The sign in principal component analysis is arbitrary, since PCA is a simple mathematical transform of your data. When you flip the direction along the axis, the variance doesn't change, and neither does the interpretation.
You can find some great answers at CrossValidated that explain this, e.g: