Impaired fear extinction in mice lacking protease nexin-1


Dr D. Monard, as above.


The serine protease inhibitor protease-nexin-1 (PN-1) has been shown to modulate N-methyl-d-aspartate receptor (NMDAR)-mediated synaptic currents and NMDAR-dependent long-term potentiation of synaptic transmission. Here, we analysed the role of PN-1 in the acquisition and extinction of classical auditory fear conditioning, two distinct forms of learning that both depend on NMDAR activity in the amygdala. Immunostaining revealed that PN-1 is expressed throughout the amygdala, primarily in γ-aminobutyric acid containing neurons of the central amygdala and intercalated cell masses (ITCs) and in glia. Fear extinction was severely impaired in mice lacking PN-1 (PN-1 KO). Consistent with a role for the basal nucleus of the amygdala in fear extinction, we found that, compared with wild-type (WT) littermate controls, PN-1 KO mice exhibited decreased numbers of Fos-positive neurons in the basal nucleus after extinction. Moreover, immunoblot analysis of laser-microdissected amygdala sub-nuclei revealed specific extinction-induced increases in the level of phosphorylated alpha-calcium/calmodulin protein kinase II in the medial ITCs and in the lateral subdivision of the central amygdala in WT mice. These responses were altered in PN-1 KO mice. Together, these data indicate that lack of extinction in PN-1 KO mice is associated with distinct changes in neuronal activity across the circuitry of the basal and central nuclei and the ITCs, supporting a differential impact on fear extinction of these amygdala substructures. They also suggest a new role for serine protease inhibitors such as PN-1 in modulating fear conditioning and extinction.


Serine proteases and their inhibitors are expressed and secreted by many cell types in the adult CNS. They play a role in the neuronal response to injuries and their expression can be regulated by neuronal activity (Melchor & Strickland, 2005; Wang et al., 2008). They have also been reported to modulate neuronal function, e.g. through controlled proteolysis of extracellular proteins or indirectly through interaction with membrane proteins, thereby affecting cell surface receptor-mediated neuronal signaling (Melchor & Strickland, 2005; Samson & Medcalf, 2006; Samson et al., 2008; Wang et al., 2008).

Protease nexin-1 (PN-1) is a serine protease inhibitor of the serpin family (Gloor et al., 1986). While constitutively expressed by glial and neuronal subpopulations, its expression is also regulated by neuronal activity (Kvajo et al., 2004). PN-1 levels influence synaptic properties, including long-term potentiation at Schaffer collateral–CA1 synapses in the hippocampus (Lüthi et al., 1997). Mice lacking PN-1 (PN-1 KO) have reduced N-methyl-d-aspartate receptor (NMDAR)-mediated synaptic currents in hippocampal CA1 and cortical layer II/III pyramidal neurons, and display impaired vibrissa sensory processing (Lüthi et al., 1997; Kvajo et al., 2004).

Another prominent area of PN-1 expression is the amygdala – a central part of the circuits assigning emotional valence to sensory stimuli (Davis, 1992; LeDoux, 2000). These circuits have been extensively investigated using the paradigm of classical auditory fear conditioning. In this model, subjects learn to associate a neutral tone, the conditioned stimulus (CS), with an aversive foot shock, the unconditioned stimulus (US). The amygdala consists of many nuclei that are extensively interconnected. The basolateral amygdaloid complex (BLA), which includes the lateral (LA) and basal (BA) nuclei, is considered to be an important site where sensory inputs converge and associations between the CS and the US are formed (Maren, 1999; LeDoux, 2000). Surrounding the BLA are γ-aminobutyric acid containing (GABAergic) interneurons of the intercalated cell masses (ITCs), which are thought to gate information flow into and out of the BLA (Paréet al., 2004; Marowsky et al., 2005; Pape, 2005). These structures influence the central nucleus of the amygdala (CEA), a major source of output neurons projecting to downstream targets (LeDoux, 2000).

Conditioned fear responses can be inhibited by repeated non-reinforced presentations of the CS – a process termed extinction (Myers & Davis, 2007). Both fear conditioning and extinction are NMDAR-dependent (LeDoux, 2000; Myers & Davis, 2007). NMDAR-dependent synaptic plasticity has been described in various nuclei of the amygdala, including the LA (Blair et al., 2001), BA (Maren & Fanselow, 1995; Chapman et al., 2003), ITCs (Royer & Paré, 2002) and CEA (Fu & Shinnick-Gallagher, 2005; Samson & Paré, 2005). As PN-1 can regulate NMDAR function and synaptic plasticity, we compared the acquisition and extinction of auditory fear conditioning in PN-1 KO mice and wild-type (WT) littermates. Then, in order to determine if the pattern of fear conditioning- and extinction-induced biochemical responses distributed over the different nuclei of the amygdala was altered in these mice, we immunohistologically analysed Fos protein expression and, using immunoblot analysis of extracts of laser-microdissected subregions, measured phosphorylated alpha-calcium/calmodulin protein kinase II (pαCamKII) levels.

Materials and methods


PN-1 heterozygote mice (Lüthi et al., 1997) and PN-1HAPN−1-lacZ/HAPN−1-lacZ (PN-1 reporter mice; Kvajo et al., 2004) were derived and backcrossed into the C57BL/6J (RCC, Füllinsdorf, Switzerland) background in our animal facility. Heterozygous mating generated PN-1−/− (PN-1 KO) and PN-1+/+ (WT) littermates. All experimental animals were male, except females were used for PN-1 immunohistology, 4–8 months old, housed on a 12-h day/night cycle with ad libitum access to food and water. Mice were singly housed for at least 2 weeks for all experiments. A total of 101 mice were used in these experiments. All animal experiments were approved by the Swiss Veterinary Authorities and carried out in accordance with the European Communities Council directive (86/609/EEC).

Fear conditioning and extinction

All studies took place during the light portion of the cycle. Mice were handled gently for 2–5 min/day for 5 days. Fear conditioning and extinction sessions took place in two different contexts, basically as described (Herry et al., 2006). Briefly, mice were submitted to fear conditioning protocols in which a 30-s tone CS (7.5 kHz, 80 dB) was systematically paired with a 1-s US (0.6 mA footstock; inter-trial interval 20–180 s). Four CS–US pairings were used in one solely behavioral experiment to determine if the extinction impairment of PN-1 KO mice depended on the number of pairings. Five pairings were used for all other experiments to ensure that WT mice showed strong freezing responses at the beginning of extinction trials. The onset of the US coincided with the offset of the CS. To score freezing behavior, we used an automatic infrared beam-detection system placed on the bottom of the experimental chambers (Coulbourn Whitehall, PA, USA). The mice were considered to be freezing if no movement was detected for 2 s. Freezing was sampled for 2 min before the CS presentation to establish baseline activity and during the 30-s CS presentations. The fear conditioning context differed from the extinction context in shape, smell and light intensity. Conditioning, early extinction and late extinction sessions took place on three consecutive days. The extinction group (ext.) was conditioned in one context, and on the following 2 days underwent early and late extinction training sessions consisting of 16 CS presentations in a different context in order to eliminate contextual conditioning effects. The no extinction group (no ext.) was conditioned as the extinction group, but only exposed to four CS presentations on the following 2 days. The freezing response to the first two CS presentations in early extinction sessions was used as the measure of fear retrieval. The CS-only group (CS-only) underwent the same regime as the no extinction group except that they were never exposed to a foot shock. Naïve control mice for Fos immunohistological staining were handled as above, but kept in their home cage and never exposed to the CS. Unless stated otherwise, behavioral data were analysed by two-way repeated measure anova and Bonferroni post hoc tests (GraphPad Prism4 software, San Diego, CA, USA), and shown as mean ± SEM total freezing time. Student’s t-test analysis was performed using GraphPad Prism4 software. For immunohistochemical and immunoblotting experiments, mice were killed 2 h after the start of Day 3 trials.


Mice were deeply anesthetized using ketamine (ml/kg, i.p.) and perfused transcardially with 50 mL 0.1 m ice cold phosphate-buffered saline, pH 7.4 and 80 mL 4% paraformaldehyde in said buffer (unless specified otherwise, reagents were from FLUKA). The brains were dissected and postfixed for 24 h. Samples for cryostat sectioning were cryoprotected in 25% sucrose for 2 days, embedded in Shandon M-1 Embedding Matrix (#1310 Thermo Electron Corporation, Thermo Fisher Scientific) and frozen in −40°C isopentane. Sections were collected either on slides (12 or 25 μm thick) or as free-floating sections (40 or 60 μm thick) in sterile 0.05 m TRIS-buffered saline-filled wells, pH 7.4 (24-well plates) and stored at 4°C until use. Free-floating sections were stained three per well. Fos stainings were performed essentially as described (Herry & Mons, 2004), except the secondary antibody used was the one supplied in the Elite Vectastain (Vector Labs Burlingame, CA, USA) kit and developed with Fast DAB (Sigma) according to manufacturer’s instructions. Some sections were also revealed by immunofluorescence (as below). Sections from PN-1 reporter mice were immunostained after a 4-h block [3% normal goat serum (Vector Labs)/0.3% Triton-X-100/0.1 m phosphate-buffered saline, pH 7.4] with antibodies to the following proteins overnight at room temperature: ß-galactosidase (mouse monoclonal, 1:1000; Promega; rabbit polyclonal, 1:1000; US Biologicals), glial fibrillary acidic protein (GFAP; rabbit polyclonal, 1:1000; DAKO), NeuN (mouse monoclonal-Alexa 468 coupled, 1:1000; Chemicon/Millipore), glutamic dehydrogenase isoform 67 (GAD67; mouse monoclonal, 1:1000; Chemicon/Millipore). Nuclei were stained with the DNA-binding fluorescent dye TOPRO-3 (1:1000; Invitrogen). Secondary antibodies included goat anti-mouse and anti-rabbit IgG conjugated-Alexa 488 and -Alexa 568 (Invitrogen), incubated for 1 h at room temperature at 1:200 for cryostat sections and 1:1000 for free-floating sections. PN-1 immunostaining (1:100, 4B3 monoclonal antibody; Reinhard et al., 1994) was performed on 12-μm-thick cryostat sections from PN-1 KO and WT mice on the Ventana Discovery XT automated stainer (Roche Diagnostics, Basel, Switzerland). Slides were pre-treated with RiboCC buffer (Ventana) and processed with the Omni-Ultra Map HRP XT (Ventana) procedure omitting DAB and Cu reagents. To detect the immunoreaction, TSA plus fluorescein (1:100, Perkin Elmer) was dropped onto the slides after the end of the run and incubated for 10 min. Sections were mounted in Kaiser’s Gelatin (Merck) or in Prolong Gold antifade reagent (Invitrogen). In all experiments, sections from WT and mutant mice were processed simultaneously. Controls for antibodies included the omission of primary and/or secondary antibodies and single primary antibodies with double secondary antibodies for colocalization experiments. Staining with 4B3 antibody gave no detectable signal on sections from PN-1 KO mice treated under the same conditions as the WT.

Images of Fos-immunostained sections were acquired with a Nikon Eclipse E600 microscope using a 10 × /0.17 lens equipped with a Nikon DX1200 camera and quantitated using ImagePro Plus software (Media Cybernetics, MD, USA). The images were converted to 8-bit gray-scale, a single threshold was chosen such that strongly stained individual nuclei were distinguishable and automatically counted. For stainings revealed by immunofluorescence, counting was performed manually, and no distinction was made between strongly and weakly labeled nuclei. The experimenter was blind to genotype and treatment. Average density was determined from at least three sections per mouse. The data were analysed by two-way anova and Bonferroni post hoc tests (GraphPad Prism4 software), and shown as mean ± SEM cells/mm2. Amygdala wide images were acquired with a Zeiss Z1 microscope using a 5 × 0.13 lens or 20 × 0.8 lens, in which case individual images were stitched together. Images of ß-galactosidase-immunostained sections for colocalization are single optical sections 0.8–0.9 μm thick, scanned with an LSM510, Axiovert 200M confocal microscope, acquired with a 20 × 0.75 lens or 40 × /1.4 oil-immersion lens. The scans were acquired sequentially to ensure that there was no carry over of signal between the channels. The images were quantified using the ImageJ cell count plugin; data are averages of at least three sections from three mice each and shown as mean ± SEM % cells.

Immunoblotting of laser-dissected tissue

Mice were briefly anesthetized with isoflurane inhalation and decapitated. The brains were dissected, frozen in −40°C isopentane and stored at −80°C till used. Frozen brains were brought to −20°C; 14-μm-thick coronal cryostat sections collected on MMI membrane slides (Molecular Machines and Industries, Glattbrugg, Switzerland), dried for 1 h and stored at −80°C till used. For laser dissection, slides were stored on dry ice, one slide at a time was stained with 1 mL bis-benzimide (Hoechst 33258, Sanofi-Aventis; 10 μg/mL in 70% isopropanol) for 1 min, briefly rinsed in 70% isopropanol, dehydrated for 1 min in 100% isopropanol and air dried. The amount of material needed for simple immunoblotting using enhanced chemiluminescence detection can consist of 1000 cells or even fewer, depending on the abundance of the protein and the sensitivity of the primary antibody. Selected areas of the amygdala were cut using MMI CellCut laser microdissection system and captured on 0.5-mL caps of collecting tubes (MMI) for a final area of 200 000 μm2, sufficient for one or two determinations. Tissue areas were collected between Bregma −1.34 and −1.58 except for ITCs, which were collected over a larger range. Generally 10–14 cuts for the medial (mITCs) and lateral ITCs (lITCs), two cuts each for the lateral (no distinctions were made between the lateral and capsular subdivisions) and medial divisions of the CEA, and one cut each for LA and BA were sufficient. Care was taken that when multiple cuts were used they did not overlap on the cap. The tubes were put on dry ice and stored at −80 °C till used. Brief hematoxylin staining was also compatible with subsequent steps of immunoblotting. The slides could be recut at least a few times without noticeable deleterious effects.

Immunoblotting was based on a previously described technique (Martinet et al., 2004), with some modifications. Briefly, proteins were extracted in 14 μL of 2 ×  NuPAGE LDS sample buffer supplemented with 1 ×  Complete Protease inhibitors (Roche Diagnostics), 1 ×  Phosphatase Inhibitors I and II (Sigma), 1 ×  NuPAGE reducing agent for 30 min at 80°C and loaded on 17-well NuPAGE 4–12% bis TRIS gradient gels, and blotted onto Immobilion P (Millipore) according to NuPAGE manual instructions (Invitrogen), except that Miniprotean transfer tanks (BioRad) were used. Samples of a specific area were divided over two gels, such that each gel was loaded with protein from two pairs of WT and KO mice from each of the CS-only, no extinction and extinction groups. Two such gels comprising a complete sampling of one area from 24 mice were run and processed together. Control brain homogenates (50, 100, 200, 400 ng total protein) of WT mice were included to verify that signal development was in the linear range. Invitrogen Plus2 pre-stained Standards were included on all gels. Membranes were blocked for at least 3 h using Amersham’s Advanced ECL kit blocking agent (GE Healthcare). Blots were incubated overnight at 4°C with primary antibodies to: activated alpha-calcium/calmodulin protein kinase II (αCamKII) (pT286), which detects primarily the α subunit (52 kD) (pαCamKII, rabbit polyclonal, 1:1000; Promega), but also weakly the ß subunits (58 kD), actin (mouse monoclonal, 1:1000; Neomarkers) and αCamKII (mouse monoclonal, 1:2000; Upstate/Millipore), and with secondary HRP-coupled biotinylated anti-mouse or anti-rabbit antibodies (1:5000 for all, except 1:10 000 for αCamKII; Thermo Fisher Scientific) for 1 h at room temperature. Blots were washed thoroughly between incubations and developed according to Advanced ECL instructions. Because αCamKII protein runs at the same position as the phosphorylated activated form, carry over signal was reduced by incubating for 15 min in 15% H2O2 after probing for pαCamKII. Signal was revealed with Amersham’s Hyperfilm for ECL (GE Healthcare).

Blots were scanned in 16-bit gray-scale mode, quantified using Odyssey software (LI-COR, Lincoln, Nebraska, USA). Multiple bands for pαCamKII were sometimes visible. Only the main band corresponding to the alpha isoform at 52 kD was used for quantitation. Data were analysed by two-way anova and Bonferroni post hoc tests (GraphPad Prism4 software), and are shown as mean ± SEM integrated density normalized to WT CS-only values. For these experiments, four male mice 3–5 months old were used for each genotype and behavioral group. The behavior of one KO mouse in the extinction group was not included in the analysis as it did not acquire conditioned fear, for a total of 12 WT and 11 KO mice.


PN-1 expression in the amygdala

PN-1 protein is widely expressed throughout the amygdala (Fig. 1A). Because the protein is secreted, it is difficult to determine the pattern of expression at the cellular level. To overcome this difficulty, we used PN-1 reporter mice (Kvajo et al., 2004), which express ß-galactosidase with a nuclear localization signal under the control of the endogenous PN-1 promoter, to identify PN-1-expressing cell populations. Sections from these mice were stained for ß-galactosidase colocalization with neuronal (NeuN or GAD67) and glial (GFAP) markers. Areas of intense GAD67 immunoreactivity were observed in the subregions of the amygdala, which are predominantly composed of inhibitory neurons, namely CEA and the ITCs (Nitecka & Ben Ari, 1987; Cassell et al., 1999; Fig. 1B, green staining). A high density of ß-galactosidase-positive cells was also evident in these areas (Fig. 1B and inserts B1 and B2). Quantification of sections costained with TOPRO-3 confirmed that PN-1-expressing cells make up a high proportion of all cells in the lateral (CEl) and medial (CEm) subdivisions of the CEA, and in the mITC and lITC (Fig. 1C and D; Table 1). PN-1 expression was predominantly neuronal in these areas as determined by the colocalization of the neuronal marker NeuN with ß-galactosidase-immunopositive cells (Fig. 1C and D; Table 1). Furthermore, as neurons in these areas are overwhelmingly GABAergic, these results indicate that PN-1 is expressed by inhibitory neurons.

Figure 1.

 PN-1 is expressed by GABAergic neurons of the CEA and ITCs and by many glial cells. (A) Immunostaining for PN-1 protein (green) shows an extracellular distribution throughout the amygdala of WT mice. Apparent holes are mostly cell soma. (B–F) Immunostaining for nuclear localized ß-galactosidase (red) under the control of the endogenous PN-1 promoter. (B) Immunostaining for the cytoplasmic GABAergic cell marker GAD67 (green). (B1 and B2) Magnified views of (B), showing a monochrome image of ß-galactosidase distribution in the CEl, CEm, mITC, BA and lITC. (B3) Magnified view of (B), arrowheads show colocalization of two ß-galactosidase-expressing cells with GAD67 immunofluorescence in the BLA. (C–F) DNA labeled with TOPRO-3 (blue). (C–E) Co-immunostaining for ß-galactosidase (red) with the nuclear neuronal marker NeuN (green). (F) Co-immunostaining for ß-galactosidase (red) with the cytoplasmic glial marker GFAP (green), arrowheads show colocalization, arrows GFAP alone. Scale bars: 200 μm (A, B); 10 μm (C); 40 μm (B1 and B2); 20 μm (B3, D–F). BA, basal nucleus of the amygdala; CEl, latero-capsular subdivision of the central amygdala; CEm, medial subdivision of the central amygdala; GAD67, glutamic acid dehydrogenase 67-kD isoform; GFAP, glial fibrillary acidic protein; IHC, immunohistochemistry; ITCs, intercalated cell masses; LA, lateral nucleus of the amygdala; lITC, lateral intercalated cell mass; mITC, medial intercalated cell mass; PN-1, protease nexin-1; WT, wild-type littermates of PN-1 KO mice.

Table 1.   PN-1 expression in the amygdala
AreaPN-1*NeuN*PN-1 + NeuNGFAP*PN-1 + GFAP
  1. PN-1 expression was determined in the PN-1 reporter mouse line with ß-galactosidase under control of the endogenous PN-1 promoter. Data are averages of at least three sections from three mice, and shown as mean ± SEM % cells; nd = not determined. *Percent of TOPRO-3-stained cells. Percent of total NeuN-immunopositive cells. Percent of total GFAP-immunopositive cells. BLA, basolateral amygdaloid complex; CEI, latero-capsular subdivision of the central amygdala; CEm, medial subdivision of the central amygdala; GFAP, glial fibrillary acidic protein; lITC, lateral intercalated cell mass; mITC, medial intercalated cell mass; PN-1, protease nexin-1.

BLA22.4 ± 3.357.5 ± 2.76.4 ± 1.926.2 ± 3.653.3 ± 1.1
CEl76.8 ± 2.665.8 ± 4.782.4 ± 0.9ndnd
CEm39.8 ± 3.451.4 ± 10.534.4 ± 2.7ndnd
mITC46.6 ± 4.567.7 ± 1.950.5 ± 6.0ndnd
lITC51.1 ± 7.075.6 ± 4.452.9 ± 5.8ndnd

The situation is different in the BLA where ß-galactosidase-positive cells represented less than a quarter of all cells. These mostly showed GFAP immunoreactivity and only a few cells were also positive for the neuronal marker NeuN (see Fig. 1E and F for BA images; Table 1 for BLA quantitation). At least some of the NeuN-positive cells were GABAergic (Fig. 1, B3).

In summary, these results show that PN-1 is strongly and widely expressed by GABAergic neurons in the CEA, less strongly but widely in the ITCs, and sparsely by neurons of the BLA. Therefore, the major source of PN-1 expression in the BLA is of glial origin, while in the CEA and ITCs it has a strong neuronal component.

Impaired fear extinction in PN-1 KO mice

To examine the acquisition and extinction of conditioned fear responses in PN-1 KO and WT littermate mice, we used freezing responses elicited by the CS to measure learned fear. During fear conditioning, PN-1 KO mice and their WT littermates displayed similar freezing responses to the US during CS presentations, showing no genotype differences in fear acquisition on Day 1 (data not shown: F1,88 = 0.02034, > 0.05; = 8 WT, 7 KO). To test fear extinction, mice were repeatedly exposed to the CS in two sessions on Days 2 and 3. Results are shown as freezing responses averaged over blocks of four CS presentations each (Fig. 2A and B). Both WT and PN-1 KO mice displayed above baseline freezing responses to the CS tone presentations during the early extinction session (trial effect F4,70 = 11.99, < 0.001; = 8 WT, 7 KO; Fig. 2A). This response decreased significantly by the 4th block of CS presentations for WT but not KO mice (1st vs. 4th CS block: WT, < 0.05; KO, > 0.05). As previously described (Herry & Mons, 2004), mice still exhibited increased freezing over pre-CS baseline values to the CS at the beginning of the late extinction session on Day 3 (trial effect: F4,70 = 19.94, < 0.0001; no tone vs. 1st CS block: WT, < 0.001; KO, < 0.001; Fig. 2B). However, while the WT mice reduced their freezing levels upon repeated exposure to the CS achieving baseline levels during the second extinction session, the PN-1 KO mice continued to exhibit high freezing levels [interaction (trial × genotype) effect: F4,70 = 3.807, = 0.0087; genotype effect: F1,73 = 16.11, = 0.0015; no tone vs. 4th CS block: WT, > 0.05; KO, < 0.001].

Figure 2.

 PN-1 KO mice show impaired fear extinction. Mice were fear conditioned on Day 1, then subjected to extinction training sessions on Days 2 (early extinction) and 3 (late extinction). (A, B) Freezing responses of mice during early (A) and late (B) extinction sessions following five paired CS–US during fear conditioning. Freezing responses to pre-CS are averaged over 4 × 30-s blocks before the first CS, and to the 16 CS presentations over four blocks of four CS each. The two genotypes were significantly different in their freezing responses to the CS presentations during late extinction (= 0.0015; = 8 WT, 7 KO). (C) Freezing responses during late extinction sessions following four paired CS–US during fear conditioning. The two genotypes were significantly different in their freezing responses to the CS presentations during late extinction (= 0.012; = 4 WT, 4 KO). Data are presented as above. (A–C) Bonferroni post hoc tests revealed significant differences between the CS blocks (WT first trial is significantly different from last trial #< 0.05, ##< 0.01, ###< 0.001; CS block significantly different from no tone: °°°P < 0.001. (D) Freezing response of no extinction and extinction group mice during early extinction. Pre-CS is the average freezing response of 4 × 30-s blocks before the first CS; 1st and 2nd CS is the average freezing response to the first two CS presentations, = 27 WT, 27 KO. CS, conditioned stimulus; KO, mouse line lacking in PN-1; US unconditioned stimulus; WT, wild-type littermates of PN-1 KO mice.

These high freezing levels displayed by PN-1 KO mice during the late extinction session indicate that the mice did not learn extinction under conditions their WT littermates did. This phenotype was manifested even with a weaker conditioning protocol of four CS–US pairings [Fig. 2C; late extinction interaction (trial × genotype) effect: F4,35 = 4.533, = 0.0072; genotype effect: F1,38 = 12.63, = 0.0120; no tone vs. 4th CS block: WT, > 0.05; KO, < 0.001; = 4 WT, 4 KO]. In order to determine whether there is a stronger initial freezing response in PN-1 KO mice that might interfere with, or occlude, extinction training, we compared the combined fear retrieval response of all the mice in both the extinction and no extinction groups. We found no significant differences between PN-1 KO and WT mice either in baseline freezing before CS presentation or in the freezing responses to the first two CS presentations of early extinction trials [Fig. 2D; significant trial effect (F1,106 = 314.8, < 0.0001), but no genotype effect (F1,106 = 0.9757), = 27 WT, 27 KO]. Taken together, our results suggest that the impaired extinction phenotype of the PN-1 KO mice is robust and not associated with a significantly stronger early freezing response.

Shifted balance of Fos protein expression in BA of PN-1 KO mice after fear conditioning and extinction

Fos protein induction is generally considered to be a marker of neuronal activation and has been used to map neuronal areas activated during learning (Tischmeyer & Grimm, 1999). In addition, it may be needed for encoding of memory (Tischmeyer & Grimm, 1999). Fos immunoreactivity is increased in the BLA after retrieval of conditioned fear responses and after extinction (Herry & Mons, 2004). The latter increase does not occur in mice resistant to extinction (Herry & Mons, 2004). Consequently, we monitored the level of Fos protein in the amygdala by immunohistological analysis as a possible indicator of an abnormal cellular response associated with the behavioral defect of PN-1 KO mice.

PN-1 KO mice show a stronger Fos induction in the BA after fear retrieval

Control naïve mice had a very low density of Fos-immunoreactive cells in the LA and BA (WT LA: 5.0 ± 2.5 cells/mm2; WT BA: 3.4 ± 1.5 cells/mm2; KO LA: 3.9 ± 1.4 cells/mm2; KO BA: 5.4 ± 2.1 cells/mm2; = 8 WT, 8 KO). Both WT and PN-1 KO mice in the no extinction group showed high freezing responses to the CS presentations on the third day (for behavioral data of the no extinction and extinction groups, see Supporting information, Fig. S1A and B). There was an increase in Fos immunoreactivity in both WT and PN-1 KO mice (Fig. 3A and B). Compared with their WT littermates, we found a significantly higher density of Fos-immunopositive cells specifically in the BA of PN-1 KO mice (genotype effect: F1,20 = 4.542, = 0.0471 and area effect: F1,20 = 24.57, = 0.0001; WT vs. KO in BA: < 0.05; = 5 WT, 6 KO).

Figure 3.

 Altered Fos expression in the BA of PN-1 KO mice following fear retrieval and extinction. Representative images (A, C) and quantification (B, D) of Fos protein immunostaining in the LA and BA of the no extinction group (A, B; n = 5 WT, 6 KO) and extinction group (C, D; = 6 WT, 6 KO) killed 2 h after the start of Day 3 session. Bonferroni post hoc tests: *< 0.05; scale bar in lower (C) 100 μm for (A, C). BA, basal nucleus of the amygdala; KO, mouse line lacking in PN-1; LA, lateral nucleus of the amygdala; WT, wild-type littermates of PN-1 KO mice.

PN-1 KO mice show a weaker Fos induction in the BA after fear extinction

After extinction acquisition, the density of Fos-immunopositive cells was also elevated in LA and BA of both WT and PN-1 KO mice (Fig. 3C and D). However, unlike the situation for the no extinction mice, PN-1 KO mice had a significantly lower Fos-immunopositive cell density in the BA than their WT littermates (genotype effect: F1,22 = 7.418, = 0.0131 and an area effect: F1,22 = 29.23, < 0.0001; WT vs. KO in BA: < 0.05; = 6 WT, 6 KO). The density of Fos cells in PN-1 KO LA was also reduced but did not reach significance. In addition, a comparison of BA no extinction and extinction groups revealed a significant increase in Fos-immunopositive cell density after extinction learning in WT but not PN-1 KO mice [interaction (genotype × treatment) effect: F1,21 = 12.32, = 0.0023; BA WT no ext. vs. ext.: < 0.01; BA KO no ext. vs. ext.: > 0.05; = 11 WT, 12 KO]. Our results indicate that the deficient extinction behavior in PN-1 KO mice is associated with altered neuronal activity in the BA.

Fos expression in the ITCs and CEA did not show behavior- or genotype-dependent changes. The average number of Fos-immunopositive cells in the ITCs was one per field, while the density of immunopositive cells in the CEA varied between 28 and 32 cells/mm2.

Extinction training, but not fear retrieval, induces a shifted pattern of pαCamKII levels in mITC and CEl of PN-1 KO mice

In order to examine longer term changes in synaptic activity and plasticity, we used another marker: αCamKII. Following Ca2+ influx through NMDARs or other calcium sources, αCamKII is activated by binding calmodulin and subsequent autophosphorylation (Fink & Meyer, 2002). As an important player in downstream signaling, it contributes to NMDAR-dependent synaptic plasticity and has been proposed to serve as a molecular switch for memory processes (Fink & Meyer, 2002; Lisman et al., 2002). Local blockade of αCamKII activity in the BLA impairs fear conditioning (Rodrigues et al., 2004), and increased levels of pαCamKII were found at LA synapses 15 min after fear conditioning (Rodrigues et al., 2004). Importantly, αCamKII and pαCamKII are present in the CEA and in the ITCs (McDonald et al., 2002; Royer & Paré, 2002; Rodrigues et al., 2004). We used laser microdissection to isolate defined amygdala nuclei and subdivisions followed by immunoblot analysis to detect discrete patterns of αCamKII phosphorylation. We chose a 2-h time point after the start of the third behavioral session as this should reflect processes downstream from the initial neuronal activation triggered by CS exposure in all behavioral groups.

Decreased mITC αCamKII phosphorylation after extinction training in PN-1 KO mice

Using extracted protein from laser-dissected tissue samples from the different behavioral groups of WT and PN-1 KO littermates (for behavioral data for these experiments, see supporting Fig. S1C and D), we analysed the changes in pαCamKII relative to αCamKII protein levels (pαCamKII/αCamKII), as well as αCamKII levels relative to actin. The results were normalized to WT CS-only control values. There were no significant differences between WT and KO αCamKII protein levels relative to actin in any of our experiments (supporting Fig. S2). We then investigated fear conditioning and extinction-induced changes in the pαCamKII/αCamKII ratio in the mITCs and lITCs (Fig. 4). In the mITC, pαCamKII/αCamKII protein ratios were changed after behavior [interaction (treatment × genotype) effect: F2,20 = 9.763, = 0.0015, and treatment effect: F2,20 = 14.80, = 0.0002; n = 12 WT and 11 KO; Fig. 4A and B]. Specifically, the level of phosphorylation increased in WT no extinction and extinction groups relative to the WT CS-only group (< 0.05 and < 0.01, respectively). The increase for the extinction group was also greater than for the no extinction group (< 0.05). This was in contrast to the situation for PN-1 KO mice. As in the case for the WT, the no extinction group showed a significant increase in phosphorylation level over the PN-1 KO CS-only mice (< 0.01); however, the extinction group did not. The WT extinction group pαCamKII/αCamKII ratios were also significantly greater than for the PN-1 KO extinction group (< 0.01). These results suggest that the mITC cells are responsive to both fear retrieval and extinction acquisition. Similarly, the decreased response in the mITC of PN-1 KO mice correlates with their impaired extinction behavior.

Figure 4.

 pαCamKII levels increase in WT mITC following fear retrieval and extinction, but decrease in PN-1 KO mice following extinction. Representative immunoblots (A, C) and quantification (B, D) of pαCamKII/CamKII ratios in mITC (A, B) and lITC (C, D). Each lane of an image is from a different mouse as indicated. Bonferroni post hoc tests: **< 0.01; compared with CS-only control, °< 0.05, °°< 0.01, °°°< 0.001; WT no ext. vs. ext., #< 0.01; = 12 WT, 11 KO. αCamKII, alpha-calcium/calmodulin protein kinase II; CS, conditioned stimulus; ext., extinction; KO, mouse line lacking in PN-1; lITC, lateral intercalated cell mass; mITC, medial intercalated cell mass; no ext., no extinction; pαCamKII, phosphorylated alpha-calcium/calmodulin protein kinase II; WT, wild-type littermates of PN-1 KO mice.

The analysis of pαCamKII/αCamKII ratios in the lITC (Fig. 4C and D) showed no behavior-dependent changes in either WT or PN-1 KO mice. The overall levels for PN-1 KO groups, however, tended to be lower than for the corresponding WT group (genotype effect: F1,21 = 6.760, = 0.0187; n = 12 WT and 11 KO).

Greater increase in CEl pαCamKII levels after extinction training in PN-1 KO mice

We also examined pαCamKII/αCamKII ratios in two subdivisions of the CEA (Fig. 5). In the CEl, the WT and PN-1 KO extinction groups showed a significant increase in phosphorylation levels over their respective CS-only controls (genotype effect: F1,21 = 12.01, = 0.0030, and treatment effect: F2,20 = 11.52, = 0.0007; n = 12 WT and 11 KO; extinction compared with CS-only group: WT, P < 0.05 and KO, P < 0.01; Fig. 5A and B). The increase shown by the PN-1 KO mice in the extinction group was significantly greater than the corresponding values for the WT extinction group (< 0.05). While there were no significant changes in the no extinction groups compared with CS controls, there was an overall trend to increased phosphorylation levels in PN-1 KO compared with the WT mice. In comparison, analysis of pαCamKII/αCamKII ratios in the CEm (Fig. 5C and D), and in the LA and BA (supporting Fig. S3) showed that neither WT nor PN-1 KO values varied with the behavioral groups. Taken together, our data indicate that extinction triggers the phosphorylation of αCamKII specifically in the mITC and CEl, and that this response is perturbed in the PN-1 KO mouse.

Figure 5.

 pαCamKII levels increase specifically in CEl following extinction and the change is greater for PN-1 KO mice. Representative immunoblots (A, C) and quantification (B, D) of pαCamKII/CamKII ratios in CEl (A, B) and CEm (C, D). Each lane of an image is from a different mouse as indicated. Bonferroni post hoc tests: *< 0.05; compared to CS control, °< 0.05, °°< 0.01; KO no ext. vs. ext., #< 0.05; = 12 WT, 11 KO. αCamKII, alpha-calcium/calmodulin protein kinase II; CEl, latero-capsular subdivision of the central amygdala; CEm, medial subdivision of the central amygdala; CS, conditioned stimulus; ext., extinction; KO, mouse line lacking in PN-1; no ext., no extinction; pαCamKII, phosphorylated alpha-calcium/calmodulin protein kinase II; WT, wild-type littermates of PN-1 KO mice.


Our behavioral results indicate that fear extinction is severely impaired in PN-1 KO mice. This deficit is accompanied by an abnormal pattern of activity-dependent signaling markers across different amygdala nuclei, including the BA, mITC and CEl.

Impaired fear extinction in PN-1 KO mice

The impaired extinction phenotype is unlikely to reflect a founder effect of the PN-1 KO line, as another line of mice with reduced PN-1 protein expression also show impaired extinction (data not shown). It may, however, have a developmental component that we cannot exclude. This impairment also cannot be considered as a general learning deficit of the PN-1 KO mice as their fear conditioning learning is comparable to their WT littermates. In addition, while we found no evidence that they are more susceptible to learning fear, we cannot exclude that the threshold for fear acquisition is lower for PN-1 KO mice.

Our study is the first demonstration as far as we know that a serpin can influence emotional learning such as fear extinction. Earlier reports have shown that serine proteases can influence fear conditioning. Acutely stressed mice lacking the protease tissue plasminogen activator exhibit reduced contextual fear learning compared with WT animals (Norris & Strickland, 2007). On the other hand, mice lacking another activity-dependent serine protease, neuropsin, display increased fear after cued fear conditioning compared with WT littermates, even in the absence of stress (Horii et al., 2008). Mice with a targeted deletion of the serine protease-activated receptor-1 (PAR-1), also known as the thrombin receptor, show reduced fear retrieval after cued fear conditioning (Almonte et al., 2007). PN-1 inhibits many of the above involved proteases and reduces PAR-1 activation (Scott et al., 1985; Stone et al., 1987; Kvajo et al., 2004; Feutz et al., 2008). In addition to a reduced proteolytic inhibition, a further impact of the absence of PN-1 could be an altered cellular signaling triggered by high molecular weight complexes between PN-1 and its target proteins (Vaillant et al., 2007; Fayard et al., 2009). Consequently, our results suggest a possible involvement of serine proteases in fear extinction as well.

Molecular correlates of impaired fear extinction in PN-1 KO mice

We evaluated short- and long-term patterns of neuronal activation in the amygdala by comparing Fos immunoreactivity and pαCamKII protein levels in the amygdala of WT and PN-1 KO mice to find cellular correlates of this behavioral deficit. We concentrated on the amygdala because of the striking pattern of PN-1 expression in GABAergic neurons as well as its central role in integrating fear inputs. It is possible that other affected brain areas contribute to the overall extinction deficit in the PN-1 KO mouse, e.g. the prefrontal cortex (Quirk & Mueller, 2008) or the hippocampus (Corcoran et al., 2005).

In WT mice, Fos immunoreactivity increased in the no extinction and extinction groups as expected in the LA and BA after fear retrieval and extinction acquisition, compared with the naive control group (Herry & Mons, 2004). The Fos-immunopositive cells possibly represent subsets of the two populations of cells recently shown to be activated differentially by fear and extinction protocols (Herry et al., 2008).

This response was shifted in PN-1 KO mice, namely the increase was higher than the WT response after fear retrieval in the no extinction group and lower than the WT in the extinction group. This shift could be the result of impaired sensory or higher brain input or as well as of local impairments resulting from the absence of PN-1 signaling in the BA. Although PN-1 is not prominently expressed by BA principal neurons, our immunohistochemical results indicate its presence in the extracellular matrix, presumably through glial secretion. Application of purified PN-1 has been shown to rescue primary cultured cerebellar granular neuron precursors derived from PN-1 KO mice, suggesting that extracellular sources of PN-1 can participate (at least in some measure) in normal neuronal signaling (Vaillant et al., 2007).

Surprisingly, PN-1 KO mice displayed a greater Fos protein expression under conditions where we would expect reduced NMDAR activity. One possible explanation for the apparently paradoxical finding is a lowered basic inhibitory activity in the BLA. Inhibitory GABAergic interneurons in the BLA exhibit NMDAR-mediated synaptic currents (Szinyei et al., 2000) and provide a strong inhibitory control over principal neurons (Lang & Paré, 1997). Reduced levels of NMDAR activity on inhibitory neurons could therefore have a proportionately greater impact on the net level of BLA activity. Concurrently, the net strength or balance of various inputs (e.g. cortical and hippocampal) to the amygdala could be affected, thereby changing the activation outcome. This altered Fos upregulation measured after fear retrieval may be an indication that the net levels of activity in the BA are abnormal in PN-1 KO mice. In fact, some of these neurons expressing cFos after fear conditioning may not be directly involved with fear expression but contribute to resistance to extinction similar to what has been described in the prelimbic cortex (Burgos-Robles et al., 2009).

No change in Fos immunoreactivity was detected in the CEA. This is unlike previous studies showing an increase in the CEA after extinction (Hefner et al., 2008; Kolber et al., 2008). One reason may be that these studies used a fear conditioning protocol with a stronger and longer foot shock US than ours.

To evaluate longer term neuronal activation, we measured the relative phosphorylation level of αCamKII by immunoblot analysis of laser-dissected amygdala subnuclei. Long-lasting increased levels of autophosphorylated αCamKII in specific brain areas have been associated with learning (Pollak et al., 2005; Singh et al., 2005). In addition, normal autophosphorylation of αCamKII has been reported to be essential for learning extinction of conditioned contextual fear (Kimura et al., 2008). We found no fear conditioning- or extinction-dependent changes in relative pαCamKII levels in the LA, BA, CEm or lITC. This may reflect an averaged sampling of heterogeneous neuronal populations. A trend of a lower pαCamKII/αCamKII ratio was, however, detected in the lITC of PN-1 KO mice.

In WT mice, behavior-dependent increases in pαCamKII levels were found in the mITC after fear retrieval and extinction, and in the CEl after extinction, suggesting that these behaviors induce distinct activity-dependent changes in cellular or synaptic function in these areas. The mITC receives excitatory input from the BA as well as other regions (Royer et al., 2000). The pattern of pαCamKII levels in the mITC correlates with the relative levels of Fos activation of the BA after fear retrieval and extinction. Moreover, as BA cells are functionally heterogeneous with distinct subpopulations active after fear conditioning and extinction (Herry et al., 2008), it is tempting to speculate that mITC neurons might exhibit a similar heterogeneity, and that the mITC might not only be involved in fear extinction (Jüngling et al., 2008; Likhtik et al., 2008) but also in the regulation of high fear states (Paréet al., 2004). In the rat brain, the CEl receives inputs from the cortex, BA and LA (Cassell et al., 1999). Therefore, the increased phosphorylation of αCamKII we detected in the WT CEl after extinction would be consistent with a sufficiently increased input from the BA as indicated by the increased density of Fos-immunopositive cells.

In contrast, PN1-KO mice exhibited a shift in the distribution of pαCamKII after extinction training relative to WT animals. The absence of a further increase over fear retrieval levels of phosphorylation in the mITC correlates with the unchanged Fos induction in the BA and is consistent with the behavioral readout of high freezing levels in PN-1 KO mice after the extinction training. The increased pαCamKII levels in the CEl of KO mice after extinction training could be explained by a reduced inhibitory input from the mITC, implied by the below WT phosphorylation level. This may serve to offset a decreased BA input, implied by the relatively low Fos immunoreactivity, leading to a net increased activation of the CEl. Indeed, connections between mITC and CEl have been described in the cat (Paré & Smith, 1993), and extracellular stimulation within the mITC was reported to activate synapses on the dendrites of CEl neurons in the rat (Delaney & Sah, 2001). Another consideration is that increased pαCamKII levels in the CEl of PN-1 KO mice might reflect activation of functionally distinct, fear-promoting subpopulations of neurons that are normally not active during extinction training.

Our study shows the usefulness of laser dissection to monitor changes in protein phosphorylation in small, specific regions of the brain and correlate them to learning. We show that WT mice, acquiring extinction with the associated reduced freezing response and increased Fos protein expression in BLA, also display corresponding increases in pαCamKII levels in mITC and CEl. PN-1 KO mice, which we show are capable of acquiring conditioned fear responses but are resistant to acquiring extinction, show impairments in these responses. Our results do not allow us to distinguish if altered NMDAR activity, denoted by the abnormal levels of activity indicators in the BA, CEl and mITC, is responsible for the behavioral impairment or is the consequence thereof. Nevertheless, these data are in line with recent studies demonstrating a link between impaired extinction learning and altered immediate-early gene expression patterns in the BA, mITC and CEA in select mouse and rat strains with inborn behavioral deficits (Hefner et al., 2008; Muigg et al., 2009) and with the recovery of conditioned fear responses in extinguished animals after attenuation of glutamatergic input to mITCs or targeted immunotoxic lesions of mITCs (Jüngling et al., 2008; Likhtik et al., 2008).

In summary, our results support a growing view that the emotional learning and memory system is not limited to the BLA but is distributed across BLA, ITC and CEA circuitry of the amygdala (Paréet al., 2004; Wilensky et al., 2006). Moreover, our study demonstrates that lack of PN-1 results in area-specific changes in signaling activity markers underlying fear extinction. Serine proteases and their inhibitors may thus represent new targets for intervention in various conditions associated with anxiety and stress.


This work was supported by grants from the Swiss National Science Foundation, the Austrian Science Fund, the Volkswagen Stiftung and by the Novartis Research Foundation. We thank Sabrina Djaffer for expert animal breeding and care, Erik Cabuy for advice about laser dissection, Sandrine Bichet for help with immunohistochemistry, Laurent Gelman for patient help with microscopy, and Cédric Fischer for help at the start of this work. We also thank Andrew Matus and Thomas Oertner for critical reading of the manuscript.


basal nucleus of the amygdala


basolateral amygdaloid complex


central nucleus of the amygdala


latero-capsular subdivision of the central amygdala


medial subdivision of the central amygdala


conditioned stimulus




γ-aminobutyric acid containing inhibitory neurons


glutamic acid dehydrogenase 67-kD isoform


glial fibrillary acidic protein


intercalated cell masses


lateral nucleus of the amygdala


lateral intercalated cell mass


medial intercalated cell mass


N-methyl-d-aspartate receptor

no ext.

no extinction


protease-activated receptor-1


mouse line lacking in protease nexin-1


protease nexin-1


phosphorylated alpha-calcium/calmodulin protein kinase II


unconditioned stimulus


wild-type littermates of PN-1 KO mice


alpha-calcium/calmodulin protein kinase II