Astrocytic microdomains confine a “molecular memory” enabling long- 1 term information storage for memory consolidation

23 Memory consolidation requires astrocytic microdomains for protein recycling; but 24 whether this lays a mechanistic foundation for long-term information storage remains enigmatic. Here we demonstrate that persistent synaptic strengthening invited astrocytic 26 microdomains to convert initially internalized (pro)-brain-derived neurotrophic factor 27 (proBDNF) into active prodomain (BDNFpro) and mature BDNF (mBDNF) for synaptic 28 re-use. While mBDNF activates TrkB, we uncovered a previously unsuspected function 29 for the cleaved BDNFpro, which increases TrkB/SorCS2 receptor complex at post- 30 synaptic sites. Astrocytic BDNFpro release reinforced TrkB phosphorylation to sustain 31 long-term synaptic potentiation and to retain memory in the novel object recognition 32 behavioral test. Thus, the switch from one inactive state to a multi-functional one of the 33 proBDNF provides post-synaptic changes that survive the initial activation (molecular 34 memory). This molecular asset confines local information storage in astrocytic 35 microdomains to selectively support memory circuits.

Introduction regulated by proBDNF, suggesting that conversion of the inactive neurotrophin precursor 86 to an active product might play a more direct role. We now ask whether these same 87 astrocytes are proficient for direct cleavage of endocytic proBDNF following LTP-88 inducing electrical stimulation. 89 Brain slices of control mice were prepared to examine the astrocytic origin of the of proBDNF proteolytic processing was analyzed 10 min after stimulation by 97 immunohistochemistry using an antibody (αBDNFpro) that specifically recognizes the 98 furin cleaved C-terminal end of the prodomain ( Fig. 1a; (24)). This epitope is unavailable 99 in both intact proBDNF and mBDNF, providing that the antibody recognized the cleaved 100 BDNFpro, leaving undetected cleavage-resistant proBDNF (proBDNF CR ) and mBDNF in 101 the Western blot analysis (Fig. 1a). BDNFpro and GFP immunoreactivity was analyzed 102 by confocal microscopy to appreciate the specific distribution of the cleaved prodomain 103 in individual astrocytes ( Fig. 1b and Supplementary Fig. 1a). GFP is a cytosolic protein 104 whose fluorescence defines the astrocyte in its entire cytoplasmic extension. This is a 105 feature that is ideal to achieve detection of BDNFpro in the astrocytic territorial volume. 106 Spatial overlap of BDNFpro and GFP was analyzed in a series of confocal stacks by 107 6 using colocalization analysis of the two signals ( Supplementary Fig. 1c). To facilitate 108 BDNFpro visualization in the astrocytic territorial volume, BDNFpro/GFP colocalization 109 was reconstructed in z-stacks. We observed sharp BDNFpro/GFP colocalization signal, 110 as detected by αBDNFpro, in astrocytes from TBS-slices ( Fig. 1c and Supplementary Fig.   111 1b). Cleaved prodomain detection was observed in small proportion at the cell body and NeuN-neuronal marker (1 ± 0.7%), demonstrating that p75-flox mice allowed astrocytes 135 selective targeting in the brain cortex in vivo. Immunoreactivity for BDNFpro was 136 examined in slices from p75-flox mice 14 days post-tamoxifen (dptm). This latency 137 7 ensured a significant depletion of p75 NTR protein following recombination in astrocytes 138 ( Supplementary Fig. 2c). In p75 NTR -deficient cells, BDNFpro/GFP signal was hardly 139 detectable in both TBS and basal stimulation (Fig. 1d). Thus, p75 NTR -mediated proBDNF 140 endocytosis fed astrocytes with a cleavable pool of proBDNF.

141
Collectively, our data indicate that proBDNF is a substrate for proteolytic processing in 142 cortical layer II/III astrocytes following LTP-mediated internalization. Given that 143 endocytic proBDNF undergoes recycling in this potentiating condition (15), our data 144 suggest that astrocytes might convert proBDNF into BDNFpro and mBDNF at peri-145 synaptic sites before routing to the secretory pathway.

148
The intricate ramifications of astrocytes allow them to tightly enwrap the synaptic 149 terminals at organized peri-synaptic structures, the microdomains (31). Astrocytic 150 microdomains can be distinguished into thick processes of micrometer scale (∼10-15 151 μm 2 ) that host endoplasmic reticulum (ER) and mitochondria capable of generating 152 inositol-3-phosphate (IP3)-dependent Ca 2+ signals and thin organelle-free structures of 153 submicron/nanometer-scale that fill the space between synapses (32-34). We speculated 154 that astrocytic proBDNF processing might be achieved on a rapid time scale at 155 microdomains and, possibly, within the same storage compartments orchestrating the 156 recycling process. Vesicular localization of the cleaved BDNFpro targeting astrocytic 157 microdomains has been evaluated to be in line with our hypothesis.
158 Subcellular localization of BDNFpro was initially resolved in super-resolution by using 159 structured illumination microscopy (SIM) (Fig. 2a). In TBS-slices from control mice, 160 BDNFpro/GFP colocalization signal appeared as a punctate pattern dispersed into cell 161 periphery of astrocytes. At higher magnification, BDNFpro/GFP colocalization was 162 present in membrane ramifications mostly shaped as finger-like extensions and flat 163 lamellar sheaths (Fig. 2a, b), which were recognized to be astrocytic structures contacting 164 the synapse. Quantification analysis confirmed that TBS induced a significant increase of 165 the colocalization signal (45 ± 6% of the total BDNFpro/GFP puncta detected in 166 astrocytes) in these specific structures. Given the nanometer scale resolution of the SIM 167 8 super-resolution microscopy (lateral res. 115 nm; axial res. 250 nm), our data suggest for 168 an enrichment of the cleaved prodomain in high membranous elaborations of the cell 169 periphery viewing the dimensions and typical morphology of microdomains.

170
To firmly assess this conclusion, we extended our investigation at ultra-structural levels 171 (15). Ultra-thin sections from TBS-slices were examined by transmission electron 172 microscopy (EM) in pre-embedding experiments (Fig. 3a). At 70,000-100,000-fold 173 magnification, immunogold labeling was observed using αBDNFpro antibody 174 (αBDNFpro-gold) in (i) vesicular structures at pre-synaptic terminals typically displaying 175 clouds of synaptic vesicles opposed to post-synaptic density structures (Fig. 3b); (ii) post-176 synaptic terminals (Fig. 3b); and (iii) astrocytes giving rise to fine astrocytic processes in 177 close proximity to the synapses (Fig. 3c). Consistent with our data that BDNFpro in 178 astrocytes is generated from endocytic proBDNF processing in response to neuronal 179 activity, TBS-slices showed a higher number of αBDNFpro-gold grains in astrocytes 180 filling the space between synapses (38 ± 7% of the total gold particles detected in 181 astrocytes) with respect to non-stimulated slices (12 ± 4% of the total gold particles 182 detected in astrocytes). Most gold particles were organized in groups of many grains and 183 were concentrated peri-synaptically (46 ± 4% of the total gold particles detected in        Thus, BDNFpro is required for a limited time window to sustain LTP. 286 Collectively, our data indicate astrocytes to be as proficient for BDNFpro release, which 287 fulfills the function to mediate the switch from early-to late-phase LTP.  dendrites was predominately concentrated in proximity of spines, appearing as membrane 320 protrusion extending from dendritic processes (Fig. 6b). Quantitative analysis confirmed 321 the increase of PLA TrkB/SorCS2 -signal in BDNFpro-treated vs. vehicle-treated cultures (Fig.   322 6c). The specificity of BDNFpro treatment was assessed by pre-treating neurons for 20 323 min with αSorCS2 (20 µg/ml) (Fig. 6b), a blocking antibody that is known to prevent 324 TrkB/SorCS2 complex formation (42). In the presence of αSorCS2 pre-treatment, 325 BDNFpro fragment could no more exert its aggregating effect and PLA TrkB/SorCS2 -signal 326 could not be seen (Fig. 6c). Moreover, when neurons were treated with recombinant 327 mBDNF (10 ng/ml), or cleavage-resistant proBDNF CR (20 ng/ml), PLA TrkB/SorCS2 -signal 328 had not changed significantly as compared to vehicle-treated neurons (Fig. 6c); that 329 indicates that, within BDNF isoforms, TrkB/SorCS2 aggregation is a unique function of 330 BDNFpro. PLA analysis was extended to cortical slices from control mice 10 min after 331 basal-or TBS-stimulation (Fig. 6d). We measured in a quantitative analysis,  Finally, high-localized TrkB/SorCS2 aggregation would result in increased TrkB activity.

344
To assess this possibility, we examined TrkB phosphorylation (pTrkB) by using α-345 phospho-TrkB (Tyr816) antibody (αpTrkB). Given that pTrkB is essentially a fraction of 346 14 the total TrkB levels, we measured, in a quantitative analysis, pTrkB that colocalizes with 347 TrkB immunoreactivity. We found that the levels of pTrkB/TrkB colocalization had 348 significantly increased in TBS-slices compared to non-stimulated slices from p75-flox 349 mice injected with LV-BDNFpro stop or TBS-stimulated slices from control littermates 350 injected with LV-GFP stop (Fig. 7b and Supplementary Fig. 7). Moreover, this increase is 351 comparable to the one observed in TBS vs. baseline condition in control littermates (Fig.   352 7b). Thus, astrocytic BDNFpro recruits TrkB expression on adjacent spines for tight 353 temporal, spatial and stimulus-dependent TrkB phosphorylation.  (Fig. 8c). Thus, BDNFpro in astrocytes could reverse the memory defect 373 exhibited by p75-flox mice, demonstrating its necessity to memory consolidation in vivo.  transfer from astrocytes to neuron) that survives to LTP induction. However, our data 406 provide further significance to this molecular activation; not only mBDNF, but also the 407 BDNFpro turned out to possess an independent function. Once generated, this byproduct 408 participates in astrocytic recycling by inducing changes in TrkB levels (Fig. 7a) as well 409 as in this receptor phosphorylation at post-synaptic sites (Fig. 7b). Thus, BDNFpro 410 operates synaptic adaptations once LTP is expressed that are relevant for its later 411 stabilization (Fig. 5a, c). Mechanistically, BDNFpro acts as an activator of TrkB/SorCS2 412 aggregation by spine targeting (Fig. 6b-d). This is a new physiological role attributed to signaling, as it is required for a "molecular memory" to maintain LTP.

452
Spine targeting of TrkB tackles the underlying changes that occur at the synapse once 453 LTP is expressed, reinforcing the common knowledge that LTP maintenance take place 454 post-synaptically (6). However, our data also provide evidence that the molecular basis 455 responsible for these changes can be confined elsewhere. To understand the molecular 456 foundation of LTP maintenance, it is therefore essential to know from which site of the 457 synapse the "molecular memory" is originated. Since the discovery of LTP, most studies  (Fig. 3). BDNFpro-gold particles were seen at pre-and post-synaptic 463 terminals and astrocytic microdomains closely interacting with these synaptic complexes.

464
Gold particles residing in these synaptic sites account for highly intricate trafficking 465 events induced by TBS: BDNFpro could be (i) synthesized at pre-and/or post-synaptic 466 sites; (ii) transported to pre-and/or post-synaptic sites; (iii) internalized in pre-and/or 467 post-synaptic compartments as well as in astrocytic microdomains. Given this 468 complexity, it is of note that BDNFpro localization at astrocytic microdomains is the only 469 relevant localization required for LTP maintenance. Although EM analysis showed a 470 variety of synapses in terms of shapes and sizes, the one thing that was constant was the 471 enrichment of BDNFpro-gold particle at peri-synaptic astrocytes (Fig. 3). This suggests 472 the general occurrence of proteolytic conversion in this specialized area. Moreover,

473
BDNFpro colocalized with the SNARE protein Vamp2 (Fig. 4 a, b) and intoxication with 474 TeTN -a protease known to cleave Vamp2 -inhibited BDNFpro secretion (Fig. 5b). This  synapses, but it is only used at synapses that have been tagged by activity. In accord to 489 this model, TrkB, the correlated molecular tag for mBDNF (60), could capture mBDNF 490 at selected synapses. Our finding that astrocytic BDNFpro increases TrkB/SorCS2 491 aggregation at the spine surface ( Fig. 6b and Fig. 7a), supports the idea that proBDNF 492 processing in microdomains may cooperate to selectively tag activated synapses. This 493 specific process will then provide the capture of mBDNF for enhancing TrkB 494 phosphorylation (Fig. 7b). Direct demonstration of the astrocytic origin of TrkB tagging 495 would need to perform a two-pathway experiment (60) in perirhinal cortex of p75-flox 496 mice; a brain area not permissive for this type of recording. However, this experiment  BDNFpro in perirhinal cortex of p75-flox mice resulted in mice that spent more time 509 exploring a novel object than a familiar one during the 24 h-test phase (Fig. 8b).

510
Astrocytic release of this "byproduct" is then correlated to long-lasting memory of the 511 task. This finding supports the idea that astrocytic microdomains participate to changes in         The authors declare that all data supporting the findings of this study are available within 769 the paper and its supplementary information files.