The problem of transmitting k independent Gaussian sources with different variances over k additive white Gaussian channels is studied. Reference system and information theoretic bounds are surveyed, and linear joint source-channel coding (JSCC) schemes are constructed. It is shown that nonlinear schemes generally outperform the best linear schemes. Examples for such systems (involving nonlinear bandwidth expansion and reduction mappings) are given. Universal schemes that do not know the variances of the sources are also considered, and such a scheme is proposed. The performance of the proposed universal scheme is only a constant (about 4dB) away from the best linear scheme, for any selection of the source variances.