There we go, now that's easier to understand. And thank you for including a spectrum for each peak, that's also helpful. Negative peaks aren't so uncommon when doing UV detection in HPLC, something is separating out the has less absorbance than initial conditions -- perhaps, some solvent component of your sample, or a contaminant with a very strong absorbance outside of your selected spectrum window.
There are some tricks you can employ. For starters, different integration software have different ways of dealing with negative peaks, if you turn on, or turn off, negative peak detection, the system may draw a more correct baseline, for the peak that precedes it. A gradient, or mobile phase change, for isocratic, may separate the peaks better, as another option.
There's an important consideration many people don't realize. The knee-jerk response for people developing UV detection is to set the peak detection at lambda max, and that's not always the best way. Sure, that gives you the most signal, but you're at the mercy of your instrument's accuracy, which may drift. You'll notice, from your spectra, that if you're outside lambda max by say, 5 nm, the peaks would all be smaller, but the effect of the negative peak on the baseline might be much less.