The data fit just fine if you assume these 6 samples being from a Denisovan subpopulation and not from a homo erectus relic population.
To me it's more likely that the 'homo erectus' variant emerged later, on top of the haplotype that carries the 'Denisovan-like' variant. So after an introgression event happened, if it did happen for this protein.
The two variants that are recovered are also found on the same protein!
If they were introgressed from homo erectus, they likely would have been linked together due to their physical proximity on the genome, and Denisovans would have inherited both.
Interesting points!
If you look at the protein data alone, there are no recovered variants that support these samples being very divergent from modern humans, e.g. 'H.erectus like'. For comparison, the H.antecessor enamel had a variant, placing it as an outgroup to nean, deni, and present day humans
Given the recent and upcoming recovery of ancient proteins from archaic human remains this would be something to keep in mind. The missmatch of autosomal, mitochondrial, Y chromosome and morphology-based trees is already complicated for these groups, proteins might add some to that complexity!
While we suspect incomplete lineage sorting as the main culprit for this result, an older admixture between Neanderthal and humans, a 'super-archaic' admixture into Denisovans or other scenarios could be at play.
In fact, in the trees of enamel, collagen and other bone/dentin proteins, humans and Neanderthals appear closer to eachother as often as Neanderthals appear closer to Denisovans. This persists even when only looking at sub-saharan African populations, who have limited archaic introgression.