Hippocampal and cortical memory circuits, which enable the capacity to remember details from one’s past, are believed to reach maturity by early adolescence. Here, we demonstrate that the transition from early to late adolescence involves extensive reorganization of the retrosplenial (RSP) cortex along with significant memory expression deficits. Specifically, the densities of perineuronal nets (PNNs) and the expression of parvalbumin (PV) established in mice RSP during early adolescence (p30) significantly declined by late adolescence (p60–p75). In parallel, using context fear conditioning, we found that memories acquired during early adolescence were significantly impaired. These cellular and memory changes could be alleviated by PNN stabilization, indicating that they were secondary to PNN loss. Consistent with this, memory expression spontaneously recovered with PNN build-up in later life. Furthermore, late adolescence mice show a decrease in the level of key PNN constituents (especially aggrecan and neurocan) and the expression of transforming growth factor beta (Tgfβ) in RSP. RSP-targeted infusion of Tgfβ2 attenuated the decrease of PNN and neurocan, supporting a role of Tgfβ2 signaling in PNN build-up. These indicate that the decrease of PNNs during late adolescence was likely due to multiple molecular adaptations. Together, our findings show that RSP ECM undergoes dynamic reorganization until adulthood and beyond, with particularly strong fluctuations during late adolescence. In addition to affecting the expression of remote memories, in susceptible individuals, the observed dynamics could interact with genetic factors, increasing the risk of late adolescent psychopathologies.

Changes of PNN densities in the RSP-DH circuit at the transition between early and late adolescence

We visualized PNNs with wisteria floribunda agglutinin (WFA) labeling, and quantified them in the DH, dorsal SUB, and RSP (Fig 1A) from pre-adolescence to late adolescence (p21–p75). The density (reported as number of cells/0.1 mm²) of PNNs in DH (Fig 1B) remained stable in CA2 and even increased with age in CA1, CA3, and dentate gyrus (DG). On the other hand, in SUB and RSP of both males and females (Figs 1C, 1D, and S1), PNN densities were high at p21 and p30, then significantly decreased between p45 and p75 (age effect: SUB: p = 0.001, RSP: p < 0.001).

These findings suggested that PNNs in RSP were established by early adolescence (p21). But as the DH progressively matured, SUB and RSP demonstrated significant PNN reductions, suggesting possible circuit instability in RSP.

Late adolescent changes of RSP PV expressions

We next investigated whether the decrease of PNNs might be related to changes of RSP PV interneurons. Quantification revealed that the density of PV⁺ interneurons closely paralleled the changes of PNNs, showing a significant decrease at p60 and further reduction by p75 (ANOVA main effect: WFA⁺: p < 0.001, WFA and PV⁺: p = 0.001, PV⁺: p = 0.003, Fig 2A).

Developmental expansion of the entire mouse cortex has been reported during adolescence [45], which could contribute to the observed decrease of PNN and PV densities. Therefore, we estimated the volume change of targeted RSP region using two independent reference datasets, the Allen Developing Mouse Brain Atlas (ADMBA) [46] and the EBRAINS Developmental mouse brain atlas (DeMBA) [47]. Interestingly, both datasets showed a decrease, rather than an expansion, in RSP volume at p56 compared with p28 (ADMBA: 6.75% decrease; DeMBA: 7.43% decrease). These findings do not support the possibility that expansion of the targeted RSP region accounts for the observed decrease in PNN and PV densities.

We further asked if the reduced PV signal is due to the developmental change of PV expression within PV interneurons or the loss of these neurons. We injected AAV vector carrying Cre-dependent GFP into the RSP of PV-Cre mice at p28, labeling PV interneurons throughout this developmental period. At p75, we observed a lower density of immunolabeled PV⁺ cells compared to GFP⁺ cells in the RSP of these animals (p = 0.001, Fig 2B). This indicates that a subset of the cells previously expressed PV showed reduced PV expression by p75, contributing to the observed reduction in PV immunoreactivity. In addition, we observed low PNN coverage among cells that previous express PV but no longer showed detectable PV immunoreactivity (GFP⁺PV⁻) compared with cells that maintained current PV expression (PV⁺, p < 0.001, Fig 2B). These results suggest that reduced PNN expressions may be associated with the developmental changes in PV expressions. The reduction of the PV expression in late adolescent RSP was specific for this cell population and did not involve similar changes of microglia, as found in the prefrontal cortex around p60 [48], whereas astrocytes and oligodendrocytes did not show a similar decline and even slightly increased between early and late adolescence (p < 0.001, S2 Fig).

Effects of late adolescent RSP reorganization on recent and remote memory expression

To better understand the potential impact of the observed RSP changes on memory functions, we measured the formation and expression of stress-induced context memories acquired during early or late adolescence. We first trained the mice in one-trial CFC (3 min CtxA exposure, 2 s footshock, 0.7 mA, constant current) on p29 or p75 and tested them every two weeks thereafter for memory-induced freezing behavior in response to context re-exposure. Mice of both sexes trained on p29 showed impaired CFC memory by p75, as revealed by a significant decrease of their freezing levels (age at test effect: p < 0.001), whereas mice trained at p75 retained their memories within a similar 6-week time frame by p120 (age at test effect: p = 0.744, Fig 3A). This effect was replicated in a follow-up experiment whose objectives were to determine: (i) whether the impairment might have been due to enhanced extinction induced by repeated testing; (ii) whether the adolescent memory lacked specificity that might have affected its persistence; and (iii) whether hippocampus-independent associative memories, such as delay tone-dependent fear conditioning, undergo similar impairments. We exposed the mice on p28 to a modified fear conditioning paradigm allowing for the determination of context specificity/discrimination [49,50] and for assessment of both CFC and delay tone-dependent fear conditioning (Fig 3B). The paradigm consisted of a 3-min pre-exposure to CtxA and CtxB. Mice were then fear conditioned by exposing themto CtxA (3 min), tone (10 kHz, 30 s), and footshock (0.7 mA constant current, 2 s), once a day over 3 days. Memory tests performed one day (p32) or one month later (p61) replicated the significant impairment of early adolescent context-shock memory (Fig 3B, middle, p32 versus p61 in CtxA: p < 0.001) without affecting the tone-shock memory (Fig 3B, right, p = 0.516), which does not depend on the DH-RSP circuit. At all ages between p25 and p65, mice showed an intact ability to differentiate between CtxA and CtxB where shock was never delivered (p32: p < 0.001; p61: p = 0.036, also see S3 Fig). The inability to express a previously acquired context memory was unlikely due to adolescent changes of fear or anxiety expression, given that mice trained during late adolescence were able to form long-lasting, context-specific memories (S4 Fig). We performed additional experiments to examine whether the observed PNN dynamics were caused by RSP-intrinsic mechanisms or by circuit adaptations caused by the maturation of hippocampal-RSP projections. Using reporter mice expressing green fluorescent protein (GFP) driven by the neurotensin (NTS) gene to label DH/SUB to RSP projections, we found that indeed, DH maturation involved changes in the connectivity between the SUB and RSP (especially disappearance of a SUB-RSP layer 5 projection, S5A–S5C Fig). However, despite these circuit adaptations, inactivation of DH-RSP projections showed no interference with the expression deficits of early adolescent memories (S5D Fig).

Two subsequent experiments were devised to determine whether the observed memory impairments reflected memory degradation, or limited memory access. In the first study, we investigated whether early adolescent context memories can be reinstated as a function of re-experiencing the footshock in a new context (CtxB). The second study was devised to provide a replicate of the reinstatement effect, with an additional, no-shock control. Data from both studies were presented together in Fig 3C, whereas findings for each experiment separately are shown in S6 Fig. Overall, the findings replicated the adolescent memory expression deficit (p30 versus p74: p < 0.001), but after subsequent CFC in CtxB, freezing to CtxA significantly increased (p74 versus p76 reinstatement test: p = 0.014), indicating reinstatement of CtxA memory. This effect was not found in no shock controls, ruling out CtxB-induced generalization (S6 Fig).

Together, these findings demonstrated that although stress-induced context memories acquired during early adolescence are initially robust and specific, they showed delayed expression deficits (relative to memories acquired during adulthood), coinciding with a decrease of RSP PNN and PV interneuron densities. The impairment was sex-independent and not due to memory loss, inability to express fear, or general impairments in retrieving negative associative memories.

Effect of RSP PNN stabilization on cellular and memory impairments during adolescence

To determine whether the late reduction of remote context memory expression and PV expressions were directly linked to PNN degradation, we next investigated whether their decrease can be prevented by stabilizing PNNs using RSP-targeted injections of hyaluronan and proteoglycan link protein 1 (HAPLN1, Fig 4A). This approach was selected based on the documented efficacy of HAPLN1 to stabilize the ECM in various in vivo models [51,52]. Injections of HAPLN1 alleviated the impairments of adolescent memory expression (Age of test × HAPLN1 interaction: p = 0.032, p60: Veh versus Hapln1: p = 0.038), the degradation of PNNs (p < 0.001), and the reduction of cells showing PV expression (p = 0.021) relative to vehicle (Veh)-injected control mice (Fig 4B and 4C). These findings indicated that PNN degradation directly contributed to the loss of PV expressions, identifying PV interneurons as a vulnerable cell population with respect to changes of cortical ECM dynamics, and that remote memory expression deficits were secondary to the PNN/PV decline.

We also explored whether early adolescent context memories might spontaneously recover as mice age. To this end, we extended the period for memory assessment and observed a significant spontaneous recovery by middle adulthood (p120–p150). Memory recovery was accompanied by substantial generalization, as revealed by similar freezing in previously shock-associated CtxA and novel context CtxB (p30 versus p90: p < 0.001, p90 versus p150: p = 0.002, p90 versus p180: p = 0.012, p180/CtxA versus p181/CtxB: p = 0.997, S7A Fig). To determine whether further PNN/PV changes could account for these effects, we quantified the densities of RSP PNNs and PV⁺ cells on p30 and p150. Unexpectedly, the densities of PNNs and PV⁺ cells on p150 not only recovered but exceeded the levels found in early adolescence. However, there was a substantial increase of the proportion of WFA-negative PV interneurons at p150 relative to p30 (p < 0.001, S7B Fig). This was not due to the formation of PNNs around other main interneuron classes (S7C Fig, p30 versus p150: PV: p < 0.001 SST: p = 0.999, VIP: p = 0.690, NPY: p > 0.999, CR: p > 0.999). This supported our findings linking PNNs to memory expression, but also suggested that the memory fidelity might be additionally dependent on the PNN/PV interneuron overlap.

RSP PNN-associated lectican levels from early to late adolescence

We next investigated possible causes of PNN loss by examining the levels of key PNN constituents-the lecticans aggrecan (Acan), neurocan (Ncan), brevican (Bcan), and phosphacan (Pcan). The detection specificity of these proteoglycans (Fig 5A) was validated by the significant correlation between WFA and Acan in RSP (R² = 0.229, p = 0.002) and SUB (R² = 0.337, p = 0.005) and with Acan (R² = 0.251, p < 0.001) and Ncan (R² = 0.349, p < 0.001) in DH (Fig 5B). Age-dependent analysis revealed a profound and persistent decrease of Acan (Fig 5C: p30 versus p75: p = 0.001) and Ncan (Fig 5D, male: p30 versus p75: p = 0.017, female: p21 versus p75 p = 0.046) in PNNs of both p75 males and females relative to pre-adolescent (p21) or early adolescent (p30) mice. Bcan levels were also reduced in females (Fig 5E, p30 versus p75: p = 0.040). In addition, we observed a decrease of Pcan across RSP layers (S8 Fig). Although lecticans in the DH also showed some age-related changes, they mainly involved increased incorporation of Acan and Ncan in PNNs (S9A Fig, age effect: Acan male: p = 0.011, Acan female: p = 0.039, Ncan male: p < 0.001), whereas SUB showed similar decreases as RSP (S9B Fig, age effect: Acan male: p = 0.022, Acan female: p < 0.001, Ncan male: p = 0.036). Pcan (age effect: DH: male: p < 0.001, female p = 0.001, SUB: male: p = 0.003, female p < 0.001) and Bcan (in females only, DH: p < 0.001, SUB: p = 0.006) showed age-related decrease in all extracellular matrix compartments.

The late adolescent decrease of lectican levels was consistent the decrease of PNN densities during late adolescence and could contribute, at least in part, to their degradation. Interestingly, although PNN levels, PV expressions, and memory deficits were similar in males and females, individual lectican levels (e.g., Bcan in females) can be influenced by sex-specific factors.

Late adolescent changes of Tgfβ and Tgfβ-mediated effects on PNN and lectican levels

Previous studies have shown that lectican gene expression is controlled through extracellular levels of active Tgfβ1 and Tgfβ2 [25,53]. We therefore examined whether this pathway could be linked to some of the observed PNN and lectican changes. A qPCR analysis revealed a significant reduction of Tgfβ, and especially Tgfb2 mRNA expression in RSP of p75 mice compared to p30, whereas no significant change was observed in DH in the same period (RSP p30 versus p75: Tgfb1: p = 0.003, Tgfb2: p < 0.001, Fig 6A). These changes were associated with epigenetic changes of the Tgfb2 gene, as revealed by multiple, bidirectional changes of Tgfb2 gene body methylation (Fig 6B). To determine whether restoring Tgfβ2 levels could reverse some of the late adolescent changes of the extracellular matrix, we performed one RSP injection of Tgfβ2 at p44 (at the time of first detection of PNN decrease, Fig 1D), and quantified the levels of PNNs, the most significantly affected lecticans (Ncan and Acan), and PV one week later. A single RSP infusion of Tgfβ2 on p44 significantly increased the PNN densities (7 d after infusion, p = 0.039), the level of Ncan (p < 0.001), and the expression of PV (p = 0.002, Fig 6C). No significant effect was observed on Acan. These findings indicate that reduced Tgfβ2 expression occurring between early and late adolescence in RSP, driven at least in part by epigenetic changes of the Tgfb2 gene, could contribute to the observed decline of PV, select lecticans and, consequently, decrease of PNN densities.

Ethics statement

All procedures were approved by Northwestern University Animal Care and Use Committee (protocols IS00002463 and IS00003359) and Albert Einstein College of Medicine Animal Care and Use Committee (protocols 00001289 and 00001268) in compliance with US National Institutes of Health standards.

Animals

Wild-type (C57BL/6N) male and female mice were obtained from Envigo (Lafayette, IN, USA). Pvalb-IRES-Cre (PC-Cre, B6.129P2-Pvalb^tm1(cre)Arbr/J, #017320) [106] and Neurotensin-IRES-tauGFP (NT-GFP; strain Nts^tm1Mom/MomJ, #006702) [107] mice were obtained from Jackson Laboratory (Bar Harbor, ME, USA) and bred in the institutional animal facility. Animals were maintained under 12 h:12 h light:dark cycle with lights on at 07:00, temperature 20–22 °C, humidity 30%–60% with ad libitum access to food and water.

Contextual fear conditioning

Contextual fear conditioning (CFC) was performed in automated system (TSE Systems, Berlin, Germany or Med Associates, Fairfax, VT, USA). Mice were exposed to Context A (CtxA, rectangular shape, metal shock grid floor, cleaned with 70% ethanol before each test) for 3 min, followed by a foot shock (0.7 mA, 2 s). On indicated days of test, freezing level of mice to the context were measured in CtxA or Context B (CtxB, triangular shape, white plastic floor, cleaned with 1% acetic acid before each test) for 3 min. Mice were removed immediately after shock delivery or upon completion of the 3 min test period. In most experiments, mice were tested once a week until p60 or p75, except for a spontaneous recovery experiment, in which mice were tested once a month until p181.

CFC discrimination/generalization

The contextual discrimination/generalization task was performed over 4 days to explicitly determine the ability of mice to discriminate between CtxA and CtxB. On day 1, the mice were allowed to explore CtxA and CtxB, each for 3 min. On day 2–4, mice were exposed to CtxA for 3 min followed by a tone (30 s, 75 dB SPL, 10 kHz, 200 ms pulse) and a foot shock (0.7 mA, 2 s, constant current) sequentially once each day. On the day of test, freezing levels were measured during exposure of mice to CtxA and CtxB for 3 min each. The same tone used on day 2–4 was played at the end of CtxB test for 30 s to measure the freezing level to the auditory cue. Exposures to CtxA and CtxB were always spaced 3 h apart and counterbalanced. To alleviate the confounding effects of generalization found with the passage of time among similar contexts [108], we used dissimilar contexts that did not show time-dependent generalization (see S4 Fig).

CFC reinstatement

Memory reinstatement was determined by measuring the effect of early adolescent CtxA CFC on late adolescence CtxB CFC. We subjected the mice to CFC in CtxA on p29 followed by 2 tests on p30 and p74. On p75 they were exposed to CtxB (as previously described, except the floor was replaced with shock grid) for 3 min and received a shock reminder (0.7 mA, 2 s) at the end of the exposure. On p76 freezing was determined in CtxA and CtxB for 3 min each, spaced 3 h apart, and counterbalanced. Interference of prior CFC in CtxA with freezing to CtxB and generalization of freezing to CtxA after CFC in CtxB served as indices of CtxA memory retention.

Stereotaxic intracranial injection

Mice were anesthetized with isoflurane and placed on a stereotaxic instrument (Model 1900, Kopf Instruments, Tujunga, CA, USA). Viral vectors carrying AAV8-hSyn-mCherry (114472-AAV8, Addgene, gift from Karl Deisseroth), AAV8-hSyn-DIO-mCherry (50459-AAV8, Addgene, gift from Bryan Roth), AAV8/rg-hSyn-Cre-dTomato (107738-AAV8/rg, Addgene, gift from Rylan Larsen), AAVDJ-CMV-DIO-eGFP (GVVC-AAV-12, Stanford University Gene Vector and Virus Core), or AAVDJ-CMV-DIO-eGFP-2A-TeNT (Cre-Dependent Tetanus Toxin, GVVC-AAV-71, Stanford University Gene Vector and Virus Core) is delivered bilaterally into DH (1.8 mm posterior, 1.2 mm lateral, 2.25 mm ventral to bregma) and/or RSP (1.8 mm posterior, 1.2 mm lateral, 1.1 mm ventral to bregma) using a Hamilton microsyringe connected with microinfusing pump (UMP-3, WPI, Sarasota, FL, USA). For each site, 0.5 μL AAV solution (2 × 10¹² GC/mL) was delivered at 0.15 μL/min, and syringes were left in place for 5 min after the injection. Recombinant HAPLN1 (2608-HP-025, R&D Systems, Minneapolis, MN, USA, 100 ng/μL in PBS), Tgfβ2 (7346-B2-005, R&D Systems, 10 ng/μL in PBS) was injected bilaterally into RSP (0.5 μL/site) as described above.

Immunohistochemistry

Mice were anesthetized with an intraperitoneal injection of 240 mg/kg Avertin and transcardially perfused with ice-cold phosphate buffer followed by 4% paraformaldehyde in phosphate buffer (pH 7.4, 50 mL each per mouse). Brains were removed and post-fixed for 24 hours in the same fixative and then cryoprotected by immersing in 20% and 30% sucrose in phosphate buffer for 24 hours each. After that the brains were frozen in OCT embedding medium and 50 µm freezing microtome sections were made for use in free-floating lectin histochemistry or immunohistochemistry. Biotinylated WFA (1:1,000 B-1355–2, Vector Laboratories, Newark, CA, USA) was used to visualize PNNs. For immunohistochemistry, primary antibodies against aggrecan (1:1,000, ab3778, Abcam, Cambridge, UK), neurocan (1:2,000, AF5800 R&D systems), brevican (1:500, 19017-1-AP, Proteintech, Rosemont, IL, USA), phosphacan (1:500, sc-33664, Santa Cruz, Dallas, TX, USA), PV (1:5,000, pvg-213, Swant, Burgdorf, Switzerland), SST (1:1,000, MAB354, Millipore, Burlington, MA, USA), VIP (1:1,000, 20077-1513001, Immunostar, Hudson, WI, USA), NPY (1:1000, 22940-1628001, Immunostar), CR (1:4,000, CG1, Swant), Iba1 (1:1,000, ab5076, Abcam), Olig2 (1:2,000, ab109186, Abcam), PDGFRa (1:2,000, AF1062, R&D Systems), and s100b (1:2,000, 287108, Synaptic Systems, Goettingen, Germany) were used followed by corresponding secondary antibodies (1:500, Alexa Fluor conjugated, Jackson IR, West Grove, PA, USA or 1:200, Biotinylated, Vector Laboratories). When using biotinylated systems, the signals were visualized with the ABC-HRP Detection Kit (PK-6100, Vector Laboratories) and TSA Fluorophores (Akoya Biosciences, Marlborough, MA, USA).

Microscopy, image analysis and quantification

For large scale analyses, sections were scanned using Axioscan Z1 microscope (Zeiss, Oberkochen, Germany) scanner under 10×/0.45 air objective. Images were stitched and converted to TIFF format. RSP findings were additionally verified following confocal microscopy under 60×/1.35 oil objective (Fluoview FV10i-LIV, Olympus, Hachioji, Tokyo, Japan). For each experiment, the same acquisition profile, including camera and excitation settings, was used across all experimental groups. All quantifications were performed with ImageJ using two methods: the multi point plug-in and analyze particles plug-in. Intensity was calculated by measuring the gray value after thresholding to eliminate background interference. Cell counting and gray value analysis were performed in a 0.1 mm² area for each ROIs. The identical threshold was applied across experimental groups. Two sections per animal were collected and the measurements were averaged. Each animal was treated as an independent sample for statistical analysis.

Volume analyses were performed within targeted RSP region starts at the first appearance of the suprapyramidal blade granular layer cells and ends at the first appearance of subcommissural organ (anterior-posterior direction, coronal plates). ADMBA p28 and p56 coronal Nissl datasets were downloaded using Allen API. Images were inspected and the distance between end plates is obtained from the metadata. Areas of RSP were measured using ImageJ at 200 µm intervals to calculate the volume. DeMBA volume and segmentation files (CCFv3, 2022 edition) were downloaded from ebrains.eu. Data were visualized with ITK-SNAP [109] to determine range of z-plates. RSP voxels within the range were counted and converted to volumes using NiBabel [110] python package.

Quantitative PCR analysis

Mice subjected to fresh tissue collection were euthanized with cervical dislocation. After that, the brain was removed and DH and RSP tissue were dissected from the brain. Total RNA was extracted using RNeasy Plus Mini Kit (74136, Qiagen). Reverse transcription was performed on 100 ng of total RNA using PrimeScript RT Reagent Kit (RR037A, Takara, Kusatsu City, Japan). Quantitative PCR analysis was performed on a QuantStudio 6 Flex instrument (Applied Biosystems, Waltham, MA, USA) using SYBR green detection system (4367659, Applied Biosystems) and primers specific for Tgfb1 (QT00145250, Qiagen), Tgfb2 (QT00106806, Qiagen), and Tgfbr2 (QT00135646, Qiagen). GAPDH (F: AACTTTGGCATTGTGGAAGG; R: ACACATGGGGGTAGGAACA) was used as endogenous control. mRNA levels were calculated using the comparative CT method (2^−ΔΔCt).

DNA methylation assay

DNA was extracted using the Qiagen AllPrep DNA/RNA/miRNA Universal Kit (Qiagen; Hilden, Germany; cat# 80224) per the manufacturer’s protocol. 500 ng of extracted DNA samples were bisulfite modified using the EZ-96 DNA Methylation Kit (ZymoResearch; Irvine, CA; cat# D5004). After that, all bisulfite-modified DNA samples were amplified using separate multiplex or simplex PCRs. Samples were run alongside established reference DNA with 0%, 5%, 10%, 25%, 50%, 75%, and 100% methylation. Sequencing libraries were prepared using a custom library preparation method created by EpigenDx. Next, library molecules were purified using Agencourt AMPure XP beads (Beckman Coulter; Brea, CA; cat# A63882). Barcoded samples were then pooled in an equimolar fashion before template preparation and enrichment were performed on the Ion Chef system using Ion 520 and Ion 530 ExT Chef reagents (Thermo Fisher; Waltham, MA; cat# A30670). Following this, enriched, template-positive library molecules were sequenced on the Ion S5 sequencer using an Ion 530 sequencing chip (cat# A27764). For data analysis, FASTQ files from the Ion Torrent S5 server were aligned to a local reference database using the open-source Bismark Bisulfite Read Mapper program (v0.12.2) with the Bowtie2 alignment.

Statistical analysis

Statistical power to detect anticipated effect sizes was determined using power analysis (calculator at http://www.stat.ubc.ca/~rollin/stats/ssize/n2.html) conducted on representative samples of previous work and pilot experiments. For all proposed experiments, minimum power is set at 0.90 to detect an α = 0.05 (two-sided test) for a difference in means from 20% to 40%, with a 15% common standard deviation. Mice are randomly assigned to groups. However, to prevent litter effects, mice from the same litter were assigned to different experimental groups. Viruses and proteins were injected by experimenters aware of the construct, but the mice were then assigned coded numbers by the laboratory technician. The code was available after quantification and before analyses. Statistical analyses were performed using GraphPad Prism. Outlier mice exhibiting abnormal behavior performance determined by Grubbs’ test or with misplaced infusion were excluded. One-way ANOVA or Two-way ANOVA followed by indicated post hoc tests were used for comparisons of three or more experimental groups (only when ANOVA was significant) whereas Student t test was used for comparison of two experimental groups. Homogeneity of variance was confirmed with Levene’s test for equality of variances. Repeated measures (RM) ANOVAs with Geisser Greenhouse correction were performed for within-subject designs. On indicated data, we performed correlation analyses and report Pearson’s r coefficients.

All comparisons were conducted using two-tailed tests and the P value for all cases was set to <0.05 for significant differences. Data are expressed as mean ± s.e.m. Statistically significant differences are indicated as *p < 0.05, **p < 0.01, and ***p < 0.001.

S1 Fig. PNN changes throughout adolescence by region and sex.

(A) Left: The standard view that human adolescence encompasses a period from roughly the age of 10–19 years (yellow) was recently revised to 10–24 years (green) to reflect changes of biological growth and other factors [57,58]. The approximate human-mouse age alignment was calculated according to Dutta and colleagues [54]. The experimental design and nomenclature were aligned with the later view; right: representative images of p21, p30, p45, p60, and p75 male brain sections stained with WFA. (B) Representative images for WFA staining in RSP (top, males) and quantification for the density of PNN⁺ cells demonstrating significant post-adolescent decreases in the density of PNNs (males bottom left; p21: n = 3, p30 and p75: n = 5, p45 and p60: n = 4; one-way ANOVA: F = 8.124, p < 0.001; Tukey’s post-hoc test: p30 vs. p45: p = 0.002, p30 vs. p60: p = 0.002; females bottom right, p21: n = 3, p60: n = 4, others: n = 5; one-way ANOVA: F = 9.032, p < 0.001; Tukey’s post-hoc test: p30 vs. p45: p < 0.001, p30 vs. p60: p = 0.002) RSP. (C) Representative images for WFA staining in SUB (top, males) and quantification for the density of PNN⁺ cells revealing similar decreases in SUB (males bottom left; p21: n = 3, p30: n = 5, others: n = 4; one-way ANOVA: F = 3.197, p = 0.044; Tukey’s post-hoc test: p30 vs. p75: p = 0.025; females bottom right, p21: n = 3, p60: n = 4, others: n = 5; one-way ANOVA: F = 3.429, p = 0.031; Tukey’s post-hoc test: p30 vs. p60: p = 0.034) SUB. (D) Representative images for WFA staining in DH (top, males) and quantification for the density of PNN⁺ cells showing stable or increased density of PNNs (males bottom left; p21: n = 3, p30 and p75: n = 5, p45 and p60: n = 4; two-way RM ANOVA: Region: F_{(1.905,30.48)} = 95.74 p < 0.001, Age: F_{(4, 16)} = 3.533 p = 0.030, Region × Age: F_(12,48) = 1.273 p = 0.265; Tukey’s post-hoc test: CA3: p21 vs. p75: p = 0.048, DG: p21 vs. p45: p = 0.035, p21 vs. p60: p = 0.003, p21 vs. p75: p < 0.001, p30 vs. p75: p = 0.036, p45 vs. p75: p = 0.048, p60 vs. p75: p = 0.020; females bottom right, p21: n = 3, p60: n = 4, others: n = 5; two-way RM ANOVA: Region: F_{(1.893,32.17)} = 149.1 p < 0.001, Age: F_{(4, 17)} = 7.984 p < 0.001, Region × Age: F_(12,51) = 0.928 p = 0.527; Tukey’s post-hoc test: CA1: p21 vs. p60: p = 0.019, CA3: p60 vs. p75: p = 0.031) CA1, CA2, CA3, and DG. Data represent mean ± s.e.m., ***p < 0.001; **p < 0.01; *p < 0.05. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003908.s001

(TIF)

S2 Fig. Developmental changes of glial cells in RSP.

Left: Density of microglia, astrocyte, oligodendrocyte precursor cell (OPC), and oligodendrocyte showing an increase for astrocyte and oligodendrocyte (n = 6; microglia: one-way ANOVA: F = 0.006640, p = 0.936; astrocyte: one-way ANOVA: F = 6.658, p = 0.009, Tukey’s post-hoc test: p30 vs. p60: p = 0.023, p30 vs. p75: p = 0.013; OPC: one-way ANOVA: F = 1.018, p = 0.385; oligodendrocyte: bottom right, F = 32.33, p < 0.001, Tukey’s post-hoc test: p30 vs. all: p < 0.001, p60 vs. p75: p = 0.024); right: representative images. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003908.s002

(TIF)

S3 Fig. Context specificity of CFC memories acquired during early adolescence (p25–p35).

Context discrimination task (top) shows that mice were able to form specific contextual fear memory after p28 (n = 9/age group and sex, bottom left, three-way RM ANOVA, Sex: F_(1,62) = 1.413, p = 0.239, Age: F_(3,62) = 0.7597, p = 0.521, Ctx: F_(1,62) = 588.2, p < 0.001, Age × Sex: F_(3,62) = 0.7293, p = 0.538, Age × Ctx: F_(3,62) = 34.50, p < 0.001, Sex × Ctx: F_(1,62) = 5.748, p = 0.020, Age × Sex × Ctx: F_(3,62) = 5.065, p = 0.003, Tukey’s post-hoc test: Ctx A vs. Ctx B in all groups other than p25 male and p25 female: p < 0.001). And that they show normal fear response to auditory cue since p21 (p32 male and p48 female: n = 8, other: n = 9, bottom right, two-way ANOVA: Age × Sex: F_(3,62) = 1.247, p = 0.300, Age: F_(3,62) = 1.069, p = 0.369, Sex: F_(1,62) = 0.3027, p = 0.584). Experimental diagrams were created in BioRender. Zhang, H. (2026) https://BioRender.com/kz4bcow. Data represent mean ± s.e.m., ***p < 0.001. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003908.s003

(TIF)

S4 Fig. Specificity of recent and remote contextual fear memory formed during late adolescence (p60–p63).

Contextual discrimination task performed on p60 mice shows a prolonged memory specificity in distinguishing dissimilar contexts (same setup used in this study for p28 mice, without tone) up to p150 (two-way RM ANOVA: Age of test: F_{(1.890, 18.90)} = 2.209 p = 0.139, Context: F_{(1, 10)} = 370.2 p < 0.001, Age of test × Context: F_{(1.897, 18.97)} = 2.532 p = 0.108). Experimental diagrams were created in BioRender. Zhang, H. (2026) https://BioRender.com/qmq1etq. Data represent mean ± s.e.m., ***p < 0.001. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003908.s004

(TIF)

S5 Fig. Effect of inactivating DH to RSP projections during late adolescence on the retrieval of previously acquired memories.

(A) Left, Delineation of brain areas showing strong NTS labeling in thalamic, hippocampal, and RSP regions during early development (p7–p14) in NTS-GFP mice [46]. Right, Representative images of NTS labeling showing two projections from SUB to RSP terminating in L2/3 and L5. (B) Disappearance of NTS from DH to RSP projections between p30 and p60. (C) Mature DH to RSP projections to RSP L1 and L2/3 (left), with excitatory projections terminating in L2/3 (middle) and lack of terminals in L5 (right). (D) Inhibition of DH to RSP projections does not affect the impaired retrieval of adolescent memories (n = 12, two-way RM ANOVA: Tettox: F_(1,21) = 0.1627 p = 0.691, Age of test: F_{(1, 21)} = 30.14 p < 0.001 Infusion × Age of test: F_(1,21) = 0.0040 p = 0.950; Šídák’s post-hoc test: Control p30 vs. p61: p = 0.002, Tettox p30 vs. p61: p = 0.002). Experimental diagrams were created in BioRender. Zhang, H. (2026) https://BioRender.com/m7zoc9h. Data represent mean ± s.e.m., **p < 0.01. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003908.s005

(TIF)

S6 Fig. Freezing behavior after shock re-exposure in a different context.

(A) Left, Experimental paradigm; Right, Shock in a novel context (Context B, CtxB) on p75 resulted in Context B-specific freezing, but also in increased freezing to the conditioned context (Context A, CtxA) (One-way RM ANOVA, F_{(2.431, 21.88)} = 15.01, p < 0.001, Tukey’s post-hoc test: p30 vs. p60: p = 0.004, p30 vs. p74: p < 0.001, p44 vs. p74: p = 0.002, p60 vs. p74: p < 0.001, p74 vs. p76: p = 0.011). (B) The experiment was repeated with a no-shock control group, and no tests performed between p30 and p74. Shock in CtxB on p75 did not increase freezing in CtxA in no shock group(n = 17 for shock group, n = 12 for no shock group; two-way RM ANOVA: Test: F_{(1.657, 44.75)} = 10.22, p < 0.001, Shock: F_{(1, 27)} = 51.12, p < 0.001, Test × Shock: F_{(1.657, 44.75)} = 15.35, p < 0.001; Tukey’s post-hoc test: shock group: p30 vs. p74: p < 0.001, p30 vs. p76: p = 0.041 p74 vs. p76: p = 0.004, no shock group: p30 vs. p76: p = 0.008, shock vs. no shock: p30: p < 0.001, p76: p = 0.005). Experimental diagrams were created in BioRender. Zhang, H. (2026) https://BioRender.com/wgbkg4j. Data represent mean ± s.e.m. ***p < 0.001; **p < 0.01; *p < 0.05. Mice that shocked on p30 and exhibited less than a 20% decline in freezing at p75 relative to p30 were excluded. See Fig 3C for data without this exclusion. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003908.s006

(TIF)

S7 Fig. Spontaneous recovery of early adolescent contextual fear memory and change of PNN-interneuron association in RSP by p150.

(A) Spontaneous recovery of adolescent aversive context memory by mid-adulthood/p150 (n = 16; one-way RM ANOVA: F = 8.025, p < 0.001; Tukey’s post-hoc test: p30 vs. p90: p < 0.001, p90 vs. p150: p = 0.002, p90 vs. p180: p = 0.012). (B) Increased proportion of RSP PNN-free PV interneuron in p150 compared to p30 (n = 6; unpaired two-tailed t test: t₁₀ = 8.539, p < 0.001). (C) Decreased co-localization between PV and PNNs was not due to increased PNN formation around other interneuron classes, which showed a significant lack of co-localization with WFA relative to PV neurons at p30 and p150 (two-way ANOVA: Interneuron type: F_{(4, 50)} = 560.8, p < 0.001 Age: F_{(1, 50)} = 33.53, p < 0.001, Interaction: F_{(4, 50)} = 39.44, p < 0.001, Šídák’s post-hoc test: p30 vs. p150: PV: p < 0.001, SST: p = 0.999, VIP: p = 0.690, NPY: p > 0.999, CR: p > 0.999). Experimental diagrams were created in BioRender. Zhang, H. (2026) https://BioRender.com/9yzhxu7. Data represent mean ± s.e.m. ***p < 0.001; **p < 0.01; *p < 0.05. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003908.s007

(TIF)

S8 Fig. Levels and developmental changes of phosphacan across RSP layers.

(A) Low magnification images demonstrating the cortical, hippocampal, and thalamic distribution of phosphacan and WFA at p21 (males n = 4, females n = 3), p30 (males, n = 5, females n = 5), and p75 (males n = 4, females n = 4). At p30, the highest levels are obtained in layer 1, containing the cortical axons. (B) High magnification images showing progressive post-adolescent down-regulation of phosphacan across layers and ECM compartments including but not restricted to PNNs. (C) Layer-by-layer quantification of phosphacan optical density demonstrating significant effects in males (two-way ANOVA: Age: F_{(2, 50)} = 42.32, p < 0.001, Layer: F_{(4, 50)} = 17.70, p < 0.001, Age × Layer: F_{(8, 50)} = 6.574, p < 0.001, Tukey’s post-hoc test: p21 vs. p30: p < 0.001, p21 vs. p75: p < 0.001, p30 vs. p75: p < 0.001) and females (two-way ANOVA: Age: F_(2,45) = 12.36, p < 0.001, Layer: F_(4,45) = 1.478, p = 0.225, Age × Layer: F_(8,45) = 0.5702, p = 0.797, Tukey’s post-hoc test: p21 vs. p30: p < 0.001, p21 vs. p75: p < 0.001). Data represent mean ± s.e.m., ***p < 0.001; **p < 0.01; *p < 0.05. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003908.s008

(TIF)

S9 Fig. Post-adolescent levels of PNN-associated lecticans in DH and SUB.

(A) Developmental decreases of PNN composition in DH (male aggrecan: p21: n = 4, other: n = 5, two-way ANOVA: Age: F_{(2, 44)} = 12.14, p < 0.001, Area: F_{(3, 44)} = 15.71, p < 0.001, Age × Area: F_{(6, 44)} = 3.146, p = 0.012, Tukey’s post-hoc test: p21 vs. p30: p < 0.001, p21 vs. p75: p = 0.011; female aggrecan: p21: n = 3, other: n = 5, two-way ANOVA: Age: F_{(2, 40)} = 3.515, p = 0.039, Area: F_{(3, 40)} = 14.21, p < 0.001, Age × Area: F_{(6, 40)} = 1.438, p = 0.225; male neurocan: p21: n = 3, other: n = 5, two-way ANOVA: Age: F_{(2, 40)} = 9.519, p < 0.001, Area: F_{(3, 40)} = 17.99, p < 0.001, Age × Area: F_{(6, 40)} = 0.2076, p = 0.972, Tukey’s post-hoc test: p21 vs. p30: p = 0.027, p21 vs. p75: p < 0.001; female neurocan: p21: n = 3, other: n = 5, two-way ANOVA: Age: F_{(2, 40)} = 2.812, p = 0.072, Area: F_{(3, 40)} = 16.57, p < 0.001, Age × Area: F_{(6, 40)} = 0.7006, p = 0.651; male brevican: p30: n = 5, other: n = 4, two-way ANOVA: Age: F_{(2, 40)} = 1.567, p = 0.221, Area: F_{(3, 40)} = 15.22, p < 0.001, Age × Area: F_{(6, 40)} = 1.302, p = 0.279; female brevican: p21: n = 2, p30: n = 4, p75: n = 5, two-way ANOVA: Age: F_{(2, 32)} = 15.37, p < 0.001, Area: F_{(3, 32)} = 4.618, p = 0.009, Age × Area: F_{(6, 32)} = 1.735, p = 0.145, Tukey’s post-hoc test: p21 vs. p30: p = 0.006, p21 vs. p75: p < 0.001, p30 vs. p75: p = 0.038; male phosphacan: p21: n = 4, other: n = 5, two-way ANOVA: Age: F_{(2, 70)} = 20.37, p < 0.001, Area: F_{(6, 70)} = 3.190, p = 0.008, Age × Area: F_{(12, 70)} = 1.307, p = 0.235, Tukey’s post-hoc test: p21 vs. p75: p < 0.001, p30 vs. p75: p < 0.001; female phosphacan: p21: n = 3, other: n = 5, two-way ANOVA: Age: F_{(2, 70)} = 7.449, p = 0.001, Area: F_{(6, 70)} = 5.615, p < 0.001, Age × Area: F_{(12, 70)} = 1.868, p = 0.054, Tukey’s post-hoc test: p21 vs. p75: p = 0.017, p30 vs. p75: p = 0.002). (B) Developmental change of PNN composition in SUB (male aggrecan: p30: n = 5, other: n = 4, one-way ANOVA: F = 5.746, p = 0.022; Tukey’s post-hoc test: p30 vs. p75: p = 0.017; female aggrecan: p21: n = 3, other: n = 5, one-way ANOVA: F = 124.9, p < 0.001; Tukey’s post-hoc test: p21 vs. p75: p < 0.001, p30 vs. p75: p < 0.001; male neurocan: p21: n = 3, p30: n = 5, p75: n = 4, one-way ANOVA: F = 4.951, p = 0.036; Tukey’s post-hoc test: p30 vs. p75: p = 0.047; female neurocan: p21: n = 3 other: n = 5, one-way ANOVA: F = 0.5794, p = 0.578; male brevican: p21: n = 3, p30: n = 5, p75: n = 4, one-way ANOVA: F = 0.2091, p = 0.815; female brevican: p21: n = 2, p30: n = 3, p75: n = 5, one-way ANOVA: F = 11.40, p = 0.006, Tukey’s post-hoc test: p21 vs. p75: p = 0.036, p30 vs. p75: p = 0.008; male phosphacan: p30: n = 5, other: n = 4, one-way ANOVA: F = 11.01, p = 0.003, Tukey’s post-hoc test: p21 vs. p75: p = 0.002; female phosphacan: p21: n = 3, other: n = 5, one-way ANOVA: F = 29.62, p < 0.001, Tukey’s post-hoc test: p21 vs. p30: p = 0.001, p21 vs. p75: p < 0.001, p30 vs. p75: p = 0.027). Data represent mean ± s.e.m., ***p < 0.001; **p < 0.01; *p < 0.05. The data underlying this Figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003908.s009

(TIF)