I assume "second-order polynomial regression" means that you want a quadratic fit.  If so, you can fit the model with the REGRESSION procedure.  Let X = dietary lipid level, and Y somatic weight gain.  Compute a new variable equal to X-squared.  Fit a model with both X and X^2 as predictors of Y.  Save the fitted (or predicted) values, and make a scatter-plot with X = lipid level and Y = fitted value.  From what you've said, I think you're expecting an inverted U in that plot. 

HTH.

Dear Professor Weaver, Thanks for your instructions.

I think using SEM-PLS (Structurel Equation Modeling-Partial Least Square) is better to get  more accurate results in regards of second-order polynomial regression.

All the best