The answer above is good but not complete. In many animals there is indeed a strong mutation bias from CpG to TpG. It should be added that in the protein coding regions you have strong natural selection, so when you need a CpG to code an aminoacid sequence it has to be there and any mutations that would normally lead to C becoming T (for example) are wiped out by natural selection. So we don't see these mutations and therfor we observe higher GC content. There is also the simpler explanation that protein coding sequences in general may require more GCs for correct encoding, no matter what mutation biases we have. 

Thanks Alexander. This is something I am actually observing in my data. Also a positive bias on the usage of GC encoded amino acids was highlighted. Thanks for your reply!