From reference genes to global mean normalization

Jo Vandesompele

professor, Ghent University

co-founder and CEO, Biogazelle

qPCR Symposium USA

November 9, 2009 – Millbrae, CA

outline

what is normalization

gold standard for mRNA normalization

global mean normalization and selection of stable small RNAs for microRNA normalization

introduction to normalization

2 sources of variation in gene expression results

biological variation (true fold changes)

experimentally induced variation (noise and bias)

purpose of normalization is reduction of the experimental variation

input quantity: RNA quantity, cDNA synthesis efficiency, …

input quality: RNA integrity, RNA purity, …

gold standard is the use of multiple stably expressed reference genes

which genes?

how many?

how to do the calculations?

normalization: geNorm solution

framework for qPCR gene expression normalisation using the reference gene concept:

quantified errors related to the use of a single reference gene

(> 3 fold in 25% of the cases; > 6 fold in 10% of the cases)

developed a robust algorithm for assessment of expression stability of candidate reference genes

proposed the geometric mean of at least 3 reference genes for accurate and reliable normalisation

Vandesompele et al., Genome Biology, 2002

geNorm software

automated analysis

ranking of candidate reference genes according to their stability

determination of how many genes are required for reliable normalization

http://medgen.ugent.be/genorm

0.003

0.0060.0210.0230.056

NF4

NF1

cancer patients survival curve

statistically more significant results

geNorm validation (I)

log rank statistics

Hoebeeck et al., Int J Cancer, 2006

mRNA haploinsufficiency measurements

accurate assessment of small expression differences

geNorm validation (II)

Hellemans et al., Nature Genetics, 2004

patient / control

3 independent experiments

95% confidence intervals

geNorm is the de facto standard for reference gene validation and normalization

> 2,000 citations of our geNorm technology

> 10,000 geNorm software downloads in 100 countries

normalization using multiple stable reference genes

global mean normalization

when a large set of genes are measured, the average expression level reflects the input amount and could be used for normalization

e.g. microarray based normalization

o lowess, mean ratio, …

SAGE / NGS sequencing counts

the set of genes must be unbiased and sufficiently large

we make use of this principle to normalize microRNA data from experiments in which we quantify a substantial number of miRNAs (450 or 650) in a given sample

global mean normalization

small-RNA controls

classic normalization strategy

small nuclear RNAs, small nucleolar RNAs

18 available from Applied Biosystems

global mean normalization

method applied for microarray data

universal: applicable for every miRNA dataset

many datapoints needed (megaplex vs. multiplex)

miRNAs/controls that resemble the mean

minimal standard deviation when comparing miRNA expression with mean ( geNorm V value, standard deviation of log transformed ratios)

compatible with multiplex assays

need to determine mean

small RNA controls

How ‘stable’ is the global mean compared to controls?

geNorm analysis using controls and mean as input variables

exclusion of potentially co-regulated controls

HY3 7q36

RNU19 5q31.2

RNU24 9q34

RNU38B 1p34.1-p32

RNU43 22q13

RNU44 1q25.1

RNU48 6p21.32

RNU49 17p11.2

RNU58A 18q21

RNU58B 18q21

RNU66 1p22.1

RNU6B 10p13

U18 15q22

U47 1q25.1

U54 8q12

U75 1q25.1

Z30 17q12

RPL21 13q12.2

miRNA expression datasets

neuroblastoma tumour samples

T-ALL samples

EVI1 deregulated leukemias

retinoblastoma tumour samples

normal tissues

normal bone marrow

0

0,2

0,4

0,6

0,8

1

1,2

1,4

1,6

1,8

exp

ressio

n s

tab

ility

T-ALL geNorm ranking

geNorm ranking

bone marrow pool normal tissues

neuroblastoma leukemia EVI1 overexpression

0

20

40

60

80

100

120

0 50 100 150 200 250 300

not normalised

stable controls

mean

miRNAs

neuroblastoma – removal of variation

removal of variation

bone marrow pool normal tissues

T-ALL leukemia EVI1 overexpression

biological validation

MYCN binds to the mir-17-92 promoter

CpG island

mir-17-92 cluster

+5 kb-5 kb

CA

TG

TG

CA

TG

TG

CA

TG

TG

CA

CG

TG

CA

CG

TG

CA

TG

TG

CA

TG

TG

A B C

0123456789

101112

A B C

Fold

en

rich

men

t

Amplicon

IMR5

WAC2

biological validation

choice of normalization strategy influences differential miRNA expression

Mir-17-92 expression in neuroblastoma tumours

0

0,5

1

1,5

2

2,5

3

3,5

stable controls

mean

miRNAs

biological validation

choice of normalization strategy influences differential miRNA expression

Mir-17-92 expression in neuroblastoma tumours

0

0,5

1

1,5

2

2,5

3

3,5

stable controls

mean

miRNAs

biological validation

choice of normalization strategy influences differential miRNA expression

Mir-17-92 expression in neuroblastoma tumours

0

0,5

1

1,5

2

2,5

3

3,5

stable controls

mean

miRNAs

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

fold

ch

an

ge

(M

YC

N a

mp

lifie

d v

s. M

YC

N s

ing

le c

op

y)

controls

mean

average FCcontrols = -0.404average Fcmean = 0.050average FCmiRNAs = 0.124

balanced differential expression

correlation MYCN downregulated genes – 2 normalization strategies

stable miRNA control normalisation

mean n

orm

alis

atio

n

strategy also works for microarray data

each sample is measured by RT-qPCR and microarray

global mean normalization

standardization per method

hierarchical clustering

samples cluster by sample (and NOT by method)

conclusions global mean normalization

novel and powerful miRNA normalization strategy

maximal reduction of technical noise

improved identification of differentially expressed genes

balancing of differential expression

universally applicable

o global mean

o multiple stable endogenous controls

Mestdagh et al., Genome Biology, 2009

most powerful, flexible and user-friendly real-time PCR data-analysis software

based on Ghent University’s geNorm and qBase technology

state of the art normalization procedures

o one or more classic reference genes

o global mean normalization

o expressed repeat normalization

detection and correction of inter-run variation

dedicated error propagation

fully automated analysis; no manual interaction required

booth 19

qbasePLUS normalization

http://www.qbaseplus.com

conclusions

proper normalization has a major impact on your results

provides statistically more significant results

enables accurate assessment of small expression differences

gold standard for mRNA gene expression analysis

geNorm evaluation of candidate reference genes

geometric mean of multiple stably expressed reference genes

global mean normalization and subsequent geNorm based selection of reference genes that resemble the mean is a valid option when measuring a large and unbiased set of genes (e.g. all miRNAs)

acknowledgments

miRNA

Pieter Mestdagh (UGent)

Frank Speleman (UGent)

Applied Biosystems

qbasePLUS

Jan Hellemans (Biogazelle – UGent)

Stefaan Derveaux (Biogazelle – UGent)

January 28-29, 2010

Ghent, Belgium

www.advances-in-genomics.org