Proteomic patterns of tumour subsets in nonsmallcell lung cancer

1
Proteomic patterns of tumour subsets in
non-small-cell lung cancer

• Yanagisawa, et al.
• Presented by Natalie Yip
• September 22, 2005

2
Outline

• Introduction
• Biological Problem
• Experimental Design
• Statistical Analyses
• Results
• Discussion and Conclusions
• Data Availability

3
Introduction

• This paper investigates the biology of non small
  cell lung cancer
• Interested in lung cancer because
• It is the leading cause of death from cancer in
  the USA for men and women
• increasing lung cancer incidence despite
  aggressive approaches to treatment and research
• 171,900 new cases and 157,200 deaths in 2003 in
  USA

4
Biological Problem

• What is the biology of cancer?
• Previous research approach
• cDNA microarray analysis
• Problems
• Poor correlation between mRNA expression and
  protein expression levels
• cannot detect post-translational changes to
  protein, which determine protein function

5
Proteomics approach

• Proteomics represent the effort to establish the
  identities, quantities, structures and
  biochemical and cellular functions of all
  proteins in an organism, organ, or organelle, and
  how these properties vary in space, time and
  physiological state http//www.genomicglossaries.c
  om/content/proteomics.asp
• Complement to the genome initiatives
• May be useful for understanding molecular
  complexities of tumor cells

6
Experimental Design

• Participant Selection
• NSCLC patients at Vanderbilt Univ. School of
  Medicine, March 1998-July 2002
• Tumor Resection
• All tumors studied preoperatively
• N2 node negative
• No preoperative therapy
• Normal tissue and tumors frozen in liquid
  nitrogen
• Patients assigned a postsurgical stage score
  based on histological group (adenocarcinoma,
  squamous-cell carcinoma, and large-cell carcinoma)

7
Experimental Design contd

• Cell and Tissue sample preparation
• Regions analyzed had tumor cellularity 70
• Cultured normal and lung cancer cells prepared in
  lysis buffer and mixed with matrix solution
• 12 mm sections cut from frozen tissue, dried in
  dessicator, and stained

8
Proteomic Analysis

• Technology used Matrix-assisted
  desorption/ionization time-of-flight mass
  spectrometry (MALDI-TOF MS)
• Directly assesses peptides and proteins in tumor
  sections
• High resolution imaging of individual
  biomolecules in tissue
• Each spectrum product of 256 laser shots over
  surface of matrix spot

9
Proteomic Analysis contd

• Signals in mass ranging between 2000 and 25,000
  M/S considered
• Each spectrums baseline was corrected by Data
  Explorer software
• Then spectra were binned using an algorithm for
  further statistical analyses

10
Statistical Challenge

• How can we analyze the results of the MALDI-TOF
  MS to learn about the biology of NSCLC?

11
Four steps of statistical analyses

• 1. Select important proteins
• Based on Kruskal-Wallis, Fishers exact, t-test,
  microarray analysis, weighted gene analysis, and
  modified information score method
• Respective cut-off points pp

12
Step 2 of Statistical Analysis

• 2. Create a prediction model
• Based on weighted flexible compound covariate
  method
• Verified proteomic patterns could be used for
  tissue classification
• Combined the most significant proteins
• Used new covariate (weighted sum of most
  important predictors) to reduce dimensionality
• Estimated misclassification rate

13
Steps 3 and 4 of Statistical Analysis

• 3. Apply the prediction model
• Test cohort
• Blinded samples
• 4. Investigate pattern of statistically
  significant discriminator proteins
• Used the agglomerative hierarchical clustering
  algorithm

14
(No Transcript)
15
Results

• over 1600 distinct protein species obtained
• 82 MS signals selected based on statistical
  criteria for normal vs. lung cancer
  classification
• Correct classification of all training (n50) and
  validation (n43) samples
• Hierarchical cluster analysis Correct
  classification of all training samples into
  histological groups
• Proteomic class-prediction model correctly
  classified all blinded test samples as primary
  and non-primary NSCLC

16
Results contd

• Pattern associated with nodal status
• Selected 2 proteins
• 85 accuracy in training cohort
• 75 accuracy in blinded validation cohort
• Protein expression profile correlation with
  patient survival
• 15 MS peaks to divide patients into good
  prognosis (median survival 33 months) and bad
  prognosis (med. Surv. 6 months)

17
Discussion

• Statistical limitations
• Small sample size
• Number of peaks based on statistical evidence,
  not minimum number of peaks needed to
  discriminate classes
• Unknown protein identities
• Identified proteins (SUMO-2 and thymosis-b4) of
  interest
• Successful use of proteomic analysis
• Great potential clinical usefulness

18
Data availability

• Unable to locate data set on web