Dear Colleagues,

I'd like to share with you with a twinning puzzle that's been
troubling me for a while.
I have a ~2.2 A data merged in p212121 space group with unit cell:
41.8300  117.4490  134.4840   90.0000   90.0000   90.0000
Translational NCS was detected but structure solution was able to be
obtained by phased molecular replacement as we have a Au derivative
data set. Obtaining experimental phase or MR alone didn't work.

However, with this model and data set, the refinement didn't work
(with R-free ~40% or higher if with more cycles).  Different twinning
tests have been carried out and no sign of twinning.  All possible
p222 space groups have been tried but no luck or even worse.
Considering that translational symmetry and perfect twinning may make
things deceiving, the data was merged into p21 and a twin-refinement
was performed in phenix.refine.  After 3 cycles of rigid-body and
group ADP, a R-free of ~30% was obtained and the output map looks OK.
It looks like we got it.  But then things get interesting when we
tested other two axis as the p21 unique axis.  With all the three
choices plus corresponding twin-refinement, we could get a similar
R-free (30%).  Besides, the three maps all look OK except for one
missing some side chains.

Then the structure was solve in p1 and Xtriage was used with the hope
that one axis would stand out.  However, the RvsR for all three axis
are almost the same.  Did I miss anything here?  Of course, we could
pick up the one axis we like best and get the structure solved ^---^.
However, this puzzle would eat me alive for a very long time.

Best,

Hua