I want to run an OLS regression on one single negatively worded item. Do I need to reverse code the item especially since I am not taking a total or average of the item with other items?
No. However, do keep in mind the polarity of the variable when it comes time to interpret the resulting regression coefficient (assuming that it is dependably different from zero).
There is no mathematical necessity, but, as the correlation will be reversed in sign, the result may be confusing. Depending on the situation and audience, I would usually recommend reversal to my students to support a coherent presentation.