1) Secondary ion mass spectrometry (SIMS) involves firing primary ions at a solid surface, which sputters secondary ions, electrons, and neutrals that can be analyzed with a mass spectrometer. 2) SIMS provides highly surface sensitive analysis but has limitations such as limited sample size and overlapping spectra. Multivariate analysis techniques like principal component analysis and canonical variates analysis are used to analyze SIMS data. 3) SIMS can produce 2D spatially resolved chemical images of sample surfaces by collecting full mass spectra at each pixel. Techniques like maximum autocorrelation factor analysis incorporate spatial information with spectral data for more detailed analysis of SIMS images.

1. SECONDARY ION
MASS SPECTROMETRY
DATA ANALYSIS IN 1D, 2D AND 3D
ALEX HENDERSON
Surface Analysis Research Centre
University of Manchester, UK

2. SIMS technique
 Fire energetic ions at a
solid – primary ions
 Solid emits (sputters)
positive ions
negative ions
electrons and neutrals
 These secondary ions
produce a mass spectrum
Secondary Ion
Mass Spectrometry

3. Comparison with other MS
 No (chromatographic) separation step so spectra are
overlapping
 Limited amount of sample available
 Relatively low throughput so no large databases of spectra
 Highly surface sensitive so detects contamination easily (pro
and con)
 Need to transport analyte to gas phase; proteins decompose,
polymers generally OK
 MS/MS being developed (here)
 Negative ion spectra available

4. 1D analysis
Collections of spectra

5. O
H 2C O C C 15H 31
O
Typical SIMS spectrum
HC O C C 15H 31
O
+
H 2C O P O CH2 CH2 N(CH 3) 3
OH
1,2-Dipalm itoyl-sn-glycero-3-phosphatidylcholine (DPPC) (positive ion)
50,000
45,000
CH3
40,000 CH2 CH N
+
CH3 O CH3
+
35,000 CH3 HO P O CH2 CH2 N CH3
OH CH3
30,000
Intensity
25,000
20,000
15,000
10,000
5,000
0
0 20 40 60 80 100 120 140 160 180 200
m/z
The Static SIMS Library, SurfaceSpectra Ltd
 100 – 250k data points
 1 – 3000 u mass range
www.surfacespectra.com

6. Spectra of bacteria
Enterococcus spp. Proteus mirabilis
4 4
x 10 En x 10 Pm
6 5
4.5
5
4
3.5
4
3
3 2.5
2
2
1.5
1
1
0.5
0 0
0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 700 800
m/z m/z
Applied Surface Science 252 (2006) 6719-6722

7. Spectra of bacteria
Enterococcus spp. Proteus mirabilis
En Pm
8000 2500
7000
2000
6000
5000
1500
4000
1000
3000
2000
500
1000
0 0
100 150 200 250 300 350 400 450 500 100 150 200 250 300 350 400 450 500
m/z m/z
Applied Surface Science 252 (2006) 6719-6722

8. PCA of bacteria
5 species of bacteria
No a priori
knowledge used
in PCA
Positive ion spectra
1-800 u, rebinned to
1 u steps
Square root of
intensity
Sum-normalised
Applied Surface Science 252 (2006) 6719-6722

9. Canonical Variates Analysis
(Discriminant Function Analysis)
 Principal components are pure (orthogonal) aspects
of the data
 Canonical variates are combinations of PCs that
best describe an a priori class structure
 Based on Fisher’s Linear Discriminant, subtly
different from Linear Discriminant Analysis (LDA): no
assumption of normally distributed classes
 CVA is also referred to as DFA

10. Canonical Variates Analysis
(Discriminant Function Analysis)
 Maximise ratio of between-group variance to within-
group variance

11. PC-CVA bacteria classification
5 species of bacteria
Class structure used
9 PCs selected using
PRESS test
Cross-validation
indicates percent
correctly classified:
Cf 75% CC
Ec 92% CC
En 100% CC
Kp 25% CC
Pm 50% CC
Applied Surface Science 252 (2006) 6869-6874

12. 2D analysis
Spatially organised collections of spectra
Hyperspectral images

13. Imaging SIMS
 Focussed ion beam stepped
from point to point
 2D data generated
 Full mass spectrum at each
point/pixel
 Beam diameter down to
~1-2 µm
(50 nm possible, but can
damage sample)

14. SIMS image - ‘chemical photograph’
4500
4000
3500
3000
intensity
2500
2000
1500
1000
500
0
0 100 200 300 400 500 600
m/z
Total ion spectrum Total ion image
Bead diameter 50 μm
Data courtesy of Penn State University, USA

15. Selected ion images
4500
4000
3500
3000
intensity
2500
2000
1500
1000
500
0
0 100 200 300 400 500 600
m/z
Data courtesy of Penn State University, USA

16. Maximum Autocorrelation Factor Analysis
(MAF)
 Incorporates spatial information with spectral
information
 Determine the correlation with the same image
offset diagonally by 1 pixel
 Autocorrelation means scaling invariant
 Computation time ~4× PCA
 Possible issue when image feature size ≈ pixel size

17. MAF versus PCA of HeLa cells
a b c
a) Total ion image, scaled and smoothed
b) Score on PC 6, scaled and smoothed
c) Score on MAF 4, scaled and smoothed
MAF captures the distinction between inner and outer cellular regions
more clearly than PCA. Comparison with the total ion image shows the
additional information available following multivariate analysis.
Surface and Interface Analysis 41 (2009) 666-674

18. MAF versus PCA of HeLa cells
Loading on PC 6 Loading on MAF 4
 MAF loadings more ‘informative’
Surface and Interface Analysis 41 (2009) 666-674

19. High mass resolution required
Nominally all at m/z 86
 Lipid (DPPC)
m/z = 86.0969692
 Unknown
 Silicon substrate [Si3H2]+
m/z = 85.9464332
Separation = 0.15 u
m/z 86
Analytical chemistry 80 (2008) 9058-9064

20. Rat kidney tissue
 Cryo-microtomed slices
 Sequential layers
 Unwashed
 Washed in methanol
 Data is
 10x10 tiles
 32x32 pixels per tile
 24000 channels per pixel
 2.4 billion datapoints
 19.2 GB (as doubles in MATLAB)
Almost all zeros, so use sparse matricies to handle the data

21. Summation of all tissue pixels

22. Clear changes in chemical distribution over small mass range
Individual images from nominal mass m/z 197
Images generated at 0.05 u increments
from 196.75 to 197.30
Images are 9 x 9 mm2 comprising
10 x 10 tiles of 32 x 32 pixels
John S Fletcher Ionoptika J105 April 2009
Analytical and Bioanalytical Chemistry 396 (2010) 85-104

23. Rat kidney section - overlay
320 x 320 pixels
9 x 9 mm
Full mass spectrum
available at each pixel
Lipid (DPPC) DAG Salt related
Analytical and Bioanalytical Chemistry 396 (2010) 85-104

24. SIMS and AFM
AFM SIMS (total ion)
Identify points of
reference between (Human cheek cells)
images
Register images
and overlay
AFM following
‘projective’ transform
and cropping in
MATLAB
Applied Surface Science 255 (2008) 1264–1270

25. SIMS and AFM
nm
µm
µm
 Use AFM z-height to offset SIMS data
 Use AFM x- and y-position to accurately scale SIMS
x- and y-positions
Applied Surface Science 255 (2008) 1264–1270

26. 3D analysis
Three-dimensional spatially organised collections
of spectra
‘Hyper-’hyperspectral images

27. HeLa-M cells
 Immortalised cervical
cancer cell line commonly
used as a model system
for biology
 Cultured on poly-L-lysine
coated substrates
 AFM after freeze drying:
30 – 50 µm diameter
0.6 – 1µm thick

28. Formalin fixed, freeze-dried HeLa cells
Lipid (DPPC) increasing ion dose
Around the nucleus
184
152
Inside the nucleus
 1st layer: evidence of chemistry having spread
outside cell area
 3rd layer: spectra show chemistry still mixed up
between nucleus and membrane
 4th layer: one cell exhibits bimodal nucleus
Rapid Commun. Mass Spectrom. 25 (2011) 925-932 indicating the dividing stage of cell cycle – mitosis

29. 3D visualisation – issues
 Problem with point of view of ‘observer’
 Mass analyser is normal to sample
 Different sample heights appear to be at same
level
 Data size:
 256 x 256 pixels x 10 layers x 11400 channels
 7.5 billion datapoints = 58 GB
 >99% of data is zero so only ~5 GB in RAM

30. 3D visualisation – our approach
 PCA of whole dataset (sparse, MATLAB)
 Identify PC that separates substrate from organic
overlayer
 Assume substrate is flat
 Slide columns of pixels to align with crossover in
score of that PC
 Discard substrate-rich pixels
 Repeat PCA to identify peaks or components
 Render results (AVS Express)
Rapid Commun. Mass Spectrom. 25 (2011) 925-932

31. Two sample preparation methods
Aim is to maintain the biological integrity of cells in vacuum system
Formalin fixed, freeze-dried cells Frozen hydrated cells
 Hela-M cells cultured on poly-  Hela-M cells cultured on one
L-lysine coated silicon wafer plate of hinged two-plate steel
 Formalin fixed (4%) substrate coated in poly-L-lysine
 Cells are washed with  Cells are washed with ammonium
ammonium formate (0.15 M) formate (0.15M) to remove salts
prior to analysis to remove  Cells trapped between plates
salts and rapidly frozen in liquid
 Freeze-dried in vacuum propane
 Sandwich fractured in vacuum at
160 K to reveal cells
Rapid Commun. Mass Spectrom. 25 (2011) 925-932

32. Frozen hydrated HeLa cells
Lipid (DPPC headgroup) @ m/z 184.1 Nucleic acid (adenine) @ m/z 136.1
Rapid Commun. Mass Spectrom. 25 (2011) 925-932

33. Formalin fixed, freeze-dried HeLa cells
Lipid (DPPC) @ m/z 184.1 Principal component indicating
Rapid Commun. Mass Spectrom. 25 (2011) 925-932 guanine & tryptophan

34. Movies available online
256 x 256 pix  Frozen hydrated HeLa cells
x 10 layers  http://www.youtube.com/watch?v=sPuMYUQPvHQ
x 11400 chans
 Visualisation of the membrane (green, m/z 184.1,
phosphocholine) and nucleus (red, m/z 136.1, adenine)
7.5 billion pts  Formalin fixed, freeze-dried HeLa cells
= 58 GB
 http://www.youtube.com/watch?v=2phMkYo7yhg
 Visualisation of the membrane (green, m/z 184.1,
>99% of data
phosphocholine) and nucleus (purple, negative scores
is zero so only on principal component 5)
~5 GB in RAM  Possible mitosis visible at lower right

35. Summary
 What SIMS lacks in ultimate species identification it
makes up for in spatial location
 New ionisation sources produce data nearer to
‘traditional’ MS, opening up resources
 Tandem MS approaches are breaking through

36.  SARC
SARC  Validation algorithms

John Vickerman
John Vickerman  Roy Goodacre
Nick Lockyer
 Nick Lockyer
John Fletcher  Yun Xu
 John Fletcher
Sadia Sheraz
(nee Rabbani)
 Elon Correa
... Sadia Rabbani (now Sheraz)
 and the rest!  Visualisation group in
 ... and the rest! Research Computer Services
Acknowledgements
www.sarc.manchester.ac.uk

37. www.sarc.manchester.ac.uk
alex.henderson@manchester.ac.uk