The advantage of using color space read is that each nucleotide in the read is read twice (except at the start and the end) and this information is coded and shared between two positions located next to each other in the colorspace read (Read this: Transforming and manipulating color space reads). So, the aligner is pretty confident about the calls for all the nucleotides except the ones in the beginning and in the end. So, after alignment when bowtie decodes the colorspace alignment it trims off the first and the last color base from the read because it is not that confident about their decoding. I think it is a good practice when using bowtie to not to use "--col-keepends" as it may reduce false-positives calling of SNPs.