PLUMED can use the PDB format in several places PLUMED can use the PDB format in several places

To read molecular structure (MOLINFO).
To read reference conformations (RMSD, but also many other methods in Distances from reference configurations, FIT_TO_TEMPLATE, etc).

The implemented PDB reader expects a file formatted correctly according to the PDB standard. In particular, the following columns are read from ATOM records

columns | content
1-6     | record name (ATOM or HETATM)
7-11    | serial number of the atom (starting from 1)
13-16   | atom name
18-20   | residue name
22      | chain id
23-26   | residue number
31-38   | x coordinate
39-46   | y coordinate
47-54   | z coordinate
55-60   | occupancy
61-66   | beta factor

PLUMED parser is slightly more permissive than the official PDB format in the fact that the format of real numbers is not fixed. In other words, any parsable real number is ok and the dot can be placed anywhere. However, columns are interpret strictly. A sample PDB should look like the following

ATOM      2  CH3 ACE     1      12.932 -14.718  -6.016  1.00  1.00
ATOM      5  C   ACE     1      21.312  -9.928  -5.946  1.00  1.00
ATOM      9  CA  ALA     2      19.462 -11.088  -8.986  1.00  1.00

Notice that serial numbers need not to be consecutive. In the three-line example above, only the coordinates of three atoms are provided. This is perfectly legal and indicates PLUMED that information about these atoms only is available. This could be both for structural information in MOLINFO, where the other atoms would have no name assigned, and for reference structures used in RMSD, where only the provided atoms would be used to compute RMSD.

Occupancy and beta factors

PLUMED reads also occupancy and beta factors that however are given a very special meaning. In cases where the PDB structure is used as a reference for an alignment (that's the case for instance in RMSD and in FIT_TO_TEMPLATE), the occupancy column is used to provide the weight of each atom in the alignment. In cases where, perhaps after alignment, the displacement between running coordinates and the provided PDB is computed, the beta factors are used as weight for the displacement. Since setting the weights to zero is the same as not including an atom in the alignement or displacement calculation, the two following reference files would be equivalent when used in an RMSD calculation. First file:

ATOM      2  CH3 ACE     1      12.932 -14.718  -6.016  1.00  1.00
ATOM      5  C   ACE     1      21.312  -9.928  -5.946  1.00  1.00
ATOM      9  CA  ALA     2      19.462 -11.088  -8.986  0.00  0.00

Second file:

ATOM      2  CH3 ACE     1      12.932 -14.718  -6.016  1.00  1.00
ATOM      5  C   ACE     1      21.312  -9.928  -5.946  1.00  1.00

However notice that many extra atoms with zero weight might slow down the calculation, so removing lines is better than setting their weights to zero. In addition, weights for alignment need not to be equivalent to weights for displacement.

Systems with more than 100k atoms

Notice that it very likely does not make any sense to compute the RMSD or any other structural deviation using so many atoms. However, if the protein for which you want to compute RMSD has atoms with large serial numbers (e.g. because it is located after solvent in the sorted list of atoms) you might end up with troubles with the limitations of the PDB format. Indeed, since there are 5 columns available for atom serial number, this number cannot be larger than 99999. In addition, providing MOLINFO with names associated to atoms with a serial larger than 99999 would be impossible.

Since PLUMED 2.4 we allow hybrid 36 format to be used to specify atom numbers. This format is not particularly widespread, but has the nice feature that it provides a one-to-one mapping between numbers up to approximately 80 millions and strings with 5 characters, plus it is backward compatible for numbers smaller than 100000. This is not true for notations like the hex notation exported by VMD. Using the hybrid 36 format, the ATOM records for atom ranging from 99997 to 100002 would read like these:

ATOM  99997  Ar      X   1      45.349  38.631  15.116  1.00  1.00
ATOM  99998  Ar      X   1      46.189  38.631  15.956  1.00  1.00
ATOM  99999  Ar      X   1      46.189  39.471  15.116  1.00  1.00
ATOM  A0000  Ar      X   1      45.349  39.471  15.956  1.00  1.00
ATOM  A0000  Ar      X   1      45.349  38.631  16.796  1.00  1.00
ATOM  A0001  Ar      X   1      46.189  38.631  17.636  1.00  1.00

There are tools that can be found to translate from integers to strings and back using hybrid 36 format (a simple python script can be found here).