Translational control of auditory imprinting and structural plasticity by eIF2α.

The formation of imprinted memories during a critical period is crucial for vital behaviors, including filial attachment. Yet, little is known about the underlying molecular mechanisms. Using a combination of behavior, pharmacology, in vivo surface sensing of translation (SUnSET) and DiOlistic labeling we found that, translational control by the eukaryotic translation initiation factor 2 alpha (eIF2α) bidirectionally regulates auditory but not visual imprinting and related changes in structural plasticity in chickens. Increasing phosphorylation of eIF2α (p-eIF2α) reduces translation rates and spine plasticity, and selectively impairs auditory imprinting. By contrast, inhibition of an eIF2α kinase or blocking the translational program controlled by p-eIF2α enhances auditory imprinting. Importantly, these manipulations are able to reopen the critical period. Thus, we have identified a translational control mechanism that selectively underlies auditory imprinting. Restoring translational control of eIF2α holds the promise to rejuvenate adult brain plasticity and restore learning and memory in a variety of cognitive disorders.

Introduction

Imprinting is a form of early learning where exposure to a stimulus becomes the triggering signal of a vital behavior (Jin et al., 2016; Horn, 2004). A particular feature of imprinting is that it occurs exclusively within a short critical period (CP) (Jin et al., 2016; Bolhuis, 1991; Nevitt et al., 1994), when structural and functional changes take place (Hensch, 2004). Imprinting drives a vigorous following behavior in chickens, key for filial attachment (Horn, 2004; Insel and Young, 2001). This rather unique and precocious behavior is advantageous for investigating experience-driven activation of molecular pathways around birth (Bredenkötter and Braun, 1997; Bock and Braun, 1999; McCabe et al., 1982, 1981). Unders tanding the biological basis of imprinting can shed light on the mechanisms of learning in newborns and create new avenues to rejuvenate adult brain plasticity by reopening CPs.

The formation of imprinted memories has been described across sensory modalities (Nevitt et al., 1994; McCabe et al., 1982; Remy and Hobert, 2005; Bock et al., 1997). Interestingly, in chickens, auditory and visual imprinting relies on different brain structures. Imprinted sounds activate the mediorostral nidopallium/mesopallium (MNM) (Bock and Braun, 1999; Wallhäusser and Scheich, 1987), where neural responsiveness increases after training (Bredenkötter and Braun, 1997). In contrast, the intermediate medial mesopallium (IMM, former IMHV) (Horn, 2004; McCabe et al., 1982) is required for visual imprinting, where neural responses shift to favor the imprinted object (Horn et al., 2001). While the brain circuits and neurophysiological changes have been uncovered (Horn, 2004; Scheich, 1987), much less is known about the molecular machinery linking experience and the formation of imprinted memories in each sensory modality.

While imprinting requires protein synthesis (Gibbs and Lecanuet, 1981), little is known about the underlying translational control mechanisms. The translation of mRNA into protein occurs in three steps: initiation, elongation and termination and can be regulated through several signaling pathways (Sonenberg and Hinnebusch, 2009). Translation initiation is believed to be the rate-limiting step and a key target for translational control (Sonenberg and Hinnebusch, 2009; Buffington et al., 2014). A major way in which translation initiation is regulated is by modulating the formation of the ternary complex via phosphorylation of the translation-initiation factor eIF2α. In rodents, protein synthesis controlled by phosphorylation of eIF2α is critically required for long-lasting forms of synaptic plasticity (Costa-Mattioli et al., 2007; Di Prisco et al., 2014) as well as long-term memory storage in several systems (Costa-Mattioli et al., 2007, 2005; Zhu et al., 2011; Stern et al., 2013; Ounallah-Saad et al., 2014; Ma et al., 2013). Here we asked whether this central translational control mechanism plays a role in imprinting in newborn chickens and can be used to restore imprinting outside of the CP.

Results

Critical periods for visual and auditory imprinting

Dark reared chickens were placed in a running wheel in front of an LCD screen and a speaker for training. Visual and auditory imprinting were tested separately 24 hr after training (Figure 1a). Stimuli consisted of animated movies showing a virtual object, and artificial sounds synchronized to movements of the object in the screen (Figure 1b, see supplementary materials). The imprinting was assessed by the preferential approach to the imprinted stimuli, either visual or auditory, compared to the approach to novel stimuli. The preference for imprinted stimuli is commonly used as an index of long-term memory storage (Horn, 2004). Individuals' preference was measured by calculating an index, where positive and negative values indicate preference for the imprinted or novel stimulus, respectively (Figure 1c and d). This index accounts for fluctuations in baseline locomotion across trials, as described in the Method section. Consistent with previous studies (Yamaguchi et al., 2012), chickens showed imprinting to either visual or auditory cues one day after hatching (P1) but not after four days (P4) (Figure 1c and d), indicating that the CP for imprinting ends before P4.

Figure 1.

Behavioral paradigm and the critical period for imprinting.

(a) Schematic sequence of behavioral experiments. Dark-reared chicks were trained in a running wheel and tested the day after for visual and auditory imprinting. (b) During imprinting training, the chickens were presented with audiovisual stimulation. An animated object moved across the screen while a sound was presented every 3 s, coupled with pulsating movements of the object. (c) Auditory imprinting (left) was assessed by comparing the approaching behavior on the wheel to the imprinted sound or a novel sound. This procedure generated robust auditory imprinting when training was performed the day after hatching (gray, n = 13) but was ineffective four days after hatching (black, n = 12) (right). (d) Visual imprinting (left) was assessed independently, by comparing the approaching behavior to the imprinted or a novel image. Similarly to auditory imprinting, visual imprinting was strong in P1 (gray, n = 13) but absent in P4 (black, n = 12) (right). Plots show mean and SEM, * indicates p<0.05 from two-sample t-test.

DOI: http://dx.doi.org/10.7554/eLife.17197.003

DOI: http://dx.doi.org/10.7554/eLife.17197.004

Preference indexes of trained chickens during (P1) or after the critical period (P4).

Protein-synthesis dependency of auditory and visual imprinting

To assess whether a protein synthesis is enhanced after imprinting in both MNM and IMM we optimized an in vivo surface sensing of translation (SUnSET) protocol (Schmidt et al., 2009) for monitoring protein synthesis in vivo in these areas. Briefly, the antibiotic puromycin (PMY) incorporated into newly synthesized proteins can be detected through immunolabeling and used to monitor translation. Because brain tissue incorporates PMY more slowly compared to other tissues (Flexner et al., 1962) pilot experiments were conducted, showing that IP-injected PMY accessed the chicken’s brain within 3–4 hr. Thus PMY was injected 1 hr before a 2 hr training and samples were collected 4 hr after injection to capture training-induced translation.

We found that imprinting training increased translation both in MNM and IMM (Figure 2b,c) after a 2 hr training session, compared to controls, which were running on the wheel but presented with an empty screen. To further estimate the time-window during which auditory and visual imprinting are sensitive to protein synthesis inhibition, we trained chickens for 1 or 2 hr on P1. Two-hour (Figure 3a, right panel) training triggered robust auditory imprinting, which was blocked by the protein synthesis inhibitor clycloheximide (CHX) injected immediately after training (Figure 3a, left panel). In contrast, one-hour training did not elicit significant auditory imprinting (Figure 3a, right panel). Interestingly, the temporal dynamics of protein synthesis dependency of visual imprinting was different. While one-hour training triggered visual imprinting that was suppressed by CHX (Figure 3b, left panel), visual imprinting after two-hour training was not blocked by post-training administration of CHX (Figure 3b right panel). Consistent with the effect on behavior, CHX effectively blocked imprinting-triggered protein synthesis in both MNM and IMM areas (Figure 3c,d). Taken together, our results show that both auditory and visual imprinting trigger new protein synthesis, which is required for both auditory and visual imprinting.

Figure 2.

Experience-dependent increase in translation assessed with SUnSET.

(a) Temporally optimized SUnSET protocol to detect changes in translation in vivo (left) induced by the imprinting training. Schematic sagittal view of the chicken forebrain, showing the position of MNM and IMM (right). (b) The auditory imprinting area MNM (left) exhibits increased puromycin incorporation (green) after imprinting training, compared with MNM samples of chickens running on the wheel but not presented with the imprinting object. S6 (red) marker was used to identify cells somas. (c) In IMM (left) translation rates were also increased in trained animals. Sample sizes: MNM untrained (six chickens, 48 images at 10X, zoom 3X); MNM trained (six chickens, 48 images at 10X, zoom 3X); IMM untrained (six chickens, 48 images at 10X, zoom 3X); IMM trained (six chickens, 48 images at 10X, zoom 3X). Bar plots show mean and SEM; * indicates p<0.05 from unpaired Mann-Whitney test.

DOI: http://dx.doi.org/10.7554/eLife.17197.005

Puromycin signal was measured in MNM and IMM.

DOI: http://dx.doi.org/10.7554/eLife.17197.006

Figure 3.

Protein synthesis requirement in auditory and visual imprinting.

(a) The protein-synthesis inhibitor cycloheximide (CHX, n = 12) injected immediately after 1 hr training (left) had no effect on the auditory preference index compared to controls (n = 15) and vehicle-injected (n = 14) groups. In contrast, 2 hr training, which induced stronger preference to the imprinted sound, was blocked by CHX-treatment (n = 11) compared to controls (n = 13) and vehicle-injected group (n = 9). (b) Visual imprinting was already robust after 1 hr training (left) in controls (n = 15) and chickens injected with vehicle (n = 14), and blocked by CHX-administration (n = 12). On the other hand, 2 hr training (right) also induced robust preference to the imprinted visual object in controls (n = 13) and vehicle-injected chickens (n = 9) but was not blocked by CHX administration (n = 12). Plots show mean and SEM, * indicates p<0.05 from two-ways ANOVA test, Bonferroni Post hoc test. (c) SUnSET protocol used to detect experience-dependent translation changes in MNM and IMM in the presence or absence of CHX. (d,e) Puromycin (green) incorporation is decreased in trained animals treated with CHX. S6 (red) was used to identify cell somas. Sample sizes: MNM trained (five chickens, 40 images at 10X, zoom 3X); MNM trained and CHX administration (six chickens, 48 images at 10X, zoom 3X); IMM trained (five chickens, 39 images at 10X, zoom 3X); IMM trained and CHX administration (six chickens, 47 images at 10X, zoom 3X). Bar plots show mean and SEM; * indicates p<0.05 from unpaired Mann-Whitney test.

DOI: http://dx.doi.org/10.7554/eLife.17197.007

DOI: http://dx.doi.org/10.7554/eLife.17197.008

SUnSET results from trained and untrained chickens.

Preference indexes and SUnSET results from control chickens and injected with cycloheximide.

eIF2α-mediated translational control selectively regulates auditory imprinting

To investigate whether the translational program controlled by eIF2α is involved in imprinting, we first measured levels of phosphorylated eIF2α (p-eIF2α) in MNM and IMM after training P1 chickens. Intriguingly, training significantly decreased p-eIF2α in the auditory area MNM (Figure 4a, left panel), but not in the visual area IMM (Figure 4a, right panel). To examine whether a reduction in eIF2α phosphorylation is required for auditory imprinting we treated chickens before training with Sal003, an inhibitor of the eIF2α phosphatase complexes (McCamphill et al., 2015), which increases p-eIF2α levels (Figure 4b and Figure 4—figure supplement 1) and decreases translation (Figure 4—figure supplement 2). Interestingly, increasing p-eIF2α with Sal003 prevented auditory imprinting, but had no effect on visual imprinting (Figure 4c). These results indicate that decreasing p-eIF2α is only required for auditory imprinting.

Figure 4.

Translational control of auditory imprinting by eIF2α.

(a) After 2 hr imprinting training, IMM and MNM were punched out for western blot analysis. The ratio of phosphorylated eIF2α (p-eIF2α) and non-phosphorylated eIF2α was measured in controls and after training in MNM (left) and IMM (right) brain tissue. Trained chicks (n = 7) exhibited decreased eIF2α phosphorylation compared to the untrained (n = 6) in MNM but not in IMM. Representative western blots are shown below each panel. * indicates p<0.05 from unpaired Mann-Whitney test. (b) Left, drugs injected for targeting the eIF2α pathway. Right, schematic effect of pharmacological manipulations on the eIF2α pathway. (c) Auditory (left) but not visual (right) imprinting is blocked by Sal003 injection (n = 12) compared to controls injected with vehicle (n = 9). (d) Auditory imprinting (left) was enhanced in chickens injected with the PKR inhibitor PKRi (n = 26), compared to controls injected with saline vehicle (n = 14). On the other hand, PKRi (n = 26) had no effect on visual imprinting (right), compared to saline injection (n = 14). (e) Auditory imprinting (left) but not visual imprinting (right) was enhanced by ISRIB administration (n = 11) compared to controls injected with vehicle (n = 13). Bar plots represent mean and SEM, * indicates p<0.05 from unpaired t-test.

DOI: http://dx.doi.org/10.7554/eLife.17197.009

DOI: http://dx.doi.org/10.7554/eLife.17197.010

Western blots for p-eIF2α and total eIF2α of brain samples obtained from chickens 2 hr after injecting Sal003 (S) or vehicle (V). Sal003 treatment increased eIF2α phosphorylation. * indicates p<0.05 from Mann-Whitney U Test.

DOI: http://dx.doi.org/10.7554/eLife.17197.011

Sal003 and ISRIB can bi-directionally regulate protein synthesis. Both in MNM and IMM Sal003 reduces puromycin (green) incorporation while ISRIB enhances it. S6 (red) was used to localize cell somas. Bar plots show mean and SEM; different letters inside bars indicate statistically significant differences (p<0.05) between groups from Kruskal-Wallis test, Dunn’s multiple comparisons test.

DOI: http://dx.doi.org/10.7554/eLife.17197.012

Figure 4—figure supplement 1.

Sal003 increases eIF2α phosphorylation.

Western blots for p-eIF2α and total eIF2α of brain samples obtained from chickens 2 hr after injecting Sal003 (S) or vehicle (V). Sal003 treatment increased eIF2α phosphorylation. * indicates p<0.05 from Mann-Whitney U Test.

DOI: http://dx.doi.org/10.7554/eLife.17197.011

Figure 4—figure supplement 2.

ISRIB and Sal003 injection modulate translation in vivo.

Sal003 and ISRIB can bi-directionally regulate protein synthesis. Both in MNM and IMM Sal003 reduces puromycin (green) incorporation while ISRIB enhances it. S6 (red) was used to localize cell somas. Bar plots show mean and SEM; different letters inside bars indicate statistically significant differences (p<0.05) between groups from Kruskal-Wallis test, Dunn’s multiple comparisons test.

DOI: http://dx.doi.org/10.7554/eLife.17197.012

Western blots of p-eIF2α/ total eIF2α ratio and behavioral pharmacology after targeting the eIF2α pathway.

We next asked whether decreasing p-eIF2α would selectively enhance auditory imprinting. To this end, we first blocked the activity of the eIF2α kinase PKR, with a specific PKR inhibitor (Zhu et al., 2011) (PKRi). PKRi-injected chickens showed significantly stronger auditory imprinting compared to controls (Figure 4d). However, PKRi failed to affect visual imprinting (Figure 4d). Given that average locomotion towards the computer screen in both treated and control conditions was similar, the changes induced by altering eIF2α phosphorylation cannot be attributed to changes in overall motor activity. To further demonstrate that auditory imprinting could be enhanced by reducing eIF2α-mediated translational control, chickens were injected with ISRIB, a compound that blocks the translational effects induced by p-eIF2α (Sidrauski et al., 2013) and increases translation (Figure 4—figure supplement 2). Consistent with the PKRi-experiments, injection of ISRIB immediately after training enhanced auditory imprinting (Figure 4e) but not visual imprinting. Hence, a reduction in p-eIF2α-mediated translational control enhances auditory but not visual imprinting.

eIF2α dephosphorylation is required only for experience-dependent structural plasticity in the auditory imprinting pathway

Plasticity in dendritic spines, the major site of excitatory inputs in neurons, is thought to be crucial during CPs (Roberts et al., 2010) and part of the cellular substrate of memory (Lamprecht and LeDoux, 2004; Bourne and Harris, 2007; Nishiyama and Yasuda, 2015). Given that (a) long-term remodeling of spines requires protein synthesis (Nishiyama and Yasuda, 2015) and (b) translational control by p-eIF2α selectively regulates auditory imprinting, we next examined the role of this translational control mechanism in structural plasticity in imprinting-relevant brain regions. To measure changes in dendritic spine number and morphology after training (Figure 5b), we used the sparse Diolistic labeling technique (Figure 5a). The spines were classified (by observers blind to treatment) in stubby, filopodia, thin and mushroom (Figure 5c), a method that is informative about the functionality and maturity of spines (Bourne and Harris, 2007) and has been used in studies of learning-related structural plasticity (Sanders et al., 2012).

Figure 5.

Translational control of experience-dependent structural plasticity.

(a) Example diolistic labeling of a type I IMM neuron (63X), used to analyze the number and the shape of dendritic spines in MNM and IMM after training. (b) Representative confocal images of dendritic segments of IMM cells from untrained animals (63X, zoom 3X). (c) Schematic length (L) and shape criteria used for spine classification. (d) Trained chickens showed an increased number of mushroom spines (red) and a decrease in thin spines (blue) in MNM. The increase in mushroom spines induced by training was blocked by Sal003. Samples size: untrained (four chickens, 12 cells, 45 dendrites); imprinted (five chickens, 15 cells, 50 dendrites); Sal003 (five chickens, 15 cells, 55 dendrites). (e) Trained chickens showed an increase in mushroom spines (red) and a decrease in thin spines (blue) in IMM. In contrast to the changes in MNM, the increase in mushroom spines was not blocked by Sal003. Sample sizes: untrained (four chickens, 11 cells, 35 dendrites); imprinted (four chickens, 10 cells, 33 dendrites); Sal003 (five chickens, 16 cells, 48 dendrites). Total number of spines did not show significant differences across groups in either region. Bar plots show mean and SEM; different letters inside bars indicate statistically significant differences (p<0.05) between groups from Kruskal-Wallis test, Dunn’s multiple comparisons test.

DOI: http://dx.doi.org/10.7554/eLife.17197.013

DOI: http://dx.doi.org/10.7554/eLife.17197.014

Dendritic spines numbers in MNM and IMM of untrained, trained and Sal003-treated chickens.

While training failed to affect the total number of spines (Figure 5d–e), it significantly increased the number of mushroom spines and decreased the number of thin spines in MNM (Figure 5d) and IMM (Figure 5e), compared to control animals with experience on the running wheel but not subject to audiovisual training. We next examined whether blocking eIF2α dephosphorization with Sal003 prevents training-induced changes in structural plasticity. Remarkably, Sal003 administration blocked the training-induced increase in the number of mushroom spines only in MNM (Figure 5d,e). These results indicate that eIF2α phosphorylation not only controls the imprinting behavior during the CP but also structural plasticity, a potential cellular substrate of memory storage (Lamprecht and LeDoux, 2004; Nishiyama and Yasuda, 2015) in a key forebrain area involved in auditory imprinting.

Blocking p-eIF2α mediated translation reopens the critical period for auditory imprinting

Identifying the mechanisms that open the CPs could lead to novel therapeutic opportunities for a variety of cognitive disorders (Hensch, 2004). Given that (a) behavioral training decreases p-eIF2α (Figure 4a), (b) blocking p-eIF2α-mediated translation enhances auditory imprinting elicited by weak-training protocol (Figure 4d–e) and (c) Sal003-mediated increase in p-eIF2α blocks auditory imprinting, we wondered whether the drugs enhancing imprinting during the CP (PKRi and ISRIB) would restore imprinting outside the CP (Figure 6a). Remarkably, treatment with either PKRi or ISRIB (Figure 6a) on P4 selectively re-opened the CP for auditory imprinting (Figure 6b), again without affecting visual imprinting (Figure 6c). Hence, by promoting brain plasticity, the reduction of p-eIF2α-mediated translational control enhances auditory imprinting.

Figure 6.

Reopening the critical period for visual and auditory imprinting through eIF2α.

(a) Chickens were trained 4 days after hatching (P4) and tested 24 hr after training. To target translational control by eIF2α, chickens were injected with PKRi or ISRIB. (b) Controls injected with vehicle (n = 12) did not show auditory imprinting at P4 but the critical period in animals treated with PKRi (n = 13) or ISRIB (n = 13) was reopened. (c) Visual imprinting was not restored in chickens injected with PKRi (n = 13) and ISRIB (n = 13) or injected with vehicle (n = 12). * indicates p<0.05 from two-ways ANOVA test, Bonferroni Post hoc test.

DOI: http://dx.doi.org/10.7554/eLife.17197.015

DOI: http://dx.doi.org/10.7554/eLife.17197.016

Preference indexes of animals trained in P4 and injected with PKRi, ISRIB or vehicle.

Discussion

Regulation of imprinting and structural plasticity by eIF2α

Imprinting allows newborns to adjust behavior in response to relevant sensory experience, immediately after birth (Horn, 2004; Bolhuis, 1991). Despite having been studied for decades, the mechanism mediating the formation of imprinted memories remains elusive. Here we showed that, although visual and auditory imprinting require newly synthesized proteins, eIF2α-mediated translational control bidirectionally regulates auditory but not visual imprinting and related changes in structural plasticity. Remarkably, targeting this translational control mechanism pharmacologically recovers auditory imprinting after the closing of the critical period.

Critical periods in the auditory system have been widely studied across species (Scheich, 1987; Riebel et al., 2002; Yang et al., 2012; Insanally et al., 2009). Yet the mechanisms engaged during the CP for auditory imprinting have not been elucidated. One major limitation has been the design of stringent experimental approaches that control for social experience and innate biases, while achieving robust auditory imprinted memories (Van and Bolhuis, 1991). We aimed to address these concerns by: (1) raising chickens in darkness and constraining social interaction, (2) imprinting chickens to more than one type of object and sound, and (3) increasing the length of training compared to other studies (Wallhäusser and Scheich, 1987; Van and Bolhuis, 1991) to achieve significant memory retention longer than 24 hr after training. These improvements, in addition to the novel custom-made audiovisual animation used for training, provided a stronger experimental design for assessing auditory and visual imprinting.

Several lines of evidence support the modality-specific role of eIF2α. First, training decreased phosphorylation of eIF2α in the auditory-imprinting relevant area MNM, but not in IMM (Figure 4a). Second, pharmacologically increasing eIF2α phosphorylation with Sal003 selectively disrupted auditory imprinting (Figure 4c). Third, inhibiting eIF2α phosphorylation with PKRi or directly blocking p-eIF2α-mediated translational control with ISRIB, enhanced auditory imprinting after weak training (Figure 4d–e). Interestingly, although Sal003 and ISRIB altered protein synthesis in IMM, these manipulations had no detectable effect on the formation of visual memories. The reason why eIF2α is not involved in visual imprinting is not yet understood. It is possible that expression of eIF2α kinase or phosphatase complexes differs between visual and auditory areas, or that upstream signaling pathways fail to engage eIF2α dephosphorylation. It would be interesting to test whether other tasks involving memory formation, such as one-trial avoidance learning (Atkinson et al., 2008), also require translational control by eIF2α. An appealing idea is that other translational control pathways such those controlled by the mechanistic target of rapamycin mTORC1 (Sonenberg and Hinnebusch, 2009) mediate the formation of visual memories.

While previous studies in adult rodents suggest that eIF2α-mediated translation regulates the two major forms of synaptic plasticity (Costa-Mattioli et al., 2007; Zhu et al., 2011; Stern et al., 2013), here we report for the first time that experience-dependent structural plasticity of dendritic spines requires eIF2a dephosphorylation. Furthermore, and consistent with our behavioral results, eIF2α-mediated translation exclusively regulated spine remodeling in the auditory but not in the visual area. This result is particularly important since structural plasticity is crucial during CPs (Roberts et al., 2010; Mataga et al., 2004) but the underlying molecular mechanisms were unknown. Different forms of structural plasticity have been linked to memory, including spine turnover and morphological changes of preexisting spines (Lamprecht and LeDoux, 2004). In this case, the structural plasticity found in IMM and MNM could be consistent with potentiation and enlargement of specific dendritic spines, favoring the detection of imprinted stimuli. While we did not observe changes in spine density, the increase in mushroom spines and decrease in thin spines may suggest coordinated structural plasticity as previously reported in hippocampal slices (Bourne and Harris, 2011). Thus, our results shed light on the biological basis of experience-dependent spine remodeling and uncovered eIF2α as a major player in spine remodeling.

Another interesting question is whether translational control by eIF2α in glial cells affects imprinting. Glutamate application induces a transient increase of eIF2α phosphorylation in glial cells in vitro (Flores-Méndez et al., 2013). This effect has been linked to glutamate removal from the synaptic cleft by glial cells (Flores-Méndez et al., 2013). However, its contribution to memory formation in vivo remains untested. In future studies, it will be important to dissect the role of p-eIF2a in memory formation at the cellular level.

eIF2α-mediated translational control: an evolutionarily conserved mechanism to rejuvenate plasticity and memory?

Behavior is shaped during sensitive periods in early postnatal life, characterized by epochs of heightened brain plasticity (Hensch, 2004; Nabel and Morishita, 2013). Reactivating such plasticity in the adult brain has the potential to rehabilitate brain function after CPs are closed (Hensch, 2004; Nabel and Morishita, 2013; Hübener and Bonhoeffer, 2014). This has been successfully achieved in the visual cortex of rodents through direct manipulation of inhibitory synaptic transmission, either pharmacologically (Hensch et al., 1998) or through transplantation of embryonic inhibitory neurons (Davis et al., 2015). Moreover, in mice and humans, releasing ‘epigenetic brakes', could reopen auditory CPs (Yang et al., 2012; Gervain et al., 2013). Our results uncover a translational control mechanism as a novel target for reopening CPs. Indeed, two different strategies, either blocking p-eIF2α-mediated translation or inhibiting the upstream kinase PKR, enabled chickens to imprint to sounds after the end of the CP, suggesting that blocking p-eIF2α-mediated translation control enhances CP-mediated plasticity. A recent report shows that reducing p-eIF2α-mediated translational control in the VTA can convert adult into adolescent mice with respect to their vulnerability to cocaine-induced changes in synaptic strength and behavior (Huang et al., 2016). Based on these results and the evolutionarily conserved nature of this process, we speculate that reopening CPs through blockade of eIF2α-mediated translational control could be used to recover plasticity in the mature brain and treat cognitive dysfunctions.

Materials and methods

Animals

We used newly hatched chicks of both sexes from the White Leghorn strain Gallus gallus domesticus (Charles River supplier). Fertilized eggs (embryonic ages E14-17) were obtained and subsequently incubated in darkness at 37–38°C under controlled humidity (Grumbach, compact S84). Upon hatching, chickens were transferred to individual compartments of a brooder maintained at 37–38°C (Brinsea, TLC-5), where they remained in darkness until each experiment. Water and food was provided. It has been shown that chickens are able to eat and drink water in the dark and that this housing does not impact visual acuity or locomotion, compared to chickens reared under light conditions(Yamaguchi et al., 2012). These experiments were approved by the institutional animal care committee (IACUC) at Albert Einstein College of Medicine (protocol 20140910).

Imprinting training and preference test

Training sessions and tests were performed in a sound proof chamber (IAC acoustics) at 37°C in the dark, except for the light coming from the monitor. All experiments and drug manipulations were performed blind to treatment. On the training day, each chicken was placed under white light for 30 min. This priming procedure has been extensively used in visual imprinting (Bolhuis et al., 2000; Nakamori et al., 2010). After priming, chickens were placed in a running wheel (internal diameter = 18 cm) in front of a computer monitor (ACER LCD, 17''). Magnets mounted on the wheel (Gibbs and Lecanuet, 1981) allowed the precise measurement of the approaching behavior by a counter (Med Associates, DIG-700G, DIG-726). Each magnet count generated a TTL signal, whose timing was stored in a computer for offline analysis.

Visual stimuli consisted of custom-made animations (Blender, http://www.blender.org/) of either a blue or red rectangular prism coupled to a sound. Both figures had exactly the same volume and followed the same rotation and movement across a virtual room (Video 1 and Video 2). This method made it possible to synthesize arbitrary movement patterns while controlling luminosity, color and shape. Objects changed shape (expansion and contraction) synchronously with sound. Two different sounds were synthesized using Audacity software (Audacity 2.1.0). The frequency range for both sounds was 0–3 KHz. Sound one consisted in frequency steps and sound two was composed of frequency sweeps (see supplementary material). Each sound was played 12 times during a minute, every 3 s. The start of each animation was commanded by software written in Matlab, which was interfaced to Med Associates equipment through a USB DAQ card (National instruments USB-6008).

Video 1.

Stimulus A presented to chickens.

This animation was played on a screen during training. For auditory and visual imprinting tests only the auditory or the visual component was presented.

DOI: http://dx.doi.org/10.7554/eLife.17197.017

Video 2.

Stimulus B presented to chickens.

This animation was played on a screen during training. For auditory and visual imprinting tests only the auditory or the visual component was presented.

DOI: http://dx.doi.org/10.7554/eLife.17197.018

Audiovisual training stimuli were presented in 4 min bouts followed by 1 min of silence and darkness. If the chicken did not move the wheel during the first half hour of exposure, the experiment was interrupted and not included in the sample. Training length varied from 0 to 120 min, depending on the protocol. To investigate long-lasting effects on imprinting we tested chickens the day after training.

Visual and auditory imprinting were tested independently in a sequential test, where the novel and imprinted stimuli are presented in alternation. While other studies have used a simultaneous choice test (Yamaguchi et al., 2012), the sequential test allowed us to randomize stimulus presentation, measure baseline locomotion and assess the response to novel and imprinted (Video 3) stimuli independently. Each test included 5 presentations of the imprinted stimulus and 5 presentations of the novel stimulus. The duration of each presentation was 1 min. Baseline locomotion was measured during 30 s between trials. Imprinted and novel stimuli were alternated over five consecutive blocks. The first stimulus that started the sequence was picked randomly. Although this method differs from previous reports where fixed sequences were used (Bolhuis et al., 2000; Town and McCabe, 2011; McCabe and Horn, 1988), randomization prevents biases and motivation changes over time emerging from fixed sequences.

Video 3.

Chicken imprinted to stimulus B approaching the screen.

This approach behavior was quantified during the presentation of imprinted or novel stimuli to compute a preference index.

DOI: http://dx.doi.org/10.7554/eLife.17197.019

Previous studies have used different criteria and indexes to quantify the strength of imprinting. Such quantifications have included differences in time spent in the proximity of the imprinted object (Yamaguchi et al., 2012), differences in locomotion toward the imprinted and novel stimulus (Bolhuis et al., 2000), differences in locomotion during the presentation of imprinted and novel objects and the absence of a stimulus (Maekawa et al., 2006), and number of chickens within a group selecting the imprinted stimulus over several trials (Wallhäusser and Scheich, 1987). In this study, we normalized differences between locomotion to novel and imprinted stimuli by the average baseline locomotion in the wheel when no stimulus was presented. Therefore, to assess imprinting, we calculated a preference index (PI), PI = ∑(Imprinted where STL indicates stimulus-triggered locomotion either during the presentation of the imprinted stimulus (Imprinted) or presentation of the novel stimulus (Novel), and baselineA refers to the average baseline locomotion across the experiment. An advantage of this quantification over previous methods is that: (1) it takes into account the fluctuations in basal locomotion before each stimulus presentation, and (2) it weights differences in approaching behavior by average locomotion.

Assessment of the sensitive period

The sensitive period for filial imprinting has been reported to close within 3–4 days after hatching(Yamaguchi et al., 2012). To ensure the training and preference tests captured this sensitivity, the ability of chickens to develop a preference to visual and auditory stimuli within the first 4 days after hatching was measured immediately and 24 hr after training.

In vivo SUnSET

We optimized previously reported in vivo SUnSET protocols in muscle fibers (Goodman and Hornberger, 2013; Goodman et al., 2011) for monitoring protein synthesis in the chick brain. It has been shown that PMY injected intravenously takes 2–4 hr to be incorporated into the brain (Flexner et al., 1962). This contrasts with the fast incorporation (approximately 30 min) into muscle (Goodman and Hornberger, 2013; Goodman et al., 2011) and other organs (Flexner et al., 1962). In pilot experiments, we determined that 3–4 hr after injecting a low dose of PMY (MP Biomedicals, 0.04 mg/g, diluted in distilled H20, IP) was the optimal time period for detecting the incorporation of PMY in newly synthesized proteins. This information was used to adjust the timing of PMY injection in our behavioral pharmacology experiments.

To simultaneously assess experience-dependent translation across sensory modalities and brain regions, in the same animal, we identified a training schedule that reliably triggered auditory and visual imprinting. Since 2 hr but not 1 hr training (Figure 3a,b) triggered both auditory and visual imprinting, we used the former schedule. Four hours after PMY injection chicks were decapitated and brains were rapidly (2–3 min) placed in cold PFA (4%) overnight at 4°C. A vibratome (Leica VT 1000S) was used for making 100 µ sagittal cross sections. After three 10 min washing with PBS, samples were incubated overnight at 4°C in a solution containing antibodies against PMY (EMD Millipore, cat# MABE343, RRID:AB_2566826) and S6 (Cell signaling, cat# 2217, RRID:AB_331355) to identify cell somas. Samples were washed in PBS (three 10 min wash) and placed for 1.5 hr in a solution containing Alexa-488 (Invitrogen, cat# A21202, RRID:AB_2535788) and Alexa-568 (Invitrogen, cat# A10042, RRID:AB_2534017) against the primary antibody host species. After washing again 3 times for 10 min in PBS, samples were covered with Prolong Gold mounting media (Molecular probes, cat# P36935).

A confocal microscope (Zeiss LSM 510 Meta Duo V2) was used to collect images from IMM and MNM (10X, zoom 3). All images were taken blind to the experimental groups. IMM is located 2.5 mm from the dorsal surface of the brain and 0.5–1 mm from the caudal edge of the forebrain, limited below and laterally by the lateral ventricle. MNM is located 0.5–1 mm lateral from the midline, 3 mm from the dorsal surface of the brain and 5 mm from the caudal edge of the forebrain, below the lateral pallial lamina that separates the hyperpallium and mesopallium (Puelles et al., 2007). All compared samples were processed the same day, using the same protocol, and images were taken with equal microscope settings. Control animals were housed in the same conditions as trained animals but presented with an empty screen.

Images were analyzed using ImageJ software (NIH, 1.50i). Threshold was adjusted by the S6 signal to select cell somas. PMY signal was detected using the selection created for the S6 channel. To compare across groups all measures were normalized to the average intensity of the control group.

Protein synthesis inhibition

To investigate the involvement of protein synthesis in long-term memory formation during imprinting we injected cycloheximide (Tocris, CHX, 1 mg/kg, IP), diluted in 0.1% DMSO and saline, immediately after training. Since 1 hr training was enough to generate visual (Figure 3a) but not auditory imprinting (Figure 3b), we injected CHX immediately after 1 hr and 2 hr training, and tested the effect on imprinting 24 hr later for each sensory modality, independently.

Manipulation of the eIF2α signaling pathway

We used the specific blocker of eIF2α phosphatases Sal003 (Sigma Aldrich,0.2 mg/Kg, diluted in 0.1% DMSO and 0.9% Saline, IP) to test whether a reduction in eIF2α phosphorylation is required for imprinting. We used 2 hr training for this experiment, which reliably triggered strong visual and auditory imprinting, and injected Sal003 before training to ensure translation was inhibited during and immediately after training.

To specifically enhance the formation of imprinted memories by reducing eIF2α–mediated translational control, we conducted two independent manipulations: animals were injected immediately after training with either the specific inhibitor of the eIF2a kinase PKR (PKRi; EMD Millipore, 0.1 mg/Kg, diluted in 0.1% DMSO and 0.9% saline) or ISRIB (Sigma Aldrich, 2.5 mg/Kg, diluted in 50% DMSO and 50% saline, IP), which blocks the translational effect induced by p-eIF2α. To avoid a ceiling effect masking the enhancement of imprinting, we used 1 hr training (weak training) and tested preference 24 hr after training.

Western blotting

Lysates of IMM and MNM (anatomical boundaries described above) were obtained from brain tissue, punched out from 0.75- to 1mm-thick sagittal brain slices collected from imprinted and control animals. We used antibodies against eIF2α (Cell Signaling Cat #9722, RRID:AB_2230924), p-eIF2α(Ser51)(Cell Signaling Cat #9721, RRID:AB_330951), following standard protocols described before(Costa-Mattioli et al., 2007). Control tissue samples were obtained from chickens that ran on the wheel towards a screen displaying only a static image of an empty room, as shown in Figure 1b (left panel).

Reopening of the CP

We tested whether reducing p-eIF2α by PKRi and ISRIB administration could reopen the CP for each sensory modality using 2 hr training on P4. Since injecting PKRi and ISRIB immediately after 1 hr training did not have an effect on visual imprinting, we injected PKRi (Stern et al., 2013; Ingrand et al., 2007) (0.1 mg/Kg, IP) and ISRIB (2.5 mg/kg) before training to control whether the lack of effect on visual imprinting was due to the time of the injection. Imprinting was assessed 24 hr after training as described above.

Dendritic spine analysis

Brains were rapidly dissected (in 2–3 min) and placed in paraformaldehyde (4%) for 1 hr, then transferred to the phosphate buffer solution. A vibratome (Leica VT 1000S) was used for making 200 uM slices. Tungsten beads coated with lipophilic dye (DiI) were delivered to each slice using a modified gene gun (Gan et al., 2000). The dye was allowed to spread overnight. The next day, each slice was mounted using ProLong Gold mounting media. A confocal microscope (Zeiss LSM 510 Meta Duo V2) was used to collect Z-stacks (63X, zoom 3) from areas of interest containing labeled dendritic branches. Images of secondary branches, within 50–75 µm from the soma, were used for spine analysis.

Dendritic spines were counted blind to experimental groups using Image J software (Version 1.50a). A multicolored lookup table (Fire) was used to reliably visualize individual spines. Two 10 μm segments were marked randomly along each secondary dendritic branch. Spines along each of the two segments were counted by a blind experimenter. The spines ‘head width, presence of neck and overall length were used for classifying them in filopodia, stubby, thin, or mushroom, using published criteria (Bourne and Harris, 2007; Sanders et al., 2012; Chakravarthy et al., 2006). Briefly, spines without clear head and neck, and shorter than 1 μm, were categorized as stubby. Spines longer than 1 μm were classified as mushroom or thin, depending on whether a head and neck were observed. Protrusions longer than 2 μm were categorized as filopodia.

To investigate if eIF2α was required for structural plasticity, we injected chickens with Sal003 (i.p., 0.2 mg/kg) and trained them for 2 hr. The day after the training, we labeled dendritic arbors and assessed dendritic spines, as described above.

Statistical analyses

Statistical analyses were performed using SigmaPlot (Systat Software). Data distribution normality was assessed using the Shapiro-Wilk and F-test to evaluate the differences of variances. When variances were significantly different the Welch’s correction was used. Statistics were based on the two-sided Student’s t test, or the two-way ANOVA and Bonferroni post-hoc test for multiple comparisons of normally distributed samples. Otherwise the Mann-Whitney or the Kruskal-Wallis and Dunn’s multiple comparisons tests were used. Within-group variation is indicated by standard errors of the mean of each distribution, which are depicted in the graphs as error bars. p<0.05 was considered significant.

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: this article was originally rejected after discussions between the reviewers, but the authors were invited to resubmit after an appeal against the decision.]

Thank you for submitting your work entitled "Translational control of auditory imprinting and structural plasticity by eIF2α" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Eve Marder as the Senior Editor. The reviewers have opted to remain anonymous.

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife.

The question of what regulates critical periods and imprinting has broad appeal, and your efforts to show a modality specific translational control mechanism underlying the critical period for imprinting were of great interest to the reviewers. As currently presented, however, the reviewers did not find your conclusions that the translational regulator eIF2α and its phosphorylation state regulate the formation of auditory memories and spine plasticity convincing. For the protein synthesis aspect, better demonstration of actual alterations in synthesis of given proteins would strengthen the study. In addition, more clearly stating the time points of injection relative to the timing of training, confirmation of the brain areas examined for sampling of spine changes, and clarification of whether the p-eIF2α signal derived from neurons or from neurons and glia by immunohistochemistry of the tissue taken for immunoblotting, would enhance your presentation.

Your attempts to distinguish effects on auditory but not visual imprinting are laudable, but are confounded by difficulties in the field in separating stimuli for auditory versus visual learning; your design should be better argued compared to previous efforts. Finally, the novel method you use for the preference index for analyzing imprinting could be better explained and compared to previously used tests, and the appropriateness of the statistical tests you used considered.

Reviewer #1:

The work by Batista et al. is very intriguing. Understanding what regulates critical periods and imprinting are indeed important questions that need to be addressed and will be of great interest to a broad community. While the title suggests that eIF2α controls translation-dependent auditory imprinting and structural plasticity, the authors do not provide any direct evidence that new proteins are synthesized. At best, the authors can claim that eIF2α is involved in auditory imprinting. Thus, the authors need to provide convincing evidence that translation, or increased protein synthesis, is required for auditory imprinting and structural plasticity. In addition, while the literature suggests eIF2α is involved in memory formation, so are other protein synthesis pathways such as mTOR, as mentioned in the Discussion. The authors should include Western blot analysis for other protein synthesis pathways, as a control.

Major comments:

Figure 2

The use of cycloheximide does not indicate that protein synthesis is required. To show convincingly that protein synthesis is involved, the authors must show that (1) total protein synthesis (or some proteins) is elevated at 2-hour training in auditory imprinting and at 1-hour training in visual imprinting, and (2) training-induced increase in protein synthesis is blocked by cycloheximide. This can be done by metabolic labeling or demonstration that levels of select proteins are increased by Western blot analysis.

Figure 3

Total eIF2α seems to be reduced in imprinted samples. The authors should measure other proteins whose levels are not affected by imprinting and normalize total eIF2α to demonstrate that imprinting affects eIF2α phosphorylation and not total levels of eIF2α.

The authors should also show that the manipulation of eIF2α phosphorylation directly affects protein synthesis. They may demonstrate this by measuring total protein (or select proteins) levels of MNM.

Figure 5

The authors use PKRi and ISRIB to investigate the critical period. It's rather dubious that Sal003 was not used to investigate the critical period. Does Sal003 fail to open the critical period? The authors should use the same manipulation/drugs for Figures 4 and 5.

Reviewer #2:

This manuscript addresses an important issue in the molecular mechanism of imprinted memories during a critical period. The authors argue that the translational control by phosphorylation of the eukaryotic translation-initiation factor 2α subunit (eIF2α) is required for auditory imprinting, and mediates changes in dendritic spines in the MNM region. They present data that the increasing phosphorylated eIF2α prevents the formation of auditory memories and spine plasticity. In contrast, the inhibition of an eIF2α kinase enhances auditory memories. Furthermore, they provide evidence that the blocking p-eIF2α mediated translational control enhances auditory memories and reopens the critical period. They propose a modality-specific translational control mechanism underlying the critical period for imprinting.

The paper contains sufficient interest and originality to merit publication, however, the impact is weakened by a technical limitation of the methods used to quantify the preference score and to analyze the statistics. While the findings presented in this study should be of interest to the readers in the field of animal behavior, it is my opinion that a rather substantial revision, based on the technical comments given below, is needed to make this manuscript suitable for publication.

Major points

My main concern is that the preference index of the authors (/baseline) differs from that of the conventional method of Horn et al. (imprinting/)(P14 L231). Did they use the background on the screen as the "baseline" image? Compared to the method of Horn et al., this method seems to be affected by the individual difference of animals. I am wondering why the authors used "baseline". The authors should clarify the reason why each score is divided by the baseline. The authors may hypothesize that more active chicks in locomotion should show bigger number of preference score than inactive chicks. But the less active chicks show the highest preference in some cases. If the data are analyzed by the method of Horn et al., it will be easier for most of the readers in the field of imprinting research to understand the significance of this paper.

As for the statistical analysis, the authors use either unpaired t-test or Mann-Whitney U test to examine the difference between two groups. The statistical tests were not performed appropriately in some cases. For example, Mann-Whitney U test should be used in the data of Figure 3D (Auditory imprinting) because the variances between two groups in Figure 3D are significantly different (F-test, p=0.0253), but the authors use unpaired t-test. On the other hand, unpaired t-test should be used for data of Figure 3A, because the variances between two groups were not different (F-test, ns). The authors should examine the difference of variances by F-test in all data to choose the appropriate statistical test.

There is controversy whether there is genuine auditory imprinting (Bolhuis, J.J. et al., Behaviour Vol. 122, (1992), pp.195-230). It is generally concluded that auditory stimuli play an important role in the formation of filial preferences, but that auditory imprinting is not as prominent when compared to visual imprinting. Although a lot of studies have demonstrated that the addition of an auditory stimulus improves following of a visual stimulus, few studies using only auditory exposure have demonstrated significant auditory learning. As is the case in this paper, the majority of studies of early auditory learning have used a compound of a visual and an auditory stimulus during training. The authors should discuss the physiological significance of auditory imprinting in relation to visual imprinting in the Introduction section.

Figure 2

I suggest the chemicals like cycloheximide should be used by direct injection into the target brain area, not by intraperitoneal injection because cycloheximide is a potent inhibitor of translation and therefore will be toxic to any kind of cells in the whole body.

Figure 3

In western blotting, the time course experiment at different time points (before training, during training, 2-3 h after training, et al.) is necessary to know whether the decrease of p-eIF2α is temporary or not. Is there any difference in the amount of p-eIF2α in the MNM region between 1 day old and 4 day old control chick?

Did the authors examine the decrease of p-eIF2α was learning dependent or locomotion dependent? The authors may know the answer by doing experiments under the condition of forced running.

As the authors punched out the region of MNM or IMM out of brain slice section, they should mention the exact location in the brain for the extraction in detail in the methods section. This definition of the brain area is very important because either the MNM or IMM region does not have the clear border from surrounding areas. Was the decrease of p-eIF2α detected specifically in the MNM region of the brain?

Figure 4

I understand the imaging study takes a lot of effort to detect the change of the properties of dendritic spines. I assume the neurons are randomly labeled in the MNM. Therefore, more than 1000 spines at least per each experimental setting should be examined to detect significant morphological changes beyond individual deviation. At the same time, the exact location of brain area they examined and the morphological definition of dendritic spine for classification in chick should be mentioned in detail in the method section

Reviewer #3:

This manuscript from Batista et al. is beautifully written and addresses the role of translational regulation during critical period plasticity in the chick. The study includes data indicating that auditory imprinting involves translation initiation regulation by eIF2α, and in particular that eIF2α is phosphorylated and inhibited to allow for plasticity in the auditory cortex during the critical period. This indicates that mRNAs containing upstream open reading frames (whose translation is upregulated when eIF2α is phosphorylated and inhibited) are of particular importance to auditory plasticity during the critical period. In contrast, the authors show that phosphorylation of eIF2α, and hence eIF2α translational regulation, is not involved in visual imprinting. The findings are intriguing and I believe important. However, I also believe that there are some significant flaws in the experimental design and interpretation that weaken the study. My principal concerns have to do with the question of how well the translational inhibition worked, and the question of the kinetics of translational regulation in the MNM (for auditory imprinting) and IMM (for visual imprinting). What were the kinetics of inhibition of protein synthesis following IP injection? Was it equivalent in MNM and IMM? It's not clear from the manuscript whether or not the IP injection was done at the end of the 1 hr and the 2 hr training. The authors write: "both auditory and visual imprinting rely on protein synthesis but following different temporal dynamics", and yet the conclusions from the translational inhibition experiments are all based on manipulations that do not appear to consider the differential temporal dynamics of imprinting (because they simply target a single time point, optimized for the effects on auditory imprinting). I also think it's really important to show that there is an effect on translation, and in particular on eIF2α-mediated translation. This is provided for Sal003 in Figure 3—figure supplement 1, but only for "brain samples," not for IMM or MNM. Without this, the results are all based on pharmacological interventions, which always have some caveats unless the expected effects are directly measured. For example it looks like there is significant variability in the ratio of peIF2α to eIF2α in Figure 3—figure supplement 1, which raises the question of whether this correlates in any way with plasticity during imprinting.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for resubmitting your work entitled "Translational control of auditory imprinting and structural plasticity by eIF2α" for further consideration at eLife. Your revised article has been favorably evaluated by Eve Marder as the Senior editor, a Reviewing editor, and two reviewers.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

The reviewers have several queries, and ask for further clarification on the following:

Reviewer 1 asks for you to explain why protein synthesis dependent visual plasticity at 1 hour of training was ignored and the focus placed on 2 hours; an explanation would help the reader move to the logic of the next set of experiments.

In addition, this reviewer wanted an explanation of the relevance of structural spine changes over density in the context of behavior.

Reviewer 2 wondered why the first trials of the novel stimulus strongly suppressed locomotion; this does not seem to occur in the reviewer's hands in their experimental design. This reviewer is concerned about whether the training for imprinting by the authors is sufficient to cause strong preference compared with controls.

Here are the comments in greater detail:

Reviewer #1:

The manuscript "Translational control of auditory imprinting and structural plasticity by eIF2α" by Batista et al. is greatly improved. This is an important manuscript describing how protein synthesis mediates structural plasticity and behavior. The finding that restoring translational control of eIF2α opens the critical period is timely and important. The authors have addressed most of my concerns.

Still missing from the manuscript is an explanation as to why the authors ignored the protein synthesis dependent visual plasticity at 1 hour of training. A sentence or two justifying the shift toward the 2 hour training period and the focus on auditory imprinting over visual imprinting would provide the reader with the logic behind the next set of experiments. Also, further discussion on the relevance of the structural spine changes over density in the context of behavior would strengthen the manuscript (see Bourne and Harris, Hippocampus, 2011).

Reviewer #2:

The revised manuscript is a great improvement on the original and now suitable for publication.

I will just add some comments about the authors' preference index in imprinting. I am a little bit surprised to hear that the first trials of the novel stimulus generated a stronger suppression of locomotion because it never happens when we use the simultaneous choice test as a method to measure the preference for visual imprinting. It seems to me that the novel stimulus causes fear for chicks in the initial trials. Usually, newly-hatched chicks show the preference to novel conspicuous moving objects. Rather in some cases, I even choose the object which attracts the intrinsic preference of chicks as a control in the test. Under the condition, the imprinted chicks still show the strong preference against the controls in the simultaneous choice test. I am worried whether or not the training for imprinting by the authors is sufficient enough to cause the strong preference against the controls. Also, there is a possibility that the increasing number of trials in the test can bring the similar effect as the imprinting training. The authors may improve the methods of the training and the test in future to strengthen their interesting findings by the molecular approach.

[Editors’ note: the author responses to the first round of peer review follow.]

The question of what regulates critical periods and imprinting has broad appeal, and your efforts to show a modality specific translational control mechanism underlying the critical period for imprinting were of great interest to the reviewers. As currently presented, however, the reviewers did not find your conclusions that the translational regulator eIF2α and its phosphorylation state regulate the formation of auditory memories and spine plasticity convincing. For the protein synthesis aspect, better demonstration of actual alterations in synthesis of given proteins would strengthen the study.

As requested by the editor, we now measured protein synthesis in vivoin response to drug treatments and training using the SUrface SEnsing of Translation (SUnSET) technique. We now show that imprinting triggers translation in imprinting-relevant areas of the forebrain (new Figure 2B) and validate all pharmacological manipulations on the eIF2α pathway as means of regulating translation (new Figure 4—figure supplement 2). These new results strengthen our conclusions on the translational control of auditory imprinting by eIF2α.

In addition, more clearly stating the time points of injection relative to the timing of training, confirmation of the brain areas examined for sampling of spine changes, and clarification of whether the p-eIF2α signal derived from neurons or from neurons and glia by immunohistochemistry of the tissue taken for immunoblotting, would enhance your presentation.

To clarify the time points of injection relative to the timing of training we added a schematic in each figure indicating when injection was performed relative to training. Also, in the methods section we provide the rationale for the timing of the injection for each experiment.

As pointed out by one of the reviewers, the boundaries of MNM and IMM are diffuse, thus now we report the coordinates used in our study to sample spine changes within those regions. In brief, since there are no- specific markers for each area, we determined the location of MNM and IMM based on anatomical landmarks. Also, while these structures are reported to be large (~1 mm3), we constrained our search for cells within a 0.5 mm radius around the center.

eIF2α is ubiquitous across cell types, therefore we find interesting the question whether experience-dependent changes in eIF2α phosphorylation emerge from glia or neurons. Indeed, even in the memory field the cell-type contribution of eIF2α phosphorylation remains to be determined. in vitrostudies have shown that changes in glial eIF2α after glutamate application are transient, lasting around 5-10 minutes. It has been speculated that this process helps glial cells to cope with the metabolic stress of removing glutamate from the synaptic cleft. We included this information at the end of the third paragraph of the Discussion section and raised this interesting and important question for future work.

Your attempts to distinguish effects on auditory but not visual imprinting are laudable, but are confounded by difficulties in the field in separating stimuli for auditory versus visual learning; your design should be better argued compared to previous efforts.

It has been shown audiovisual stimulation enhances both visual and auditory imprinting tested separately (Bolhuis et al., 1992). Moreover, investigations on visual imprinting have commonly used audiovisual stimulation during training. However, most studies have focused on the mechanisms underlying visual imprinting because studying auditory imprinting carries additional experimental challenges, such as effect of previous social interaction and the difficulty in eliciting robust auditory memory. To clarify how we overcame these difficulties we revised the second paragraph of the Discussion section. In brief, our behavioral paradigm aimed to: 1) restrict social interaction and raise chicks in darkness, 2) generate a balanced data set where chicks were trained with two different synthetic sounds, and 3) increase the training length compared to previous studies to generate a robust long-lasting auditory memory.

Finally, the novel method you use for the preference index for analyzing imprinting could be better explained and compared to previously used tests, and the appropriateness of the statistical tests you used considered.

Imprinting strength has been assessed differently across studies. We now mention the different approaches used in the past before introducing ours. As pointed out by one of the reviewers, the classic preference score (PS) was the ratio between locomotion during imprinted trials and total locomotion, PS= locomotion during imprinted trials / total locomotion. This preference index was used previously to assess imprinting strength in a 4-trial testing paradigm with stimuli presented in ‘fixed’ sequence (either imprinted- novel-novel-imprinted or novel-imprinted-imprinted-novel). Upon preliminary analysis, we noticed that considering only the first 4 trials for computing the classic preference score overestimated the strength of the imprinting (Author response image 1). This was mainly because the first trials of the novel stimulus generated a stronger suppression of locomotion, compared to subsequent presentations of the same stimulus. For this reason, we decided to test imprinting strength in 10 pseudo-random trials. In addition, total locomotion varied largely from first to last trials mainly because locomotion during the novel trials increased over time (Author response image 1). Because baseline locomotion was the most stable parameter, on average, we decided to use it to normalize the difference in locomotion between imprinted and novel trials. We believe our computation of the preference score is advantageous for randomizing the stimulus presentation (thus ruling out biases due to sequence), normalizing by a more stable variable (baseline) and increasing the number of trials to 10 to assess not only novelty detection but also the maintenance of the preference over time.

Author response image 1.

) Classic preference score (PS) for auditory and visual imprinting computed as PS=locomotion during imprinted trials/ total locomotion, in chickens trained for 2 hours (n=13), b) locomotion before and during the presentation of the imprinted stimulus (n=26, in each trial) and c) locomotion before and during the presentation of the novel stimulus (n=26, in each trial)

DOI: http://dx.doi.org/10.7554/eLife.17197.020

We appreciate the comments on the statistical analysis. We revised the tests used, considering the differences in variance between groups, and applied the appropriate tests for each comparison. The statistical significance persisted after the revised analysis.

In summary, we now show that: a) training triggers an increase in translation in vivothat can be blocked with cycloheximide, b) translation is decreased in animals treated with Sal003, which increases eIF2α phosphorylation and blocks auditory imprinting, and c) translation is increased in animals treated with ISRIB, an enhancer of eIF2α- dependent translation that reopens the critical period for auditory imprinting. In addition, we clarified the timing of the injection, included a discussion of the putative role of p-eIF2α translational control in glial and justified our experimental design with reference to previous studies and a detailed analysis of the chickens’ behavior.

Reviewer #1:

The work by Batista et al. is very intriguing. Understanding what regulates critical periods and imprinting are indeed important questions that need to be addressed and will be of great interest to a broad community. While the title suggests that eIF2α controls translation-dependent auditory imprinting and structural plasticity, the authors do not provide any direct evidence that new proteins are synthesized. At best, the authors can claim that eIF2α is involved in auditory imprinting. Thus, the authors need to provide convincing evidence that translation, or increased protein synthesis, is required for auditory imprinting and structural plasticity. In addition, while the literature suggests eIF2α is involved in memory formation, so are other protein synthesis pathways such as mTOR, as mentioned in the Discussion. The authors should include Western blot analysis for other protein synthesis pathways, as a control.

We thank the reviewer for this comment. As requested, we now measure translation in vivoin response to training and tested the effect of different drugs. As shown in the new Figure 2B, training increased translation in imprinting-relevant areas. In addition, we demonstrate that this increase in translation can be blocked with cycloheximide (Figure 3D, C). Also, we validated our pharmacological manipulations and demonstrated that Sal003 reduces and ISRIB enhances protein synthesis in vivo(new Figure 4—figure supplement 2).

We acknowledge that other protein synthesis pathways such as mTORC1 are also involved in memory formation. However, very little is known about the translational control pathways involved in imprinting. In this paper we provided gain- and loss- of function evidence that the translational program mediated by eIF2α is crucial for imprinting. The role of other translational pathways in imprinting is currently under investigation by our group.

Major comments:

Figure 2

The use of cycloheximide does not indicate that protein synthesis is required. To show convincingly that protein synthesis is involved, the authors must show that (1) total protein synthesis (or some proteins) is elevated at 2-hour training in auditory imprinting and at 1-hour training in visual imprinting, and (2) training-induced increase in protein synthesis is blocked by cycloheximide. This can be done by metabolic labeling or demonstration that levels of select proteins are increased by Western blot analysis.

Using SUnSET we now show that training induces an increase in protein synthesis (see new Figure 2). Moreover, we also show that this experience-dependent increased in translation is blocked by CHX (Figure 3D)

Figure 3

Total eIF2α seems to be reduced in imprinted samples. The authors should measure other proteins whose levels are not affected by imprinting and normalize total eIF2α to demonstrate that imprinting affects eIF2α phosphorylation and not total levels of eIF2α.

We performed western blots against total eIF2α and normalized to ß-tubulin to address this concern. As shown in Author response image 2, there are no significant differences in the total levels of eIF2α between untrained and trained animals.

Author response image 2.

) total eIF2α levels normalized to ß-tubulin levels show no difference across untrained (n=7) and trained (n=6) animals, b) representative images of western blots performed for quantification.

DOI: http://dx.doi.org/10.7554/eLife.17197.021

The authors should also show that the manipulation of eIF2α phosphorylation directly affects protein synthesis. They may demonstrate this by measuring total protein (or select proteins) levels of MNM.

As requested by the reviewer, we now show that we can bidirectionally manipulate translation rates using either ISRIB or Sal003 (Figure 4—figure supplement 2).

Figure 5

The authors use PKRi and ISRIB to investigate the critical period. It's rather dubious that Sal003 was not used to investigate the critical period. Does Sal003 fail to open the critical period? The authors should use the same manipulation/drugs for

We apologize for the lack of clarity regarding the action of Sal003. We have included a schematic showing that Sal003 inhibits the PP1-GADD34, PP1-Crep complexes. GADD34 and Crep are PP1 cofactors that render the phosphatase specific to eIF2α. Thus, we show that Sal003-treatment increases eIF2α phosphorylation (new Figure 4—figure supplement 1), reduces protein synthesis (new Figure 4—figure supplement 2) and blocks auditory imprinting (new Figure 4C) during the critical period. Therefore, Sal003 was not expected to re-open the critical period.

Reviewer #2:

This manuscript addresses an important issue in the molecular mechanism of imprinted memories during a critical period. The authors argue that the translational control by phosphorylation of the eukaryotic translation-initiation factor 2 α subunit (eIF2 α) is required for auditory imprinting, and mediates changes in dendritic spines in the MNM region. They present data that the increasing phosphorylated eIF2 α prevents the formation of auditory memories and spine plasticity. In contrast, the inhibition of an eIF2 α kinase enhances auditory memories. Furthermore, they provide evidence that the blocking p-eIF2 α mediated translational control enhances auditory memories and reopens the critical period. They propose a modality-specific translational control mechanism underlying the critical period for imprinting.

The paper contains sufficient interest and originality to merit publication, however, the impact is weakened by a technical limitation of the methods used to quantify the preference score and to analyze the statistics. While the findings presented in this study should be of interest to the readers in the field of animal behavior, it is my opinion that a rather substantial revision, based on the technical comments given below, is needed to make this manuscript suitable for publication.

We thank reviewer #2 for the comments. We agree that our computation of the preference index should be more detailed and justified. As stated above, imprinting preference scores are influenced both by novelty and persistence of a preference for the imprinted stimulus. Thus, we designed our index taking both components into account, to avoid biasing the scores by disproportionate differences in the first trials due to novelty responses. The new version of the manuscript elaborates on this issue and compares our index with the ones in previous studies.

Major points

My main concern is that the preference index of the authors (/baseline) differs from that of the conventional method of Horn et al. (imprinting/)(P14 L231). Did they use the background on the screen as the "baseline" image? Compared to the method of Horn et al., this method seems to be affected by the individual difference of animals. I am wondering why the authors used "baseline". The authors should clarify the reason why each score is divided by the baseline. The authors may hypothesize that more active chicks in locomotion should show bigger number of preference score than inactive chicks. But the less active chicks show the highest preference in some cases. If the data are analyzed by the method of Horn et al., it will be easier for most of the readers in the field of imprinting research to understand the significance of this paper.

We clarified and justified our metric for assessing imprinting strength. This index was generated to reflect imprinting across a larger number of trials and obtain a more accurate measure of memory strength. In contrast with previous studies, we did not use fixed presentation sequences and opted to pseudo randomize presentation order. We would like to argue that this paradigm better controls for the possible effect of stimulus presentation order. In addition, a fine trial-by-trial analysis of the locomotor response to novel and imprinted stimuli showed that the preference score captures the influence of both novelty detection and the persistence of increased locomotion towards the imprinted stimulus. Therefore, while classic preference scores are computed over 4 trials, we decided to measure preference across 10 trials. Based on the same analysis, we noticed that baseline (locomotion when only a background image is presented), was the most stable parameter during the experiment. Thus we decided to normalize differences to baseline.

As for the statistical analysis, the authors use either unpaired t-test or Mann-Whitney U test to examine the difference between two groups. The statistical tests were not performed appropriately in some cases. For example, Mann-Whitney U test should be used in the data of

We specially thank the reviewer for this comment. Now, in addition to whether or not the data fit a normal distribution, we also selected the statistical test that was the most appropriate based on differences in variances. When an unpaired t-test was used and statistically significant differences in variances were detected, Welch’s correction was applied as in new Figure 4C and D (3C and D in previous version).

There is controversy whether there is genuine auditory imprinting (Bolhuis, J.J. et al., Behaviour Vol. 122, (1992), pp.195-230). It is generally concluded that auditory stimuli play an important role in the formation of filial preferences, but that auditory imprinting is not as prominent when compared to visual imprinting. Although a lot of studies have demonstrated that the addition of an auditory stimulus improves following of a visual stimulus, few studies using only auditory exposure have demonstrated significant auditory learning. As is the case in this paper, the majority of studies of early auditory learning have used a compound of a visual and an auditory stimulus during training. The authors should discuss the physiological significance of auditory imprinting in relation to visual imprinting in the Introduction section.

The definition of imprinting has evolved to not only comprise the acquired social preference of precocial birds but also a variety of early learning processes occurring during a critical period. Based on this, we defined auditory imprinting as a memory formation process that occurs exclusively during the critical period. However, to address the critiques raised by Bolhuis, J.J. et al., Behaviour Vol. 122, we followed a stringent experimental design that is now described in the second paragraph of the Discussion section.

Figure 2

I suggest the chemicals like cycloheximide should be used by direct injection into the target brain area, not by intraperitoneal injection because cycloheximide is a potent inhibitor of translation and therefore will be toxic to any kind of cells in the whole body.

We are aware of the limitations of systemic injections but we think it is adequate for most of our experiments since we are looking into differences across brain regions and sensory modalities. We reasoned that an approach where drugs access both regions at a similar rate from a unique source facilitates the interpretation of the results. In addition, local injections in MNM and in IMM are relatively close. Thus, spatial control of local injections would be unreliable.

A toxic effect of cycloheximide is unlikely. For example, in new Figure 3D, cycloheximide had no effect on auditory imprinting but specifically inhibited visual imprinting, demonstrating that even when cycloheximide was administered, chicks could perform correctly the visual task and arguing against a toxicity confound.

Figure 3

In western blotting, the time course experiment at different time points (before training, during training, 2-3 h after training, et al.) is necessary to know whether the decrease of p-eIF2α is temporary or not. Is there any difference in the amount of p-eIF2α in the MNM region between 1 day old and 4 day old control chick?

To address the temporal dynamics of eIF2α phosphorylation we performed new western blots from 1- day old and 4-days old brain samples and found no differences across developmental stages (Author response image 3). This result was very insightful since it indicates that the closure of the critical period is not achieved through an increase in eIF2α phosphorylation. It is likely that a mechanism upstream eIF2α is responsible for closing the critical period. Future investigation will address this interesting question.

Author response image 3.

Quantification of p-eIF2α/eIF2α ration between P1 and P4 in MNM (blue, nP1=7, nP4=7) and IMM (green, nP1=7, nP4=7).

No significance was detected in both areas across developmental time points.

DOI: http://dx.doi.org/10.7554/eLife.17197.022

Did the authors examine the decrease of p-eIF2 α was learning dependent or locomotion dependent? The authors may know the answer by doing experiments under the condition of forced running.

We apologize for not making this point clear. All the controls were animals that ran towards a screen containing an empty background image without sound.

As the authors punched out the region of MNM or IMM out of brain slice section, they should mention the exact location in the brain for the extraction in detail in the methods section. This definition of the brain area is very important because either the MNM or IMM region does not have the clear border from surrounding areas. Was the decrease of p-eIF2α detected specifically in the MNM region of the brain?

To further confirm the specificity of eIF2α dephosphorylation we performed western blots of p-eIF2α and eIF2α in the caudolateral nidopalium (NCM), an area involved in auditory learning in songbirds (Bolhuis et al., Eur J Neurosc., 2001). We did not find significant changes in eIF2α phosphorylation after training in this area (Author response image 4).

Author response image 4.

Quantification of p-eIF2α/eIF2α ratio in NCM comparing untrained (n=7) and trained (n=7) chickens.

No significant difference was detected between both areas across treatments.

DOI: http://dx.doi.org/10.7554/eLife.17197.023

Figure 4

I understand the imaging study takes a lot of effort to detect the change of the properties of dendritic spines. I assume the neurons are randomly labeled in the MNM. Therefore, more than 1000 spines at least per each experimental setting should be examined to detect significant morphological changes beyond individual deviation. At the same time, the exact location of brain area they examined and the morphological definition of dendritic spine for classification in chick should be mentioned in detail in the method section

The analysis method of dendritic spines is an important issue that we considered before performing the experiments. One advantage of the ‘diolistic’ technique is that it granted stochastic labeling of neurons both in MNM and IMM. Thus the results were not affected by labeling biases.

We agree that a large number of spines and dendrites should be analyzed to detect biologically meaningful differences between experimental groups. To assess this we carried out a power analysis to determine the minimum sample size per group (ndendrites=42, for statistical power= 0.85) and aimed to a total number of spines per group (average for our study= 673.33) within the range of previous studies in birds (~450-1500 spines, Roberts et al., 2010). In all analyzed groups our sample size is above 42 dendrites.

As mentioned above, we now included a more detailed description of the coordinates of the areas. We also include a more detailed description of the criteria in the methods section and a schematic of the categories in the new Figure 5C.

Reviewer #3:

This manuscript from Batista et al. is beautifully written and addresses the role of translational regulation during critical period plasticity in the chick. The study includes data indicating that auditory imprinting involves translation initiation regulation by eIF2α, and in particular that eIF2α is phosphorylated and inhibited to allow for plasticity in the auditory cortex during the critical period. This indicates that mRNAs containing upstream open reading frames (whose translation is upregulated when eIF2α is phosphorylated and inhibited) are of particular importance to auditory plasticity during the critical period. In contrast, the authors show that phosphorylation of eIF2α, and hence eIF2α translational regulation, is not involved in visual imprinting. The findings are intriguing and I believe important. However, I also believe that there are some significant flaws in the experimental design and interpretation that weaken the study. My principal concerns have to do with the question of how well the translational inhibition worked, and the question of the kinetics of translational regulation in the MNM (for auditory imprinting) and IMM (for visual imprinting). What were the kinetics of inhibition of protein synthesis following IP injection? Was it equivalent in MNM and IMM? It's not clear from the manuscript whether or not the IP injection was done at the end of the 1 hr and the 2 hr training. The authors write: "both auditory and visual imprinting rely on protein synthesis but following different temporal dynamics", and yet the conclusions from the translational inhibition experiments are all based on manipulations that do not appear to consider the differential temporal dynamics of imprinting (because they simply target a single time point, optimized for the effects on auditory imprinting). I also think it's really important to show that there is an effect on translation, and in particular on eIF2α-mediated translation.

We agree with the reviewer that ensuring that drugs have a similar temporal dynamic to alter protein synthesis is crucial for our interpretations. Using SUnSET, we now show that we can alter translation bidirectionally, in both IMM and MNM, with pharmacology (new Figure 4—figure supplement 2).

This is provided for Sal003 in Figure 3—figure supplement 1, but only for "brain samples," not for IMM or MNM. Without this, the results are all based on pharmacological interventions, which always have some caveats unless the expected effects are directly measured. For example it looks like there is significant variability in the ratio of peIF2α to eIF2α in Figure 3—figure supplement 1, which raises the question of whether this correlates in any way with plasticity during imprinting.

There is some intrinsic variability in p-eIF2α levels that cannot be correlated with performance, so far, because to carry out such experiments, quantification of p-eIF2α needs to be done before testing the animals. in vivomeasures of p-eIF2α, which are required to overcome this difficulty, are not yet available.

[Editors’ note: the author responses to the re-review follow.]

[…] The reviewers have several queries, and ask for further clarification on the following:

Reviewer 1 asks for you to explain why protein synthesis dependent visual plasticity at 1 hour of training was ignored and the focus placed on 2 hours; an explanation would help the reader move to the logic of the next set of experiments.

We decided to use 2-hour training to evaluate experience-dependent changes in translation because it allowed us to assess protein synthesis across brain regions in the same animal. Using 1-hour training did not induce reliable auditory imprinting. The interesting difference in time course is conveyed to readers in the Results section ‘Protein-synthesis dependency of auditory and visual imprinting (Figure 3A, B). We now further clarify the motivation for the 2-hour training in the methods as follows:

“To simultaneously assess experience-dependent translation across sensory modalities and brain regions, in the same animal, we identified a training schedule that reliably triggered auditory and visual imprinting. Since 2-hour but not 1-hour training (Figure 3A, B) triggered both auditory and visual imprinting, we used the former schedule.”

In addition, this reviewer wanted an explanation of the relevance of structural spine changes over density in the context of behavior.

We thank Reviewer 1 for suggesting references, which we now cite. To elaborate on the possible relevance of the observed structural plasticity on behavior and memory formation, we added the following paragraph to the Discussion:

“Different forms of structural plasticity have been linked to memory, including spine turnover and morphological changes of preexisting spines (32). In this case, the structural plasticity found in IMM and MNM could be consistent with potentiation and enlargement of specific dendritic spines, favoring the detection of imprinted stimuli. While we did not observe changes in spine density, the increase in mushroom spines and decrease in thin spines may suggest coordinated structural plasticity as previously reported in hippocampal slices (42)”

Reviewer 2 wondered why the first trials of the novel stimulus strongly suppressed locomotion; this does not seem to occur in the reviewer's hands in their experimental design. This reviewer is concerned about whether the training for imprinting by the authors is sufficient to cause strong preference compared with controls.

We thank Reviewer 2 for sharing personal experience.

We chose the sequential scheme in a running wheel because it allowed us to: 1- randomize the stimulus order, 2- measure baseline locomotion, 3- assess independently the response to novelty and to the imprinted stimulus. We have not been the first using this training schedule and reporting suppression by novel stimuli. Initial suppression of locomotion by the novel stimulus using a similar setup to ours has been reported (Figure 1C; Maekawa et al., 2006). In addition, Maekawa et al. (2006) showed that aversion to novelty emerged after training. Thus behavioral suppression by novel stimuli is not inconsistent with innate attraction to conspicuous moving objects. To contextualize our methodology we now highlight in the methods our use of a sequential test and acknowledge the simultaneous choice test, used in other studies:

“Visual and auditory imprinting were tested independently in a sequential test, where novel and imprinted stimuli are presented in alternation. While other studies have used a simultaneous choice test (23), the sequential test allowed us to randomize stimulus presentation, measure baseline locomotion and assess the response to novel and imprinted stimuli independently.”

Regarding the reviewer’s concern about whether training for imprinting is sufficient to induce strong preference, we would like to argue that the response to the imprinted stimulus was substantially above baseline across trials. We enclose a figure from the previous rebuttal letter supporting this claim (Author response image 1). The magnitude of imprinting induced by our training provided us a dynamic range to perform both loss- and gain-of-function manipulations.