vulnerability (7, 5, 8, and 11, ranked in the order ofABA severity) exhibited strong correlation between FAA runningand their 24 h activity during the first ABA induction (R = 0.9524;P < 0.05) as well as a marginal correlation with their total wheelactivity, spanning from the baseline days to the end of the secondABA (R = 0.8771; P = 0.12).

To determine whether animals would reduce the running be-havior to gain more food during the period of limited food access(7 PM to 9 PM) under FR, thewheel running activity during 7 PM to9 PM in baseline, first ABA, re-acclimation, and second ABAweremeasured. No difference in the ABA group mean values of wheelrunning was found in first ABA and second ABA, compared withbaseline (Fig. 4D). This indicates that animals voluntarily chose torun on the wheel, rather than eat, even during the hours of foodavailability, after they had become entrained to the restrictivefeeding schedule and had lost, on average, 18% of their bodyweight.

If the emergence of increased wheel activity during the 6-hpreceding feeding is a reflection of animals’ entrainment to thelimited hours of food access, then those animals that exhibitheightened FAA might also be the ones to maximize feeding bysuppressing wheel running during the hours of 7 PM to 9 PM,when food is available. However, no correlation was foundbetween individual animals’ FAA and wheel activity duringfood access (R = 0.22; P = 0.63).

GAD Terminal Contacts onto Cell Bodies of Layer VPyramidal Neurons in the PFC of Animals Following ABA

We assessed the extent of axo-somatic contacts formed by GADterminals onto layer V pyramidal neurons by quantifying the pro-portion of the perikaryal plasmamembrane contacted directly by

axon terminals immunoreactive for GAD, the rate-limiting en-zyme for the synthesis of GABA (Fig. 1B). These GAD-immunor-eactive axon terminals formed only symmetric synapticjunctions, characterized by the parallel alignment of the pre- topostsynaptic plasmamembranes and vesicles on the presynapticside (Figs 1–3). With the use of EM, these contacts were verified tobe direct, that is, not interposed by astrocytic or other axonal pro-cesses that can be as fine as 30 nm in diameter (Figs 1 and 2).Symmetric synaptic specializations (synaptic junctions) weretypically flanked by plasma membranes that lacked the clearlyparallel pre-to-postsynaptic alignment, but nevertheless main-tained direct contact. Measurements of the GAD terminal con-tacts included these non-parallel flanking portions, so long asthe 2 plasma membranes were not interposed by astrocytic pro-cesses (Figs 1–3).

EM revealed that in all 3 regions of the PFC of CON animals,GAD terminals contacted 11.9 ± 0.95% of the cell body plasmamembrane lengths. In contrast, the extent of contacts formedby GAD terminals onto cell bodies of the ABA animals was 17.3 ±0.64%, reflecting a 45% increase, relative to the CON values [t(12) =4.762, P < 0.05]. This difference was significant (P < 0.05), as weredifferences when comparing each PFC area separately (Fig. 5A,all P < 0.05). This difference, in percentage, could have arisendue to changes in the presynaptic GAD terminals or the postsy-naptic perikaryal plasma membranes. The mean lengths ofthe ultrastructurally analyzed perikaryal plasma membraneswere 26710.4 ± 1080.45 nm for the ABA group and this was notsignificantly different from the values of the CONgroups (27466.8± 1544.15 nm), whether pooled across the 3 groups or analyzedseparately for each PFC area (Fig. 5B, P > 0.05 for all), therebypointing to the change being presynaptic.

Differences in the proportion of plasmamembrane covered byGAD terminals across the groups could have arisen due to differ-ences in GAD terminal contact sizes and/or the frequency of GADterminals contacting the somata. The mean number of GADterminals encountered along the perikaryal plasma membranewas 4.07 ± 0.18 for the PFC of the ABA group and this was not sig-nificantly different from the values for the CON group (3.47 ±0.22), whether analyzed across the 3 regions together (P > 0.05)or separately (P > 0.05; Fig. 4). The possibility that percentagedifferences in the lengths of contact formed by GAD terminals

Figure 3.Ultrastructure of synaptic profiles re-captured in an adjacent ultrathin section. Contacts 3 (PanelA) and 1 (Panel B), captured from an ultrathin section adjacent to

the one shown in Figure 2, verified that these contacts are associated with synaptic specializations, based on the clearer visualization of synaptic vesicles, due to the

lightened DAB reaction product intensity (contact 3), and of the more clearly parallel alignment of the plasma membranes of the axon terminals and of cell body

(arrows, contacts 3 and 1). The portions of the plasma membrane with synaptic specializations contrast the longer contact lengths (arrowheads at the lateral-most

borders) that include plasma membranes lacking the definitively parallel alignments (arrowheads). Calibration bar = 500 nm applies to both panels.


at New

York U



ownloaded from


onto cell bodies reflected differences in GAD terminal contactsizes was tested by analyzing the mean lengths and cross-sec-tional areas of the boutons formed by GAD terminals directlycontacting the somatic plasma membrane of layer V pyramidalcells. The mean value of individual GAD terminal contact profilelengths was 4381.8 ± 154.9 nm for the ABA group, representing a40% enlargement compared with the value from the CON tissue(3125.6 ± 216.9 nm). Each PFC area also exhibited significant dif-ference between ABA and CON groups (Fig. 5C, all P < 0.05). Cor-respondingly, the mean cross-sectional area of GAD boutonsforming contacts onto cell bodies of pyramidal cells was 1 816

382 ± 70886.55 nm2 for the ABA group, representing a >60% en-largement compared with the value from the CON tissue (1 137852 ± 128575.4 nm2). The group difference was also significantfor each of the 3 PFC regions that were analyzed separately(Fig. 5D).

This difference in GAD contact terminal lengths and areas atsites of contact with cell bodies of layer V pyramidal neuronsindicates that enlargement of GAD terminals is likely to havecontributed most significantly to the differences in the propor-tion of the perikaryal plasma membrane receiving GAD terminalcontact.

Figure 4. Individual differences in vulnerability to ABA. (A) Mice varied in their baselinewheel activity, but all exhibited increased activity in response to the first ABA. The

y-axis depicts the averagewheel activity in kmper dayacross 2 days,while each group of 3 bars depicts thewheel activity exhibited byan individual animal. The bar graphs

are presented in an order of increasing vulnerability to ABA, from left to right. The average of increaseswheel running during the first and secondABA, relative to baseline,

is indicated as thewhite star on each gray bar. The degree of change in activity following FR, compared with baseline, fell into 2 groups, with some exhibiting an increase

that was >7 km per day on average between the first ABA and baseline (animals 7, 5, 8, and 11), and the remaining animals exhibiting an increase that wasmodest (<4 km

per day). (B) The total time that animals dwelled on the wheel was positively correlated with the total distance that animals ran on the wheel. Correlation reveals

significant effect at 0.0001 level (two-tailed). (C) The wheel running during FAA (1 to 7 PM) and during the hours other than the FAA was increased significantly after

FR. Lines connect the same animal before and after FR. Bar graphs show mean values. *P < 0.05. (D) To determine whether wheel running was changed during the

hours of food access (7–9 PM) after FR, the wheel running activity during food access of baseline, first ABA, recovery, and second ABA were measured. No difference

was found after FR. Lines connect the same animal before and after FR.

8 | Cerebral Cortex

at New

York U



ownloaded from


Correlation between the Running Wheel Activityand GAD Terminals’ Axo-Somatic ContactLengths on Layer V Pyramidal Neurons

Previous studies indicated that the anterior cingulate cortex (ACg)(Cg1), located at the dorsal-most region of the PFC, is linked tomotor behaviors (Heidbreder and Groenewegen 2003; Vertes2006). We investigated whether the inter-animal differences inbehavioral responses to the FR might correlate with differencesin the inhibitory input onto layer V pyramidal neurons in Cg1.

The relationship between lengths of GAD terminal contactwiththe somatic plasmamembrane of Cg1 layer V pyramidal cells andrunning wheel activity in the ABA paradigm was examined usingthe Pearson correlation analysis. Therewasno correlation betweenrunning activity prior to FR and GAD terminal contact lengths(Fig. 6A). In contrast, GAD terminal contact lengthswere negativelycorrelated with the wheel-running activity during the first ABA(Table 1 andFig. 6B),meaning that theanimals exhibiting the great-est increase inwheel activitywere the ones that exhibited the low-est increase of GAD terminal contact lengths at layer V cell bodies.The negative correlationwas even stronger during the second ABAexposure (Table 1 and Fig. 6C). The negative correlation betweenGAD terminal contact length and wheel running was also strongfor the total running activity (sum of running from baselinethrough the first as well as the second ABA; Table 1 and Fig. 6D).

GAD terminal size analysis further confirmed that the cyto-plasmic area of GAD terminals at the site of contact with cell bod-ies was also negatively correlated with the total wheel-runningactivity (Table 1).

In contrast to Cg1, which is linked to motor behaviors, theventral regions of the PFC areas PrL and Mo are associated withdiverse emotional, cognitive, and mnemonic processes and lessto motor behaviors (Heidbreder and Groenewegen 2003; Vertes2006). The strong correlations observed between the contactby GAD terminals onto cell bodies of layer V pyramidal neuronsand running activity were specific for the Cg1, in that the PrL(Fig. 6E) and MO (Fig. 6F) did not exhibit these correlations(Table 1).

Correlation between the FAA, Running during FoodAccess, and GAD Terminal Contact Lengths with SomaticPlasma Membranes of Layer V Pyramidal Neurons

Although axo-somatic contact lengths formed by GAD terminalsin the PrL and MO did not exhibit correlation with wheel runningduring the first or second ABA, a possibility remained that someother subcomponent of the wheel activity might. Indeed, correl-ationwas also found between GAD terminal contacts onto soma-ta in the PrL and the running during the hours of food access,from 7 PM to 9 PM, of the second ABA (Table 1 and Fig. 7B). The

Figure 5.Quantification of GAD-immunoreactive axon terminals contacting cell bodies of layer V pyramidal neurons in PFC areas following ABA. (A) Quantification of the

proportion of somatic plasmamembrane contacted by GAD-immunoreactive axon terminals in 3 different regions of prefrontal areas—Cg1 (cingulate cortex, area 1), PrL

(prelimbic cortex), andMO (medial orbital cortex). Comparisons of the somatic profiles revealed that the plasmamembrane from the ABA animals weremore extensively

covered by GAD-immunoreactive axon terminals thanwere the profiles frombrains of CON animals across all 3 regions (n = 5–7 for each group, representing the number of

animals). (B) The mean lengths of the ultrastructurally analyzed perikaryal plasma membranes in 3 regions. No difference was found between the values of the CON

groups and ABA animals. (C) GAD terminal lengths at sites of contact with the somatic plasma membrane of layer V pyramidal neurons. The same profiles analyzed

for (A) were re-analyzed to determine the mean lengths of GAD terminals forming contact with the somatic plasma membrane of pyramidal cells in prefrontal areas.

Statistically significant enlargement of GAD terminal contacts onto somatic membrane of the ABA, relative to the CON tissue, was detected in all 3 regions of

prefrontal areas. Graph represents mean + SEM values. *P < 0.05. (D) GAD terminal areas at sites of contact with the somatic plasma membrane of layer V pyramidal

neurons. Statistically significant enlargement of GAD terminal area of the ABA, relative to the CON tissue, was detected in all 3 regions of prefrontal areas. Graph

represents mean + SEM values. *P < 0.05.


at New

York U



ownloaded from


correlation was negative and significant, meaning that animalsexhibiting the least wheel activity during the food access periodwere the ones that exhibited the greatest increase of GAD termin-al contact lengths. GAD terminal size analysis further confirmedthat the area of GAD terminals at axo-somatic contact sites in thePrL was also negatively correlated with the running during thehours of food access (Table 1). In contrast, GAD terminal contactlengths onto cell bodies in the Cg1 and MO did not correlate sig-nificantly with wheel running during food access in the secondABA (Fig. 7A for Cg1; Fig. 7C for MO; and Table 1). None of the 3PFC regions exhibited correlation between GAD terminal contactlengths onto cell bodies of layer V pyramidal cells and runningduring the hours of food access of the first ABA (Table 1).

Interestingly, correlation was also found between GAD ter-minal contact lengths in the PrL and FAA, spanning the 6 h thatpreceded feeding during the first ABA (Fig. 7E and Table 1). Thecorrelation was “positive,” meaning that animals exhibiting thegreatest wheel activity during the FAA period were the onesthat exhibited the greatest increase of GAD terminal contact

lengths at cell bodies. MO, the most ventral part of PFC, also ex-hibited marginally positive correlation between GAD terminalcontact lengths at cell bodies and FAA of the first ABA (Fig. 7Fand Table 1). In contrast, GAD terminal contact lengths onto pyr-amidal cells in layer V of Cg1 did not correlate significantly withFAA during the first ABA (Fig. 7D).

DiscussionThe present study is consistent with the idea that inductionof FR-evoked hyperactivity on the wheel (i.e., ABA) caused en-largement of GAD axon terminals targeting layer V pyramidalcell bodies in the PFC. However, contrary to our expectation, thiscircuitry change is not likely to have supported the FR-evokedhyperactivity, since the 2 parameters correlated negatively andstrongly (R =−0.87, P < 0.01): Individuals with the most enlargedGAD terminal contacts onto somata exhibited the greatest sup-pression of FR-evoked hyperactivity, whereas those animalswith terminal contact sizes no different from controls’ exhibited

Figure 6. Correlation between the running wheel activity and GAD terminal lengths at sites of contact with the somatic plasmamembrane of layer V pyramidal neurons.

(A) No correlationwas found betweenGAD terminal length inCg1 and the averagewheel activity in 3 days baseline. Here and in other panels, GAD terminal lengths of CON

animals are shown for comparisonwith the values of ABA animals. (B) There is a negative correlation between GAD terminal length in Cg1 and the averagewheel activity

during the 3 days of the first ABA. Those animals that exhibited the greatest mean value of GAD axon terminal length ran the least. By the Pearson correlation analysis,

correlation is significant at 0.0489 level (two-tailed). (C) There is a negative correlation betweenGAD terminal length inCg1 and the averagewheel activity during the 3 days

of exposure to the second ABA. Correlation reveals significant effect at 0.009 level (two-tailed). (D) A negative correlation was found between GAD terminal length in Cg1

and total running activity at 0.0267 significant level (two-tailed). Total running activitywasmeasured as the sumof baseline, during thefirst ABA induction, and during the

second ABA induction. (E) No correlation was found between GAD terminal length in PrL and total running activity (P = 0.94). (F) No correlation was found between GAD

terminal length in MO and total running activity (P = 0.21).

10 | Cerebral Cortex

at New

York U



ownloaded from


the greatest hyperactivity. FR-evoked hyperactivity can lead todeath (Chowdhury et al. 2013; Gutierrez 2013). Therefore, sup-pression of FR-evoked hyperactivity is clearly adaptive. We pro-pose that the enhancement of axo-somatic inhibition of layer Vpyramidal neurons in the PFCmay contribute to this adaptive be-havior. In support of this idea, it has been shown by others thatGABAergic synapses exhibit activity-dependent structural plasti-city, including enlargement of synapse sizes [review by FloresandMendez (2014)], and that this enlargement can be accompan-ied by increased expression of GABAergic receptors at synapses(Nusser et al. 1998). Unfortunately, our attempt to quantify α1and β2/3 subunits of GABAA receptors at axo-somatic synapseshas not been successful, so we have yet to determine whetherABA induction increases GABAA receptor density at these contactsites.

Another unexpected finding was that although individual dif-ferences in GAD terminal contact sizes at cell bodies correlatedmost strongly with those in hyperactivity during the secondABA induction, this correlation was already detectable 10 daysprior, during the first ABA induction. Although identification ofthe molecular cascades underlying this change was beyond thescope of this study, an idea compatible with the current findingis that someyet-to-be-identified individual difference supportingactivity-dependent changes in GABAergic synapses existed priorto ABA induction. A number of previously published findingspoint to individual differences in the activity-dependent brainderived neurotrophic factor (BDNF) expression as a possiblemechanism linking the FR-evoked stress or heightened exerciseto the upregulation of GABA receptors and expansion of contactsites formed by GAD terminals onto somata (Neeper et al. 1996;Gomez-Pinilla et al. 2002; Lee et al. 2002; Duman 2005; Stranahanet al. 2007, 2009; Jiao et al. 2011).

Functional Significance of the Increased Lengths ofContacts Formed by GABAergic Axon Terminals ontoLayer V Pyramidal Neurons in the PFC

PFC is a site of integration of multiple sensory modalities in add-ition to signals associated with visceral states, rewards, socialconflicts, stress, attention, and “affective qualities” [reviewed byVertes (2006)]. The PFC converts these signals into efferent sig-nals that contribute toward behavior that are in accordancewith internal goals. The PFC has been described as contributingto executive functioning which, for rodents, includes attentionalselection, resistance to interference, reduction in impulsive be-havior, and ultimately generating behavioral flexibility for plan-ning and decision-making (Dalley et al. 2004; Erlich et al. 2011).Because layer V pyramidal neurons are the major efferent neu-rons of the PFC to motor centers (Berendse et al. 1992; Gabbottet al. 2005; Riga et al. 2014), increased GABAergic terminals’ con-tact lengths onto layer V pyramidal neurons may be the cellularmechanism that facilitates response selection through inhibitionof impulsive behavior. For rodents, an impulsive initial responseto starvation is to enhance foraging behavior. It has been hy-pothesized that the increased wheel activity that was elicitedby all 7 of our experimental animals during the initial ABA induc-tion may be a substitute for the foraging behavior elicited byhungry rodents in captivity (Gutierrez 2013). Conversely, thedampened wheel activity that was observed by some of themice during the second ABA induction may reflect a learnedbehavior of suppressing what turned out to be an unproductivebehavior during the first ABA induction, mediated through en-hanced GABAergic inhibition of excitatory outflow from the PFCto themotor centers. As for themechanismsupporting enhancedT

able

1Su

mmaryofco

rrelationco

efficien

tsan

dP-va

lues

betw

eenru

nningac

tivities

andco

ntact

size

s

Brain

region

Runningduringba

seline

Runningduringth

efirstABA

Runningduringth

ese

condABA

Totalru

nningac

tivity

Cg1

PrL

MO

Cg1

PrL

MO

Cg1

PrL

MO

Cg1

PrL

MO

Contact

size

:length(nm)

Pearso

nr=0.11

Pearso

nr=0.33

Pearso

nr=0.36

Pearso

nr=−0.76

Pearso

nr=0.1

Pearso

nr=0.71

Pearso

nr=−0.87

Pearso

nr=−0.08

Pearso

nr=−0.13

Pearso

nr=−0.81

Pearso

nr=−0.03

Pearso

nr=−0.53

P=0.82

P=0.48

P=0.43

P=0.04

P=0.84

P=0.07

4P=0.01

P=0.86

P=0.78

P=0.02

P=0.95

P=0.21

Contact

size

:area

(nm

2)

Pearso

nr=0.13

Pearso

nr=0.13

Pearso

nr=−0.1

Pearso

nr=−0.69

Pearso

nr=−0.37

Pearso

nr=0.72

Pearso

nr=−0.37

Pearso

nr=−0.19

Pearso

nr=0.34

Pearso

nr=−0.78

Pearso

nr=−0.28

Pearso

nr=−0.28

P=0.79

P=0.78

P=0.84

P=0.08

P=0.41

P=0.07

P=0.41

P=0.68

P=0.46

P=0.04

P=0.55

P=0.54

FAA

duringth

efirstABA

Runningduringfoodac

cess

inth

efirstABA

Runningduringfoodac

cess

inth

ese

condABA

Cg1

PrL

MO

Cg1

PrL

MO

Cg1

PrL

MO

Contact

size

:length(nm)

Pearso

nr=−0.51

Pearso

nr=0.75

Pearso

nr=0.69

Pearso

nr=−0.68

Pearso

nr=−0.25

Pearso

nr=0.56

Pearso

nr=0.09

Pearso

nr=−0.84

Pearso

nr=−0.17

P=0.24

P=0.05

P=0.08

P=0.09

P=0.58

P=0.19

P=0.86

P=0.01

P=0.72

Contact

size

:area

(nm

2)

Pearso

nr=−0.69

Pearso

nr=0.48

Pearso

nr=0.69

Pearso

nr=−0.66

Pearso

nr=−0.71

Pearso

nr=0.59

Pearso

nr=0.22

Pearso

nr=−0.79

Pearso

nr=0.02

P=0.09

P=0.27

P=0.08

P=0.11

P=0.07

P=0.16

P=0.64

P=0.03

P=0.97

Bold

values

indicatestatistica

llysign

ifica

nt,P<.05.

ABA,a

ctivity-ba

sedan

orexia;

Cg1

,cingu

late

cortex

,area1;

PrL,

prelim

bicco

rtex

;MO,m

edialo

rbital

cortex

;FAA,foodan

ticipatory

activity.


at New

York U



ownloaded from


GABAergic inhibition, elevated levels of BDNF in the PFC havebeen hypothesized for another learned behavior—the extinctionof fear memory (Bredy et al. 2007), and may be at work for theABA-induced mice as well.

Functional Significance of the Regional Differences in theCorrelation between GAD Terminals’ Axo-SomaticContact Lengths and Wheel Activity

Cg1, PrL, andMOare all parts of themedial PFC, but receive differ-ent afferents (Hoover and Vertes 2007). Specifically, the dorsallylocated Cg1 receives primarily sensorimotor cortical input,whereas themore ventrally located PrL and MO receive primarilylimbic inputs that are both cortical and subcortical, includingthe amygdala, hippocampus, basal forebrain area, the mono-aminergic nuclei, and themidline thalamus. All 3 of thesemedialPFC regions exhibited increased GAD terminal sizes in responseto ABA induction. However, we also noted individual differencesin GAD terminal contact sizes among the ABA-inducedmice and

discovered that GAD terminal sizes within specific subregionscorrelate with distinct components of the FR-evoked hyperactiv-ity. Specifically, within the area PrL, GAD terminal sizes corre-lated negatively and strongly with wheel activity during the 2 hof food availability of the food-restricted days. In otherwords, an-imals with the largest GAD terminals in the PrL were the onesthat were able to suppress wheel activity the most during the2 h of food availability. Although the PrL region is part of themed-ial PFC, as is the Cg1, highly localized lesion studies within themedial PFC have revealed that there are behavioral tasks linkedmore to the PrL than to the Cg1. These include the delayednon-matching-to-sample response (Delatour and Gisquet-Verrier 1999), extra-dimensional and cross-modal set-shifting,and shifts between new strategies or rules (Dalley et al. 2004).As was described earlier, cage-housed mice that are resilientand survive ABA induction are those that can learn a new rule,namely that wheel running does not yield a new source of foodand is detrimental within the confines of a cage. The enlarge-ment of GAD terminal contact sizes at the PrL cell bodies may

Figure 7. Correlation between the FAA, running during food access, and GAD terminal lengths at sites of contact with the somatic plasmamembrane of layer V pyramidal

neurons. (A) No correlation was found between GAD terminal contact length in Cg1 and running during food access in the second ABA (P = 0.86). Here and in other panels,

GAD terminal contact lengths of CON animals are shown for comparison with the values of ABA animals. (B) There was a negative correlation between GAD terminal

contact length in PrL and running during food access in the second ABA. Pearson correlation was significant at P = 0.016 level (two-tailed). (C) No correlation was found

between GAD terminal contact length in MO and running during food access in the second ABA (P = 0.71). (D) No correlation was found between GAD terminal contact

length in Cg1 and FAA during the first ABA (P = 0.24). (E) There was a positive correlation between GAD terminal contact length in PrL and FAA during the first ABA.

Pearson correlation was significant at P = 0.05 level (two-tailed). (F) A marginal positive correlation was found between GAD terminal contact length in MO and FAA

during the first ABA, at P = 0.08 level (two-tailed).


at New

York U



ownloaded from


have contributed toward the animal’s learned behavior and/orshifting to this new rule, even if running is intrinsically reward-ing (Meijer and Robbers 2014) and helpful for foraging.

GAD terminal sizes in Cg1 also correlated with individuals’wheel activity. However, in contrast to the PrL, GAD terminalsizes in the Cg1 correlated most strongly with the animals’ totalwheel activity during the second ABA, and not specifically to the2 h of food availability. Although weaker, Cg1 GAD terminal con-tact sizes also correlated significantly with the 24-h activity dur-ing the first ABA, even though these preceded euthanasia bymore than 10 days. Within the medial PFC, the most dorsally lo-cated Cg1 (corresponding to ACg in rats; Paxinos and Watson1998; Ongur and Price 2000) projectsmost prominently to the dor-sal striatum (Gabbott et al. 2005). Based on this trajectory, it is notsurprising that inhibition of these excitatory outflows from thePFC would also suppress locomotor activity.

Among the 3 animals—3, 4, and 6—that exhibited thegreatest resilience to ABA induction by reducing wheel activity,it was Animal 4 that exhibited the largest GAD terminal contactlengths and bouton areas in PrL and the least wheel activityduring the hours of food availability. The other 2 animals(3 and 6) exhibited the largest GAD terminal contact lengthsand areas in Cg1 and minimal overall wheel activity duringthe FR days. These observations indicate that the cellular re-sponse for attaining resilience to ABA induction varied acrossanimals—enlargement of GAD terminal contacts in PrL or inCg1. For both the Cg1 and PrL, the correlation between GADterminal contact lengths and wheel activity suppression wasmore robust for the second ABA than the first. Perhaps, theGAD terminal enlargement was cumulative through the firstABA, the recovery phase, and the second ABA.

It is reasonable to consider that those animals exhibiting themost heightened FAA are the onesmost strongly entrained to therestricted feeding schedule and have undergone the most robustalteration in PFC circuitry. This idea is supported by our observa-tion that a positive correlation exists between FAA and GAD ter-minal sizes—those individuals with largest GAD axo-somaticcontact lengths in the PrL and MO are the ones that exhibitedthe highest FAA. Interestingly, there was no correlation betweenFAA and suppression of activity during the non-FAA period, in-cluding the 2 h of food access. This indicates that entrainmentto the feeding schedule (as expressed by the heightened FAA)was not sufficient to elicit behavior according to the new rule—to suppress the intrinsically rewarding wheel activity duringthe hours of food availability, when food access is limited. Thelack of correlation between FAA and wheel running suppressionduring the non-FAA hours is explained by the diversity amongthe mice in terms of attaining resilience. As described above,one attained resilience by reducing wheel activity during the2 h of food availability, while 2 attained resilience by reducingoverall activity during the FR days. The remaining 4 did not attainresilience because they did not reduce activity during the non-FAA hours, even though their FAAs were high. These 4 that failedto elicit behavior according to the new rule were the ones withsmaller GAD terminal sizes, no different from controls’.

Prior to this study, we showed that ABA induction during ado-lescence evokes a change in the GABAergic systemwithin 2 otherbrain regions—the hippocampal CA1 and the basolateral amyg-dala. The change evoked in the hippocampal CA1was to increaseGABAergic inhibition through increased contact lengths byGABAergic axon terminals onto cell bodies and dendrites of pyr-amidal cells (Chowdhury et al. 2013). After just one exposure toABA, we also observed increased plasmalemmal expression ofnon-synaptic GABAergic receptors at spines in the hippocampus

(Aoki et al. 2012, 2014). Since upregulation of GABAergic terminalcontact sizes onto cell bodies and of non-synaptic GABAergic re-ceptors at spines was greater among animals exhibiting suppres-sion of the FR-evoked hyperactivity (Chowdhury et al. 2013; Aokiet al. 2014), these changes to the hippocampal GABAergic systemmay have contributed, together with the medial PFC, toward re-ducing the FR-evoked hyperactivity. Whether other corticalareas also undergo changes in the GABAergic system and maycontribute toward suppression of hyperactivity or whether thechanges evoked in the hippocampus and/or PFC are sufficientto alter behavior remains to be determined.

Basolateral amygdala, a hub of emotion circuits in the brain(LeDoux 2000), was analyzed within brains of rats, immediatelyafter they underwent single ABA induction (Wable et al. 2014).Here, the change evoked by ABA induction was an increase inthe expression of non-synaptic GABA receptor expression by in-hibitory interneurons, which would be expected to increase theexcitatory outflowand possibly contribute toward increased anx-iety and the initial FR-evoked hyperactivity (Wable et al. 2015). Inthe current study, we observed a modest positive correlation be-tween GABAergic axon terminal contact sizes (lengths and areas)impinging upon layer V pyramidal neurons of the MO and FAAwheel activity. MO is a visceromotor limbic cortex with projec-tions to the dorsalmotor nucleus of vagus, nucleus of the solitarytract and amygdala, through which it is proposed to modulateautonomic activity and emotion (Fisk and Wyss 2000; Vogtet al. 2004). Assuming the murine basolateral amygdala under-goes similar GABAergic reorganization as observed in rats, the re-modeling of the GABAergic system inMOmay be a compensatorymechanism to dampen the heightened anxiety brought about bythe increased excitatory outflow from the basolateral amygdala.

ConclusionsIt is well established that food deprivation evokes robust and re-liable increases in voluntary wheel running by rodents (Sherwin1998). Among themany ideas about why food deprivation evokeswheel running in rodents, one prevalent idea is that this reflectsthe adaptive behavior of foraging within the wild by hungry ani-mals, but could also be related to the maladaptive, over-exerciseresponse elicited by a large proportion of patients diagnosedwithAN (Epling et al. 1983; Davis et al. 1997, 1999; Hebebrand and Bulik2011; Zunker et al. 2011; Gutierrez 2013). This idea is the basis forusing ABA as an animalmodel for understanding and developingtreatments for AN (Gutierrez 2013). AN is a complex psychiatricillness with biological bases. Twin studies indicate that there isa genetic basis for vulnerability to AN (Bulik et al. 2007), and ani-mal models have pointed to a link between genes modulatinganxiety and ABA vulnerability (Gelegen et al. 2006, 2007, 2008,2010). However, unlike other mental and neurological illnessessuch as schizophrenia, autism, epilepsy, or depression, the lar-gest scale search for a genetic link to AN has yet to identify can-didate genes (Boraska et al. 2014). In our opinion, one importantcontributionwe havemade to the field of ABA is the utility of cor-relating individual differences in vulnerability to ABAwith cellu-lar and molecular markers.

Our findings indicate that ABA induction among adolescentfemalemice has an impact on PFC circuitry and could be contrib-uting toward prevention of weight loss caused by FR-evokedhyperactivity The individual differences in response to ABA in-duction and AN vulnerability may arise due to GAD gene poly-morphism, which exists (Addington et al. 2005; Hettema et al.2006; Zhao et al. 2007), or epigenetic regulations of the GADgene promoter (Kundakovic et al. 2007; Satta et al. 2008; Zhang


at New

York U



ownloaded from


et al. 2010) in themedial PFC. More studies are needed to identifythemolecularmechanisms underlying the differential responsesof GAD terminals in the PFC and whether it is mediated throughthe activity-dependent release of BDNF or other neurotrophicfactors, such as NGF1-A (Zhang et al. 2010). What is also neededis a direct test of whether increased GABAergic inhibition of layerV pyramidal neurons in the PFC will enable individuals to exertcognitive control over the impulsive desire to exercise, ratherthan to eat.

FundingThis study was supported by the National Institutes of Health[NEI Core grant EY13079, 1 R21MH091445-01, 1 R21MH105846-01, R01NS066019-01A1, and R01NS047557-07A1 to C.A., T32MH019524 to G.S.W., and the UL1 TR000038 from the NationalCenter for the Advancement of Translational Science (NCATS)to T.G.C.]; The Fulbright Graduate Study Grant (Fulbright Taiwan,to Y.-W.C.); and NYU’s Research Challenge Fund to C.A.

NotesWe thank Kei Tateyama for her help with electron microscopy.Conflict of Interest: None declared.

ReferencesAddington AM, Gornick M, Duckworth J, Sporn A, Gogtay N,

Bobb A, Greenstein D, Lenane M, Gochman P, Baker N, et al.2005. GAD1 (2q31.1), which encodes glutamic acid decarboxyl-ase (GAD67), is associated with childhood-onset schizophre-nia and cortical gray matter volume loss. Mol Psychiatry.10:581–588.

American Psychiatric Association. 2013. Diagnostic andStatistical Manual of Mental Disorders: DSM-5. Washington,DC: American Psychiatric Publishing.

Andersen SL, Thompson AT, Rutstein M, Hostetter JC,Teicher MH. 2000. Dopamine receptor pruning in prefrontalcortex during the periadolescent period in rats. Synapse.37:167–169.

Aoki C, Sabaliauskas N, Chowdhury T, Min JY, Colacino AR,Laurino K, Barbarich-Marsteller NC. 2012. Adolescent femalerats exhibiting activity-based anorexia express elevated levelsof GABA(A) receptor alpha4 and delta subunits at the plasmamembrane of hippocampal CA1 spines. Synapse. 66:391–407.

Aoki C, Wable G, Chowdhury TG, Sabaliauskas NA, Laurino K,Barbarich-Marsteller NC. 2014. alpha4betadelta-GABAARs inthe hippocampal CA1 as a biomarker for resilience to activ-ity-based anorexia. Neuroscience. 265:108–123.

Berendse HW, Galis-de Graaf Y, Groenewegen HJ. 1992. Topo-graphical organization and relationship with ventral striatalcompartments of prefrontal corticostriatal projections in therat. J Comp Neurol. 316:314–347.

Birmingham CL, Su J, Hlynsky JA, Goldner EM, Gao M. 2005. Themortality rate from anorexia nervosa. Int J Eat Disord.38:143–146.

Boraska V, Franklin CS, Floyd JA, Thornton LM, Huckins LM,Southam L, Rayner NW, Tachmazidou I, Klump KL,Treasure J, et al. 2014. A genome-wide association study ofanorexia nervosa. Mol Psychiatry. doi:10.1038/mp.2013.187.

Bredy TW,WuH, Crego C, Zellhoefer J, SunYE, BaradM. 2007. His-tone modifications around individual BDNF gene promotersin prefrontal cortex are associated with extinction of condi-tioned fear. Learn Mem. 14:268–276.

Bulik CM, Slof-Op’t Landt MC, van Furth EF, Sullivan PF. 2007. Thegenetics of anorexia nervosa. Annu Rev Nutr. 27:263–275.

Chattopadhyaya B, Di Cristo G, Higashiyama H, Knott GW,Kuhlman SJ, Welker E, Huang ZJ. 2004. Experience and activ-ity-dependent maturation of perisomatic GABAergic innerv-ation in primary visual cortex during a postnatal criticalperiod. J Neurosci. 24:9598–9611.

Chowdhury TG, Rios MB, Chan TE, Cassataro DS, Barbarich-Marsteller NC, Aoki C. 2014. Activity-based anorexia duringadolescence disrupts normal development of the CA1pyramidal cells in the ventral hippocampus of female rats.Hippocampus. doi:10.1002/hipo.22320.

Chowdhury TG,Wable GS, SabaliauskasNA, Aoki C. 2013. Adoles-cent female C57BL/6 mice with vulnerability to activity-basedanorexia exhibit weak inhibitory input onto hippocampal CA1pyramidal cells. Neuroscience. 241:250–267.

Dalley JW, Cardinal RN, Robbins TW. 2004. Prefrontal executiveand cognitive functions in rodents: neural and neurochemicalsubstrates. Neurosci Biobehav Rev. 28:771–784.

Davis C, Katzman DK, Kaptein S, Kirsh C, Brewer H, Kalmbach K,Olmsted MP, Woodside DB, Kaplan AS. 1997. The prevalenceof high-level exercise in the eating disorders: etiological im-plications. Compr Psychiatry. 38:321–326.

Davis C, Katzman DK, Kirsh C. 1999. Compulsive physical activityin adolescents with anorexia nervosa: a psychobehavioralspiral of pathology. J Nerv Ment Dis. 187:336–342.

Delatour B, Gisquet-Verrier P. 1999. Lesions of the prelimbic-infralimbic cortices in rats do not disrupt response selectionprocesses but induce delay-dependent deficits: evidence fora role in working memory? Behav Neurosci. 113:941–955.

Dellava JE, Thornton LM, Hamer RM, Strober M, Plotnicov K,Klump KL, Brandt H, Crawford S, Fichter MM, Halmi KA,et al. 2010. Childhood anxiety associated with low BMI inwomen with anorexia nervosa. Behav Res Ther. 48:60–67.

Dixon DP, Ackert AM, Eckel LA. 2003. Development of, and recov-ery from, activity-based anorexia in female rats. PhysiolBehav. 80:273–279.

Duman RS. 2005. Neurotrophic factors and regulation of mood:role of exercise, diet and metabolism. Neurobiol Aging. 26(Suppl 1):88–93.

Epling WF, Pierce WD, Stefan L. 1983. A theory of activity-basedanorexia. Int J Eat Disord. 3:27–46.

Erlich JC, Bialek M, Brody CD. 2011. A cortical substrate for mem-ory-guided orienting in the rat. Neuron. 72:330–343.

Fisk GD, Wyss JM. 2000. Descending projections of infralimbiccortex that mediate stimulation-evoked changes in arterialpressure. Brain Res. 859:83–95.

Flores CE, Mendez P. 2014. Shaping inhibition: activity dependentstructural plasticity of GABAergic synapses. Front CellNeurosci. 8:327.

Gabbott PL, Warner TA, Jays PR, Salway P, Busby SJ. 2005. Pre-frontal cortex in the rat: projections to subcortical auto-nomic, motor, and limbic centers. J Comp Neurol. 492:145–177.

Gelegen C, Collier DA, Campbell IC, Oppelaar H, Kas MJ. 2006.Behavioral, physiological, and molecular differences in re-sponse to dietary restriction in three inbred mouse strains.Am J Physiol Endocrinol Metab. 291:E574–E581.

Gelegen C, Collier DA, Campbell IC, Oppelaar H, van den Heuvel J,Adan RA, Kas MJ. 2007. Difference in susceptibility toactivity-based anorexia in two inbred strains of mice. EurNeuropsychopharmacol. 17:199–205.

Gelegen C, Pjetri E, Campbell IC, Collier DA, Oppelaar H, Kas MJ.2010. Chromosomal mapping of excessive physical activity


at New

York U



ownloaded from


in mice in response to a restricted feeding schedule.Eur Neuropsychopharmacol. 20:317–326.

Gelegen C, van den Heuvel J, Collier DA, Campbell IC, Oppelaar H,Hessel E, Kas MJ. 2008. Dopaminergic and brain-derivedneurotrophic factor signalling in inbred mice exposed to a re-stricted feeding schedule. Genes Brain Behav. 7:552–559.

Goddard E, Ashkan K, Farrimond S, Bunnage M, Treasure J. 2013.Right frontal lobe glioma presenting as anorexia nervosa: fur-ther evidence implicating dorsal anterior cingulate as an areaof dysfunction. Int J Eat Disord. 46:189–192.

Gomez-Pinilla F, Ying Z, Roy RR, Molteni R, Edgerton VR. 2002.Voluntary exercise induces a BDNF-mediated mechanismthat promotes neuroplasticity. J Neurophysiol. 88:2187–2195.

Gutierrez E. 2013. A rat in the labyrinth of anorexia nervosa:contributions of the activity-based anorexia rodent model tothe understanding of anorexia nervosa. Int J Eat Disord.46:289–301.

Gutierrez E, Baysari MT, Carrera O, Whitford TJ, Boakes RA. 2006.High ambient temperature reduces rate of body-weightloss produced by wheel running. Q J Exp Psychol (Hove).59:1196–1211.

Hare BD, D’Onfro KC, Hammack SE, FallsWA. 2012. Prior stress in-terfereswith the anxiolytic effect of exercise in C57BL/6Jmice.Behav Neurosci. 126:850–856.

Hebebrand J, Bulik CM. 2011. Critical appraisal of the provisionalDSM-5 criteria for anorexia nervosa and an alternative pro-posal. Int J Eat Disord. 44:665–678.

Heidbreder CA, GroenewegenHJ. 2003. Themedial prefrontal cor-tex in the rat: evidence for a dorso-ventral distinction basedupon functional and anatomical characteristics. NeurosciBiobehav Rev. 27:555–579.

Hettema JM, An SS, Neale MC, Bukszar J, van den Oord EJ,Kendler KS, Chen X. 2006. Association between glutamicacid decarboxylase genes and anxiety disorders, major de-pression, and neuroticism. Mol Psychiatry. 11:752–762.

Hodes GE, Shors TJ. 2005. Distinctive stress effects on learningduring puberty. Horm Behav. 48:163–171.

Hoover WB, Vertes RP. 2007. Anatomical analysis of afferent pro-jections to themedial prefrontal cortex in the rat. Brain StructFunct. 212:149–179.

Hudson JI, Hiripi E, Pope HG Jr, Kessler RC. 2007. The prevalenceand correlates of eating disorders in the National ComorbiditySurvey Replication. Biol Psychiatry. 61:348–358.

Jiao Y, Zhang Z, Zhang C, Wang X, Sakata K, Lu B, Sun QQ. 2011.A key mechanism underlying sensory experience-dependentmaturation of neocortical GABAergic circuits in vivo. Proc NatlAcad Sci USA. 108:12131–12136.

KayeWH, Bulik CM, Thornton L, Barbarich N,Masters K. 2004. Co-morbidity of anxiety disorderswith anorexia and bulimia ner-vosa. Am J Psychiatry. 161:2215–2221.

Kaye WH, Fudge JL, Paulus M. 2009. New insights into symptomsand neurocircuit function of anorexia nervosa. Nat RevNeurosci. 10:573–584.

Kullmann S, Giel KE, Hu X, Bischoff SC, Teufel M, Thiel A, Zipfel S,Preissl H. 2014. Impaired inhibitory control in anorexianervosa elicited by physical activity stimuli. Soc Cogn AffectNeurosci. 9:917–923.

Kundakovic M, Chen Y, Costa E, Grayson DR. 2007. DNA methyl-transferase inhibitors coordinately induce expression of thehuman reelin and glutamic acid decarboxylase 67 genes.Mol Pharmacol. 71:644–653.

Kvitsiani D, Ranade S, Hangya B, Taniguchi H, Huang JZ, Kepecs A.2013. Distinct behavioural and network correlates of two inter-neuron types in prefrontal cortex. Nature. 498:363–366.

LeDoux JE. 2000. Emotion circuits in the brain. Annu RevNeurosci.23:155–184.

Lee J, Duan W, Mattson MP. 2002. Evidence that brain-derivedneurotrophic factor is required for basal neurogenesis andmediates, in part, the enhancement of neurogenesis by dietaryrestriction in the hippocampus of adult mice. J Neurochem.82:1367–1375.

LozsaA. 1974.Uranyl acetateasanexcellentfixative for lipoproteinsafter electrophoresis on agarose-gel. Clin Chim Acta. 53:43–49.

Meijer JH, Robbers Y. 2014. Wheel running in the wild. Proc BiolSci. 281. doi:10.1098/rspb.2014.0210.

Neeper SA, Gomez-Pinilla F, Choi J, Cotman CW. 1996. Physicalactivity increases mRNA for brain-derived neurotrophicfactor and nerve growth factor in rat brain. Brain Res.726:49–56.

NelsonM. 1985. Level of living and fertility among a rural popula-tion of the Philippines. Stud Comp Int Dev. 20:31–46.

Nusser Z, Hajos N, Somogyi P, Mody I. 1998. Increased number ofsynaptic GABA(A) receptors underlies potentiation at hippo-campal inhibitory synapses. Nature. 395:172–177.

Ongur D, Price JL. 2000. The organization of networks withinthe orbital and medial prefrontal cortex of rats, monkeysand humans. Cereb Cortex. 10:206–219.

Paxinos G, Franklin KBJ. 2001. The mouse brain in stereotaxiccoordinates. 2nd ed. San Diego: Academic Press.

Paxinos G, Watson C. 1998. The rat brain in stereotaxiccoordinates. 4th ed. San Diego: Academic Press.

Perdereau F, Faucher S, Wallier J, Vibert S, Godart N. 2008. Familyhistory of anxiety andmood disorders in anorexia nervosa: re-view of the literature. Eat Weight Disord. 13:1–13.

Peters A, Jones EG. 1984. Cellular components of the cerebralcortex. New York (NY): Plenum Press.

Peters A, Palay SL, Webster HF. 1991. The fine structure ofthe nervous system: neurons and their supporting cells.New York (NY): Oxford University Press.

Riga D, Matos MR, Glas A, Smit AB, Spijker S, Van den Oever MC.2014. Optogenetic dissection of medial prefrontal cortex cir-cuitry. Front Syst Neurosci. 8:230.

Romeo RD. 2010. Adolescence: a central event in shaping stressreactivity. Dev Psychobiol. 52:244–253.

Sato Y, Saito N, Utsumi A, Aizawa E, Shoji T, Izumiyama M,Mushiake H, Hongo M, Fukudo S, et al. 2013. Neural basis ofimpaired cognitive flexibility in patients with anorexia ner-vosa. PLoS ONE. 8:e61108.

Satta R, Maloku E, Zhubi A, Pibiri F, Hajos M, Costa E, Guidotti A.2008. Nicotine decreases DNAmethyltransferase 1 expressionand glutamic acid decarboxylase 67 promoter methylationin GABAergic interneurons. Proc Natl Acad Sci USA. 105:16356–16361.

Sherwin CM. 1998. Voluntary wheel running: a review and novelinterpretation. Anim Behav. 56:11–27.

Spear LP. 2000. The adolescent brain and age-related behavioralmanifestations. Neurosci Biobehav Rev. 24:417–463.

Stranahan AM, Khalil D, Gould E. 2007. Running induceswidespread structural alterations in the hippocampus andentorhinal cortex. Hippocampus. 17:1017–1022.

Stranahan AM, Lee K, Martin B, Maudsley S, Golden E, Cutler RG,Mattson MP. 2009. Voluntary exercise and caloric restrictionenhance hippocampal dendritic spine density and BDNFlevels in diabetic mice. Hippocampus. 19:951–961.

Sullivan PF. 1995. Mortality in anorexia nervosa. Am J Psychiatry.152:1073–1074.

Terzakis JA. 1968. Uranyl acetate, a stain and a fixative.J Ultrastruct Res. 22:168–184.


at New

York U



ownloaded from


Thornton LM, Dellava JE, Root TL, Lichtenstein P, Bulik CM. 2011.Anorexia nervosa and generalized anxiety disorder: furtherexplorations of the relation between anxiety and body massindex. J Anxiety Disord. 25:727–730.

van Kuyck K, Casteels C, Vermaelen P, Bormans G, Nuttin B, VanLaere K. 2007. Motor- and food-related metabolic cerebralchanges in the activity-based rat model for anorexia nervosa:a voxel-based microPET study. Neuroimage. 35:214–221.

Vertes RP. 2006. Interactions among themedial prefrontal cortex,hippocampus and midline thalamus in emotional and cogni-tive processing in the rat. Neuroscience. 142:1–20.

Vogt BA, Vogt L, Farber NB. 2004. Cingulate Cortex and DiseaseModels. In: Paxinos G, editor. The Rat Nervous System.3rd ed. San Diego (CA): Elsevier. p. 705–727.

Wable GS, Barbarich-Marsteller NC, Chowdhury TG,Sabaliauskas NA, Farb CR, Aoki C. 2014. Excitatory synapseson dendritic shafts of the caudal basal amygdala exhibit ele-vated levels of GABAA receptor alpha4 subunits followingthe induction of activity-based anorexia. Synapse. 68:1–15.

Wable GS, Min JY, Chen Y-W, Aoki C. Forthcoming 2015. Anxiety iscorrelatedwith running in adolescent femalemice undergoingactivity-based anorexia. Behav Neurosci. 129(2):170–182.

White EL, Keller A. 1989. Cortical circuits: synaptic organizationof the cerebral cortex—structure, function, and theory.Boston: Birkhauser.

Zhang TY, Hellstrom IC, Bagot RC, Wen X, Diorio J, Meaney MJ.2010. Maternal care and DNA methylation of a glutamic aciddecarboxylase 1 promoter in rat hippocampus. J Neurosci.30:13130–13137.

Zhao X, Qin S, Shi Y, Zhang A, Zhang J, Bian L, Wan C, Feng G,Gu N, Zhang G, et al. 2007. Systematic study of associationof four GABAergic genes: glutamic acid decarboxylase 1gene, glutamic acid decarboxylase 2 gene, GABA(B) receptor1 gene and GABA(A) receptor subunit beta2 gene, with schizo-phrenia using a universal DNAmicroarray. Schizophr Res. 93:374–384.

Zhou Y, Danbolt NC. 2013. GABA and glutamate transporters inbrain. Front Endocrinol (Lausanne). 4:165.

Zipfel S, Lowe B, Reas DL, Deter HC, Herzog W. 2000. Long-termprognosis in anorexia nervosa: lessons from a 21-year fol-low-up study. Lancet. 355:721–722.

Zunker C, Mitchell JE, Wonderlich SA. 2011. Exercise interven-tions for womenwith anorexia nervosa: a review of the litera-ture. Int J Eat Disord. 44:579–584.


at New

York U



ownloaded from


enlargement of axo-somatic contacts formed by gad-immunoreactive axon terminals onto layer v...

Documents