Cellular decision-making bias: the missing ingredient in cell functional diversity Bradly Alicea http://www.msu.edu/~aliceabr/ http://syntheticdaisies.blogspot.com
Cellular decision-making bias: the missing ingredient in cell functional diversity
Bradly Aliceahttp://www.msu.edu/~aliceabr/
http://syntheticdaisies.blogspot.com
Typical four factors reprogramming (e.g. iPS) is inefficient and highly variable (e.g. stochastic dynamics). Rais et.al
discover a way to make process deterministic.
Rais et.al Deterministic direct reprogramming of somatic cells to pluripotency. Nature (2013)
Mbd3+/- iPS lines (DOX-inducible cassette)
Host Blastocyst(mouse)
Differentiate into MEFs
Reprogrammed to iPS(with latency)
In Rais et.al (2013), “inefficiency” (the presence of un-reprogrammed cells) is characterized as a rate-limiting barrier.
Success!(efficiency)
But what about these?(1-efficiency)
How do you overcome rate-limiting factors?1) Deplete Mbd3 (nucleosome remodeling and deacetylation repressor complex).
2) Promotion of naïve pluripotency conditions.
Reprogramming factors exist in a dynamic equilibrium: * Reactivate endogenous pluripotency networks (positive signal).
* Directly recruits Mbd3/NuRD repressor complex (negative feedback signal for reactivating this network).
Mbd3+/- iPS lines (DOX-inducible cassette)
Host Blastocyst(mouse)
Differentiate into MEFs
Reprogrammed to iPS(with latency)
Reprogramming Latency(per Hanna, 2009 and Rais, 2013)
EarlyReprogrammers
LateReprogrammers
t(μ)
Mbd3f/- is necessary but not sufficientto achieve deterministic reprogramming
time (δ)
EL
ITE
DE
MO
CR
AT
IC
STOCHASTIC
DETERMINISTIC
B Cells, Hanna et.al, 2009Fibroblasts, Alicea et.al, 2013
MUSE Cells, Dezawa et.al, 2013
MEFsRais et.al, 2013
Differences in cellular identity
Differences in pathway regulation
Mbd3 is depleted, reprogramming efficiency promoted (using floxed
and negative allele).
Mbd3 is expressed normally, efficiency is low and/or highly
variable.
“Gas and Brakes” model: Figure 5, frame F
For more information, see: McDonel, P., Costello, I., and Hendrich, B. Keeping things quiet: Roles of NuRD and Sin3 co-repressor complexes during mammalian development. International Journal of Biochemistry and Cell Biology, 41(1), 108-116 (2009).
From a systems perspective
Core Pluripotency
Factors
Mbd3/NuRD repressor complex
( + )( - )
“Gas and Brakes” model: Figure 5, frame F
For more information, see: McDonel, P., Costello, I., and Hendrich, B. Keeping things quiet: Roles of NuRD and Sin3 co-repressor complexes during mammalian development. International Journal of Biochemistry and Cell Biology, 41(1), 108-116 (2009).
Yet epigenetic regulation does not tell the whole story. Are there higher-level
organizational factors at play?
Buganim et.al, Cell, 150(6), 1209-1222 (2012).
Difference between early and late reprogramming:
* early phase = core genes in pluripotency network exhibit mass upregulation (genes act independently).
* late phase = core genes in pluripotency network exhibit hierarchical dependence (above).
Rais et.al assumption: all cells reprogram to iPS, and occurs
with uniform latency (no intrinsic differences in cell
population).
Violation of assumption: what happens when cells exhibit
variation? Or when one subpopulation is favored?
Question to keep in mind:
Is there a necessary relationship between the presence of a favored subpopulation and reprogramming being a uniformly-distributed event?
iSM
The creation of “deterministic reprogrammers” relies upon minimizing the variability in regulatory mechanisms (e.g. industrial process).
* This is not normally found in nature, but systematic variation may exist between conversion regimens (e.g. iN, iSM).
* I/O problem: transcription factor induction (input) and destination phenotype (output).
* are all forms of conversion equal, or are certain types of conversion (iPS, iN, iSM, iCM) easier to achieve?
Reprogramming bias: tendency for some cell lines to favor a certain destination phenotype upon reprogramming.
Reprogramming BiasPhenotypic (H1):
* induced phenotype A vs. induced phenotype B (e.g. iNC, iSMC).
Genomic (H2 and H3):
* pre-existing bias, gene expression in different cell types before the transformative process.
* induced bias, gene expression after a transformative process has occurred.
Extrinsic (H4):
* tied to survivability of cells, does signal spectrum of a phenotype overlap with that of cells put under defined (survival) conditions?
Building a signal spectrum (histogram):
* requires experimental replicates.
* rank-order frequency method.
Sparse histogram:
* provides a multimodal distribution for further analysis.
Classical SDT
Signal and Noise are distinct
Signal and Noise overlap
Overlap = d’
Signals are distinct
Signals overlap
Cellular SDT
Overlap = O(n,m)
O(N,M) = Σ MAX(Ni,Mi) - ||Ni – Mi||
OVERLAP(N and M)
MAXIMUM (ith elementN, ith element M)
Reprogramming Bias
Taken from a rank-order frequency spectrum for same cell lines.
FR
EQ
UE
NC
Y
RANK ORDER (CELL LINES IN ANALYSIS)
KIDNEY HEART
OVERLAP(N and M)
O(N,M) = Σ MAX(Ni,Mi) - ||Ni – Mi||
Reprogramming Bias
Cell lines from some tissues (kidney, skeletal muscle) show bias for one type of conversion
over another.
O(N,M) = Σ MAX(Ni,Mi) - ||Ni – Mi||
Reprogramming Bias
Cell lines from some tissues (kidney, skeletal muscle) show bias for one type of conversion
over another.
PROCESS DIAGRAM
Pre-existing BiasFibroblasts from 13 mouse fibroblasts cell lines known to exhibit differential reprogramming between muscle and neuron.
* high-throughput case (two breast and one lung line) exhibit no distinct pattern of bias, interesting (single probe) local differences.
Distributions are uniform with no tails, smear into one another (e.g. no bias).
Induced BiasHuman Fibroblasts under various drug
treatments
Translatome (Blue), Transcriptome (Red) A = COL1A, B = Fibronectin, C = UTF
All three genes: significant overlap for both fractions of RNA:
* differences between genes: high-rank skew for COL1A, low-rank skew for UTF.
* COL1A, UTF: intermittent expression?
High-throughput case (fibroblasts under Vitamin C treatment):
* differences are inconclusive.
O(S,M) = Σ MAX(Si,Mi) - ||Si – Mi||
OVERLAP(S and M,S and N)
MAXIMUM (ith elementS, ith element N or M)
Survivability
Taken from a rank-order frequency spectrum for same cell lines under
survival conditions.
O(S,M) = Σ MAX(Si,Mi) - ||Si – Mi||
OVERLAP(S and M,S and N)
MAXIMUM (ith elementS, ith element N or M)
Survivability
Taken from a rank-order frequency spectrum for same cell lines under
survival conditions.
FR
EQ
UE
NC
Y
RANK ORDER (CELL LINES IN ANALYSIS)
KIDNEY HEARTOVERLAP(S and M)
2-dimensional Genotype Space
Naïve ground state
iPS
iSMiN
BIAS
BIAS
Schematic of a Random Walk, step size based on non-uniform distribution (semi-Levy Flight).
Stochasticity w.r.t. time
(δ)
12d
Reprogramming Model of Rais et.al, 2013 (inducible factors)
4d
12d
Theoretical MaximumEfficiency (e.g. 40%)
Kurtosis = efficiency of process (rate-limiting factors).
Skew = variability inprocess.
time (δ)
12d
Reprogramming Model of Rais et.al, 2013 (inducible factors)
4d
δ
12d
Model used here assumes that reprogramming events over time can be drawn from a Gaussian (e.g. uniform) probability distribution.
For each day, a certain proportion of cells convert. Above, 12d sees the maximum number of conversions.
Theoretical MaximumEfficiency (e.g. 40%)
Kurtosis = efficiency of process (rate-limiting factors).
Skew = stochasticity inprocess.
4d
12dIs reprogramming according to a uniform distribution a reasonable assumption?
* model matches observations of reprogramming using inducible factors, but perhaps this has little relevance to the biology of process.
time (δ)
Con
vers
ion
Rat
e
Infectability Data (inducible YFP signal)
Mouse Cell Lines
4d
12d
4d
12dIs reprogramming according to a uniform distribution a reasonable assumption?
* model matches observations of reprogramming using inducible factors, but perhaps this has little relevance to the biology of process.
time (δ)
Con
vers
ion
Rat
e
Infectability Data (inducible YFP signal)
Mouse Cell Lines
4d
12d
4d
12dIs reprogramming according to a uniform distribution a reasonable assumption?
* model matches observations of reprogramming using inducible factors, but perhaps this has little relevance to the biology of process.
Converting to iN and iSM phenotypes results in variable distributions.
This suggests the reprogramming process should be modeled using a exponential rather than a Gaussian.
time (δ)