As a proxy for asking which bacterial proteins are membrane-bound, we can ask which proteins harbor transmembrane domains. A popular tool for detecting transmembrane domains is TMHMM, which looks for transmembrane helices in protein sequences using a hidden Markov model.1
For E. coli strain K12, UniProt reports 4,391 protein sequences corresponding to unique genes. Of these, TMHMM predicts 1,072 proteins have at least 1 transmembrane helix (24.4%), and 775 proteins have 2 or more transmembrane helices (17.6%). Expanding this analysis to the E. coli pan proteome, which contains 165,127 non-redundant protein sequences, TMHMM predicts 28,369 proteins have at least 1 transmembrane helix (17.2%), and 16,105 proteins have 2 or more transmembrane helices (9.8%).
For B. subtilis strain 168, UniProt reports 4,260 protein sequences corresponding to unique genes. Of these, TMHMM predicts 1,175 proteins have at least 1 transmembrane helix (27.6%), and 845 proteins have 2 or more transmembrane helices (19.8%). Looking at the pan proteome for this species, which contains 18,046 non-redundant protein sequences, TMHMM predicts 3,967 proteins have at least 1 transmembrane helix (22.0%), and 2,463 proteins have 2 or more transmembrane helices (13.6%).
The relative depletion of transmembrane domain-containing proteins in the pan genomes of both species relative to their reference genomes could point to a functional skew of accessory genomes towards proteins that are less likely to be membrane-associated, such as strain-specific virulence factors, antibiotic resistance genes, integrated phage proteins, and proteins involved in niche-specific metabolism.2,3 Additionally, the relative enrichment of transmembrane domain-containing proteins in B. subtilis compared to E. coli could reflect the different membrane compositions of Gram positive and Gram negative bacteria, though it's hard to say without sampling more species.
One important caveat to TMHMM -- predicted transmembrane segments in the N-terminal region are sometimes just signal peptides. Signal peptides are likely to appear as single transmembrane domains, which is why I've also reported the fraction of proteins with 2 or more transmembrane helices, above. A more complete analysis may run the same protein sets through SignalP 4 and cross-reference the TMHMM hits to identify transmembrane domain-containing proteins that also contain signal peptides.
Note that the cited publication by Almen et al. uses TMHMM and directly references the results of the original TMHMM paper,1 which reports transmembrane protein abundance in the proteomes of several species.
One of the most referenced papers regarding the percentage of membrane proteins in proteomes is from 2001 where the membrane topology prediction method TMHMM was applied on a number of proteomes from different species to estimate the membrane protein content, for example, Caenorhabditis elegans (31%), Escherichia coli (21%) and Drosophila melanogaster (20%).
- Krogh A, Larsson B, von Heijne G, Sonnhammer EL. Predicting transmembrane protein topology with a hidden Markov model: application to complete genomes. J Mol Biol. 2001 Jan 19;305(3):567-80.
- Kung VL, Ozer EA, Hauser AR. The accessory genome of Pseudomonas aeruginosa. Microbiol Mol Biol Rev. 2010 Dec;74(4):621-41.
- Álvarez VE, Quiroga MP, Galán AV, Vilacoba E, Quiroga C, Ramírez MS, Centrón D. Crucial Role of the Accessory Genome in the Evolutionary Trajectory of Acinetobacter baumannii Global Clone 1. Front Microbiol. 2020 Mar 18;11:342.
- Nielsen H, Brunak S, von Heijne G. Machine learning approaches for the prediction of signal peptides and other protein sorting signals. Protein Eng. 1999 Jan;12(1):3-9.