1. The FXM, as of May 2004
An auspicious
beginning . . .

2. The FXM, as of May 2004
but . . .
Knotty
problems
abound

3. We can (and do) congratulate ourselves for . . .
New cell, new experimental conditions
Parts list and network map
Primary assays: Ca2+, p-Akt, PH-Akt
. . . in cell populations and single cells
RNAi works, at (relatively) high throughput

4. RNAi Knockdowns (KDs) so far
37 KD lines, against 31 targets
KD robust: >99% in one third, >90% in two thirds,
>80% in almost all
Throughput: prepare and assay 4 lines per week
Many KDs produce stable Ca2+ response
phenotypes; of these, many are not expected
*KDs of 28 targets change ~30 % of Ca2+
responses to FXM ligands

5. FXM: Challenges, questions
Weak IgG2a responses
KDs without phenotypes
Need to validate RNAi
phenotypes
Need more/better assays
for network intermediates
Modeling is just beginning

6. Multiple KD phenotypes: delight vs. disaster
Many phenotypes are unexpected, often with gain of
function rather than loss
Are we . . .
Heaven?
Hell?
Uncovering unsuspected
complexity and generating
fascinating puzzles? or
Opening a Pandora’s box
of misleading, biologically
irrelevant phenomena?

7. Validating knockdowns: the questions
Are shRNAi KDs reliable, in general and in
individual cell lines?
Can an shRNAi exert off-target effects?
Are we selecting clones with compensatory
mutations or long-term adaptations?
(Do we want to study such adaptations?)
What are other sources of variability? How
should we deal with them?
What should we do about any/all of these?

8. Validating knockdowns: compensatory mutations/adaptations
To make such compensations less likely,
knock down the target faster . . .
Antisense RNA vs. the same target
Transiently transfect siRNA vs. the same target
and/or
Replicate the phenotype with a KD down-
stream (to rule out compensation at sites
between the first and second targets)

9. Validating knockdowns: coping with variability
Early days! We don’t know yet how much
variation to expect, from any/all sources
Initially, with several ‘unexpected’ phenotypes:
Replicate cell lines with different shRNAi
sequences (some already replicate
the phenotype)
Multiple determinations of responses, to assess
general experimental variability
mRNA arrays, antisense, siRNAi, as above
Devise/apply better statistical criteria for
comparing responses

10. Validating knockdowns: reverse the phenotype
Express the target protein in the shRNAi,
line, using a cDNA it cannot affect (e.g.,
human vs. mouse DNA sequence)
Reversal of the shRNAi phenotype will
indicate that the phenotype was indeed
produced by KD of the target protein*
*But will not rule out compensatory mutations/adaptations

11. Validating knockdowns: test a good hypothesis
(What we always want, of course!)
An shRNAi phenotype is more likely to be due to KD of
the target protein if it is predictably
affected by a second perturbation
E.g., the PTEN KD* appears to increase
the Ca2+ response to C5a
Hypothesis 1: Effect is due to elevated PIP3
PI3K inhibitor should reverse
Hypothesis 2: Elevated PIP3 increases Ca2+
response by targeting PLCg to membrane
PLCg KD should reverse
*Caution:
Reproducibility of PTEN KD phenotype needs to be confirmed

12. Magical inductionism vs. needlepoint nihilism
Get more
data!
Test more
Hypotheses!
From unbiased data
the truth will accrue
Data without ideas
= ignorance
‘Creative tension’

13. Hypothesis center Relieve your creative tension! In the AfCS
Each hypothesis will include . . .
AfCS data
Hypothesis
One experiment that would disprove it

14. Intermediate signals Pressing need to assay many more intermediate
variables
PIP3, IP3, DAG vexingly hard to measure
Phosphorylation disappointing: few, often not robust
Plan/hope: SILAC, AQUApeptide technologies
XFP translocations
Screening under way
Lipids, PIP3
FRET assays

15. Network models From the modelers we ask a lot
Construct a model network that . . .
Represents a comprehensive set of molecular
interactions responsible for key responses
Can vary strengths of interactions & activities,
in silico, to simulate responses
Predicts and evaluates responses in the cell
Easily incorporates (& even suggests) new
hypotheses (feedbacks, connections, nodes)
Evaluates experimental tests of these
new hypotheses

16. Network models The bad news The good news
A difficult task, likely to remain so
Good precedents are rare, but
not unknown
The good news
Will model responses of cell
populations AND of single cells
Abundant data kindles modelers’
enthusiasm

17. It’s a new day!
Overcast . . .
but full of promise

19. IgG2a responses This tyrosine-phosphorylation pathway makes an
immensely attractive target to study, and . . .
Ca2+ & p-Akt responses are quantitatively
similar to C5a responses
But:
IgG2a elicits little detectable tyrosine
phosphorylation (because Syk is poorly
expressed?)
Single cell responses are weak, not yet
reproducible

20. IgG2a responses On the one hand . . .
Signaling mechanisms differ from those of GPCR pathways
Already see potentially interesting (& unexpected)
shRNAi phenotypes
But . . .
How can we begin to understand an IgG2a-
triggered network without measuring
phosphotyrosine responses So
Adapt more sensitive technology (SILAC
or AQUApeptide?)
And ?

21. KDs without phenotypes
E.g. IP3R KDs (so far)
KD ineffective: assess by western, RT-PCR; try
alternative shRNAi sequences
Redundant isoforms: double (& ? triple) KDs with
multiple lentiviruses
Redundant signals: regulation predominantly by a
different pathway (which we must find)

22. Validating knockdowns: off-target effects
Can an shRNAi exert off-target effects?
Probably yes, as already reported with siRNA
But how frequently? In a specific cell line?
To estimate how often this occurs . . .
Immunoblots against unrelated target proteins
mRNA arrays in multiple control vs.
shRNAi-expressing lines

23. Intermediate signals Ligand
We need to measure these to understand information flow
through the network i m n
Ca2+/
PIP3
Ligand h j o k p
What will a KD at i, j, or k do to a signal transmitted at
nodes n, o, or p?