Normal VH domains without a partner VL domains are difficult to express in functional form because they are highly aggregation prone. VHH domains are based on an exceptionally stable domain core, mutations in the VL/VH dimer interface render it less hydrophobic and they carry very long CDR-H3 loops that at least partially mask the VL/VH dimer interface . In humanized VHH/nanobody libraries, some of these features have been introduced into the human VH3 sequence to produce synthetic libraries that combine a human-like core with the adaptations found in camelid VHH domains. While the rabbit VH domains are closer to huVH3 than to other, less stable germline families, they still have the very hydrophobic VL/VH dimer interface. Libraries containing a large fraction of aggregation-prone members are difficult to select from, because they produce a much higher background of non-specific binding.

Actually, check out this paper:

Article Efficient generation of single domain antibodies with high a...

Dear Dr Honegger,

Thanks you very much for your help, this will help me a lot. Seems that the temperature is a key factor and, after selection I can "camelized" the best candidates.