Science - USA (2022-01-21)

Zhuet al.,Science 375 , eabg9765 (2022) 21 January 2022 2 of 11

A Ideal capabilities for a synthetic multistable circuit

B Natural multistable circuits use dimerization and autoregulation

C MultiFate-2 circuit

Endodermal genes

(including Sox17)

Pluripotent genes

(including Sox2)

Sox 2 Oct4 Sox17

other TFs other TFs other TFs

Id E47 MRF

Myogenesis genes

(including MRFs)

High protein stability Low protein stability

High protein stability Intermediate protein stability Low protein stability

TF A

A

AA

AB AC BC

AA BBB

TF B

B

BB

CCC

TF C

C

CCC

inactive complexes

Type II tristability Bistability

Type II septastability Hexastability *Type I quadrastability

(Tristability experimentally)

Transcription

Factors (TFs)

Cellular

States

TF A

TF B

Switch states

TF A

TF B

Stabilize/destabilize states

A

TF A

TF C

TF B

Expand states

TF = transcription factor

D MultiFate-3 circuit

Non-dimensionalized parameters

maximal activated protein

production rate

basal protein production rate

Nullclines

Separatrix

Attractor basins

Phase portrait legends

TF A

A

AA

AB

AA B

TF B

B

BB

inactive

complex

B

Upstream signals Upstream signals

Myogenesis

Fig. 1. The naturally inspired MultiFate architecture generates diverse types of
multistability in the model.(A) A hypothetical synthetic multistable circuit is
represented by colored cell cartoons (top) and attractors in a transcription factor
phase space (bottom; axes represent transcription factor concentrations of TF A, TF
B, and TF C). An ideal synthetic multistable circuit should generate multiple stable
states, support control of state switching (left) and state stability (center), and
allow easy expansion of states by addition of more transcription factors (right).
(B) Competitive protein-protein interactions and autoregulatory feedback are
prevalent in natural multistable circuits that control myogenesis (left) and
endodermal differentiation (right), as shown by these simplified and abridged
diagrams. Blue arrows indicate competitive protein-protein interactions, which can
involve higher-order multimerization. Orange dashed arrows indicate direct or

indirect positive transcriptional feedback. (CandD) Models of the MultiFate-2 circuit

and MultiFate-3 circuit (Box 1) ( 25 ) generate diverse types of multistability in

different parameter regimes (indicated above plots). In the model of the MultiFate-3

circuit, low protein stability generates four stable states (type I quadrastability), but

the state in which all transcription factors are minimally expressed is unstable in

the presence of biological noise (fig. S22), consistent with experimental results in

Fig. 5B, low-TMP columns. See figs. S1 and S2 for complete lists of multistability

regimes. All models used here are symmetric and nondimensionalized, with rescaled

dimerization dissociation constantKd= 1 and Hill coefficientn= 1.5 (Box 1). In both

(C) and (D), each axis represents the dimensionless total concentration of each

transcription factor. Note that in the nondimensionalized model, changing protein

stability is equivalent to multiplyingaandbwith the same factor (Box 1).

RESEARCH | RESEARCH ARTICLE