What he did: He ran control DNA (from cells grown with lots of phosphate and no arsenate) in two CsCl gradients and collected fractions (~ 100 µl fractions from gradients with total volumes of 1 or 2 ml). He then measured the volume and DNA concentration of each fraction. This showed a nice DNA peak in each gradient (green is the 1 ml gradient, red is the 2 ml gradient). He then pooled the high-DNA fractions of each gradient, desalted them to remove the CsCl, and digested the DNAs in preparation for mass spectrometry (LC-MS).
What I've done: Arithmetic to calculate how much purification these gradients would have accomplished.
The green data: 6778 ng of DNA (89% of the total DNA recovered) is in four fractions with a total volume of 300 µl (37% of the volume recovered). This means that the concentrations of soluble contaminants not bound to the DNA will have been reduced to about 40% of what they were.
The red data: 5135.3 ng of DNA (68% of the total DNA recovered) is in two fractions with a total volume of 310 µl (17% of the total volume recovered). This means that the concentration of soluble contaminants will have been reduced to about 25%.
Hmm, that's not very efficient purification. Larger gradient volumes and longer spins might help. And of course the desalting step should have removed much more of the soluble contaminants.
But this arithmetic may not matter much. The real advantage of the CsCl step is not that it's removing soluble contaminants. Instead, it's fractionating on completely independent principles than the other steps we use, and so it is expected to reduce or remove contaminants that the other methods might not remove. It should remove contaminants that might have coprecipitated with the DNA when it was spooled out of 70% ethanol, and ones that might elute with the DNA in the desalting column because they're insoluble and soluble under the same combinations of conditions as DNA (we typically have to treat these conditions as 'secret sauce' because the manufacturers of the desalting columns don't like to reveal how they work).
In both gradients, contaminants that weren't soluble in the CsCl solution will probably have either pelleted to the bottom of the tube or risen to the top of the tube, depending on their density relative to the CsCl. (As I recall from my undergrad experiences, RNA pellets but proteins rise.) Whether these would have recontaminated the fractions as they were collected depends on how the collection was done. Fractions collected as drops from the bottom of the tube would have avoided contaminants that had risen to the top but would have encountered any pelleted contaminants on their way out.
Do we need to also consider contaminants that might have banded at a specific density in the gradient? The centrifugation is powerful enough to cause the heavy Cs+ ions to move down in the tube, might it also affect the distribution of other ions? What does Wikipedia say? (Ah, the correct term is 'isopycnic centrifugation'.) Nothing about other ions. CsCl gradients have typically been used to separate DNAs with different base compositions from each other (e.g. nuclear DNA from mitochondrial or plastid DNA); I don't know if anyone ever used them to separate DNA from soluble contaminants.
Bottom line: If the LC-MS data shows arsenic in the DNA, we can polish up these DNA purification steps. If it doesn't, we won't need to bother.