Analysis of Heterogeneous, Single-cell Kinase Signaling Dynamics

Patrick Kinnunen

Jennifer Linderman lab + Gary and Kathy Luker lab

​

IMAG/MSM WG

8/4/2022

​

​

Link to talk bibliography

Intro to Cell Signaling

• Chemical messengers: Kinase proteins
• ERK + Akt – control proliferation, survival, migration -> common cancer targets

​

2

Kahn Academy, Garay et. al. 2014 MBoC, AbCam

Bulk, endpoint measurement: Western blot

Dynamics direct cell behavior

3

Adapted from Traverse et. al. 1994

Differentiation into neurons

Proliferation

Dynamic data reveals correlation between behaviors

4

Stimulus A

Stimulus B

Protein expression

Endpoint data

Protein expression

A -| B

Acquiring Dynamic, Single-cell data

• Express fluorescent reporter

​

​

• Fluorescence microscopy

​

​

• Automated image processing

5

ERK-KTR: Low ERK

ERK-KTR: High ERK

ERK activation

Initial signal

5 minutes after EGF

2 hours after EGF

Regot et. al Cell 2014

Final Data Output

6

Cell 1

Cell 3

Cell 2

Time

Kinase Activity

Akt

ERK

Cell 1

Cell 2

Cell 3

What kind of data do we get?

• Signaling response (rise and/or fall)

​

​

• Multiple signaling responses

​

​

• Continuous signaling (oscillations and pulses)

​

​

• Discrete event (cell birth/death, migration, differentiation)

​

​

7

Time

Kinase activity

Paradigms of Cellular Heterogeneity

Heterogeneity due to noise

​

Heterogeneity due to pre-existing state

8

Kinase activity

Time

Free Energy

Creative analysis of single-cell signaling data

9

Mechanistic modeling – Where does heterogeneity come from?

Dimensionality reduction – How much heterogeneity is there?

Statistical Modeling – how do conditions affect heterogeneity?

Information theory – How much does the heterogeneity affect communication?

Spinosa et. al. Sci. signaling 2019

Verma et. al. PNAS 2021

Sampattavanich Cell Systems et. al. 2018

fPCA

Time

Signal

Source

Destination

Noise

Cheong et. al. Science 2011

Signal

Distribution

Increasing dose

Mechanistic Models of Heterogeneity

• Is heterogeneity in metastasis due to heterogeneity in signaling?
• Where does heterogeneity in signaling come from?

10

At t = 0, CXCL12 added

Modeling heterogeneous cell responses

• Model including receptor ligand dynamics, kinase signaling network, and KTR dynamics

11

Spinosa, PK et. al. Cell Mol. Bioeng 2021

Time

log₂(C/N)

Fitting Single Cells

Hypothesis – some model nodes would be affected by other cellular processes not in the model

- Could account for heterogeneity in response

12

Spinosa, PK et. al. Cell Mol. Bioeng 2021

Other receptors

Cell Cycle

Metabolic state

Model Maps Pre-existing Cell State

13

Single-cells

MDA-MB-231

Time

Matching experimental cells to model cells gives each experimental cell a position in 3D space (Ras, PI3K, mTORC1)

k_PI3K (1/min)

k_mTORC1 (1/min)

control conditioning

low

high

Occupancy

Akt activation

Cell state inference was repeatable, controllable, and predicted ERK and Akt drug response

Spinosa et. al. Sci. Signaling 2019/Spinosa, PK et. al. Cell Mol. Bioeng 2021

ERK activation

Cell state predicts multiple ligands

• State variables are internal to cell – not ligand/receptor specific
• Hypothesis – any receptor signaling to ERK and Akt should be affected.

14

Stimulate with CXCL12

Infer state variables (Ras, PI3K, mTORC1)

Predict EGF responses (different LR kinetics)

Test EGF responses

Statistical Modeling of Signaling

Observation:

• Pulsatile signaling dynamics can affect cell migration and proliferation

15

ERK activity

Approach:

• Model pulse probability as a self-exciting, coupled, point process.

Data: Albeck et. al. Mol. Cell 2013, Analysis: Verma et. al PNAS 2021

Modeling Pulsatile ERK Activity

16

Observed dynamics

time

Simplest model – constant intensity

time

Self-exciting model – pulse increases likelihood by a, for time b

Verma et. al PNAS 2021

Multicellular correlations between

Full model: Given signaling pulse data for multiple cells and locations, can infer:

μ: Baseline rate

a_self, b_self: Self excitation and duration

a₁₂, b₁₂: Neighbor excitation and duration

​

17

Verma et. al PNAS 2021

Model identifies perturbations

• TAPI-1: interferes with cell-cell communication

18

Verma et. al PNAS 2021

• Lower baseline and neighbor coupling under TAPI-1

μ (Baseline pulse rate)

a₁₂ (Neighbor coupling)

Cell Signaling as a Communication Problem

Information theory quantifies how many different inputs can be distinguished by a population of cells

19

Information source: Concentration of ligand in environment

Communication Channel: Signaling pathway

Destination: Kinase activation

Quantifying cellular information transfer

• Units: Bits - log₂(distinct outputs)
• Ex: 2 bits of information = 4 different outputs

20

1 ng/ml

10 ng/ml

100 ng/ml

1000 ng/ml

Population distribution

ERK activity

[EGF]

Perfect communication, I= 2 bits

Population distribution

ERK activity

Larger variance/lower dynamic range, I ~= 1 bit

Quantifying cellular information transfer

21

Selimkhanov et. al. Science 2014

Dynamic Measurements: EGF activating ERK

Dynamic measures of information transfer = much higher

Any single-time measurement = low information transfer

Considering multiple timepoints = higher information transfer

Novel method suggests high information transfer

22

0.1 uM

1 uM

10 uM

100 uM

[Ach]

[Ach]

time

Different cells have markedly different behaviors

Some cells have very high information capacity

Keshalva et. al. Nat Comm 2018

[Ach] (uM)

Information capacity (bits)

Frequency

Calcium response

Dimensionality reduction and clustering

23

Functional principal component analysis extends PCA to temporal data

CODEX: Neural network based dimensionality reduction, clustering, and motif identification

Left: Sampattavanich et. al. Cell Systems 2018, right: Jacques et. al. Mol Syst. Biol. 2021

Different methods for different questions

24

Free Energy

Explaining heterogeneity with cellular noise

Explaining heterogeneity with pre-existing cell state

Information theory

Statistical Modeling

Mechanistic modeling

The Clinical Utility of Cell State

25

Target subpopulations based on cell state

Manipulate cell state to make cells more treatable

Looking ahead

Understand sources of heterogeneity and leverage manipulations for cancer treatment

26

Cancer cell signaling dynamics and modeling

Migration

Proliferation

Resistance

Cell relationships / genetics

Pathway inhibitors

Tumor microenvironment

Signaling Landscape

Acknowledgements

• Jennifer Linderman lab
• Phil Spinosa
• Ray Asare

​

• Gary and Kathy Luker lab
• Kenneth Ho

​

​

27

• Funding
• NIH
• MIDAS and MICDE at the University of Michigan
• W.M. Keck Foundation

​

​

Questions?

28

Mechanistic modeling – Where does heterogeneity come from?

Dimensionality reduction – How much heterogeneity is there?

Statistical Modeling – how do conditions affect heterogeneity?

Information theory – How much does the heterogeneity affect communication?

Spinosa et. al. sci. signaling 2019

Verma et. al. PNAS 2021

Sampattavanich et. al. 2018

fPCA

Time

Signal

Source

Destination

Noise

Cheong et. al. 2011

Signal

Distribution

Increasing dose