1. Lecture 14: The Biology of Learning References: H Shouval, M F Bear, L N Cooper, PNAS 99, 10831-10836 (2002) H Shouval, G Castellani, B Blais, L C Yeung, L N Cooper, Biol Cybernetics 87, 383-391 (2002) W Senn, H Markram, M Tsodyks, Neural Computation 13, 35- 67 (2001) Dayan and Abbott, Sects 8.1, 8.2

2. Learning = long-term synaptic changes Long-term potentiation (LTP) and long-term depression (LTD)

3. Learning = long-term synaptic changes Long-term potentiation (LTP) and long-term depression (LTD) CA1 region of rat hippocampus

4. Learning = long-term synaptic changes Long-term potentiation (LTP) and long-term depression (LTD) CA1 region of rat hippocampus Requires NMDA receptors, postsynaptic depolarization (not necessarily postsynaptic firing)

5. Timing dependence Spike-timing dependent plasticity (STDP)

6. Timing dependence Spike-timing dependent plasticity (STDP) (Markram et al, 1997)

7. Timing dependence Spike-timing dependent plasticity (STDP) (Markram et al, 1997)(Zhang et al, 1998)

8. Model I: Ca control model Shouval et al:

9. Model I: Ca control model Shouval et al: Everything depends on Ca concentration

10. Model I: Ca control model Shouval et al: Everything depends on Ca concentration Ca flows in through NMDA channels

11. Model I: Ca control model Shouval et al: Everything depends on Ca concentration Ca flows in through NMDA channels “Back-propagating” action potential (BPAP) after postsynaptic spike (with slow tail)

12. Model I: Ca control model Shouval et al: Everything depends on Ca concentration Ca flows in through NMDA channels “Back-propagating” action potential (BPAP) after postsynaptic spike (with slow tail) Ca dynamics:

13. Ca control model (2) NMDA channel current (after spike at t = 0 ):

14. Ca control model (2) NMDA channel current (after spike at t = 0 ):

15. Ca control model (2) NMDA channel current (after spike at t = 0 ):

16. Ca control model (2) NMDA channel current (after spike at t = 0 ):

17. Ca control model (3) Synaptic strength (conductance) obeys

18. Ca control model (3) Synaptic strength (conductance) obeys

19. Ca control model (3) Synaptic strength (conductance) obeys Back-propagating action potential:

20. Possible basis of equation for synaptic changes AMPA receptors – in membrane (active) and in cytoplasm (inactive)

21. Possible basis of equation for synaptic changes AMPA receptors – in membrane (active) and in cytoplasm (inactive) Kinetic equations:

22. Possible basis of equation for synaptic changes AMPA receptors – in membrane (active) and in cytoplasm (inactive) Kinetic equations:

23. Possible basis of equation for synaptic changes AMPA receptors – in membrane (active) and in cytoplasm (inactive) Kinetic equations:

24. Possible basis of equation for synaptic changes AMPA receptors – in membrane (active) and in cytoplasm (inactive) Kinetic equations:

25. Possible basis of equation for synaptic changes AMPA receptors – in membrane (active) and in cytoplasm (inactive) Kinetic equations:

26. Possible basis of equation for synaptic changes AMPA receptors – in membrane (active) and in cytoplasm (inactive) Kinetic equations: where

27. Possible basis of equation for synaptic changes AMPA receptors – in membrane (active) and in cytoplasm (inactive) Kinetic equations: where

28. How it works

29. Need the slow tail of the BPAP

30. LTD if presynaptic spike is too far in advance of postsynaptic one

31. (unavoidable consequence of model assumptions)

32. Model II (2 second messengers) (Senn, Markram, Tsodyks, 2001)

33. Model II (2 second messengers) (Senn, Markram, Tsodyks, 2001) Markram-Tsodyks experiments (rat barrel cortex, exc-exc):

34. Model II (2 second messengers) (Senn, Markram, Tsodyks, 2001) Markram-Tsodyks experiments (rat barrel cortex, exc-exc): What is changed, (at least on the 1-hour timescale) is the probability of transmitter release

35. Model II (2 second messengers) (Senn, Markram, Tsodyks, 2001) Markram-Tsodyks experiments (rat barrel cortex, exc-exc): What is changed, (at least on the 1-hour timescale) is the probability of transmitter release (recall (Lect 6) treatment of synaptic facilitation: y = P(release|vesicle)

36. Model II (2 second messengers) (Senn, Markram, Tsodyks, 2001) Markram-Tsodyks experiments (rat barrel cortex, exc-exc): What is changed, (at least on the 1-hour timescale) is the probability of transmitter release (recall (Lect 6) treatment of synaptic facilitation: y = P(release|vesicle) Here (SMT notation): call it

37. Model II (2 second messengers) (Senn, Markram, Tsodyks, 2001) Markram-Tsodyks experiments (rat barrel cortex, exc-exc): What is changed, (at least on the 1-hour timescale) is the probability of transmitter release (recall (Lect 6) treatment of synaptic facilitation: y = P(release|vesicle) Here (SMT notation): call it Actual changes in build up slowly over ca 20 min,

38. Model II (2 second messengers) (Senn, Markram, Tsodyks, 2001) Markram-Tsodyks experiments (rat barrel cortex, exc-exc): What is changed, (at least on the 1-hour timescale) is the probability of transmitter release (recall (Lect 6) treatment of synaptic facilitation: y = P(release|vesicle) Here (SMT notation): call it Actual changes in build up slowly over ca 20 min,

39. Model II (2 second messengers) (Senn, Markram, Tsodyks, 2001) Markram-Tsodyks experiments (rat barrel cortex, exc-exc): What is changed, (at least on the 1-hour timescale) is the probability of transmitter release (recall (Lect 6) treatment of synaptic facilitation: y = P(release|vesicle) Here (SMT notation): call it Actual changes in build up slowly over ca 20 min, But  changes faster, on the scale of ~1 s or less

40. Model II (2 second messengers) (Senn, Markram, Tsodyks, 2001) Markram-Tsodyks experiments (rat barrel cortex, exc-exc): What is changed, (at least on the 1-hour timescale) is the probability of transmitter release (recall (Lect 6) treatment of synaptic facilitation: y = P(release|vesicle) Here (SMT notation): call it Actual changes in build up slowly over ca 20 min, But  changes faster, on the scale of ~1 s or less Here we try to describe the dynamics of

41. 2-messenger model (2)

42. NMDA receptors Have 3 states

43. 2-messenger model (2) NMDA receptors Have 3 states 2 nd messenger #1

44. 2-messenger model (2) NMDA receptors Have 3 states 2 nd messenger #2 2 nd messenger #1

45. NMDA receptors Kinetic equations:

46. NMDA receptors Kinetic equations:

47. NMDA receptors Kinetic equations:

48. NMDA receptors Kinetic equations:

49. NMDA receptors Kinetic equations:

50. 2 nd messengers Activation driven by N u,d

51. 2 nd messengers Activation driven by N u,d

52. 2 nd messengers Activation driven by N u,d

53. 2 nd messengers Activation driven by N u,d

54. Effect on release probability

57. where are active concentrations of 2 nd messengers right after post/pre spikes

58. Effect on release probability where are active concentrations of 2 nd messengers right after post/pre spikes Finally,

59. State diagram:

60. Qualitative summary

61. Pre followed by post:

62. Qualitative summary Pre followed by post: move N to up state (pre)

63. Qualitative summary Pre followed by post: move N to up state (pre) activate S u (post)

64. Qualitative summary Pre followed by post: move N to up state (pre) activate S u (post) upregulate P dis (post)

65. Qualitative summary Pre followed by post: move N to up state (pre) activate S u (post) upregulate P dis (post) Post followed by pre:

66. Qualitative summary Pre followed by post: move N to up state (pre) activate S u (post) upregulate P dis (post) Post followed by pre: move N to down state (post)

67. Qualitative summary Pre followed by post: move N to up state (pre) activate S u (post) upregulate P dis (post) Post followed by pre: move N to down state (post) activate S d (pre)

68. Qualitative summary Pre followed by post: move N to up state (pre) activate S u (post) upregulate P dis (post) Post followed by pre: move N to down state (post) activate S d (pre) downregulate P dis (pre)

69. Simulation vs expt Pre/post vs post/pre: modelexpt

70. Simulation vs expt (2) modelexpt