Rare suppressor effect?

Hey, I recently discovered a somewhat surprising result in a hierarchical regression:

I have three predictors (C M N) and one outcome (D).

Model 1 Regression (also with backgroun variables):
C: B = -.06, p <.05

Model 2 Regression:
C: B = .05, p<.01
M: B = -26, p<.01
N: B = .56, p<.01

If i run the same regression but including only one of the two other predictors in model 2, the same pattern emerges, but C is not significant.

Scatter C and D:



My thoughts is that C has variance both negatively and positively related to the outcome, although more negatively related (hence, the correlation coefficient). The same negative variance is also shared partly by N and partly by M, which "removes" this variance in the calcuation of the coefficient.

Any thoughts, interpretations?

PS: Maybe I've been too nice with the scatter plot when i was eye-balling linearity?