This paper proposes a multiple signal classification (MUSIC) framework for direction-of-arrival estimation by combining multiple circular arrays in the circular-harmonics domain. We jointly transform the received signals of all sub-arrays into the circular harmonics domain to generate a set of sound field coefficients containing the direction information of the sound sources. These coefficients are then formulated in a form in which MUSIC algorithm can be applied. Compared with the conventional circular-harmonics-domain localization methods, which are based on a single circular array, the proposed method can provide sufficient spatial resolution over different frequency ranges by adjusting the distribution of sub-arrays. Furthermore, the mean square error of the estimated sound field coefficients is derived for guiding this adjustment. Numerical simulation results indicate that a reasonable distribution of sub-arrays can effectively avoid the performance degradation caused by the zeros of Bessel functions, which is an inherent problem of the modal array signal processing. Simulation and experimental results with different configuration parameters demonstrate that the proposed method provides a better localization performance compared to the state-of-the-art methods.