That depends on your experimental design. If you have an unpaired design (intervention on one group and not on another) use ttest2. If you have a paired design (before an after an intervention on the same group), use ttest.
Of course this requires that your data are normally distributed and meet the other assumptions of the t-test for both groups. If they do not, or you do not know the distribution, use the ranksum and signrank tests, respectively for unpaired or paired data.