Predicting not to predict too much: How the cellular machinery of the brain

1. Predicting not to predict too much: How the cellular machinery of the brain may anticipate the uncertain future Yadin Dudai Department of Neurobiology The Weizmann Institute of Science

2. The textbook account of the biography of a memory: Items mature from a short-term into a long-term, stable state, via consolidation, which occurs just once per item. Storage LTM Consolidation (The dual-trace hypothesis) (The consolidation hypothesis) STM Time (Combining Ribot 1882, Muller & Pilzecker 1900, Hebb 1949, McGaugh 2000, Dudai 2004)

3. An alternative account: • Active Memory Inactive Memory (After Lewis 1973, Dudai 2004, Nader 2005) (The recurrent phases hypothesis) (The reconsolidation hypothesis) What do the data indicate?

4. 100 LiCl ip Hrs Hrs - Yrs 80 Toxicosis US Aversion CR Taste CS 60 Aversion Index NaCl ip 40 20 0 Central gustatory area, insular cortex An Experimental System: Conditioned Taste Aversion (CTA) (Same magnitude of memory even if first test performed after several months)

5. The algorithm for the formation of long-term memory in cortex and its molecular implementation consolidation Triggering memory encoding consolidation (Berman, Lamprecht, Rosenblum, Naor, Bahar, Kobilo, Shema & Dudai, 1996-2009) Prima facie, this model fits the classic, unidirectional, account of LTM But do the data support such ‘deterministic’ interpretation? I will now demonstrate 4 ways to alter long-term CTA memory in cortex

6. A. Molecular changes in cortex following encoding ERK1/2 PSD95 Elk-1 0.5 LTM STM Memory Consolidation 180 240 (Fits, but does not prove, the dual-trace hypothesis) Magnitude of observed effect B. Blockade of long-term memory in cortex Time 0.5 Scopolamine Anisomycin 96h 30 60 90 120 Time after training (min) Method No. 1: Disruption of consolidation Taste memory consolidates in the insular cortex (Note time windows of consolidation) (A, taste exposure; B, CTA training. Data compiled from Rosenblum et al. 1993, Naor & Dudai 1996, Berman et al. 1998, Kobilo-Moav & Dudai unpublished, Elkobi et al. 2008)

7. Method No. 2: Experimental ‘extinction’ Retrieval Training Non-reinforced retrieval LTM STM 1 4 5 6 Days Days Memory Consolidation Time (untested control doesn’t decline in 4 months) (Agnostic to the dual-trace hypothesis) (Eisenberg et al. 2003, Stehberg & Dudai unpublished) Long-term taste memory extinguishes with use in the insular cortex

8. Retrieval Retrieval LTM Training Anisomycin into cortex STM 1 2 5 6 7 8 Days Memory Consolidation Reconsolidation Vehicle Time (Does not fit the original dual-trace- and consolidation hypotheses) Anisomycin Post-retrieval test days (Eisenberg et al. 2003) Method No. 3: Disruption of ‘Reconsolidation’ Long-term taste memory reconsolidates in the insular cortex Boundary conditions on reconsolidation: Extinction – No, New encoding - Yes (Background: Misanin 1968, Sara 2000, Nader et al. 2000)

9. consolidation Method No. 4: Disruption of persistence It now appears that there might be potential plastic opportunities even in the absence of retrieval (Berman, Lamprecht, Rosenblum, Naor, Bahar, Kobilo, Shema & Dudai, 1996-2009)

10. PKMz can be inhibited by the pseudosubstrate peptide, ZIP Protein kinase M zeta (PKMz) is an autonomous form of the atypical PKC, PKCz, synthesized from PKMz mRNA z (after Hernandez et al. 2003)

11. ZIP ZIP Memory Memory LTM Training Training STM Time Memory Consolidation 100 (Demonstrates that very-LTM requires persistent enzyme activity and is capable of rapid alterations) Time Aversion index 60 20 Ctrl 1 Month Ctrl 3 Months Time from ZIP to test Time from training to ZIP (Data compiled from Shema, Sacktor & Dudai. 2007, Shema, Sacktor, Hazvi & Dudai 2009) Back to Method No. 4: Disruption of persistence Long-term taste memory is quickly erased by an inhibitor of PKMz (Background: Pastalkova et al. 2006)

12. Effect of ZIP on CTA memory in cortex: • Once erased, memory can be reacquired – and re-erased • Not prevented by very intensive training • Not rescued by spontaneous recovery and US-reinstatement • Applies to multiple past taste-associations • Does not apply to taste recognition • Undetected before the taste association is established • Undetected within the first hour after training (i.e., consolidates) • Not replicated by a general serine/threonine kinase inhibtior (H7) • Not replicated by microinfusions into the hippocampus • A manifestation of a physiological regulatory process?

13. The weakest link? A cellular model: PKMz increases insertion of AMPA receptors into the membrane via phosphorylation of scaffold/trafficking proteins (Drawing modified from Hernandez et al. 2003)

14. Take home message • There are multiple opportunities for a • long-term memory trace to change in cortex: • in consolidation, • in extinction (where it is mostly the expression that changes), • in post-retrieval reconsolidation, • and possibly also in the absence of retrieval. • This modifiability of the cellular substrate of memory seems a basic attribute of memory. • It is tempting to propose that it reflects the need • to update and integrate new information into old, i.e., the need of memory systems to prepare for • the unpredictable demands of the future.

15. Memory models: which is closer to reality? The dual-trace model: Consolidation just once per item, deterministic The cyclic model: multiple windows of plasticity, trace modifiable in reactivation (reconsolidation) Item accessibility The extended cyclic model: multiple windows of plasticity, trace modifiable in the presence and in the absence of reactivation

16. WIS Reut Shema Shoshi Hazvi Efrat Furst Orit Furman Kelly Ludmer Daniel Levy Avi Mendelsohn Uri Nili Dana Bezalel Aya Ben-Yakov Micah Edelson Yossi Chalamish Yaara Yeshurun Shiri Ron NYU Joe LeDoux Lila Davachi Nava Rubin SUNY Todd Sacktor