The discovery of proteins translated from non-canonical open reading frames has expanded the recognized human proteome into a so-called ghost proteome composed of alternative proteins (AltProts). These small proteins—hereafter referred to as MicroAltProts when under 200 amino acids—are challenging to detect because of their size, diverse physicochemical properties, and poor annotation in reference databases. The authors aimed to develop an integrated bioinformatic and proteomic workflow to benchmark detection of reference proteins (RefProts), isoforms, and MicroAltProts in colorectal cancer cells and to characterize methodological detection biases.
The experimental comparison used four protein extraction protocols applied to colorectal cancer cells: HCl extraction; RIPA buffer; RIPA combined with chloroform; and RIPA followed by 30 kDa filtration. These protocols were chosen to sample a wide range of physicochemical fractions and to test whether different extraction chemistries bias identification toward specific MicroAltProt types.
High-resolution data-independent acquisition (DIA) mass spectrometry on an Orbitrap Astral instrument was used to analyze extracts. Across the methods the workflow identified 66,438 peptides mapping to 12,584 distinct protein groups. The authors report that RIPA-based extraction approaches provided the most comprehensive proteome coverage among the tested protocols.
To focus on small alternative proteins and reduce redundancy inherent to comprehensive resources, the OpenProt database was curated by removing known isoforms and longer proteins. This curation yielded a non-redundant dataset comprised of 183,937 MicroAltProts for downstream classification, clustering, and experimental interrogation.
The study derived eight ProtParam-related features for each MicroAltProt and applied K-means clustering to group sequences into four physicochemical clusters. Clustering captured distinctions in properties relevant to detection and biophysical behaviour, including predicted disorder content, net charge (alkalinity), hydrophobicity, and transmembrane propensity. These clusters were used to interpret which MicroAltProts were more likely to be captured by each extraction protocol.
From the experimental dataset, 43 MicroAltProts shorter than 200 amino acids were validated by mass spectrometry. These experimentally detected MicroAltProts were classified into tiers following recently recommended international guidelines referenced in the study, providing a structured evidence framework for the reported identifications.
Cluster assignment of the detected MicroAltProts highlighted method-dependent detection biases. The HCl extraction protocol favored identification of MicroAltProts with disordered, alkaline sequence characteristics. In contrast, RIPA-based protocols enhanced recovery and identification of MicroAltProts predicted to be membrane-associated and to contain amphipathic alpha-helical segments. These findings indicate that extraction chemistry materially influences which regions of the ghost proteome are experimentally accessible.
Structural predictions for experimentally detected MicroAltProts indicated diverse folding determinants across the set. Predicted features included transmembrane helices, intrinsically disordered regions, and motifs resembling nucleic acid-binding segments. Such structural heterogeneity supports the possibility of varied cellular roles for MicroAltProts, including membrane association and interactions with nucleic acids, although the study reports predictions and does not claim functional validation beyond structural inference.
The authors present an integrated bioinformatic and proteomic framework that benchmarks simultaneous detection of RefProts, isoforms, and AltProts and documents extraction-dependent biases. Key outcomes include a curated non-redundant MicroAltProt catalogue, a four-cluster physicochemical classification, experimental validation of 43 MicroAltProts, and evidence that RIPA-based protocols give broader proteome coverage while HCl extraction preferentially isolates disordered alkaline MicroAltProts. Collectively, the work provides a methodological roadmap to improve and interpret detection of the human ghost proteome and suggests a broader functional repertoire for MicroAltProts based on their physicochemical and predicted structural diversity.
The authors declared no competing interests. Funding sources reported include grants from the Ministerio de Ciencia, Innovación y Universidades and projects supported by the Instituto de Salud Carlos III. The article is a preprint and has not been certified by peer review.