Their LC-MS-MS data was only semi-quantitative as gathered, meaning they could compare relative levels of one protein to another in the same sample, but not necessarily between different samples with a high degree of confidence. So, they spiked known amounts of BSA, an extremely well-characterized protein frequently used in the lab, into their samples and created a standard curve across different concentrations that correlated pretty well with the input amounts. They could then use this curve to better quantify the other proteins in their samples among all the different sample types. This in turn allowed them to say with greater certainty that, for example, Protein X was expressed at three-fold higher levels in Sample A than in Sample B. Otherwise, they would only have been able to say that Sample A expression of Protein X was higher than Sample B.