The small-scale interactions between turbulence, chemistry, and soot have a profound effect on the soot formation, growth, and destruction processes in turbulent flames. The nucleation of soot occurs from Polycyclic Aromatic Hydrocarbons (PAH), whose chemistry, unlike the major products of combustion, is slower and more sensitive to the turbulent environment. In addition, measurements and two-dimensional Direct Numerical Simulation (DNS) have shown that soot is confined to relatively thin structures which are very intermittent, and accounting for these small-scale structural characteristics will be vital in predicting the mean behavior of the soot field. In this work, subgrid-scale (SGS) models for Large Eddy Simulation (LES) of sooting turbulent flames will be analyzed a priori using two tools: analysis of a DNS database and Lagrangian flamelet analysis. Analysis of the 2D-DNS database shows that including the effect of turbulent fluctuations on the scalars describing the soot population dramatically improves the prediction of the soot intermittency and evaluation of the soot source terms. Lagrangian flamelet analysis shows that PAH concentrations are slow to respond to changes in the scalar dissipation rate, and accounting for this transient response is vital in predicting the formation rate of soot which depends on the square of the PAH concentration.