This document discusses measuring off-target effects of oligonucleotide drug candidates. It finds that on average, antisense oligonucleotides (ASOs) impact the transcriptome to a similar extent as small molecule drugs by changing expression of a comparable number of genes. The best way to predict potential off-target effects is through a combination of in silico sequence analysis and in vitro validation of predicted off-targets. While current models only explain a small portion of measured off-targets, experimental transcriptomics allows refinement of prediction algorithms. Careful preclinical studies in relevant animal models and controlled human trials are ultimately needed to fully understand safety.
Human Factors of XR: Using Human Factors to Design XR Systems
Measurement and Prediction of Hybridization Effects of Oligo Candidates
1. Measurement and Prediction of
Hybridization-induced Off-target Effects
of Oligonucleotide Drug Candidates
morten lindow, ph.d,
associate director, informatics
santaris pharma A/S
adjunct associate professor, bioinformatics
university of copenhagen
morten.lindow@gmail.com
2. The views and opinions expressed in the following PowerPoint slides are
those of the individual presenter and should not be attributed to Drug
Information Association, Inc. (“DIA”), its directors, officers, employees,
volunteers, members, chapters, councils, Special Interest Area
Communities or affiliates, or any organization with which the presenter is
employed or affiliated.
These PowerPoint slides are the intellectual property of the individual
presenter and are protected under the copyright laws of the United States of
America and other countries. Used by permission. All rights reserved. Drug
Information Association, DIA and DIA logo are registered trademarks or
trademarks of Drug Information Association Inc. All other trademarks are
the property of their respective owners.
2www.diahome.orgDIA
3. DIA www.diahome.org 4
Does antisense oligonucleotides
perturb the transcriptome more or
less than small molecule drugs?
4. Measuring drug induced changes to the human
transcriptome
DIA www.diahome.org 5
Connectivity map: Small
molecules
Antisense oligonucleotides
Database of 1309 small
molecules applied
systematically in 6100 cell
culture experiment
Mining of Gene Expression
Omnibus and Santaris internal
data
Stratifiable by drug type 24 different oligos (both
antimiRs and gapmers)
Cells subjected to
pharmacological dose
Cells subjected to
pharmacological dose (intended
target is knocked down)
Affymetrix microarrays Affymetrix microarrays
Science. 2006 Sep 29;313(5795):1929-35
5. Compare transcriptome changes induced by ASOs to
those induced by approved drugs
DIA www.diahome.org 6
Comparing across multiple expression
experiments is not straightforward
Took the path of minimal data
transformation:
• All compounds compared directly to
their designated vehicle control
• Compare number of genes that
change expression by more than 50%
(up or down)
• Tried a range of other
thresholds, conclusion is the same
Hagedorn et al., in preparation
Transcriptschangingmorethan50%
2
5
10
20
50
100
200
500
1000
2000
5000
ASO
SMC(rest)
SMC(L+P)
ns ***
Compounds
L+P=
anticancer and
antiparasite drugs
6. Drug induced changes to transcript levels
DIA www.diahome.org 9
drug
interaction with target
(protein or RNA)
interaction with
non-target proteins
Change of cell
state
expression
changes
Change of cell
state
expression
changes
Disease
improvement
Unwanted
pharmacology
Change of cell
state
expression
changes
interaction with
non-target RNA
Possible adverse
effects and toxicity
non-target RNA
relatively predictable
hybridization-dependentUnpredictable
hybridization-independent
7. Paper from OSWG subcommitee on off-targets
Candidate
oligonucleotide drug
In silico off-target
screen
Database of
transcripts
off-target present in
tox species?
Penultimate test: Preclinical toxicity studies in vivo
In vitro validation of
critical putative off-
targets.
Relative IC50 in
human cell line
Proceed to human
testing
Case by case evaluation of putative off-targets may include:
Technology
and
mechanism
based
algorithms
Comparison of tissue
expression of off-
target with tissue
accumulation of drug
candidate
Function of putative
off-target if known,
e.g. phenotype of
genetic knock-out
Flow chart from Lindow et al 2012: OSWG off-target committee recommendations
10. For each possible oligonucleotide against
the intended target (~ 20 000 * modification variants)
Evaluate activity determinants against all
possible target sites in the transcriptome
(1.4E9 sites)
Ideal exhaustive in silico specificity evaluation
DIA www.diahome.org 13
NOT FEASIBLE!
11. • Sequence search to choose oligo-
sequences with minimal number of close
sequence matches to non-target RNAs
What is feasible?
DIA www.diahome.org 14
Late discovery phase:
a few candidates
transcriptome
sequences
~1E9 nt>5 yrs ago: search with
BLAST or FASTA
Design phase:
~tens of thousands
of possible oligo
sequences
faster computers, more RAM,
suffix arrays, BW-transforms,
hashing
12. in silico paradigms employed in practice
• Complete-with-mismatches
• Alignment score cutoff: plus for a match,
minus for a mismatch/indel
• Hybridization energy cutoff
character based
energy based
13. Number of off-targets
decrease with length
Number of off-targets
increase with length
Number of off-targets
increase with length
Complete with mismatches
Alignment score cut-off
Hybridization energy cut-off
14. Aim of sequence search and selection
DIA www.diahome.org 18
affinity
- G
potency of
(off-)target
down-regulation perfect full target site
closest imperfect sites
in non-targets
G
Oligonucleotide
with too high-
affinity!
more matches -> higher affinity
mismatches, indels -> lower affinity
modifications affect affinity
neighbouring bases affect affinity (stacking)
Prediction of affinity is
possible with nearest
neighbour models
17. Oligo1 against ApoB Oligo2 against Apob
Disentangle downstream pharmacological effects and class effects
from sequence specific off-target effects
Manuscript in preparation
18. • Only small overlap between current in silico
predictions and measured off-targets
• Global transcriptomics measurements allows
data driven refinement of algorithms
– we use regression methods to combine
determinants
• our current best model includes two determinants
– predicted binding affinity between oligonucleotide and
(off-)target site
– predicted RNA structural accessibility of (off-)target site
Lessons from transcriptomics measurement of
specificity
Morten Lindow www.diahome.org 22
19. Summary
DIA www.diahome.org 23
Transcriptschangingmorethan50%
2
5
10
20
50
100
200
500
1000
2000
5000 ASO
SMC(rest)
SMC(L+P)
ns ***
Compounds
ASOs on par with small molecules:
• On average same size of impact on
transcriptome
• Penultimate test for toxicology is in relevant
animals models
• Understanding that the only way to truly test
for human responses is in carefully controlled
and monitored clinical trials
Sequence analysis for specificity allows:
• Risk minimization
• Guide exploratory toxicology
Experimental design to measure off-target
pertubation
20. • OSWG off-target committee
• Peter Hagedorn, research bioinformatician
• Danish Strategic Research Council
Acknowledgements
DIA www.diahome.org 24
Editor's Notes
First I will try to put the issue into perspective by asking:
In an attempt to look at this we mined the connectivity map, a database of more than thousand small molecules applied to cell cultures and subsequent transcriptome measurements. We compared this to in-house and public dataset of how oligonucleotides perturb transcriptomes of some of the same cell lines.
Such an analysis is not straight forward and has many caveat: different cell lines, different doses, different number of replicates, different times of treatment. It is therefore bound to be pretty rough and we could talk and discuss the methodollogy for hours. For now I just want to show the simplest analysis we could think about. Count the number of genes that change more than 50% up or down as a result of he treatment. If we look we notice that theres is quite a large spread in the number of transcripts that change. Also we found that oncology and antiparasitic drugs stuck out and changed more than the rest. Oligodrugs are not significantly diiferent from this rest-group of small molecules.
Bennett and collegues nicely summarized the obseved types of toxicities and whether they had been observed preclinically or clinically. What I will be focusing on is effects due to watson/crick hybridization to unintended target. When there is such unwanted activity is of significance it can be expected to change the expression level of the off-target RNA. Phrased in general terms it would change the transcriptome of the target cells
Let us review the various ways that drugs in general could lead to changes in transcript levels. It would (hopefully) interact with the intended target, which would lead to some desiredable change in cell stage and accompanying expression changes. This would lead to downstream effects of either disease improvement or unwanted pharmacology. The drug might also interact with non-target proteins that also leads to change of cell state. Interactions with non-target proteins are generally unpredictable. oOligonucleotides can also interact with non/target RNAs. The interactions, however, ought to relatively predicatable, because they are partially governed by basepairing interactions
In 2012 the OSWG published a position paper summarizing the consensus thinking about how these potential unintended effects should be assessed. That paper is summarized by this flow diagram
In the following I am going to elaborate on especially the in silico part of this flow chart. Focusing on a technology that we like to employ in Santaris, namely that of RNAseH recruiting single stranded oligonucleotides