Abstract

   Under physiological conditions, strength and persistence of memory must
   be regulated in order to produce behavioral flexibility. In fact,
   impairments in memory flexibility are associated with pathologies such
   as post-traumatic stress disorder or autism; however, the underlying
   mechanisms that enable memory flexibility are still poorly understood.
   Here, we identify transcriptional repressor Wilm’s Tumor 1 (WT1) as a
   critical synaptic plasticity regulator that decreases memory strength,
   promoting memory flexibility. WT1 is activated in the hippocampus
   following induction of long-term potentiation (LTP) or learning. WT1
   knockdown enhances CA1 neuronal excitability, LTP and long-term memory
   whereas its overexpression weakens memory retention. Moreover,
   forebrain WT1-deficient mice show deficits in both reversal, sequential
   learning tasks and contextual fear extinction, exhibiting impaired
   memory flexibility. We conclude that WT1 limits memory strength or
   promotes memory weakening, thus enabling memory flexibility, a process
   that is critical for learning from new experiences.

   Subject terms: Forgetting, Hippocampus
     __________________________________________________________________

   Impairments in memory flexibility are associated with neuropsychiatric
   disorders such as PTSD and autism. Here, the authors report that the
   transcriptional repressor Wilm's Tumor 1 regulates synaptic plasticity
   leading to weakening of memory strength and enabling memory
   flexibility.

Introduction

   Learning produces long-term memory retention and storage by activating
   molecular mechanisms that consolidate and strengthen an initially
   labile experience representation. The process of memory strengthening
   must be regulated in order to remain within the physiological ranges;
   excessively weak or excessively strong memories are in fact maladaptive
   and pathological. Weak memories can result from impairments in any of
   several different processes—storage, retrieval, or consolidation (the
   stabilization process that forms long-term memories) or by an
   overactive forgetting process^[58]1–[59]3. All these processes likely
   play important roles in memory disorders, in Alzheimer’s disease,
   aging-related memory loss, and neurodevelopmental cognitive
   impairments. Conversely, an excessive memory consolidation, and/or
   impaired forgetting may produce excessively strong and inflexible
   memories, possibly leading to diseases such as posttraumatic stress
   disorder (PTSD), autism spectrum disorder, schizophrenia, and obsessive
   compulsive disorder (OCD). Therefore, the ability to regulate the
   intensity of memory consolidation and strengthening is of great
   importance for adaptive behaviors and mental health.

   The biological mechanisms required for promoting memory consolidation
   and strengthening have been investigated in many species and types of
   memory, identifying roles for a variety of signaling
   networks^[60]4,[61]5, transcription factors^[62]6,[63]7, and epigenetic
   changes^[64]8. However, little is known about mechanisms that reduce
   memory consolidation and strengthening in order to enable behavioral
   flexibility. A key question is whether consolidated memories are
   weakened through a passive decay process, and/or by a learning-induced,
   active mechanisms that serves to promote memory flexibility. In other
   words, do signaling pathways that are activated during experience not
   only support consolidation, but also include counteracting molecular
   regulators that can decrease memory strength and favor
   forgetting^[65]3, such as the Rho family of GTPases signaling G
   proteins (Rac)^[66]9,[67]10, scribble scaffolds^[68]11, DAMB dopamine
   receptors^[69]12, inhibition of AMPA receptor recycling^[70]13, and
   neurogenesis^[71]14?

   We therefore tested the hypothesis that memory flexibility results from
   an active process that occurs in parallel with memory consolidation and
   strengthening. If this is the case, then mechanisms enabling memory
   flexibility should be activated and/or induced by learning.

   Memory consolidation engages complex regulation of genes transcription
   activation and repression^[72]4. Whereas the role of transcription
   activators, such as members of the CREB, C/EBP, AP1, NFkB, Rel, Egr 1
   and 2, and Nurr families have been more extensively documented as
   promoters of memory consolidation and
   strengthening^[73]2,[74]5,[75]7,[76]15–[77]18, less is known about the
   role of transcription repressors^[78]4,[79]19,[80]20. A few
   transcription repressors that directly bind to promoter/enhancer DNA
   sequences in memory formation have been documented: CREB^[81]21,[82]22,
   MeCP2^[83]23, DREAM (downstream regulatory element antagonistic
   modulator)^[84]24, myocyte enhancer factor-2 (MEF2)^[85]25. The
   literature thus far suggests that induction of transcription activation
   correlates with memory strengthening, whereas induction of
   transcription repression correlates with memory weakening or
   forgetting^[86]3,[87]4,[88]19,[89]20.

   To search for transcription repressors of plasticity, we screened for
   transcription factors activated by induction of LTP at hippocampal
   excitatory synapses, a cellular model of learning and memory^[90]26. We
   identified Wilm’s tumor 1 (WT1), a protein that is important for kidney
   and gonads development^[91]27. WT1 is a form of kidney cancer that
   primarily affects children ages 3–4. Interestingly one of the health
   conditions due to Wt1 germline mutations is the WAGR syndrome, a
   disorder characterized by Wilm’s Tumor (W), aniridia (A), genitourinary
   anomalies (G) and mental retardation (R). Patients with WAGR syndrome
   have difficulties in learning, processing, and responding to
   information; they may develop behavioral and cognitive abnormalities
   such as anxiety, OCD, depression, attention deficit hyperactivity
   disorder, and autism^[92]28. While Wt1 gene has been well characterized
   for its role in kidney development and function, its role in the brain
   is not fully understood. WT1 has been linked to neurodegeneration
   associated with Alzheimer disease^[93]29, and a recent study has showed
   that during early neuronal development its transcriptional activity is
   repressed to allow normal neuronal differentiation^[94]30. In this
   study, we use different types of genetic and molecular manipulation to
   investigate the functional role of WT1 in memory consolidation and
   strengthening and its ability to regulate memory flexibility and new
   learning. We also investigate WT1 hippocampal physiology by examining
   the effects of ablating WT1 on pyramidal cell excitability, synaptic
   plasticity, and regulation of entorhinal cortex-hippocampus circuitry.
   In addition, we identify numerous transcriptional targets of WT1 in the
   hippocampus and functionally characterize one of these genes in
   plasticity experiments. Our data indicate that the transcriptional
   repressor WT1 is a key regulator of synaptic plasticity, memory
   strength, and memory flexibility in the hippocampus.

Results

Learning-induced WT1 decreases memory strength

   To identify transcription factors activated or induced by long-term
   plasticity, we employed a protein-DNA binding array on rat hippocampal
   slices in which long-term potentiation (LTP) was induced by strong
   high-frequency stimulation (Strong-HFS) of the Schaffer
   collaterals^[95]26. We identified nearly 40 transcription factors whose
   binding was increased (Fig. [96]1a and Supplementary Fig. [97]1a). One
   of these transcription factors, WT1, is a transcriptional repressor
   shown to be involved in regulating kidney development^[98]27 and in
   mRNA transport and translation in several cell lines^[99]31,[100]32.

Fig. 1.

   [101]Fig. 1
   [102]Open in a new tab

   WT1 expression and DNA-binding activity are induced by synaptic
   plasticity and learning. a Protein–DNA binding assay comparing rat
   hippocampal CA1 extracts from control tissue versus extracts obtained
   from tissue where LTP was induced. WT1 is circled in red; numbers in
   parentheses indicate two different DNA probes with WT1 consensus sites.
   b EMSA showing increased in vitro WT1 binding to a DNA consensus
   sequence (arrow indicates the WT1/DNA complex) 10 and 30 min after
   induction of LTP in hippocampal CA1 region (Stim) compared with
   unstimulated control (C). The specificity of DNA–protein binding was
   verified by incubation with excess unlabeled cold probe (CP). c EMSA
   showing increased WT1 binding to DNA (arrow indicates the WT1/DNA
   complex) at different time points after CFC (S, shocked group; C,
   context only controls). The specificity of DNA-protein binding was
   verified by incubation with excess unlabeled cold probe (CP). d Bar
   graph of the top ten transcription factors predicted to regulate gene
   expression profiles in rat tissue obtained 90 min after a stimulation
   that produced LTP. e Expression of WT1 was significantly increased in
   rat CA1 region 30 min after LTP induction (paired t test: *p = 0.0495).
   f WT1 expression in the dorsal hippocampus of rats trained in CFC
   (Paired) compared with non shocked rats (Ctx only) (unpaired t test:
   *p = 0.0385). g Expression of WT1 was significantly increased in the
   dorsal hippocampus of rats trained in an IA task. Protein expression
   was measured 30 min after training and compared with naïve rats
   (unpaired t test: *p = 0.0187). Data are expressed as mean ± s.e.m

   Strong-HFS as well as contextual fear conditioning (CFC) learning
   increased the binding of WT1 to its DNA consensus sequence in the
   hippocampus of rats (Fig. [103]1b, c), providing functional evidence
   for an active involvement of WT1 in these functions. Furthermore, we
   found independent evidence for WT1 activation in mRNA-seq experiments
   that identified increased expression of transcripts 90 min after LTP
   induction. Enrichment analysis of this transcriptomic data (see
   Supplementary Data [104]1 for complete list of differentially expressed
   transcripts) predicted WT1 as the second most likely candidate to
   regulate LTP-induced gene expression followed by members of the CREB
   family (ATF2 and ATF4) (Fig. [105]1d; see Supplementary Data [106]2 for
   predicted transcription factors analysis).

   In addition both LTP induction—but not LTD—(LTP, Fig. [107]1e; LTD,
   Supplementary Fig. [108]2a) as well as contextual fear learning in two
   independent tasks, contextual fear conditioning (CFC, Fig. [109]1f;
   behavioral data shown in Supplementary Fig. [110]2b) and inhibitory
   avoidance (IA, Fig. [111]1g; for behavioral data see reference
   Taubenfeld et al.^[112]18), resulted in significant increases in the
   expression levels of WT1 protein within 30 min. These data suggest that
   induction of WT1 is due to the learning process and not to the
   presentation of the aversive stimulus itself as we did not observe any
   significant change in WT1 expression using an unpaired CFC protocol
   (Supplementary Fig. [113]2c, d).

   To determine the functional role of WT1 in memory formation, we
   knockdown WT1 protein expression using bilateral injections of
   antisense oligodeoxynucleotides (WT1-AS) into rat dorsal hippocampus
   (Fig. [114]2a) and tested the effect on memory retention using two
   different hippocampal tasks, one aversive (CFC) and one nonaversive
   (novel object location, NOL). As shown, WT1-AS compared with control
   scrambled oligodeoxynucleotides (SC-ODN) significantly decreased WT1
   protein levels in dorsal hippocampus and resulted in a significantly
   enhanced CFC memory retention 24 h after training (Fig. [115]2a). Rats
   injected with either WT1-AS or SC-ODN did not differ in locomotor
   activity suggesting that the significant difference in CFC freezing was
   not due to mobility alteration (Supplementary Fig. [116]3a). Similar
   results were obtained with NOL, as rats injected with WT1-AS exhibited
   increased memory at 24 h after training (Fig. [117]2a). Furthermore
   WT1-AS injected rats showed short-term memory retention, at 1 h after
   training, comparable to SC-ODN injected controls (Fig. [118]2a),
   indicating that WT1 in the hippocampus selectively affects long-term
   memory. The WT1-AS or SC-ODN groups did not exhibit any difference in
   total object exploration time (Supplementary Fig. [119]3b). These
   findings, based on two distinct hippocampus-dependent tasks, suggest
   that WT1, whose expression and DNA binding activity increase following
   training, decreases memory retention.

Fig. 2.

   [120]Fig. 2
   [121]Open in a new tab

   WT1 represses long-term memory consolidation. a Left: Change in WT1
   expression after double injection (2 nmoles/each, 2 h apart) of WT1-AS
   into CA1 (paired t test: *p = 0.0362; t = 2.687, df = 6). Right: Scheme
   of behavioral experiments in WT1 knockdown rats. WT1-AS-injected rats
   increased freezing time 24 h after training in CFC (unpaired t test
   **p = 0.0086). WT1 acute knockdown did not affect memory retention in
   NOL 1 h after training (unpaired t test p = 0.1685, n = 8–12 rats). In
   contrast, 24 h after training, WT1-AS-injected rats showed better
   memory than SC-ODN injected ones (unpaired t test: **p = 0.0011. Dashed
   line indicates 50% preference). b Left: Wt1∆ mice showed enhanced
   freezing 24 h and 30 days after training in CFC (unpaired t test: for
   24 h, **p = 0.0088; for 30 days, *p = 0.0104). Right: Both Control and
   Wt1∆ groups showed preference for the new location when tested 1 h
   after training in NOL while only Wt1∆ mice showed significant
   preference for the new location 24 h after training (unpaired t test:
   **p = 0.0033; n = 8–10 rats. Dashed line indicates 50% preference). c
   Top: Immunostaining and immunoblot showing WT1 overexpression in rats.
   Bottom: Scheme of the behavioral experiments: green highlight line
   indicates time window for AAV-induced full expression of WT1. WT1
   Overexpression reduced levels of freezing both during acquisition
   (unpaired t test **p = 0.0061) and 7 days after training (unpaired t
   test ***p = 0.0004) compared with CTR-AAV controls. d Top:
   Immunostaining and immunoblot showing WT1 overexpression via HSV virus
   in rats. Bottom: Scheme of pre- and post-training behavioral
   experiments. Green highlight line indicates time window for HSV-induced
   full expression of WT1. Pre-training: WT1-HSV group showed a
   significant difference in freezing compared with CTR-HSV group both
   during acquisition (unpaired t test **p = 0.0042) and 7 days after
   training (unpaired t test: **p = 0.0049). Post-training: rats were
   tested 4 days after HSV injection. WT1-HSV group showed significantly
   reduced levels of freezing compared with CTR-HSV group (unpaired t
   test: *p = 0.0444). Data are expressed as mean ± s.e.m

   To extend the investigation of the role of WT1 on synaptic plasticity
   and memory to different species, we generated genetically modified mice
   with forebrain expression of an in-frame internal Wt1 deletion, which
   produces a truncated WT1 protein that lacks zinc fingers 2 and 3
   (Wt1^fl/fl; Camk2a-Cre mice, referred thereafter as Wt1Δ mice,
   Supplementary Fig. [122]4a). These protein domains are essential for
   WT1 DNA and RNA binding activity^[123]33.

   Wt1Δ mice were viable, of normal size and weight, and did not show any
   gross alteration in hippocampal morphology compared with wild type
   littermates (referred thereafter as Control mice; Supplementary
   Fig. [124]4b). The transgenic mice also were similar to Control mice
   with respect to protein levels in peripheral tissue, as well as in
   their urine and blood chemistry (metabolic enzyme and electrolyte
   panel; Supplementary Fig. [125]4c, d).

   Similar to rats in which WT1 was knocked down in the hippocampus, Wt1Δ
   mice compared with Control mice showed enhanced memory retention 24 h
   as well as 30 days after CFC training (Fig. [126]2b). They also showed
   enhancement in NOL retention 24 h, but not 1 h following training
   (Fig. [127]2b). The open field activity, pain response and total object
   exploration time of Wt1Δ mice were similar to those of Control mice
   (Supplementary Fig. [128]5a–c), indicating that the effect of the
   genotype on NOL and CFC were not due to changes in locomotion, pain
   sensitivity, or exploration. In contrast, when tested in the elevated
   plus maze, a paradigm used to measure anxiety-like behavior, Wt1Δ mice
   spent significantly more time in the closed arm and made a significant
   lower number of entries in the open arm (Supplementary Fig. [129]5d),
   compared with Control mice, suggesting that forebrain deletion of WT1
   may affect also anxiety behavior regulation.

   Collectively these results indicate that the expression and functional
   activation of the transcriptional repressor WT1 is increased in the
   hippocampus by learning and that WT1 acts to suppress memory.

   To test WT1 function as a memory suppressor, we overexpressed wild type
   WT1 using either an AAV or HSV virus injected into the dorsal
   hippocampus of rats (WT1-AAV or WT1-HSV; Fig. [130]2c, d respectively).
   AAV-GFP or HSV-GFP viruses were used as controls (CTR-AAV or CTR-HSV).
   Rats bilaterally injected into the hippocampus with either viruses were
   trained in CFC either 4 weeks (AAV) or 3 days (HSV) after infection,
   times that correspond to the respective peaks of viral expression for
   the two viruses. As shown in Fig. [131]2c, d (pre-training) both
   WT1-AAV- and WT1-HSV-injected rats had a significantly decreased memory
   retention 7 days after CFC training. These data indicate that WT1
   overexpression is sufficient for reducing memory retention. Notably,
   because overexpression of WT1 significantly reduced the acquisition of
   the task (Fig. [132]2c, d), we tested the effect of viral injection
   following training (post-training). As shown in Fig. [133]2d
   (post-training), WT1-HSV compared with control virus decreased memory
   retention tested 7 days after training. Overall our data indicate that
   overexpression of WT1 is associated with decreased memory retention of
   an aversive memory.

WT1 controls excitability of hippocampal CA1 neurons

   Immunohistochemical staining of rat and mouse hippocampus obtained from
   naïve animals revealed that WT1 is predominantly localized within the
   nuclei of pyramidal neurons with a weaker immunoreactivity in the
   proximal apical dendrites. WT1 immunoreactivity was not detected in
   astrocytes marked by glial fibrillary associated protein
   (GFAP-positive) (Fig. [134]3a–c).

Fig. 3.

   [135]Fig. 3
   [136]Open in a new tab

   WT1 effect is mediated by enhanced activity and excitability of CA1
   neurons. a In the mouse hippocampus WT1 localizes predominantly within
   the cell bodies layer. Scale bar = 500 μm. b Immunostaining of the
   mouse CA1 region shows WT1 expression mainly in cell bodies but also in
   proximal dendrites. Scale bar = 50 μm. c In the rat CA1 region WT1 is
   expressed in pyramidal neurons and not in GFAP positive astrocytes.
   Scale bar = 50 μm. d Scheme for the electrophysiology experiments.
   Reduction in WT1 expression after a single intrahippocampal injection
   of WT1-AS (paired t test: *p = 0.0202). A weak stimulus (delivered at
   arrow) induced LTP in WT1-AS group (two-way ANOVA RM:
   F[(1,12)] = 10.58, **p = 0.0069). Calibrations: 0.5 mV/10 ms. e,
   Bicuculline (10 μM) did not block WT1-AS-mediated LTP enhancement
   (two-way ANOVA RM: F[(1,9)] = 6.039, *p = 0.0363). Calibrations:
   0.5 mV/10 ms. f Whole-cell patch recordings in rats CA1 pyramidal
   neurons. Right inset: probability of evoking at least one spike in
   response to a weak (20–50 pA) or a stronger (60–90 pA) current step in
   WT1-depleted or control groups (two-tailed Chi-square test,
   **p = 0.0041). Resting membrane potential and input resistance measured
   −63.75 ± 3.15 mV and 105.8 ± 21.76 MΩ in the WT1-AS group, and
   −60.80 ± 2.85 mV and 109.3 ± 20.04 MΩ in the SC-ODN group. Left inset:
   representative traces in cells from WT1-AS or SC-ODN. Calibration:
   50 mV/100 ms. g Upon weak stimulus (delivered at arrow) LTP was induced
   in Wt1∆ mice but not in control group (two-way ANOVA RM:
   F[(1,23)] = 5.125, *p = 0.0333). Calibrations: 0.5 mV/10 ms. h Wt1∆
   mice showed increased basal synaptic efficiency (left panel:
   input/output; linear regression unpaired t test, **p = 0.0077) but did
   not affect paired-pulse ratio (right panel; two-way ANOVA RM,
   p = 0.0878). Representative fEPSPs graphs show traces recorded during
   baseline and 60 min post-HFS. Dot blot graphs display final 10 min of
   fEPSP slope. Data are expressed as mean ± s.e.m

   Given that the effect of decreasing WT1 expression in the hippocampus
   enhances memory retention, here we tested whether WT1 knockdown affects
   hippocampal LTP induction and/or maintenance. Western blot analyses
   showed that single intrahippocampal injection of WT1-AS significantly
   decreased WT1 levels in hippocampal slices compared with control slices
   injected with SC-ODN (Fig. [137]3d). This WT1 knockdown did not alter
   basal synaptic transmission (Supplementary Fig. [138]6a), nor did it
   affect the induction or maintenance of LTP elicited by Strong-HFS
   (Supplementary Fig. [139]6b). However, a role for WT1 in synaptic
   plasticity emerged at synapses activated with a weak high-frequency
   stimulation (Weak-HFS) protocol, which produced decremental
   potentiation in control slices but stable LTP in slices from animals
   injected with WT1-AS (Fig. [140]3d).

   To assess whether WT1 knockdown might enhance LTP indirectly through an
   effect on interneuron function^[141]34, we stimulated slices with
   Weak-HFS in the presence of the GABA[A]- receptor antagonist
   bicuculline. Under these conditions, hippocampal slices from WT1
   knockdown rats still showed enhanced LTP, indicating that WT1 likely
   regulates synaptic plasticity through direct effects on pyramidal
   neurons (Fig. [142]3e).

   We therefore hypothesized that WT1 knockdown might enhance LTP by
   increasing pyramidal cell excitability, since postsynaptic spiking
   during stimulation facilitates LTP induction^[143]35. To test this
   hypothesis, whole-cell recordings were obtained from pyramidal neurons
   in area CA1 of rat hippocampus. In recordings from WT1-AS injected
   hippocampi, weak depolarizing currents (20–50 pA) were more likely to
   evoke action potentials than in neurons of scrambled ODN-injected
   hippocampi (Fig. [144]3f) indicating that WT1 knockdown increased
   excitability. In contrast, in response to relatively strong
   depolarizing currents (70–100 pA), neurons from WT1-AS slices fired
   significantly fewer action potentials than those treated with scrambled
   ODN (mean number of spikes = 2.1 ± 0.173 and 1.375 ± 0.125,
   respectively; unpaired t-test: *p = 0.0146; t = 3.394, df = 6). No
   significant differences were observed in the amplitude, frequency or
   inter-event interval in both spontaneous and mEPSCs (Supplementary
   Fig. [145]7a, b).

   In agreement with these data in rat hippocampus, slices from Wt1Δ mice
   also showed sustained hippocampal LTP following Weak-HFS, while slices
   from Control mice produced only transient potentiation (Fig. [146]3g).
   When compared with their control littermates, Wt1Δ mice showed
   increased basal Schaffer collateral—CA1 synaptic efficiency with no
   difference in paired-pulse ratio (Fig. [147]3h), indicating that WT1
   regulates synaptic efficiency through a postsynaptic mechanism.

   Collectively, these results suggest that WT1 acts as a synaptic
   plasticity repressor that dampens the postsynaptic response to a weak
   stimulus, while preserving the normal dynamic range of the response to
   super threshold stimuli.

WT1 regulates the computational properties of CA1 cells

   The role of the CA1 region in memory processing involves the
   circuit-level integration of information arriving from the entorhinal
   cortex via two major inputs: (1) the direct temporoammonic (TA)
   pathway, in which entorhinal neurons of the perforant path synapse on
   distal apical dendrites of CA1 pyramidal neurons, and (2) an indirect
   input, in which entorhinal activity provides phase-delayed information
   to proximal apical dendrites in CA1 through a series of three synapses:
   perforant path→dentate gyrus, mossy fibers→CA3, and Schaffer
   collaterals (SC)→CA1. The CA1 pyramidal neuron functions as a
   coincident detector, integrating these temporally segregated streams of
   information from cortical activity^[148]36. This coincidence detection
   function can be studied in hippocampal slices, where the two inputs are
   activated independently (Fig. [149]4a; wild type animal)^[150]37. We
   reasoned that WT1 levels could regulate the need for convergent
   activity of both inputs to induce LTP at the Schaffer collateral—CA1
   synapse. Depletion of WT1 might allow SC stimulation alone to induce
   LTP without the added information provided by the TA input
   (Fig. [151]4a; WT1 knockdown animal). We tested this hypothesis by
   stimulating CA1 with theta-burst stimulation (TBS) at both the TA and
   SC inputs, with SC stimulation phase-delayed relative to TA. In
   hippocampi from control rats injected with scrambled ODN, induction of
   LTP required activation of both inputs (Fig. [152]4b). However, the TA
   input became dispensable in WT1-depleted hippocampus, so that SC
   stimulation alone was as effective in producing LTP as dual pathway
   stimulation (Fig. [153]4b). Thus, the “normal” level of WT1 imposes a
   requirement for circuit-level computation in the CA1 neuron, leading to
   LTP. In contrast, in WT1-depleted hippocampus circuit-level computation
   no longer is necessary: SC→CA1 activity can induce LTP without
   confirmatory input from TA→CA1. Combined with our findings of increased
   pyramidal cell excitability and altered spike encoding of
   depolarization in WT1-depleted hippocampus, this result indicates that
   WT1 activity plays an important role in determining the computational
   properties of CA1 pyramidal cells.

Fig. 4.

   [154]Fig. 4
   [155]Open in a new tab

   Circuit mechanism of WT1 action. a Scheme of WT1 depletion effect on
   corticohippocampal input to CA1. Left panel (wild type animal):
   normally, activation of both the direct temporoammonic pathway (blue)
   and the trisynaptic pathway (green) are required for LTP induction at
   the Schaffer collateral (SC) → CA1 synapse. Right panel (WT1 knock-down
   animal): in WT1-depleted hippocampus, enhanced basal efficiency of SC →
   CA1 signaling and/or CA1 excitability enable trisynaptic pathway
   activity alone to induce LTP. EC = entorhinal cortex; DG = dentate
   gyrus; TA = temporoammonic pathway. b Theta burst stimulation (TBS,
   delivered at arrow) of the SC induced stable LTP in slices from rats
   injected with SC-ODN only when combined with phase-delayed TBS at the
   TA pathway (left and right panels). Conversely, in slices from
   WT1-AS-injected hippocampi, the same TBS of SC alone induced LTP, which
   did not differ from that induced by dual-pathway TBS (center and right
   panels). Representative fEPSPs show superimposed traces recorded during
   baseline and 60 min post-TBS. Calibrations: 0.5 mV/10 ms. Data are
   expressed as mean fEPSP ± s.e.m. Statistical analysis by two-way ANOVA
   RM: SC stimulation comparing SC-ODN vs WT1-AS ODNs: F[(1,10)] = 6.931,
   *p = 0.0250; SC-ODN comparing SC stimulation vs SC + TA:
   F[(1,9)] = 7.112, *p = 0.0258. No significant effect was observed when
   comparing WT1-AS SC vs WT1-AS SC + TA: F[(1,10)] = 1.437, p = 0.2582

WT1 downstream targets genes

   To identify the target genes of WT1 in hippocampal synaptic plasticity
   and memory, we compared mRNA-seq profiles of Wt1∆ and Control mice. We
   identified 193 differentially expressed transcripts (Table [156]1 and
   Supplementary Data [157]3).

Table 1.

   Top 40 differentially expressed genes whose mRNA expression was
   regulated in the Wt1∆ mice. Numbers indicate the log[2]-fold change for
   each gene comparing Wt1∆ mice with wild type littermates. For the list
   of all differentially expressed genes, their gene symbols as well as
   their extended names see Supplementary Data [158]3
   Sample name NCBI official symbol NCBI gene description log2(fold
   change)
   Wt1∆ Ttr Transthyretin 3.755
   Wt1∆ Eif3j1 Eukaryotic translation initiation factor 3, subunit J1
   3.323
   Wt1∆ Folr1 Folate receptor 1 (adult) 2.907
   Wt1∆ Slc4a5 Solute carrier family 4, sodium bicarbonate cotransporter,
   member 5 2.786
   Wt1∆ 2900040c04rik RIKEN cDNA 2900040C04 gene 2.744
   Wt1∆ Kcne2 Potassium voltage-gated channel, Isk-related subfamily, gene
   2 2.247
   Wt1∆ 1500015o10rik RIKEN cDNA 1500015O10 gene 2.172
   Wt1∆ Cldn2 Claudin 2 2.128
   Wt1∆ Otx2 Orthodenticle homeobox 2 2.059
   Wt1∆ Clic6 Chloride intracellular channel 6 1.931
   Wt1∆ Calml4 Calmodulin-like 4 1.792
   Wt1∆ Hba-A1 Hemoglobin alpha, adult chain 1 1.680
   Wt1∆ Prlr Prolactin receptor 1.642
   Wt1∆ Ccl28 Chemokine (C–C motif) ligand 28 1.640
   Wt1∆ Eps8l1 EPS8-like 1 1.617
   Wt1∆ Igf2 Insulin-like growth factor 2 1.594
   Wt1∆ Wdr86 WD repeat domain 86 1.560
   Wt1∆ Drc7 Dynein regulatory complex subunit 7 1.557
   Wt1∆ Aqp1 Aquaporin 1 1.532
   Wt1∆ Kcnj13 Potassium inwardly-rectifying channel, subfamily J, member
   13 1.507
   Wt1∆ Enpp2 Ectonucleotide pyrophosphatase/phosphodiesterase 2 1.484
   Wt1∆ Gdf1 Growth differentiation factor 1 1.479
   Wt1∆ 4833420g17rik RIKEN cDNA 4833420G17 gene 1.439
   Wt1∆ Tmem72 Transmembrane protein 72 1.434
   Wt1∆ Abca4 ATP-binding cassette, sub-family A (ABC1), member 4 1.419
   Wt1∆ Col8a2 Collagen, type VIII, alpha 2 1.398
   Wt1∆ Rdh5 Retinol dehydrogenase 5 1.372
   Wt1∆ Sema3b Sema domain, immunoglobulin domain (Ig), short basic
   domain, secreted, (semaphorin) 3B 1.358
   Wt1∆ Trpv4 Transient receptor potential cation channel, subfamily V,
   member 4 1.298
   Wt1∆ Tcea3 Transcription elongation factor A (SII), 3 1.249
   Wt1∆ Sulf1 Sulfatase 1 1.244
   Wt1∆ Wfdc2 WAP four-disulfide core domain 2 1.236
   Wt1∆ Sostdc1 Sclerostin domain containing 1 1.235
   Wt1∆ Ace Angiotensin I converting enzyme (peptidyl-dipeptidase A) 1
   1.214
   Wt1∆ Gbgt1 Globoside alpha-1,3-N-acetylgalactosaminyltransferase 1
   1.213
   Wt1∆ Kl Klotho 1.201
   Wt1∆ Slc6a12 Solute carrier family 6 (neurotransmitter transporter,
   betaine/GABA), member 12 1.167
   Wt1∆ Spp1 Secreted phosphoprotein 1 1.158
   Wt1∆ Lbp Lipopolysaccharide binding protein 1.145
   Wt1∆ Igfbp2 Insulin-like growth factor binding protein 2 1.123
   [159]Open in a new tab

   While transcripts encoding for plasticity and memory-related immediate
   early genes, such as the activity-regulated cytoskeletal-associated
   protein (Arc) and the FBJ osteosarcoma oncogene (Fos), were
   significantly downregulated, we found that several genes belonging to
   the retinoic acid signaling pathway were instead upregulated. These
   include retinol dehydrogenase 5 (Rdh5), cellular retinoic acid binding
   protein 2 (Crabp2), aldehyde dehydrogenase family 1, subfamily A2
   [(Aldh1a2; also known as retinaldehyde dehydrogenase 2 (Raldh2)] (see
   Table [160]1 and Supplementary Data [161]3 for complete list).
   Interestingly, another upregulated transcript encodes transthyretin
   (TTR), a protein that is involved with transport of retinol in the
   plasma and which plays an important role in neuroprotection^[162]38 as
   well as memory consolidation and neurogenesis in the
   hippocampus^[163]39,[164]40. Furthermore, TTR has also been shown to
   upregulate hippocampal expression of insulin-like growth factor
   receptor I (IGF-IR) and its nuclear translocation^[165]41. Notably we
   found that Igf2 was ranked sixteenth in our list (and the insulin-like
   growth factor binding protein 2, known as IGFBP2, was ranked fortieth),
   as one of the top differentially regulated genes. This is in agreement
   with previous literature on kidney and cell lines (human fetal kidney
   and HepG2 cells) reporting that WT1 suppresses the expression of
   Igf2^[166]42,[167]43.

IGF-2 can mediate WT1 effects on synaptic plasticity

   In the brain, IGF-2 is required for long-term memory consolidation in
   the hippocampus, and it has been shown that administration of
   recombinant IGF-2 significantly enhances memory as well as
   LTP^[168]44,[169]45.

   Using quantitative real time RT-PCR, we confirmed that acute WT1
   knockdown using WT1-AS significantly increased IGF-2 mRNA expression in
   dorsal hippocampus (Fig. [170]5a). Based on this finding, we examined
   whether Igf2 mediates the enhanced synaptic plasticity produced by
   WT1-depletion. In hippocampal slices, application of an IGF2
   receptor-blocking antibody significantly inhibited LTP enhancement in
   WT1-deficient mice and rats (Fig. [171]5b, c) consistent with the
   hypothesis that, similarly to the kidney, Igf2 is one of the key
   downstream targets of the transcriptional repressor WT1. Thus, we
   conclude that the effects on plasticity observed when WT1 is knocked
   down or ablated rely on derepression of the Igf2 gene.

Fig. 5.

   [172]Fig. 5
   [173]Open in a new tab

   WT1 effects on hippocampal plasticity are mediated via IGF2. a
   Quantitative real time PCR showed that WT1 acute knockdown in rats
   significantly increases IGF2 mRNA expression (unpaired t test
   *p = 0.0260). b LTP induced by weak-HFS in Wt1∆ slices was abolished by
   bath application of IGF2 receptor antibody (IGF2-R Ab, 5μg/ml).
   Superimposed traces showing representative fEPSPs recorded during
   baseline and 60 min post-HFS. Calibrations: 0.5 mV/10 ms. Summary of
   the final 10 min of recording showed that LTP in hippocampal slices
   from Wt1∆ mice was significantly reduced by bath application of IGF2-R
   Ab (two-way ANOVA RM, *p = 0.0430, F[(1, 13)] = 5.03). For ease of
   comparison data for the Control and the Wt1∆ group in the bar graph are
   the same as in Fig. [174]3g. c WT1-AS-mediated LTP enhancement was
   blocked by bath application of IGF2 receptor antibody (IGF2-R Ab,
   5 μg/ml). Representative fEPSPs show superimposed traces recorded
   during baseline and 60 min post-HFS. Calibrations: 0.5 mV/10 ms. Final
   10 min of recording showed that LTP in WT1-AS injected slices was
   significantly reduced by IGF2-R Ab (two-way ANOVA RM, **p = 0.0017;
   F[(1, 11)] = 16.85). For ease of comparison data for the SC-ODN and the
   WT1-AS groups, in both the time course and bar graph, are the same that
   is shown in Fig. [175]3d. Data are expressed as mean ± s.e.m

WT1 enables memory flexibility

   A possible role for WT1 is that it limits memory consolidation and
   strengthening to promote memory flexibility. If this were true,
   eliminating WT1 function, which results in enhanced CFC, should reduce
   the ability of the animals to adapt behavioral responses to a changing
   environment. Thus, we tested whether WT1 depletion affects extinction
   of CFC memory, reversal learning, repetitive compulsive behavior,
   and/or sequential learning. Compared to control littermates, Wt1Δ mice
   showed deficient CFC extinction (Fig. [176]6a), a hippocampal-dependent
   task by which the animals learn to decrease the conditioned response to
   fear^[177]46. Wt1Δ mice also showed reduced spontaneous alternation in
   a Y maze (Fig. [178]6b), a paradigm widely used to test active
   retrograde working memory, based on the general trend of mice to
   explore the least recently visited arm and thus to alternate their
   visits among the three arms^[179]47. Furthermore, when compared to
   Control, Wt1Δ mice showed enhanced memory in the acquisition phase but
   impairment in the reversal learning phase of the Y maze (Fig. [180]6c),
   indicating that WT1 limits the ability to adapt to previously learned
   responses. Finally, Wt1Δ mice also differ from controls in the marble
   burying test, a paradigm used to measure repetitive behavior
   (Fig. [181]6d).

Fig. 6.

   [182]Fig. 6
   [183]Open in a new tab

   WT1 controls memory flexibility. a Wt1∆ mice exhibited a lower rate of
   fear extinction than their control littermates measured at day 5 of
   extinction; % freezing at day 5 was normalized to freezing at day 1
   (unpaired t test: *p = 0.0196). b Wt1∆ mice showed impaired spontaneous
   alternation (% alternation) in a Y maze test (unpaired t test:
   *p = 0.0178). c Wt1∆ mice compared to Control mice performed
   significant different during the acquisition and reversal phase of the
   reversal learning task in a Y maze. Data are expressed as % correct arm
   entry (baited arm). A-day 1 and A-day 2: acquisition sessions 1–2;
   R-day 1 and R-day 2: reversal sessions 1–2 (two-way ANOVA RM;
   *p = 0.0348; F[(1,17)] = 5.263 for acquisition phase; *p = 0.0120;
   F[(1,17)] = 7.916 for reversal phase). d, Wt1∆ mice exhibited an
   increase in repetitive behavior as indicated by the number of marbles
   buried in the marble burying test (unpaired t test: *p = 0.0297). Data
   are expressed as mean ± s.e.m

   Together these results indicated that the enhanced memory of Wt1Δ mice
   impacts the abilities of these mice to learn new experiences and
   flexibly modify their behavior to adapt toward changing environments.
   These results suggest that sequential learning would be impaired in
   Wt1Δ mice.

   To obtain further experimentally testable predictions about possible
   effects of WT1 on sequential memory, we developed a toy control
   theory-based model of an information processing and response system. In
   the model, we postulated that experience activates two parallel
   pathways: a memory-strengthening pathway that includes transcription
   factors like CREB and EGR1, and a memory-weakening pathway that
   includes transcription factors such as WT1. Together the pathways
   control the activity level of effectors to regulate balance that
   dictates the level of memory retention. A priori, the total number of
   effectors could either be in excess of that needed to encode multiple
   experiences, or they could limit the encoding capacity of the
   cortico-hippocampal circuit (Fig. [184]7a). We used the computational
   toy model to run simulations to study the effect of varying the
   activity of the memory weakening pathway for a fixed stimulus. The
   results of the simulations are shown in Supplementary Fig. [185]8a. The
   model predicts that if the memory capacity of the cortico-hippocampal
   circuit is limiting, then over-representation of the first experience
   could interfere with the ability to acquire subsequent experiences.
   Alternatively, if effectors were not limiting, then reducing WT1 levels
   could enhance the ability to memorize both experiences (this is shown
   schematically in the bar graphs in Fig. [186]7a and in the simulation
   results in Supplementary Fig. [187]8b; refer to methods section
   Table [188]2).

Fig. 7.

   [189]Fig. 7
   [190]Open in a new tab

   Consequences of WT1∆-mediated impaired memory flexibility. a Proposed
   mechanism of WT1’s effect on memory regulation. An initial experience
   such as Task 1 (NOL) activates both pro-memory strengthening and
   pro-memory weakening pathways. When the memory weakening pathways are
   inhibited by depletion of WT1, there is prolonged memory for Task 1.
   Retention of Task 1 memory may or may not interfere with the ability to
   remember a Task 2 (CFC) based on the availability of effectors
   (limiting vs in excess). b Schematic representation for short-interval
   sequential training in mice (top panel). Wt1∆ mice showed increased
   time spent exploring the new location when first trained in NOL and
   tested 24 h after training (left panel, unpaired t test: *p = 0.0422.
   Dashed line indicates 50% preference). In the next day after being
   tested in NOL mice were trained on CFC and Wt1∆ mice spent
   significantly less time freezing than Control littermates when tested
   24 h after training (right panel, unpaired t test: *p = 0.0161). c
   Schematic representation for long-interval sequential training (top
   panel). Wt1∆ mice showed increased time spent exploring the new
   location when first trained in NOL and tested 24 h after training (left
   panel, unpaired t test: **p = 0.0036. Dashed line indicates 50%
   preference). Nine days after being tested in NOL, Wt1∆ mice were
   trained on CFC and compared with control group, they spent comparable
   amount of time freezing when tested 24 h after training (right panel,
   unpaired t test: p = 0.3816). Data are expressed as mean ± s.e.m

Table 2.

   Model parameters
   Parameter Description Value Comments
   τ [1] Memory-strengthening signaling time constant 0.5 h Should be
   faster than memory-weakening; not affected by WT1 knockdown
   τ [2] Memory-weakening signaling time constant 36 h control; 144 h WT1
   knockdown Slower than memory-strengthening, prolonged by WT1 knockdown
   K[1], K[2] Steady-state gains 3 control; 7.2 WT1 knockdown N/A
   u Step input magnitude 0.125 nominal Range 0.025 to 0.15 for parameter
   variation Applies to all memory tests, and all animals (control or WT1
   knockdown)
   [191]Open in a new tab

   We therefore tested whether WT1 depletion, which enhances memory for
   one learning, would interfere with new learning in a sequential
   behavioral paradigm. We first trained mice in NOL, which does not yield
   long term memory (LTM) at 24 h after training in Control mice, but does
   so in Wt1Δ mice (whereas Control mice shows memory retention at earlier
   time points, e.g., one hour after training; Fig. [192]2b). We then
   exposed the mice to a second learning experience, CFC, which normally
   does induce LTM (as shown in Fig. [193]2b). As depicted in Fig. [194]7b
   left panel, as expected, Wt1Δ mice had significant LTM retention for
   NOL at 24 h after training, while Control mice did not. However, when
   Wt1Δ mice that first underwent the NOL experience, were exposed one day
   later to CFC training, they showed significantly reduced LTM for CFC at
   24 h compared with Control mice (Fig. [195]7b, right panel), indicating
   that the first experience learned in the absence of WT1 impacts
   subsequent learning. In line with these finding, we observed that Wt1Δ
   mice showed significant preference for the new location when tested
   48 h after NOL training (Supplementary Fig. [196]9a), suggesting a
   memory interference effect. To determine the duration of this active
   learning interference, we tested the effect of extending the interval
   of time between NOL and CFC learning to 10 days. Consistent with
   previous experiments, Wt1Δ mice showed enhanced 24 h retention for NOL
   (Fig. [197]7c, left panel). The 10 days delay between the two
   sequential experiences resulted in no difference between the two groups
   in LTM for CFC (Fig. [198]7c, right panel), indicating that the
   interference effect is a decaying function of the process induced by
   the first learning experience and it is temporarily limited. Animals
   from both groups showed similar exploration time during NOL training
   for both experiments (Supplementary Fig. [199]9b, c). This suggests
   that strengthening memory by removing WT1 limits behavioral flexibility
   and that this effect is temporarily restricted.

Discussion

   A better understanding of mechanisms of forgetting is critical for
   understanding memory storage and persistence. In this study, we
   identified an important role for the transcriptional repressor WT1 in
   limiting memory strength by promoting forgetting, which is required for
   normal flexibility in forming sequential memories. Since WT1, like
   activator transcription factors including C/EBP, cFos, and
   Zif268^[200]7, is induced by LTP and behavioral training, we conclude
   that the cascade of gene expression that is engaged during learning,
   and required for long-term memory, requires specific transcriptional
   repressors in addition to activators. Surprisingly, not many
   transcription factors have been studied in the context of forgetting
   and memory flexibility; one example is the transcription activator
   XBP1, which like WT1 is induced by learning^[201]48, but acts
   conversely as a positive regulator of hippocampal long-term memory and
   flexibility through transcriptional upregulation of brain-derived
   neurotrophic factor^[202]49. Given that the role of WT1 is to actively
   promote forgetting, transcriptional repression via specific DNA binding
   factors adds to other recently identified mechanisms of active
   forgetting (processes that counteract memory consolidation and
   strengthening), which include neurogenesis and Rac1-, dopamine-, and
   Cdc42-mediated AMPA receptor endocytosis^[203]3,[204]50. Notably, the
   process of WT1-mediated active forgetting will occur via the function
   of its target genes, including several members of the retinoic acid
   signaling pathway (Rdh5, Crabp2, and Ttr), the immediate early genes
   Arc and Fos, as well as Igf2 which our data indicate to signal through
   the IGF-2 receptor (IGF-2R). IGF-2R, also known as cation-independent
   mannose-6-phosphate receptor, binds IGF-2 and other ligands and targets
   them to lysosomal degradation. Hence, it is possible that lysosomal
   degradation serves to rebalance and complement the de novo protein
   synthesis and structural changes induced by learning. Of note, Igf2 is
   one of the best characterized WT1 target genes^[205]42, and it has been
   shown to enhance synaptic plasticity and long-term memory and to
   prevent memory loss^[206]44,[207]51, mimicking some of our findings
   related to WT1-ablated animals. However, there are divergences as IGF-2
   injected mice show enhanced fear extinction with intact memory
   flexibility^[208]45,[209]52, suggesting that induced IGF2 expression
   can explain only some of the behavioral effects observed in WT1-ablated
   animals (enhanced plasticity and long-term memory).

   One additional observation is that WT1, by regulating its targets,
   might be involved in the regulation of homeostatic plasticity or
   synaptic scaling^[210]53, which is the ability of neurons to respond to
   periods of reduced or excessive activity by increasing or decreasing,
   respectively, their synaptic efficiency. Synaptic scaling in excitatory
   neurons occurs through enhancement, or decrease of, AMPA
   receptor-mediated transmission, which is in turn regulated by several
   molecular players^[211]53. In this regard, active forgetting has also
   been linked to cytoskeleton targeting mechanisms of synaptic
   remodeling^[212]11,[213]54,[214]55 and AMPA receptor
   recycling^[215]50,[216]56. The regulation mechanisms underlying
   homeostatic plasticity and AMPA receptor recycling are still only
   partially known, but they include some of the WT1 target genes, such as
   Arc^[217]57, retinoic acid^[218]58, and IGF-2^[219]44. Notably, both
   retinoic acid and IGF-2 bind to the IGF-2 receptor, which regulates
   endocytosis and endosomal trafficking, in addition to lysosomal
   degradation^[220]59. We suggest that this regulation may influence AMPA
   receptor trafficking and surface expression, and therefore contribute
   to synaptic scaling as well as memory-related plasticity. Our data,
   combined with what is known from the literature, indicate that there
   are likely to be multiple downstream molecular mechanisms underlying
   the effects of WT1. Further studies will be needed to determine which
   mechanisms are operative in a particular context.

   Our finding that the expression of nonfunctional WT1 impairs subsequent
   learning after a first learning experience suggests anterograde
   interference due to aberrantly strong representation of the first
   learning experience. The nature of the first task appears important for
   the time window during which the memory interference effect occurs, as
   in the sequential learning protocol we found that Wt1Δ mice still
   showed preference for the new location 48 h after training (see
   Supplementary Fig. [221]9a), time point at which CFC was performed (see
   Fig. [222]7b). However, when the two tests were separated by 10 days,
   Wt1Δ animals performed similarly to Control mice in CFC. Different
   groups have been providing compelling evidence that strong LTP induced
   by learning can limit the ability to induce further LTP, a phenomenon
   known as occlusion^[223]60–[224]62. LTP-occlusion has been shown to
   impair subsequent learning (leading to memory interference) in the
   hippocampus as well as motor cortex^[225]61–[226]63. Furthermore, it
   has been suggested that memory interference is caused by competition
   for neural resources and that it can persist for hours or days before
   the capacity of the neurons to undergo LTP is again restored^[227]64.
   Based on our data that Wt1Δ mice showed enhanced LTP and anterograde
   memory interference, we cannot rule out the possibility that a similar
   mechanism of LTP-occlusion plays a role in the effect observed here.
   Further investigation is needed to address this question.

   The mechanisms that counteract memory strengthening and consolidation
   are critical for normal memory formation. In fact, these mechanisms, by
   preventing over-consolidation of memories, allow learning flexibility
   that supports the ability of the organism to adapt to changing
   conditions. Particularly important for pathologies in the area of
   trauma and anxiety was the observation that WT1-depleted mice trained
   in CFC show decreased extinction and an increased anxiety response, as
   measured by elevated plus maze. These behaviors are typical of anxiety
   disorders including PTSD, in which it is well known that memories of
   the aversive experience and traumas have been
   over-consolidated^[228]65. As WT1 ablation in the hippocampus does not
   affect short-term memory, we suggest that the role of WT1 in forgetting
   is either to counteract memory consolidation or to impair retrieval,
   and further studies will be needed to understand this issue.

   Accurate consolidation of long-term memories in the cortico-hippocampal
   circuit relies on coordinated activity in two major inputs, both
   originating in the entorhinal cortex but activating hippocampal CA1
   neurons either directly, or through the trisynaptic pathway. At the
   level of CA1 neuron, a nonlinear response to synaptic input might
   underlie its capacity to function as a coincidence detector that
   appropriately processes the coherent effects of activity in both
   pathways^[229]66. We reported here that depletion of WT1 from the CA1
   pyramidal neurons leads to a significant increase in excitability
   (Fig. [230]3f and h), to LTP enhancement (Fig. [231]3d and g), and to
   alteration of the intra-hippocampal circuit response (Fig. [232]4).
   Depletion of WT1 interfered with the ability of CA1 neurons to perform
   this circuit level computation, as dual input to CA1 neurons was no
   longer needed to produce LTP.

   Lastly, we speculate that the identification of WT1 as a new
   transcriptional regulator of memory persistence and memory flexibility
   may have potential implications for the treatment of those neurological
   conditions where memory is inflexible and excessively resistant to
   disruption, such as PTSD and OCD.

Methods

Replication, blinding, and statistical analysis

   Experiments were run at least three separate times. For the Wt1∆-mRNA
   seq experiment, the results represent two different biological
   replicates. For behavior experiments the results are obtained from
   pulling together multiple animals from at least two different cohorts.
   Details of replicates are provided in each experiment. No statistical
   methods were used to predetermine sample sizes, but our sample sizes
   are similar to those reported in previous publications.

   For all the electrophysiology and behavior experiments, the
   experimentalists were blind to the mice genotype or to the type of
   oligonucleotide or AAV/HSV virus treatment during the entire data
   gathering process. Only after the data were pooled and analyzed was the
   coding for the different groups revealed.

   Unless otherwise stated, data are represented as mean ± s.e.m. All the
   statistical analyses were run in GraphPad Prism 7.02.

Research animals

   All animal experiments were performed according to ethical regulations
   and protocols approved by the internal Animal Care and Use Committee at
   Icahn School of Medicine at Mount Sinai.

Transcription factor activation arrays

   Nuclear and cytosolic extracts were isolated according to standard
   procedures using low speed centrifugation. All buffers contained
   protease and phosphatase inhibitors. Tissue was lysed using a motorized
   Potter–Elvehjem homogenizer (~10 strokes) in Buffer A (20 mM HEPES (pH
   7.4), 40 mM NaCl, 3 mM MgCl, 0.5 % NP-40, 10% glycerol, 1 mM DTT).
   Homogenized tissue was left for 10 min on ice, and lysates were spun at
   500 g for 10 min at 4 °C to pellet nuclei. Nuclei were washed gently in
   Buffer B (20 mM HEPES (pH 7.4), 40 mM NaCl, 3 mM MgCl, 0.32 M Sucrose,
   1 mM DTT) and spun at 500 g for 10 min at 4 °C. Nuclei were then
   resuspended using equal volumes of Buffer C (20 mM HEPES (pH 7.4),
   40 mM NaCl, 1.5 mM MgCl, 25% glycerol, 1 mM DTT) and of Buffer D (20 mM
   HEPES (pH 7.4), 800 mM KCl, 1.5 mM MgCl, 1% NP-40, 25% Glycerol, 0.5 mM
   EGTA, 1 mM DTT). Samples were then rotated at 4 °C for 30 min to
   extract nuclear proteins and the resulting lysates were then spun at
   13,000 RPM for 20 min at 4 °C.

   The supernatant containing nuclear proteins was used to study
   transcription factors activation using the Panomics Combo Protein-DNA
   Array (Affymetrix, MA1215, now sold by Isogen Life Science). Each array
   membrane is spotted with 345 oligonucleotides that correspond to
   consensus binding sites for different transcription factors. The
   location on the array of each consensus binding site, as well as the
   complete protocol are available in the manufacturer’s website
   [233]http://www.isogen-lifescience.com/tf-protein-dna-array). Five
   micrograms of nuclear extract was incubated with the biotinylated probe
   mix from the array kit for 30 min at 15 °C. These probes are also
   transcription factor consensus binding sites that are complementary to
   the oligonucleotides spotted on the array. Probes that bound to
   transcription factors in the nuclear extract were purified by spin
   column separation, and bound probes were further purified from the
   transcription factors according to the manufacturer’s instructions. The
   purified probes were boiled for 3 min and hybridized overnight at 42 °C
   to the array containing 345 oligonucleotide transcription factor
   consensus binding sites. The array was then washed, blocked, incubated
   with Streptavidin-HRP, and visualized by enhanced chemiluminescence.
   The blot was scanned and spot intensities were quantified using Image
   J.

   For each condition (control and stimulated 30 min), ten CA1 regions
   were dissected from hippocampal slices obtained from at least three
   different animals and were pooled together in order to obtained
   sufficient nuclear extracts (5–10 µg). We compared extracts from
   unstimulated (control) slices with extracts from slices that were
   stimulated with Strong-HFS (see field recordings section within
   electrophysiology methods) and collected 30 min after stimulation.

Gel shift assay-EMSA

   DNA probes were prepared by annealing complementary single-stranded
   oligonucleotides with 5′GATC overhangs (Genosys Biotechnologies, Inc.)
   and labeled by filling in with [α-32P]dGTP and [α -32P]dCTP using
   Klenow enzyme. For the CFC experiment, DNA probes were prepared using
   the LightShift Chemiluminescent EMSA Kit (Thermo Scientific) where
   complementary single-stranded transcription factor binding consensus
   sequence was first biotinylated using the Biotin 3′ End Labeling Kit
   (Thermo Scientific) and then annealed. In both experiments nuclear
   extracts were incubated with labeled DNA probes for 30 min at room
   temperature (22–24 ^°C). For the LTP experiment DNA-binding complexes
   were separated by electrophoresis on a 5%
   polyacrylamide-Tris/glycine-EDTA gel which was dried and exposed to
   X-ray film. For the CFC experiment protein/DNA complexes were separated
   using a 6% DNA retardation gel (Invitrogen) that was electroblotted
   into a Biodyne B membrane (Thermo Scientific), incubated with
   Streptavidin-HRP (Thermo Scientific) and visualized by ECL according to
   the manufacturer’s instructions. The consensus sequence used for WT1
   was: 5′-AATTCGGGGGCGGGGGCGGGGGCGGGGGAGGGGCGC-3′ and its complementary
   sequence. For the CFC experiment binding was confirmed using an
   additional consensus sequence 5′- TCCTCCTCCTCCTCTCCC-3′.

   For the LTP experiments, slices were stimulated using Strong-HFS
   protocol (see field recordings section within electrophysiology
   methods); for the CFC experiment, animals were trained using three
   footshocks protocol (2 s, 0.65 mA, 1 min apart). The control animals
   (indicated as “C”) remained in the conditioning chamber for the same
   amount of time as the ones receiving the shock (indicated as “S”) but
   without receiving any footshock.

Real time quantitative RT–PCR

   Hippocampal total RNA was extracted with TRIzol (Invitrogen) and 1 µg
   of total RNA was reverse-transcribed using SuperScript III First-Strand
   Synthesis System (Invitrogen, ThermoScientific, catalog #18080–051).
   Real-time PCR was performed using 7500RT PCR System (Applied
   Biosystems). 1 µl of the first-strand cDNA was subjected to PCR
   amplification using a QuantiTect SYBR Green PCR kit (Qiagen). IGF-II
   primers (forward: 5′-CCCAGCGAGACTCTGTGCGGA-3′; reverse,
   5′-GGAAGTACGGCCTGAGAGGTA-3′); Forty cycles of PCR amplification were
   performed as follows: denaturation at 95 °C for 30 s, annealing at
   55 °C for 30 s and extension for 30 s at 72 °C. GAPDH (forward,
   5′-TGCACCACCAACTGCTTAGC -3′; reverse, 5′-GGCATGGACTGTGGTCATGA -3′) was
   used as internal control. To determine the relative quantification of
   gene expression the cycle threshold method (C[T]) was used.

Immunohistochemistry

   Rats and mice were deeply anesthetized, perfused using 4%
   paraformaldehyde and coronal or hippocampal brain sections were
   obtained using a vibratome (Leica VT 1000S vibratome; 40 μm) or a
   cryostat (Leica CM1850; 15 μm). Brain slices were then blocked with 3%
   normal goat serum (Vector), 0.3% Triton X-100 (Sigma-Aldrich), 1% BSA
   (Sigma-Aldrich) for 2 h at room temperature and incubated with the
   appropriate primary antibody: rabbit monoclonal WT1 (for staining in
   Fig. [234]3a Santa Cruz, catalog #SC-192 (C-19); for staining in
   Fig. [235]3b Novus Biological, catalog #NBP1–40787; for staining in
   Fig. [236]3c Abcam, catalog #ab52933); mouse monoclonal glial
   fibrillary acidic protein, GFAP (Cell Signaling, catalog #3670); mouse
   monoclonal β-tubulin (Cell Signaling, catalog #86298). An antibody
   against green fluorescent protein-GFP (chicken anti-GFP, from Aves Labs
   Inc., catalog #GFP-1020) was used to check viral spread in WT1
   overexpression experiments using AAV and HSV viruses (Fig. [237]2c, d).
   After incubation with primary antibodies, sections were washed and
   incubated with secondary antibodies complexed to either Alexa Fluor 568
   or Alexa Fluor 488 dyes (Invitrogen, ThermoFisher). Please refer to
   Supplementary Data [238]4 for complete list of antibodies used for this
   study. After washing, Hoechst 33342 (Invitrogen) was used to label
   nuclei. Sections were then mounted and imaged using a confocal
   microscope (Zeiss LSM 880).

Western blotting

   We used either CA1 regions dissected from 400 μm-thickness hippocampal
   slices or dorsal hippocampus, homogenized in proportional volumes of
   ice-cold lysis buffer using a motorized Potter–Elvehjem homogenizer
   (~10 strokes). The lysis buffer consisted of 50 mM Tris-HCl, pH 7.4,
   100 mM NaCl, 1 mM EDTA, 0.5% sodium deoxycholate, 1% NP-40, 0.1% SDS,
   0.5 mM PMSF, 1 μM mycrocystine, 1 µg/ml benzamidine, 2 mM
   dichlorodiphenyltrichloroethane (DTT), 1 mM sodium orthovanadate, 2 mM
   sodium fluoride, 1 mM EGTA; protease inhibitor cocktail (Sigma-Aldrich)
   and phosphatase inhibitor cocktails 2 and 3 (Sigma-Aldrich) were added
   according to manufacturer’s instructions. Lysates were cleared by
   centrifugation at 14,000 RPM for 10 min. Protein concentration was
   determined using Bradford reagent (Biorad). 20–50 µg of total protein
   was loaded per well, into 10% SDS-PAGE and transferred to supported
   nitrocellulose membranes (pore size 0.2 µm, Biorad), followed by
   western blotting and chemiluminescence detection. The following
   antibodies were used: rabbit polyclonal WT1 (custom-made, against a
   synthetic rat-specific peptide; GeneScript), rabbit monoclonal WT1
   (Novus Biological, catalog #NBP1–40787), mouse monoclonal β-tubulin
   (Cell Signaling, catalog #86298), mouse monoclonal β-actin
   (Sigma-Aldrich, catalog #A4700), mouse monoclonal GAPDH (Sigma-Aldrich,
   catalog #G8795). Please refer to Supplementary Data [239]4 for complete
   list of antibodies used for this study. We either use films that were
   scanned and signal intensity analyzed using either ImageJ or Odyssey.
   Uncropped blots are provided as Supplementary Fig. [240]10.

   For both electrophysiology (Fig. [241]1e, Supplementary Fig. [242]2a)
   and behavior (Fig. [243]1f, g, Supplementary Fig. [244]2d) experiments
   time point chosen was 30 min after the delivery of Strong-HFS, LFS or
   the aversive stimulus (shock). Total protein lysates were collected
   from CA1 region (electrophysiology) or from dorsal hippocampus
   (behavior) and processed as described above. For the IA experiment,
   trained animals were compared with Naive ones. For additional
   information about the protocol used for LTD and CFC experiments
   reported in Supplementary Fig. [245]2, refer to the electrophysiology
   and behavioral assay sections respectively.

Hippocampal injections of ODNs or HSV/AAV viruses in rats

   Animals were anesthetized with a solution containing a mix of ketamine
   (100 mg/kg) and xylazine (20 mg/kg) (10 mg/kg, intraperitoneal), and a
   stainless-steel guide cannulae were bilaterally implanted targeting the
   dorsal hippocampus (4.0 mm posterior to bregma, 2.6 mm lateral from
   midline, and 2.0 mm ventral). The rats were returned to their home
   cages and allowed to recover from surgery for 7–10 days.

   For WT1 knock-down experiments all hippocampal injections consisted of
   2 nmol in 1 μl per side (unless otherwise specified) of either WT1
   antisense oligodeoxynucleotide combo (WT1-AS = 1 nmol of WT1 antisense
   1 + 1 nmol of WT1 antisense 2) or scrambled oligodeoxynucleotide combo
   (SC-ODN = 1 nmol Scrambled 1 + 1 nmol Scrambled 2) both diluted in
   phosphate-buffered saline (PBS) at pH 7.4. The sequences used were the
   following: WT1 antisense 1: TCGGAACCCATGAGGTGCGG; WT1 antisense 2:
   TCGGAACCCATGGGGTGC; Scrambled 1: GGTGGTAGAACGCCGTACCG; Scrambled 2:
   GGTGGTAGAACGCCGTCC. The scrambled oligonucleotides, which served as a
   control, were designed to lack homology to any rat sequence in GenBank,
   and contained the same base composition but in a randomized order. Both
   antisense and scrambled oligonucleotides were phosphorothioated on the
   three terminal bases of both 5′ and 3′ ends to increase their stability
   and were reverse phase purified (GeneLink). For electrophysiology
   experiments, male Sprague–Dawley rats were used. Animals received a
   single injection of oligonucleotides 2 h before being sacrificed, and
   their brains were dissected (see Fig. [246]3d for schedule diagram).
   For electrophysiology experiments one side of the brain was always
   injected with WT1-AS and the other side of the brain with SC-ODN. For
   all the behavior experiments either male Sprague-Dawley or Long-Evans
   rats were used and no differences between the strains were observed.
   For behavior experiments, animals received two injections of
   oligodeoxynucleotides 2 h apart and 2 h before training (see
   Fig. [247]2a for schedule diagram); animals were injected bilaterally
   with either WT1-AS or SC-ODN.

   For overexpression of WT1 via HSV, we used a p1005 based HSV vector
   co-expressing GFP and WT1-IsoformD (WT1-HSV). In this system, GFP
   expression is driven by a cytomegalovirus (CMV) promoter, while the
   WT1-isoformD is driven by the IEF4/5 promoter. HSV virus expressing GFP
   alone was used as a control (CTR-HSV). We injected 2 μl of HSV vectors
   in each hemisphere (titer 0.5 × 10^9 infectious unit/ml, Virovek,
   Hayward, CA) using a 28-gauge needle that extended 1.5 mm beyond the
   tip of the guide cannula and connected via polyethylene tubing to a
   Hamilton syringe. The infusions of HSV viruses were delivered at a rate
   of 0.33 μl min^−1 using an infusion pump (Harvard Apparatus). The
   injection needle was left in place for 10 min after the injection to
   allow complete diffusion of the solution. Rats were randomized to
   different treatments.

   For WT1 overexpression via AAV, we used AAV8.2-EF1a-WT1-PP2A-GFP
   (WT1-AAV) and AAV8.2-EF1a-PP2A-GFP (CTR-AAV; both vectors were
   1 × 10^13 vg/ml, Virovek, Hayward, CA) as a control. AAV vectors were
   injected using a 33 Ga needle attached to a 5 µl syringe (Hamilton)
   2 μl in each hemisphere over a 10 min period. The needle was left in
   place for 10 min to allow for efficient diffusion before removal. Rats
   were randomized to different treatments.

   To verify proper placement of cannula implants or viral injection, rats
   were deeply anesthetized and perfused (20 mL/min) with 4% of
   paraformaldehyde (PFA) in PBS, their brains removed and fixed with 10%
   (vol) buffered formalin in PBS for 48 h. Brains were then sliced in
   coronal sections (40 µm) and the hippocampus region was examined under
   a light microscope (for cannulae placement) or confocal microscope (for
   viral injection). Animals where cannulae were misplaced, viral
   expression was mostly spread outside of the hippocampus, and serious
   tissue damage was observed were excluded from the experimental groups.

Generation of functionally deficient WT1 mice

   Forebrain-specific deletion of Wt1 was achieved by crossing animals
   homozygous for the conditional Wt1 knockout allele (Wt1^fl/fl)^[248]67
   with a transgenic line, Camk2a-Cre, (B6.Cg-Tg(Camk2a-cre)T29–1Stl/J;
   Jackson Lab: [249]http://jaxmice.jax.org/strain/005359.html) in which
   Cre recombinase expression is driven by the 7.8 kb promoter of
   Ca^2+/calmodulin-dependent protein kinase II alpha subunit^[250]68.
   Progeny were crossed to obtain Wt1^fl/fl; Camk2a-Cre (referred through
   the paper as Wt1∆ mice) and littermate control animals (referred
   through the paper as Control mice). Expression of Cre recombinase
   resulted in the in-frame deletion of of exons 8 and 9 [see Fig. 1e of
   Gao et al.^[251]67], and generated a truncated allele encoding a
   shortened non functional WT1 protein lacking zinc fingers 2 and 3.
   Expression of the recombined Wt1 allele was detectable in the mouse
   forebrain (see Fig. [252]3a of Gao et al.^[253]67), and its detection
   was performed using the following primers: Primer WT1 Delta Forward 5′
   GCT AAC ATA TGG GAG ACA TT 3′ and Primer WT1 Delta Reverse 5′ TGC CTA
   CCC AAT GCT CAT TG 3′. As reported by others, heterozygous Wt1 mice
   develop kidney nephropathy and glomerulosclerosis^[254]69, which we
   have not observed at any time in the Wt1Δ mice. To further address this
   issue, we evaluated proteinuria since loss of kidney function is
   associated with increased levels of proteins in the urine. Using
   Chemstrips (Roche), we found that there was no significant difference
   between proteinuria levels of Wt1∆ mice compared with their control
   littermates as indicated by the color of the top strips (Supplementary
   Fig. [255]4c). We further confirmed that kidney function was normal and
   that there was no significant difference in the enzymatic values of
   Wt1∆ mice through a pathology screening of their blood samples
   performed at the Comparative Pathology Center of Mount Sinai
   (Supplementary Fig. [256]4d).

   To genotype the animals, we used the following primers for the LoxP
   allele: Primer LoxP Forward 5′ CCT TTT ACT TGG ACC GTT TG 3′ and Primer
   LoxP Reverse 5′ GGG GAG CCT GTT AGG GTA 3′. For the Cre allele we used
   the following primers: Cre Primer Forward 5′ GCG GTC TGG CAG TAA AAA
   CTA TC 3′ and Cre Primer Reverse 5′ GTG AAA CAG CAT TGC TGT CAC TT 3′
   (as indicated in the genotyping section by Jackson lab at
   [257]http://jaxmice.jax.org/strain/005359.html).

   Wt1∆ animals were viable and had a normal life span, normal body
   weight, normal fertility and a normal growth rate compared with control
   littermates.

   Throughout the study control wild-type littermates are indicated as
   Control and they comprise the following subgroups: Wt1^+/+; Camk2a-Cre
   positive, Wt1^+/+; Camk2a-Cre negative, Wt1^fl/+; Camk2a-Cre negative,
   Wt1^fl/fl; Camk2a-Cre negative. These were grouped together for both
   electrophysiology and behavior experiments, since they were no
   statistically different between the genotypes.

Hematoxylin and eosin (H&E) staining

   H&E staining was performed in order to verify if there was any
   macroscopic abnormality in brain tissue of Control and Wt1∆ mice.
   Animals were deeply anesthetized with a solution containing
   ketamine + xylazine and perfused transcardially with ice-cold 10%
   formalin. The brains were embeded in paraffin and sliced into 2 μm
   thick sections for staining. The sections were de-paraffinized in
   xylene, rehydrated in graded ethanol series, stained with Mayer’s
   Haemalaun (Carl Roth, Karlsruhe, Germany) for 5 min, washed again, and
   stained with 1% eosin (Carl Roth, Karlsruhe, Germany for 2 min). The
   sections were washed in water, dehydrated in graded ethanol series,
   treated with xylene and mounted for imaging (Supplementary
   Fig. [258]4b).

Contextual fear conditioning and extinction

   Mice or rats were handled for 3 min per day for 5 days before training.
   The conditioning chamber consisted of a rectangular Perspex box
   (VFC-008: 30.5 × 24.1 × 21.0 cm, Med Associates) with a metal grid
   floor (Model ENV-008 Med Associates) through which the footshock was
   delivered. The experiment was conducted in a sound-attenuated room,
   with low levels of light and white noise background.

   For animals that underwent paired fear conditioning, the training
   session consisted of 2 min exploring the context prior to the delivery
   of either one footshock (2 s, 0.65 mA) or three footshocks (2 s,
   0.65 mA, 1 min apart); after that animals remained in the chamber for
   additional 2 min before returning to their home cages. Control animals
   were allowed to explore the context for exactly the same time as the
   shocked groups without receiving any footshock. Rats trained in the
   unpaired fear conditioning were placed in the context and after 5 s one
   footshock (2 s, 0.65 mA) was delivered. Animals stayed in the box for
   additional 20 s. Control group was exposed to the context for 25 s
   without receiving any footshock. Animals from all groups were tested
   24 h after training and 30 days after training (mice only). Test
   session consisted in placing the animals back into the conditioning
   chamber for 5 min in the absence of any footshock and freezing behavior
   was recorded. For mice memory extinction experiment, 24 h after CFC
   training, animals were placed into the conditioning chamber for five
   consecutive days, 5 min each day in the absence any footshock and
   freezing was scored. Sessions were recorded using a digital video
   camera, and freezing behavior defined as lack of movement besides heart
   beat and respiration, was scored every 10 s by trained observers blind
   to the experimental conditions. The number of scores indicating
   freezing (reported in Fig. [259]2a only) were calculated as a
   percentage of the total number of observations. Ethovision (Noldus
   Information Technology) was used to measure percentage of freezing time
   in all experiments involving mice and rat experiments with WT1
   overexpression and in those reported in Supplementary Fig. [260]2b, c.

Inhibitory Avoidance (IA)

   IA was carried out as described previously^[261]44. Briefly the IA
   chamber (Med Associates) consisted of a rectangular Perspex box divided
   into a safe compartment and a shock compartment. The safe compartment
   was white and illuminated, whereas the shock compartment was black and
   dark. The chamber was located in a sound-attenuated, non-illuminated
   room. Footshocks were delivered though the grid floor of the shock
   chamber via a constant current scrambler circuit. During training
   sessions, each rat was placed in the safe compartment with its head
   facing away from the door. After 10 s, the door separating the
   compartments was automatically opened, allowing the rat access to the
   shock compartment; the rats usually enter the shock (dark) compartment
   within 10–20 s of the door opening. As soon as rats stepped into the
   shock compartment a mild footshock was delivered. (2 s, 0.60 mA). For
   the western blot experiment (Fig. [262]1g) using IA extracts, animals
   were euthanized 30 min after training using halothane and their brains
   dissected. Dorsal hippocampi from trained animals were compared with
   dorsal hippocampi obtained from naive controls (animals that remained
   in their home cages).

Novel Object Location (NOL)

   For both mice and rats experiments, animals were allowed to familiarize
   with the arena for 5 min each day for three consecutive days before
   training. The arena consisted of a opaque box
   (44.4 cm × 44.4 cm × 31.5 cm for rats and 28 cm × 28 cm × 20 cm for
   mice). Arena was placed in a room with a low level of light and
   sound-proof. During training session two identical objects (Lego^®)
   were placed into the arena side by side and animals were allowed to
   freely explore them for 10 min, returning to their home cages
   afterwards. During testing animals were placed back into the arena for
   5 min and one object was moved to a different location which was
   counterbalanced between animals. Object exploration was defined as the
   orientation of the animal’s nose towards the object at a distance ≤2 cm
   or as the animal placing its forepaws on the object; climbing on the
   object was not considered exploration. The objects and the arena were
   cleaned with 70% ethanol between animals to avoid olfactory cues. For
   NOL experiments with rats the sessions were videotaped and scored by an
   experimenter blind to experimental conditions; for NOL experiments with
   mice the sessions were scored using Ethovision (Noldus Information
   Technology). Memory retention was measured as % Preference calculated
   as the time spent exploring the object in the new location (N) relative
   to the total exploration time (N + familiar (F)) (%
   Preference = (N/(N + F)*100)^[263]70.

Open field

   For the locomotion experiment in rats, animals were allowed to freely
   explore for 5 min an open field arena (44.4 cm × 44.4 cm × 31.5 cm)
   divided into 16 imaginary quadrants. Locomotion was calculated as total
   number of crossings in the open field. An observer blind to
   experimental procedures scored the experiments. For mice, they were
   allowed to explore an empty arena (34 cm × 34 cm × 23 cm) for 10 min
   during which the total distance traveled as well as the time spent in
   the center or periphery of the arena were recorded using a video
   tracking system (Ethovision, Noldus Information Technology).

Spontaneous alternation and reversal learning in a Y maze

   Spontaneous alternation and reversal learning were performed as
   described previously^[264]51. Briefly the Y-maze consisted of three
   white opaque arms (Med Associates) with sliding doors at the entrance
   of each arm. During spontaneous alternation test animals were allowed
   to freely explore the three arms from the center of the maze for 10 min
   and spontaneous alternation was defined as successive entries into each
   of the arms on overlapping triplets sets (e.g., ABC, BCA, CAB, etc).
   The percentage of alternation was calculated by as the ratio of total
   alternations to possible alternation (total arm entries −2) × 100. For
   the reversal learning experiment mice were single housed, food
   restricted and monitored daily until they reached 85% of their original
   weight before starting the experiment and during testing. They were
   given 1/2 food pellet (LabDiet 5053) and one fruit loop (Kellog’s) each
   day. The habituation phase was identical to spontaneous alternation.
   During the acquisition phase, one arm of the maze was chosen as the
   “correct arm” and baited with half of a fruit loop. The animals were
   initially restrained in the “start arm” for 1 min and then allowed to
   explore between the two arms. The acquisition phase consisted of 10
   consecutive trials per day for 2 days (each day divided in 2 blocks of
   5 trials each). Memory was calculated as the percentage of correct
   choice over each block of trials. During the reversal learning phase
   the “correct arm” was switched. The “correct arm” was counterbalanced
   between animals. Both experiments were scored by an observer blind to
   the experimental conditions and analyzed manually.

Marble burying test

   Regular rat cages were used and filled with ~5 cm deep bedding tamped
   down to make a flat, even surface. A regular pattern of 20 glass
   marbles was positioned on the surface of the bedding, spaced regularly,
   about 4 cm apart one from the other. Each animal was left in the cage
   for 30 min and the % marbles buried was calculated as the number of
   marbles buried to ~2/3 of their depth over the total number of marbles
   × 100.

Elevated plus maze

   The elevated plus maze consisted of black Plexiglass fitted with white
   bottom surfaces to provide contrast and was placed 60 cm above the
   floor. The four arms (2 open and 2 closed) were interconnected by a
   central platform. Mice were placed at the center of the maze and were
   allowed to freely explore it for 5 min under red-lighting conditions.
   Time that each animal spent in the open and closed arms as well as the
   number of entries in the closed and open arms were recorded and further
   analyzed using Ethovision (Noldus Information Technology).

Plantar test (Hargreaves method)

   To assess mice nociceptive response, animals were placed in a clear
   plastic chamber (45 cm × 40 cm, divided in 12 small animal enclosures,
   IITC Life Science) with a glass floor and allowed to acclimatize to the
   room and to the apparatus for 2 h. After the acclimation period, the
   radiant heat source (infrared beam) was positioned under the glass
   floor directly beneath one of the animal’s hind paws. The radiant heat
   source creates a 4 × 6 mm intense spot on the paw. The paw withdrawal
   latency was determined using an electronic stopwatch coupled to the
   infrared source that switches off when the animal feels discomfort and
   withdraws its paw; a cutoff of 20 s for paw withdrawal was set up.

Electrophysiology

   Field recording: Male Sprague-Dawley rats (6–8 weeks old) or ~3 months
   old mice (either Control or Wt1∆ mice) were deeply anesthetized with
   isoflurane and decapitated. The brain was rapidly removed and chilled
   in ice-cold artificial cerebro spinal fluid (ACSF) containing (in mM)
   118 NaCl, 3.5 KCl, 2.5 CaCl[2], 1.3 MgSO[4], 1.25 NaH[2]PO[4], 24
   NaHCO[3], and 15 glucose, bubbled with 95% O[2]/5% CO[2]. Transverse
   slices of dorsal hippocampus (400 μm thick) were made on a tissue
   chopper at 4 °C, and then placed in an interface chamber (ACSF and
   humidified 95% O[2]/5% CO[2] atmosphere), where they were maintained at
   room temperature for at least 2 h. For recording, slices were
   transferred to a submersion chamber and superfused with ACSF at
   31 ± 1 °C. Monophasic, constant-current stimuli (100 μs) were delivered
   with a bipolar stainless steel electrode positioned in stratum radiatum
   of area CA3, and field EPSPs (fEPSPs) were recorded in stratum radiatum
   of area CA1, using electrodes filled with ACSF (Re = 2–4 MΩ). For all
   slices, initial spike threshold exceeded 2 mV. Signals were low-pass
   filtered at 3 kHz and digitized at 20 kHz, and analyzed using pClamp 9
   (Molecular Devices). Two HFS protocols were used: Weak-HFS, consisting
   of two trains separated by 20 s, each consisting of 100 stimuli
   delivered at 100 Hz at an intensity that initially evoked a fEPSP
   measuring 20% of spike threshold; and Strong-HFS, identical to Weak-HFS
   but delivered at an intensity that initially evoked a fEPSP of 75–80%
   of spike threshold. For the LFS protocol, 900 pulses at 1 Hz were
   delivered at an intensity that initially evoked a fEPSP of 100% of
   spike threshold. In all experiments, the stimulation protocol was
   delivered at least 30 min after transfer of the slices to the recording
   chamber, when the basal fEPSP had been stable for at least 20 min.
   Control slices were placed in the recording chamber and subjected only
   to test stimuli (0.033 Hz). Drug preincubations, when used, were
   performed at room temperature in submersion maintenance chambers
   containing ACSF saturated with bubbling 95% O[2]/5% CO[2]. Drugs were
   prepared as stock solutions and diluted to final concentrations in ACSF
   before use.

   In slices where both the TA→CA1 and SC→CA1 inputs were activated,
   stimulating electrodes were placed both in proximal stratum radiatum
   near the CA1/CA2 border (to activate Schaffer collaterals) and in the
   lacunosum moleculare within CA1 (to activate the perforant path). For
   the baseline period, slices were stimulated every 30 s, alternating
   between Schaffer collaterals and perforant path. The perforant path was
   activated with theta-burst stimulation (TBS) consisting of 10 bursts at
   5 Hz, 4 pulses per burst at 100 Hz, using 250 µA stimuli. The Schaffer
   collaterals were stimulated with the same TBS pattern, delayed 20 ms
   delay relative to the perforant path, at an intensity that initially
   evoked 90% of the spike threshold. Recording electrodes were positioned
   in stratum radiatum and stratum lacunosum-moleculare. All slices had a
   spike threshold of at least 1.8 mV in stratum radiatum.

   For recordings in the presence of bicuculline, the brain was rapidly
   removed and chilled in ice-cold ACSF containing (in mM) 118 NaCl, 2.5
   KCl, 4 CaCl[2], 4 MgSO[4], 1.25 NaH[2]PO[4], 24 NaHCO[3], and 15
   glucose, bubbled with 95% O[2]/5% CO[2]. Transverse slices of dorsal
   hippocampus (400 μm thick) were made on a tissue chopper at 4 °C, and
   then placed in an interface chamber (ACSF and humidified 95% O[2]/5%
   CO[2] atmosphere), where they were maintained at room temperature for
   at least 1 h. The CA3 region was then dissected from CA1 region and
   slices were placed in a submersion chamber for 0.5–2.5 h before being
   transferred to the recording chamber. A Weak-HFS was delivered at a
   stimulus strength that evoked a fEPSP measuring 25–30% of spike
   threshold in bicuculline. All other conditions were as described above.
   Bicuculline was suspended in water to 10 mM and diluted to 10 μM in
   ACSF immediately before the experiment began.

   Whole-cell recording: Adult male Sprague-Dawley rats (250–300 g) were
   deeply anesthetized with isoflurane and transcardially perfused with
   ice-cold ACSF. For experiments on excitability (Fig. [265]3f), the ACSF
   contained (in mM): NaCl (128), d-glucose (10), NaH[2]PO[4] (1.25),
   NaHCO[3] (25), CaCl[2] (2), MgSO[4] (2), and KCl (3), bubbled with 5%
   CO[2] /95% O[2] (pH = 7.3, 290–300 mOsM). Following perfusion, the
   brain was rapidly removed and chilled in ice-cold sucrose-ACSF
   containing (in mM): sucrose (254), D-glucose (10), NaH[2]PO4 (1.25),
   NaHCO[3] (25), CaCl[2] (2), MgSO[4] (2), and KCl (3) (pH = 7.3, 290–310
   mOsM). Coronal slices of dorsal hippocampus (200 μm thick) were
   prepared using a vibratome in ice-cold sucrose-ACSF, and were allowed
   to recover submerged in bubbled ACSF for 45 min at 33 ± 1 °C, and
   thereafter at room temperature. Slices were transferred to a submersion
   recording chamber and perfused with ACSF (2 mL/min) at room
   temperature. CA1 pyramidal neurons were identified using IR DIC optics,
   and whole-cell recordings were obtained with an Axopatch 1D amplifier.
   Signals were low-pass filtered at 2 kHz and digitized at 20 kHz, and no
   adjustment was made for pipette junction potential. Membrane
   excitability was tested in current clamp mode using pipettes containing
   (in mM): K gluconate (115), KCl (20), MgCl[2] (1.5),
   phosphocreatine-Tris (10), Mg-ATP (2), Na-GTP (0.5), and Hepes (10)
   (pH = 7.3, 280–285 mOsM; 3.5–4.5 MΩ). The membrane was depolarized with
   a series of ten 200 ms-long current steps, increasing from 10 to 100 pA
   from a holding potential of −70 mV.

   For recording spontaneous and miniature EPSCs (mEPSCs) (Supplementary
   Fig. [266]7a, b), slice preparation and recordings were performed in
   modified ACSF containing (in mM): NaCl (128), d-glucose (10),
   NaH[2]PO[4] (1.25), NaHCO[3] (25), CaCl[2] (2), MgCl[2] (2), and KCl
   (3) (pH = 7.3, 290–300 mOsM), using pipettes filled with (in mM):
   Cs-methanesulfonate (130), HEPES (10), EGTA (0.5), NaCl (8), TEA-Cl
   (5), Mg-ATP (4), Na-GTP (0.4), Na-phosphocreatine (10), and N-ethyl
   lidocaine (1) (pH = 7.3, 280–285 mOsM; 3.0–4.5 MΩ). mEPSCs were
   recorded in the presence of D,L-2-amino-5-phosphonovaleric acid (APV;
   50 μM), gabazine (5 μM), and tetrodotoxin (0.5 μM). Spontaneous events
   were recorded in the absence of inhibitors. 3–5 min after breakthrough,
   gap-free recordings were obtained for 10 min. Only cells with stable
   input resistances (<20% change as measured before and after the
   gap-free period) were included in the analysis. Template-based event
   detection was performed using Clampfit 10.3 (Molecular Devices).
   Templates were generated by averaging 5–10 events for each file, and
   the automated search results were verified manually.

Molecules and inhibitors used in electrophysiology

   Bicuculline was purchased from Tocris (catalog #2503) and resuspended
   in ACSF to reach a final concentration used 10 μM. The antibody against
   the IGF2 Receptor (IGF2-R Ab) was purchased from R&D solutions (catalog
   #AF2447) and used at a final concentration of 5 μg/ml.

Transcriptomic profiling by mRNAseq

   For the mRNAseq experiments, total RNA was extracted using Trizol
   (Thermo Fisher) from CA1 regions isolated from rat hippocampal slices
   (Control vs LTP 90 min). A pool of ~10 CA1 regions collected from
   hippocampal slices of at least three different animals were necessary
   in order to obtain ~1 µg of total RNA for each condition. For the
   experiment relative to Wt1∆ mice versus wild type littermates, dorsal
   hippocampus from naïve untrained animals were used. For the experiment
   relative to acute WT1 knockdown in rats (WT1-ODN vs Scrambled-ODN),
   dorsal hippocampus tissue surrounding the injection site was used. For
   all the mRNA sequencing experiments RNA integrity was checked by either
   the Agilent 2100 Bioanalyzer using the RNA 6000 Nano assay (Agilent,
   CA, USA). All processed total RNA samples had RIN value≥9. The seq
   library was prepared with the standard TruSeq RNA Sample Prep Kit v2
   protocol (Illumina, CA, USA). Briefly, total RNA was poly-A-selected
   and then fragmented. The cDNA was synthesized using random hexamers,
   end-repaired and ligated with appropriate adapters for seq. The library
   then underwent size selection and purification using AMPure XP beads
   (Beckman Coulter, CA, USA). The appropriate Illumina-recommended 6 bp
   barcode bases are introduced at one end of the adapters during PCR
   amplification step. The size and concentration of the RNAseq libraries
   was measured by the Agilent 2100 Bioanalyzer using the DNA 1000 assay
   (Agilent, CA, USA) before loading onto the sequencer. The mRNA
   libraries were sequenced on the Illumina HiSeq 2000 System with 100
   nucleotide single-end reads, according to the standard manufacturer’s
   protocol (Illumina, CA, USA).

   For the RNA-Seq data analysis Tophat 2.0.13^[267]71, bowtie
   2.1.0^[268]72, samtool 0.1.7^[269]73 and cufflinks 1.3.0^[270]74 were
   used. The rn5-bowtie2 index was generated with the command
   “bowtie2-build rn5.fa rn5”. The “rn5.fa”-file was downloaded from the
   UCSC genome browser. The mm10-bowtie2 index was downloaded from
   [271]http://bowtie-bio.sourceforge.net/bowtie2/manual.shtml. RefSeq
   geneTracks and GTF-files for the rn5 and mm10 genome assembly were
   downloaded from UCSC genome browser. Common gene ids in the GTF-files
   were matched to individual transcript_ids using the corresponding
   official symbols obtained from the geneTracks files.

   The likelihood to detected a lowly to moderately expressed gene in a
   particular sample depends on the total number of sequenced reads,
   especially in case of lower reads counts (<30,000,000)^[272]75.
   Therefore it could happen that more genes are detected in a sample with
   a higher read count than in a sample with a lower read count. This
   experimental artifact might distort normalization including total reads
   normalization as well as upper quartile normalization that is applied
   in this study. Both normalization methods only change the number of
   reads that are associated with a gene, but not the number of identified
   genes. In consequence, the same number of reads might be distributed
   over a different number of (by chance) experimentally identified genes
   in two samples, introducing gene expression differences between the two
   samples that do not exist. To prevent such experimental artifacts reads
   we applied an additional computational step before read alignment and
   differentially expressed genes detection. Under the assumption that
   during the seq process every fragment has the same chance to be
   sequenced, we ensured that each sample had the same number of total
   read counts by randomly removing reads from those samples with higher
   read counts than the minimum read count.

   Reads were aligned to the rn5 or mm10 genome using Tophat with the
   option “--no-novel-juncs” and the refSeq-GTF-file (the option
   “--solexa1.3-quals” was additionally chosen in case of the rat
   samples). Differentially expressed genes were identified using Cuffdiff
   with the options “--upper-quartile-norm”, “--frag-bias-correct” against
   the rn5 genome and “--multi-read-correct” and the refSeq-GTF-file.

   In each analysis all differentially expressed genes (DEGs) that were
   statistically significant (FDR = 5%) were considered. DEGs with a
   minimum fold change of
   log[2]((FPKM[condition1] + 1)/(FPKM[condition2] + 1)) > = ±log[2](1.3)
   were submitted to pathway enrichment analysis as described below.

Analysis of transcriptomic data

   Enrichment analysis using mRNA seq data was performed similarly as
   previously described^[273]76. The “Transfac and jaspar pwms” library
   was downloaded from the EnrichR website^[274]77. All human
   transcription factor gene associations were kept. Human target genes
   and transcription factors were replaced by their rat homologs based on
   the mouse informatics database (Mouse Genome Informatics,
   [275]http://www.informatics.jax.org, 5/24/2013) and the National Center
   for Biotechnology Information homologene database
   ([276]http://www.ncbi.nlm.nih.gov/homologene/, 06/01/2018). Mouse
   gene-transcription factor associates were removed from the database.

   To increase the statistical accuracy we removed all gene symbols in
   both databases that are not part of the RefSeq rn5 gene annotation and
   therefore could not be identified as differentially expressed.
   Similarly, we removed all differentially expressed genes that were not
   part of the “Transfac_and_jaspar_pwms” library. Right tailed fisher’s
   exact test was used for enrichment analysis and the negative logarithms
   to the basis 10 of the p-values were calculated.

Control theory-based toy model of WT1 function

   Input of an experience to the hippocampus is represented as a
   rectangular pulse. Neuronal activity in the hippocampus converts this
   pulse into a more long lasting output with respect to the time scale of
   the experiments (days), which we represent as a time integrator. Thus,
   the area under the rectangular pulse becomes a step function as inputs
   to memory-strengthening and memory-weakening pathways. We are unaware
   of any experimental data to suggest reasonable values for the magnitude
   of this step input, u, hence arbitrary values were chosen and u was
   subsequently varied to make a range of predictions (Supplementary
   Fig. [277]7). We model memory-strengthening and memory-weakening
   signaling as two first order processes in parallel. A first order
   process is governed by the following equation:
   [MATH:
   <mi>τ</mi><mfrac><mrow><mi>d</mi><mi>x</mi></mrow><mrow><mi>d</mi><mi>t
   </mi></mrow></mfrac><mo>=</mo><mo>-</mo><mi>x</mi><mo>+</mo><mi>K</mi><
   mo>⋅</mo><mi>u</mi><mrow><mo>(</mo><mrow><mi>t</mi></mrow><mo>)</mo></m
   row><mo>.</mo> :MATH]
   1

   Here, τ is the time constant, K is the steady state gain, u is the
   input strength, t is time, and x is the dependent variable (in this
   case memory-strengthening or memory-weakening signal strength). We
   denote memory-strengthening with the subscript 1 and memory-weakening
   with the subscript 2. Because activation of one cell’s signaling could
   affect other non-activated cells, we take both gains (K[1] and K[2]),
   to be 3, reflecting signal amplification. However, in the model the
   effects of these gains and the input magnitude are indistinguishable,
   so our parameter variation exercise effectively explored both of these
   avenues. Additionally, we estimated from electrophysiological data that
   lack of functional WT1 induces an ~2.4-fold increase in the input
   signal strength, so in the case of Wt1∆ mice, we take the gains as 7.2.
   The time constants τ[1] and τ[2] were tuned to be consistent with the
   data in Figs. [278]1–[279]3. Thus, τ[1] for memory-strengthening
   signaling was taken as fast (0.5 h) and not affected by lack of
   functional WT1, whereas τ[2] for memory-weakening signaling was taken
   as slow (36 h) and took a different value for Wt1∆ animals (144 h).
   These model parameters are summarized in the below Table [280]2:

   The difference of these two process outputs was passed through a
   saturation function (based on neurobiological reasoning presented in
   the main text), to be fixed between 0 and 1, which we call “Pathway
   Activity”. Thus,
   [MATH:
   <mi>P</mi><mi>a</mi><mi>t</mi><mi>h</mi><mi>w</mi><mi>a</mi><mi>y</mi><
   mspace
   width="0.16em"></mspace><mi>A</mi><mi>c</mi><mi>t</mi><mi>i</mi><mi>v</
   mi><mi>i</mi><mi>t</mi><mi>y</mi><mo>=</mo><mi>s</mi><mi>a</mi><mi>t</m
   i><mrow><mo>(</mo><mrow><msub><mrow><mi>x</mi></mrow><mrow><mn>1</mn></
   mrow></msub><mo>-</mo><msub><mrow><mi>x</mi></mrow><mrow><mn>2</mn></mr
   ow></msub><mo>,</mo><mn>0</mn><mo>,</mo><mn>1</mn></mrow><mo>)</mo></mr
   ow><mo>.</mo> :MATH]
   2

   This “Pathway Activity” variable coarsely represents an amalgamated
   capacity for learning new events in the short-term. Based on the
   assumption of a finite amount of downstream effectors that interpret
   pathway activity, we define
   [MATH:
   <mi>E</mi><mi>f</mi><mi>f</mi><mi>e</mi><mi>c</mi><mi>t</mi><mi>o</mi><
   mi>r</mi><mi>s</mi><mspace
   width="0.16em"></mspace><mi>A</mi><mi>v</mi><mi>a</mi><mi>i</mi><mi>l</
   mi><mi>a</mi><mi>b</mi><mi>l</mi><mi>e</mi><mo>=</mo><mn>1</mn><mo>-</m
   o><mi>P</mi><mi>a</mi><mi>t</mi><mi>h</mi><mi>w</mi><mi>a</mi><mi>y</mi
   ><mspace
   width="0.16em"></mspace><mi>A</mi><mi>c</mi><mi>t</mi><mi>i</mi><mi>v</
   mi><mi>i</mi><mi>t</mi><mi>y</mi><mo>.</mo> :MATH]
   3

   We specify that “Memory” is a function of pathway and effectors
   dynamics by the following logic. In the absence of any past event, we
   can calculate the peak of Pathway Activity elicited by a particular
   event. This peak value is taken as the amount of capacity required to
   fully learn, which we call “need”. Then, we can calculate the Effectors
   Available elicited by a particular event as a function of time, given
   that other events may have already occurred previously, which we call
   “have”. Memory at each time point is defined as the Pathway Activity
   attributable to a particular event, divided by its maximum value, but
   weighted by the fraction have/need. Specifically,
   [MATH:
   <mi>M</mi><mi>e</mi><mi>m</mi><mi>o</mi><mi>r</mi><mi>y</mi><mo>=</mo><
   mfrac><mrow><msub><mrow><mrow><mo>(</mo><mrow><mi>P</mi><mi>a</mi><mi>t
   </mi><mi>h</mi><mi>w</mi><mi>a</mi><mi>y</mi><mspace
   width="0.16em"></mspace><mi>A</mi><mi>c</mi><mi>t</mi><mi>i</mi><mi>v</
   mi><mi>i</mi><mi>t</mi><mi>y</mi></mrow><mo>)</mo></mrow></mrow><mrow><
   mi>i</mi></mrow></msub></mrow><mrow><mo>max</mo><msub><mrow><mrow><mo>(
   </mo><mrow><mi>P</mi><mi>a</mi><mi>t</mi><mi>h</mi><mi>w</mi><mi>a</mi>
   <mi>y</mi><mspace
   width="0.16em"></mspace><mi>A</mi><mi>c</mi><mi>t</mi><mi>i</mi><mi>v</
   mi><mi>i</mi><mi>t</mi><mi>y</mi></mrow><mo>)</mo></mrow></mrow><mrow><
   mi>i</mi></mrow></msub></mrow></mfrac><mfrac><mrow><mi>h</mi><mi>a</mi>
   <mi>v</mi><mi>e</mi></mrow><mrow><mi>n</mi><mi>e</mi><mi>e</mi><msub><m
   row><mi>d</mi></mrow><mrow><mi>i</mi></mrow></msub></mrow></mfrac><mo>,
   </mo> :MATH]
   4

   where subscript i here denotes a particular learning input event. Thus,
   if there were not enough “Effectors Available” at the time of an
   event’s stimulus, have/need is reduced, and thus Memory is lowered.

   All simulations were performed in MATLAB (The Mathworks, Natick, MA)
   and the code is available upon request.

Reporting summary

   Further information on research design is available in the [281]Nature
   Research Reporting Summary linked to this article.

Supplementary information

   [282]Supplementary Information^ (37.1MB, pdf)
   [283]Reporting Summary^ (81.2KB, pdf)
   [284]41467_2019_11781_MOESM3_ESM.pdf^ (173.8KB, pdf)

   Description of Additional Supplementary Files
   [285]Supplementary Data 1^ (19.5KB, xlsx)
   [286]Supplementary Data 2^ (13.9KB, xlsx)
   [287]Supplementary Data 3^ (22.9KB, xlsx)
   [288]Supplementary Data 4^ (13.2KB, xlsx)

Acknowledgements