Scientific papers often have poor graphics, and peers often seem to focus on text rather than the figures.

R can export any graphic as PDF including text as moveable objects. Which can then be placed in an optimal manner using either paywalled software packages such as Adobe Illustrator or CorelDraw but also the free, LINUX-based Inkscape.

Labels should always be readable and not be partly covered by other objects or overlapping with other labels.

Regarding what can be seen in the graph and is written in the legend.

Given that the 1st and 2nd component only explain 55% variation: Are the percentages different enough from those that would be expected with a broken stick distribution for random data? Not a statistician myself but I guess a scree plot would have been not a bad idea to put the PCA into proper perspective.

One can depict each four triangles, squares and circles but the legend is not clear about what they indicate, are these the primary data points included for PCA? If so, what is the purpose of the "95% confidence" ellipses? With four datapoints per set, they all must lie inside, so it's just containing ellipses.