Reading the mind of a worm 0.1 Global dynamics embed the motor - - PowerPoint PPT Presentation

SLIDE 1

Reading the mind of a worm

Saul Kato (IMP Vienna) Stanford, 2016-04-04

−0.2 0.2 −0.1 0.1 −0.05 0.05 0.1 PC1 PC2 PC3

Global dynamics embed the motor command sequence of C. elegans

SLIDE 2

Harris Kaplan, IMP Tina Schrödel, IMP Manuel Zimmer, IMP

funding EMBO Simons Collaboration on the Global Brain European Research Council NIH collaborators Yifan Xu, Rockefeller Christine Cho, Rockefeller Cori Bargmann, Rockefeller Larry Abbott, Columbia

SLIDE 3

Today

• 1. Introduction
• 2. Single cell dynamics
• 3. Network dynamics
• 4. What’s next

SLIDE 4

my motivating hypothesis

Some aspects of higher cognition, such as flexible reasoning, may have originated in the production of variable but controlled behavioral sequences in simpler animals.

SLIDE 5

what would a “complete” story of how a nervous system generates behavior look like?

SLIDE 6

a complete story of neurons-to-behavior should, at the minimum:

cannot rely on a homunculus or components

• utside of the model

explain single trials be self-contained explain behavior as as a time series

• rganism behavior must be

competent on a single trial basis behavior is an continuous,

• nline output of a nervous

system

SLIDE 7

a one neuron animal

Chemotaxis in E. coli Analysed by 3D Tracking Berg & Brown, Nature 1972

A0 = −kmA(1 − I) − kbA A + Kmm + kpGI

G0 = km(1 − I)A − kpIG + kr

SLIDE 8

a one neuron animal

Robustness and adaptation in simple biochemical networks Barkai & Leibler, Nature 1997

A0 = −kmA(1 − I) − kbA A + Kmm + kpGI

G0 = km(1 − I)A − kpIG + kr

p(CW → CCW) = f1(G) p(CCW → CW) = f2(G)

Larry Abbott

SLIDE 9

• C. elegans

302 neurons ~6000 synapses ~900 gap junctions

no classical action potentials. Na+ channels lost during nematode evolution

Virtual Worm Project

• cell lineage fully mapped (Sulston et al., 1983)
• connectome fully mapped (White et al., 1986)
• genome sequenced (C. elegans Consortium, 1998)

SLIDE 10

worm ethology

SLIDE 11

describing high-level behavior

state transition diagram

SLOWING DORSAL TURN FORWARD RUN VENTRAL TURN REVERSE RUN

SLIDE 12

the worm connectome

interneurons

0.025 0.02 0.015 0.01 0.005 0.005 0.01 0.015 0.02 2 1.5 1 0.5 0.5 1 1.5 2 VB05 VB03 DD03 DD02 VB04 DD04 VD07 VB06 VD05 VD06 VD04 DB03 VC01 VC02 VA07 VD08 DB02 AS04 VA04 VD03 VA06 VC03 DB04 DD05 VA05 DA04 DA03 VD09 AS03 AS05 AS06 VB08 DD01 VD02 DA05 VA09 VA08 AS02 VA03 VB02 DB01 PDB VA02 VB09 VD10 PDA RID DA02 VB07 DA06 AS11 VD13 VD01 VA12 DD06 VC04 DA09 VD12 DVB PVDL AS09 VA11 DA08 DA07 VB11 AS08 PVDR VA10 PHCL PHBL VA01 AS01 VB10 AS07 DB07 PVCR PHBR AS10 PVCL AVAL DA01 LUAL DB05 DB06 AVAR PVWR PLML VD11 PHCR SABD LUAR PQR PVWL AVDL PVNR AVL AVDR AVBR PHAL AVBL PLMR PHAR FLPR AVG PVNL FLPL SABVR AVM SABVL AVJL PVPR DVC VC05 PVPL VB01 AVJR BDUR HSNR PDER PDEL PVM DVA AVFR RIFL AVFL RIFR AQR AVHR AVHL PVR SDQL ALMR BDUL AVKL ALA PVQR PVT ALML ADAL HSNL SDQR AVEL

normalized Laplacian eigenvector 2

ASJR ADER SAADL RMFR ADAR AVER ADLR SAAVR SAAVL ADLL AIML AVKR ASHR RIMR AIMR RIML ASHL PVQL SIBVL ASJL RMFL ASKR SAADR RIGL AIBR RICL RIS SMBVR AIBL PLNL ADEL ASKL PLNR RIGR SMBDL ALNL AIAR RICR SMBDR AIAL SMBVL RMGR ALNR AWBR RIR BAGR BAGL RMGL ASGL AUAL URXL ASGR AIZR AWBL SMDDR RIBL AIZL ASIR URYVL SIBVR URBL URYVR ADFR RMHL RIBR URYDR ASER AWAR ASIL AINR RIVL OLLL AUAR ADFL AWCR SMDDL RMHR RIVR SIBDL AWCL SIADL ASEL AINL AWAL URYDL CEPDL CEPVL OLLR RMDR URXR SIAVR CEPDR SIADR AIYL AIYR SIAVL SIBDR AFDR AFDL SMDVR SMDVL URBR RMDL CEPVR RIAL RIAR OLQDR OLQVL OLQDL RMDVR IL2L RMDDL RMED RMEV RIH RMDVL RMDDR IL1L IL2R IL1R URADL OLQVR IL1DL IL1VL RIPL IL1DR URADR RMEL IL1VR RIPR IL2DL URAVL IL2VL RMER IL2DR IL2VR URAVR

processing depth

interneurons sensory neurons motor neurons motor neurons sensory neurons Chen 2011 processing depth 2

• 2

SLIDE 13

part of the parts list

CHEMOSENSORY NEURONS MECHANOSENSORY PHARYNX AMPHIDS (LR) INNER LABIA (6) OUTER LABIA (6) DEIRIDS HEAD (4) BODY TAIL PHASMIDS (LR) MOTONEURONS INTERNEURONS RING BODY

L R

ASE

L R

ASK

L R

ASJ

L R

ASG

L R

AWC

L R

AWB

L R

URX

L R L R

URY

L R L R

URA

L R L R

RME

L R L R

SMD

L R L R

SMB

L R L R

SAA

L R L R

SIB

L R L R

SIA

L R

FLP

L R

BAG AQR

L R

PHA

L R

PHB

L R

PHC

L R

M2

L R

M3

L R

I2

L R

I1

L R

IL1

DL DR VL VR

CEP

DL DR VL VR

OLQ

L R

OLL

L R

PDE

L R

ADE

L R

PLM

L R

PVD

L R

ALM

L R

ALN

L R

AVM

L R L R L R

IL2

L R L R L R

AWA

L R

LUA

L R

BDU

L R

PVQ

L R

PVP

L R

PVN

L R

PVW

L R

ADE

L R

ASH

L R

AVA

L R

AVF

L R

AVH

L R

AVJ

L R

AVK

L R

RIA

L R

RIF

L R

RIG

L R

RIC

L R

RIB

L R

URB

L R

SDQ

L R

AVB

L R

PVC

L R

AVD

L R

AVE

L R

ADA

L R

HSN AS 1 2 4 8 10 11 11

L R

AIA

L R

AIB

L R

AIM

L R

AIN

L R

AIY

L R

AUA

L R

AIZ

L R

RIM

L R

RMF

L R

RMH

L R

RMG

L R

RIP

L R

RIV

L R

AFD

L R

ASI

L R

ADF

L R

ADL PQR M1 M4 I3 I4 I5 I6 M5 MI MC NSM 3 5 6 9 7 VD 1 2 4 8 12 10 11 13 3 5 6 9 11 7 VB 1 2 4 8 10 11 11 3 5 6 9 7 VA 1 2 4 8 10 11 3 5 6 9 7 DA 1 2 4 8 9 3 5 6 7 DB 1 2 4 7 3 5 6 DD 1 2 4 3 5 6 VC 1 2 4 3 5 6 12

L R

RMD

L R L R D

SAB

D L D R awar awcr awcl avar aval avfr avfl avhr avhl avjr avjl avkr avkl riar riaL rifr rifl rigr rigl ricr ricL ribr ribL urbr urbl sdqr sdql avbr avbl pvcr pvcl avdr avdl aver avel adar adal hsnr hsnl as11 as01 vd13 vd01 vb11 vb01 va12 va01 da09 da01 db07 db01 dd06 dd01 vc06 vc01 aiar aial aibr aibl aimr aiml aimr aiml aiyr aiml auar aual aizr aizl rimr riml rmfr rmfl rmhr rmhl rmgr rmgl ripr ripl rivr rivl awal luar lual bdur bdul pvql pvqr pvpr pvpl pvnr pvnl pvwr pvwl ader adel

AVG

avg

ALA

ala

RID

rid

RIR

rir

RIS

ris

RIH

rih

DVA

dva

DVC

dvc

DVB

dvb

PVM

pvm

PDB

pdb

PVR

pvr

PVT

pvr

PDA

pda

AVL

avl

AVM

avm

SLIDE 14

Today

• 1. Introduction
• 2. Single cell dynamics
• 3. Network dynamics
• 4. Next steps

SLIDE 15

AWC

SLIDE 16

sensory responses circa 2009

Chalasani, Kato, et al Nat Neuro 2011

120 60 120 60 WT nlp-1 Odor Odor ΔF/F 200% time (s ) time (s)

GCaMP1.0

SLIDE 17

responses to complex input

30 60 90 120 150 180 210 1 2 time (s) F/F

AWC Kato et al, Neuron 2014

GCaMP3.0

SLIDE 18

trials

1 2

20 s

sensory neurons are highly reliable

Kato et al, Neuron 2014

SLIDE 19

AWC

4 8 12 16 20 24

normalized magnitude lag (s)

y ( F/F) x

L N input

• utput

K(t) F(x) x

u(t) x(t) y(t)

a simple model predicts sensory responses with high fidelity

Kato et al, Neuron 2014

SLIDE 20

an ODE model of a neuron

input

• utput

A F S

kaf kas ka ks kf

dA dt = kaA+ input dS dt = ksS - k A =F S +

• utput

dF dt = kfF + kaf

as

A

Kato et al, Neuron 2014

10 20 30 40 Lag (s) 10 20 30 40 Lag (s)

F S

SLIDE 21

summary part 1

• C elegans sensory neurons can be highly reliable

signal transducers.

• the analog GCaMP signal can be used for

quantitative dynamical studies.

SLIDE 22

Today

• 1. Introduction
• 2. Single cell dynamics
• 3. Network dynamics
• 4. Next steps

SLIDE 23

the c elegans brain

SLIDE 24

the c elegans brain

head ganglia

SLIDE 25

• C. elegans has a stereotypic neuroanatomy

adapted from White, 1986

Retrovesicular ganglion Ventral ganglion Ventral ganglion Dorsal ganglion Anterior ganglion

10 µm

SLIDE 26

how to do high quality whole-brain Ca2+ imaging

microfluidics spinning disk confocal 10 Z @ 2.9 volumes/s volumetric microscopy nuclear-localized GCaMP5K

M13$ GFP$ CaM$ NLS$ NLS$ N+$ +C$

SLIDE 27

nucleus-localized GCaMP nuclear localized GCaMP for resolving single cell activity

SLIDE 28

AVAR AVAL RIMR RIML AVER VA01 SABVL

SLIDE 29

(OLQVR/URYVR) (OLQDL/URYDL) AIBL AIBR 42 108 38 (OLQDR/URYDR) 9 20 93 40 ALA 31 47 22 50 AVFL 48 76 18 55 52 77 94 62 90 5 85 49 10 14 1 11 (RIFR/AVG/DD01) 51 88 83 35 105 (SMBDR/SMBVR/SMDDR/RMFR) 64 74 7 25 34 86 3 6

60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 1080

Time (s)

SLIDE 30

neural dynamics exhibit a widely shared, cyclical signal

Neuron Time (s) 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 ΔF/F0 −0.2 0 1 1.5 AVAR AVAL RIMR RIML AVER VA01 SABVL OLQVL DB01 VB01 DB02 RMER RMEL RID AVBR RIBL VB02 RMED RMEV AVBL SMDVL SMDVR RIVL RIVR AIBL OLQVR AIBR OLQDL OLQDR RIFR SMBVR PC1 PC2 PC3 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 3 2 1 Time (s) TPC#

PCA TVD

SLIDE 31

60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 3 2 1 Time (s) TPC#

12345678910 20 40 60 80 PC Variance explained (%)

neural dynamics exhibit a widely shared, cyclical low-dimensional signal

SLIDE 32

60 120 180 240 300 360 420 480 540 600 660 3 2 1 Time (s) TPC#

1 brain cycle

neural dynamics exhibit a widely shared, cyclical low-dimensional signal

SLIDE 33

60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 3 2 1 Time (s) TPC#

neural dynamics live on a low-dimensional manifold

SLIDE 34

individual neurons are active in different regions

• f the neural manifold

−0.2 0.2 −0.1 0.1 −0.05 0.05 0.1 PC1 PC2 PC3

AVAL ↑ SMDV ↑ RMED ↑

0.2 −0.1 0.1 PC1 PC2 −0.2

SLIDE 35

neuron activations tend to have stable phase

RIVL RIVR RMED RMEL RMER RMEV SABVL (SMBDR/SMBVR/ SMDDR/RMFR) SMDVL SMDVR VA01 VB01 VB02 AIBL AIBR ALA AVAL AVAR AVBL AVBR AVER AVFL DB01 DB02 (OLQDL/ URYDL) (OLQDR/ URYDR) (OLQVL/ URYVL) (OLQVR/ URYVR) RIBL RID (RIFR/AVG/ DD01) RIML RIMR

SLIDE 36

30 Hz epifluorescence 10 Hz infrared

experimental techniques

• I. Whole-brain imaging
• II. Free-moving single neuron imaging
• Good signal-to-noise
• High frame rate (30 Hz)
• Simultaneous behavioral recording

Faumont et al, 2011

SLIDE 37

∆ R / R0 1 1.5 2 2.5 1 2 3 4 1 2 3 4 Time from reversal start (s) 1 2 ∆ R / R0 ∆ R / R0 ∆ R / R0

• 2
• 1

1 2

AVA RIM AVE AIB

Time from reversal end (s)

• 2
• 1

1 2 .5 1 −1 1 −1 1 1

REVERSAL

n>100

free-moving imaging of single neurons gives neural correlates of behavioral commands

SLIDE 38

4-state coloring analysis

FALL HIGH RISE LOW

π 3π π 2 2 RISE HIGH FALL LOW

60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 3 2 1 Time (s) TPC#

SLIDE 39

4-state coloring analysis

FALL HIGH RISE LOW

π 3π π 2 2 RISE HIGH FALL LOW

AVAL

60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 0.2 0.4 0.6 0.8 1 1.2 1.4 time (s)

60 s

SLIDE 40

sub-trajectories registered by time warping

SLIDE 41

geometry preserved across multiple trials

n=5

π 3π π 2 2 RISE HIGH FALL LOW

SLIDE 42

state coloring applied to whole dataset

AIBL SMDVL SABVL

SLIDE 43

many neurons show PC1 synchrony with subtle differences

AVAR AVAL RIMR RIML AVER VA01 SABVL AIBL AIBR

60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020

OLQVL

SLIDE 44

clustering sub-trajectories by relative transition timing

high-to-fall transition timing

1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 AIBL AIBR AVAL AVAR AVER OLQDL OLQDR OLQVL OLQVR RIML RIMR SABVL SMDVL VA01 avbl avbr avfl db01 db02 ribl rid rifr rivl rivr rmed rmel rmer rmev smbvr smdvr vb01 vb02

t=0 none

+6

• 6

seconds

SLIDE 45

Trajectory bundles can be clustered by neuronal timing patterns alone

π 3π π 2 2 RISE HIGH FALL LOW

1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 AIBL AIBR AVAL AVAR AVER OLQDL OLQDR OLQVL OLQVR RIML RIMR SABVL SMDVL VA01 avbl avbr avfl db01 db02 ribl rid rifr rivl rivr rmed rmel rmer rmev smbvr smdvr vb01 vb02

FALL2 FALL1

−0.2 −0.15 −0.1 −0.05 0.05 0.1 0.15 0.2 0.25 0.3 −0.15 −0.1 −0.05 0.05 0.1 0.15 −0.05 0.05 0.1 PC1 PC2 PC3

FALL1 FALL2

SLIDE 46

Trajectory bundles can be clustered by neuronal timing patterns alone

1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 AIBL AIBR AVAL AVAR AVER OLQDL OLQDR OLQVL OLQVR RIML RIMR SABVL SMDVL VA01 avbl avbr avfl db01 db02 ribl rid rifr rivl rivr rmed rmel rmer rmev smbvr smdvr vb01 vb02

FALL2 FALL1

SLIDE 47

FALL clusters correspond to dorsal vs ventral turns

1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 AIBL AIBR AVAL AVAR AVER OLQDL OLQDR OLQVL OLQVR RIML RIMR SABVL SMDVL VA01 avbl avbr avfl db01 db02 ribl rid rifr rivl rivr rmed rmel rmer rmev smbvr smdvr vb01 vb02

FALL2 FALL1

GCaMP/mCherry 0.5 ∆ R/R0 head-bend angle 30°

SMDV

V D Time (seconds) 30 60 90

D V REV

Free moving imaging

SLIDE 48

∆ R / R0 1 1.5 2 2.5 1 2 3 4 1 2 3 4 Time from reversal start (s) 1 2 ∆ R / R0 ∆ R / R0 ∆ R / R0

• 2
• 1

1 2

AVA RIM AVE AIB

Time from reversal end (s)

• 2
• 1

1 2 .5 1 −1 1 −1 1 1

REVERSAL

n>100

free-moving imaging of single neurons gives neural correlates of behavioral commands

Time from end (s)

• 2
• 1

1 2 Time from end (s)

• 2
• 1

1 2 Ventral Dorsal Reverse Reverse

SMDV

TURNS

SLIDE 49

clustering of brain cycle transitions allow decoding of behavioral command state

PC1

• 0.2

0.2

PC2

• 0.1

0.1

SUSTAINED REVERSAL REVERSAL II REVERSAL I VENTRAL TURN DORSAL TURN FORWARD SLOWING

SLIDE 50

trajectory clustering allows bundle averaging

PC1

• 0.2

0.2

PC2

• 0.1

0.1

REVS REV1 REV

2

DORSAL TURN VENTRAL TURN SLOWING FWD

BRAIN STATE GRAPH

• 0.2

0.2

• 0.1

0.1

PC1 PC2

SLIDE 51

the behavioral graph is embedded in neural dynamics

BRAIN STATE GRAPH

REVS REV1 REV

2

DORSAL TURN VENTRAL TURN SLOWING FWD

BEHAVIORAL STATE

SLOWING DORSAL TURN FORWARD RUN VENTRAL TURN REVERSE RUN

REVERSAL DORSAL TURN VENTRAL TURN SLOWING FWD

SLIDE 52

the command sequence pattern is preserved despite decoupling from motor output

60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 3 2 1 Time (s) TPC#

−0.2 −0.1 0.1 −0.15 −0.1 −0.05 0.05 0.1 0.15 PC1 PC2

Pokala, PNAS 2014

AVA:: HisCl

SLIDE 53

manifold structure is preserved despite salient stimuli

−0.2 0.2 0.4 −0.2 −0.1 0.1 −0.1 0.1 PC1 PC2 PC3

Fall Rise

4% O 21% O BAG

2 2

Pre

Reversal comand state probability Time (s) O2 4% 4% 4% 4% 4% 4% 21% 60 120 180 240 300 360 420 480 540 600 660 720 0.25 0.5 0.75 1

n=13

SLIDE 54

• an autonomous, distributed

“neural dynamics engine” generates the motor command sequence

summary part 2

another C. elegans first

• continuous, all-time neural

decoding of behavior has been achieved in a model

• rganism
• the neural manifold is a

scaffold for detailed studies

• f behavior generation


Kato et al, Cell 2015

input

• utput

SLIDE 55

Today

• 1. Introduction
• 2. Single cell dynamics
• 3. Network dynamics
• 4. What’s next

SLIDE 56

What’s next

• Decision making
• Origins of lawful dynamics
• Beyond the worm

SLIDE 57

branching dynamics reflect decision making

ventral turn dorsal turn reversal ?

SLIDE 58

−3π −2π −π π 10

−5

10

−4

10

−3

10

−2

10

−1

10 Global phase Fall sub-trajectories

branching dynamics reflect time course of decision execution

Clustering p-value

π 3π π 2 2 RISE HIGH FALL LOW

Kato et al, Cell 2015

SLIDE 59

branching dynamics reflect time course of decision execution

−3π −2π −π π 10

−5

10

−4

10

−3

10

−2

10

−1

10 Global phase Fall sub-trajectories −2π −π π 2π 10

−5

10

−4

10

−3

10

−2

10

−1

10 p−value Clustering Global phase Rise sub-trajectories

π 3π π 2 2 RISE HIGH FALL LOW

Kato et al, Cell 2015

REVS REV1 REV

2

DORSAL TURN VENTRAL TURN SLOWING FWD

SLIDE 60

ventral turn dorsal turn reversal ?

honing in on decision execution

SLIDE 61

honing in on decision execution

• Characterize brain-state-dependent stimulus

responses

• Determine the locus of downstream execution

control

Closed-loop experiments

SLIDE 62

Sensory interactions with brain state

4% O2 21% O2 30 s

s e n s

• r

y

SLIDE 63

SLIDE 64

backup slides

SLIDE 65

0.2

PC1

• 0.2
• 0.1

0.1 0.1

PC2 PC3 PC1

• 0.2

0.2

PC2

• 0.1

0.1 drive (a.u.) no value

• 1
• 0.5

0.5 1

manifold encoding of analog parameters of behavior

SLIDE 66

2 4 6 8 10 12 14 16 18 20 >20 0.5 LOW phase probability 2 4 6 8 10 12 14 16 18 20 >20 0.2 0.4 RISE phase probability 2 4 6 8 10 12 14 16 18 20 >20 0.05 0.1 HIGH phase probability 2 4 6 8 10 12 14 16 18 20 >20 0.1 0.2 FALL phase probability state duration (s)

state durations in restrained chip

SLIDE 67

voltage -> nuclear calcium

time (s)

10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170

Vm

• 30
• 20
• 10

Mag (a.u.)

• 1
• 0.5

0.5 1

time (s)

30 60 90 120 150 180 210 240 270 300

Vm

• 70
• 60
• 50
• 40
• 30

Mag (a.u.)

• 1
• 0.5

0.5 1

time (s)

30 60 90 120 150 180 210 240 270 300 330 360 390

Vm

• 50
• 40
• 30
• 20
• 10

Mag (a.u.)

• 1
• 0.5

0.5 1

time (s)

30 60 90 120 150 180 210 240 270

Vm

• 60
• 50
• 40
• 30
• 20

Mag (a.u.)

• 1
• 0.5

0.5 1

c

SLIDE 68

SLIDE 69

Mark Alkema