Contribution of neurotrypsin to synaptic plasticity in juvenile mice
First, we investigated the importance of NT-dependent spinogenesis for functional long-term synaptic plasticity underlying learning and memory. We used a “spaced” protocol for the induction of LTP and applied two 1-h spaced TBSs (Kramár et al., 2012). The first TBS was supposed to promote filopodia generation, while the second TBS may convert silent synapses into functional synapses by recruitment of AMPA receptors. As NT is expressed in the hippocampus, we tested spaced LTP in NT+/+ and NT-/- juvenile mice at CA3-CA1 synapses (Figure 1).
We found that after induction of LTP by one TBS, the levels of potentiation were similar in both genotypes (p genotype = 0.383, F (1,19) = 0.798; p genotype x time = 1.000, F (717, 13623) = 0.610; Figure 1C, D). Application of the second TBS (TBS2) further increased the level of LTP in NT+/+ but not in NT-/- mice (p genotype = 0.010, F (1,17) = 8.280; p genotype x time <0.001, F (537,9129) = 2.902; Fig.1C, D). Overall, these results indicated that mice lacking NT showed no additional potentiation after TBS2, unlike control mice. These observations suggested that a new population of synapses induced by the NT-agrin signaling pathway by TBS1 could be potentiated by TBS2.
Role of neurotrypsin in contextual fear conditioning in juvenile mice
Previous experiments have implicated NT in hippocampal plasticity (Matsumoto-Miyai et al., 2009). To match the spaced LTP protocol at the behavioral level, we designed a protocol of spaced contextual fear conditioning (CFC) and extinction in which 6 foot shocks were applied in the conditioned context (CC+), divided into two learning sessions (3x + 3x) with a 1 h-interval delay between both sessions (Figure 2A). To evaluate fear memory, we measured the freezing response (total freezing time divided to trial duration) in the conditioned context (CC+) and in the neutral context (CC-) on day 2 (recall d2). To examine memory extinction, we analyzed freezing levels at day 9 (recall d9) after 9x extinction sessions in CC+ to erase conditioned fear. Monitoring the freezing response in the CC- allowed us to measure the mice's ability to differentiate between both contexts.
The results indicated that NT deficiency did not affect the level of spontaneous freezing before CFC at day 0 (training) in either the CC- or CC+ (p genotype = 0.405, F (1,21) = 0.720; p genotype x context = 0.417, F (1,20) = 0.688; Figure 2C). This freezing level shown by mice before CFC was typical for the exploration of novel environments in mice. Moreover, we assessed anxiety levels and general locomotor activity in the open field test. The results showed that NT deficiency altered neither general locomotor activity (p = 0.540; NT-/-: 14.4 ± 1.1 m vs. NT+/+: 15.3 ± 1.0 m; Figure S1D) nor anxiety levels (p = 0.479; NT-/-: 34.2 ± 3.9 % vs. NT+/+: 30.6 ± 3.0 %; Figure S1A). Additionally, the levels of freezing immediately after unconditioned stimuli were not different between genotypes (p genotype = 0.359, F (1,21) = 0.880, p genotype x stimulus =0.250, F (5,105) = 1.347; Figure 2D).
Analysis of memory recall revealed that both groups of mice were able to distinguish between the CC+ and CC- contexts on d2 (two-way repeated measures ANOVA: p context < 0.0001, F (1,21) = 92.929; p genotype = 0.651, F (1,21) = 0.210; p time x genotype = 0.031, F (1,21) = 5.323; p < 0.001 for both NT+/+ and NT-/- mice, respectively, Holm-Sidak post hoc test; Figure 2F left). During extinction sessions from day 5 to day 7, the freezing time in CC+ of both group gradually reduced (p session <0.001, F (8,168) = 7.13), but there was not statistical difference detected between genotypes (p genotype = 0.469, F (1,21) = 0.545; p genotype x session = 0.245, F (8,168) = 1.303; Figure 2E). The fear response reduced to similar levels in all genotypes (p genotype = 0.416, F (1,21) = 0.688) and contexts (p context = 0.312, F (1,21) = 1.074) after the fear extinction protocol at recall on d9 (Figure 2F right). However, NT-/- mice appeared to be less efficient in discriminating between contexts during fear memory retrieval on day 2 (p = 0.026, Holm-Sidak post hoc test; Figure 2G). To test whether impaired context discrimination was due to impaired visual function in NT-/- mice, we subjected these mice to the novel object recognition test (NORT) that critically depends on visual discrimination. As NT-/- mice stayed longer near novel rather than familiar objects in NORT (73.6 ± 13.6 s vs. 39.9± 12.5 s, p = 0.0073; Figure S1G), similar to NT+/+ mice (65.7 ± 8.6 s vs. 38.8± 6.8 s, p = 0.00007; Figure S1G left), it appeared that they have normal visual function and form object recognition memory normally (discrimination ratio: NT-/-: 42.1 ± 7.8 % vs. NT+/+: 42.8 ± 4.6 %, p = 0.937; Figure S1H).
In these experiments, we used a strong CFC protocol, which resulted in only a mild difference between genotypes in the discrimination between contexts. To clarify whether this was due to a ceiling/saturation of CFC in both genotypes, we designed a milder CFC protocol with 3x foot shocks and single-context testing (Figure 2H). This second CFC protocol was designed without spaced learning sessions; however, several studies suggest that sleep plays an active role in the replay of information and memory consolidation (Dudai et al., 2015; Graves et al., 2003; Rauchs et al., 2011). In this sense, acquisition followed by consolidation can be viewed as spaced stimulation. Importantly, the freezing time in the conditioned context, CC+, was significantly less 24 h after conditioning in the NT-/- mice (31.9 ± 3.2 % vs. 50.5 ± 4.7 %; p = 0.005; Figure 2I), suggesting that NT deficiency impairs the formation and/or retrieval of contextual fear memory. As in spaced CFC, both genotypes similarly decreased their freezing levels after the fear extinction protocol, indicating that extinction of contextual fear memory was not altered by NT deficiency (Figure 2I).
Neurotrypsin is important for social interaction in juvenile mice
Next, we investigated social behavior in NT-/- mice. Alterations in social behavior are symptoms of several neuropsychiatric and neurological diseases. In particular, mental retardation is generally accompanied by a functional deficit in adaptive behavior, such as social skills and communication (Bieleck & Swender, 2004).
To evaluate sociability, we performed a three-chamber sociability test (Figure 3A). We observed that NT-/- mice spent more time near box with stimulus mouse inside (74.9 ± 10.8 s vs. 40.2 ± 5.7 s, p = 0.017; Figure 3B) compare to their control NT+/+ littermates (153.9 ± 28.9 s vs. 34.9 ± 4.9 s, p = 0.0029; Figure 3B). But the NT-/- mice showed less preference to the stimulus mouse. This result was confirmed by the analysis of the discrimination ratio to remove individual variability in total exploration time (NT-/-: 25.4 ± 9.6 % vs. NT+/+: 55.5 ± 7.3 %, p = 0.0267, t-test; Figure 3C).
Age-persistent and new defects in mature neurotrypsin-deficient mice
Next, we asked if impaired CFC and sociability in juvenile NT-/- mice would persist after animals matured and aged and if additional defects could appear. Hence, we evaluated the behavioral phenotype of 1- to 2-year-old age-matched NT-/- and NT+/+ mice in a battery of behavioral tests (Figure S2A).
In the open field test, the total distance traveled by NT-/- mice was not significantly different compared with NT+/+ littermates (p = 0.355, 32.3 ± 1.6 m vs. 36.4 ± 3.8 m; Figure S2B). There was also no difference between genotypes in cumulative time spent in the central (p = 0.285, NT-/-: 111.2 ± 14.6 s; NT+/+: 134.7 ± 15.5 s) and peripheral areas (p = 0.284, NT-/-: 488.9 ± 14.6 s; NT+/+: 465.3 ± 15.5 s) (Figure S2C, D). Due to a slight trend for NT-/- mice to spend less time in the central area, the elevated plus-maze was additionally used to verify whether these mice were more anxious. However, NT+/+ andNT-/- littermates spent similar amounts of time in both open arms (p = 0.827, 112.4 ± 13.8 vs. 108.5 ± 11.7 s) and enclosed arms (p = 0.855, 419.9 ± 16.0 s vs. 423.7 ± 12.7 s). There was no significant difference in discrimination ratio for the time spent in the arms (p = 0.820; NT-/-: 57.6 ± 5.3%; NT+/+: 59.2 ± 4.5%; Figure S2E-G). These results suggest that NT deficiency influences neither locomotor activity nor anxiety status.
Next, we performed a series of cognitive tests in which animals with a normal ability to memorize objects/animals during the encoding phase should spend more time exploring a new object/animal or familiar objects in a new location during the test phase in the novel object recognition test or novel object location test, respectively. In the novel object recognition test, NT+/+ animals indeed spent more time exploring a novel object than a familiar one (p = 0.0003; 31.5 ± 2.5 s vs. 14.8 ± 1.5 s), whereas NT-/- mice spent similar amounts of time exploring novel and familiar objects (p = 0.460; 23.6 ± 2.2 vs. 20.4 ± 2.5 s; Figure S3A). There was a significant difference in the discrimination ratio between genotypes (p = 0.024; NT-/-: 8.0% ± 9.1%; NT+/+: 35.1% ± 6.6%; Figure S3E). In the novel object location task, NT+/+ mice showed a tendency to explore an object with a changed spatial position during the retrieval phase (p = 0.083; 23.8 ± 2.9 vs. 18.2 ± 2.8 s), while NT-/- mice spent the same time exploring both objects (p = 0.474; 23.9 ± 1.0 s vs. 26.3 ± 2.7 s; Figure S3B). The difference in discrimination ratio between genotypes was close to statistical significance (p = 0.068, Mann-Whitney test; Figure S3F).
To test the ability of NT-/- mice to memorize the temporal sequence of events, we performed a temporal order recognition task. However, both NT-/- and NT+/+ mice at this age failed to spend more time exploring the object less recently shown than they spent exploring the other object (p=0.346 and 0.799, respectively; Figure S3C). There was no difference in the discrimination ratio between genotypes (p = 0.278; NT-/-: -15.6% ± 8.6%, NT+/+: -2.0% ± 8.5%; Figure S3G). To test the capacity of NT-/-miceto recognize and memorize other mice, we performed the social recognition test. Here, mice of both genotypes exhibited some preference for novel mice (NT-/-: 12.8% ± 10.0%, NT+/+: 18.9% ± 7.7%; Figure S3D). The exploration time that NT+/+ animals spent around novel animals tended to be higher than the time spent around familiar animals (p = 0.053; 53.4 ± 6.2 s vs. 36.1 ± 4.7 s; Figure S3D, H)., Thus, aged NT-/- mice showed at least mild impairment in all recognition memory tasks compared to age-matched NT+/+ mice, indicating deficits in memory formation.
In line with data obtained in juvenile mice, aged NT-/- mice failed in the sociability test, showing an almost equal level of interest in a “stimulus” mouse and control objects (p = 0.705; 46.6 ± 5.8 vs. 43.3 ± 4.4 s), whereas NT+/+ mice spent more time engaging in social communication than object exploration (p = 0.0027; 53.5 ± 5.8 vs. 43.3 ± 4.4 s; Figure 4A). The discrimination ratio in this test differs significantly between genotypes (p = 0.015; KO: -1.9% ± 8.9% vs. NT+/+: 25.9% ± 5.6%; Figure 4B).
On day 0 of CFC, both NT-/- and NT+/+ littermates exhibited freezing less than 3% of the total time in the CC+ and CC- before foot shock (Figure 4C). Two-way repeated measures ANOVA revealed a significant increase in freezing time in CC+ immediately after foot shocks (p context < 0.001, F (1,19) = 32.728; Figure 4D) with no difference between genotypes (p genotype= 0.899, F (1,19) = 0.017; p context x genotype= 0.530, F (1,19) = 0.409; Figure 4D). Thus, NT-/- mice perceived the unconditioned stimulus and had the same ability as NT+/+ mice to form and express fear memory. On day 1, fear memory was evaluated in a recall session. As shown in Figure 4F, animals spent more time freezing in the CC+ than in the CC- (p context < 0.001, F (1,19) = 69.398), as revealed by the Holm-Sidak post hoc test within the NT-/- (p < 0.001, 47.4% ± 4.9% vs. 21.3% ± 2.6%) and NT+/+ groups (p < 0.001, 44.3% ± 3.1% vs. 21.5% ± 1.8%). The discrimination ratios were not different between genotypes (p = 0.701, NT-/-: 37.3% ± 5.7%; NT+/+: 34.8% ± 3.3%; Figure 4H middle). Thus, after maturation, NT-/- and NT+/+ mice have a similar ability to discriminate contexts and recall fear memory.
From day 2 to day 4, animals experienced 9 sessions (3x each day) in the CC+ to induce contextual fear memory extinction. Two-way repeated measures ANOVA revealed a statistically significant interaction between genotype and extinction sessions (p test phase x genotype = 0.0072, F (8,152) = 2.754; Figure 4E); post hoc analysis indicated a significant difference between NT-/- and NT+/+ mice in session 8 (p = 0.014, 30.7% ± 4.1% vs. 19.3% ± 3.5%) and session 9 (p = 0.014, 29.0% ± 3.9% s vs. 17.7 ± 2.8%). Thus, NT-/- mice failed to exhibit contextual fear memory extinction. On day 5, another recall test was done to evaluate animals’ performance in both the CC- and CC+. Two-way repeated measures ANOVA revealed a statistically significant difference in freezing time between genotypes (p genotype = 0.005, F (1,19) = 9.853; Figure 4G). While NT+/+ mice spent an almost equal amount of time freezing in the CC+ and CC- (p = 0.948, 16.4% ± 2.2% vs. 16.2% ± 2.2%), NT-/- mice still spent more time freezing in the CC+ than in the CC- (p = 0.030, 26.9% ± 2.1% vs. 21.1% ± 1.9%). Due to the large variance, the difference in the discrimination ratio did not reach statistical significance (p = 0.184, 12.6%± 5.0% vs. -4.5% ± 10.9%; Figure 4I). Nevertheless, considering all differences between genotypes on days 2-5, we conclude that NT-/- mice (unlike NT+/+ mice) failed to exhibit contextual fear memory extinction.
Neurotrypsin regulates spine density in juvenile mice in vivo
As LTP-dependent formation of filopodia is abolished in mice lacking NT (Matsumoto-Miyai et al., 2009) and dendritic filopodia are thought to be direct precursors of new dendritic spines (Jontes & Smith, 2000; Yuste & Bonhoeffer, 2004; Ziv & Smith, 1996), we addressed the question of how NT affects spinogenesis and spine morphology in naïve conditions and upon learning. For this purpose, we crossbred NT mice with Thy1-EGFP mice, then analyzed spine density and morphology in CA1 secondary apical dendrites (Figure 5A) in naïve conditions 24 h after contextual fear conditioning and 24 h after fear memory extinction.
The results revealed striking differences between the two genotypes. NT-/- mice showed significantly reduced spine density in naïve and extinction conditions compared with their control NT+/+ littermates (two-way ANOVA: p condition =0.621, F (2, 241) = 0.478; p genotype = 0.0076, F (1, 241) = 7.252; p condition x genotype = 0.0111, F (2, 241) = 4.590; p naive = 0.0010, p extinction =0.0422, Holm-Sidak post hoc test; Figure 5B, subpanel 1). Cumulative frequency curves showed that the spine density distribution shifted towards lower values in naïve NT-/- mice (KS-test: p = 0.0149, Figure 5B, subpanel 2). However, no difference in cumulative frequency curves was found between genotypes after acquisition or extinction of fear conditioning (Figure 5C, Figure 5B, subpanels 3,4).
Interestingly, morphological analysis determined that the percentage of thin/filopodia-like spines was significantly reduced in NT-/- mice (two-way ANOVA: p condition = 0.101, F (2, 241) = 2.318; p genotype < 0.001, F (1, 241) = 28.661; p condition x genotype = 0.370, F (2, 241) = 0.998; p naive = 0.011; p CFC = 0.013, p extinction < 0.001, Holm-Sidak post hoc test; Figure 5C, subplanel 4), whereas the proportion of mushroom spines was higher in this genotype than in wild-type mice (two-way ANOVA: p condition = 0.386, F (2, 241) = 0.957; p genotype < 0.001, F (1, 241) = 48.813, p condition x genotype = 0.257, F (2, 241) = 1.365; p naive = 0.005; p CFC < 0.001, p extinction < 0.001, Holm-Sidak post hoc test for comparisons shown in Figure 5C, subpanel 3). In agreement with these observations, we found a statistically significant reduction in spine head size in NT+/+ mice (two-way ANOVA: p condition = 0.0099, F (2, 241) = 4.708; p genotype < 0.001, F (1, 241) = 26.921; p condition x genotype = 0.005, F (2, 241) = 5.402; p naive < 0.001, p extinction <0.001, Holm-Sidak post hoc test). However, this reduction was not present after contextual fear conditioning (Figure 5C, subpanel 1). Regarding stubby spines, no significant differences were found between genotypes and conditions (Figure 5C, subpanel 2).
Taken together, it appeared surprising that the spine head diameter was not smaller in the control group after CFC, as in all conditions (also in CFC), the percentage of thin spines was higher in the control group and the percentage of mushroom spines was larger in the mutant group. Initially, we speculated that this could be due to mature mushroom spines in the control group being larger and more mature after CFC, thus compensating for this difference. However, we observed that the mushroom spine head size was very similar in both groups of mice (p =0.286, 0.551 (0.520, 0.613) vs. 0.532 (0.518, 0.573); as median (quartile 1, quartile 3), Figure S4A, C). Consequently, we measured the head diameter in the thin spines after CFC. Interestingly, we observed that the thin spines were larger and most likely more mature in NT+/+ mice (p = 0.0003, 0.2849 ± 0.006 vs. 0.2579 ± 0.004; Figure S4B), suggesting that NT deficiency may specifically affect the maturation of thin/filopodia-like spines. In agreement with this, a cumulative frequency plot of spine head diameter of thin spines revealed that NT-/- mice had a leftward shift in the cumulative curve, indicating a reduction in the head diameter for this spine type in the mutant group of mice (Figure S4D).
Finally, we studied the learning-induced changes in spatial distribution of spines following CFC and extinction. We found that spine density linearly decreases over the length of traced dendritic branches (Figure 5D). In NT+/+ mice this non-uniformity of spine distribution was mostly visible in naïve mice, abolished after CFC session and partially restored after extinction sessions (two-way ANOVA: p session x distance = 4.91*10-9, F (2, 692) = 19.671; Fig 5D, subpanel 4). For NT-/- mice we also observed a negative correlation between spine density and spine distance to the beginning of the traced dendritic branches. Nevertheless, it was less prominent and NT-/- mice did not show any learning-induced modulation in the spatial distribution of spines (two-way ANOVA: p session x distance = 0.146, F (2, 709) = 1.93; Fig 5D, subpanel 4).
Neurotrypsin-dependent cleavage of agrin plays a major role in regulating dendritic spine formation and clustering of synapses
As agrin is the only substrate of NT identified so far and its 22 kDa cleavage fragment is critically important for activity-dependent filopodia formation (Matsumoto-Miyai et al., 2009), we addressed the question of whether agrin cleavage is responsible for the putative effects of NT deficiency. To test this hypothesis, we aimed to deliver agrin-22 in the hippocampus of NT-/- mice and evaluate its effect on dendritic spines. For this purpose, we designed an AAV expressing agrin-22 construct specifically in neurons. The DNA construct also included a secretion signal sequence (Aricescu et al., 2006) and the red fluorescent protein scarlet as a reporter. As a control, we used a shorter version of agrin-22, agrin-15, which was shown to act as an agrin antagonist in hippocampal and cortical cultures (Hoover et al., 2003) and in acute-slice preparations (Hilgenberg et al., 2006) (Figure S5).
We injected either pAAV-Syn-Agrin22-Scarlet (AAV-Ag22) or pAAV-Syn- Agrin15-Scarlet (AAV-Ag15) at postnatal day 7 (P7) into the hippocampus of NT-/- mice. Subsequently, we collected samples for spine imaging at P24 after 3 consecutive days of habituation (see Figure 6A for a general scheme of the experiment) to follow the same protocol as that used in previous spine imaging experiments. As shown in Figure 6B, injection of AAV-Ag22 increased the spine density in NT-/- mice to similar levels as those previously observed in their control wild-type littermates (NT-/- + Ag22:1.539 ± 0.050 vs. NT+/+: 1.647 ± 0.051). However, there was no increase in spine density when mice were injected with AAV-Ag15 (1.299 ± 0.043). The difference between mice treated with AAV-Ag22 and AAV-Ag15 was significant (p = 0.001).
In addition to its effect on spine density, injection of AAV-Ag22 in NT-/- mice showed larger spine head size compared with AAV-Ag15 treated group (p = 0.030; NT-/- + Ag22: 0.478 (0.441, 0.515) vs. NT-/- + Ag15: 0.432 (0.412, 0.480) µm; as median (quartile 1, quartile 3), Figure 6C). Moreover, filopodia density was apparently slightly increased in mice injected with AAV-Ag22 compared with those injected with AAV-Ag15 (0.577 ± 0.08 vs. 0.437 ± 0.05), but did not reach statistical significance (p = 0.275, Figure 6D). Additionally, we performed analysis of the spatial density of spines, however, no difference between agrin-15 and agrin-22 injected mice was detected (two-way ANOVA: p agrin x distance = 0.408, F (2, 506) = 0.898).
To ensure proper injections, each hippocampal slice was imaged to detect Ag22-scarlet or Ag15-scarlet expression, and only those animals with positive expression were selected for subsequent spine analysis. Both Ag22-scarlet and Ag15-scarlet labeling were diffusely distributed over neuronal somas from the stratum pyramidale and showed puncta distribution in the stratum radiatum. Distinctly, all mice injected with AAV-Ag15 exhibited a lower intensity of “agrin puncta” in the stratum radiatum, presumably because of less stability, as agrin-15 is not found under physiological conditions in animals (Figure S6).
To determine whether Ag22-scarlet bound its potential neuronal receptor a3NKA (a3 Na+/K+ ATPase) (Hilgenberg et al., 2006), CA1 slices from brains injected with AAV-Ag22 were labeled with the a3NKA monoclonal antibody. Consistent with a previous in vitrostudy (Hilgenberg et al., 2006), we observed an extensive overlap between a3NKA and “agrin puncta” expression in the stratum radiatum (Figure S7A). To confirm that AAV-Ag22 was properly delivered and expressed at synaptic sites, we stained CA1 slices from brains injected with AAV-Ag22 with the excitatory presynaptic marker VGLUT1. Colocalization of scarlet-tagged Ag22 and immunostained VGLUT1 confirmed that agrin-22 was concentrated at synapses (Figure S7B). Altogether, these observations provide strong evidence that virally expressed agrin-22 is properly delivered to synaptic sites in the stratum radiatum and overlaps with its physiological neuronal receptor a3NKA.
Next, we investigated the effect of Ag22-scarlet on VGLUT1 puncta. Interestingly, we observed that VGLUT1-positive presynapses colocalizing with Ag22-scarlet were significantly larger than those without Ag22 scarlet colocalization (0.512 ± 0.027 mm2 vs. 0.120 ± 0.019 mm2; p < 0.0001; Figure 7A-C). In agreement with this, a cumulative frequency plot of VGLUT1-immunopositive puncta area revealed that the presynapses colocalizing with Ag22-scarlet had a rightward shift in the cumulative curve, indicating an enlargement of VLGUT1 presynapses colocalizing with Ag22-scarlet (Figure 7D). This striking observation suggests that Ag22-scarlet may induce synapse formation or aggregation of VGLUT1-positive synaptic vesicles at presynapses.
Finally, to examine whether these large structures are composed of pre- but also postsynaptic specializations, we stained CA1 slices from brains injected with AAV-Ag22 with the excitatory presynaptic marker VGLUT1 and the excitatory postsynaptic marker PSD95. Interestingly, we observed complex synapses with multiple postsynaptic densities contacting agrin-22 clusters (see the magnification in the upper-right corner of Figure S8).