The trials, tribulation and potential of FTIR spectroscopy in cervical cancer research

1
The trials, tribulation and potential of FTIR
spectroscopy in cervical cancer research
2
ICAVS July 2009 Melbourne The Worlds most
Liveable City
3
(No Transcript)
4
Outline

• Cervical cancer
• Early FTIR studies on exfoliated cells and
  confounding variables
• FTIR studies on cervical tissues
• Insights from 4D imaging
• Discussion points
• Acknowledgements

5
Significance

• Each year over 400,000 new cases of invasive
  cervical cancer are diagnosed world wide
• HPV strains 16 and 18, which together cause about
  70 of cervical cancer cases
• Current screening is via thin-prep PAP smear and
  PapNet
• Cervarix has been approved in Australia for women
  from 15-45 y.o. on May 24 2007

6
Cervical Histology
Mucosal Surface
Superficial
Intermediate
Parabasal
Basal
7
Architectural Features of Cervical Neoplasia
8
Infrared spectroscopy of exfoliated human
cervical cells Evidence of extensive structural
changes during carcinogenesis
P. T. T. WONG, RITA K. WONG, T. A. CAPUTO, T. A.
GODWIN and B. RIGAS Proc. Nat. Acad. Sci. USA
Vol. 88, pp. 10988-10992, December 1991

• Significant changes in the intensity of the C-O
  stretching bands at 1155 cm-1, 1047 cm-1 and 1025
  cm-1.
• Shifts of the peaks normally appearing at 1082
  cm-1, and 1244 cm-1 and an additional band at
  970 cm-1.

Changes in C-O stretching bands due to changes
in cellular metabolism Changes in phosphodiester
vibrations due to difference in nuclei DNA
9
Infrared Imaging
Whole cells 1. Exfoliates (eg. Cervical
cells) Many cell types present Remove unwanted
cells and debris if necessary white cell lysis
buffer red cell lysis buffer size selected
filtration For cervical smears ThinPrep does
this in a single step Deposition of a
single cell layer
Cells can be fixed to retain morphology or air
dried
10
Infrared Imaging
2. Cultured cells centrifugation as for
exfoliated cells tends to compact and shrink
cells sub structure affected good
adhesion/contact obtained culture directly on
Kevley slides or IR substrates good
adhesion/contact obtained fix using formalin
seems to be the best ( Gazi et al
Biospectroscopy, 77, 18-30 2005) unfixed
rapid drying or centrifugation drying cell
growth on ZnSe can be a problem Tobin et al
Faraday Discussions, 126, 27-38 2004
11
An Investigation into FTIR Spectroscopy as a
BiodiagnosticTool for Cervical CancerB. R.
WOOD, M. A. QUINN, F. R. BURDEN, and D.
McNAUGHTONBiospectroscopy, Vol. 2,143-153 (1996)
Category 1. All 6 spectra exhibit Type 1
profile Category 2. At least 1 spectrum exhibits
Type 2 profile from 250 samples
12
An Investigation into FTIR Spectroscopy as a
Biodiagnostic Tool for Cervical Cancer
13
Infrared Spectroscopy of Human Tissue. I.
Differentiationand Maturation of Epithelial
Cells in the Human CervixL. CHIRIBOGA, P. XIE,
H. YEE, V. VIGORITA, D. ZAROU, D. ZAKIM, M.
DIEMBiospectroscopy, Vol. 4, 4753 (1998)
1. Spectral data of single exfoliated cells,
obtained via infrared microscopy, indicate the
glycogen concentration of single cells can vary
enormously
2. The spectral features of nucleic acids do not
seem to contribute strongly to the spectral
features of the cells, in particular in the case,
of fully mature cervical cells
3. Thus we conclude that the nucleus is opaque
or nearly opaque in the mid-IR region Diem
Hypothesis!!!
14
GV oocytes
Experimental Wash 3 in PBS Deposit on
substrate (CaF2) with micro pipette Fix
with ethanol Wash 3 in water Removes phenyl
red, EDTA, oil
(35-120 µm)
Sample GV3
Amide I 1645 cm-1
Nucleic acids 1241 cm-1
Lipid 2950 cm-1
High histone levels in nucleus
High NA in cytoplasm
High lipid in cytoplasm
B.R.Wood, T. Chernenko, M. Diem, O.
Lacham-Kaplan, A. Trounson, C. Chong, D.
McNaughton, M. Tobin (In preparation for special
edition of ICAVS Vib. Spec.)
15
GV oocytes
Amide I 1643cm-1
16
GV oocytes
Lipid 2950 cm-1
17
IR beamline of Australian synchrotron
GV Oocyte
5 µm x 5 µm aperture 97 mA current 16 scans 6cm-1
18
IR beamline of Australian synchrotron
Nucleus and Cytoplasm
Wavenumber (cm-1)
5 µm x 5 µm aperture 97 mA current 16 scans 6cm-1
19
IR beamline of Australian synchrotron
MII Oocyte
(a)
(b)
20
A spy from Down Under !!!
Biospectroscopy, 4, 55-59 (1998)
Biospectroscopy, 4, 75-91 (1998)
21
Confounding variables Spies or no Spies !!!!
22
(No Transcript)
23
Unidentified Thread-Like Object (UTLO)
24
Our Seminal Study !!!!
25
Infrared Spectroscopy of Human Tissue. V.
InfraredSpectroscopic Studies of Myeloid
Leukemia (ML-1)Cells at Different Phases of the
Cell CycleS. BOYDSTON-WHITE, T. GOPEN, S.
HOUSER, J. BARGONETTI, MAX DIEMBiospectroscopy,
Vol. 5, 219227 (1999)

• DNA is detectable by infrared spectroscopy mainly
  when
• the cell is in the S phase, during the
  replication of DNA.
• In the G1 and G2 phases, the nucleic acid
  spectral contributions in the G1 and G2 phases
  would be mostly be that of cytoplasmic RNA.
• Results indicate differences between normal and
  abnormal cells may have been due to different
  distributions of cells within the stages of the
  cell division cycle

26
Removal of Blood Components from Cervical
Smears Implications for Cancer Diagnosis Using
FTIR Spectroscopy M. J. ROMEO, B. R. WOOD, M.
A. QUINN and D. MCNAUGHTON Biospectroscopy Vol.
72, 6976 (2003)
27
Observing the cyclical changes in cervical
epitheliumusing infrared microspectroscopyM.
J. Romeo, B. R. Wood and D. McNaughton
Vibrational Spectroscopy 28 (2002) 167175
28
FTIR mapping and Imaging of Tissue sections

• Can we determine cancer spectroscopic markers in
  tissue sections?
• Can we use FTIR imaging as a routine diagnostic
  for tissue sections?
• Are we really detecting chemical differences or
  are we really detecting physical/chemical
  differences?
• What do we need to do to standardise the
  procedure?
• Why we need to go to into the 4th Dimension !!!

29
3 Important Breakthroughs in FTIR tissue analysis

• Ag/SnO2 coated infrared reflective slides (Kevley
  Technologies)
• FTIR mapping/imaging capability

Peter Lasch
30
Infrared Imaging

• Sample preparation
• Tissue sectioning
• 1. Requires rapid fixation after biopsy or
  experiment
• 1 part phosphate buffered saline (PBS)
• 3 parts 16 formalin (6.4 formaldehyde)
• glutaraldehyde is also common
• cervical tissue fixed for 72 hours

Arrests autolysis and bacterial decomposition
Stabilizes the cellular and tissue constituents
so they withstand the subsequent stages of tissue
processing. Cross-links the proteins and lipids
preserving the basic macromolecular chemistry.
31
Infrared Imaging

• 2. Processing and storage
• a. paraffin blocks
• requires a series of alcohol immersions to
  soften and dehydrate the tissue for wax
  impregnation and exchange the ethanol with
  xylene
• An overnight procedure that removes lipids
• final step in warm wax that impregnates the
  tissue
• b. for cryo-sectioning
• embed in OCT (Optimal Cutting Temperature)
  matrix and freeze
• OCT tends to streak over section when cut
  and has strong IR spectrum
• epoxy resin or plastic is used by pathologists
  when 1 mm or less sections are required

32
Infrared Imaging
3. Sectioning and mounting For transmission
spectra require 8-10 mm For absorption/reflectio
n require 4-5 mm This needs an expert and the
right equipment NB Waxed sections can be cut
to 2 mm Cryo-sections harder to cut
typicallygt5 mm The cut wax ribbon is then
floated on water on to the substrate.
4. Dewaxing wash the tissue section three times
in clean xylene. NB. The standard wax removal in
most labs is not sufficient NB. If the area of
spectral interest is not covered by paraffin
absorptions then this step is not required
33
Methodology for Cervical sections
2
Tissue imbedded in paraffin
HE stained
2
Fixed with a series of alcohol dilutions and
washed in xylene
Ag/SnO2 Coated glass slide
1
2
X
Y
Photographed and compared with multivariate
cluster map
unstained
HE stained
34
Methods of Analysis. Pseudo-color images from
unsupervised hierarchical cluster analysis (HCA)
Correlation matrix CLM for all spectral pairs L
and M, summed over N spectral data points
merge items
Data hypercube ? dendrogram ? pseudo-color map
merge items
Wood, Chiriboga, Yee, Quinn, McNaughton Diem,
Gynecologic Oncology 93(4), 5968 (2004)
35
Fourier transform infrared (FTIR) spectral
mapping of the cervicaltransformation zone, and
dysplastic squamous epithelium
B.R. Wood, L. Chiriboga, H. Yee, M.A. Quinn, D.
McNaughton, and M. DiemGynecologic Oncology 93
(2004) 5968
36
FTIR Map 59 of CIN III sample recorded with 20
?m2 aperture 1000 x 1200 ?m2 area with 20 ?m step
size
1080 cm-1
1244 cm-1
37
Cluster map of cancer sample
m
m
m
m
300
1200
1200
m
m
300
m
m
38
The tenth cluster (orange) highlights two
potential foci of dysplasia (pre-malignant
cells)
39
FT-IR microscope
40
Rapid scan and Analysis

• FPA images can be obtained to 5 mm spatial
  resolution
• Pixel aggregation and tile mosaics allow larger
  tissue areas to be imaged
• A typical four tile (2?2) image mosaic at 4 pixel
  aggregation has 10 mm resolution, covers 0.6 ?
  0.6 mm tissue section and can be scanned in under
  10 minutes
• The resulting 64 ? 64 pixel image can be
  processed with HCA in under 10 minutes

41
Infrared Spectroscopic Imaging with a Focal Plane
Array (FPA)
Spectral Hypercube 64 x 64 FPA array X,Y
coordinate for each pixel wavenumber and
intensity values for each pixel
42
2nd Derivate spectra considerations effect of
smoothing Little smoothing gives images based
on instrumental differences only
7
9
11
Eg. Water vapour Noise Tiling due to
pixels on the periphery badly illuminate Smoothin
ggt11 required
15
13
17
43
Normal cervical tissue(Sample 1)

Epithelial tissue
44
Univariate maps of different spectral regions
Absorbance at 1651 cm-1
Integrated area of PO2 sym str signal
45
Hierarchical Cluster AnalysisWhole spectral
region 1800 950 cm-1
46
Hierarchical Cluster AnalysisAmide 1 region
1700 1570 cm-1
47
Hierarchical Cluster AnalysisPO2 sym str region
1300 1200 cm-1
48
Clusters calculated from 2nd derivative spectra
5 clusters Whole spectral region
Wavenumber (cm-1)
49
Cervical tissue exhibitingHPV, CIN1 and
endocervical glands (Sample 2)
50
Hierarchical Cluster AnalysisWhole spectral
region 1800 950 cm-1
51
Hierarchical Cluster AnalysisAmide 1 region
1700 1570 cm-1
52
Hierarchical Cluster Analysis PO2 sym str region
1300 1200 cm-1
53
Mean representative spectra corresponding to 5
cluster analysis(Sample 2)
1800 950 cm-1
1700 1570 cm-1
1300 1200 cm-1
Wavenumber (cm-1)
54
Artificial Neural Networks (ANN) as a method of
analysing FTIR images of thin cervical tissue
sections
55
IR spectra can be used to train a neural network
to classify individual cell types
Red blood cell
White blood cell
Connective tissue
Normal
Squamous cells
Pre-cancer
Columnar cells
cancer
Dyspaltic cells
cancer cells
56
Feed forward back propagation algorithm

• Choose the design parameters so obtain the best
  possible model with as few parameters as
  possible.
• Start with a linear model (neural networks with
  no hidden layers) then move to 1 layer (no more
  than 5-10 neurons)
• Use only a minimum of 2 input layers

57
Pruned ANNs
58
Cluster Spectra
59
Database of spectra for training artificial
neural networks (ANN)
60
Stuttgart Neural Network Simulator (SNNS)
http//www-ra.informatik.uni-tuebingen.de/SNNS/
Supervised feedforward backpropagation net
spectra
classifications
61
ANN results
62
ANN results
63
Windows retained in Pruned network
64
Immunoselection

• In tumour cells, there can be downregulation or
  complete loss of HLA class I molecules
• Escape recognition by T cells
• Therefore are able to proliferate become
  majority of the cells in tumour

Presenting HLA CI with tumour peptides
T cell
Proliferate
Kills tumour cells expressing Class I tumour
peptide
NOT Presenting HLA CI with tumour peptides
65
Column 1 FTIR Unsupervised hierarchical cluster
maps of non-stained sections of melanoma tissue
Column 2 FTIR Neural Network generated images of
non-stained sections of melanoma tissue. The NN
was trained with spectra of expressing and
non-expressing HLA class I cells
Column 3 HLA stained sections of tissue with
antibody directed to class I HLA molecules (red)
66
Journey into the Fourth dimension!!

• Enables visualization in 3 spatial and 1 spectral
  dimension
• Provides information on shape and depth of
  penetration of histopathological feature
• Images can be rotated and made translucent
• Enables whole cell in tissues to be imaged
• Minimizes orientation effects

67
Orientation effects
68
Software
www.cytospec.com
http//www.sci.utah.edu
69
2D Images

• There are potentially many different types of 2D
  imaging/mapping techniques that could form the
  starting point for generating 3D visualisations
  of FTIR data from tissue samples.
• Current work is exploring 2 promising starting
  points derived from CytoSpec software.

• (Physico-)Chem images can be obtained from IR
  band amplitudes, integrated areas or band ratios
• 2D Unsupervised Hierarchical Cluster maps can be
  computed from FPA hyperspectral data cubes

70
Sample exhibiting villoglandular papillary
adenocarcinoma and micro-glandular hyperplasia
71
2D Images

• FTIR images are stitched together using a program
  we have written using MatLab

• UHCA may be performed on the whole set in
  CytoSpec.

72
3D Images
May be constructed from stacked 2D scalar field
data in ASCII format
OR may be constructed directly from stacked 2D
images
73
SciRun Software
Something about how it can work with 2D image
files or with scalar field data files?
74
Adenocarcinoma in 4D
75
Adenocarcinoma in 4D semi-transparent mode
76
Mean extracted spectra form 3 major clusters
This seems just a bit over the top!!!!
0.8
0.6
Absorbance Units
0.4
0.2
0
1000
1100
1200
1300
1400
1500
1600
1700
1800
Wavenumber (cm-1)
77
FTIR imaging in 4-Dimensions Application to
heart diseaseB. R. Wood, T. Cox, J. Drenkhahn,
K. R. Bambery and D. Mcnaughton
78
Mouse Heart in 3D from 5 sections
Protein map-Highest concentration of protein in
red
79
Mouse Heart in 3D from 5 adjacent sections
As before except weakest protein absorption
areas removed
80
Mouse Heart in 3D from 5 sections
Chem image Integrated amide I peak (1620-1680
cm-1) As before except only strongest amide I
absorption areas retained
81
Build 4D NN map of an entire ovary (120 sections)
82
Discussion points

• Is shift in Amide I indicative of dispersion
  related effects due to different tissue density
  (protein density)?
• Can this information be diagnostically useful?
• Is it valid to perform linear decomposition
  methods prior to NN?
• Need to go 4D Neural-Network image reconstruction
  (with colour) to compete in market place
• Test subject for Instrument Companies (UHCA)

83
Acknowledgements

• Centre for Biospectroscopy
• Don McNaughton
• Michael Quinn
• Brian Tait
• Keith Bambery
• Phil Heraud
• Tony Eden
• Connie Chong

• CUNY/N.E. Uni.
• Max Diem
• Luis Chiriboga
• Tatyana Chernenko
• Melissa Romeo