1 of 80

Multiscale and multimodal reconstruction of cortical structure and function

Nick Turner & “Tony” Runzhe Yang

11-5-2020

1

2 of 80

Zebrafish hindbrain

Vishwanathan et al., bioRxiv, 2020

Mouse S1

Motta et al., 2019

Full adult drosophila brain

Dorkenwald et al., bioRxiv, 2020

Zebrafinch area X

Kornfeld et al., bioRxiv, 2020

Drosophila hemibrain

Scheffer et al., 2020

2

3 of 80

Zebrafish hindbrain

Vishwanathan et al., bioRxiv, 2020

Mouse V1

Lee et al., 2016

Mouse retina

Bae et al., 2018

Zebrafish olfactory bulb

Wanner et al., 2020

3

4 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

4

5 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

5

6 of 80

250 x 140 x 90 µm³

TEM Volume

Mouse V1 Layer 2/3

6

Dorkenwald et al., bioRxiv, 2019

7 of 80

363 (/416)

Pyramidal cells

34 Inhibitory cells

169 non-neuronal cells

↑ have proofread morphology

7

8 of 80

> 3 million synapse

predictions

~85% prec & rec overall

Synapses between 363 PyCs were proofread

Accurate graph of

1981 synapses in

1752 connections

8

Dorkenwald et al., bioRxiv, 2019

9 of 80

9

Over 96,000 orphans possess at least 10 predicted synapses,

where each orphan’s synapses are uniformly presynaptic or postsynaptic

7% merge error rate

(out of 200)

“Orphans”

Cells w/ soma

10 of 80

10

Automated mitochondria segmentation

93.6% precision

92.7% recall

Qualitatively, ~10%

dendritic mitos split,

some “weak” output in somas

11 of 80

11

Zhou et al., bioRxiv, 2020

2-Photon calcium imaging for

112 Pyramidal cells

12 of 80

12

16 directions

(pseudorandomized)

30 “trials”

~ 30 minutes

recording

https://neuromancer-seung-import.appspot.com/#!%7B%22layers%22:%5B%7B%22source%22:%22precomputed://gs://neuroglancer/alex/calcium/pinky/stimulus/scan2_stimulus%22%2C%22type%22:%22image%22%2C%22blend%22:%22default%22%2C%22shaderControls%22:%7B%7D%2C%22name%22:%22scan2_stimulus%22%7D%2C%7B%22source%22:%22precomputed://gs://neuroglancer/alex/calcium/pinky/stimulus/scan3_stimulus%22%2C%22type%22:%22image%22%2C%22blend%22:%22default%22%2C%22shaderControls%22:%7B%7D%2C%22name%22:%22scan3_stimulus%22%2C%22visible%22:false%7D%5D%2C%22navigation%22:%7B%22pose%22:%7B%22position%22:%7B%22voxelSize%22:%5B4%2C4%2C40%5D%2C%22voxelCoordinates%22:%5B85.02117919921875%2C49.476383209228516%2C177%5D%7D%7D%2C%22zoomFactor%22:1.0632118363557066%7D%2C%22showAxisLines%22:false%2C%22perspectiveOrientation%22:%5B-0.6823164820671082%2C0.6458501815795898%2C0.2724331319332123%2C0.2076101005077362%5D%2C%22perspectiveZoom%22:69.4078518387552%2C%22showSlices%22:false%2C%22layout%22:%22xy%22%7D

13 of 80

13

Automated segmentation (w/ some proofreading)

Automated synapses (w/ some proofreading)

Automated mitochondria

Functional data

https://neuromancer-seung-import.appspot.com/#!%7B%22layers%22:%5B%7B%22source%22:%22precomputed://gs://neuroglancer/alex/calcium/pinky/stimulus/scan2_stimulus%22%2C%22type%22:%22image%22%2C%22blend%22:%22default%22%2C%22shaderControls%22:%7B%7D%2C%22name%22:%22scan2_stimulus%22%7D%2C%7B%22source%22:%22precomputed://gs://neuroglancer/alex/calcium/pinky/stimulus/scan3_stimulus%22%2C%22type%22:%22image%22%2C%22blend%22:%22default%22%2C%22shaderControls%22:%7B%7D%2C%22name%22:%22scan3_stimulus%22%2C%22visible%22:false%7D%5D%2C%22navigation%22:%7B%22pose%22:%7B%22position%22:%7B%22voxelSize%22:%5B4%2C4%2C40%5D%2C%22voxelCoordinates%22:%5B85.02117919921875%2C49.476383209228516%2C177%5D%7D%7D%2C%22zoomFactor%22:1.0632118363557066%7D%2C%22showAxisLines%22:false%2C%22perspectiveOrientation%22:%5B-0.6823164820671082%2C0.6458501815795898%2C0.2724331319332123%2C0.2076101005077362%5D%2C%22perspectiveZoom%22:69.4078518387552%2C%22showSlices%22:false%2C%22layout%22:%22xy%22%7D

14 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

14

15 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

15

16 of 80

Pyramidal Cell Subgraph

16

113 neurons w/ >100 μm axons.

“Less truncated neurons” are mainly concentrated at small y (shallow in depth) area.

17 of 80

Pyramidal Cell Subgraph

17

762 synapses among this subset of PyCs.

A directed simple graph consisting of 113 nodes and 666 edges.

18 of 80

Null Model: Erdős-Rényi Graphs

18

Directed edges are drawn independently with the same probability.

The expected number of edges matches the observation.

19 of 80

Degree Sequences

19

In-degree = # presynaptic pyramidal cells in the graph.

The observed degree distribution is “wider” (has larger variance) than the expected degree distribution of the ER model.

20 of 80

Degree Sequences

20

Out-degree = # postsynaptic pyramidal cells in the graph.

The observed degree distribution is “wider” (has larger variance) than the expected degree distribution of the ER model.

21 of 80

Two-Cell Motifs

21

5,691

608

29

22 of 80

Null Model: Configuration Model

22

The “simplest” random model which preserves degree sequences.

A switch-and-hold algorithm samples graphs with same degree sequences uniformly (unbiased motif counts).

23 of 80

Two-Cell Motifs Frequencies

23

The overrepresentation of bi-directional connections is significant with respect to both ER and CFG.

24 of 80

Two-Cell Motifs Frequencies

24

The overrepresentation of bi-directional connections is roughly consistent with Song et. al. 2005

25 of 80

Three-Cell Motifs

25

171,860

51,793

2,531

2,436

1,521

2,891

255

358

12

355

75

16

23

8

2

0

26 of 80

Null Model: Generalized ER Model

26

Edges are drawn independently at random for all pairs of nodes (e.g., A and B):

A→B w/ probability p_uni
A←B w/ probability p_uni
A↔B w/ probability p_bi
A B w/ probability(1-2p_uni-p_bi )

The expected number of both uni-directional and bi-directional edges matches the observation.

27 of 80

Three-Cell Motifs Frequencies

27

The overrepresention of motifs 4, 10, 11, 12 is significant w.r.t. generalized ER.

28 of 80

Three-Cell Motifs Frequencies

28

The overrepresentation of motifs 4, 10, 11, 12 w.r.t. Generalized ER is roughly consistent with Song et al.

The overall level of deviation is smaller than Song et al.

29 of 80

Three-Cell Motifs Frequencies

29

The overrepresentation is gone when comparing with CFG.

30 of 80

Three-Cell Motifs Frequencies

30

CFG predicts frequencies better than the generalized ER for most motifs.

31 of 80

Clustering Coefficient

31

CFG predicts clustering coefficients closer to observation.

32 of 80

Common Neighbor Rule

32

CFG also implies “common neighbor rule” (Perin et al. 2011)

33 of 80

Main Conclusions

33

Consistent with previous studies, we found the frequencies of two-cell and three-cell motifs derivate significantly from the ER model.

However, with a degree-preserving configuration model, our analysis suggests that the “non-randomness” of cortical connections are more subtle than previously supposed.

34 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

34

35 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

35

J. Alexander Bae

36 of 80

36

37 of 80

37

38 of 80

“In-connection density”

2

Length ( )

38

39 of 80

39

Lien & Scanziani, 2013; Also see Li et al., 2013

40 of 80

“In-connection density”

2

Length ( )

40

41 of 80

41

Pyramidal cells receiving more connections from nearby cells

show stronger responses.

r=0.46, p=0.005

42 of 80

42

43 of 80

43

Response

Trial

44 of 80

44

Response

Trial

45 of 80

Intermittency

45

46 of 80

Intermittency

46

47 of 80

Intermittency

47

48 of 80

48

Intermittency is correlated with in-connection density, while it is NOT correlated with total in-synapse density.

Connections within pyramidal subgraph

Synapses within�the entire graph

r=−0.49, p=0.003

r=0.05, p=0.757

49 of 80

49

Pyramidal cells receiving more connections from nearby cells show stronger and more reliable responses.

Connections within pyramidal subgraph

Synapses within�the entire graph

50 of 80

50

Okun et al., 2015

51 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

51

52 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

52

53 of 80

125/364 validated

53

125 cells

54 of 80

54

55 of 80

55

56 of 80

56

See Chang, Honick and Reynolds, 2006; Lewis et al., 2018

57 of 80

57

58 of 80

58

Li et al., 2004

Drp1-K38A

Drp1-wt

Control

GFP MitoDsRed Merge

59 of 80

59

r=0.62, p<3.3 × 10-8

r=0.34, p=0.005

60 of 80

60

Somatic mitochondria are intermediate in size between axonal and dendritic. Mitochondrial density covaries with input synapse density.

61 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

61

62 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

62

Casey Schneider-Mizell

63 of 80

63

64 of 80

64

65 of 80

65

66 of 80

66

67 of 80

67

68 of 80

The resource

Motifs

Mitochondria

Functional Data

Interneurons

68

69 of 80

69

70 of 80

70

Barak Nehoran

Seun Ogedengbe

Sergiy Popovych

Ben Silverman

Will Silversmith

Marissa Sorek

Amy Sterling

Sebastian Ströh

Nick Turner

Ashwin Vishwanathan

Adrian Wanner

Ryan Willie

Kyle Patrick Willie

Alyssa Wilson

Jingpeng Wu

Runzhe (Tony) Yang

Szi-chieh Yu

Zhihao Zheng

Jonathan Zung

Clay Reid

Nuno da Costa

Forrest Collman

Andreas Tolias

Jake Reimer

Sebastian Seung

Alex Bae

Douglas Bland

Austin Burke

Manuel Castro

Celia David

Sven Dorkenwald

Daniela Gamba

Jay Gager

Akhilesh Halageri

Eric Hammerschmith

James Hebditch

May Husseini

Zhen Jia

Devon Jones

Chris Jordan

Nico Kemnitz

Selden Koolman

Kai Kuehner

Kisuk Lee

Ran Lu

Kyle Luther

Thomas Macrina

Claire McKellar

Merlin Moore

Shanka Subhra Mondal

Sarah Morejohn

Shang Mu

71 of 80

Multiscale and multimodal reconstruction of cortical structure and function

Nick Turner & “Tony” Runzhe Yang

11-5-2020

71

72 of 80

Supplementary Slides

72

73 of 80

Truncation Effect

73

74 of 80

Degree Sequences

74

75 of 80

Subgraph of PyCs (> 0μm axon)

75

229 nodes, 1264 edges.

As the threshold increases, the overrepresentation w.r.t. ER decreases, and CFG becomes closer to ER.

76 of 80

Subgraph of PyCs (> 300μm axon)

76

100 nodes, 622 edges.

As the threshold increases, the overrepresentation w.r.t. ER decreases, and CFG becomes closer to ER.

77 of 80

Full Graph (363 neurons)

77

78 of 80

Null Model Comparison

78

79 of 80

Better Fit from CFG is Non-Trivial

79

80 of 80

Clustering Coefficients

80