The characterization of the optimal joint source-channel coding scheme for transmission of correlated sources over multiple-access-relay channels (MARCs) is an open problem. Here, this problem is studied in the presence of arbitrarily correlated side information at both the relay and the destination. Since each transmitter observes only one of the sources, the admissible joint distributions of the sources and channel inputs must satisfy a Markov relationship which constrains their statistical dependence. This observation is used together with the new data processing inequality derived by [Kang and Ulukus, 2011] to obtain two new sets of single-letter necessary conditions. These new conditions are shown to be at least as tight as the previously known ones, and strictly tighter than the cut-set bound.