HATRIC-based identification of receptors for orphan ligands.

Cellular responses depend on the interactions of extracellular ligands, such as nutrients, growth factors, or drugs, with specific cell-surface receptors. The sensitivity of these interactions to non-physiological conditions, however, makes them challenging to study using in vitro assays. Here we present HATRIC-based ligand receptor capture (HATRIC-LRC), a chemoproteomic technology that successfully identifies target receptors for orphan ligands on living cells ranging from small molecules to intact viruses. HATRIC-LRC combines a click chemistry-based, protein-centric workflow with a water-soluble catalyst to capture ligand-receptor interactions at physiological pH from as few as 1 million cells. We show HATRIC-LRC utility for general antibody target validation within the native nanoscale organization of the surfaceome, as well as receptor identification for a small molecule ligand. HATRIC-LRC further enables the identification of complex extracellular interactomes, such as the host receptor panel for influenza A virus (IAV), the causative agent of the common flu.

controlled, this control seems random. New datasets with compound-free virus competition or quenched virus would seem better controls. 2. Fig. 2e. Compound labeling of viruses can strongly affect infectivity. The authors should perform control experiments, in which they compare titers of virus before and after labeling. Moreover the effect of labeling on the specific infectivity (infectious particle / genome copy number) should be measured. 3. Fig. 2e. Compound labeling of small enveloped viruses such as influenza A virus may affect its entry route. The authors should experimentally demonstrate that the entry pathway into A549 cells is not altered after compound labeling of the virus particles using inhibitory compounds and/or imaging techniques. 4. Fig. 2 and lines 316-331: The authors should explain the filtering for cell surface molecules in the main manuscript, not only in the methods. They should disclaim, which fraction of the identified proteins was cell surface associated according to e.g. GO annotation. 5. Fig. 2e. Why was a nuclear pore protein (NUP210) identified despite the surfaceome filtering? What is the leakiness of the method towards cytoplasmic or nuclear proteins? 6. Line 177: Multiple RNA interference (RNAi) screens on influenza have been published, with some overlap. It is recommended that the authors discuss in more detail why on the one hand the published RNAi hits were not discovered in their HATRIC experiment and on the other hand, why their MS hits were vice versa not previously identified in any of the influenza host factor searches. 7. A differentiated discussion on the limitations of the technology is missing. Can any small ligand be linked to the HATRIC compound without affecting receptor affinity? What are the requirements of organic compounds to be successfully fused to HATRIC by synthesis? 8. The full MS datasets should be disclosed in supplementary tables and deposited in public online repositories such as the EMBL/EBI IntAct database. In particular for the influenza A virus experiment.
Minor comments: 1. Full protein names are not mentioned. Please write out the full names at first mentioning of a protein abbreviation, such as FOLR1. 2. Supplementary table 3: The human surfaceome should be presented with separate columns for gene name, protein name and Uniprot accession number for easier accessibility. 3. Fig. 2e,f. The gene/protein names do not match between Fig. 2e, Fig. 2f, Tab. S2 and Tab S4. If the authors decide to use protein names in Fig. 2e and gene names in Fig. 2f, it is advisable to include both -protein names and gene names -in Tab S2 and S4 to allow the reader to match the datasets. 4. Certain proteins, which were silenced (Fig. 2f), are not included in Tab. S2 or annotated differently. Examples are SLC19A1, NUP210, ABCC4.

Reviewer #3 (Remarks to the Author):
In the manuscript from Sobotzki et al., the authors demonstrate their development of nextgeneration LRC method. Having been the leading developers of the first-generation reagents, TRICEPS-LRC, the Wollscheid laboratory is well-suited to evolve this useful technology for improved coverage, applicability, and sensitivity. The updated methodology, termed HATRIC, still employs the key step of receptor sugar alcohol to aldehyde periodate oxidation, and subsequent coupling to the hydrazine-containing probe. However, the authors optimized the periodate oxidation to achieve high efficiency at neutral pH. In addition, the authors introduced Click chemistry in the HATRIC reagent. These optimizations directly contribute to the improved sensitivity of the approach, with a minimum requirement of between 1 -2 orders of magnitude less cellular material. The authors experimentally demonstrated the results of HATRIC-LRC with 1 million cells, though as mentioned in the comments below, the explanation of this experiment in the manuscript could be improved. The work nicely demonstrates the broad application of the method to a range of ligands, including the small molecule folate, the polypeptide EGF, and the intact virus, influenza A. The authors convincingly demonstrated that their technology could identify biologically relevant cell surface receptors of IAV by validation with siRNA knockdown of candidate IAV cell surface receptors during infection. However, as mentioned in the main comments section, the authors did not fully discuss why none of the known IAV receptors were identified.
• Receptor identification of orphan ligands remains a challenging area and 22 advancements in this area would be of interest to many bio-researchers. The 23 HATRIC crosslinker itself is quite similar to the TRICEPS reagent previously 24 described -the major functional difference being the replacement of the biotin group 25 for an azide which would allow purification on an affinity resin, without additional 26 protein contamination from streptavidin. HATRIC also has a different protecting group 27 on the hydrazide functional group than TRICEPS, though the authors do not mention 28 whether this has any functional consequences, or was simply a choice made for 29 ease-of-synthesis or other considerations. Several of the major advantages of 30 HATRIC that are highlighted in the manuscript by the authors have been previously 31 described in work on the ASB crosslinker (reference 4 in this manuscript) -the ASB 32 procedure as described also allows identification based on tryptic peptides from the 33 entire protein, rather than focusing on the N-glycopeptides. Although not discussed in 34 detail, the ASB procedure described also appears to use a catalyst and ligand 35 binding is at pH 8.0. Since these appear to be the major advantages cited by the 36 authors for HATRIC, the novelty aspect of HATRIC over TRICEPS may be lessened. 37 There would be definite advantages of HATRIC over ASB -including the simplified 38 ligand labelling and the enrichment using alkyne-beads rather than streptavidin 39 beads. The authors have not described the previous work on ASB in this manuscript, 40 nor compared it to HATRIC. 41 42 • We would like to thank the reviewer for the valuable suggestion to add 43 information about similarities and differences compared to ASB. In principle, it 44 is very good for the community that complementary technologies are 45 available to decode ligand receptor interactions. There is a wealth of ligands 46 out there in search for receptors and having different strategies and 47 chemistries available is certainly of advantage for the community. The 48 HATRIC-based LRC strategy is indeed a protein-based workflow and this is 49 similar in parts to the ASB strategy. However, the chemistry used for the 50 HATRIC-based approach is novel and makes the difference. The next 51 generation HATRIC sporting the acetone-protected hydrazide functionality in 52 combination with click-chemistry and the catalyst, allowing for reactions in 53 different ligand receptor interaction suitable pH ranges, enables now new 54 applications and delivers results with unprecedented sensitivity, as shown in 55 the manuscript. Furthermore, this new combination of chemistries within the 56 HATRIC-LRC workflow allows for the first time a significant reduction of 57 cellular starting material needed for the discovery of receptors compared to 58 ASB and TRICEPS-based LRC workflows. HATRIC-LRC can be routinely 59 performed with 1x 150mm dish vs. 5-7x 150mm plates in ASB and 4x 150mm 60 plates in TRICEPS-LRC. In addition, the catalyst-enhanced HATRIC-LRC 61 never required us to increase the sodium periodate concentration beyond 1.5 62 mM (compared to up to 10mM in ASB) which is a clear advantage in respect 63 to cell viability during the process of labeling, especially with primary cells. 64 Finally, HATRIC-LRC -for the first time -enabled the receptor 65 capture/identification with a small molecule compound which was never 66 before demonstrated on cell surface proteins. 67 ○ We added new text as detailed below to the introduction and 68 discussion section and after completion of the suggested edits, the 69 revised manuscript has benefitted from an improvement in the overall 70 presentation and clarity. 71

72
• Regarding your comment related to the functional consequences of changing 73 the hydrazide protection group in HATRIC we would like to provide you with 74 more context and insights. Investigating the pH as a critical factor during the 75 receptor capture reaction, we tested the impact of different protection groups 76 on the yield of hydrazone formation on live cells at higher pH (pH 7.6). We 77 employed the first generation of TRICEPS compounds bearing a NHS group 78 coupled to a biotin and a hydrazide group and studied two different TRICEPS 79 versions bearing either a a trifluoroacetyl-protected (PbP Figure 1A) or 80 acetone-protected (PbP Figure 1B) hydrazide. When comparing hydrazone 81 formation of these two TRICEPS versions on the cell surface, we detected  82  much brighter cell surface labeling with the acetone-protected hydrazide-83   containing compound compared to the Tfa-protected under the same  84 conditions (visualized by Streptavidin-FITC) at both pH 6.5 and pH 7.6 on live 85 A2.01 cells (PbP Figure 1C). These experiments, conducted in the absence 86 of the catalyst, indicate higher reactivity in the cell surface micro-environment. 87 The possibility to conduct the experiments at different pH levels, supported in 88 addition kinetically by the the catalyst, turned out to be a major advantage for 89 studying pH-sensitive ligand-receptor interactions, such as between folate 90 and folate-receptor alpha: Folate-based receptor capturing was never 91 successful at pH 6.5, but only at pH 7.4. technology partly overcame these difficulties and enabled the identification of 101 ligands for orphan N-glycoprotein-receptors using the tri-functional reagent 102 TRICEPS (Frei et al. 2012(Frei et al. , 2013 and modifications thereof in ASB (Tremblay 103 and Hill 2017). Application of TRICEPS-LRC and ASB in different biological 104 systems, however, revealed the need to redesign the first-generation 105 technologies: TRICEPS-LRC was intentionally designed to enable the 106 identification of ligand-bound receptors solely based on formerly N-107 glycosylated peptides. O-glycosylated receptors and N-glycosylated receptors 108 whose deamidated peptides were not detectable by mass spectrometry were 109 eventually missed by this strategy. However, this peptide-based strategy 110 benefitted from the ability and quality to be able to filter for deamidated 111 receptor peptides as indicators of direct TRICEPS-crosslinking and ligand-112 binding. In contrast, in ASB, tryptic digestion is performed directly on 113 Streptavidin beads, which enables protein-level affinity purification, enabling, 114 in principle, the identification of receptors through non-glycopeptides. 115 However, direct digestion of proteins bound to Streptavidin beads leads to 116 major contaminations with streptavidin peptides, impairing identification and 117 label-free quantification of receptor peptides. Furthermore, ASB requires 118 performing a two-step reaction in order to couple the ligand to the cross-119 linker, and biotin transfer from ligand to receptor is mediated by reduction of a 120 disulfide bond, making its application sensitive to reductive environments. 121 Furthermore, the ASB strategy utilizes a catalyst to catalyze oxime formation 122 on the cell surface at pH 8. Similar to first generation TRICEPS-LRC, ASB 123 requires high amounts of starting material (50 million cells or 5-7 150mm 124 plates) and captures ligand-receptor interactions at pH 8 compared to pH 6.5 125 for TRICEPS LRC. The pH of the microenvironment directly influences the 126 affinity between a ligand and its receptor, exemplified by ligands that are 127 internalized upon receptor binding: The affinity for the receptor is high at pH 128 7.4 on the surface of living cells, but decreases upon acidification (pH 6.5) in 129 the endosome, leading to release of the ligand from the receptor. A prime 130 example of this is folate, which has an affinity for folate receptor alpha 131 (FOLR1) that is 2000 times lower at pH 6.5 than at pH 7.4 (Yang et al. 2007 (Khan et al. 1999). Aniline-140 derived water-soluble catalysts have been described that substantially 141 improve catalysis of hydrazone formation, but none had been tested in 142 biological systems (Crisalli and Kool 2013). Evaluation of a number of aniline 143 derivatives regarding their solubility, cytotoxicity and capability to enhance 144 hydrazone formation between aldehydes on cell surface proteins and the 145 HATRIC-hydrazide on living cells led to identification of 5-methoxyanthranilic 146 acid (5-MA, Fig. 1c, Supplementary Fig. 1). 5-MA catalyzed hydrazone 147 formation at a non-toxic concentration at pH 7.4 more efficiently than 2-148 amino-4,5-dimethoxy benzoic acid (ADA). Additionally, replacing the original 149 Trifluoroacetyl-protection group of TRICEPS by an acetone-derived protection 150 group in HATRIC enabled higher yield of hydrazone formation on live cells 151 (data not shown). Last, we confirmed that under the chosen conditions, 152 HATRIC does not penetrate cells avoiding contamination with intracellular 153 proteins ( Supplementary Fig. 2). 154

155
The manuscript is well-written and clearly presented. 156 157 • Thank you very much & the comment is very well appreciated. 158 159 Scientifically and statistically the work presented in this manuscript appears to be generally 160 solid and interesting. However, some details are lacking and the discussion/interpretation of 161 the experiments and methods is quite limited, perhaps due to space constraints (? abundances of all identified proteins without any filtering. The quantitative 208 comparison will help to hide the majority of "unspecific" proteins in the scatter 209 plot as not specifically enriched, as these are somewhat equally identified 210 across samples. This approach may be sufficient to identify highly abundant 211 or large cell surface proteins or cell surface proteins that are highly 212 soluble/MS detectable, but it highly neglects proteins that are small, of lower 213 abundance or have many, hardly soluble transmembrane-spanning peptides. 214 It is well known that cell surface proteins are notoriously difficult to identify by 215 MS and our strategy enables the identification of hundreds of cell surface 216 proteins using a chemoproteomic strategy. Therefore, this approach is 217 inadequate when one is interested in these typically underrepresented 218 species. To increase the informative value of such screens, we recommend to 219 filter HATRIC-LRC data sets with our surfaceome filter to enable the 220 identification of low abundant proteins that are typically overlooked and push 221 them over the significance value against the background of "nonspecific" 222 proteins with many peptides. Taken together, filtering doesn't change the 223 fold changes of proteins across samples, but significantly affects p-224 values. At the same time, the screening protocol is by no means 100% 225 efficient and considerable losses of peptides are expected during glycan 226 oxidation, aldehyde capturing, affinity purification and tryptic peptide release, 227 as well as peptide purification. Therefore, in our experience, the "nonspecific" 228 peptides are essential to "chaperone" the membrane protein-derived peptides 229 to the MS. Further, we would like to point out politely that filtering is commonly 230 performed in screens, such as filtering for proteins that are identified with a 231 minimum number of peptides (ASB) or that carry specific sequence motifs 232 such as the N[115]-X-S/T signature in TRICEPS-LRC. In cases, where no cell 233 surface filter list is available (e. g. more exotic mammals), we recommend to 234 include one further step in the protocol and release N-glycosylated peptides 235 from the beads using PNGase F and limit quantification to proteins that were 236 identified in the N-glycopeptide fraction. 237 whereas only 1 peptide was quantified and scored for EGF). In order to 277 overcome this bias, we used a filter for known and predicted cell surface 278 proteins prior to statistical scoring to rescue receptor candidates where most 279 peptides are hardly detectable via MS (e. g. due to decreased solubility) 280 ( Supplementary Fig. 3 Table 3). Reports of direct interactions between these 288 proteins and EGF are not available, but it was shown before that SLC16A3 289 co-locates with CD147 in breast cancer cells   there any data to suggest that the TRICEPS method would not work with 1 million MDA-231 307 cells with these ligands? 308 309 • We tried to identify EGFR and TFR1 using anti-EGFR antibody and holo-310 transferrin (hTF) on 1 million MDA-MB-231 by TRICEPS-LRC, but failed 311 repeatedly (PbP Fig. 5, PbP Fig. 7). In parallel, we conducted TRICEPS-LRC 312 on 50 million MDA-MB-231 cells and successfully identified EGFR as receptor 313 for EGF (PbP Fig. 4, PbP Fig. 6). However, we were also not able to identify 314 TFR1 for receptor of hTF in this particular experiment. This might be 315 explained by the fact that transferrin is released from the cell at pH 5. purification, more than one peptide is commonly identified per protein, such 353 as exemplified by EGFR (Supplementary Fig. 4). Therefore, we investigated 354 the HATRIC-LRC detection limit with respect to the amount of starting 355 material needed for successful receptor identification. From as little as one 356 million MDA-MB-231 cells per sample, we were able to unambiguously 357 identify EGFR as the receptor for HATRIC-coupled anti-EGFR antibody and 358 transferrin receptor protein 1 (TFR1) as the receptor for HATRIC-coupled 359 Holo-transferrin (TRFE) (Fig. 2b) Fig. 1c, Supplementary Fig. 1) 6. Interpretation of the alternative candidate EGF receptors needs to be handled with 439 some caution. Is there any evidence that some of these 'candidate receptors' truly bind to 440 EGF? Is it possible that these proteins may simply be co-localized on the cell surface with 441 the true receptor, leading to enriched proximity-based crosslinking via HATRIC? As written, 442 some biologists may mistakenly take the proteins in Supp Table 1 as 'proven' EGF  443 receptors.  Table 3-4). Reports of direct interactions between these 485 proteins and EGF are not available, but it was shown before that SLC16A3 486 co-locates with CD147 in breast cancer cells  For the viral work, an interesting follow-on functional study is shown. For this work, 508 the authors should show the level of depletion achieved by the siRNAs for each of these 509 targets. For interpretation of this data, it is important for the reader to know if all of the 510 candidate receptors were successfully depleted and, if so, by how much? 511 512 • We would like to thank the reviewer for the comment and we have addressed 513 this now in our revised manuscript. To this end, we performed real time RT-514 PCR for all 21 genes and quantified the gene depletion level (Pbp Figure  515 10/ Supplementary Figure 8). The experiment was repeated twice with 516 similar results. Twenty genes showed above 70% depletion of the respective 517 mRNA i.e. >90%, 9 genes; >80%, 6 genes; >70%, 5 genes. A single gene, 518 CRTAP, showed no reduction upon siRNA treatment. We conclude that IAV 519 infection in CRTAP siRNA-treated cells were reduced to unknown off-target 520 effects (see original manuscript (Fig. 2F)). Thus, we removed CRTAP from 521 the infection data figure (Fig. 2F). impact IAV entry, we depleted A549 cells of 21 of these proteins using short 542 interfering RNA (siRNA) and analyzed infection efficiency. siRNA-mediated 543 depletion of more than 70% was confirmed by real time RT-PCR in 20 genes. 544 We excluded cartilage-associated protein (CRTAP) from further analysis as 545 siRNA treatment failed to deplete it (Supplementary Fig. 8). Depletion of four 546 proteins, phospholipase D3 (PLD3), ribophorin I (RPN1), folate transporter 1 547 (SLC19A1) and vesicular integral-membrane protein VIP36 (LMAN2) reduced 548 IAV infection by more than 50% relative to cells treated with control siRNA 549 (Fig. 2f)  Minor comments: 582 1. The information provided on the MS results is minimal. While it is great that the MS 583 raw files have been made available, some minimal information should be provided in the 584 manuscript/supplementary info. For example, no peptide-level results are shown or provided. 585 At a minimum, the number of unique peptides identified/quantified for each protein should be 586 provided in the manuscript. Ideally, some information on the quantitative variability seen 587 between different peptides from the same protein should also be provided. 588 589 • We added tables containing the complete information on peptides used for 590 quantification for each data set (Progenesis output tables, Supplementary  591   tables 1A, 4A, 5A, 6A, 7A, 9A) and the outcome of our statistical analysis 592 containing all information necessary to create volcano plots (Supplementary  593   tables 1B, 4B, 5B, 6B, 7B, 9B). This information will provide a transparent 594 overview on the quality of the data. 595 596 2.
More details on the statistical methods used would be helpful. How were protein-level 597 p-values determined? How was quantitative data from individual peptides combined? How 598 were the different technical replicates used for this calculation? What modules from MSstats 599 were used? 600 601 • Thank you for noticing, it was indeed very short and we rectified it now. Major comments: 675 676 1. Fig. 2e. Why was insulin used as control ligand? While the first three experiments 677 were well controlled, this control seems random. New datasets with compound-free virus 678 competition or quenched virus would seem better controls. 679 680 • This is a valid and appreciated argument raised from this reviewer and we 681 agree with the reviewer that on the first glance, the choice of this ligand 682 appears random. However, we would like to politely point out, that we 683 deliberately chose insulin as a technical control ligand in the virus-receptor 684 capture experiment. in contrast to the other experiments reported in the 685 paper, we didn't know which receptors to expect for influenza. Given the 686 rather long protocol and the risk of bias in the result due to differential sample 687 processing, we wanted to use a ligand with known receptor specificity that 688 would allow us to come to a distinct decision if the experiment was successful labeling. Moreover the effect of labeling on the specific infectivity (infectious particle / 700 genome copy number) should be measured. 701 702 • Please see a combined response below. 703 704 3. Fig. 2e. Compound labeling of small enveloped viruses such as influenza A virus 705 may affect its entry route. The authors should experimentally demonstrate that the entry 706 pathway into A549 cells is not altered after compound labeling of the virus particles using 707 inhibitory compounds and/or imaging techniques. 708 709 710 • The response is combined for the above two points: We thank the reviewer 711 for these comments. It is indeed possible that compound labeling with 712 HATRIC (albeit at 2 HATRIC molecules per virion) could affect infectivity and 713 could alter the entry pathway of IAV particles. We performed IAV endocytosis, 714 primary infection, and multi-step growth assays using IAV labeled with 715 HATRIC or incubated with buffer alone (PbP Figure 12 Figures 6 & 7). We 763 conducted H3N2-based HATRIC-LRC on to 20 million human lung 764 adenocarcinoma (A549) cells and compared to the control ligand insulin. We 765 identified 24 virus-interacting candidates (Fig. 2e, Supplementary Table 7 filtering? What is the leakiness of the method towards cytoplasmic or nuclear proteins? 778 779 • As pointed out earlier, we filter our data for proteins that are annotated to be 780 located at the cell surface. We identified NUP210 here because it is in this  , Fig. 1c). 5-MA catalyzed hydrazone formation at a non-toxic 804 concentration at pH 7.4 more efficiently than 2-amino-4,5-dimethoxy benzoic 805 acid (ADA) (Fig. 1c, Supplementary Fig. 1). Additionally, replacing the 806 original Trifluoroacetyl-protection group of TRICEPS by an acetone-derived 807 protection group in HATRIC enabled higher yield of hydrazone formation on 808 live cells (data not shown). Last, we confirmed that under the chosen 809 conditions, HATRIC does not penetrate cells, to avoid contamination with 810 intracellular proteins (Supplementary Fig. 2) Fig. 9). We also had 4 strong 890 hits (i.e. increased or decreased infection by more than 70%) out of 20 891 validated genes -a hit rate of 20% -which is considerably higher compared 892 to the genome-wide screens (-1% Certain proteins, which were silenced (Fig. 2f) termed HATRIC, still employs the key step of receptor sugar alcohol to aldehyde periodate 974 oxidation, and subsequent coupling to the hydrazine-containing probe. However, the authors 975 optimized the periodate oxidation to achieve high efficiency at neutral pH. In addition, the 976 authors introduced Click chemistry in the HATRIC reagent. These optimizations directly 977 contribute to the improved sensitivity of the approach, with a minimum requirement of 978 between 1 -2 orders of magnitude less cellular material. The authors experimentally 979 demonstrated the results of HATRIC-LRC with 1 million cells, though as mentioned in the 980 comments below, the explanation of this experiment in the manuscript could be improved. 981 The work nicely demonstrates the broad application of the method to a range of ligands, 982 including the small molecule folate, the polypeptide EGF, and the intact virus, influenza A. 983 The authors convincingly demonstrated that their technology could identify biologically 984 relevant cell surface receptors of IAV by validation with siRNA knockdown of candidate IAV 985 cell surface receptors during infection. However, as mentioned in the main comments 986 section, the authors did not fully discuss why none of the known IAV receptors were 987 identified. 988 989 Overall, this is a strong methodological study with significant application to biomedical and 990 pharmaceutical research, particularly in contributing to the characterization of orphan 991 receptors. The authors do have a few outstanding and several minor points to address; 992 however, if these can be addressed, I would recommend the manuscript for publication. 993 994 Main Points 995 1.
A general main point is the lack of discussion related to novel identified candidates or 996 lack of identification for known candidates in the case of IAV. For instance, in addition to 997 identifying the known receptors for the EGF and folate ligands, the authors found several 998 other putative candidates, which the authors did not discuss. 999 1000 • We would like to thank the reviewer for the valuable suggestion to add 1001 information about putative receptor candidates for the ligands EGF and folate. 1002 The lack of some details is mainly due to the initial space constraints of the 1003 format. We now added more details in the text and in the supplementary 1004 information. 1005 is used for stimulation experiments) or (4) the identified candidate is a false positive.
Our experiments do not allow us to delineate right away which type of interaction was observed, but the validation experiments and the cited data clearly underline the relevance of the identified proteins. The analysis pipeline was optimized to allow for identification and ranking of receptor candidates. However, the resulting data have to be analyzed carefully and more stringent receptor spaces can be defined based on the identification of positive control receptors or the ligand (e.g. EGF). Identified candidates need validation in tailor-made follow-up experiments, such as siRNA-based approaches. These approaches cannot be generalized and for every LRC application the type of follow-up experiment will depend on the type of ligand, the biological context, and the tools available for the system under study. However, we would also like to point out that the biological relevance of the neighboring proteins is not to be underestimated either. Proteins that are in close proximity of the target receptor might interfere with the activity of the actual target and are therefore relevant for future studies of the lateral cell surface interactome. HATRIC-LRC could potentially also be used to generate candidates for such studies -another exciting application of HATRIC-LRC for life science research.

Changes to the manuscript:
[…] We incubated the folate-HATRIC conjugate with 20 million HeLa Kyoto cells at pH 7.4.
In the control, we added six-fold excess of unmodified folate. We detected interactions with FOLR1 and with a small set of further receptor candidates (Fig. 2c, d; Supplementary Table   7). None of these receptors were previously described to interact directly with folate. At the same time, we didn't identify any other known folate receptors. We speculate that other folate receptors (e. g. FOLR2) were not identified as their affinity towards folate is lower than the affinity of FOLR1 or because they are not expressed in HeLa Kyoto cells 19 . Related approaches studied methotrexate-based labeling of FOLR1, but didn't investigate if the compound also binds to other proteins 18 .
[...] Applying this filter prior to statistical analysis, we correctly identified EGF significantly enriched and identified five other EGF receptor candidates that have not been described before (Supplementary Table 3-4), namely monocarboxylate transporter 4 (SLC16A3), filamin-A (FLNA), peroxisomal 3-ketoacyl-CoA thiolase (ACAA1), transmembrane emp24 domaincontaining protein 7 (TMED7) and sarcoplasmic/endoplasmic reticulum calcium ATPase 1 (AT2A1) (Supplementary Table 3-4). Reports of direct interactions between these proteins and EGF are not available, but it was shown before that SLC16A3 co-locates with CD147 in breast cancer cells  , which in turn is associated with EGFR in similar lipid