In-source fragmentation in mass spectrometry-based proteomics: prevalence, impact, and strategies for mitigation

13 Peptide-level analyses are becoming increasingly popular in mass spectrometry-based proteomics and 14 are being applied, for example, in immunopeptidomics, structural proteomics, and analyses of post-15 translational modifications. In such analyses, peptides that are not biologically meaningful but instead 16 arise as artifacts prior to mass spectrometry analysis pose the risk of data misinterpretation. Here, we 17 describe an approach based on retention time analysis and precise chromatographic peak matching to 18 identify peptides generated by in-source fragmentation (ISF), which occurs between chromatographic 19 separation of peptide mixtures and the first mass filter of a tandem mass spectrometer (MS). To 20 understand the prevalence and properties of ISF, we generated 13 proteomics datasets and analyzed 21 them along with additional 25 previously published datasets spanning a broad range of sample types, 22 MS, and proteomics approaches including classical bottom-up proteomics, immunopeptidomics, 23 structural proteomics, and phosphoproteomics. We found that, in typical trypsin-digested samples on 24 average 1 % of fully-tryptic peptides and 22 % of semi-tryptic peptides originated from ISF. However, 25 we observed large variations between datasets, and in-source fragments exceeded, in some cases, a 26 third of the total peptide identifications. The extent of ISF was dependent on the peptide sequence, the 27 instrument, method parameters, and sample complexity. Although ISF did not impair relative 28 quantification across samples, it generated peptides that could be misinterpreted qualitatively, inflated 29 peptide identifications, and comprised up to 37 percent of peptides shorter than 9 amino acids in 30 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 2 immunopeptidomics datasets. We propose that, for peptide-centric applications, our open-source ISF 31 detection approach be used to re-annotate peptides generated by ISF and remove them to avoid 32 misinterpretation of data. ISF is an increasing concern with improving mass spectrometers, as they 33 enable detection of an ever-increasing number of m/z features, including low abundance features like 34 ISF products. Our work thus addresses a growing issue in proteomics and presents solutions to mitigate 35 the impact of in-source fragment peptides. In the future, improved feature detection algorithms may 36 enable elucidation of new ISF patterns affecting side chains that have been missed so far, which could 37 contribute to explaining the vast space of as-yet unannotated proteomics data. 38 39

40 proteomics / in-source fragmentation / mass spectrometry / retention time / 41 immunopeptidomics / phosphoproteomics 42 43

44 Mass spectrometry (MS) has emerged as a standard tool for the analysis of complex metabolite1,2 45 and protein 3,4 samples. While MS has been broadly applied in scientific and industrial research 46 for decades, an increasing number of MS methods are now applied in clinical diagnostics as well5. 47 In MS, analytes such as metabolites and peptides are ionized prior to measurement 6, with 48 electrospray ionization (ESI)7 being one of the most frequently used ionization methods in current 49 approaches. During ESI, a liquid phase containing analytes is sprayed through an injection needle 50 and subjected to high temperatures (25 – 500 °C) 8 and electric potentials (500 - 4500 V) 6. This 51 leads to the ionization of the analytes and evaporation of the liquid phase. Although ESI is often 52 described as a soft or gentle ionization technique 6,9–11, analytes can be chemically modified or 53 partially fragmented during ESI11–13. Such in-source fragmentation (ISF) is typically considered an 54 artifact because it creates new molecular species that can be misinterpreted as biologically 55 relevant. 56 In the metabolomics field, ISF has been extensively studied 10–18, and depending on the study, 5 57 up to 70 % of the data have been reported to originate from ISF10,11,13,15–18. Several bioinformatic 58 tools and strategies have been developed to annotate in-source fragments in metabolomics 59 data10,11,15,19. If analytes are separated by chromatography before MS measurements, in-source 60 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 3 fragments can readily be identified because they share the same retention time and 61 chromatographic elution profiles as their parental molecule 11,19. In cases where retention times 62 are not available, for example in high-throughput metabolomics approaches that rely on flow 63 injection analyses20, network approaches can identify in-source fragments13,15. 64 In contrast to the metabolomics field, ISF has been much less studied in the context of 65 proteomics. A possible reason for this is that the primary goal of classical bottom-up proteomics 66 has been the differential abundance analysis of proteins, which was dominated by data 67 dependent acquisition (DDA) MS methods and by peptide search engines that mostly focus on 68 identifying fully-tryptic peptides. In this context, the probability of detecting fully-tryptic peptides 69 as artifacts arising from ISF is low because for such a mis-annotation to occur, a longer parental 70 peptide arising from missed trypsin cleavage events would have to fragment into a fully-tryptic 71 peptide of sufficient intensity to be selected for MS2 measurements by DDA. In protein 72 abundance analysis, potential effects of ISF on the accuracy of peptide quantification would be 73 diminished during aggregation of peptide quantities to protein quantities. Therefore, ISF has not 74 been considered of major concern so far. In support of this, a previous study identified ISF as a 75 major source of half-tryptic peptides in DDA MS, but showed that it had a low impact for complex 76 samples measured, as in-source fragments represented only 1 - 3 % of the analyzed DDA data 21. 77 However, data independent acquisition (DIA) MS methods have gained much in popularity in 78 recent years 22,23. Together with new mass spectrometers with drastically increased sensitivity, 79 they enable the detection of an ever-increasing number of peptides, and it is unclear whether 80 such approaches are more prone to identifying in-source fragments than previous methods. 81 Further, it is known that instrument parameters have an impact on ISF, but the extent of these 82 effects and the relative impact of different parameters is unclear. To complicate this matter, the 83 number of proteomics applications that rely on peptide-centric analyses to quantify half- or non-84 tryptic peptides, or post-translational modifications (PTMs), is steadily increasing 24. In these 85 analyses, the quantitative impact of ISF is expected to be more pronounced as peptide quantities 86 would not be aggregated to protein quantities. In addition, qualitative effects of ISF on peptide 87 identification could substantially compromise data interpretation. For instance, in mass 88 spectrometry-based immunopeptidomics or structural proteomics approaches using limited 89 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 4 proteolysis25,26, biological findings rely on the precise characterization of every peptide from a 90 given protein, including those with semi- or non-tryptic termini. It remains unclear how ISF 91 impacts peptide-centric analyses, quantitatively and qualitatively. 92 We have therefore revisited the prevalence, characteristics, and impact of ISF in the context of 93 proteomics focusing on DIA datasets acquired with state-of-the-art mass spectrometers. We 94 systematically evaluated a range of instrument parameters and sample types, with particular 95 emphasis on assessing the implications of ISF for peptide-centric proteomics approaches. 96 Further, we provide practical solutions on how to mitigate ISF in proteomics. 97 98

99 Detecting in-source fragment peptides in proteomics data 100 To analyze the prevalence and impact of ISF products, we first sought to develop a bioinformatic 101 pipeline for their detection and quantitative analysis. Peptides are usually separated by liquid 102 chromatography prior to electrospray ionization and mass spectrometry in bottom-up 103 proteomics (Supplementary Fig. 1). Since ISF occurs after chromatographic peptide separation 104 and prior to the first mass filter, in-source fragments and their respective parent peptides have 105 the same retention time profiles despite being of different molecular mass. Fragment peptides 106 also share their amino acid sequence with their parents. We therefore developed an algorithm 107 to detect peptide ISF by identifying peptides with shared sequences that co-elute, or, more 108 accurately for in-source fragments, have the same retention time (RT). To illustrate our approach, 109 we consider a theoretical example with 8 different peptides that share a sequence (Fig 1.a), which 110 yield 28 unique peptide pairs. For each pair, the difference between the apex retention times 111 (∆RT) was calculated by subtracting the RT of the shorter (or equally sized) peptide (peptide 1) 112 from the RT of the longer peptide (peptide 2) (Fig. 1.b). Peptides that resulted from ISF, which 113 were peptides I, II, VI, and VII in our example, should have small or negligible ∆RTs if paired with 114 their respective ISF parent peptides. We thus identified these peptides by defining a ∆RT cutoff 115 and inspecting which peptide pairs have a ∆RT that fell below it. 116 As RT distributions, chromatographic peak shapes, and resolution can vary between instrumental 117 setups or runs, we determined the “co-elution” ∆RT cutoff for each sample individually. To obtain 118 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 5 this cutoff, we analyzed the data at the peptide precursors level, as throughout this study if not 119 stated otherwise, and used the ∆RT distribution of peptides that differ only in charge and not in 120 mass, which were peptides IV and V in our example. Because these peptides must have the same 121 RT, they provide ground truth for the natural variation of ∆RT values of co-eluting peptides (Fig. 122 1.c). As false-discovery-rates are commonly set at 1 % for peptide searches, a small fraction of 123 peptide pairs will contain a falsely annotated peptide resulting in a continuous uniform 124 distribution of ∆RT values corresponding to these false positives. The ∆RT cutoff for co-eluting 125 peptides is defined as the ∆RT at which the normal distribution of true-positives intersects with 126 the continuous uniform distribution of the false-positives. In the next step, peptide pairs were 127 classified as pairs with ISF evidence, if their ∆RT was below the ∆RT cutoff, or as pairs without ISF 128 evidence (Fig. 1.d). In the last step of our ISF detection, peptide pairs were uncoupled and, in 129 case of ISF evidence, shorter, type 1 peptides annotated as fragment peptides, and longer, type 130 2 peptides as parent peptides (Fig. 1.e). If there was no ISF evidence, peptides were referred to 131 as non-ISF peptides. 132 In case of a parent peptide yielding more than one fragment peptide, the intermediary fragments 133 will still be annotated as fragments but not as parent peptides, implying that parent and fragment 134 peptide annotations are mutually exclusive. Multiple in-source fragments of a parent peptide 135 also enable the construction of ISF networks, in which each peptide is a node and each edge a 136 peptide pairing with a ∆RT. These fragmentation networks enable the identification of in-source 137 fragments that share the same parent and can be used to define new peptide groups, for which 138 all individual peptide quantities of each group can be aggregated to single peptide group 139 quantities. We further identify C-terminal ISF if the N-terminal part of the parent is detected as 140 a fragment, and vice versa for N-terminal ISF. We denote the first amino acid at the C-terminal 141 side of the ISF site F1, and the first amino acid at the N-terminal side F1’, analogous to the 142 nomenclature of proteolytic cleavage sites (P1, P1’, etc.). While the detection of F1 and F1’ 143 implies an ISF that alters the amino acid sequence of the parent peptide, it is in principle possible 144 that an ISF does not change the amino acid sequence but affects a side chain or PTM. As some 145 PTMs like oxidation of methionine are included in most peptide searches by default, we also 146 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 6 consider PTM in-source fragmentations (PTM ISF), defined as an ISF event that detaches the PTM 147 without altering the amino acid sequence. 148 We tested our ISF detection algorithm on DIA data from a Saccharomyces cerevisiae 288c sample 149 spiked with 8 proteins, which we used as a reference sample throughout this study. We chose 150 this sample as a reference because we suspected that peptides with high concentrations were 151 more likely to yield measurable in-source fragments, and this sample yielded such peptides (from 152 the spiked in proteins) in a controlled fashion without compromising on overall sample 153 complexity. Our model identified a ∆RT cutoff of ± 0.129 min for co-eluting peptides with a recall 154 of 97.1 % for the true positive peptide pairs, which were the peptides of equal mass but different 155 charge states (Supplementary Fig. 2a). This ∆RT cutoff was more conservative than the median 156 full width at half maximum (FWHM), which was 0.19 min and is often used to determine peak 157 separation. Next, we used the ∆RT cutoff to determine ISF peptides and found that 24.4 % of all 158 peptide pairs that differed in molecular mass fell within the cutoff (Supplementary Fig. 2b). This 159

provides the first evidence that in-source fragmentation is prevalent in DIA proteomics 160 data. 161 In the ∆RT distribution, we further observed a broad peak of peptide pairs around -10 min. We 162 found that this was due to peptides carrying an oxidation group as PTM (Fisher’s exact test 163 checking for enrichment between -13 and -5 min, right-tailed, P-value 0, which was expected because peptide 165 pairs that were not due to ISF typically have a longer type 2 peptide that is often stronger retained 166 on the chromatographic column than the shorter type 1 peptide. 167 To our knowledge, there are no published, open access software tools to detect ISF in DIA 168 proteomics data to which we could compare our ISF detection algorithm. However, we compared 169 it to another, commercially available and unpublished algorithm that was developed in parallel 170 to our study as a feature for Spectronaut 20 (Biognosys). For our yeast reference sample, 97.7 % 171 of all ISF annotations across n = 4 technical replicates (4716 out of 4828) were identical, showing 172 broad agreement between both algorithms (Supplementary Fig. 3) and providing a cross-173 validation. 174 In-source fragmentation is abundant 175 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 7 Next, to determine how abundant peptide ISF is in data from typical trypsin-digested bottom-up 176 proteomics samples measured by DIA, we analyzed 16 published datasets, acquired with 9 177 different mass spectrometers models, from different research groups (Supplementary Tab. 1) 27–178 42. These datasets comprised 317 samples from a broad range of sources, including Escherichia 179 coli, Saccharomyces cerevisiae , Arabidopsis thaliana , Mus musculus , and Homo sapiens, and 180 comprising both laboratory cultures and human biopsies. The number of detected protein groups 181 ranged from 3 to 6,180. In-source fragments made up 3.4 ± 3.2 % of the total peptide precursor 182 identifications (mean and standard deviation, n = 317 samples) across the datasets, and 4.8 ± 4.2 183 % of the total peptide intensity (Fig. 1a). However, we observed a large variation in the extent of 184 ISF, ranging from 0.2 ± 0.1 % by intensity (PXD062917, n = 9 samples; 0.5 ± 0.5 % by 185 identifications) to 15.5 ± 0.3 % by intensity (PXD069457, n = 6; 13.1 ± 0.1 % by identifications). 186 Increasing sample complexity generally yielded lower shares of in-source fragments within a 187 dataset (% by identifications and by intensities), although not every sample with a low complexity 188 had high levels of ISF. To better understand how sample complexity affects ISF, we prepared the 189 following 7 trypsin-digested samples of ascending complexity and compared the prevalence of 190 ISFs: (1) mix of 11 peptides, (2) mix of 8 proteins, (3) E. coli BW25113 lysate, (4) affinity-191 purification MS (APMS) from a H. sapiens kinase pull-down (5) S. cerevisiae S288c lysate, (6) H. 192 sapiens HEK-293 lysate, and (7) the tribrid proteome sample obtained by combining samples 3, 193 5, and 6. We detected ISF peptides in all samples (Fig. 2.b). The tribrid proteome sample (n = 4 194 technical replicates) had the lowest level of ISF, with fragments representing 4.01 ± 0.06 % of 195 total peptide intensity (2.591 ± 0.005 % of identifications, mean and standard deviation), while 196 the mix of 8 proteins had the highest value (24.1 ± 1.3 % by intensity; 31.0 ± 0.5 % by 197 identifications). Across all samples, both the intensities as well as the numbers of fragment 198 peptide identifications increased with decreasing sample complexity. The absolute number of 199 parent peptides was on par with the number of fragments, except for the low complexity samples 200 (peptide and protein mixes), which had more fragments than parents (Fig. 2.c). Although we 201 observed fewer parent peptides than non-ISF peptides, the intensities of the parent peptides 202 were in the same range as the non-ISF peptides (Fig. 2.c). This also meant that at least 32 % ( H. 203 sapiens) and up to 95 % (mix of peptides) of the total peptide intensity could be associated with 204 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 8 ISF, combining the intensities of fragments and parents (Fig. 2.d). This was consistent with our 205 observations in the published datasets, in which on average 37 ± 22 % of the total peptide 206 intensity could be associated with ISF, with large variation between individual datasets. Across 207 all datasets (published and our new data reflecting a range of complexity), 82 % of the in-source 208 fragments were semi-tryptic peptides (Fig. 2.e), and in-source fragments represented 0.2 – 83.3 209 % of semi-tryptic peptides, depending on the sample (Fig. 2.f). 210 Taken together, these results showed that ISF varies strongly between datasets, depends on 211 overall sample complexity, and can account for a large share of the mass spectrometry signal. ISF 212 is especially prevalent in low complexity samples, highlighting the importance of ISF detection in 213 those cases. 214 215 Characterization of ISF peptides 216 To learn general properties of the observed ISF peptides, we analyzed combined data with 217 2,407,805 identified peptides from all the (published and newly generated) datasets described 218 above. 30.3 % of the peptides shared their sequence with another peptide and formed around 219 13.7 million peptide pairs (Fig. 3.a). 1,222,589 of the peptide pairs (8.9 %) showed evidence of 220 ISF, of which 67.9 % had sequence-altering ISF, and 26.6 % were fragment-to-fragment pairs. 221 Most fragment peptides originated from N-terminal ISF of fully-tryptic peptides and had a C-222 terminal arginine or lysine (Supplementary Fig. 4.a). Further, non-ISF peptides had mostly lower 223 intensities than in-source fragments and parents, with parent peptides having higher intensities 224 than fragments (Fig. 3.b and Supplementary Fig. 4.b). This indicated that the detected in-source 225 fragments and parents either derived from abundant proteins, were especially amenable to 226 measurement by MS, or both. As expected, in-source fragment peptides often had a lower 227 number of charges than their respective parent peptides and were shorter than the parent and 228 non-ISF peptides, as 47.3 % of the ISF peptide pairs showed losses in charge and 92.5 % showed 229 losses of amino acids due to ISF (Supplementary Fig. 4.c, d, e, and f). 230 To understand if ISF events had a bias for certain amino acid bonds, we analyzed how often an 231 amino acid occurred at F1’ or F1 of an ISF site. Valine followed by leucine, isoleucine, and alanine 232 were most abundant at F1’, while proline was most abundant at F1, followed by glycine and 233 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 9 alanine (Fig. 3.c). An enrichment analysis of peptides bonds at ISF sites, using all possible peptide 234 bonds of detected peptides as null distribution (two-tailed Fisher’s exact test), showed that 56 235 out of 80 possible peptide bonds with short, hydrophobic amino acids (valine, isoleucine, leucine, 236 and alanine) at F1’ showed a significant enrichment (Benjamini-Hochberg 43 adjusted P-values < 237 e-20), and 17 out of 20 bonds with proline at F1 were enriched at ISF sites (Fig. 3.d). The 238 aspartate-proline bond is known to be susceptible to non-enzymatic cleavage under acidic 239 conditions44. Here, we did not observe an enrichment of the aspartate-proline bond at 240 fragmentation sites but rather a depletion. However, we detected enrichment of the aspartate-241 proline bond at non-tryptic termini of semi-tryptic non-ISF peptides (Supplementary Fig. 5). 242 As they made up over a quarter of the ISF peptide pairs, fragment-to-fragment pairs indicated 243 that multiple ISF events of peptides are abundant. We thus constructed ISF networks with 244 sequence-altering ISFs to better understand this effect and its prevalence but also to assign ISF 245 peptides to new peptide groups. Merging information from peptides with multiple charges, we 246 observed that 72.4 % of the ISF networks had two unique peptide sequences and 72.5 % a one-247 to-one ratio between in-source fragment and parent peptides (Fig. 3.e), implying that the 248 majority of ISFs were from a single ISF parent to a single fragment peptide. However, the 249 remaining 27.6 % of the ISF networks contained more than two unique peptide sequences, with 250 the largest ISF network having 23 unique peptide sequences (Supplementary Fig. 6). 27.3 % of 251 the ISF networks also had an imbalance between the number of in-source fragments and parent 252 peptides (Fig. 3.e). Detecting both N- and C-terminal fragments from a single ISF event appeared 253 to be challenging as we only observed 9 cases across the 23 datasets, in which the combined 254 sequences of two in-source fragments matched the parent sequence. 255 These results show that peptide ISF mostly occurs at specific peptide bonds prone to fragmenting 256 in the gas phase and often creates a complex pattern of peptide products and m/z features, which 257 can be deciphered by network analysis. 258 259 In-source fragmentation of PTMs 260 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 10 If ISF only removes the PTM group of a peptide, the resulting fragment peptide can appear as an 261 unmodified peptide, which could be mistaken for an authentic identification of the unmodified 262 peptide. We analyzed how often this effect occurs for several PTMs. 263 Methionine oxidation and N-terminal acetylation are commonly observed PTMs in proteomics 264 data; we included them as variable modifications in the peptide searches of our combined 265 dataset and studied if they undergo ISF. We detected that ISF of these PTMs comprised 5.5 % of 266 the total ISF events (Fig. 3.a) and that the intensity differences between in-source fragments and 267 parents were much smaller for PTM ISF than for sequence-altering ISF (Supplementary Fig. 4.b). 268 Most of the cases involved methionine oxidation, and only a few N-terminal acetylation (0.4 % of 269 all PTM ISF) (Fig. 4.a). While there was a large variation in the number of methionine oxidation 270 ISF between the datasets, we did not observe a correlation with the number of peptides carrying 271 a methionine oxidation (Fig. 4.b). This indicated that the amount of PTM ISF was dependent on 272 other factors, e.g. instrument parameters. 273 It is known that phosphorylated metabolites like ATP tend to lose phospho-groups due to in-274 source fragmentation 11, and we thus assessed the prevalence of ISF events affecting 275 phosphopeptides. We first directly re-searched our S. cerevisiae S288c dataset for 276 phosphopeptides. Second, we analyzed S. cerevisiae S288c digests after phosphopeptide 277 enrichment by titanium dioxide chromatography 45, and third, we included two additional, 278 published datasets that used various phosphopeptide enrichment strategies (PXD014525 46 and 279 PXD04448247). As expected, the number of phosphopeptides were much higher in the three 280 phospho-enriched datasets, where they made up 68 - 91 % of all identified peptides as compared 281 to 0.6 % in the standard proteomics dataset (Fig. 4.c). Similarly, the number of ISF events leading 282 to loss of a phospho-group was higher in the phosphoproteomics datasets (0.17 - 0.63 % of in-283 source fragments by identifications, 0.46 – 1.06 % by intensity, mean values, number of samples 284 n = 4 (phospho-enriched yeast dataset), n = 18 (PXD014525), and n = 10 (PXD044482)) than in 285 the standard proteomics dataset (0.0048 ± 4e-5 % by identifications, 0.0037 ± 1e-4 % by intensity, 286 mean and standard deviation, n = 4 technical replicates) (Fig. 4.d). In the two published datasets, 287 we observed a bias of phospho-group ISFs towards phosphorylated tyrosines (Y) indicating that 288 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 11 this amino acid is more prone to phospho-group ISF than phosphorylated serines (S) and 289 threonines (T) (Supplementary Fig. 7). 290 In conclusion, PTM ISF occurs at a detectable level in proteomics data and varies in a dataset and 291 PTM-dependent manner, although it was less prevalent than ISF of regular peptide bonds. 292 293 Peptide ISF is instrument and parameter dependent 294 The ion funnel design and its radiofrequency (RF) are known to have an impact on the ISF of 295 peptides48, the temperature of the ion transfer capillary and the RF affect lipid ISF 49, and 296 increasing accumulation times during trapped ion mobility spectrometry (TIMS) time-of-flight 297 mass spectrometry (TOF) also increases ISF50. Based on this information, we expected ISF to vary 298 between different mass spectrometers. To assess the extent of such variation, we measured a 299 trypsin-digested yeast reference sample mixed with 8 pure proteins on three Thermo Scientific 300 mass spectrometers, an Orbitrap Exploris 480, an Orbitrap Astral, and a Q Exactive Plus Hybrid 301 Quadrupole-Orbitrap (QE+), each using LC and DIA methods based on previously established 302 workflows51,52. With 1.07 ± 0.02 and 1.51 ± 0.02 % (mean and standard deviation, n = 4 technical 303 replicates), the QE+ and the Astral mass spectrometers yielded a much lower fraction of peptide 304 identifications corresponding to in-source fragments than the Exploris 480 (5.68 ± 0.03 %) 305 (Supplementary Fig. 8). This was also reflected in the total intensity of in-source fragments (QE+: 306 2.2 ± 0.3 %, Exploris 480: 8.8 ± 0.2 %, Astral: 3.70 ± 0.05 %). Thus, ISF was indeed highly dependent 307 on the instrument. 308 As the RF and the ITC temperature were shown to affect ISF48,49, we subsequently quantified the 309 impact of these two parameters as well as the electrospray voltage on peptide ISF using the 310 Orbitrap Exploris 480. We used the yeast reference sample again and measured it at a spray 311 voltage fixed at 2500 V with seven variations of the funnel radio frequency and six variations of 312 the ITC temperature. We also measured the same sample with a relative funnel radio frequency 313 fixed at 50 % with six variations each of the other two parameters (ITC temperature and spray 314 voltage), which amounted to a total of 72 unique sets of parameters tested. 315 The chosen parameters resulted in a broad range of ISF, from low (0.2 % of peptide precursor 316 identifications across all samples, 301 fragments at 100 °C, 2500 V, 10 % RF, n = 1 replicate) to 317 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 12 high (4.9 %, 6566 fragments at 350 °C, 2500 V, 60 % RF) (Fig. 5.a and Supplementary Fig. 9.a). 318 Increasing the funnel RF increased both the relative intensity and the number of in-source 319 fragments (Fig. 5.b). Simultaneously, it only increased the overall signal up to a RF of 40 %, 320 indicating that higher funnel RF values are disadvantageous. The effect of the ITC temperature 321 on ISF was dependent on the funnel RF and led to increases in ISF (in signal and identifications) 322 at high RF values but had little effect on ISF at low RF values. In contrast to both other parameters, 323 the electrospray voltage had little effect on overall signals, number of peptide identifications, 324 and ISF, except for the very low setting of 1500 V that decreased overall signal drastically. 325 The number of precursor identifications is often used as a quality control measure during data 326 acquisition. Because instrument parameters increase the number of in-source fragments while 327 decreasing the number of identifications of genuine peptides with no evidence of fragmentation, 328 not accounting for ISF will lead to an overestimation of identifications and can thus lead to 329 choosing suboptimal conditions (Fig. 5.a). 330 In conclusion, these results show that ISF was strongly dependent on the instrument, funnel RF, 331 and ITC temperature, indicating that proteomics experiments can be optimized to avoid high ISF 332 rates. Since ISF can lead to an overestimation of the total peptide identifications, minimizing it 333 should generally be part of method optimization efforts alongside the optimization of other 334 parameters like total peptide identifications and signal intensities. 335 336 ISF does not impair relative quantification 337 To test if ISF has an impact on relative quantification of peptide precursors, peptide groups, and 338 protein groups, we conducted an experiment in which we mixed, at equal volumes, a yeast 339 sample at a fixed concentration with 8 samples each with a different concentration of E. coli 340 peptides (Fig. 6.a). Peptide precursor fold changes were calculated relative to the sample with 341 the highest E. coli proteome concentration. 342 For all samples, the yeast proteome remained stable around a fold change of 1, whereas the E. 343 coli peptide precursors followed the dilutions (Fig. 6.b). At lower concentrations of the E. coli 344 proteome, the peptide precursor fold changes did not match the theoretical values as accurately 345 as at higher concentrations, which was probably due to ion suppression and poor peptide 346 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 13 detectability. However, comparing the data between non-ISF, fragment, and parent peptide 347 precursors revealed that the non-ISF peptide precursors had broader fold change distributions 348 than the other peptide precursors (Fig. 6.b and Supplementary Fig. 10) and that coefficients of 349 variation (n = 3 technical replicates) of non-ISF peptide precursors were often larger than those 350 of in-source fragments and parents (Fig. 6.c, left plot, P < e-200 for all pairwise comparisons, 351 Wilcoxon rank sum test, two-tailed). Because the dilution factors were known, we could calculate 352 a relative error between the measured fold changes of the precursor peptide quantities and the 353 expected, theoretical fold changes. These relative errors showed that the non-ISF precursor 354 peptides recovered the expected, theoretical values better than the in-source fragments by only 355 a small margin (0.015 difference in the median relative error); interestingly, the ISF parents 356 recovered the theoretical values the best, which is probably due to their generally good 357 measurability and high signals (Fig. 6.c, right plot, P < e-13 for all pairwise comparisons, Wilcoxon 358 rank sum test, two-tailed). These results show that ISF has little effect on relative quantification 359 of peptide precursors. 360 So far, our analyses have been at the peptide precursor level but most applications in proteomics 361 rely on peptide or protein level analyses, in which data of peptide precursors that only differ in 362 charge or even PTMs are combined to improve quantification and reduce redundancy. Since ISF 363 only had a minor impact on relative peptide precursor quantification, we tested if assigning in-364 source fragments to their parental peptide groups or protein groups provide an advantage, e.g. 365 by reducing errors, or if in-source fragments should simply be filtered out. 366 First, we used network analysis to group in-source fragments with their respective ISF parents, 367 and calculated peptide quantities by summing all peptide precursors intensities of each peptide 368 group. Subsequently, we compared the peptide groups of ISF parents with and without added in-369 source fragments as well as all the peptide groups unaffected by ISF with each other. The CVs of 370 peptide groups that contain in-source fragments were smaller than the same peptide groups 371 without in-source fragments, although the difference was minor (Fig. 6.d, P = 0.013, Wilcoxon 372 rank sum test, two-tailed). The absolute relative errors showed that parent peptides (without 373 merging with in-source fragments) much better followed the theoretical fold changes than non-374 ISF peptides (Fig. 6.d). However, these data also revealed that merging in-source fragment data 375 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 14 with their parental peptide groups worsens quantification ( P = 2.6e-14, Wilcoxon rank sum test, 376 two-tailed) but again the effect is small. 377 Second, we calculated protein-level quantities by summing up all peptide precursors that belong 378 to a certain protein, once with in-source fragments and once without them. Comparing the CVs 379 and relative errors revealed that in-source fragments had close to no effect on the protein-level 380 quantities (Fig. 6.e, P > 0.4, Wilcoxon rank sum test, two-tailed). 381 Taken together, these results show that ISF has little effect on relative quantification, that in-382 source fragments and parent peptides are among those peptides that are usable for relative 383 quantification, and that ISF parents were the best peptides for quantification. These findings are 384 likely because in-source peptides have typically higher intensities than non-ISF peptides. 385 Regrouping in-source fragments with their parent peptides or including them in protein groups 386 also had little effect, with a small tendency to reduce CVs of the new peptide groups but at the 387 cost of distorting relative quantities across samples. In sum, while in-source fragments present 388 no major challenge for quantitative analyses in general, they are only a proxy of their parent 389 peptides, with slightly worse quantitative performance (CVs and relative errors), and we 390 therefore recommend filtering them out. 391 392 Impact of ISF on immunopeptidomics 393 Since ISF of tryptic peptides produces mostly peptides with semi-tryptic sequences 394 (Supplementary Fig. 4), we wondered how ISF would impact peptide-centric proteomic 395 approaches relying on relaxed trypsin specificity searches, such as immunopeptidomics. During 396 the immune response, peptides produced by proteolytic in vivo-processes are presented to killer 397 T cells as antigens25,53. Identifying those antigens is crucial for a better understanding of immune 398 responses but also for the development of new medical treatments. Misidentification of an in-399 source fragment as true biological antigen can thus have costly consequences. The peptide 400 antigens show narrow length distributions, typically between 9 and 14 amino acids, depending 401 on the peptide class, and they do not show specific terminal amino acids 54. Therefore, 402 immunopeptidomics requires unspecific searches for peptide identification, which causes an 403 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 15 exponential increase of the search space and increases the risk of confusing in-source fragments 404 with biologically relevant peptides. 405 Here, we used four datasets from the literature (PXD022950, PXD051490, PXD054417, and 406 PXD05880)55–58 to assess the impact of ISF on immunopeptidomics. Across the datasets, we 407 observed large variation in the number of in-source fragments (Fig. 7.a), and similarly to our 408 previous results, most ISFs occurred at the N-terminus (Fig. 7.b). As expected, the length 409 distribution of peptides from immunopeptidomes was much narrower than in our data of trypsin-410 digested proteomes, but in-source fragments were again typically 1 to 2 amino acids shorter than 411 parent peptides (Fig. 7.c). In the datasets PXD058880 and PXD022950, in-source fragments made 412 up 16.6 and 36.8 % of all peptides below 9 amino acids (2.6 and 2.3 % by intensity) but only 0.07 413 and 0.92 % of longer peptides (0.04 and 0.8 % by intensity), respectively. The other datasets that 414 we analyzed (PXD054417 and PXD05149) had much less ISF; the fragment identifications were 415 again higher for peptides below 9 amino acids (0.53 and 0.90 % by identifications, 0.003 and 0.17 416 % by intensity) than for longer peptides (0.11 and 0.07 % by identifications, 0.06 and 0.01 % by 417 intensity). 418 These results show that, in immunopeptidomics based on HLA-I peptides, ISF is likely to impact 419 peptides shorter than 9 amino acids. Therefore, a simple strategy to mitigate this would be to 420 exclude peptides shorter than 9 amino acids before further analyses. Alternatively or in addition, 421 application of our RT-based detection algorithm to immunopeptidomics data will enable 422 identifying in-source fragments of any length that contaminate the pool of biologically relevant 423 peptides. 424 425 In-source fragmentation in limited proteolysis data 426 Limited proteolysis coupled with mass spectrometry (LiP-MS) is another peptide-centric 427 proteomic approach, which enables the global analysis of protein structural changes in proteome 428 extracts26. It relies on proteases with broad specificity that cleave proteins for a brief period of 429 time such that structural properties govern cleavage events. These primary cleavages can 430 produce large protein fragments that may not be directly amenable for MS and are thus further 431 digested by trypsin under denaturing conditions in a second step. LiP-MS therefore intrinsically 432 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 16 features many more semi-tryptic peptides than other proteomic datasets. As most in-source 433 fragments appear as semi-tryptic peptides based on our analyses, we asked how many of the 434 identified semi-tryptic peptides in a typical LiP experiment were due to ISF and therefore could 435 mistakenly be identified as structurally relevant peptides. 436 We used three published LiP datasets to study this (PXD022297, PXD033826, and PXD034606)59–437 61 and found that 0.6 % to 2.0 % of all identified peptides were due to ISF (1.3 to 3.6 % of the total 438 intensity) (Fig. 7.d). However, the in-source fragments made up only 0.9 to 3.7 % of the semi-439 tryptic peptides (2.1 to 5.9 % by intensity), showing that most semi-tryptic peptides were indeed 440 true-positives. In these LiP datasets, only 7.9 to 18.1 % of the in-source fragments appeared were 441 non-tryptic (Fig. 7.f), corresponding to 0.08 – 0.12 % of all identifications (0.07 – 0.14 % by 442 intensity). 443 These results show that ISF had only little impact on LiP datasets as the vast majority of semi-444 tryptic peptides were indeed caused by limited proteolysis and not by ISF. While the inherently 445 high amounts of semi-tryptic peptides in LiP data did yield a small number of in-source fragments, 446 detection of these peptides requires unspecific searches, which are typically not performed for 447 standard LiP datasets. Overall, our data indicates that ISF is unlikely to be an issue for LiP-MS 448 analyses. 449 450 Current peptide-centric DIA analyses underestimate the extent of ISF 451 Our analyses so far focused on DIA datasets because DIA offers higher confidence than DDA in 452 assessing co-elution of peptides and therefore in detecting ISF. DIA data however come with an 453 important caveat: most peptide-centric DIA algorithms assume that a peptide elutes as a single 454 peak at a specific retention time. However, if ISF occurs, it is possible that a peptide has more 455 than one peak because it could stem from ISF in addition to a biological source that gives rise to 456 a genuine peptide. Therefore, current search engines will either miss the identification of a 457 genuine peptide or underrepresent ISF in cases in which a peptide has multiple peaks 458 (Supplementary Fig. 11.a and b). In contrast to current DIA analysis approaches, DDA analysis 459 does not make any assumption regarding the elution of peptides but reports all RTs at which a 460 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 17 sequence is identified, although RT determination can be imprecise due to dynamic peak 461 exclusion settings (Supplementary Fig. 11.C). 462 To understand the extent to which current “single-peak” peptide-centric DIA algorithms 463 underestimate ISF, we analyzed DIA datasets using an approach that relies on modified DDA 464 libraries62 (also see Methods part M.5). In brief, peptides that were identified in multiple DDA 465 windows (each two or more minutes apart) were assigned to different bins of spectral assays 466 using unique identifiers. Searches of DIA data with such modified spectral libraries would thus 467 retain assays of the same peptide that are separated by RT and allow analysis of multiple peaks. 468 We used this strategy to generate a modified spectral library from the DDA files acquired for the 469 LiP dataset PXD022297 59, which contains many genuine semi-tryptic peptides due to limited 470 proteolysis. The library had 44,046 peptide precursor assays and included 2,961 (6.7 %) assays of 471 precursors that were present in more than one RT window. Next, we used this library to search 472 the DIA data from the same dataset and compared the library-based peptide peak identifications 473 to those obtained by a “direct” peptide search of the DIA data, which did not rely on a spectral 474 library from DDA. This revealed that the library-free DIA extraction identified 1,505 (57 %) 475 precursors of the 2,961 assays with multiple peaks from our modified spectral library at the 476 expected RT, misidentified 940 assays at a different RT, and completely missed 211 assays. Using 477 the ISF parent and fragment annotations of the library-free DIA extraction showed that the 478 (direct) peptide-centric DIA analysis underestimated the number of ISF peptides by 218 and the 479 number of genuine peptide sequences by 933. 480 Using a spectral library for DIA extraction introduces the usual biases of DDA measurements, such 481 as undersampling low abundant signals or underrepresenting singly charged precursors. 482 Therefore, our results on the multiple peak extractions still present an underestimation of ISF 483 and the true number of cases in which a peptide sequence occurs multiple times in a gradient. 484 However, this analysis shows that current peptide-centric DIA analysis tools that only account for 485 a single RT per peptide sequence systematically underestimate both the extent of ISF as well as 486 the number of genuine peptides, which are not created by ISF. These results thus encourage the 487 development of peptide-centric algorithms that account for multiple peaks. 488 489 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 18

490 It is known that ISF affects peptides and can be a source of peptides with non-tryptic termini in 491 bottom-up proteomics data21. The proteomics field has evolved in past years, with an increasing 492 use of peptide-centric proteomics approaches like structural proteomics and 493 immunopeptidomics that rely on semi- and non-tryptic peptides. Further, DIA methods that 494 enable detection of low intensity m/z features, as opposed to DDA methods that likely miss many 495 low intensity fragments, have found mainstream adaptation, and the development of the new 496 mass spectrometers with vastly increased sensitivity has introduced new instrument geometries 497 and designs that could affect the prevalence of ISF. It is therefore timely to conduct an 498 assessment of ISF in the context of DIA proteomics, with a particular focus on peptide-based 499 approaches. 500 We described an approach to re-annotate peptides based on RT patterns. The novelty of our 501 approach is to use intra-sample information from peptides with multiple charge states to 502 estimate natural distributions of ∆RT values and, based on these, determine ∆RT cutoffs for ISF 503 detection. This approach can therefore account for differences in LC performance between runs, 504 batches, or HPLCs. In the future, our ISF detection could even be further improved by using 505 dynamic ∆RT cutoffs that depend on the RT of the peptide peaks. This could address dynamically 506 varying ∆RT value distributions in non-linear gradients. 507 We observed that ISF can affect a large share of a proteomics dataset acquired by DIA and 508 account for more than 50 % of the detected semi-tryptic peptides. We also found that the 509 numbers of detected in-source fragments anticorrelate with sample complexity; this has 510 implications for MS approaches such as crosslinking MS, hydrogen/deuterium exchange mass 511 spectrometry (HDX-MS), affinity purification coupled to MS, or (co-)fractionation MS that often 512 rely on the analysis of low complexity samples63–66. These observations match well with previous 513 reports based on DDA data21. Further, our analysis of the intensity distributions showed that the 514 observed ISF parents often had very high intensities resulting also in fragments with higher 515 intensities than of non-ISF peptides. This observation could be the result of a bias of the search 516 engine against in-source fragments with low intensities. 517 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 19 Generally, we observed a large variation in the amount of detected ISF among different datasets, 518 which can be explained by different samples complexities, instruments, and chosen parameters, 519 all of which we have shown here to have an impact on the prevalence of ISF. The combinatorial 520 effect of all these factors makes it difficult to estimate whether ISF could be a problem for a 521 particular sample or dataset and, considering that ISF can in some cases account for more than 522 30 % of the peptide identifications, we suggest that ISF products should be detected and 523 annotated as a routine data analysis step. 524 The current peptide-centric DIA analysis tools assume that a peptide sequence occurs only once 525 across a chromatogram. Using a RT-based splitting strategy to generate spectral libraries from 526 DDA data, we were able to analyze peptides occurring multiple times across the LC gradient in 527 DIA data. Our analysis showed that both the extent of ISF and the number genuine, non-ISF 528 peptides were underestimated depending on the peak selection. Underestimating ISF might not 529 present a problem for typical applications but, whenever an in-source fragment is identified, it 530 can come at the cost of losing an identification of a genuine peptide that shares its sequence with 531 the in-source fragment. With current peptide search strategies for DIA data, ISF can thus mask 532 the identification of genuine peptides and lead to loss of information. Furthermore, for peptides 533 with high signals, there is a reasonable chance that an in-source fragment can be observed in the 534 data. This effect could be used in future peptide search engines to improve peptide identification 535 and estimation of false-discovery rates while increasing overall annotations of proteomics data. 536 In immunopeptidomics, peptides are not generated by in vitro digestion and are often identified 537 by unspecific peptide searches. Therefore, immunopeptidomics data could be much more prone 538 to interference by ISF, which often generates peptides with non-tryptic termini. Our results show 539 that in immunopeptidomics data with HLA-I peptides, most in-source fragments are shorter than 540 9 amino acids and can account for more than a third of all peptides of such short length. This 541 raises the question of how many of the immunopeptides shorter than 9 amino acids are truly 542 authentic peptides because current approaches still underestimate how many of the detected 543 peptides are in-source fragments. We recommend using our ISF detection to minimize the risk of 544 falsely identifying peptides. Alternatively, or in addition, immunopeptides smaller than 9 amino 545 acids can be filtered out to strongly reduce the impact of ISF on the data. 546 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 20 As in-source fragments have the same RT as their parent peptides, the RT of the in-source 547 fragment does not correctly reflect the peptide separation by chromatography and is different 548 from the RT of a non-ISF peptide with the same amino acid sequence. Peptide ISF can therefore 549 have an impact on the RT realignment during peptide searches. Depending on the MS approach, 550 the strength of this effect is likely to vary, and will be more impactful in cases where a few 551

peptides are chosen for realignment, like analyses by single reaction monitoring (SRM), 552 or when the prevalence of ISF is particularly high. 553 We observed little PTM ISF for phosphorylation, oxidation of methionine, and N-terminal 554 acetylation, but we cannot exclude that other PTMs such as glycosylation, which is known to be 555 prone to ISF 67, might be affected to a much larger extent. Some PTMs might even be so labile 556 that we only detect the in-source fragments, which would require exceptionally mild ionization 557 and ion routing approaches to study them. Sequence-altering ISF of parent peptides that carry a 558 PTM could also explain some of the detected semi-tryptic peptides in typical trypsin-digested 559 bottom-up proteomics samples, which have not been associated with an ISF parent so far, 560 because many peptides with PTMs are very challenging to detect in non-enriched datasets by 561 current search engines. Other cases of ISF that have not been addressed so far are those in which 562 fragmentation occurs not at peptide bonds (between C1 and N2) or at PTM-peptide bonds but 563 rather at side chains. Detecting these cases presents a major challenge but potentially could 564 explain many of the detected but uncharacterized m/z features in proteomics data and help 565 separate these from more biologically meaningful uncharacterized portions of the proteome 68. 566 Moreover, in-source chemical reactions that add to the mass of a molecule have been observed 567 for metabolites before at a high prevalence 13 and, in principal, could also affect peptides. Taken 568 together, this poses the question: how many of the m/z features in a proteomics dataset that are 569 unannotated can be explained by ISF or in-source reactions? Answering this question will result 570 in new peptide groups without annotation, for which the mode of fragmentation or reaction 571 could provide insights into the identity of the parent peptides. 572 In summary, in-source fragments inflate the number of identifications and, especially in low 573 complexity samples, can make up most of the signal and lead to data misinterpretation. Further, 574 since ISF varies substantially dependent on numerous parameters, predicting the impact of these 575 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 21 fragments a priori in a given analysis is challenging. Based on our findings in this study, we thus 576 conclude that ISF should be identified by default in current proteomics approaches and the in-577 source fragments filtered out bioinformatically. 578 579 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 22 Figures 580 581 Fig. 1. Illustration of the ISF detection algorithm 582 a, Graph showing hypothetical chromatograms of different peptide precursors that share amino acid sequences and 583 thus form peptide pairs. The axes display the retention time, the m/z, and the signal intensity as indicated. The 584 peptides form three coeluting groups. 585 b, Heatmap illustrating all pairwise combinations of peptides from the example in A and their respective ∆RTs. The 586 ∆RT is calculated by subtracting the apex RT of the shorter peptide from that of the longer peptide. Peptide pairs 587 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 23 that do not show co-elution and have ∆RTs larger than a cutoff are blue. Peptide pairs with evidence of ISF are red. 588 Pairs of peptides that only differ in charge are yellow. 589 c, Diagram illustrating a theoretical ∆RT distribution of peptides pairs that only differ in charge. An observed ∆RT 590 distribution (blue line) is the sum of two sub-distributions: a normal distribution of true-positive peptide pairs (blue 591 shade) centered around zero and a uniform distribution of false-positive peptide pairs (red shade), of which at least 592 one peptide per pair is falsely annotated. The ∆RT cutoff is the ∆RT at which the value of the normal distribution 593 equals the uniform distribution. 594 d, Graph showing a theoretical ∆RT distribution of all peptide pairs that are not part of C. The ∆RT cutoff is used to 595 classify peptide pairs into those with ISF evidence, if they fall within the cutoff, and those without ISF evidence. 596 e, Illustration of peptide annotation: peptide pairs with ISF evidence are deconvoluted, and peptide precursors 597 assigned as in-source fragments or parent peptides. Peptides that show ISF often form complex fragmentation 598 networks, in which edges indicate ISF evidence (labelled with the respective ∆RT) and nodes are peptides. 599 Fragmentation networks can be used to create new peptide groups. Amino acid sequence-altering ISF can occur N-600 terminally, C-terminally, or at both ends (omitted in the illustration). ISF of exclusively PTM groups are examples of 601 amino acid sequence-conserving ISF. 602 603 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 24 604 Fig. 2. Evidence for abundant ISF across proteomics data of different complexity 605 a, Dot plot showing the share of in-source fragments (% by intensity) per number of detected protein groups 606 (minimum of 5 unique peptide sequences per group) in 317 samples across 16 published proteomics datasets (as 607 indicated by color and symbol shape). 608 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 25 b, Dot plot showing the share of in-source fragments (dots: % by intensity, squares: % by identifications) per number 609 of identified protein groups (minimum of 5 unique peptide sequences per group) across 7 datasets (as indicated by 610 colors) that ranged from a mix of 11 peptides up to a mix of three whole proteomes. Dots and squares are mean 611 values (n = 4), vertical lines indicate standard deviations. 612 c, Bar graphs showing the number of identifications and the absolute intensity of in-source fragments, ISF parents, 613 and peptides without ISF evidence for the datasets shown in B. Bars indicate mean values (n = 4), vertical lines 614 standard deviations. 615 d, Bar plot showing the percentage of total intensity covered by in-source fragments and parents combined. 616 e, Pie chart showing the share (%) of semi- and fully-tryptic peptides among in-source fragments across all the 617 datasets from A. 618 f, Bar plot depicting the average share of semi-tryptic in-source fragments (%) among all detected semi-tryptic 619 peptides per dataset from A. Blue bars are intensity-based representations (yellow bars: number of identifications-620 based). Bars indicate mean values (n = 4), horizontal lines the standard deviation. 621 622 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 26 623 Fig. 3. ISF peptide characterization 624 a, Three pie charts illustrating: (1) the fractions of peptides with and without peptide pairing (in %), (2) the number 625 of peptide pairs with and without evidence of ISF (%), and (3) the number of peptide pairs with ISF evidence that 626 have a sequence-altering ISF, are fragment-to-fragment pairs, and feature PTM ISF (%). The data used for this figure 627 stems from 23 datasets (also see Fig. 2) that were merged. 628 b, Graph showing the z-scored (modified) log 2-intensity distributions (in %) of in-source fragment (red) and parent 629 (yellow) peptides as well as peptides without ISF evidence (blue). Box whisker plots above the distributions indicate 630 the median values and interquartile ranges. Modified z-scores are calculated with median values and median 631 absolute deviations. 632 c, Dot plot showing the abundance of amino acids at the F1’ position at ISF sites over occurrences at F1. 633 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 27 d, Heatmap showing the adjusted P-values from an enrichment analysis (Fisher’s exact test, two-tailed, P-value 634 adjustment by Benjamini-Hochberg procedure 43) of all peptide bonds observed at IFS sites. Blue squares indicate a 635 negative odds ratio, yellow a positive ratio. 636 e, Histograms showing the number of ISF networks (in %) with the indicated number of unique peptides per network 637 (left) or with the indicated in-source fragment to parent ratio (right). 638 639 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 28 640 Fig. 4. PTM in-source fragmentation 641 a, Bar plot showing the number of peptide pairs with a PTM ISF (as labelled) among all pairs with ISF evidence from 642 23 combined datasets (also see Fig. 3.). 643 b, Dot plot showing the number of ISF of methionine oxidation (%) per number of precursors with a single 644 methionine oxidation (%). Colors and symbols indicate 23 different datasets. 645 c, Bar plot showing the percentages of phosphopeptides (red) and other peptides (blue) in four different datasets. 646 The two S. cerevisiae datasets were acquired from the same sample source: one as a standard yeast proteomics 647 dataset and the other as phosphoproteomics dataset (including phosphopeptide enrichment). The two literature 648 datasets (PXD014525 and PXD044482) relied on phosphopeptide enrichment. 649 d, Bar graph displaying the number of in-source fragments (% by identifications) in the four datasets of C. 650 651 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 29 652 Fig. 5. Impact of the funnel RF, electrospray voltage, and ITC temperature on peptide ISF. 653 a, Bar plot displaying the number of detected peptide precursors that are non-ISF and parent peptides (blue bars) 654 and that are in-source fragments (red bars) (% of all identifications across all samples) in a yeast reference sample 655 acquired at different funnel radiofrequencies (%), electrospray voltages (V), and ITC temperatures (°C) (n = 1, except 656 5 pairs of technical replicates as indicated). The samples are sorted from left to right in ascending order of increasing 657 ISF. 658 b, Dot plots showing the relative intensity (%) (upper left ), the total intensity (lower left), the relative number of 659 identifications (%) (upper right ), and absolute number of identifications (lower right ) of in-source fragments in the 660 samples from A. 661 662 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 30 663 Fig. 6. Impact of ISF on quantification 664 a, Whole proteome digests of S. cerevisiae S288c and E. coli BW25113 were mixed at different ratios as indicated as 665 dilution factors below the blue bars. The yeast proteome concentration was kept constant. The E. coli proteome 666 concentration was varied, and volumes compensated with water. 667 b, Dot plots showing the mean log 2-precursor fold changes (relative to the sample with the undiluted E. coli 668 proteome) per mean log2-precursor intensity (n = 3 technical replicates). Red dots indicate in-source fragments, blue 669 dots ISF parents, and grey dots non-ISF peptides. 670 c, Box whisker plot showing the median coefficient of variation (CV, %, n = 3 technical replicates) and the median 671 relative error between theoretical and measured fold changes (n = 3 technical replicates) of all peptide precursors 672 from the samples in B. Data of in-source fragments is indicated in red (parents: blue, non-ISF precursors: grey). Boxes 673 depict interquartile ranges, and whiskers extend to 1.5-fold of the interquartile ranges. 674 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 31 d, Box whisker plot showing the median coefficient of variation (CV, %, n = 3 technical replicates) and the median 675 relative error between theoretical and measured fold changes (n = 3 technical replicates) of all peptide groups from 676 the samples in B. Peptide groups of all ISF parents without (blue) and with (yellow) merging with their respective in-677 source fragments are compared with all peptide groups without ISF evidence (green). Boxes depict interquartile 678 ranges, and whiskers extend to 1.5-fold of the interquartile ranges. P-values were calculated using the two-tailed 679 Wilcoxon rank sum test. Peptide quantities were calculated by summing up precursor intensities. 680 e, Same as D but depicting protein groups that contain (yellow) or omit (green) in-source fragments. Protein 681 quantities were calculated by summing up precursor intensities. 682 683 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 32 684 Fig. 7. ISF in immunopeptidomics and limited proteolysis datasets 685 a, Bar plot showing the average percentage of in-source fragments in immunopeptidomics datasets (PXD058880: n 686 = 12, PXD054417: n = 12, PXD022950: n = 6, PXD051490: n = 12), by peptide intensities and numbers of 687 identifications. 688 b, Bar plot showing the number of precursor pairs with evidence of ISF (per dataset in A) (%) that feature N-terminal 689 ISF, C-terminal ISF, or ISF at both termini. 690 c, Dot plots displaying the number of peptides (%, normalized per group) of different lengths (number of amino 691 acids) in each dataset of A. Non-ISF peptides are indicated in blue (ISF parents: yellow, in-source fragments: red). 692 d, Bar graph showing the mean fraction of in-source fragments within each indicated LiP dataset (yellow bars: % by 693 intensity, blue bars: % by the number of identifications) (n = 3). Data results from trypsin unspecific peptide searches. 694 e, Bar plot showing the percentage of in-source fragments among semi-tryptic peptides (green bars: % by intensity, 695 violet bars: % by identifications) in the indicated LiP datasets. 696 f, Bar plot showing the percentage of in-source fragments by their apparent trypticity in the indicated LiP datasets. 697 Red bars indicate in-source fragments that appear as non-tryptic peptides (yellow bars: semi-tryptic in-source 698 fragments, blue bars: fully-tryptic in-source fragments). 699 700 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 33

701 M.1 Sample preparation for MS-based proteomics 702 M.1.1 Saccharomyces cerevisiae S288c whole-proteome sample 703 S. cerevisiae S288c (ATCC #204508) was streaked from cryo stock onto an YPD agar plate and 704 incubated at room temperature. After single colonies became visible, the plate was stored at 4 705 °C. 25 mL synthetic defined (SD) medium containing 20 g/L D-glucose (Merck, Sigma-Aldrich 706 #G7021), 5 g/L ammonium sulfate (Fluka #09982), and yeast base (Merck, Millipore #Y1251) was 707 inoculated from a single colony and incubate in 100 mL Erlenmeyer flasks at 30 °C under shaking 708 of 200 rpm for circa 20 h. 40 mL SD medium (2 w/v-% glucose) in 500 mL Erlenmeyer flasks was 709 inoculated from the previous 25 mL-culture to a start optical density (OD at 600 nm) of 0.004 and 710 incubated at 30 °C under shaking at 200 rpm. 300 mL SD medium (2 w/v-% glucose) in 2 L 711 Erlenmeyer flasks were inoculated from the previous 40 mL-culture to a start OD of 0.08 and 712 incubated at 30 °C under shaking at 200 rpm. The OD was measured regularly to calculate growth 713 rates and exclude growth defects. At a final OD of circa 0.6, samples for proteomics were 714 collected using a method that is analogous to a phosphoproteomics sampling approach 45 and 715 relies on quenching enzyme activity by trichloroacetic acid (TCA): 275 mL of culture were 716 transferred to 500 mL harvesting tubes on ice. 18.3 mL 4 °C-cold 100 w/v-% TCA (final 717 concentration 6.2 % TCA; Merck, Sigma-Aldrich, #91228) was added. The cell suspension was 718 incubated for 10 min on ice and, subsequently, centrifuged for 5 min at 4 °C and 3428 g. The 719 supernatant was discarded, and the cell pellet resuspended in 40 mL 4 °C-cold acetone (Merck, 720 Supelco, #1.00014). The suspension was transferred to 50 mL reaction tubes and centrifuged for 721 5 min at 4 °C and 3,428 g. The supernatant was discarded, the cell pellet resuspended in 40 mL 4 722 °C-cold acetone, and the suspension centrifuged for 5 min at 4 °C and 3,428 g. The supernatant 723 was discarded, and the cell pellet resuspended in 5 mL 4 °C-cold acetone. The cell suspensions of 724 4 independent biological replicates (= 4 different colonies) were pooled and distributed to 1.5 mL 725 screw cap-reaction tubes. After 5 min of centrifugation at 3,400 g and 4 °C, the supernatant was 726 discarded, and the remaining cell pellets flash frozen in liquid nitrogen and stored at -80 °C. Cell 727 pellets were resuspended in 500 µL lysis buffer, which was 8 M urea (Merck, Empure Essential, 728 #1.08486.1000) 100 mM ammonium bicarbonate (Merck, Sigma-Aldrich, #A6141) at pH 7.8. After 729 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 34 adding an equal volume of glass beads (0.5 mm diameter, Merck, Sigma-Aldrich # G8772), cells 730 were lysed at 4 °C using 8 cycles of bead beating at 6.5 m/s for 30 s with 200 s pause. After bead 731 removal, lysates were centrifuged for 15 min at 17,000 g and 4 °C, and all supernatants pooled 732 for further processing. After tryptic digest and peptide clean-up, the reference yeast sample was 733 obtained by mixing 800 µL of the S. cerevisiae S288c sample with 200 µL of the mix of 8 proteins 734 sample. 735 736 M.1.2 Escherichia coli BW25113 whole-proteome sample 737 E. coli BW25113 (DSMZ, #27469) was streaked from cryo stock onto a LB agar plate and incubated 738 at 37 °C overnight. The plate was stored at 4 °C. 25 mL M9 minimal medium was inoculated from 739 a single colony and incubate in 100 mL Erlenmeyer flasks at 37 °C under shaking of 200 rpm 740 overnight. M9 minimal medium contained: 5 g/L D-glucose (Merck, Sigma-Aldrich, #G7021), 6 g/L 741 Na2HPO4 (Carl Roth, #P030.2), 3 g/L KH 2PO4 (Carl Roth, #3904.1), 0.5 g/L NaCl (Carl Roth, 742 #9265.1), 1.5 g/L (NH 4)2SO4 (Merck, Sigma-Aldrich, #A3920), 1.8 mg/L ZnSO 4x 7 H 2O (Fluka, 743 #96500), 1.2 mg/L CuCl2 x 2 H2O (Fluka, #61174), 1.2 mg/mL MnSO4 x H2O (Merck, Sigma-Aldrich, 744 #M7631), 1.8 mg/L CoCl2 x 6 H2O (Riedel de Haën, #12914), 2.8 mM thiamine-HCl (Merck, Sigma-745 Aldrich, #T4625), 1mM MgSO 4 x 7 H 2O (Merck, Sigma-Aldrich, #63138), 100 µM CaCl 2 x 2 H 2O 746 (Merck, Sigma-Aldrich, #C8106), and 100 µM FeCl 3 x 6 H2O (Merck, Sigma-Aldrich, #31232). 300 747 mL M9 minimal medium (0.5 w/v-% glucose) in 2 L Erlenmeyer flasks were inoculated from the 748 previous 25 mL-culture to a start OD of circa 0.004 and incubated at 37 °C under shaking at 200 749 rpm. The OD was measured regularly to calculate growth rates and exclude growth defects. At a 750 final OD of circa 0.6, samples for proteomics were collected: 275 mL of culture were transferred 751 to 500 mL harvesting tubes on ice. 18.3 mL 4 °C-cold 100 w/v-% TCA was added. The cell 752 suspension was incubated for 10 min on ice and, subsequently, centrifuged for 5 min at 4 °C and 753 3,428 g. The supernatant was discarded, and the cell pellet resuspended in 40 mL 4 °C-cold 754 acetone. The suspension was transferred to 50 mL reaction tubes and centrifuged for 5 min at 4 755 °C and 3,428 g. The supernatant was discarded, the cell pellet resuspended in 40 mL 4 °C-cold 756 acetone, and the suspension centrifuged for 5 min at 4 °C and 3,428 g. The supernatant was 757 discarded, and the cell pellet resuspended in 5 mL 4 °C-cold acetone (Merck, Supelco, #1.00014). 758 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 35 The cell suspensions of 4 independent biological replicates (= 4 different colonies) were pooled 759 and distributed to 1.5 mL screw cap-reaction tubes. After 5 min of centrifugation at 3,400 g and 760 4 °C, the supernatant was discarded and remaining cell pellets flash frozen in liquid nitrogen. 761 Samples were stored at -80 °C. Cells were lysed like S. cerevisiae S288c samples. 762 763 M.1.3 Homo sapiens HEK-293 whole-proteome sample 764 H. sapiens HEK-293 cells (ATCC, #CRL-1573) were thawed from cryo stock and transferred to a 765 cell culture-grade, sterile, 10 cm-diameter petri dish. 10 mL Dulbecco’s Modified Eagle Medium 766 (DMEM, Thermo Fisher Scientific, Gibco #41965039) containing 10 v/v-% heat inactivated foetal 767 calf serum (FCS, BioConcept, #2-01F16-I) and 1 v/v-% penicillin-streptomycin solution (P/S, 768 10,000 U/mL, Thermo Fisher Scientific, Gibco #15140122) were added, and cells incubated for 3 769 days at 37 °C, 5 %CO 2, and a relative humidity of 85 %. The supernatant was removed, 10 mL 770 phosphate buffered saline solution (PBS, pH 7.4, Thermo Fisher Scientific, Gibco #10010015) 771 added to wash surface adherent cells. The supernatant was removed and 1 mL 0.25 % Trypsin-772 EDTA solution (Thermo Fisher Scientific, Gibco #25200056) added. 24 mL of fresh medium were 773 added to a 15 cm-diameter petri dish, trypsin-suspended cells transferred to the plate with fresh 774 medium, and cells incubated for 4 days. The cell culture was tested negative for mycoplasma 775 contamination by the MycoGenie Rapid Mycoplasma Detection Kit (AssayGenie, #MORV0011-776 50). The supernatant was removed and cells washed with 25 mL PBS. 3 mL 0.25 % Trypsin-EDTA 777 solution added, and 5 new plates each with 23 mL fresh medium prepared. 8 mL medium were 778 added to the trypsinated culture, each 2 mL of the cell suspension added to a new plate, and cells 779 incubated for 2 days. Proteomics samples were taken by transferring the plates onto ice, remove 780 the supernatant, and, for each plate, adding 10 mL 4 °C-cold PBS containing 1 mM EDTA (Merck, 781 Sigma-Aldrich, #03677). The cell suspension was transferred to 15 mL reaction tubes and 782 centrifuged for 4 min at 300 g and 4 °C. The supernatant was removed, and cells resuspended in 783 1 mL of the 4 °C-cold PBS-EDTA solution. The cell suspensions from 4 plates were pooled, 784 distributed to 2 mL reaction tubes, and centrifuged for 4 min at 300 g and 4 °C. The supernatant 785 was discarded, and cell pellets flash frozen in liquid nitrogen and stored at -80 °C. Cell pellets 786 were resuspended in 500 µL lysis buffer and lysed by vortexing twice for 30 s with a pause of at 787 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 36 least 30 s on ice. Lysates were centrifuged for 15 min at 17.000 g and 4 °C. Supernatants were 788 used for further processing. 789 790 M.1.4 Mix of 8 proteins 791 Separate solutions of bovine catalase (Uniprot ID: P00432, Merck, Sigma-Aldrich, #C40), rabbit 792 creatine kinase (P00563, Merck, Sigma-Aldrich, #C3755), rabbit fructose-bisphosphate aldolase 793 A (P00883, Merck, Sigma-Aldrich, #A2714), bovine lactoferrin (P24627, Merck, Sigma-Aldrich, 794 #L9507), chicken ovotransferrin (P02789, Fluka, #27695), rabbit pyruvat kinase (P11974, Merck, 795 Sigma-Aldrich, #P9136), bovine serotransferrin (Q29443, Merck, Sigma-Aldrich, #T1408), and 796 bovine serum albumin (P02769, Merck, Sigma-Aldrich, #A7638) were prepared at 1 µg protein/µL 797 in 8 M urea 100 mM ammonium bicarbonate. After tryptic digestion and sample clean-up, 798 peptides from the individual proteins were pooled together yielding the mix of 8 proteins sample. 799 800 M.1.4 Tryptic digest and sample clean-up 801 Protein concentrations in cell lysates were measured using a bicinchoninic acid-based assay 802 (Thermo Scientific, Pierce BCA Protein Assay Kit, #23225). At a protein concentration of 1 µg/µL, 803 disulfide bonds were first reduced by incubation with 5 mM tris(2-carboxyethyl)phosphin -804 hydrochlorid (TCEP, Merck, Sigma-Aldrich, #C4706) for 30 min at 37 °C under 200 rpm of shaking 805 and second alkylated by incubation with 12 mM iodoacetamide (IAA, Merck, Sigma-Aldrich, 806 #I1149) for 15 min in the dark at 37 °C under 200 rpm of shaking. Samples were diluted with 100 807 mM ammonium bicarbonate (Merck, Sigma-Aldrich, #A6141) to a final urea concentration of 1 808 M. Sequencing-grade trypsin (Promega, #V5111) was added at a protease to protein ratio of 809 1:100, and samples incubated overnight at 37 °C. The digestion was stopped with formic acid (FA, 810 Merck, Sigma-Aldrich, #33015) at a final concentration of 2 %. Samples were desalted using C18 811 columns (Waters Sep-Pak Vac 3cc, 500 ng). After washing the columns with 2.5 mL methanol, 812 twice with 2.5 mL Buffer B (50 % 0.1 % FA), and thrice with 2.5 mL Buffer A (0.1 % FA), samples 813 were loaded onto the columns, and columns washed thrice with 2.5 mL Buffer A. Peptides were 814 eluted with 2 mL Buffer B, dried by vacuum centrifugation at 40 °C, and resuspended in Buffer A 815 at a concentration of ca. 1 µg/µL. 816 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 37 817 M.1.5 Affinity purification samples 818 A H. sapiens HEK-293 cell line expressing the Strep-HA-tagged human kinase SRPK3 (Uniprot ID: 819 Q9UPE1) under an inducible promoter was obtained from Varjosalo et al.,69 and cells seeded at 820 a density of 10 7 in 25 mL DMEM supplemented with 10 % FCS and 1 % P/S in a 15 cm dish and 821 incubated overnight at 37 °C. Protein expression was induced with 1.3 μg/mL doxycycline for 24 822 hours. Per sample, cells from two 15 cm dishes were harvested by scraping off in 4 °C-cold PBS. 823 Cells were washed once with PBS, and cell pellets flash frozen in liquid nitrogen. Pellets were 824 lysed by resolubilizing in 2 mL HNN buffer (50 mM HEPES pH 8.0, 150 mM NaCl, 50 mM NaF) 825 supplemented with 0.5 % IGEPAL, 400 nM Na 3VO4, 1 mM phenylmethylsulfonyl fluoride, 0.2 % 826 protease inhibitor cocktail (Merck, Sigma-Adrich, #P8340) and 0.5 μL/mL benzonase (Merck, 827 Sigma-Aldrich, #1.01695). The lysate was incubated on ice for 10 min and centrifuged at 18,000 828 g for 20 min at 4 ˚C. The supernatant was added to 80 μL StrepTatcin Sepharose resin (IBA 829 Lifesciences, #2-1201) and incubated on a rotary shaker for 60 min at 4 ˚C. The beads were loaded 830 onto a 1 μm glass filter column and washed 3 times with 1 mL HNN buffer supplemented with 831 0.5 % IGEPAL and 400 nM Na 3VO4, 3 times with 1 mL HNN buffer, and 3 times with 100 mM 832 ammonium bicarbonate. Samples were eluted twice by incubation with 30 μL of 0.5 mM biotin 833 in 100 mM ammonium bicarbonate for 15 min. To the samples, 60 μL 8 M urea in 100 mM 834 ammonium bicarbonate for a final concentration of 4 M urea. Samples were reduced with 5 mM 835 TCEP for 40 min at 37 °C and 200 rpm and alkylated with 40 mM IAA for 30 min at 30 °C in the 836 dark, at 200 rpm. Samples were diluted with 100 mM ammonium bicarbonate to a urea 837 concentration of 1 M. To each sample, 1 μg Lys-C and 1 μg trypsin were added, and samples were 838 incubated overnight at 37 °C and 200 rpm. The digestion was stopped with 2 % formic acid. 839 Samples were desalted using HNFR S18V desalting plates (Nest Group). The resin was activated 840 using 200 μL methanol and washed 3 times with 200 μL Buffer B. The resin was equilibrated 3 841 times with 200 μL Buffer A. Samples were loaded and washed 3 times with 200 μL Buffer A. 842 Samples were eluted twice with 100 μL Buffer B and dried at 40 °C in a vacuum centrifuge. 843 844 M.1.6 Mix of 11 peptides sample 845 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 38 We used a commercially available mix of 11 peptides (Biognosys iRT Kit, Bruker Daltonics, 846 #1816351). 847 848 M.1.7 Phosphoproteomics samples 849 Phosphopeptides of S. cerevisiae S288c samples were enriched based on a protocol described in 850 Bodenmiller et al. 45. Between 0.5 and 1 mg of desalted peptides were incubated in a rotary 851 shaker for 1 h with 1.25 mg of TiO 2 resin (GL Sciences, Japan), preequilibrated twice with 500 µL 852 of methanol, and twice with 500 µL of a solution saturated with phthalic acid (Merck, Sigma-853 Aldrich, #402915). Peptides bound to the TiO2 resin were washed twice with 500 µL phthalic acid 854 solution, twice with 80 % acetonitrile 0.1% formic acid solution, and finally twice with 0.1% formic 855 acid. The phosphopeptides were eluted from the beads twice with 150 µl of 0.3 M ammonium 856 hydroxide at pH 10.5 and immediately acidified with 50 µL 5% trifluoroacetic acid to reach about 857 pH 2.0. The enriched phosphopeptides were desalted on microspin columns (The Nest Group, 858 USA), dried using a vacuum centrifuge, and resolubilized in 10 µl of 0.1 % formic acid. 859 860 M.2 LC-MS 861 If not stated otherwise, an EASY-nLC 1200 (Thermo Scientific) coupled to an Exploris 480 (Thermo 862 Scientific) was used to measure proteomics samples in DIA mode. A sample volume 863 corresponding to 1 µg of peptides was injected. A 40 cm x 0.75 µm (inner diameter) column (New 864 Objective, 10 µm tip, PicoFrit, #PF360-75-10-N-5) packed with 3 μm C18 beads (Dr. Maisch, 120 865 Å pore size, 300 m² surface area, #Reprosil-Pur 120) was used for separating peptides over a 866 linear 120 min gradient. Buffer A was 0.1 % FA, and Buffer B 50 % ACN 0.1 % FA. The gradient 867 started at 3 % Buffer B and ended at 30 % Buffer B. The flow rate was 300 nL/min. Peptides were 868 measured in positive ionization mode. The electrospray voltage was 2500 V, the ion transfer 869 capillary temperature was 275 °C, and the funnel radiofrequency 50 % if not indicated else. To 870 reduce contamination of the MS, the electrospray voltage was set to 1500 V during the initial 8 871 min of the method and to 0 V during column washes. The MS1 full scan range was from 350 to 872 1150 m/z at a resolution of 120,000. with an automatic gain control (AGC) target of 200 % and a 873 maximum injection time of 264 ms. Precursors were fragmented in the higher-energy collisional 874 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 39 dissociation (HCD) cell at 30 % relative collision energy. For DIA, 41 variable-size MS2 scan 875 windows were used between 350 and 1,150 m/z, each at a resolution of 30,000, an AGC target 876 of 200 %, a maximum injection time of 66 ms, and a window overlap of 1 Da. 877 For measurements with an Q Exactive Plus (Thermo Scientific), which was also coupled to an 878 EASY-nLC 1000 (Thermo Scientific), the same parameters were used as for the Exploris 480, if 879 possible. MS1 full scans differed by the resolution (70,000), the AGC target (3e6), and the 880 maximum injection time (120 ms). DIA MS2 measurements differed by the resolution (35,000), 881 AGC target (3e6), and the maximum injection time (60 ms). 882 For measurements with an Astral (Thermo Scientific) coupled to a Vanquish Neo LC system 883 (Thermo Scientific), the flow rate was 400 nL/min, and a 41.8 min gradient was used. Starting 884 with Buffer B at 3 %, buffer B was changed in linear steps to 32 % at 25.5 min, 45 % at 30.5 min, 885 95 % at 32.5 min, where it was kept constant for 8 min, and 3 % at 41.8 min. MS1 full scans were 886 acquired between 350 and 1400 m/z with a resolution of 240,000. The AGC target was 5e6. For 887 DIA, 524 fixed-size (2 Da) MS2 scan windows between 350 and 1400 m/z were used with a 888 maximum injection time of 3 ms, a relative HCD collision energy of 27 %, the RF lens at 40 %, and 889 an AGC target of 5e4. 890 891 M.3 Peptide searches and data processing 892 Spectronaut 20 (Biognosys, Schlieren, Switzerland) was used to search DIA data for peptides 893 using “directDIA”. The data was searched against different proteome databases appropriate to 894 each sample and related FASTA-files were obtained from UniProt 70. The enzyme specificity was 895 set to semi-specific with cleavage rules for trypsin (or if applicable also for LysC). Peptide lengths 896 between 7 and 52 amino acids, two missed cleavages, and a maximum of 5 variable modifications 897 were permitted. The immunopeptidomics and LiP datasets were searched for peptides with 898 unspecific cleavage sites and 5 to 16 amino acids in length. N-Acetylation at protein N-termini 899 and oxidation of methionine were set as variable modifications, and carbamidomethylation of 900 cysteine as a fixed modification. Searches for phosphopeptides included phosphorylation of 901 serine, threonine, and tyrosine. Data from the parameter test (also see Fig. 4) was searched 902 twice: (1) all samples were searched individually to obtain an accurate representation of the 903 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 40 number of peptide precursor identifications, and (2) all samples were searched together to 904 obtain accurate quantitative data (intensities). Supplementary Table 2 provides an overview of 905 which samples of the literature datasets (also see Supplementary Tab. 1) were used in this study 906 and included in peptide searches. 907 Data from peptide searches were analyzed with MATLAB (R2023b, 23.2.0.2409890). Venn 908 diagrams were created using the MATLAB function “venn” 71. The data was analyzed at the 909 peptide precursor level if not specified else. Precursors with identification Q-values greater than 910 10-3 and precursors without quantities or “FGMS2RawQuantity” smaller than 2 were filtered out, 911 except for the algorithm comparison (also see Supplementary Fig. 3) for which no filter was 912 applied. 913 914 M.4 ISF detection 915 ISF were detected in each sample separately. First, all peptide precursors were paired up against 916 each other, for which one precursor’s amino acid sequence contained the sequence from the 917 other precursor. Then, looping over all precursors and their pairings from small to large (length), 918 ∆RTs were calculated by subtracting the peak apex retention times of the smaller peptide from 919 the larger peptide. Subsequently, ∆RT cutoffs were determined by fitting the model function 920 (EQ1) to the ∆RT distribution of precursor pairs that only differ in charge. The function consisted 921 of two components. One of which was a normal distribution with mean value µ and standard 922 deviation σ that was scaled with parameter a. The other component was the uniform distribution 923 k0. ∆RT cutoff was the ∆RT, for which the absolute difference between the normal distribution 924 and k0 was smallest. 925 𝑦(𝑥) = 𝑎 ∙ ௘ ష (౮ షഋ )మ మ഑ మ ఙ √ଶగ + 𝑘଴ (EQ1) 926 The distribution of measured ∆RTs was estimated ten times with different bin sizes and, 927 accordingly, the model function fitted to each estimation using least square optimization. The 928 final ∆RT cutoff was the median across the ten ∆RT cutoffs from the ten individual fits. If the 929 number of datapoints of false-positive precursor pairs that only differ in charge was insufficient 930 to model the uniform distribution, the ∆RT cutoffs could not be accurately determined and 931 typically became very large. In such cases, in which our model-based approach yields a ∆RT cutoff 932 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 41 larger than 0.15 min, the full width at half maximum (FWHM) of the normal distribution was 933 used. In extreme cases, in which the data was even insufficient to model the normal distribution 934 of true positives, a ∆RT cutoff of 0.1 min was taken. The lower boundary of the ∆RT cutoff was 935 the smallest ∆RT value of the respective sample. 936 937 M.5 Multiple peak extraction from DIA data 938 To analyze peptide precursors with multiple peaks in DIA data, we followed a previously 939 published strategy 62. If the same peptide sequence is identified in DDA beyond a certain time 940 span, which is typically longer than the average elution peak width (e.g. 2 min, which we used 941 here), the peptide-spectrum matches were binned into separate spectral libraries. Subsequently, 942 for each of the peptides, a local average retention time and consensus MS/MS fragmentation 943 spectrum were calculated. To force the DIA analysis software to extract multiple identical peptide 944 sequences at different retention times, we “anonymized” the peptides with a set of characters 945 that cannot be properly parsed by the software (e.g. adding numbers or special characters). 946 When faced with such non-parsable sequences, the search tool defers back to the only other 947 information available to extract the DIA data: (1) the m/z and charge state of the precursor, (2) 948 the m/z, charge states and relative intensities of the fragments, and (3) the peptide retention 949 time. This strategy enabled us to estimate how often a peptide sequence could be identified 950 multiple times in DIA file. 951 952 Data and code availability 953 Proteomics data will be made available on PRIDE and relevant MATLAB code will be provided on 954 GitHub and Zenodo upon publication. The source data of figures will be provided as 955 supplementary information upon publication. 956 957 Acknowledgments 958 We thank Oliver Bernhardt, Tejas Gandhi, Roland Bruderer, and Monika Pepelnjak for 959 discussions. We thank Biognosys for early access to the ISF detection feature of Spectronaut 20. 960 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 42 P. P. was funded by the Promedica Stiftung (2025-0022/M) and by an ETH Zurich Grant (25-2 ETH-961 027). 962 963 Author contributions 964 Thorben Schramm: Conceptualization, Methodology, Software, Validation, Formal analysis, 965 Investigation, Resources, Data Curation, Writing - Original Draft, Writing - Review & Editing, 966 Visualization, Project administration 967 Ludovic Gillet: Conceptualization, Methodology, Software, Validation, Formal analysis, 968 Investigation, Resources, Writing - Original Draft, Writing - Review & Editing 969 Viviane Reber: Resources 970 Natalie de Souza: Writing - Review & Editing 971 Matthias Gstaiger: Supervision 972 Paola Picotti: Conceptualization, Writing - Original Draft, Writing - Review & Editing, Supervision, 973 Project administration, Funding acquisition 974 975 Declaration of interests 976 P. P. is a scientific advisor for the company Biognosys AG (Schlieren, Switzerland) and an inventor 977 of a patent licensed by Biognosys AG that covers the LiP-MS method. The remaining authors 978 declare no competing interests. 979 980

981 1. Alseekh, S. et al. Mass spectrometry-based metabolomics: a guide for annotation, quantification 982 and best reporting practices. Nat. Methods 18, 747–756 (2021). 983 2. Collins, S. L., Koo, I., Peters, J. M., Smith, P. B. & Patterson, A. D. Current Challenges and Recent 984 Developments in Mass Spectrometry–Based Metabolomics. Annu. Rev. Anal. Chem. 14, 467–487 985 (2021). 986 3. Guo, T., Steen, J. A. & Mann, M. Mass-spectrometry-based proteomics: from single cells to clinical 987 applications. Nature 638, 901–911 (2025). 988 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 43 4. Aebersold, R. & Mann, M. Mass-spectrometric exploration of proteome structure and function. 989 Nature 537, 347–355 (2016). 990 5. Thomas, S. N., French, D., Jannetto, P. J., Rappold, B. A. & Clarke, W. A. Liquid chromatography–991 tandem mass spectrometry for clinical diagnostics. Nat. Rev. Methods Primer 2, 96 (2022). 992 6. Glish, G. L. & Vachet, R. W. The basics of mass spectrometry in the twenty-first century. Nat. Rev. 993 Drug Discov. 2, 140–150 (2003). 994 7. Yamashita, M. & Fenn, J. B. Electrospray ion source. Another variation on the free-jet theme. J. Phys. 995 Chem. 88, 4451–4459 (1984). 996 8. Prabhu, G. R. D., Williams, E. R., Wilm, M. & Urban, P. L. Mass spectrometry using electrospray 997 ionization. Nat. Rev. Methods Primer 3, 23 (2023). 998 9. Daniel, J. M., Friess, S. D., Rajagopalan, S., Wendt, S. & Zenobi, R. Quantitative determination of 999 noncovalent binding interactions using soft ionization mass spectrometry. Int. J. Mass Spectrom. 1000 216, 1–27 (2002). 1001 10. Guo, J., Shen, S., Xing, S., Yu, H. & Huan, T. ISFrag: De Novo Recognition of In-Source Fragments for 1002 Liquid Chromatography–Mass Spectrometry Data. Anal. Chem. 93, 10243–10250 (2021). 1003 11. Xu, Y.-F., Lu, W. & Rabinowitz, J. D. Avoiding Misannotation of In-Source Fragmentation Products as 1004 Cellular Metabolites in Liquid Chromatography–Mass Spectrometry-Based Metabolomics. Anal. 1005 Chem. 87, 2273–2281 (2015). 1006 12. Domingo-Almenara, X. et al. Autonomous METLIN-Guided In-source Fragment Annotation for 1007 Untargeted Metabolomics. Anal. Chem. 91, 3246–3253 (2019). 1008 13. Farke, N., Schramm, T., Verhülsdonk, A., Rapp, J. & Link, H. Systematic analysis of in-source 1009 modifications of primary metabolites during flow-injection time-of-flight mass spectrometry. Anal. 1010 Biochem. 664, 115036 (2023). 1011 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 44 14. Chen, L. et al. Widespread occurrence of in-source fragmentation in the analysis of natural 1012 compounds by liquid chromatography–electrospray ionization mass spectrometry. Rapid Commun. 1013 Mass Spectrom. 37, e9519 (2023). 1014 15. Schmid, R. et al. Ion identity molecular networking for mass spectrometry-based metabolomics in 1015 the GNPS environment. Nat. Commun. 12, 3832 (2021). 1016 16. Lu, W. et al. Improved Annotation of Untargeted Metabolomics Data through Buffer Modifications 1017 That Shift Adduct Mass and Intensity. Anal. Chem. 92, 11573–11581 (2020). 1018 17. Giera, M., Aisporna, A., Uritboonthai, W. & Siuzdak, G. The hidden impact of in-source 1019 fragmentation in metabolic and chemical mass spectrometry data interpretation. Nat. Metab. 6, 1020 1647–1648 (2024). 1021 18. El Abiead, Y. et al. Discovery of metabolites prevails amid in-source fragmentation. Nat. Metab. 7, 1022 435–437 (2025). 1023 19. Senan, O. et al. CliqueMS: a computational tool for annotating in-source metabolite ions from LC-1024 MS untargeted metabolomics data based on a coelution similarity network. Bioinformatics 35, 1025 4089–4097 (2019). 1026 20. Fuhrer, T., Heer, D., Begemann, B. & Zamboni, N. High-Throughput, Accurate Mass Metabolome 1027 Profiling of Cellular Extracts by Flow Injection–Time-of-Flight Mass Spectrometry. Anal. Chem. 83, 1028 7074–7080 (2011). 1029 21. Kim, J.-S., Monroe, M. E., Camp, D. G. I., Smith, R. D. & Qian, W.-J. In-Source Fragmentation and the 1030 Sources of Partially Tryptic Peptides in Shotgun Proteomics. J. Proteome Res. 12, 910–916 (2013). 1031 22. Gillet, L. C. et al. Targeted Data Extraction of the MS/MS Spectra Generated by Data-independent 1032 Acquisition: A New Concept for Consistent and Accurate Proteome Analysis*. Mol. Cell. Proteomics 1033 11, O111.016717 (2012). 1034 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 45 23. Gillet, L. C., Leitner, A. & Aebersold, R. Mass Spectrometry Applied to Bottom-Up Proteomics: 1035 Entering the High-Throughput Era for Hypothesis Testing. Annu. Rev. Anal. Chem. 9, 449–472 (2016). 1036 24. Ting, Y. S. et al. Peptide-Centric Proteome Analysis: An Alternative Strategy for the Analysis of 1037 Tandem Mass Spectrometry Data*. Mol. Cell. Proteomics 14, 2301–2307 (2015). 1038 25. Chong, C., Coukos, G. & Bassani-Sternberg, M. Identification of tumor antigens with 1039 immunopeptidomics. Nat. Biotechnol. 40, 175–188 (2022). 1040 26. Feng, Y. et al. Global analysis of protein structural changes in complex proteomes. Nat. Biotechnol. 1041 32, 1036–1044 (2014). 1042 27. Kalxdorf, M., Müller, T., Stegle, O. & Krijgsveld, J. IceR improves proteome coverage and data 1043 completeness in global and single-cell proteomics. Nat. Commun. 12, 4787 (2021). 1044 28. Midha, M. K. et al. A comprehensive spectral assay library to quantify the Escherichia coli proteome 1045 by DIA/SWATH-MS. Sci. Data 7, 389 (2020). 1046 29. Jayavelu, A. K. et al. The proteogenomic subtypes of acute myeloid leukemia. Cancer Cell 40, 301-1047 317.e12 (2022). 1048 30. Salovska, B. et al. Peroxiredoxin 6 protects irradiated cells from oxidative stress and shapes their 1049 senescence-associated cytokine landscape. Redox Biol. 49, 102212 (2022). 1050 31. Gunter, H. M. et al. A universal molecular control for DNA, mRNA and protein expression. Nat. 1051 Commun. 15, 2480 (2024). 1052 32. Bradley, D. et al. The fitness cost of spurious phosphorylation. EMBO J. 43, 4720–4751 (2024). 1053 33. Guzman, U. H. et al. Ultra-fast label-free quantification and comprehensive proteome coverage with 1054 narrow-window data-independent acquisition. Nat. Biotechnol. 42, 1855–1866 (2024). 1055 34. Kattelus, R. et al. Phenotypic profiling of human induced regulatory T cells at early differentiation: 1056 insights into distinct immunosuppressive potential. Cell. Mol. Life Sci. 81, 399 (2024). 1057 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 46 35. Zare, A. et al. Axonal tau reduction ameliorates tau and amyloid pathology in a mouse model of 1058 Alzheimer’s disease. Transl. Neurodegener. 14, 39 (2025). 1059 36. Zhang, H. et al. Heterochromatome wide analyses reveal MBD2 as a phase separation scaffold for 1060 heterochromatin compartmentalization and composition. Nucleic Acids Res. 53, gkaf1380 (2025). 1061 37. Romero-Pérez, P. S. et al. Protein surface chemistry encodes an adaptive tolerance to desiccation. 1062 Cell Syst. 16, 101407 (2025). 1063 38. Arroyo-Gomez, J. et al. Functional landscape of ubiquitin linkages couples K29-linked ubiquitylation 1064 to epigenome integrity. EMBO J. 44, 6944–6978 (2025). 1065 39. Botella, J. et al. Sprint interval exercise disrupts mitochondrial ultrastructure driving a unique 1066 mitochondrial stress response and remodelling in men. Nat. Commun. 17, 71 (2025). 1067 40. Pereyra, G. et al. SFRP1 upregulation causes hippocampal synaptic dysfunction and memory 1068 impairment. Cell Rep. 44, 115535 (2025). 1069 41. Gharibi, B. et al. Post-gastrulation amnioids as a stem cell-derived model of human extra-embryonic 1070 development. Cell 188, 3757-3774.e20 (2025). 1071 42. Su, J. et al. Polymerization-mediated SRFR1 condensation in upper lateral root cap cells regulates 1072 root growth. Plant Cell 38, koaf292 (2026). 1073 43. Benjamini, Y. & Hochberg, Y. Controlling the False Discovery Rate: A Practical and Powerful 1074 Approach to Multiple Testing. J. R. Stat. Soc. Ser. B Methodol. 57, 289–300 (1995). 1075 44. Hinterholzer, A. et al. Detecting aspartate isomerization and backbone cleavage after aspartate in 1076 intact proteins by NMR spectroscopy. J. Biomol. NMR 75, 71–82 (2021). 1077 45. Bodenmiller, B. et al. Phosphoproteomic Analysis Reveals Interconnected System-Wide Responses 1078 to Perturbations of Kinases and Phosphatases in Yeast. Sci. Signal. 3, rs4–rs4 (2010). 1079 46. Bekker-Jensen, D. B. et al. Rapid and site-specific deep phosphoproteome profiling by data-1080 independent acquisition without the need for spectral libraries. Nat. Commun. 11, 787 (2020). 1081 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 47 47. Wang, Y. et al. GABAA receptor π forms channels that stimulate ERK through a G-protein-dependent 1082 pathway. Mol. Cell 85, 166-176.e5 (2025). 1083 48. He, Y. et al. Evaluation of the Orbitrap Ascend Tribrid Mass Spectrometer for Shotgun Proteomics. 1084 Anal. Chem. 95, 10655–10663 (2023). 1085 49. Criscuolo, A., Zeller, M. & Fedorova, M. Evaluation of Lipid In-Source Fragmentation on Different 1086 Orbitrap-based Mass Spectrometers. J. Am. Soc. Mass Spectrom. 31, 463–466 (2020). 1087 50. Yu, F. et al. Fast Quantitative Analysis of timsTOF PASEF Data with MSFragger and IonQuant. Mol. 1088 Cell. Proteomics 19, 1575–1585 (2020). 1089 51. Reber, V. et al. Paradoxical non-catalytic kinase functions are driven by inhibitor-induced 1090 displacement of autoinhibitory domains. 2025.11.06.687012 Preprint at 1091 https://doi.org/10.1101/2025.11.06.687012 (2026). 1092 52. Elsässer, F. et al. Limited proteolysis-coupled mass spectrometry captures proteome-wide protein 1093 structural alterations and biomolecular condensation in living cells. Mol. Syst. Biol. (2026) 1094 doi:10.1038/s44320-025-00182-6. 1095 53. Abelin, J. G. & Cox, A. L. Innovations Toward Immunopeptidomics. Mol. Cell. Proteomics 23, 100823 1096 (2024). 1097 54. Purcell, A. W., Ramarathinam, S. H. & Ternette, N. Mass spectrometry–based identification of MHC-1098 bound peptides for immunopeptidomics. Nat. Protoc. 14, 1687–1707 (2019). 1099 55. Pak, H. et al. Sensitive Immunopeptidomics by Leveraging Available Large-Scale Multi-HLA Spectral 1100 Libraries, Data-Independent Acquisition, and MS/MS Prediction. Mol. Cell. Proteomics 20, 100080 1101 (2021). 1102 56. Kessler, A. L. et al. HLA I immunopeptidome of synthetic long peptide pulsed human dendritic cells 1103 for therapeutic vaccine design. Npj Vaccines 10, 12 (2025). 1104 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 48 57. Dorvash, M., Illing, P. T., Croft, N. P., Ramarathinam, S. H. & Purcell, A. W. Deep Exploration of the 1105 Immunopeptidome of a Pancreatic Cancer Cell Line: Implications for Clinical Immunopeptidomics 1106 and Immunotherapy. Mol. Cell. Proteomics 24, 101030 (2025). 1107 58. Tanuwidjaya, E. et al. SAPrIm 2.0: a semi-automated protocol for mid-throughput soluble HLA 1108 immunopeptidomics. Front. Immunol. 16, (2025). 1109 59. Cappelletti, V. et al. Dynamic 3D proteomes reveal protein functional alterations at high resolution 1110 in situ. Cell 184, 545-559.e22 (2021). 1111 60. Mehta, V. et al. Structure of Mycobacterium tuberculosis Cya, an evolutionary ancestor of the 1112 mammalian membrane adenylyl cyclases. eLife 11, e77032 (2022). 1113 61. Li, K. et al. A peptide-centric local stability assay enables proteome-scale identification of the 1114 protein targets and binding regions of diverse ligands. Nat. Methods 22, 278–282 (2025). 1115 62. Schubert, O. T. et al. Building high-quality assay libraries for targeted analysis of SWATH MS data. 1116 Nat. Protoc. 10, 426–441 (2015). 1117 63. Dunham, W. H., Mullin, M. & Gingras, A.-C. Affinity-purification coupled to mass spectrometry: Basic 1118 principles and strategies. PROTEOMICS 12, 1576–1590 (2012). 1119 64. O’Reilly, F. J. & Rappsilber, J. Cross-linking mass spectrometry: methods and applications in 1120 structural, molecular and systems biology. Nat. Struct. Mol. Biol. 25, 1000–1008 (2018). 1121 65. Goel, R. K., Bithi, N. & Emili, A. Trends in co-fractionation mass spectrometry: A new gold-standard 1122 in global protein interaction network discovery. Curr. Opin. Struct. Biol. 88, 102880 (2024). 1123 66. Malta, C. F. et al. Pushing the limits of hydrogen/deuterium exchange mass spectrometry to study 1124 protein:fragment low affinity interactions. Commun. Chem. 8, 405 (2025). 1125 67. Harvey, D. J. Analysis of Protein Glycosylation by Mass Spectrometry. in Analysis of Protein Post-1126 Translational Modifications by Mass Spectrometry 89–159 (John Wiley & Sons, Ltd, 2016). 1127 doi:10.1002/9781119250906.ch3. 1128 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint 49 68. Cardon, T., Fournier, I. & Salzet, M. Chasing the Ghost Proteome in the Dark Matter. Mol. Cell. 1129 Proteomics 24, 101076 (2025). 1130 69. Varjosalo, M. et al. The Protein Interaction Landscape of the Human CMGC Kinase Group. Cell Rep. 1131 3, 1306–1320 (2013). 1132 70. The UniProt Consortium. UniProt: the Universal Protein Knowledgebase in 2025. Nucleic Acids Res. 1133 53, D609–D617 (2025). 1134 71. Gamble, D. venn (https://ch.mathworks.com/matlabcentral/fileexchange/22282-venn), MATLAB 1135 Central File Exchange. (2026). 1136 1137 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 30, 2026. ; https://doi.org/10.64898/2026.03.27.714398doi: bioRxiv preprint