Applications of Proteomics to the Management of Lung Cancer

Lung cancer is the third most common cancer in the United
States, yet it causes more deaths than breast, colon, pancreas,
and prostate cancer combined. 1 Attempts to alter
these statistics have been challenging in all respects. From
risk assessment to diagnosis, staging, assessment of response
to therapy, and prognostication, there is much room for improvement
in this disease (Fig. 9.1). Rapid developments in
technology have allowed the direct analysis of patterns of
protein expression in tumors and blood samples, and these
proteomic analyses promise to assist in making progress in
each of these areas.
Proteins are ultimately responsible for the function of the
vast majority of biological systems, and it is clear that many of
the crucial proteins in a cell are primarily regulated by posttranslational
modifications such as proteolytic processing,
phosphorylation, or acetylation. Thus, a full knowledge of
the derangement in the expression, modification, and function
of proteins in cancer cells is likely to be more informative
than study of DNA or RNA alone. The aim of proteomics
is therefore the characterization of proteins to obtain a more
integrated view of the biology. In order to further understand
the molecular biology of lung cancer, we need to probe these
tissues and related biological materials with tools that address
the molecular complexity of the proteome in lung cancer.
New technologies are being rapidly developed to allow the
increasingly thorough, systematic, and simultaneous analyses
of thousands of proteins in cancer cells. In particular, these
studies give us a unique insight into the biology of cancer,
can yield important new therapeutic targets, and may enable
the identification of novel biomarkers to differentiate tumor
from normal cells and predict individuals likely to develop
lung cancer.
In this chapter, we will review the progress made in clinical
proteomics as it applies to the management of lung cancer.
We will focus our discussion on how this approach may
advance the areas of early detection, response to therapy, and
prognostic evaluation.
PROTEOMICS TECHNOLOGIES
Sample Preparation With proteomics strategies, one
strives to identify novel proteins and understand their structure,
function, interaction with proteins and other molecules, and to
bring this knowledge to the clinic by means of new diagnostic
and predictive biomarkers and well as identification of therapeutic
targets. The rapid advance of mass spectrometry (MS)
and related technologies offers powerful new tools to analyze
the proteome. In contrast to standard protein biochemistry, proteomics
is defined as the study of the proteome, the complete
set of proteins produced by a species, using the technologies of
large-scale protein separation and identification (Table 9.1). 2
Protein biochemistry has long been exploited to understand
how biological systems function and lead to a cancer phenotype.
Early efforts were geared to study primarily one protein at a time
with biochemical methods of increasing power and sensitivity. The
world of immunoassays brought and continues to bring major
contributions to the field alone or in combination with MS.
In proteomic analysis, the isolation and preparation of
samples for analysis is of critical importance, and the precise
technique chosen depends on the scientific question being
addressed, whether it may be a comprehensive expression analysis,
evaluation of secreted proteins, nuclear proteins, proteins with a
particular modification (e.g., phosphorylation), or those that bind
to other proteins. Many approaches are available: some gel-based,
some based on separation by reverse phase chromatography, affinity,
size exclusion, ion exchange, and isoelectric focusing. The
separation/purification strategies all have the advantage of separating
the targets of interest from very abundant proteins in the
milieu. The trade-off is between the addition of complexity and
variability to the analysis and increased sensitivity. For example,
proteomics of blood samples is complicated by the fact that the
vast majority of the proteins in blood are made up of albumin
and immunoglobulin, but the proteins of interest may be 7 to 10
orders of magnitude less abundant, and much more readily found
if the abundant proteins are removed. An example of successful
separation strategy is immunoaffinity phosphoproteomics, which
combines immunoaffinity purification with tandem MS. Using
this approach, Rikova et al. 3 discovered oncogenic kinases such as
platelet-derived growth factor receptor (PDGFR)-alpha and discoidin
domain receptor family, member 1 (DDR1) that had not
been implicated in the pathogenesis of lung cancer.
The Mass Spectrometer A mass spectrometer analyzes
proteins after their conversion to gaseous ions, based on their
mass to charge ratio. It is essentially made of three basic elements:
an ion source for converting them to gaseous ions, a
mass analyzer for separating the ions by mass, and a detector
for detecting the ionized proteins (Fig. 9.2).
Tandem mass spectrometry (MS/MS) is a major analytic
tool used for evaluating proteins and protein complexes. With
this approach, protein samples are first digested with proteases
into a mixture of peptides and analyzed. Peptide ions are separated
in the first stage, then each peptide is fragmented in the
collision cell, and the fragments are then separated again to
identify them. The precise measurement of the mass of these
fragments allows the reconstruction of the identity and composition
of the original peptide (Fig. 9.3). There are many modes
of MS/MS. Different mass analyzers currently used for MS/
MS analysis are quadrupole ion trap (QIT), triple quadrupole
(TQ), quadrupole time of flight (QTOF), or Fourier transform
ion-cyclotron resonance (FTICR).
Specific mass spectrometers have specific applications. For
example, electron transfer dissociation (ETD) MS allows detailed
analysis of phosphorylated peptides that is of optimal
quality. 4 The LTQ integrates the steps of mass analyzer, collision
cell, and then another mass analyzer actually does it
tandem in time, meaning that three steps occur in the same
location—in an ion trap. An LTQ tandem MS is particularly
well equipped for excellent throughput, good sensitivity, MSn
capabilities, and a robust instrument. Each clinical question
needs to be addressed separately and proteomics may provide
tools to address them. The suite of technologies available to
researchers is ever increasing and their selection very much depends
on the goals and the type of samples to analyze.
Analysis of Complex Protein Mixtures The proteome
has multiple layers of complexity. The composition of
the proteome is not static like DNA. The structure is at least
an order of magnitude more complex than the genome, the
dynamic range in protein concentrations in given biological
specimens is huge (10 12 ), we have no target amplification
method such as PCR for genomic analysis, and the methods
used still have limitations in sensitivity primarily because of our
ability to separate protein complexes in subgroups pure enough
for analysis. Quantitative analysis is also challenging, but new
methodologies are being developed to address this challenge.
High-throughput analysis of the proteome without compromising
reproducibility is difficult as well.
The analysis of the complex mixture can be conceptualized
in two major ways, the top-down and the bottom-up
approaches. In the top-down approach, one starts with a specific
protein candidate; it is then separated, purified, and its
structure is identified. Recent technologies, such as FITRC
MS, increase the resolution and allow the analysis of larger
peptide fragments. However, the bottom-up approach takes
the challenge of embracing complexity from the start and directly
analyzes complex mixtures with a large number of proteins,
and uses computational peptidomics to reconstruct the
identities of the proteins in the mixture. This later approach
is less intuitive and may benefit from higher throughput.
It is recently being facilitated by modern bioinformatics
tools, enabling the analysis of proteomic digestion with different
enzymes than trypsin and therefore increasing the likelihood
of detecting increasing number of peptides mapping
to the same protein therefore improving the confidence of
identification.
Biomarkers The assumption underlying the concept of
proteomic biomarkers is that certain characteristics of proteomes
are highly correlated with specific clinically relevant
biological states. These characteristics include changes in expression
levels of proteins and the presence of specific modified
protein forms (Table 9.2). Specific effort has been exerted
in cancer proteomics to develop biomarkers for the early
detection of disease by analysis of plasma or serum proteins.
Detection of cancers at early stages maximizes survival, and
identification of blood-borne markers would lead to minimally
invasive tests.
The best biomarkers are those that are reproducibly measured,
related to the disease process, and trigger a clinical decision
resulting in improved clinical outcomes. Despite an intense search
for such biomarkers in the last 20 years, there are none currently
available for early diagnosis of lung cancer. 5 One reason for such a
lack of success thus far is the enormous challenge offered by lung
cancer development. The onset of the disease process is extremely
slow (months to years) and we have no means of evaluating the
rate of progression. Therefore, there is a critical need for new
biomarkers that are related to the disease process and that can
be measured early, easily, and repeatedly to assess progression of
the process.
Approaches to Biomarker Discovery Using Proteomics
Biomarker identification has been addressed by
multiple proteomic technologies (Fig. 9.4). MALDI profiling
is rapid, high throughput, but detects only the most abundant
proteins of relatively low molecular weight, and does
not enable direct identification when applied to complex
proteomes. Two-dimensional (2D) gel-based analysis suffers
problems of interlaboratory reproducibility and throughput.
More recent in-depth proteomic analyses are trying to overcome
these limitations and are summarized here.
High-Throughput Profiling Techniques The rapid proteomic
profiling of blood, tissue, or urine with minimal sample
preparation, using the peak pattern as a diagnostic tool, has generated
great enthusiasm and yet has been minimally successful
at providing robust signatures to translate to the clinic. In this
approach, the focus is on the use of MS peak patterns of abundant
proteins or peptide fragments that correlate with an early
disease stage but are usually not part of the disease mechanism.
MALDI TOF MS is capable for very high throughput where
a sample can be analyzed in seconds and has higher tolerance
for salts, buffers, and other biological contaminants. Because of
these qualities, MALDI MS has been utilized to study proteins/
peptides in serum, 6–10 urine, 11 tissue extracts, 12,13 whole cells, 14
and laser-captured microdissected cells. 15
Profiling using this technology in biological fluids or tissue
samples is not without challenges. The enormous complexity of
the sample composition, the large dominance of few proteins
in the sample, and their ability to mask lower abundance limits
the informativity of this approach. Truly tumor-derived markers
are likely to be present at low levels in blood, similar to levels
of the thousands of other proteins in blood that derive from
normal tissue leakage. Thus, the dynamic range of protein concentrations
adds a new dimension of technical considerations
to successful analysis of the serum/plasma proteome. 16 These
profiling experiments have been applied to a series of biological
specimens. Yet, reproducibility between platforms and institutions
remains a problem such that none of the profiling experiments
have yet made an impact in clinical medicine. This is in
contrast with greater early steps in the translation of genomic
signatures to the clinic. 17,18
Finally, protein arrays have recently been developed
and offer a series of targets printed onto different surfaces.
Proteins, 19 peptides, antibodies, 20,21 or lysates 22 will then be
detected by antibodies, serum, or multicolor detection systems.
The Swedish Human Protein Atlas (HPA) program proposes
a systematic analysis of the human proteome using antibodybased
proteomics combining affinity-purified antibodies with
protein profiling assembled in tissue microarrays. 23
In-Depth Proteomics Analysis The analysis of the
plasma proteome has made great progress in the last few years.
http://www.hupo.org/research/hppp/. This is largely a consequence
of novel methods of serum fractionation and MS-based
protein identification techniques; the number of plasma proteins
now includes major categories of proteins in the human
proteome. 24 The list confirms the presence of a number of
interesting candidate marker proteins in plasma and serum. 25
The detection of low-abundance proteins in the plasma requires
combinations of powerful technologies. The identification
of proteins whose expression levels are altered with
the disease state progression (2DE, MS, shotgun proteomics)
requires methodological improvements over the profiling
experiments. Methods related to separation of ions and ionization
have moved the field forward.
Two technology platforms have been developed to enable
unbiased discovery of candidate markers from tissues and biofluids
and verification of candidate markers by targeted analysis.
Unbiased discovery employs a shotgun proteomics platform
based on isoelectric focusing of peptides from tissue protein digests,
followed by reverse phase LC-MS-MS on Thermo LTQ
or LTQ-Orbitrap instruments. Verification is done by targeted
quantitation of peptides derived from biomarker candidate
proteins using liquid chromatography–multiple- reaction
monitoring MS (LC-MRM-MS). 26–28
In shotgun analyses, protein mixtures are digested to peptides,
which then are analyzed, most commonly by multidimensional
LC-MS-MS. MS-MS spectra encode the sequences
of peptides, as well as the masses and sequence positions of any
modifications (Fig. 9.5). Matching of MS-MS spectra to database
sequences enables identification of the peptides and the
proteins from which they were derived. Shotgun analyses by
LC-MS-MS also result in direct identification of the peptides
detected and provide for quantitative analysis of protein components.
Shotgun proteomics has proven the most versatile
and effective method for dissecting multiprotein complexes,
signaling networks, and complex subcellular proteomes. 29
Shotgun analyses can confidently identify 3000 to 5000 proteins
from a 200- g protein sample. Shotgun analyses have
generated the most complete proteomic inventories to date
of major eukaryotic subcellular organelles, whole cell and tissue
proteomes, and proteomes of human biofluids, including
plasma and serum. 28,30–33
Targeted quantitative analysis of top candidates can be
done by LC-MRM-MS analysis, as a first level of verification
of the shotgun results. Briefly, tissue lysates are run on
a NuPAGE gel and peptides are extracted, then injected in a
TSQ Quantum Ultra mass spectrometer. Peptides are loaded
and desalted and resolved in reverse phase chromatography,
eluted with a linear gradient. For MRM, four transitions are
recorded, and chromatographic peak areas for the transitions
are summed and compared to summed peak areas for beta
actin. Differences between peaks are evaluated for statistical
significance. Targeted LC-MRM-MS analyses can analyze up
to 100 candidates per run in individual tissue or plasma specimens.
Moreover, the application of both stable isotope tagging
and label-free quantitation has enabled the application of
shotgun proteomics to quantitative comparisons of complex
proteome samples. 34–39
Proteomic Data Analysis Analysis and interpretation
of the data derived from MS-based proteomic technologies
represents unique challenges as well. From MALDI MS experiments,
Spectra are generated in the mass-to-charge (m/z) 3000
to 50,000. Internal calibration is performed using internal or
external calibrants. The data processing consists of internal
calibration, smoothing, baseline correction, normalization to
the total ion current, feature selection with a signal-to-noise
ratio, and binning of features. This processing results in 100 to
300 m/z peaks per spectrum on average, using conservative parameters.
Statistical analyses of these data for biomarkers focus
on the selection of MS features and differential expression levels
between the study groups and on building class prediction
models based on the selected features. 40–44 The misclassification
rate is typically estimated using the leave one out crossvalidation.
From tandem MS analysis, raw data is extracted for individual
spectra with filters applied to remove obvious background
ions and low-quality spectra. These spectra yield a list
of peptide sequences and the frequency that each peptide is
detected. These sequences are searched against the NCBI protein
database to generate candidate proteins from which they
may have come. This list is filtered in various ways to reduce
the likelihood of false matches, and the protein and hit count
lists from different study groups are compared to generated
candidate biomarkers.
In summary, while genes carry the genetic information,
proteins are principal actors of vital regulatory processes.
Proteomics has many theoretical advantages over the more
established genomic and transcriptomic approaches to these
questions, but has been hampered by a number of technical
problems (Table 9.3). Another major challenge is the need for
extensive validation in using these novel global proteomics
research platforms prior to routine clinical applications. Once
interested in applying any new technology, one should remember
to carefully frame a biological question, understand the
literature, consider the limitations of each approach, and use
the most appropriate technology to answer the question.
EARLY DETECTION
Biomarker discovery for early detection of disease is made challenging
by the fact that patients are often identified late in the
course of disease, while it is abundantly clear that those treated
during early disease stages have a much better prognosis. Access
to samples before diagnosis is very difficult. Another major challenge
is the target sample to be analyzed. Complex samples such
as serum or urine although readily accessible add to the complexity
of the task. Tumor biomarkers for lung cancer can be
categorically classified into serum biomarkers, tissue biomarkers,
and sputum. Exhaled breath condensate is an interesting source
of material, but has not proven feasible yet. Serum biomarkers
stand out as being the most attractive at this time because of
their easy and routine accessibility. See also Chapter 22.
Biomarkers of early detection of lung cancer are still at an
early stage of development. 45 The Early Detection Research
Network (National Cancer Institute, division of cancer prevention)
has proposed a stepwise method for evaluating biomarkers,
and to identify people at risk (http://www.cancer.gov/edrn). 46
None of current biomarkers for the early detection of lung cancer
have passed the early validation (phase II). While genetics
has provided considerable insights into the molecular biology
of lung cancer, 47 the overall correlation between level of expression
of the messenger RNA molecules and protein expression is
relatively poor. 48 It is possible that proteomic technologies offer
a new avenue for biomarker discovery.
Proteomics-based early detection strategies for cancer
diagnosis include the analysis of complex mixtures such as
tissue samples, serum, plasma, sputum, and exhaled breath
condensate. The inherent analytical advantages of MS, including
sensitivity and speed, promise to make MS a mainstay of
biomarker discovery. The optimal use of these technologies depends
on the desired goal, such as protein identification, identification
of posttranslation modification, or determination of
protein–protein interactions.
The direct analysis of serum proteomes to detect disease
markers has attracted widespread interest and intense scrutiny.
Early work using either MALDI or a proprietary MALDI
variant termed surface-enhanced laser desorption ionization
(SELDI) demonstrated that spectra of crude serum protein
mixtures or subfractions displayed differences in spectral features
that appeared to correlate with disease status. 49–51 The
application of multivariate analytical methods generated models
that correlated spectral features with disease status.
To determine the diagnostic accuracy of MALDI mass spectrometric
analysis of serum in lung cancer, we used MALDI-MS
to analyze undepleted and unfractionated serum from a total
of 288 NSCLC patients and matched controls divided into
training (92 cases and 92 controls) and test (50 cases and
56 controls) sets. 10 In the training set, they defined a sevensignal
proteomic signature distinguishing lung cancer serum
from matched controls with an overall accuracy of 78%, a sensitivity
of 67.4%, and a specificity of 88.9%. In the test set, the
signature reached an overall accuracy of 72.6%, a sensitivity of
58%, and a specificity of 85.7%. As diagnosis of early stage lung
cancer is important, authors searched for a protein signature
discriminating stage I lung cancers from controls and found
a six-signal signature reaching 70.8% sensitivity and 84.4%
specificity in the training set, and 57.1% sensitivity and 71.4%
specificity in the test set (Fig. 9.6).
With a multivariate logistic regression model applied on a
total of 223 cases and controls, they showed that the serum signature
was associated with lung cancer diagnosis independently
of gender, smoking status, smoking pack-years, and C-reactive
protein levels, and had the strongest association with lung cancer
diagnosis among all the covariates in the model. Using SELDITOF-
MS on serum samples from 158 lung cancer patients and
50 controls, Yang et al. 52 reported a five-signal protein signature
distinguishing lung cancer cases from controls with 86.9%
sensitivity and 80.0% specificity in the validation set.
Initial reports suggesting that these features comprise new
families of biomarkers were followed by considerable critical
analysis, which pointed out several problems. 53–55 First,
MALDI and SELDI analyses were subject to systematic bias
due to inconsistent sample collection, processing, and poor
instrument calibration. Second, the analyses of small numbers
of samples displaying large numbers of spectral features led to
models that “overfit” the data and did not scale effectively to
larger sample numbers. Third, spectral features that correlated
to disease state were found to be poorly reproducible between
different laboratories. Finally, MALDI- and SELDI-based
methods did not enable direct identification of the protein and
peptide species that constituted putative markers. Eventual
identification of some of the species associated with spectral
signals revealed that they were all abundant blood proteins or
their proteolysis products, many of which were produced ex
vivo during sample handling or serum preparation. 56,57
Moreover, some of the most characteristic markers (e.g.,
serum amyloid A) were associated with multiple cancers. 10,58
Recent work by Villanueva et al. 59 demonstrated that blood
from patients with different cancer types yielded characteristic
sets of proteolysis products of abundant blood proteins during
serum preparation.
The proteomic characteristics observed in MALDI and
SELDI analyses are interesting but may ultimately be of questionable
value in the detection and diagnosis of cancer in
human populations. Clearly, the majority of identified species
do not arise from cancers themselves, but rather from systemic
responses to disease. The distribution of protein proteolysis
products is exquisitely sensitive to sampling methods and
processing, and the different proteolysis products can only be
distinguished by MS instruments. Most importantly, the relationship
of these putative markers to cancer is unclear, as is
their ability to distinguish different cancers.
Ultimately, the early diagnosis of lung cancer will be addressed
in an integrated way after gathering information from
clinical, biological, imaging, and molecular data (Fig. 9.7) .
PROTEOMICS FOR CLASSIFICATION
OF PROGNOSIS
The clinical behavior of individual patients with lung cancer is
extremely diverse. Some tumors progress rapidly with widespread
metastases, and some grow only very slowly over the course of
years and never result in clinical symptoms. Knowledge of this
propensity when selecting initial therapy would be extremely useful
in determining whether treatment is needed at all and how
aggressive an approach is indicated. Deciding on the need for
adjuvant chemotherapy is an example. Presumably, this diversity
results from variability in the precise molecular makeup of individual
tumors, and a pattern of molecular features associated
with clinical course, independent of therapy, is referred to as a
prognostic signature. Features associated with the benefit or lack
of benefit from a specific intervention are considered to make up
a predictive signature, and will be addressed in the next section.
As discussed previously, many of the functionally important
molecules in the development of cancer are regulated by
posttranslational modifications, implying that direct assessment
of the proteins themselves might be more informative
than RNA or DNA. Attempts have been made to apply proteomic
technologies to uncovering an accurate prognostic signature
for lung cancer. Gharib et al.60 used 2D gels to study
the differences in protein expression patterns for 93 resected
lung adenocarcinomas and 10 normal lung samples, and assessed
their association with survival. In this study, they found
that two of the five cytokeratin 7 isoforms, one of eight CK8
isoforms, and one of three CK19 isoforms were associated with
survival. One of these, CK19, had independently been found
to be a useful tumor marker for lung cancer. Also uncovered
in this dataset was the significant downregulation of seleniumbinding
protein 1, also strongly associated with survival. 61
Another study used 2D gels to study 20 squamous cell lung
cancers with matched normal tissues 62 and identified tumor/
normal differential expression of mdm2, c-jun, and EGFR,
and found 26 proteins reactive to patient autoantibodies, but
no survival analysis was done. The finding that not all of the
CK19 isoforms conveyed prognostic significance underscores
the power of proteomic technologies to uncover these associations.
Unfortunately, because specific antibodies for individual
isoforms are rarely available, it also underscores a problem with
this approach in the practical clinical measurement of these
features.
MALDI-TOF has also been applied to this problem, and
in a study of 79 tumors and 14 normal lung tissues, classifiers
were defined that were able to discriminate tumor from normal,
and lung cancer from metastatic disease. 40 A 15-protein classifier
was constructed that was able to predict survival after
resection These proteins included thymosin beta4 and SUMO-2,
both with interesting biological implications. Thymosin beta4
has been associated with inhibition of caspase-3 in taxolinduced
tumor cell death, 63 as well as stabilization of
HIF-1 alpha, 64 and its expression is regulated by hMLH1. 65
SUMO-2 is also associated with multiple important cancer
pathways. 66–70 A subsequent study of a larger cohort of 174
NSCLC tumors with a longer follow-up derived a 25 signal
classifier overlapping with the previous classifier, and also able
to divide patients into high- and low-risk groups. 71 However,
MALDI-TOF is limited by nonstandardized or transportable
analytic platforms and its low sensitivity and selectivity for
low–molecular-weight proteins. It is also difficult to identify
the precise protein detected, making translation to clinical
practice difficult.
Analysis of serum by SELDI-TOF has also uncovered an
unidentified 4628 Da protein associated with survival in 87
advanced stage NSCLC patients. 72 However, this was evaluated
only in a cross-validation approach and not an independent
test set, and needs additional validation.
LC-MS/MS analysis of lung cancer tumors has great
promise for the discovery of tumor signatures, as it is capable
of detecting many more proteins than MALDI-TOF and gives
the identity of the observed protein features. In one very preliminary
study, 73 24 surgically resected adenocarcinomas were
analyzed by SDS-page gel followed by in-gel digestion and C-18
LC separation and MS/MS analysis. They identified 51 candidate
signals and selected two of the corresponding proteins,
myosin IIA and vimentin, as biomarkers. When evaluated by
immunohistochemistry, they were able to use these two markers
to classify good and poor prognosis groups (Fig. 9.9).
None of these markers have reached routine clinical practice
to date, at least partly due to the difficult steps of taking a
research lab–based assay for large numbers of candidate markers
and translating that into a reproducible and accurate highthroughput
commercial product capable of validating them
in a prospective fashion in large enough cohorts of patients.
Work continues in this direction, however.
RESPONSE TO THERAPY
More important than simple prognostic classification is predicting
response to therapy. Identification of patients destined to do
poorly regardless of therapy is much less interesting than identification
of specific therapies capable of assisting in the selection
of the optimal intervention able to alter a patient’s outcome.
Several proteomic studies have attempted to define protein
signatures capable of predicting benefit from specific
interventions. In its simplest form, single immunohistochemical
marker studies are being tested for their utility in defining patients
who will benefit from specific therapies. For example, immunohistochemical
expression of excision repair cross- complementing
rodent repair deficiency, complementation group 1 (ERCC1) is
associated with lack of benefit from platinum-based therapy, 74
and high thymidylate synthase expression in squamous cell lung
cancer predicts lack of benefit from pemetrexed. 75 Some studies
are attempting to use proteomic technologies to measure the
levels of several candidate markers and cytokines in the blood and
correlate these with response. One such study 76 looked at biomarkers
associated with benefit from bevacizumab, and antibody
against VEGF, and found that baseline intercellular adhesion
molecule (ICAM) levels were prognostic for survival and predictive
of response to chemotherapy with or without bevacizumab,
and that VEGF levels were predictive of response to bevacizumab
but not survival. Other studies have attempted to find bloodbased
biomarkers of response to VEGFR tyrosine kinase inhibitors.
77 Using high-throughput platforms, several groups are now
studying the utility of measuring 100 or more such markers simultaneously.
78,79
2D gel analysis of squamous cell lung cancer has identified
candidate proteins associated with resistance to mitoxantrone, 80
and taxanes, 81 but no clinical candidate markers have been identified
in these datasets. A study of H322 and H1299 lung cancer
cells using 2D gels identified thioredoxin reductase to be associated
with resistance to the histone deacetylase inhibitor depsipeptide.
82 2D differential in-gel electrophoresis (DIGE) technology
also identified nine proteins associated with response to the EGFR
tyrosine kinase inhibitor gefitinib. 83 These proteins included fatty
acid binding protein and glutathione-S- transferase P, and application
of this signature to an external sample set confirmed the
predictive ability, but only involved 14 patients.
To date, MS/MS proteomic analysis of lung cancer tumors
has also failed to find markers predictive of response to
uracil-tegafur, 73 but other studies looking at other chemotherapies
are in progress. The potential power of tandem MS not
only for identification of markers predictive of benefit from
therapy, but also new targets for therapy in specific tumors is
demonstrated in a recent study of the phosphoproteome of
lung cancer cells. 3 This study evaluated tyrosine phosphorylated
peptides in 41 NSCLC cell lines and over 150 tumors
and found known therapeutic target phosphorylation as well
as activation of potential but not previously identified kinases
such as DDR1. In spite of this progress, to date, none of the
markers from 2D gels or mass spectrometric analysis of lung
tumors has advanced to careful clinical testing.
Probably most surprising is the report that a serum protein
signature has been reported that could accurately define patients
with good or poor survival after treatment with gefitinib or
erlotinib. 84 It is more intuitive that an accurate predictive signature
is more likely to be defined from tumor protein expression
patterns than from patterns in the blood, but this group defined
a signature of eight proteins that was capable of classifying good
and poor outcomes in a training cohort of 139 second-plus line
patients treated with gefitinib and applied this classifier in a
blinded way to two independent test cohorts, one second-plus
line treated with gefitinib, and one a cooperative group trial of
erlotinib in first-line therapy (E3503) (Fig. 9.10).
Remarkably, highly statistically significant classification was
achieved for both progression-free survival and overall survival
in these cohorts, and not in three control cohorts treated with
chemotherapy or surgery alone. None of these patient cohorts
was from a randomized trial, however, so it is impossible to determine
if this classifier is truly specifically predictive of benefit
from erlotinib. This test has been commercialized and prospective
testing is in progress.

The rapid development of proteomic technologies has provided
a large amount of novel information leading to the assembly
of large protein inventories and a better understanding of how
they interact, the role of specific posttranslational modifications,
and advances in biology. Proteomic analysis has the potential to
profile differences between lung tumor and no tumor, between
different stages and histology of cancer, and between different
cancer samples at the same stage of progression. The ability to
identify important proteins involved in the transformation process
may lead to early markers for detection of specific types of
cancers and treatments based on the molecular profile of lung
cancer. Molecular profiling may assist in identifying high-risk
populations and offers a unique opportunity to study early carcinogenesis
and potentially to reduce cancer mortality through
its integration with genomics. The importance of clinical proteomics
comes from the fact that it will have a fundamental impact
on our understanding into complex disease processes, such
as lung cancer, and will offer new opportunities in the diagnosis,
prognosis, and therapy of disease. The development of specific
and sensitive diagnostic biomarkers using biological fluids,
such as sputum and serum, should improve screening, early detection,
monitoring of disease progression, treatment response,
and surveillance for recurrence. Proteomic biomarker discovery
analysis is still early in this process and will benefit from
these technologies to detect, identify, and specifically quantify
protein markers. The biological amplification of protein signals
through the immune system may also claim autoantibodies
as potential biomarkers. The development of immunoaffinity
assays to validate candidate biomarkers is required. Finally, a
major challenge is the need for extensive validation in using
these novel global proteomics research platforms prior to routine
clinical applications.