Joaquín Dopazo

Computational Genomics Department,

Centro de Investigación Príncipe Felipe (CIPF),

Functional Genomics Node, (INB),

Bioinformatics in Rare Diseases (BiER-CIBERER),

Valencia, Spain.

Multigenic (mechanistic) biomarkers

http://bioinfo.cipf.es http://www.babelomics.org @xdopazo

SEQC2, Advancing Precision Medicine and Cancer Genomics,

Bethesda, MD, September 13‐14, 2016

Modular nature of human genetic

diseases (and other relevant traits)

• With the development of systems biology, studies have shown that phenotypically

similar diseases are often caused by functionally related genes, being referred

to as the modular nature of human genetic diseases (Oti and Brunner, 2007; Oti

et al, 2008).

• This modularity suggests that causative genes for the same or phenotypically

similar diseases may generally reside in the same biological module, either a

protein complex (Lage et al, 2007), a sub-network of protein interactions (Lim et

al, 2006) , or a pathway (Wood et al, 2007)

• Perturbed modules will account for disease better than individual perturbed genes

Disease genes are close in the interactome

Goh 2007 PNAS

Same disease

in different

populations is

caused by

different genes

affecting the

same functions

Fernandez, 2013, Orphanet J Rare Dis.

However, personalized treatments are mostly based on single-gene biomarkers

http://www.fda.gov/drugs/scienceresearch/researchareas/pharmacogenetics/ucm083378.htm

With a few exceptions: MammaPrint, an example of successful breast cancer decision support test based on a multigenic biomarker

The strength of this approach is that it is unbiased: there are no assumptions about which genes are likely to be involved in the process of interest. For example, in a data-driven study of the prognosis of patients with breast cancer, little was known about the function of 15 of the 70 genes that were found to constitute a prognostic gene-expression signature4. A drawback of this approach is that the outcome relies solely on the quality of the data (and the samples). By contrast, using the knowledge-driven approach, genes that are thought to be relevant to a particular cancer trait are selected on the basis of the scientific literature.

Finding genes

1 2

Assessing functions

Risk is calculated as a function of the 70 gene expression levels

risk= f(gene1, gene2, … gene70)

By historic reasons genes were first selected and their functionality (mechanism according to the disease) was assessed afterwards.

Change in the paradigm MammaPrint, OncoType and other multigenic

biomarkers: bottom up, from genes to

functions

Models of cell functionality: top-down

mechanism-based biomarkers, from functions

(functional modules) to genes

Bottom-up models rely on the observation of perturbed gene

activity present in the training set.

Top-down models are based on the observation of perturbed

functional activity.

An unobserved gene activity leading to the same perturbation in

functional activity would be missed in bottom-up approach,

causing lack of reproducibility (initial problem of MammaPrint)

How realistic are models of

functional modules?

Beyond static biomarkers—The

dynamic response potential of signalling

networks as an alternate biomarker? Fey et al., Sci. Signal. 8, ra130 (2015).

ODE used to solve the dynamics of a model

from the expression values of their

components

Problem:

ODE can

efficiently solve

only small systems

Two problems: defining

functional modules and

modeling their behavior Gene ontology:

descriptive;

unstructured

functional labels

Interactome:

relationships among

components but

unknown function

Pathways:

relationships among

components and

their functional roles

Models

Enrichment methods. GO, etc. (simple statistical tests)

Connectivity models. Protein-protein, protein-DNA and protein-small molecule interactions (tests on network properties)

Low resolution models. Models of signalling pathways, metabolic pathways, regulatory pathways, etc. (executable models)

Detailed models. Kinetic models including stoichiometry, balancing reactions, etc. (mathematical models)

How a functional module is defined?

What “pathway activity” detected by

enrichment methods means?

Does it make sense?

Different and often

opposite gene actions

are triggered by the

same pathway. E.g.:

death and survival

The same gene can trigger different

(and often opposite) responses,

depending on the stimulus

Survival

Death

How the behavior of a functional

module is defined? Transforming gene expression levels into a different metric that accounts for a function. Easiest example of modeling function: signaling pathways. Function: transmission of a signal from a receptor to an effector

Receptors Effectors

Important assumption:

collective changes in gene

expression within the

context of a signaling

circuit are proxies of

changes in protein

activation

Important fact: when the

signal reaches the end of a

circuit triggers a function

A B D

E

F

H Signal

1 0.7

0.1 0.9

0.5

1 0.6

1

A Signal

1 0.7

0.07

0.5

0.6

0.7

0.7

0.7

0.063

0.483

0.516 0.36

0.65

0.825

0.65

0.325

0.65

0.175

0.50 0.50

C

C F

G

H Function

Function

E

D B

G

0.50

Modeling the behavior of signaling

pathways (signal propagation)

Proxies for protein activity

Signal transmission

Transforming gene expression data into

circuit activation levels

Cases / controls

Norm

aliz

ed g

enes

Ps1 344 344 4556 667 88

Ps2 543 67 88 90 12 36

Ps3 36 833 78 38 99 00

Ps4 59 73 336 677 00 31

Ps1 344 344 4556 667 88

Ps2 543 67 88 90 12 36

Ps3 36 833 78 38 99 00

Ps4 59 73 336 677 00 31

…….

…….

…….

Cases / controls

Circuits

Ps1 344 344 4556 667 88

Ps2 543 67 88 90 12 36

Ps3 36 833 78 38 99 00

Ps4 59 73 336 677 00 31

Ps1 344 344 4556 667 88

Ps2 543 67 88 90 12 36

…….

…….

…….

Circuits within

pathways Cases / controls

Raw

data

Ps1 344 344 4556 667 88

Ps2 543 67 88 90 12 36

Ps3 36 833 78 38 99 00

Ps4 59 73 336 677 00 31

Ps1 344 344 4556 667 88

Ps2 543 67 88 90 12 36

Ps3 36 833 78 38 99 00

Ps4 59 73 336 677 00 31

…….

…….

…….

Downstream analysis (differential expression, clustering to find subgroups, survival,

predictors, etc.)

What would you

predict about the

consequences of

gene activity changes

in the apoptosis

pathway in a case

control experiment of

colorectal cancer?

The figure shows the

gene up-regulations

(red) and down-

regulations (blue)

The effects of changes in gene

activity are not obvious

Apoptosis

inhibition is

not obvious

from gene

expression

Two of the three possible sub-

pathways leading to apoptosis

are inhibited in colorectal

cancer. Upper panel shows the

inhibited sub-pathways in blue.

Lower panel shows the actual

gene up-regulations (red) and

down-regulations (blue) that

justify this change in the activity

of the sub-pathways

Signaling activity trigger cell functions

directly related to cancer progression

Estimations of signal intensity received by the effectors

that trigger a cancer-related function can be related to

clinical parameters, such as survival

Actually, signal activity triggers

all the cancer hallmarks

Hanahan, Weinberg, 2011

Hallmarks of cancer: the next

generation. Cell 144, 646

Negative regulation of release of cytochrome c

from mitochondria (inhibition of apoptosis)

Different cancer use different

gene expression programs to

activate the same functions

Important: similar observation was made

across patients of the same cancer

Mechanistic biomarkers

show high specificity and

sensitivity

Models used for obtaining

mechanistic biomarkers

can integrate different

omics data (e.g. mutations)

Mechanistic biomarkers

can be used in the context

of prediction

Specificity Sensitivity

Some interesting features of mechanistic

biomarkers derived from models of pathway

activity

Bio

logic

al variabili

ty

With

in t

echnolo

gie

s

Even more interesting: multigenic

biomarkers increase reproducibility Human neuroblastoma cell lines.

RNA-seq vs microarrays

Errors observed in individual

genes are cancelled across

signaling circuits

Technical variability

Across technologies

Future prospects: Actionable models

The real advantage of models is that, the same way they can be used

to convert omics data into measurements of cell functionality that

provide information on disease mechanisms and drug MoA, they can

be used to test hypothesis such as “what if I suppress (or over-

express) this gen?” This lead to the concept of actionable models.

By simulating changes of gene expression/activity it is easy to:

• Direct study of the consequences of induced gene over-expressions

or KOs

• Reverse study of genes that need to be perturbed to change cell

functionalities, such as:

• Reverting the “normal” functional status of a cell

• Selectively kill diseased cells without affecting normal cells

• Enhancing or reducing cell functionalities (e.g., apoptosis or

proliferation, respectively, to fight cancer)

• Etc.

Actionable pathway models

Transcription

We can inhibit EGFR (target of Afanatib) by

reducing its activity value (0.56 in cancer).

Absolute KO value = 0

Estrogen signaling pathway http://pathact.babelomics.org/

Actionable pathway models

http://pathact.babelomics.org/

The inhibition of the transcription

sought has been attained, but

six more pathways have been

affected in different ways

Six

pathways

have

affected

circuits

Actionable pathway models

Cell cycle inhibited in

Proteoglycans in cancer pathway

Transcription, angiogenesis and other

are inhibited in ErbB signaling pathway

Cell cycle is inhibited in Oxytocin

signaling pathway

http://pathact.babelomics.org/

Simulating drug inhibition

“Ideal” KO of

EGFR affects 7

pathways

Real inhibition with

Afanatib affects 11

pathways

Inhibition with

broader spectrum

Trastuzumab

affects 13 pathways

The use of new algorithms that enable the transformation of genomic

measurements into cell functionality measurements that account for

disease mechanisms and for drug mechanisms of action will ultimately

allow the real transition from today’s empirical medicine to precision

medicine and provide increasingly personalized medicine

The real transition to precision medicine

Intuitive Based on trial

and error

Identification of probabilistic

patterns

Decisions and actions based on knowledge

Intuitive Medicine Empirical Medicine Precision Medicine

Today Tomorrow

Degree of personalization

The Computational Genomics Department at the Centro de

Investigación Príncipe Felipe (CIPF), Valencia, Spain, and…

...the INB, National Institute of Bioinformatics (Functional Genomics Node) and the BiER (CIBERER Network of Centers for Research in Rare Diseases)

@xdopazo @bioinfocipf Follow us on twitter