**Summary** pdb2gmx generates correctly the topology file for a single head-to-tail cyclic peptide. But if there is another chain in the PDB file, for example a protein-peptide complex, then this works only, if the cyclic peptide appears before the protein in the PDB file. If the protein appears first in the PDB file, then the topology of the cyclic peptide is not generated correctly (missing bond N-ter-C-ter, three protons on N-ter, etc). **GROMACS version** gmx 2023.3 and gmx 2024.4 For gmx 2023.3: 2023.3-Ubuntu_2023.3_1ubuntu3 **Steps to reproduce** You can download these example files: https://dropsu.sorbonne-universite.fr/s/QNQfFK5CwXC2TDa From a cyclic peptide-protein complex (PDB 5N99) I've extracted three structures of the same cyclic peptide (chains C, E and F), as the asymmetric unit contains three copies of the biological complex => file 5n99_chainCF.pdb All three structures are separated in space from each other. The residue DPR (for D-proline) had to be replaced by PRO in the PDB files here. The command line: gmx pdb2gmx -f 5n99_chainCF.pdb -o 5n99_chainCF.gro -p 5n99_chainCF.top -ignh -water tip3p yields correct head-to-tail bonds for all three cyclic peptides (see in *.itp files the bond between atom 1 and 85). I selected the Amber96 force field (choice 3) here, but the bug seems independent on the force field. In a second test, I truncated the first peptide (chain C) by removing the last residue from it, in order to generate a linear peptide without head-to-tail cyclisation. The other two cyclic peptides (chain E and F) stayed the same => file 5n99_chainCF_truncated.pdb This time the command line : gmx pdb2gmx -f 5n99_chainCF_truncated.pdb -o 5n99_chainCF_truncated.gro -p 5n99_chainCF_truncated.top -ignh -water tip3p gives *.itp topology files, without any head-to-tail bond for ALL three cyclic peptides, as can be seen also with the addition of two protons the N-ter at the beginning of the *.itp files. No error or warning message appeared here. **What is the current bug behavior?** See above **What did you expect the correct behavior to be?** As all three chains are independent here, the cyclisation should also be treated independently, or the outcome of the first chain (linear peptide or protein) affects the outcome for all the remaining chains. **Possible fixes** in the pdb2gmx.cpp file, lines 2434 to 2458 (latest gromacs version): for (int i = 0; i < cc->nterpairs; i++) { find_nc_ter(pdba, cc->chainstart[i], cc->chainstart[i + 1], &(cc->r_start[j]), &(cc->r_end[j]), residueTypeMap, logger); if (cc->r_start[j] >= 0 && cc->r_end[j] >= 0) { if (checkChainCyclicity( pdba, pdbx, cc->r_start[j], cc->r_end[j], rtpFFDB, rtprename, long_bond_dist_, short_bond_dist_)) { cc->cyclicBondsIndex.push_back(cc->r_start[j]); cc->cyclicBondsIndex.push_back(cc->r_end[j]); cc->r_start[j] = -1; cc->r_end[j] = -1; } else { j++; } } } there is either a problem with find_nc_ter or checkChainCyclicity or elsewhere in this big loop (not shown here) over all chains. Somehow the state of the first iteration (=first chain) affects the results in all further iterations (chains). Would be nice, if pdb2gmx could be compiled and tested independently of all the rest of GROMACS.