{"paper_id":"2d550d86-df96-46fb-b03c-3ce479d50d62","body_text":"Tumour specimen cold ischemia time impacts molecular cancer\nd\nrug target discovery\nSilvia von der Heyde 1, Nithya Raman 1, Nina Gabelia 1, Xavier Matias-Guiu 2, Takayuki Yoshino3, Yuichiro\nTsukada4, Gerry Melino 5, John L. Marshall 6, Anton Wellstein 7, Hartmut Juhl 1, and Jobst Landgrebe 1,8\n1Indivumed GmbH, F alkenried 88d, 20251 Hamburg, Germany\n2Dept. of Pathology , Hospital Universitari Arnau de Vilanova, Universitat de Lleida, IRBLLEIDA, Lleida, Spain\n3Dept. of Gastrointestinal Oncology , National Cancer Center Hospital East (NCCE), Kashiwa, Japan\n4Dept. of Colorectal Surgery , National Cancer Center Hospital East (NCCE), Kashiwa, Japan\n5Dept. of Experimental Medicine, University T or V ergata, Rome\n6The Ruesch Center for the Cure of Gastrointestinal Cancers, Georgetown University , W ashington, D.C., USA\n7Dept. Oncology & Pharmacology , Lombardi Comprehensive Cancer Center, Georgetown University , W ashington, D.C., USA\n8T o whom correspondence should be addressed ( landgrebe.jobst@indivumed.com)\nTumour tissue collections are used to uncover pathways associated with\ndisease outcomes that can also serve as targets for cancer treatment, ideally\nby comparing the molecular properties of cancer tissues to matching normal\ntissues. The quality of such collections determines the value of the data and\ninformation generated from their analyses including expression and modifi-\ncations of nucleic acids and proteins. These biomolecules are dysregulated\nupon ischemia and decomposed once the living cells start to decay into inan-\nimate matter. Therefore, ischemia time before final tissue preservation is the\nmost important determinant of the quality of a tissue collection. Here we\nshow the impact of ischemia time on tumour and matching adjacent normal\ntissue samples for mRNAs in 1,664, proteins in 1,818 and phosphoproteins\nin 1,800 cases (tumour and matching normal samples) of four solid tumour\ntypes (CRC, HCC, LUAD and LUSC NSCLC subtypes). In CRC, ischemia\ntimes exceeding 15 minutes impacted 12.5% (mRNA), 25% (protein) and\n50% (phosphosites) of differentially expressed molecules in tumour versus\nnormal tissues. This hypoxia- and decay-induced dysregulation increased\nwith longer ischemia times and was observed across tumour types. Interest-\ningly, the proteomics analysis revealed that specimen ischemia time above 15\nminutes is mostly associated with a dysregulation of proteins in the immune\nresponse pathway and less so with metabolic processes. We conclude that\nischemia time is a crucial quality parameter for tissue collections used for\ntarget discovery and validation in prognostic cancer research.\n1\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n1 Introduction\nT\nhe identification rate of novel protein targets in the context of solid cancer therapy has\ndecreased over the last ten years.\nOne of the major concepts in identifying functional proteins as oncogenic targets is\nthe detection of alterations that lead to aberrant signalling in cancer-relevant pathways.\nA core principle of today’s onco-pharmacological targeting approach is detecting and ap-\nplying therapeutics that antagonize oncogenic protein function and thereby significantly\nimprove overall survival [1]. An important strategy to systematically identify new target\nproteins for cancer treatment is to build a cancer registry combining tissue and meta-\ndata collection, a strategy pursued, for example, by the TCGA consortium. 1 To this\nend, cancer and ideally also matching normal adjacent tissue are collected and analysed\nto compare the molecular characteristics of the two tissue types. The two parameters\nwhich have the strongest impact on the quality of tissue registries are the asservation\nmethod and the so-called cold ischemia time (in short ‘ischemia time’), which denotes\nthe time it takes to freeze the tissue for permanent storage after its removal from the\nbody during surgery.\nThere is an ongoing debate about the consequences of ischemia on the quality of\nbiomolecules, especially in the context of characterising the molecular entirety of tissues\nor cells (so-called ‘omics’). The corresponding findings vary widely [2, 3, 4] due to\ninsufficient numbers of samples which leads to an insufficient power to detect the impact\nof ischemia time on the molecular composition of the materials. Hence, the impact of\nischemia time on the molecular characteristics of the tissue under investigation remains\nunclear. Unlike any previous study, we use adequate sample sizes and statistical methods\nto investigate the impact of ischemia time on target discovery.\nHere, we focus on the impact of ischemia time on differential expression of mRNAs,\nproteins and phosphoproteins using a carefully curated collection of patient-derived fresh-\nfrozen tissue samples and related multiomic data. This data base contains the breadth\nof molecular characteristics of tumour and normal adjacent tissues from multiple can-\ncer types. Specimens were analysed at the genomic, transcriptomic (TRX), proteomic\n(PTX) and phosphoproteomic (PPX) level with focus on colon cancer (CRC). Analyses\nwere expanded to hepatocellular carcinoma (HCC) and non-small cell lung epithelial can-\ncer (NSCLC), covering lung adenocarcinoma (LUAD) and lung squamous cell carcinoma\n(LUSC).\nTo generate a baseline in our studies we set an initial filter to capture biomolecules\ndifferentially expressed in the group of samples with shortest ischemia time available\n(t < 10 min.). We then analyse the changes in expression of these biomolecules over\ntime with an emphasis on the impact of ischemia time on the target identification process,\nrather than trying to model the tissue decay under ischemia.\nWe found that DNA characteristics are to a large extent unaffected even by the longest\nischemia times of samples collected. In contrast, the relationship of tumour and normal\ntissue mRNA, protein and phosphoprotein expression is affected at increasing severity\n1h ttps://www.cancer.gov/tcga\n2\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n(mRNA < p rotein < phosphoproteins). Based on the analyses we propose an ischemia\ntime cut-off threshold at 12 minutes that enables a sufficient amount of tissues to be\ncollected while avoiding a dilution of signals for biomolecules of interest.\n2 Results\n2.1 Cancer samples and differential expression\nData sets and samples used in the analysis are shown in Table 1, tumour stages are\nshown in Table 2. Details on the samples can be found in the methods section.\nWe defined the ischemia time reference group as the group of samples with ischemia\ntimes less than 10 minutes, the shortest time possible for the collection and storage of a\nsufficient number of samples. To obtain differential biomolecule expression for mRNAs,\nproteins and phoshoproteins for the shortest ischemia time group, we normalised the ex-\npression data and selected the differentially expressed biomolecules using non-parametric\nWilcoxon tests for paired samples to take into account the non-normal distribution of\nthe data [5] as described in the methods part. We adjusted the resulting P-values for\nmultiple testing and selected the biomolecule sets for mRNA, protein, and phosphopro-\ntein using an αfdr level of 0. 01 and an effect boundary using the 5th and 95th percentiles\nof the effect distribution. This yields 1,948 differentially expressed mRNAs, 794 proteins\nand 1,846 phosphosites on which we focussed in the CRC cohort. Analogously, we in-\nferred 1,870 differentially expressed mRNAs, 523 proteins and 388 phosphosites in the\nHCC cohort. In the LUAD cohort, we inferred 1,951 differentially expressed mRNAs,\n805 proteins and 2,217 phosphosites, and in the LUSC cohort, we inferred 1,950 differ-\nentially expressed mRNAs, 798 proteins and 1,919 phosphosites on which we focussed\nin the subsequent analyses.\n2.2 Differential expression over time in CRC\nWe performed a detailed analysis of differential (tumour versus normal adjacent tissue)\nexpression over time in CRC, and then applied the most relevant analyses to the other\ncancer types listed in Table 1.\nTo evaluate whether DNA sequences remain unaffected by ischemia times as reported\nby others [6], we analysed protein sequences for affecting mutations (PAM), the presence\nor absence of gene deletions or amplification as well as gene truncations by comparing\nthe shortest available ischemia time tissue group with the longer ischemia time groups\nfor each genomic locus (see Methods). At αfdr = 0.01, only 0.09% of the proteins showed\na signal in the PAM submodality, and no change was found in the other DNA-derived\nsubmodalities. This is likely based on random somatic difference in the genomic sequence\nof the patients of the various time groups (see Discussion).\nTo obtain an overview of the effect of ischemia time on the selected biomolecules, we\ninitially partitioned the samples into groups of T ′\n1 : t < 10, T ′\n2 : 10 ≤ t ≤ 14, T ′\n3 : 15 ≤ t ≤\n19, T ′\n4 : 20 ≤ t ≤ 24, T ′\n5 : t ≥ 25 minutes of ischemia time duration intervals. We then\nused hierarchical clustering to assess the changes in biomolecule expression for the three\n3\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nTable 1: Number of tumour samples per ischemia time interval (min.) with specific omic data\nCohort\nT1 : t < 10 T2 : 10 ≤ t ≤ 12 T3 : 13 ≤ t ≤ 15 T4 : 16 ≤ t ≤ 18 T5 : 19 ≤ t ≤ 20 t > 20 Total\nTRX PTX PPX TRX PTX PPX TRX PTX PPX TRX PTX PPX TRX PTX PPX TRX PTX PPX TRX PTX PPX\nCRC 181 188 188 223 223 217 111 102 101 44 47 47 18 19 19 36 54 54 613 633 626\nHCC 48 41 41 35 36 36 11 9 9 15 15 16 6 6 6 32 38 38 147 145 146\nLUAD 215 258 255 119 152 153 90 95 94 51 50 48 17 26 26 35 44 44 527 625 620\nLUSC 134 157 153 95 103 102 56 58 58 40 37 35 20 23 23 32 37 37 377 415 408\nTable 2: Number of tumour samples per cancer stage group with specific omic data\nCohort\nI II III IV NA Total\nTRX PTX PPX TRX PTX PPX TRX PTX PPX TRX PTX PPX TRX PTX PPX TRX PTX PPX\nCRC 65 60 60 229 251 248 195 201 199 121 118 116 3 3 3 613 633 626\nHCC 31 33 34 18 22 22 18 17 17 7 7 7 73 66 66 147 145 146\nLUAD 232 293 291 129 145 144 133 143 141 24 33 33 9 11 11 527 625 620\nLUSC 126 151 147 125 130 130 116 119 116 6 8 8 4 7 7 377 415 408\n4\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nTable 3: Alterations in phosphosite expression over time across tissue t ypes\nA: Long duration ischemia time groups\nS\nignificant ( αfdr = 0.05) effect for\nCluster Tissue Time Interaction\n1a 147 45 0\n2a\n169 63 3\n3a 177 61 6\n4a 369 129 2\n5a 317 101 12\n6a 188 68 8\n7a 54 11 2\n8a 52 16 0\n9a 97 35 0\n10a 242 76 6\nTotals 1812 605 39\nB: Refined ischemia time groups\nS\nignificant ( αfdr = 0.05) effect for\nCluster Tissue Time Interaction\n1b 115 16 0\n2b\n293 51 2\n3b 283 35 2\n4b 95 11 0\n5b 136 25 0\n6b 89 10 0\n7b 442 80 1\n8b 189 21 2\n9b 74 15 0\n10b 105 20 2\nTotals 1821 284 9\nomic modalities as described in the methods section. Figure 1 shows the results for the\nphosphoprotein modality for k = 10 clusters (the corresponding plots for the other two\nexpression modalities are shown in Supplementary Figure S1).\nThe effect magnitude and significance seem to increase for times over 20 minutes. To\nconfirm this impression, we tested for differences in differential expression over time in\neach modality using a 2 × 5 factorial ANOV A-like design with the tissue types (tumour\nand normal) as first factor and the time groups T ′\n1 . . . T′\n5 as second factor using the\nnon-parametric Scheirer-Ray-Hare test also assessing interaction effects between the two\nmain effects. Table 3A shows the phosphosites per cluster with significant tissue, time\nor interaction effects. The three effects on display are the two main effects of tissue and\ntime difference and the interaction effect of the two variables.\nTable 3 confirms that across all clusters (note that the clusters in both panels do\nnot contain identical phosphosites since the data vectors per phosphosite differ between\nthe panels) of the original time groups (panel A), 33% (605/1814) of the phosphosites\ndisplay a significant time effect. We therefore repartitioned the available samples into\nthe shortest ischemia group (samples with ischemia time shorter than 10 minutes), and\ninto further groups of 3 minute intervals from t ≥ 10 to 20 minutes ( T1 . . . T5) as shown\nin Table 1 in order to quantify which ischemia time point to use as cut-off (we call\nthese refined time groups). With these refined groups (Table 3, panel B) which exclude\nlonger ischemia times, the time effect halves to 16% (284/1823) of the phosphosites. 2\nThere are also less interaction effects. The trends for mRNA and protein modalities are\ncomparable, though less pronounced (see Supplementary Figure S1 and Supplementary\nTables S1 and S2). But even though the effect is least pronounced in mRNA, to perform\nvalid multiomics analyses, we need an ischemia time limit that is the same for all modal-\n2N ote that the differing denominator between the two groupings is due to reassigned samples and a\nsubsequently altered pattern of missing values.\n5\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nFigure 1: Differential expression of the selected phosphosites in 5-mi nute intervals (ab-\nscissa) grouped into ten clusters (ordinate) in CRC. Colours indicate log 2-fold\nmean expression differences (phosphosite-wise standardised) between tumour\nand normal tissue. Red: upregulation in tumour, blue: downregulation.\n6\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nTable 4: Loss L of biomolecules outside the ( Q2.5, Q97.5) interval\nModality Tissue Biomolecule loss [%] vs. T1\nT2 T3 T4 T5\nmRNA Normal 27 32 39 21\nT\numour 23 28 45 27\nProtein Normal 24 25 36 56\nT\numour 20 23 27 42\nPhosphoprotein Normal 49 64 72 94\nT\numour 46 60 64 81\nities. Thus, we aim for a limit that is optimal for the most sensitive ty pe of molecule\n(phosphate groups), which must be under 20 minutes.\nBefore investigating differential biomolecule expression, we analysed the distribution\nof the underlying expression levels in normal and cancer tissue samples of the groups\nT1 . . . T5. We specifically focused on biomolecules with extreme expression values which\ncould play an important role in cancer. We regarded biomolecules with expression values\noutside the (Q 2.5, Q97.5) interval as extremely expressed. We determined the relative\nloss of such extreme biomolecules of reference group T1 in other ischemia time groups.\nIn detail, we counted how many biomolecules with extreme expression values outside\n(Q2.5, Q97.5) of T1 were not detected outside the ( Q2.5, Q97.5) intervals of the other groups\nT2 . . . T5 any more. For example for T1 and T2, this relative loss is\nL = |δ(T1) \\ δ(T2)|\n|(δ(T1)| ·100,\nw\nhere δ computes the biomolecules outside the interval (Q 2.5, Q97.5) and | · · · |gives the\nset size. Table 4 shows the loss rates for tumour and normal tissues for the time group\naverage. The vanishing biomolecules fall from outside ( Q2.5, Q97.5) into the interval in\nthe time groups T2 . . . T5.\nThe loss of biomolecule sets in group T2 is striking for all types of analysed biomolecules,\nbut especially for phosphosites. Notably, the loss of the most highly expressed mRNAs\nis slower than for proteins and even more so than for phosphoproteins, where almost\n50% of the most highly expressed sites are lost in a short interval time between 10 and\n12 minutes of ischemia. In mRNA, there seems to be a recovery of the initial expression\npattern after the longest ischemia duration, an artefact which may be explained by the\ndegradation and altered detection of the molecules. This has to be taken into account\nwhen looking at differential biomolecule expression.\n2.2.1 Temporal patterns of refined time series\nTo evaluate the influence of ischemia on differential expression patterns, we next analysed\nthe shorter interval groups T1 . . . T5 using a Dirichlet process Gaussian process mixture\n7\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nTable 5: Pathways with differential protein expression Tk+ 1 − Tk\nNumber of proteins\nPathways Upregulated Downregulated\nImmune response 34 1\nM\netabolic processes 7 7\nTransport of molecules 5 2\nRegulation of cell signalling 8 3\nCell structure and adhesion 6 3\nTotal 60 16\nmodel (DPGP, details see Material and methods). This non-parameteri c time series\nanalysis technique jointly models data clusters with a Dirichlet process and temporal\ndependencies with a Gaussian process. 3\nFigure 2 shows the eleven clusters obtained in CRC for the protein modality, which we\nselected out of the three available expression modalities because it allows a deeper bio-\nlogical interpretation (analogous time series analyses can be generated for each modality\nand tumour type).\nWe expect various patterns of protein expression variance due to ischemia with proteins\nlosing or gaining expression levels under the effect of the ischemia-induced decay of the\ntissue.\nBiological cluster interpretation To interpret the biological meaning of the protein\nexpression patterns, we used a Wilcoxon test to identify proteins with differential ex-\npression from one time group to another ( µT2 − µT1, . . . , µTk+1 − µTk , k = 2 . . . 4) excluding\neffects inside the interval (2 − 1/2, 2\n1/2).\nWe then mapped the genes related to these differentially expressed proteins to KEGG 4\nand performed a GO enrichment analysis for biological pathways [7]. We could not find\ncluster-specific relevant biological patterns, maybe because ischemia regulation cannot\nbe revealed by DPGP-clustering on our pseudo-time-series (see footnote 5), but we still\nfound interesting overall patterns. The analysis revealed acute inflammatory response\nand metabolism as the most significantly up- and downregulated pathways, respectively.\nTable 5 shows the amount of significantly differentially expressed proteins per pathway.\nQuite surprisingly, the data show a strong upregulation of immune response related\nproteins. Notably, differential expression changes in the proteins involved in immunity\ncan already be observed after 10 – 12 minutes of ischemia (group T2), which shows that\nthe alterations occur rapidly once the cells are put under ischemic conditions. These\nchanges in immune response were mostly (20 of 34 proteins) due to a synchronous up-\nregulation of the protein expression in both tumour and normal tissue from one time\n3N ote that since it is impossible to obtain patient-wise time series data based on surgical tissue\nremoval, this is merely a pseudo-time-series.\n4https://www.genome.jp/kegg/mapper/search.html\n8\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nFigure 2: Differential expression of the selected proteins in refined time groups ( T1 . . . T5\nas 3-minute intervals on the abscissa) grouped into 11 clusters. The ordinates of\neach cluster show the normalised differential expression. The blue line shows\nthe cluster mean expression. The red lines indicate the individual protein\nexpression levels. The shaded blue area indicates the cluster mean ± 2 ·σ.\n9\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\npoint group to the next. There is almost no differential expression effe ct in this group\nbased on a tissue effect. These pathways reveal a coping mechanism involving decrease in\nexpression of proteins that are either not survival-critical under metabolic stress induced\nby ischemia (such as proteins involved in drug metabolism like UGT2B17 or UGT1A8)\nor those that are energy intensive for the cell (ribosome biogenesis). Ischemia very inter-\nestingly appears to slowly breach through this coping mechanism by causing a decrease\nin tumour survival-critical proteins such as NUDT1, ELOVL, HPGDS or DZIP3 [8, 9,\n10, 11].\n2.2.2 Confounder analysis\nWe next analysed the influence of ischemia time on differential biomolecule expression\ncompared to other independent variables of known clinical importance by performing\na classical confounder analysis using multiple linear regression. We computed a multi-\nvariate linear model with a Gaussian error distribution at the biomolecule level for each\nomic type and each of the differentially expressed biomolecule sets obtained from the\ninitial selection step described in section 2.1.\nThe variation of the fold-change between tumour and normal tissue expression was\nmodelled as dependent variable to be explainable by the independent variables (details\nsee section 4). This analysis is only a rough approximation because at the biomolecule\nlevel both statistical assumptions for Gaussian linear models of normal error distribution\nand linear variable relations are not fully met as revealed by statistical testing (not\nshown). Nevertheless, the analysis revealed informative trends when we determined the\ndistributions of variable effect estimates for biomolecules with any statistically significant\n(p < 0.01) variable. The corresponding plots of regression coefficient estimate means,\nsplit by positive and negative effects, for the CRC cohort are shown in Figure 3. Note\nthat variables which did not have any significant effect on any biomolecule are not shown\nin the confounder plots (details see section 4).\nThe figure shows that for CRC, disregarding the ischemia time variable effects (for\nwhich T1 is the reference variable), the tumour grade and stage as well as the alcohol\nconsumption status of the patients relative to their respective references are the strongest\npredictors of differential biomolecule expression as expected [12, 13]. High grade has\nnot only high coefficient estimates, but also the highest number of biomolecules with\nsignificant effects in each modality (491, 79, and 72, for mRNA, protein and phosphosite,\nresp.).\nWhat about the effect of ischemia time? With the exception of protein expression,\nthe ischemia group T2 shows the lowest T1-relative impact on the outcome compared to\nthe other time groups, but overall, the influence of ischemia on differential biomolecule\nexpression increases with longer ischemia times. If the ischemia time exceeds 12 minutes,\nthe average coefficient estimate of this variable is at least as strong as the stage-IV\nestimates, and above 15 minutes it is much higher (also higher than grade). The number\nof biomolecules with significant T5 estimates also surpasses the number of biomolecules\nwith significant stage-IV estimates, but does not exceed the number of biomolecules\nsignificantly affected by grade - which demonstrates the high relevance of this variable\n10\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n(a) Coefficient estimates (mean ± s td. dev.) of important predictors in\nthe linear model for the differentially expressed genes in CRC.\nFigure 3: Regression coefficient estimate means and standard deviations of important\npredictors (linear model T-statistic with p < 0.01) for the differentially ex-\npressed biomolecules of the three omic modalities in CRC. Numbers related to\neach variable denote the number of biomolecules with any significant positive\n(red) or negative (blue) effect. For categorical variables, the reference variables\nof the linear model are indicated in section 4. For example, T1 is the reference\ngroup for the other ischemia times. (a) Transcriptome, (b) Proteome, (c)\nPhosphoproteome.\n11\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n(b) Coefficient estimates (mean ± s td. dev.) of important predictors in\nthe linear model for the differentially expressed proteins in CRC.\n(c) Coefficient estimates (mean ± s td. dev.) of important predictors in\nthe linear model for the differentially expressed phosphosites in CRC.\nFigure 3: Regression coefficient estimate means and standard deviations (cont.).\n12\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nto explain the difference between tumour and normal tissue.\nT\nhe ischemia effects gain importance from transcriptome to proteome and are strongest\nin phosphoproteome. But in all modalities, the magnitude of the ischemia time estimates\nwipes out the impact of grade and stage, which are clinically the most important pre-\ndictors of cancer survival. Though this analysis is only indicative due the lack of full\nmodelling adequacy, it provides important insights into the influence of ischemia time\non biomolecule expression.\nAs a next step, we analysed the data to identify an ideal ischemia time cut-off.\n2.3 Ischemia time cut-off\nTo find an ischemia time cut-off that optimises the trade-off of collecting the maxi-\nmum amounts of samples while not impeding the identification of differentially expressed\nbiomolecules, we investigated how many biomolecules which are differentially expressed\nin the short ischemia time group ( < 10 min.) get lost with rising ischemia time. Figure\n4 shows the relative biomolecule loss in proportion to ischemia time in the CRC cohort.\nAs described in the methods section, we computed the number of differentially ex-\npressed biomolecules exclusive to the shortest ischemia time group as compared to the\nother groups for the three omic modalities, and observed overall a stronger biomolecule\nloss in groups of longer ischemia durations. For mRNA, the loss at 20 minutes is 22%,\nbut is much stronger pronounced at protein (53%) and phosphoprotein (86%) level in\nthe CRC cohort.\n2.4 Results for other cancer types\nWe next compared the most important results from the CRC analysis to the effects of\nischemia time on differential biomolecule expression in HCC, LUAD, and LUSC tumour\nversus matching normal tissue. Because of small group sizes ( T3 and T5) in HCC (see\nTable 1), we adjusted the time groups we used for the other cancer types in order to\nenable a reasonable statistical pseudo-time series analysis. Table 6 shows the adapted\nclassification scheme which was applied in the HCC cohort; note that due to the different\nassignment to the groups, there is no group T5. Figure 5 shows the confounder analysis\nfor HCC.\nInterestingly, unlike in CRC, the ischemia estimates do not surpass those of the impor-\ntant clinical predictors in any modality in general, though more biomolecules tend to be\nsignificantly affected by ischemia time than by stage or grade. As in CRC, the effects are\nmore pronounced for protein and phosphoprotein modalities than for the transcriptome.\nIn the biomolecule loss analysis for HCC (Supplementary Figure S4), the T4 group does\nnot show a stronger loss of deregulated proteins or phosphoproteins compared to the\nprevious time group T3. This could be explainable by the smaller group sizes of HCC\nsamples. Nevertheless, there is an analogous trend of losing deregulated biomolecules\nfrom T1 to T3 in a monotone manner.\nThe confounder analysis in LUAD (Supplementary Figure S2) shows a pattern similar\nto CRC. Over time, the ischemia effect gains the same importance as the effect of the\n13\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n(a) Percentage of differentially expressed genes exclusively de-\nt\nected in T1 in CRC.\nFigure 4: Relative biomolecule loss in proportion to ischemia times in CRC. Dot plots\nshowing the percentage of differentially expressed biomolecules exclusively de-\ntected in the shortest ischemia time group T1 (details see section 4) for each\nmodality. The ordinate indicates the time group compared to T1, ordered\nfrom shortest to longest time interval, the abscissa indicates the percent of\nbiomolecule loss.\n14\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n(b) Percentage of differentially expressed proteins exclusively\nd\netected in T1 in CRC.\n(c) Percentage of differentially expressed phosphosites exclu-\ns\nively detected in T1 in CRC.\nFigure 4: Relative loss of differentially expressed biomolecules in proportion to ischemia\ntimes in CRC. (cont.). 15\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nTable 6: Number of tumour samples per adapted ischemia time interval (mi n.) with\nspecific omic data for HCC\nTRX PTX PPX\nT1 : t < 10\n48 41 41\nT2 : 10 ≤ t ≤ 12\n35 36 36\nT3 : 13 ≤ t ≤ 18\n26 24 25\nT4 : 19 ≤ t ≤ 30\n23 28 28\nt > 30\n15\n16 16\nTotal\n147\n145 146\nworst stage (mRNA) or surpasses it (protein and phophoprotein). The biomolec ule loss\nin LUAD is comparable to CRC as well (see Supplementary Figure S5).\nThe confounder analysis in LUSC (Supplementary Figure S3) also shows the ischemia\neffect gaining importance over the clinical predictors with longer time, but in the protein\nand phosphoprotein modalities, there is an interesting exception, the stage IV variable.\nLate stage of LUSC alters differential biomolecule expression in these modalities much\nmore than any other predictor. This is not visible in the other cancer types, though tu-\nmour grade is a very strong predictor in HCC for differential phosphosite expression. The\nbiomolecule loss in LUSC is comparable to CRC and LUAD as well (see Supplementary\nFigure S6).\n3 Discussion\nThere is an ongoing debate about the influence of ischemia time on data from various\nhigh-dimensional molecular characterisation modalities (‘omics’) used in the analysis of\ncancer tissues. This discussion is important because these data are the foundation of the\nidentification of indicators of disease outcomes (prognosis) and novel cancer treatment\ntargets leading to potential therapeutics.\nAccording to some authors, ischemia times of 30 to 60 minutes can be tolerated still\nyielding acceptable gene expression data [2] ( n = 6 samples). Others reported that most\nphosphoproteins are stable over time [3] ( n = 3 samples) or that RNA does not degrade\nfor two hours at room temperature [4] ( n = 18 samples). All of these studies have an\nextremely limited number of samples in common which, given the high dimensionality\nof molecular characteristics, leads to a very low statistical power to detect ischemia\ntime effects. On the other hand, several studies have shown that longer ischemia times\nreduce the quality of the measured molecular characteristics such as mRNA, protein\nand phosphoprotein significantly [14, 15, 16, 17, 18, 19]. However none of these studies\n16\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n(a) Coefficient estimates (mean ± s td. dev.) of important predictors in\nthe linear model for the differentially expressed genes in HCC.\n(b) Coefficient estimates (mean ± s td. dev.) of important predictors in\nthe linear model for the differentially expressed proteins in HCC.\nFigure 5: Mean coefficient estimates and standard deviation of important predictors\n(linear model T-statistic with p < 0.01) for the differentially expressed\nbiomolecules in HCC. Numbers related to each variable denote the number\nof biomolecules with any significant positive (red) or negative (blue) effect.\nFor categorical variables, the reference variables of the linear model are indi-\ncated in section 4. (a) Transcriptome, (b) Proteome, (c) Phosphoproteome.\n17\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n(c) Coefficient estimates (mean ± s td. dev.) of important predictors in\nthe linear model for the differentially expressed phosphosites in HCC.\nFigure 5: Mean coefficient estimates and standard deviation of important predictors\n(linear model T-statistic with p < 0.01) for the differentially expressed\nbiomolecules of the three omic modalities in HCC. Numbers related to each\nvariable denote the number of biomolecules with any significant positive (red)\nor negative (blue) effect. For categorical variables, the reference variables of\nthe linear model are indicated in section 4. (a) Transcriptome, (b) Proteome,\n(c) Phosphoproteome (cont.).\n18\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nhave been able to use sufficient numbers of samples and therefore have an insufficient\npower to define a cut-off value for detecting the impact of ischemia time on the molecular\ncomposition of the materials, in particular for multi-omics data-driven target discovery.\nTherefore, so far, the published results on ischemia time have not been conclusive.\nOur research demonstrates the impact of ischemia time on the relative and absolute\nquantities and properties of DNA, mRNA, protein and phosphoprotein molecules using\na large tissue sample collection between 145 (HCC) and 633 cases (CRC) of four different\ncancer types in total including matching adjacent normal tissue. Because our aim is to\nunderstand the impact of ischemia time on differential biomolecule expression, we first\nestablished a set of differentially expressed biomolecules at the shortest ischemia time\ngroup T1, which most closely mimics the in vivo status. We used this baseline to map\nchanges in these biomolecules over time.\nWe first consider the underlying changes of the absolute biomolecule expression in\nnormal and tumour tissue under ischemic conditions. Table 4 shows the drastic loss\nin expression levels of biomolecules which are among the 5% least or most strongly\nexpressed ones in mRNA, protein and phosphoprotein omic modalities separately for\ntumour and normal tissue. We must imagine this decomposition of molecules in the\ndying cells as highly chaotic non-ergodic complex process, during which an animate\nsystem is transformed to inanimate, decaying biomolecule matter. Fundamentally, the\ndecay process is comparable in both tumour and normal tissues as is also evidenced by the\nlow proportion of interaction effects (Table 3B). Therefore, the differential biomolecule\nexpression, which compares tumour and normal tissue expression, is less affected by\nischemia than the individual expression levels per tissue type.\nFrom Figure 4 it is clear that the effect of losing differential mRNA expression over\ntime is weaker than for proteins and phosphoproteins, though at the separate tissue\nlevels, mRNA and protein decay in a similar manner. This could be caused by a higher\nregularity of the mRNA decomposition between the tissue types.\nOur main findings on the influence of ischemia time on differential biomolecule ex-\npression in the CRC cohort differ between the specific omic modalities. First of all, we\ndo not see any effect on genomic DNA, as is expected from the biochemical properties\nof this molecule type which can even be recovered from paleontological fossils to obtain\ngenetic sequences.\nFor the other modalities, as we see for phosphoprotein (Figure 1, Table 3), but also for\nmRNA (Supplementary Table S1) and protein (Supplementary Table S2), ischemia times\nover 25 minutes make analyses of rapidly decaying molecules scientifically unattractive.\nRegarding the impact of the tissue type on biomolecule expression, i.e. tumour vs.\nnormal tissue, and the ischemia time group or the combination of both variables (inter-\naction effect), the tissue main effect dominates the results since the biomolecules were\nselected based on group T1. Together with group T2 (10 ≤ t ≤ 12), which overall has\nan expression pattern very similar to T1 at least for mRNA and proteins, these time\ngroups cover roughly 2/3 of the samples (cf. Table 1), hence dominating the main effect\non biomolecule expression. The time and interaction effects in both panels of Table\n3 reflect the influence of ischemia time on differential biomolecule expression. In our\n19\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\ntissue collection, only 1/3 of the samples have longer ischemia times and their impact\non expression observed at the shorter times is limited. Thus we do not see too many\nbiomolecules with time effects due to the high quality of our tissue collection.\nOur enquiry concerning shorter ischemia times shows that the interpretation of time\nseries patterns of differential expression is challenging. The most clear pattern we ob-\nserve across the time series clusters relates to immune-response proteins and metabolic\nfunction bearing proteins.\nTo reveal independent variables which could dominate differential expression in ad-\ndition to ischemia time, we performed a systematic confounder analysis. As Figure 3\nshows, increasing ischemia times even surpass the effect estimators of alcohol consump-\ntion, grade, and cancer stage in the CRC cohort, which are known to strongly affect\nbiomolecule expression in cancer tissue [12, 13]. However, the number of biomolecules\nfor which the grade-covariable is significanctly correlated to differential expression is not\nsurpassed by the ischemia effect. While this can be related to the composition of our co-\nhort (with relatively few samples with longer ischemia times, see above), the confounder\nanalysis highlights the biological importance of grade, a predictor that is debated and\nsometimes underestimated in clinical practice [20].\nThe results obtained in the other two epithelium-descendend cancer types we anal-\nysed (LUAD and LUSC) are very similar to CRC. The number of biomolecules whose\ndifferential expression is significantly correlated to grade is also high in these cancer\ntypes.\nInterestingly, in HCC, a parenchymatous cancer type, the influence of ischemia time on\nthe differential expression outcome is weaker than in the adeno-carcinomata. However,\nas shown in the loss analysis (see Figure S4), there is a considerable loss of differentially\nexpressed biomolecules in hepatic tissue was well. Taken together our findings confirm\nthat HCC and liver normal tissue are more stable against ischemia than epithelial tissues\nand the cancers derived from them, so that even under longer ischemia time, there is\nsomewhat less loss of information.\nIn summary, our experiments show that ischemia times below 12 minutes are rec-\nommended to obtain optimal differential biomolecule expression data. If samples with\nlonger times are still to be included for specific reasons, they should be limited to a small\nproportion of the collection in order to obtain data of relevant scientific value.\n4 Material and methods\n4.1 Tissue sources and preparation\nIndivumed GmbH has a tissue collection of fresh frozen tumour samples with matching\nnormal tissues. These samples were frozen after different ischemia times after removal\nfrom the situs of surgery.\nTissue samples were collected by Indivumed’s clinical partners using a standardized,\nIRB approved protocol, focusing on minimal ischemia time. They were processed and\npathologically assessed as previously described [21]. Nucleic acid extraction, library\n20\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\npreparation and NGS were performed as previously described [21]. Protein extraction\nand MS analysis were performed as previously described [22, 23].\nThe data for CRC with matching colon mucosa from the same patient include 613\nTRX (mRNA), 633 PTX (protein) and 626 PPX (phosphoprotein) data sets, for HCC\nwith matching normal liver tissues 147 mRNA, 145 protein and 146 phosphoprotein data\nsets, and for LUAD 527 mRNA, 625 protein and 620 phosphoprotein data sets and for\nLUSC 377 mRNA, 415 protein and 408 phosphoprotein data sets with matching normal\nlung epithelium analyses.\nTumour stages, which are relevant for the confounder analysis we performed (see Fig.\n3 and corresp. text), match the expected distribution for colon cancer surgery specimens\n(10% stage I, 33% II, 33% III and 20% IV). The proportions in HCC are roughly 4 : 2.5\n: 2.5 : 1 for stages I to IV, and the proportions for LUAD are similar. For LUSC, they\nare roughly 3 : 3 : 3 : 1, inbetween CRC and HCC proportions.\n4.2 Data processing and analysis\n4.2.1 Processing and normalisation\nFor DNA sequences obtained from whole genome sequencing, we analysed protein amino-\nacid sequence affecting somatic mutations (PAM, counts per locus), copy number vari-\nations (binomial distribution per genomic locus for amplification or deletion, in two\nmatrices resp.), and the presence of truncations (binomial data). A Kruskal-Wallis test\nwas used to identify genes with a significant mutation load on the PAM count data,\nFisher’s exact test was used for this purpose on the binomial data. For each genomic\nlocus, the resulting P-Value for each ischemia time interval was compared to the data\nof the shortest time interval. For mRNA sequencing data, we adjusted batch effects of\nsequencing providers using ComBat-seq [24]. The normalised counts were then trans-\nformed to standard TPM values (transcripts per million) accounting for the length of\nthe transcripts as described in [25]. Protein MS/MS signals were transformed to counts\nusing DIA-NN [26], phosphoprotein MS/MS using Spectronaut 13 (Biognosys). For both\nmodalities, a median normalisation was performed per analysis run to scale the intensity\nvalues and make the runs comparable [27].\nThis preprocessing and normalisation results for each cancer type (CRC, HCC, and\nNSCLC) in three matrices, i.e. one matrix per omic data type (mRNA, protein and\nphosphoprotein), with biomolecules (genes, proteins, phosphosites) in rows and samples\nin columns. When needed, the expression values of the paired samples were aggregated to\nlog2-fold changes (tumour versus normal tissue), yielding matrices with half the number\nof columns.\n4.2.2 Statistical analysis\nDifferential biomolecule expression was calculated for the short ischemia time group T1\nusing the two-sided Wilcoxon test for paired samples. P-values were adjusted for multiple\ntesting via the method of Benjamini and Hochberg [28] which was the default correction\n21\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nmethod if not reported otherwise. The effect was computed as the mean di fference of\nthe log 2 transformed expression values between tumour and normal tissue.\nTo assess differences in differential expression between tissue types (tumour versus\nnormal) across ischemia time-groups, we used the non-parametric Scheirer–Ray–Hare\ntest for 2-factorial designs with the normalised expression values as outcome.\nTo perform hierarchical clustering on the differentially expressed biomolecules in the\nshort ischemia time group T1, for each modality the matrix of mean log 2-fold changes\n(tumour versus normal tissue) was row-wise normalised (mean-centered and scaled by\ndividing by standard deviation). We then applied hierarchical clustering with complete\nlinkage (using the Euclidean distance metric d(xi, yi) = ∥yi − xi∥2 =\n√∑P\nj=\n1(yij − xij)2,\nwith xi and yi two matrix rows and j = 1 . . . P the matrix columns) to the matrix rows\nand displayed the results as heatmaps with dendrograms. The number of clusters was\ndetermined by visual inspection. The clusters were used to refine the time groups and\nlimit them to 20 min.; they are merely illustrative and irrelevant for the further analyses.\nFor the clustering of these deregulated biomolecules in the refined ischemia time groups ,\nwe used a Dirichlet process (DP) Gaussian process (GP) mixture model [29]. This non-\nparameteric time series analysis technique jointly models data clusters with a Dirichlet\nprocess and temporal dependencies with Gaussian processes. 5 The DPGP software 6\nwas parameterised with concentration parameter α = 0.1, the number of empty clusters\nin each iteartion m = 12, and the shape and scale parameters of the inverse gamma\ndistribution set to αIG = 4 and βIG = 2, respectively.\nThe confounder analysis was performed by computing biomolecule-wise linear mod-\nels with Gaussian error distribution for each modality using the differential biomolecule\nexpression as response variable and age, gender, alcohol consumption and red meat con-\nsumption anamnesis, histological grade, tumour stage and the refined ischemia time\ngroups as independent predictor variables. The values of the resulting effects as average\nchanges in the log odds of the response variable associated with a one unit increase in each\npredictor variable were visualised, also including standard deviation, for the biomolecules\nwith at least one significant predictor variable (t-statistic derived p < 0.01). Apart from\nthe numeric continuous variable age, all other predictor variables were categorical ones\nwith the following factor levels:\nˆ gender : female (reference), male,\nˆ alcohol consumption : inactive (reference), active,\nˆ meat consumption (days per week): low (0-3 days, reference), high (4-7 days),\nˆ histological grade:low (G1-G2, reference), high (G3-G4),\nˆ tumour stage : I (reference), II, III, IV,\n5N ote that since it is impossible to obtain patient-wise time series data based on surgical tissue\nremoval, this is merely a pseudo-time-series.\n6https://github.com/PrincetonUniversity/DP GP cluster\n22\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nˆ r efined ischemia time : T1 (reference), T2, T3, T4, T5.\nThe optimal ischemia time cut-off was inferred within a set-difference analysis for\nthe refined time groups. This set-difference analysis was based on the results of the\ndifferential expression analysis. In detail, we identified the significantly differentially\nexpressed biomolecules per time group Ti, i = 1 . . . 5 ( αfdr = 0.01). We then determined\nthe set differences T1 \\ Ti of deregulated biomolecules in the shortest ischemia time group\nand the other groups. The percentage of the number of biomolecules in the set differences\nin relation to the number of differentially expressed biomolecules in the shortest ischemia\ntime group was visualised in dot plots.\n23\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nEthics declarations\nC\nompeting interests\nThis study was performed under the control of Indivumed; SvH, NR, NG, JL and HJ\nare employees of Indivumed. The authors declare no other competing financial interests.\nGM is a member of the editorial board of Cell Death Disease.\nEthics statement\nThe study was approved by the local Ethics Committees, and the patients signed the\nappropriate informed consent.\nAuthor contributions\nJL and SvH performed the mathematical analyses and wrote the paper. NR, A W and\nJLM performed the biological interpretation. Sample and clinical data collection as well\nas related preparation was conducted by NG, XM, TY, YT, GM and HJ.\n24\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\nReferences\n1.\nWellstein A. The pharmacological basis of therapeutics. Ed. by Brunton L and Knoll-\nmann B. 14th. McGraw Hill, 2023 :1335–450\n2. Blackhall FH, Pintilie M, Wigle DA, Jurisica I, Liu N, Radulovich N, Johnston\nMR, Keshavjee S, and Tsao MS. Stability and heterogeneity of expression profiles\nin lung cancer specimens harvested following surgical resection. Neoplasia 2004;\n6:761–7\n3. Buffart TE, Oord RA van den, Berg A van den, Hilhorst R, Bastiaensen N, Pruijt\nHF, Brule A van den, Nooijen P, Labots M, Goeij-de Haas RR de, et al. Time de-\npendent effect of cold ischemia on the phosphoproteome and protein kinase activity\nin fresh-frozen colorectal cancer tissue obtained from patients. Clinical Proteomics\n2021; 18:1–12\n4. Guo D, Wang A, Xie T, Zhang S, Cao D, and Sun J. Effects of ex vivo ischemia\ntime and delayed processing on quality of specimens in tissue biobank. Molecular\nMedicine Reports 2020; 22:4278–88\n5. Li Y, Ge X, Peng F, Li W, and Li JJ. Exaggerated false positives by popular dif-\nferential expression methods when analyzing human population samples. Genome\nbiology 2022; 23:79\n6. Pak MG and Roh MS. Influence of cold ischemia time and storage period on DNA\nquality and biomarker research in biobanked colorectal cancer tissues. Kosin Med-\nical Journal 2020; 35:26–37\n7. Thomas P, Ebert D, Muruganujan A, Mushayahama T, Albou L, and PANTHER\nHM. Making genome-scale phylogenetics accessible to all. Protein Sci. 2022; 4218:8–\n22\n8. Gad H, Koolmeister T, Jemth AS, Eshtad S, Jacques SA, Str¨ om CE, Svensson LM,\nSchultz N, Lundb¨ ack T, Einarsdottir BO, et al. MTH1 inhibition eradicates cancer\nby preventing sanitation of the dNTP pool. Nature 2014; 508:215–21\n9. Centenera MM, Scott JS, Machiels J, Nassar ZD, Miller DC, Zinonos I, Dehairs J,\nBurvenich IJ, Zadra G, Chetta PM, et al. ELOVL5 is a critical and targetable fatty\nacid elongase in prostate cancer. Cancer research 2021; 81:1704–18\n10. Shao F, Mao H, Luo T, Li Q, Xu L, and Xie Y. HPGDS is a novel prognostic\nmarker associated with lipid metabolism and aggressiveness in lung adenocarci-\nnoma. Frontiers in Oncology 2022; 12:894485\n11. Kolapalli SP, Sahu R, Chauhan NR, Jena KK, Mehto S, Das SK, Jain A, Rout M,\nDash R, Swain RK, et al. RNA-binding RING E3-ligase DZIP3/hRUL138 stabilizes\ncyclin D1 to drive cell-cycle and cancer progression. Cancer research 2021; 81:315–\n31\n12. Na HK and Lee JY. Molecular basis of alcohol-related gastric and colon cancer.\nInternational journal of molecular sciences 2017; 18:1116\n25\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n13. Huo T, Canepa R, Sura A, Modave F, and Gong Y. Colorectal cancer stages tran-\ns\ncriptome analysis. PLoS One 2017; 12:e0188697\n14. David KA, Unger FT, Uhlig P, Juhl H, Moore HM, Compton C, Nashan B, D¨ orner\nA, Weerth A de, and Zornig C. Surgical procedures and postsurgical tissue process-\ning significantly affect expression of genes and EGFR-pathway proteins in colorectal\ncancer tissue. Oncotarget 2014; 5:11017\n15. Freidin MB, Bhudia N, Lim E, Nicholson AG, Cookson WO, and Moffatt MF.\nImpact of collection and storage of lung tumor tissue on whole genome expression\nprofiling. The Journal of molecular diagnostics 2012; 14:140–8\n16. Gajadhar AS, Johnson H, Slebos RJ, Shaddox K, Wiles K, Washington MK, Herline\nAJ, Levine DA, Liebler DC, and White FM. Phosphotyrosine signaling analysis\nin human tumors is confounded by systemic ischemia-driven artifacts and intra-\nspecimen heterogeneity. Cancer research 2015; 75:1495–503\n17. Mertins P, Yang F, Liu T, Mani D, Petyuk V A, Gillette MA, Clauser KR, Qiao JW,\nGritsenko MA, Moore RJ, et al. Ischemia in tumors induces early and sustained\nphosphorylation changes in stress kinase pathways but does not affect global protein\nlevels. Molecular & cellular proteomics 2014; 13:1690–704\n18. Spruessel A, Steimann G, Jung M, Lee SA, Carr T, Fentz AK, Spangenberg J,\nZornig C, Juhl HH, and David KA. Tissue ischemia time affects gene and protein\nexpression patterns within minutes following surgical tumor excision. Biotechniques\n2004; 36:1030–7\n19. Unger FT, Lange N, Kr¨ uger J, Compton C, Moore H, Agrawal L, Juhl H, and David\nKA. Nanoproteomic analysis of ischemia-dependent changes in signaling protein\nphosphorylation in colorectal normal and cancer tissue. Journal of translational\nmedicine 2016; 14:1–15\n20. Chen K, Collins G, Wang H, and Toh JWT. Pathological features and prognosti-\ncation in colorectal cancer. Current Oncology 2021; 28:5356–83\n21. Yang X, Smirnov A, Buonomo OC, Mauriello A, Shi Y, Bischof J, Woodsmith J,\nBove P, Rovella V, Scimeca M, Sica G1TG, Wang Y, Servadei F, Melino G, Candi\nE, et al. A primary luminal/HER2 negative breast cancer patient with mismatch\nrepair deficiency. Cell Death Discovery 2023; 9:365\n22. Han Y, Rovella V, Smirnov A, Buonomo OC, Mauriello A, Perretta T, Shi Y,\nWoodmsith J, Bischof J, Bove P, Juhl H, Scimeca M, Sica G, Tisone G, Wang Y,\nGiacobbi E, Materazzo M, et al. A BRCA2 germline mutation and high expression\nof immune checkpoints in a TNBC patient. Cell Death Discovery 2023; 9:370\n23. Bruderer R, Sondermann J, Tsou CC, Barrantes-Freer A, Stadelmann C, Nesvizh-\nskii AI, Schmidt M, Reiter L, and Gomez-Varela D. New targeted approaches for\nthe quantification of data-independent acquisition mass spectrometry. Proteomics\n2017; 17:1700021\n26\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint \n\n24. Zhang Y, Parmigiani G, and Johnson WE. ComBat-seq: batch effect adjustment\nf\nor RNA-seq count data. NAR Genomics and Bioinformatics 2020; 2:lqaa078\n25. Wagner GP, Kin K, and Lynch VJ. Measurement of mRNA abundance using RNA-\nseq data: RPKM measure is inconsistent among samples. Theory in Biosciences\n2012; 131:281–5\n26. Demichev V, Messner CB, Vernardis SI, Lilley KS, and Ralser M. DIA-NN: neu-\nral networks and interference correction enable deep proteome coverage in high\nthroughput. Nature methods 2020; 17:41–4\n27. Kultima K, Nilsson A, Scholz B, Rossbach UL, F¨ alth M, and Andren PE. Develop-\nment and evaluation of normalization methods for label-free relative quantification\nof endogenous peptides. Molecular & Cellular Proteomics 2009; 8:2285–95\n28. Benjamini Y and Hochberg Y. Controlling the false discovery rate: a practical and\npowerful approach to multiple testing. Journal of the Royal statistical society: series\nB (Methodological) 1995; 57:289–300\n29. McDowell IC, Manandhar D, Vockley CM, Schmid AK, Reddy TE, and Engelhardt\nBE. Clustering gene expression time series data using an infinite Gaussian process\nmixture model. PLoS computational biology 2018; 14:e1005896\n27\n.CC-BY 4.0 International licensemade available under a \n(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 \nThe copyright holder for this preprintthis version posted May 28, 2024. ; https://doi.org/10.1101/2024.05.23.595517doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}