Coordination of meristem and boundary functions by transcription factors in the SHOOT MERISTEMLESS regulatory network

ABSTRACT The Arabidopsis homeodomain transcription factor SHOOT MERISTEMLESS (STM) is crucial for shoot apical meristem (SAM) function, yet the components and structure of the STM gene regulatory network (GRN) are largely unknown. Here, we show that transcriptional regulators are overrepresented among STM-regulated genes and, using these as GRN components in Bayesian network analysis, we infer STM GRN associations and reveal regulatory relationships between STM and factors involved in multiple aspects of SAM function. These include hormone regulation, TCP-mediated control of cell differentiation, AIL/PLT-mediated regulation of pluripotency and phyllotaxis, and specification of meristem-organ boundary zones via CUC1. We demonstrate a direct positive transcriptional feedback loop between STM and CUC1, despite their distinct expression patterns in the meristem and organ boundary, respectively. Our further finding that STM activates expression of the CUC1-targeting microRNA miR164c combined with mathematical modelling provides a potential solution for this apparent contradiction, demonstrating that these proposed regulatory interactions coupled with STM mobility could be sufficient to provide a mechanism for CUC1 localisation at the meristem-organ boundary. Our findings highlight the central role for the STM GRN in coordinating SAM functions.


Model description
The model incorporates the key interactions between proteins and mRNA, as described in the main text and shown in Figure 5 of the main text. In all cells, the transcription of STM mRNA is promoted by CUC1, whereas transcription of CUC1 mRNA is promoted by STM, creating a positive feedback loop. STM promotes the production of the microRNA miRNA164c, which inhibits CUC1 protein production by increasing the degradation rate of CUC1 mRNA. To maintain STM levels in the centre of the meristem, where CUC1 protein levels are thought to be low, we include include STM upregulation of its own transcription and/or basal transcription of STM in the model as optional effects. Both STM and CUC1 mRNA are translated into protein at a constant rate, and the proteins have constant basal degradation rates. However miRNA164c greatly increases the degradation rate of CUC1 mRNA, reducing CUC1 mRNA levels in the centre of the primordium.
Explaining how primordium cells are specified is beyond the scope of the current model; this specification is thought to involve auxin transport and regulation of the PIN auxin efflux carriers and has previously been modelled [3]. As a proxy for this, we introduce a Primordium Identity Factor (PRIF). We prescribe [PRIF] i = 1 for cells within the primordium, and [PRIF] i = 0 for the remaining cells; this prescribed distribution specifies the primordium's position. Production of TCP is promoted by PRIF. For simplicity we combine the transcription and translation steps of TCP production. TCP further promotes the production of microRNA miRNA164a. As there is no basal production both TCP and miRNA164a are restricted to the primordia. As for miRNA164c, miRNA164a acts through increasing the degradation rate of CUC1 mRNA. TCP is negatively regulated by STM and we suppose that TCP decays at a constant rate.
As the details of transport of STM between neighbouring cells in the meristem are not fully understood (see [5]), we choose a simple model and suppose that the net flux of STM between neighbouring cells is proportional to the concentration difference between them.

Model equations
We can represent the interactions described above using the a system of 7N ordinary differential equations which indicate how the concentrations of each species within each cell (as listed in Table 1) change with time: where i = 1, · · · , N. The model depends on 28 parameters, as listed in Table 2.
We specify the three cells at the left-hand side of the file to be primordium cells, setting levels of the primordium interaction factor to be

Parameter values and initial conditions
The model predictions depend on a number of parameters (summarised in Table 2). As we are primarily interested in the steady-state behaviour of this system, and in particular the relative spatial expression patterns, rather than in the absolute concentrations of the network components, all parameter values are dimensionless. We investigated how the choice of parameters affects the predicted distributions of the network components; we found that the proposed network of interactions could qualitatively mimic the biologically observed distributions using the parameter values listed in Table 2. These parameters were selected from a heuristic search of parameter space, selecting parameter values which generated large differences in CUC1 expression levels between the boundary region and the centre of the primordium. As initial conditions, the STM protein and mRNA concentrations are set equal to 1 for all cells, and the concentrations of all other species set to zero:

Numerical solution
Steady-state solutions were obtained by the numerical integration of equations (1), with initial conditions (2) using standard ODE solvers ( [2]), until T = 10 8 . Python code to reproduce the simulation outputs shown in the paper is attached as a Supplementary File.

Parameter sensitivity analysis
To assess the sensitivity of simulation outputs to parameter choices, we performed a basic local parametric sensitivity analysis [1]. For each parameter p, we calculated the mean normalized sensitivityS p from  For simulations with STM autoregulation and no basal STM transcription, as shown in Figure 1, simulation results were particularly sensitive to δ STM , α STM , λ STM , η STM , k STM and β 1 ; these parameters control the overall level of STM protein. Simulations are also sensitive to k mc and n c , which regulate the sharp response of miR164c to STM. Without STM autoregulation, but with basal STM transcription, the results of the sensitivity analysis are shown in Figure 2. Simulation outputs are now sensitive to the basal STM transcription rate C STM , but here β 1 = 0.    Table S1. Genes identified using GO analysis.