After struggling a while on my +-25V current sensing circuit which uses an LT1990 for such purpose, I realize that the NMOS proposed on the datasheet for boosting the closed loop gain from x1 to x10 were affecting the behaviour of my circuit.
I found that the drain-source junction of one of the 2N7002 was clamping the negative voltages the output could take. I couldn't reproduce this on LTSpice and my guess is that the output impedance of the gain selector pins is not modelled.
Regards