Abstract—In this paper, we present a simple method for the
processing and quantification of multiplexed Quantum Dot
(QD) labeled images of clinical cancer tissue samples. QDs
provide several features which make them ideal for reliable
quantification, including long-term signal stability, high signal-
to-noise ratios, as well as narrow emission bandwidths.
Deconvolution of QD spectra is accomplished in a batch mode in
which unmixing parameters are preserved across samples to
allow for quantitative and reproducible comparisons. After
unmixing the QD images, we segment each one to exclude
acellular regions. We use a simple average intensity to quantify
the level of QD staining for each image. We illustrate the
viability of this approach by testing it on 28 tissue samples using
a tissue microarray. We show that using as few as two QD
protein targets (MDM-2, and B-actin), the Renal Cell
Carcinoma (RCC) samples are distinguishable from adjacent
normal tissue samples. A simple linear discriminant results in
100% classification of 25 RCC samples and 3 normal samples.
This suggests that multiplexed QDs can be used to properly
diagnose RCC from otherwise healthy tissue. We expect to
apply this work to larger panels of more robust QD biomarker
targets to aid in clinical decision-making for the diagnosis and
prognosis of diseases, such as cancer.
I. INTRODUCTION
ORE than 50,000 men and women in the United States
are estimated to be diagnosed with kidney cancer in
2008 . Renal cell carcinoma (RCC) exists in a variety of
subtypes each of which has a different prognosis,
progression, and recommended treatment strategy [1].
Recent advances in high-throughput genomics, largely due to
Manuscript received April 16, 2008. This work was supported in part by
the National Institutes of Health (Bioengineering Research Partnership
R01CA108468, P20GM072069, Center for Cancer Nanotechnology
Excellence U54CA119338), Georgia Cancer Coalition (Distinguished
Cancer Scholar Award to MW) Microsoft Research, National Science
Foundation (GRFP Fellowship to RM), and a Georgia Tech Institute of
Bioscience and Bioengineering Seed Grant.
M. L. Caldwell is a BS candidate at the Georgia Institute of Technology,
Atlanta, GA 30318 USA (e-mail: mcaldwell@gatech.edu).
R. A. Moffitt is a Ph.D. candidate at the Georgia Institute of Technology
and Emory University, Atlanta, GA 30322 USA (e-mail:
r.moffitt@gatech.edu).
J. Liu is a Postdoctoral fellow at Emory University (e-mail:
jliu44@emory.edu).
R. M. Parry is a Postdoctoral fellow at Georgia Institute of Technology
(e-mail: parry@bme.gatech.edu).
Y. Sharma is a Ph.D. candidate at the Georgia Institute of Technology
(e-mail: ysharma3@mail.gatech.edu).
M. D. Wang is an Assistant Professor at the Georgia Institute of
Technology, and Emory University (corresponding author; phone: 404-385-
5059; e-mail: maywang@bme.gatech.edu).
the maturation of RNA microarray technology, have
provided insights into the unique molecular profiles
exhibited by these renal cancer subtypes [2]. This in turn has
caused a shift in emphasis away from generalized cancer
therapies and towards personalized and subtype specific
treatments [1].
Current methods of RCC molecular subtyping rely almost
entirely on the use of traditional organic dyes to label
relevant biomarkers. However, traditional dyes present a
number of obstacles to accurate quantification. First, many
common dyes fluoresce over a broad range of wavelengths.
This property makes it difficult to multiplex biomarkers as
the emission spectra of different dyes often overlap with
each other as well as excitation sources, and are difficult to
unmix spectrally. In addition, most organic dyes are
susceptible to photobleaching, which limits their usefulness
as a quantitative tool as the intensity of the fluorescence
degrades quickly with repeated measurements.
The lack of multiplexing ability for organic dyes has led to
a focus on more nonspecific and qualitative methods of
cancer image classification, mainly analyzing morphology
using hematoxylin and eosin (H&E) staining. Computational
methods to automatically assess such images have had some
success [3] but have not yet been widely implemented.
Multiplexed bioconjugated QD staining is an emerging
technology which has the potential to greatly increase the
accuracy of cancer diagnosis and subtyping through
molecular profiling. This technology has several properties
which help overcome the known issues with current methods
based on organic dyes. Quantum dots are nano-scale
semiconductor crystals with broad excitation spectra and
relatively narrow emission spectra which are controlled by
the size of the crystals [4]-[9]. These properties allow the
simultaneous excitation of multiple QDs in the same image
with one light source as well as the simultaneous and
separable observation of 5 or more different QDs in the
visible and infrared range [5,8]. Furthermore, QDs are
significantly brighter than most organic dyes and extremely
resistant to photobleaching [4]-[9]. This combination of
properties gives QDs the potential to be a robust and easily
quantifiable tool for molecular profiling.
Despite the promise of QDs as a quantitative pathology
tool, significant obstacles still remain to their use in a clinical
setting. One issue is the a priori identification of the relevant
regions in the tissue sample before quantification, a task that
Simple Quantification of Multiplexed Quantum Dot Staining in
Clinical Tissue Samples
Matthew L. Caldwell Student Member, IEEE, Richard A. Moffitt, Graduate Student Member, IEEE,
Jian Liu, R. Mitchell Parry, Yachna Sharma, and May D. Wang, Member, IEEE
M
30th Annual International IEEE EMBS Conference
Vancouver, British Columbia, Canada, August 20-24, 2008
978-1-4244-1815-2/08/$25.00 ©2008 IEEE. 1907