1. Correlations (part I)

2. 1.Mechanistic/biophysical plane: - What is the impact of correlations on the output rate, CV,... Bernarder et al ‘94, Murphy & Fetz ‘94, Shadlen & Newsome ’98, Stevens & Zador ’98, Burkit & Clark ’99, Feng & Brown ‘00, Salinas & Sejnowski ’00, Rudolf & Destexhe ’01, Moreno et al ’02, Fellows et al ‘02, Kuhn et al ’03, de la Rocha ’05 - How are correlations generated? Timescale? Shadlen & Newsome ’98, Brody ’99, Svirkis & Rinzel ’00, Tiensinga et al ’04, Moreno & Parga, 2. Systems level: - do correlations participate in the encoding of information? Shadlen & Newsome ’98, Singer & Gray ’95, Dan et al ’98, Panzeri et al ’00, DeCharms & Merzenich ’95, Meister et al, Neimberg & Latham ‘00 - linked to behavior? Vaadia et al, ’95, Fries et al ’01, Steinmetz et al ’00, …

3. Type of correlations: 1.- Noise correlations: stimulus sresponse: r 1, r 2,..., r n 2.- Signal correlations:

4. Quantifying spike correlations 1. Stationary case: t 1 i,r t 2 i,r t 3 i,r... trial index cell index Corrected cross-correlogram:

5. Quantifying spike correlations 2. Non-stationary case: Joint Peristimulus time-histogram: Time averaged cross-correlogram:

6. Quantifying spike correlations Correlation coefficient: Spike count: if T >>

7. How to generate artificial correlations? 1. Thinning a “mother” train. “mother” train (rate R) deletion rate = p * R

8. How to generate artificial input correlations? 2. Thinning a “mother” train (with different probs.) “mother” train (rate R) deletion rate = p * R rate = q * R

9. How to generate artificial input correlations? 3. Thinning & jittering a “mother” train “mother” train (rate R) deletion + jitter random delay: exp(-t/  c ) /  c rate = p * R

10. How to generate artificial input correlations? 4. Adding a “common” train “common” train (rate R) summation rate = r + R

11. How to generate artificial input correlations?: 5. Gaussian input 1 Diffusion approximation

12. White input: 1 2 How to generate artificial input correlations?: 5. Gaussian input

13. Impact of correlations

14. Impact of correlations on the input current of a single cell: 1 2 NENE … 1 2 NINI … Excitation Rate = E Poisson inputs; Zero-lag synchrony Inhibition Rate = I

15. Impact of correlations on the output rate: Balanced state Unbalanced state

16. The development of correlations: a minimal model. 1. Morphological common inputs.

17. The development of correlations: a minimal model. 2. Afferent correlations

18. The development of correlations: a minimal model. 3. Connectivity

19. Development of correlations: common inputs Shadlen & Newsome ‘98

20. p p p p R(1-x) Rx Rp(1-x) Rp 2 x+Rpx(1-p) Independent: Rp(1-x)+Rpx(1-p)= Rp(1-xp)= R eff (1-x eff ) Common: Rp 2 x = R eff x eff where we have defined: Effective rate: R eff = R p Effective overlap: x eff = x p Development of correlations: common inputs & synaptic failures

22. Development of correlations: common inputs. Experimental data.

23. Development of correlations: correlated inputs.

29. Biophysical constraints of how fast neurons can synchronize their spiking activity Rubén Moreno Bote Nestor Parga, Jaime de la Rocha and Hide Cateau

30. Outline I. Introduction. II. Rapid responses to changes in the input variance. III. How fast correlations can be transmitted? Biophysical constrains. Goals: 1. To show that simple neuron models predict responses of real neurons. 2. To stress the fact that “qualitative”, non-trivial predictions can be made using mathematical models without solving them.

31. I. Introduction. Temporal changes in correlation Vaadia el al, 1995 deCharms and Merzenich, 1996

32. I. Temporal modulations of the input. mean variance Diffusion approximation: white noise process with mean zero and unit variance

33. I. Temporal modulations of the input. common variance

34. I. Problems Problem 1: How fast a change in  and  can be transmitted? Problem 2: How fast two neurons can synchronize each other? Leaky Integrate-and-Fire (LIF) neuron common source of noise

35. II. Probability density function and the FPE Description of the density P(V,t) with the FPE V  P(threshold)=0 P(V) Firing rate:

36. II. Non-stationary response. Fast responses predicted by the FPE V  P(threshold, t)=0 P(V,t) 1.1. Firing rate response time Mean input current 2.2. Firing rate response time Variance of the current Described by the equation

37. II. Rapid response to instantaneous changes of  (validity for more general inputs) Silberberg et al, 2004 Moreno et al, PRL, 2002. Change in the mean Change in the input correlations for dif. correlation statistics Change in input variance

38. II. Stationary rate as a function of  c  >0  <0  =0 Moreno et al, PRL, 2002.

39. III. Correlation between a pair of neurons 0 time lag = t1 - t2 Cross-correlation function = P( t1, t2 ) Moreno-Bote and Parga, submitted

40. III. Transient synchronization responses. Exact expression for the cross-corr. The join probability density of having spikes at t1 and t2 for IF neurons 1 and 2 receiving independent and common sources of white noise is: Total input variances at indicated times for neurons 1 and 2. Joint probability density of the potentials of neurons 1 and 2 at indicated times

41. III. Predictions 1.Increasing any input variance produces an “instantaneous” increase of the synchronization in the firing of the two neurons. 2. If the common variance increases and the independent variances decrease in such a way that the total variance remains constant, the neurons slowly synchronize. 3. If the independent variances increase, there is a sudden increase in the synchronization, and then it reduces to a lower level. … biophysical constraints that any neuron should obey!!

42. III. Slow synchronization when the total variances does not change.

43. III. Fast synchronization to an increase of common variance.

44. III. Fast synchronization to an increase of independent variance, and its slow reduction.