clustering from structural variation in endometrial...
TRANSCRIPT
Clustering from structural variation in endometrial cancer Julian Stanley, Shreepriya Das, Michele Cummings, Nicolas Orsi, & Jeremy Gunawardena
Department of Systems Biology, Harvard Medical School, Boston, MA 02115
Picture courtesy of AffymetrixPicture courtesy of Leica Biosystems
Endometrial cancers are the fourth most common cancer in womenworldwide, and identification of molecular characteristics that define subtypesof the cancer may complement traditional diagnostic techniques. We obtained95 hysterectomies of different grades of endometrial cancer and analyzedstructural variation events from formalin-fixed samples. Hierarchical clusteringof copy number variation (CNV) revealed two clusters of endometrial cancerthat also correspond to survival. Clustering can also be accomplished by asimple technique of plotting patients in terms of aggregate copy number gainand loss. Further work needs to be done to identify genetic markers that mayaid diagnosis, and how those markers compare across cancer types.
Grade 1 or 2 Endometroid
Grade 3 Endometroid
Non-Endometroid
Better Prognosis
Worse Prognosis
A simplified view of endometrial cancer subtypes
Grade 1 or 2 Endometroid
Grade 3 Endometroid
Non-Endometroid
Better Prognosis
Worse Prognosis
A simplified view of endometrial cancer subtypes
Grade 1 or 2 Endometroid“Type-1”
Grade 3 Endometroid“Type-1” or “Type 2”
Non-Endometroid“Type 2”
Better Prognosis
Worse Prognosis
A simplified view of endometrial cancer subtypes
Picture courtesy of Cancer Research UK
Figure courtesy of Watkins et al. 2014 in Breast Cancer Research. Green bar presents copy number gain, red bar represents copy number loss. Figures represent: (left) copy number gain leading to allelic imbalance, (center) copy number loss leading to LOH, (right) recombination leading to copy neutral LOH.
Copy number variation (CNV), allelic imbalance, and loss of heterozygosity (LOH) are types of structural variation
Structural variation analysis
CN
Lo
ss
Allelic
Imb
alance
LOH
CN
Gain
Grade 1 Endometroid (11)
Grade 2 Endometroid (21)
Non-Endometroid (31)
Grade 3 Endometroid (32)
Figure 1. Aggregate (a) copy number gain and loss and (b) allelic imbalance and loss of heterozygosity in 95 endometrial cancer FFPE samples
a b
Cluster 1 (68 Patients) 2 (27 Patients)1 (68 Patients) 2 (27 Patients)
Cluster 1 (Low Copy Number)
Cluster 2 (High Copy Number) Log-rank P = 0.0025
Figure 2.
Method: Unsupervised hierarchical clustering based on Euclidean distance, Ward linkage, and silhouette optimization.
a: Copy number variation separates patients into two clusters. Green represents CN loss and red represents CN gain.
b: Clusters have a statistically and clinically significant difference in survival.
Figure 3. Patient clusters in copy number space, labeled by clinical diagnosis.
K-means clustering on a simple metric shows clusters that have significant difference in survival (not shown).
CNV-based clusters
Does the method also apply to ovarian cancer?
No
rmal
ized
co
py
nu
mb
er g
ain
Frac
tio
n s
urv
ived
Time since diagnosis (days)
Normalized copy number loss
Clear Cell: 19
Low Grade EDM: 14
Mucinous: 16
High Grade Serous: 19
Low Grade Serous: 18
Clear Cell: 19
Low Grade Endometroid: 14
Mucinous: 16
High Grade Serous: 19
Low Grade Serous: 18
Figure 4. Ovarian cancer patient clusters in copy number space.
❖ Low grade endometroid are least severe Low copy number❖Mucinous is intermediate, but based on CN we predict less severe❖ Low grade serous and clear cell are intermediate; they distribute normally❖ High grade serous are most serious High copy number
Figure 5. Copy number calls were made and analyzed by Nexus ExpressSoftware. Breast and renal cell cancers were used as comparisons withrespectively similar and different risk factors as endometrial and ovariancancer. a: CN gain (blue) and CN loss (red) and b: allelic imbalance (purple)and LOH (brown) shown as a proportion of samples with the given variation.
DiscussionBoth endometrial and ovarian cancer are extremely heterogenous diseasesthat affect hundreds of thousands of women worldwide. Microarray analysisis a powerful tool that may provide information that can lead to morepersonalized cancer treatments.
The Oncoscan platform, based on the molecular inversion probe technique,can consistently detect variation heavily-degraded FFPE clinical samples.
Clustering techniques and comparative analyses are necessary to understandwhat differences between cancers are visible from genomic analyses, andhow different molecular profiles can complement traditional diagnostics.
Acknowledgements
References
I am extremely grateful to HMS Systems Biology and the Summer Internshipin Systems Biology program for the welcoming working environment. Thanksto Shreepriya Das and Jeremy Gunawardena for their invaluable mentorshipduring this short internship, and to Nicolas Orsi and Michele Cummings forproviding samples and helpful clinical expertise.
1. Levine, DA et al. 2013 Integrated genomic characterization of endometrial carcinoma. Nature 497, 67-73.2. Zhang, J. 2018 CNTools: Convert segment data into a region by sample matrix to allow for other high levelcomputational analyses. R package version 1.36.0.3. Kaveh, F. et al. 2016 A systematic comparison of copy number alterations in four types of female cancer. BMC Cancer16:913.4. Jung, H. et al. 2017 Utilization of the Oncoscan microarray assay in cancer diagnostics. Applied Cancer Research 37:1.
Contact: [email protected]
No
rmal
ized
co
py
nu
mb
er g
ain
Normalized copy number loss
Moving forward: comparative analyses
Better Prognosis
Worse Prognosis
Grade 3 Endometroid
Grade 1 Endometroid
Grade 2 Endometroid
Non-Endometroid
Less severe
More severe
Intermediate
DNA extracted from FFPE tissue was analyzed using the Oncoscan Copy Number assay
% of samples with CN gain
% of samples with CN loss
% of samples with allelic imbalance
% of samples with LOH
Breast Cancer (11)
Endometrial Cancer (95)
Ovarian Cancer (89)
Renal Cell Cancer (15)
Breast Cancer (11)
Endometrial Cancer (95)
Ovarian Cancer (89)
Renal Cell Cancer (15)
Chromosome: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 18 20 2217 19 21 X
Agg
rega
te %
of
sam
ple
sA
ggre
gate
% o
f sa
mp
les
Chromosome: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 18 20 2217 19 21 X