Presented at SIGBIO38, Okinawa, Japan, 26 June 2014 Full paper http://www.ncbi.nlm.nih.gov/pubmed/25521548

1. TINAGL1 and B3GALNT1 are potential therapy target
genes to suppress metastasis in non-small cell lung cancer
Hideaki Umeyama,
Department of Biological Science, Chuo University
MItsuo Iwadate,
Department of Biological Science, Chuo University
＊Y-h. Taguchi,
Department of Physics, Chuo University

2. Introduction
Purpose of the present research:Purpose of the present research:
Applying recently proposed “PCA based
unsupervised FE” to drug discovery, by combining
with FAMS/choseLD pipeline.
Target in this presentation:Target in this presentation:
Metastasis of non-small cell lung cancer (NSCLC)
MaterialsMaterials::(GSE52144)
Two NSCLC cell lines (HTB56/A549), each
with/without metastasis (in total, four types
(classes)).

3. MeasurementsMeasurements:(GSE52144)
mRNA expression + promoter methylation
(microarray) (microarray+NGS)
Plan:Plan:
Target gene identification using integrated analysis of
mRNA/expression and promoter methylation, i.e.,
11. aberrant mRNA expression associated with metastasis
22. aberrant promoter methylation associated with
metastasis
33. Negative correlation between mRNA expression and
promoter methylation
→→ 1&2&3 : target genes1&2&3 : target genes

4. ProblemsProblems::
11. Difficulty of detection of negative correlation.
Simple application Pearson correlation analysis
+ adjusted P-values considering multiple comparison
→ P<0.05 : only one gene
22. Difficulty of detection of aberrant mRNA expression /
promoter methylation
Simple t test: with vs without metastasis
+ adjusted P-values considering multiple comparison
mRNA:(P<0.05)
A549 No genes / HTB56 434 genes
Promoter methylation:(P<0.05)
A549 No genes / HTB56 No genes

5. Using PCA based unsupervised FE
Embeddings of Probes to 2D surface
PC1
PC2 mRNA methylation

6. PC1:no sample dependence
A549 without metastasis
A549 with metastasis
HTB56 without metastasis
HTB56 with metastasis
mRNA 97% Methylation 87%

7. PC2: distinction between two cell lines
mRNA 1.7% Methylation 9.6%

8. mRNA 0.2% Methylation 1.5%
PC3: distinction between with and without
metastasis for HTB56 only
Top 100 outliers Top 100 outliers
Common seven genes (P=1x10-4
)

9. mRNA 0.6% Methylation 0.09%
PC5 VS PC4: distinction between with and
without metastasis for A549 only
Common four genes (P=1x10-8
)

10. mRNA
methylation PC1 PC2
PC3 PC4 vs PC5

11. PCA based unsupervised FE could
successfully selected genes ….
1 aberrant mRNA expression
2 aberrant promoter methylation
3 negative correlation between mRNA
expression and promoter methylation
11 genes:
HOXB2, CCDC8, ZNF114, DIO2, LAPTM5,
RGS1, B3GALNT1, TINAGL1, PMEPA1,
CX3CL1, ICAM1

12. 11 genes:
HOXB2, CCDC8, ZNF114, DIO2, LAPTM5,
RGS1, B3GALNT1, TINAGL1, PMEPA1,
CX3CL1, ICAM1
Gendoo Server:Gendoo Server:
Many cancer related diseases association
Target of TF AHRTarget of TF AHR:
AHR: reported to be cancer causing TF,
including lung cancer
As a result, it was successful also from
biological point of views

13. Since we could seem to identify metastasis causing
genes in NSCLC, next we try to investigate if
these are therapy (drug) targets.
We downloaded corresponding amino acid
sequence from UniProt, and unloaded to profile
base protein structure inference server, FAMS.
FAMS: “Full Automatic Modeling System”
developed by Profs. Umeyama and Iwadate.
Predicted structures were also compared with the
predictions by public domain prediction server
Phyre2 and was confirmed that there were no
significant differences.

14. ExampleExample:
HOXB2: Structure predicted partially by FAMS
HOXB2 has Homeobox structure (see below) that
binds to DNA/RNA region.
If we can find some
compounds that
“blcok” DNA/RNA
binding of HOXB2,
the compound can
be a drug.

15. We have investigated eleven genes from
this point views, and proposed two possible
drug target genes.
TINAGL1 B3GALNT1

16. TINAGL1 is usually classified as tumor
supressor, but was also reported to be
upregulated in metastasis.
Amino Acid Sequence
PDB
Tertiary Structure Template Ligands
ChEMBL Homologous Protein
Ligand (Drug) Candidate
BLAST
BLAST
FAMS
ChooseLD*
* profile based ligand docking prediction software

17. Drug candidate Compounds identified from ChEMBL
Binding modes of two compounds inferred by chooseLD

18. Independent Binding affinity evaluation by Cyscore
→ Not very good, only acceptable
Known ligand (PDB)
Drug candidate (ChEMBL + chooseLD)

19. ConclusionConclusion
Using integrated analysis of mRNA
expression and promoter methylation with
PCA based unsupervised FE, we
successfully identified eleven possible
metastasis causing genes.
Using FAMS/chooseLD, we identified
possible drug candidate compounds for two
of eleven genes.

20. This research was supported by bothThis research was supported by both
Chuo University Joint Research Project Grant
“Drug discovery based on protein functional inference by
FAMS”
and
Grant-in-Aid for Scientific Research on Innovative Area
“Initiative for high-dimensional data-driven science based
on sparse modeling”
It is also accepted as Oral presentation at InCob2014 (not
yet paper acceptance)