Most experienced cancer genomic folks will agree that pathology estimates of tumor content are often inaccurate. If the genomic sequencing tells you that the VAFs max out at 20, then your sample is likely 40% tumor (and 60% non tumor - normal contamination).

That said, there may be cases where you don't pick up the founding clone, especially when you're doing targeted or exome sequencing, or in tumors with very low mutation rates. You'll have to assess all the available evidence to decide which is more likely to be the case.

That said, there may be cases where you don't pick up the founding clone, especially when you're doing targeted or exome sequencing, or in tumors with very low mutation rates.

That's an interesting thought. For my samples I have exome sequencing data of 300x depth. At such a high depth I would expect the mutations present in all tumor cells (which would be the founding clone) to be detected, as you write in this paper:

Despite additional complexity from factors like contamination of non-tumor cells, tumor heterogeneity, and aneuploidy, cancer genomes have generally been sequenced to comparable depths, typically between 30x and 50x mean coverage (Borad et al., 2014 and Mardis, 2012). While this is enough coverage to discover SNVs in the founding clones of high-purity tumors, most tumors are not pure

But if what you say is true, I still think it strange that in two consecutive samples from the same patient I cannot detect a founding clone in both cases. I could miss it in one because of technical issues regarding how the biopsy was made, and the sequencing, but I don't think it would make sense to miss it in both. Unless that founding clone consists of mutations that occured only in the non-coding region of the genome. I could not find examples of that in the litterature, but I might of course have overlooked something.

To my mind, the only solution that makes sense both biologically and mathematically is that one you propose: the samples I am struggling with must have a much lower tumor purity than first estimated by the pathologist.

I have two hypothesis on such case. Not sure if it is logically sound...

Maybe more than one tumorigenesis network / pathway being activated in your sample and not all these mutants contribute to single one pathway.

Maybe only one oncogenic pathway being activated but different mutants on different genes contribute to this pathway

I think both of these situation could result in low VAF across all mutants in the sample.

Of course like Chris mentioned, the founding alteration may not be a SNV.