UT-SCC-7 (RRID : CVCL_7868) cell line was established from surgically removed metastatic cutaneous squamous cell carcinoma (cSCC) of the skin in Turku University Hospital, and human head and neck SCC cell line UT-SCC-2 (RRID : CVCL_7820) was established from primary SCC of the oral cavity in Turku University Hospital (18, 19). UT-SCC-7, UT-SCC-2 and HepG2 (RRID : CVCL_0027, hepatocellular carcinoma) cells were a kind gift from Professor Veli-Matti Kähäri (Department of Dermatology, University of Turku and Turku University Hospital, Turku, Finland). A549 (RRID: CVCL_0023, lung carcinoma), MDA-MB-231 (RRID : CVCL_0062) and MCF7 (RRID : CVCL_0031) (breast adenocarcinomas) cells were purchased from ATCC. SKOV3 (RRID : CVCL_0532) and Caov-3 (RRID : CVCL_0201) (ovarian adenocarcinomas) cell lines were a kind gift from Professor Klaus Elenius (Institute of Biomedicine, University of Turku, Finland). Caco-2 (RRID : CVCL_0025, colorectal adenocarcinoma) cells were kindly provided by Professor Diana Toivola (Faculty of Science and Engineering, Åbo Akademi University, Turku, Finland). Human primary prostate cancer fibroblasts (CAF) were established as previously described (20), and used up to passage 10. Human primary gingival fibroblasts (GFB) were purchased from ATCC and used up to passage 10. Primary human adult skin fibroblasts (SFB) were from the cell line collection of the Medical Biochemistry/the University of Turku and a kind gift from Professor Risto Penttinen. The fibroblasts were from male donor aged 24 years and they were used up to passage 12. A549, SKOV3, Caov-3, MDA-MB-231, MCF7, Caco-2 and HepG2 cell lines were authenticated by short tandem repeat DNA profiling. Genotyping was performed by the Institute for Molecular Medicine Finland FIMM Technology Centre, University of Helsinki.

All cell lines were grown in Dulbecco’s modified Eagle’s medium (DMEM with 4.5 g/L glucose; 12-614F, Lonza) supplemented with 10% fetal calf serum (FCS), L-glutamine (6 nmol/L), penicillin (100 U/ml) and streptomycin (100 μg/ml). 1 x MEM non-essential amino acids (11140-035, Gibco) were added to UT-SCC-7 and UT-SCC-2 medium. All cell lines were routinely tested to be negative for mycoplasma contamination using MycoAlert PLUS Mycoplasma Detection Kit (LT07-710, Lonza).

The information below on spheroid preparation and growth conditions is based on MISpheroID recommendations (21). 3D spheroids were made in micro-molds according to the manufacturer’s instructions (MicroTissues 3D Petri Dish micro-mold spheroids, Sigma-Aldrich) with 2.5 × 10⁵ cells in one mold (monocultures; 7000 cells/spheroid) or 5.0 × 10⁵ cells in one mold (cocultures; 14 000 cells/spheroid). In cocultures the cell ratio was 1:1 (cancer cells and fibroblasts, respectively). The spheroids were grown in serum-free DMEM medium for 5 days at 37°C in an incubator environment of 20% O₂ and 5% CO₂. Ascorbic acid (50 µg/ml) was added daily. The spheroids were washed out from the molds with hypotonic lysis buffer (10 mM Tris-HCl pH 7.4, 1 mM EDTA pH 8.0, 10 µg/ml DNase I (DN25, Sigma-Aldrich)) and collected into Eppendorf tubes. The spheroids were then washed twice with hypotonic lysis buffer and centrifuged between washes 2 min/1000 x g. After that, the spheroids were incubated with hypotonic lysis buffer o/n in +4 °C in a rotator. The next day, spheroids were washed twice with hypotonic lysis buffer as described above. The pellets were frozen in -20 °C for further mass spectrometric analysis.

The spheroids were grown in micro-molds for 6 days as described above. Ascorbic acid (50 µg/ml) was added daily. The spheroids were washed out from the molds with citrullination buffer (40 mM Tris-HCl pH 7.4, 5 mM CaCl₂, 150 mM NaCl) and collected into Eppendorf tubes. The spheroids were washed twice with citrullination buffer and centrifuged between washes 2 min/1000 x g. Next, PAD4 (SAE0086, Sigma-Aldrich; 2 units/sample in citrullination buffer) was added to the spheroids and incubated 2h/+37 °C. Control samples were treated with citrullination buffer only. The spheroids were then washed twice with hypotonic lysis buffer and centrifuged between washes 2 min/1000 x g. After that, the spheroids were incubated with hypotonic lysis buffer o/n in +4 °C in a rotator. The next day, spheroids were washed twice with hypotonic lysis buffer and centrifuged between washes 2 min/1000 x g. The pellets were frozen in -20°C for further mass spectrometric analysis.

The hypotonically lysed pellets were resuspended in 50 mM triethylammonium bicarbonate (TEAB), 5% SDS, pH 7.5. The dissolved samples were then clarified at 13,000g for 8 minutes. Each sample was digested in an S-Trap micro column (ProtiFi, USA) according to the standard protocol provided by the manufacturer. Briefly, the proteomic mixture in the supernatant was reduced with 20 mM dithiothreitol at 95°C for 20 minutes and then at 60°C for 2 hours. The reduced proteins were alkylated with 40 mM iodoacetamide in dark for 30 minutes. The samples were then treated with 50 mM TEAB, 5% SDS, 7.5 M urea, pH 7.5 to solubilize the matrisome proteins. These ECM solubilized samples were acidified with 1.2% aqueous phosphoric acid and mixed with 6× volumes of 90% methanol, 100 mM TEAB, pH 7.1 (S-Trap binding buffer). A protein solution of 200 µl at a time was loaded onto the filter of the S-Trap micro column and centrifuged at 4000×g for 30 seconds. The flow-through was discarded each time and the samples were washed 9 times with the S-Trap binding buffer. Proteins trapped in the S-Trap column were digested with Trypsin/Lys-C Mix (Promega, USA) at 47°C for 1 hour at a protein-to-enzyme ratio of 1:25 (w/w). The digested peptides were eluted with buffers in order: 50 mM TEAB pH 8.0, 0.2% aqueous formic acid, and 50% acetonitrile/0.2% aqueous formic acid. The eluted fractions were pooled, SpeedVac (Thermo Fisher Scientific, USA) dried, and desalted with C18 tips (Empore, USA). Finally, the desalted peptides were again SpeedVac dried and dissolved in 0.1% formic acid for liquid chromatography-mass spectrometry analysis.

The peptide analysis was performed using a nanoflow HPLC system (Easy-nLC1000, Thermo Fisher Scientific) coupled with a Q Exactive HF mass spectrometer (Thermo Fisher Scientific). The peptide mixture was separated using a 15cm C18 column (75 µm × 15 cm, ReproSil-Pur 3 µm 200 Å C18-AQ, Dr. Maisch HPLC GmbH, Ammerbuch-Entringen, Germany) with a gradient consisting of solvents A (0.1% formic acid) and solvent B (acetonitrile/water (95:5(v/v)) with 0.1% formic acid) from 8–35% B in 50 minutes, 35-100% B in 2 minutes and held at 100% B for 8 minutes. A full mass scan was acquired over the m/z range of 300-1750 at a resolution of 120,000 with a maximum allowed injection time of 50 milliseconds followed by data dependent acquisition of isolation window of 2.0 m/z and dynamic exclusion time of 20 seconds. The top 12 intense peaks from full scan were fragmented with higher-energy collisional induced dissociation and scanned over the m/z range of 200-2000 m/z at a resolution of 15,000. A new survey scan of the full mass spectrum was initiated after the completion of the fragmentation scan, and the process was repeated until the end of the gradient run.

Raw mass spectrometry data of different cancers and normal tissues were collected from PRIDE (https://www.ebi.ac.uk/pride/), MassIVE (https://massive.ucsd.edu/ProteoSAFe/static/massive.jsp) and CPTAC (https://cptac-data-portal.georgetown.edu/) databases ( Table 1 ). All data had been acquired using liquid chromatography-coupled tandem mass spectrometry with data dependent acquisition and QExactive, Orbitrap Lumos or LTQ Orbitrap mass spectrometers.

Proteome Discoverer 2.4 (Thermo Scientific) with MS Amanda 2.0 search engine (39) was used to analyze all raw data, including in-house generated, against Human reviewed Uniprot sequences including isoforms (release 2020_02). Trypsin-digested peptides with a maximum of 3 missed cleavages were allowed. Parent ion tolerance of 5 ppm and product ion tolerance of 0.02 Da (for data from QExactive or Orbitrap Lumos) or 0.2 Da (for data from LTQ Orbitrap) were allowed. Oxidation (K, M, P) and deamidation (N, Q, R) were set as dynamic modifications. With raw data from TMT or iTRAQ labeled samples, dynamic modification with TMT or iTRAQ label (N-terminus, K) was added. Carbamidomethylation (C) was set as a static modification. Only the peptide spectrum matches (PSMs) with concatenated target-decoy database search-derived false discovery rate below 0.05 were included.

From the potentially citrullinated PSMs, those with C-terminal citrulline were removed, and the spectra matching to matrisome protein-derived peptides according to the human matrisome protein list at http://matrisomeproject.mit.edu (40) were manually inspected and validated. Only the spectra with at least one b or y ion (from collision-induced dissociation) or c or z ion (from electron-transfer dissociation) peak matching to citrullinated peptide fragment with or without neutral loss of isocyanic acid from citrulline residue (-43 Da) were included. Citrullination was determined invalid if (i) citrullination could not be distinguished from N or Q deamidation or (ii) peptide mass had been erroneously calculated from second isotopic peak instead of monoisotopic peak of precursor ion or (iii) all citrullinated peptide fragments could be matched to a second isotopic peak of the same but unmodified fragments or (iv) the retention time was similar or shorter than that of identical unmodified peptide or (v) the citrullinated peptide was the only peptide matching to the corresponding protein or (vi) the peptide contained also unlikely modifications (such as hydroxylation of K or P in non-collagenous proteins). Quantitation of TMT-labeled subset of CPTAC colon adenocarcinoma dataset PDC000116 was done by using MaxQuant, version 1.6.10.43 (41).

Correlation analysis was performed using SPSS Statistics (version 26, IBM). GOnet tool at https://tools.dice-database.org/GOnet/ (42) was used for gene ontology classification.

The mass spectrometry proteomics data generated in this study have been deposited to the ProteomeXchange Consortium via the PRIDE (43) partner repository with the dataset identifier PXD033555 and 10.6019/PXD033555. Accession information of publicly available data analyzed in this study can be found in Table 1 . All other data in this article are available by contacting the corresponding author on reasonable request.

Citrullination of an arginine residue causes 0.98 Da mass increase, which can be detected by mass spectrometry. However, the risk to falsely interpret a peptide as citrullinated is high. The mass shift caused by citrullination is identical to that of spontaneous deamidation of asparagine or glutamine. Thus, it is crucial to include deamidation of asparagine and glutamine along with citrullination as variable modifications in the database search. We also included oxidation of proline and lysine as variable modifications, since we were specifically interested in matrisome proteins, including collagens. We used several criteria for rejecting the spectra matching to citrullinated peptides ( Figure 1 ).

After initial database search and selection of peptide-spectrum matches (PSMs) with false discovery rate less than 0.01, we categorically rejected the PSMs with C-terminal citrullination. It has been shown that trypsin is about 10⁵ times slower in hydrolyzing benzoyl-L-citrulline-methyl ester than benzoyl-L-arginine methyl ester (44). Therefore, it is reasonable to assume that citrullination leads to missed cleavage, and thus the presence of citrulline residue in the peptide C- terminus would be possible only in C-terminal peptides derived from proteins with C-terminal arginine.

Since our focus was in matrisome proteins, we included only the spectra matching to proteins listed in the matrisome database (http://matrisomeproject.mit.edu). The relatively low number of the remaining spectra allowed us to carefully examine them manually. Citrullination was determined invalid if it could be confused with deamidation of asparagine or glutamine. Moreover, since the mass shift of one citrullination (0.98 Da) is close to mass of neutron (1.01 Da), the search program occasionally determined the second isotopic peak as monoisotopic peak, leading to false identification of citrullinated peptide. We also noticed that the fragmentation spectra contained second (and third) isotopic peaks of the b and y ions, and occasionally the search program interpreted the second isotopic peak as a monoisotopic peak, leading to falsely detected citrullination. This together with erroneous determination of monoisotopic precursor peak or confusing with deamidation could result in false identification of citrullinated peptide by the database search.

Citrullination increases hydrophobicity of the peptide leading to longer retention time compared to the identical unmodified peptide. Moreover, deamidation of asparagine or glutamine causes smaller shift of retention time compared to that of citrullination (45). Thus, we determined citrullination invalid if the retention time of the identical unmodified peptide was similar or shorter.

It is a common practice and recommended by e.g., in the guidelines of Molecular and Cellular Proteomics journal (https://www.mcponline.org/mass-spec-guidelines) in mass spectrometry-based proteomics that a protein is determined as identified only if it has at least two peptide identifications. Therefore, we did not accept spectra that matched to a protein with no other peptide identifications. In addition, we rejected the spectra that matched to a peptide containing unlikely modifications such as proline hydroxylation in a non-collagenous protein.

Fragmentation by collision-induced dissociation causes neutral loss of isocyanic acid (43 Da) from citrullinated peptide. Although peaks representing y and b ions with 43 Da loss can be frequently seen in the fragmentation spectra of citrullinated peptides, this is not always the case (45). Therefore, we decided not to include the absence of neutral loss of 43 Da as a criterion of falsely determined citrullination.

To study the abundance of ECM citrullination in human tumors, we analyzed cancer proteomics data sets in PRIDE, MassIVE and CPTAC ( Supplementary Tables 1, 2 ) for citrullination of matrisome proteins. The data sets contained both pooled cancer samples and samples derived from individual patients ( Table 1 ). Citrullinated matrisome proteins were detected in 16 out of 24 cancer related data sets and in 3 out of 5 data sets of normal tissue samples. However, only in three data sets the relative citrullination of matrisome proteins was higher than 4%. These data sets represented a liver metastasis from one patient with a colorectal cancer, CPTAC glioblastoma and CPTAC head and neck squamous cell carcinoma samples ( Figure 2 ). Thus, the conclusion was that although small numbers of citrullinated matrisome proteins can be detected in many tumors, it is a relative rare phenomenon. Still some tumors may contain higher numbers of citrullinated matrisome proteins, and in these cases, it is also possible to speculate that citrullination contributes to pathogenic mechanisms. The most often citrullinated matrisome protein was fibrinogen alpha chain (FGA; Table 2 ). Citrullinated peptides derived from FGA were found in 14 out of 29 data sets ( Table 2 ). Previously, citrullination of fibrinogen has been connected to inflammatory diseases, such as rheumatoid arthritis (11, 46). Other most frequently citrullinated matrisome proteins included coagulation factor II (F2, thrombin; 9/29 data sets), cathepsin G (CTSG), fibronectin 1 (FN1), and periostin (POSTN, osteoblast specific factor) (7/29 data sets; Table 2 ). Citrullination of fibronectin is also an abundant phenomenon in inflammation (11, 46–48).

Our analysis of cancer proteomics data sets revealed that some malignant tumors seem to contain significant numbers of citrullinated matrisome proteins. However, it is unclear which cells in the tumor are required for citrullination. To investigate this, we used three-dimensional spheroid-type cell cultures that are widely used in in vitro studies focused on stromal interactions in cancer (49). When cancer cells are cocultured with fibroblasts the basic composition of ECM is quite similar as in the corresponding tumors (20). In a spheroid most of the ECM proteins are produced by fibroblasts, whereas cancer cells have usually a minor contribution only (20). Here we used eight different human cancer cell lines, namely SKOV3 and Caov3 (ovarian adenocarcinoma), A549 (lung carcinoma), MDA-MB-231 and MCF7 (breast adenocarcinoma), UT-SCC2 (head and neck squamous cell carcinoma), Caco2 (colon adenocarcinoma), and HepG2 (hepatocellular carcinoma), which were cocultured (1:1) with human cancer associated fibroblasts (CAF). Spheroids were grown for 5 days in serum-free medium. Ascorbic acid (50 µg/ml) was added daily to allow the synthesis of proper, hydroxyproline and hydroxylysine containing collagen. After lysis of the spheroids, the proteins in the insoluble fraction were isolated and digested, and the resulting peptides were analyzed by mass spectrometry. We could not recognize any citrullinated peptides derived from ECM proteins ( Table 3 ). Thus, we concluded that factors derived from cancer cells or fibroblasts may not be responsible for extracellular citrullination in tumors.

In another set of experiments, we used a ninth cancer cell line, UT-SCC-7 cutaneous squamous cell carcinoma cells, together with human skin fibroblasts (SFB) and UT-SCC-2 cells together with gingival fibroblasts (GFB) in spheroid cocultures. Spheroids were grown for 6 days in serum-free medium. Ascorbic acid was added daily. When ECM protein related peptides were analyzed by mass spectrometry, we did not identify any citrullinated peptide derived from matrisome proteins ( Table 4 ).

We also treated the spheroids with PAD4 to enzymatically citrullinate ECM proteins in vitro. We could not detect any citrullinated matrisome-derived peptides after PAD4 treatment, except for one peptide derived from fibroblast growth factor binding protein 1 (FGFBP1), which is not an integral part of ECM ( Table 4 ). We could detect from two- to seven-fold increase in the number of spectra matching to citrullinated non-matrisome-derived peptides after enzymatic citrullination ( Table 4 ), which indicates that PAD4 had functioned as expected. Classification of the citrullinated proteins in PAD4-treated spheroids by selected cellular compartment (CC) gene ontology terms using GOnet Tool (https://tools.dice-database.org/GOnet/) revealed that most of the citrullinated non-matrisome derived peptides in PAD4-treated spheroids were intracellular ( Supplementary Figure S1 ), which suggests that externally added PAD4 had citrullinated proteins derived from degraded or necrotic cells, which are common in spheroid cultures. The fact that externally added PAD4 could not citrullinate matrisome proteins in spheroids is in disagreement with the fact that PADs can citrullinate e.g., structural ECM proteins in chronic inflammation. However, at the site of inflammation ECM is often intensively degraded by proteolytic enzymes, such as matrix metalloproteinases (MMPs), and the intact matrix in spheroids may protect matrisome proteins against PAD4.

Our observations based on spheroid cell cultures failed to find any evidence that cancer cells or stromal fibroblasts could actively citrullinate matrisome proteins. Previously, citrullination of extracellular proteins has been well described during chronic inflammation (9). Inflammatory cells are also frequently found in malignant tumors. Furthermore, some of the proteins that we found to be most frequently citrullinated in cancer data sets, e.g., fibrinogen and fibronectin, are often citrullinated in inflammation (11, 46). Thus, it is reasonable to speculate that the citrullination of matrisome proteins in cancer may also be due to PADs released by inflammatory cells. To test this hypothesis in the case of colorectal cancer that has been suggested to be dependent on citrullination for metastatic growth (4), we selected only the experiments where several primary or metastatic colorectal cancer samples had been analyzed individually (22). The data from these samples contained spectra matching to inflammation-related proteins, such as PAD enzymes, MMPs (MMP1, MMP2, MMP7, MMP8, MMP9, MMP11, MMP12, MMP14, MMP19 and MMP23) and also inflammatory cell-specific receptors, namely leukocyte integrins. We found a statistically significant correlation between citrullinated matrisome PSMs and PAD2 + PAD4 PSMs (p=0.03), between citrullinated matrisome PSMs and MMP PSMs (p=0.004) and between citrullinated matrisome PSMs and leukocyte integrin PSMs (p=0.03) ( Figure 3A ). Specifically, it appears that an exceptionally high citrullination level is connected to high level of inflammation markers.

To confirm that the relation between citrullination and inflammation markers is not restricted to only one dataset, we analyzed a subset of tumor tissue samples (n = 22) of CPTAC colon adenocarcinoma dataset PDC000116 (50) for citrullination and used MaxQuant software (41) to quantitate the TMT labelled samples. In this subset, the samples with high citrullination level had also high level of leukocyte integrins and MMPs ( Figure 3B ). There seemed to be no clear relation between citrullination and PADs, which may be explained by the fact that, unlike the samples in Figure 3A , CPTAC samples have not been enriched for ECM and thus contain also intracellular PADs. Nevertheless, these results suggest that in malignant tumors citrullination of matrisome proteins is associated to inflammation-related release of PADs and matrix degradation rather than direct action of cancer cells or stromal fibroblasts. Furthermore, the level of citrullination varies substantially between metastasis samples of colorectal cancer ( Figure 2 ), which further support the idea that citrullination is a side effect of inflammation rather than a prerequisite for metastatic growth.