Could I say some peaks in column 1 and 2 are closed meanwhile these peaks are opened in column 3 and 4?

All this really lets you say is that those peaks have different levels of signal between your samples. z-scores can be misleading, as very robust relative changes for stable features (peaks) with a z-score of 2-3 can actually obfuscate very modest absolute differences in signal for any given feature if variability for that feature across samples is low.

That's not to say that's necessarily the case here - z-scored heatmaps just should not be the only way you look at such data to determine if groups of features are meaningfully different. As an aside, removing the rownames from your Heatmap will make it render much more quickly.

For that paper/figure:

We plotted the Log2 fold change for each peak and generated a heatmap from this data to visualize the overall differences in accessibility after insulin treatment. In general, peaks that were more closed in the vehicle-treated samples were more open in the insulin-treated samples (Fig. 1B).

Emphasis mine. Making a heatmap of log2 fold changes doesn't make sense to me, since you could just plot the actual fold changes and have a more clear effect. I expect the text may be inaccurate, but it's confusing at minimum.

I'd take a peak at DiffBind for some additional visualization and analysis methods.