Molecular Harvesting of Proteins with Electroporation In Vivo Facilitates the Proling of Spatial Differential Protein Expression in Tumors

Background. Excision tissue biopsy, while central to cancer treatment and precision medicine, presents risks to the patient and does not provide a suciently broad and faithful representation of the heterogeneity of sampled solid tumors. Methods. Here we introduce e-biopsy – a novel concept for molecular proling of solid tumors using molecular extraction with electroporation. As e-biopsy provides access to the molecular composition of a solid tumor, it potentially facilitates tumor diagnostics without tissue resection. Furthermore, thanks to its less invasive characteristics, e-biopsy enables probing the solid tumor multiple times in several distinct locations in the same procedure, thereby enabling the spatial proling of tumor molecular heterogeneity. Results. We demonstrate e-biopsy in vivo, using the 4T1 breast cancer model in mice to assess its performance, as well as the inferred spatial differential protein expression. In particular, we show that proteomic proles obtained via e-biopsy in vivo distinguish the tumors from healthy tissue and reect spatial tumor differential protein expression. Conclusions. E-biopsy provides a completely new molecular cartography modality for solid tumors, providing information that potentially enables more sensitive detection, at lesser risk, as well as more precise personalized medicine. to the best of our knowledge it so far proposed for extracting molecules for tissue molecular proling, including in vivo. The goal of this work is to test molecular harvesting by electroporation (e-biopsy) in vivo and to assess the spatial differential expression at the proteomic level, observable through this novel sampling method. We also compare the proteomic molecular proles obtained through e-biopsy with state-of-the-art solid tissue lysis buffer extraction. In particular, we show that proteomic proles obtained by e-biopsy from 4T1 mice tumors in vivo are tissue specic, consistent, reect tumor protein expression heterogeneity, and align with proteomic proles obtained using standard lysis buffers from excised tissue samples. of distinguishing between 4T1 tumor and healthy murine breast tissue, regardless of sampling site. Differential analysis of protein expression (paired two-sided t-test with n = 5, the two replicates from each location were averaged) was performed for three pairs of e-biopsy extracts: 4T1 tumor center vs healthy breast, 4T1 tumor periphery vs healthy breast, and 4T1 tumor middle vs healthy breast. We found 13 proteins harvested by e-biopsy (Table 1) strongly overexpressed (p-value < 0.01, Methods) in all sampled locations in the tumor vs normal breast (no underexpressed proteins were identied under the same criteria). An intersection of this size has a p-value < 1.5e-06. These 13 proteins therefore represent an FDR of 3.5e-04 (Methods). Moreover, releasing the p-value cutoff to 0.05 results in a set of 242 (238 overexpressed and 4 under-expressed) differentially expressed proteins, corresponding to FDR of 2.3e-03. In further analysis we call these 242 proteins – potential 4T1 biomarkers. transition 52 . Long non-coding RNA (lncRNA), including VIM-AS1, and AGAP2-AS1 regulate Vim’s expression. Vim overexpression was reported in breast tumors in previous studies 53 . In addition, Hnrnpa21b overexpression was reported in endocrine-resistant LCC9 breast cancer cells. 54 Furthermore, triple-negative breast cancer patients face resistance to the drug trastuzumab by the active involvement of Polyadenylate-binding protein 1 (Pabpc1), expression of which is induced by overexpression of SNHG14 55 . For Serpinh1, also known as Hsp47, expression activation was reported during breast cancer development and progression 56 . Previous studies also demonstrated overexpression of hnRNPA1 during breast cancer 28 to 95%, and 25 minutes at 95% acetonitrile with 0.1% formic acid in water at ow rates of 0.15 µl/min. Mass spectrometry was performed using Q-Exactive Plus mass spectrometer (ThermoFischer Scientic, CA) in a positive mode using a repetitively full MS scan followed by collision-induced dissociation (HCD) of the 10 most dominant ions selected from the rst MS scan. The mass spectrometry data from all biological repeats were analyzed using MaxQuant software 1.5.2.8 vs. the mouse proteome from the UniProt database with 1% FDR. The data were quantied by label-free analysis using the same software, based on extracted ion currents (XICs) of peptides, enabling quantitation from each LC/MS/MS run for each peptide identied in any of the experiments.


Background
Based on molecular pro les of tumors and other tissues, personalized medicine aims to optimize medical care and preventative measures on an individual patient basis. In cancer therapy and care, a clear potential advantage has been demonstrated to the personalized approach as compared to traditional therapies [1][2][3] . Accurate diagnosis is a critical component of personalized medicine. An important component of molecular diagnostics in patient samples, including tumors, is the pro ling of DNA, RNA, proteins, glycans or metabolites, to identify molecular biomarkers that are predictive of tumor type [4][5][6][7] and of potential patient response 8, 9 . To enable tumor pro ling, current methods use tissue biopsy, which involves the physical resection of a small tissue sample. This procedure leads to localized tissue injury, bleeding, in ammation, neural damage, and stress 10,11 , the effect of which is not well understood 12 . Moreover, standard tissue biopsy could increase the potential for tumor growth and metastasis [12][13][14] . In addition, because of the negative effects, only a few biopsies can be performed at a time, limiting the scope of the spatial mapping of the sampled site, and leading to misdiagnosis if the tumor is missed. Furthermore, some authors even concluded that due to solid tumor heterogeneity, information from a single biopsy is not su cient for guiding treatment decisions 15,16 .
Indeed, recent literature identi ed the absence of e cient technologies for characterizing tumor molecular heterogeneity 17 as a major limitation of the personalized medicine approach in cancer 18 . Signi cant genomic evolution often occurs during cancer progression, creating variability within primary tumors as well as between the primary tumors and their metastases 16, [19][20][21] . Recent studies have shown that while a positive result (both successful biopsy and a decisive detection of markers) appears to reliably indicate the presence of the high-risk disease 15 , a negative result does not reliably rule out the presence of high-risk clones 22 . This is partly because a harvested tissue sample may not capture the most aggressive clone of a given tumor or tumor site 15,[23][24][25] . Despite the signi cant improvement in molecular characterization technologies in recent decades, thanks to the introduction of new high-resolution sequencing and bioinformatics methods, these technologies remain limited by tissue sampling methods 16,26 . Thus, tissue sampling constitutes a critical limitation of personalized medicine 15,16,27 . New approaches to probing and pro ling several regions in the tumor at the molecular level, termed molecular cartography, are expected to be useful in this context 28,29 .
To address these issues, and to extend the state-of-the-art of technologies that will potentially enable precision diagnosis and therapy, we developed a novel approach to molecular tissue sampling using electroporation 30 . Electroporation-based technologies have been successfully used to non-thermally irreversibly or reversibly change permeabilization of the cell membrane in vivo, enabling a wide set of applications ranging from tumor ablation to targeted delivery of molecules 30 . We and others previously developed protocols for targeted delivery of electric eld to tissues to induce focused electroporation at predetermined regions in organs [31][32][33][34][35][36] . More recently, we showed that electroporation technologies selectively extract proteins and ash from biomass [37][38][39] . Although electroporation has been used to deliver molecules to tissues and to ablate multiple tumors and metastatic sites, to the best of our knowledge it has not so far been proposed for extracting molecules for tissue molecular pro ling, including tumors in vivo.
The goal of this work is to test molecular harvesting by electroporation (e-biopsy) in vivo and to assess the spatial differential expression at the proteomic level, observable through this novel sampling method. We also compare the proteomic molecular pro les obtained through e-biopsy with state-of-the-art solid tissue lysis buffer extraction. In particular, we show that proteomic pro les obtained by e-biopsy from 4T1 mice tumors in vivo are tissue speci c, consistent, re ect tumor protein expression heterogeneity, and align with proteomic pro les obtained using standard lysis buffers from excised tissue samples. Table 1 List of proteins extracted with e-biopsy that differentiate 4T1 tumor from healthy breast tissue samples for all tumor sampling locations. This is a list of genes that are overexpressed in all 3 tumor locations together (Center/Midway/Peripheral; with both tumor location replicas averaged into a single value) compared to healthy breast samples (paired t-test p-value < 0.01 at each location). A two-sided, paired t-test was applied to each of the 4,519 known proteins sampled from 5 mice via e-biopsy. No underexpressed proteins were identi ed using the same criterion, namely p<0.01 in all three comparisons. The distribution of the observed differential expression scores, computed from comparing protein measurements in the healthy breast versus the different 4T1 tumor locations, is statistically signi cant, manifesting an overabundance of differentially expressed proteins. As shown in the overabundance plots (Fig. 2), there are more differentially expressed proteins observed in our data than expected under a random null model 43 .
Gene Ontology (GO) analysis of 4,519 e-biopsy-extracted proteins was used to further examine the differential expression between control (healthy breast) and three 4T1 tumor locations in terms of cellular processes, functions, and components. Here we present the most signi cant (p-value < 1E-06 at all locations simultaneously) differentially regulated processes, functions, and components (Tables 2-7) and discuss several interesting observations. All identi ed processes, functions, and components that are simultaneously enriched in all positions with p-value < 1E-06 are presented online in https://github.com/GolbergLab/eBiopsy4T1.
Notably, the analysis of the GO terms revealed several signi cant (p-value < 1E-06 in all locations) down-regulated cellular processes in 4T1 tumor ( Table 2). Among them there are many immunoglobulin-related processes, which is consistent with earlier reports 44 , who found that intratumoral injection of allogeneic IgG combined with other factors induced nearly complete eradication of lung metastases from 4T1. The downregulation of various peptidase inhibitors-related cellular functions (Table 3) in 4T1 is also consistent with previous works 45 . Moreover, the tumor's extracellular component in all three locations was downregulated compared to the healthy breast (Table 4), which is expected in aggressive and invasive tumors such as 4T1 46 . Furthermore, we found that many biosynthesis-related cellular processes are up-regulated in 4T1 (Table 5), which is consistent with the tumor's need for enhanced replication rates 47 . These ndings are also supported by many upregulated cellular functions (Table 6) and cellular components ( Table 7).
These data show that in vivo e-biopsy extraction of proteins yields statistically signi cant and biologically different pro les when comparing various locations in 4T1 tumors to healthy breast tissue in mice.     Reproducibility of in vivo molecular harvesting with e-biopsy To study the reproducibility of our in vivo e-biopsy extraction method, we harvested liquids from 6 positions: 2 in the center, 2 in the middle, and 2 at the periphery ( Fig. 1) from 4T1 tumors in vivo in 5 mice. In total, 4,262 proteins (with positive LFQ intensity in at least one e-biopsy sample) out of 4,519 total proteins were considered in this analysis. We found that the expression levels of proteins extracted from all locations in the tumor are highly correlated when comparing the location replicates (Table 8). In vivo e-biopsy supports mapping of 4T1 intratumor proteome spatial heterogeneity To study the intratumor heterogeneity, we compared (two-sided, paired t-test) expression levels of proteins extracted in vivo by e-biopsy from three different tumor locations -center, midway, and periphery -in ve animals. We found ( Next, we intersected the genes from the above analysis (41 from the center, 113 from the middle and 117 from the periphery) with the set of 242 potential 4T1 biomarkers (over/under-expressed in each of three 4T1 tumor locations compared to healthy breast with p-value < 0.05, Methods and text before Table 1). We found (Table 9) 2 such genes in the center, 3 in the middle and 3 in the periphery. This represents FDR of 7.1e-4, 4.7e-4 and 4.7e-4 respectively (Methods).
Speci cally, we found ( Table 9) that the gene Glrx is signi cantly overexpressed in the 4T1 tumor center compared to the tumor's middle area and periphery (Fig. 3). We also found that Rbb7, Pkn1, and Ppme1 are overexpressed in the middle area of the tumor compared to its center and periphery (Fig. 3). No uniquely overexpressed potential tumor biomarker genes have been identi ed at the tumor periphery.
In the opposite direction, we found ( Table 9) that the gene Ppme1 is signi cantly underexpressed in the 4T1 tumor center compared to tumor middle and tumor periphery (Fig. 3). Moreover, we found that Prkcsh, Tra2a, and Shoc2 are underexpressed at the tumor's periphery compared to its center and mid-zones (Fig. 3). No uniquely underexpressed potential tumor biomarker genes have been identi ed in the middle area tumor zone. The distribution of the observed differential expression scores, computed from comparing protein measurements between 4T1 tumor locations, is statistically signi cant (Fig. 4). The overabundance plots show that more differentially expressed proteins are observed in our data than would be expected under a random null model 43 . We also show that in vivo measurements obtained by e-biopsy are consistent with those obtained from the standard ex vivo lysis method (Supplementary information).
Gene Ontology (GO) analysis of 4,519 e-biopsy-extracted proteins was used to further examine the differential expression between all three locations in the 4T1 tumor in terms of various cellular processes, functions, and components. Here we present the most signi cant (p-value < 1e-06 at each location simultaneously) differentially regulated processes, functions, and components (Table 10) and discuss several interesting observations. All the identi ed processes, functions, and components that differ with p-value < 1e-06 are presented online in https://github.com/GolbergLab/eBiopsy4T1. In the comparison between Center and Periphery, no signi cant overexpressed function and components were identi ed in the Center. In the comparison between Middle and Periphery, no signi cant overexpressed processes, functions, and components were identi ed in the Middle (all three were overexpressed at the Periphery). Notably, in analyzing GO terms we found a signi cant decrease in ribosomal activity toward the 4T1 tumor's central region. This nding is consistent with previous work that showed that tumorigenicity was associated with profound alterations in ribosomal biogenesis and function, leading to the decreased translation of mRNA of tumor suppressor p53 and the reduced control of translational delity 48 . Also, GO analysis yielded a signi cant increase in blood coagulation toward the center of the tumor, consistent with increased vacularisation 49 .
The ndings above show that proteomic pro ling of in vivo e-biopsy samples can detect and potentially characterize 4T1 tumor heterogeneity. Speci cally, the differences in protein expression pro les for the different sampled tumor regions are statistically signi cant. In addition, we identi ed enriched biological changes in cellular processes, functions, and components when comparing the 4T1 tumor's center, middle, and periphery regions.

Discussion
Current cancer treatment decisions are often based on the information obtained from an aspiration needle biopsy or a surgical excision. These excised samples are evaluated for histopathology. Sometimes molecular tests are used to obtain more precise diagnostic results 25 . Standard treatment of patients with metastatic disease is usually based on predictive biomarkers detected with the original biopsy, which often does not fully re ect the status of disease progression 25 . Moreover, multiple recent studies suggest that tumor biopsies may vastly underrepresent tumors' heterogeneity and, therefore, may miss the drug-resistant clones 24,25,50,51 . In this work, we report a new method to probe tumors using molecular harvesting with electroporation, termed e-biopsy. Electroporation changes the permeabilization of the cell membrane, consequently increasing the accessibility of intracellular compounds 30 . In this work, we show that in vivo e-biopsy extraction of proteins yields a characteristic signature of 4T1 tumors vs healthy breast tissue in mice. We moreover show that point e-biopsy can detect various proteomic signatures in various geographical locations of the same tumor, thus increasing our understanding of the tumor sub-clonal spatial composition (Fig. 3, Fig. 4).
As e-biopsy is potentially less aggressive than the current standard excision-based biopsy method, this technology can serve as a basis for new diagnostic approaches that will better address tumor heterogeneity, by probing tumors in multiple locations.
Molecular harvesting by e-biopsy can distinguish between 4T1 tumor and healthy breast tissues, regardless of sampling location. We found 13 strongly overexpressed proteins (Table 1) in all sampled 4T1 tumor locations simultaneously. Some of these 13 proteins are known to have profound roles in breast cancer. For example, Vimentin (Vim), is considered a marker for epithelial-to-mesenchymal transition 52 . Long noncoding RNA (lncRNA), including VIM-AS1, and AGAP2-AS1 regulate Vim's expression. Vim overexpression was reported in breast tumors in previous studies 53 . In addition, Hnrnpa21b overexpression was reported in endocrine-resistant LCC9 breast cancer cells. 54 Furthermore, triplenegative breast cancer patients face resistance to the drug trastuzumab by the active involvement of Polyadenylate-binding protein 1 (Pabpc1), expression of which is induced by overexpression of SNHG14 55 . For Serpinh1, also known as Hsp47, expression activation was reported during breast cancer development and progression 56 . Previous studies also demonstrated overexpression of hnRNPA1 during breast cancer progression 57 . Slc25a24, Cnpy2 and BAX overexpression was reported in breast cancer 58 . Similar to our work, a recent study reported on the overexpression of Plin3 in the breast cancer tissues in comparison with normal breast 59 . HMGB1 is considered a ubiquitous protein, which has a role as a nuclear cofactor in the regulation of transcription 60 . HMGB1 overexpression in breast cancer tissue indicates metastasis, TNM stage, and differentiation 61 . HMBG1 has a promising role in breast cancer management as it affects chemotherapy, immunotherapy, and radiotherapy 62 . Mat2b overexpression was observed in the triple negative breast cancer 63 . In summary, this comparison to literature analysis showed that 11 out of 13 proteins that were extracted with e-biopsy in vivo and measured as overexpressed have been reported as overexpressed in breast cancer in previous studies using other extraction and quanti cation methods.
Drug resistance is one of the major hurdles in cancer treatment 64,65 . There are several known resistance mechanisms, with heterogeneity in tumors being one of the most important amongst them 26 . Our work, using 4T1 as a model, shows that e-biopsy may help in charting and quantifying the heterogeneity in tumors, mapping over-and underexpressed genes spatially, and thereby leading to in vivo low resolution molecular tumor cartography. In Table 9, Table 10, Fig. 3, and Table S4, we show an example of such a map based on differentially expressed genes in three spatial zones of a 4T1 tumor.
In addition to individual gene expression analysis, we also performed a gene ontology (GO) enrichment analysis of the measured proteome extracted by e-biopsy and of the inferred differential proteomics. GO analysis revealed that overexpression and underexpression of biological processes ( and therefore suggests tumor heterogeneity for these studied samples ( Table 9, Table 10). Similar to our work, published literature on 4T1 66,68 corroborates these ndings. Altogether, the pathway enrichment analysis suggests that the proteomic pro le detected by e-biopsy is corroborated by similar reports in the literature using other extraction methods. We, therefore, expect e-biopsy sampling to potentially yield biological information which is equivalent, at the level of gene sets or pathways, to that which would be inferred by other sampling technologies.

Conclusions
In the current work we introduce e-biopsy, a novel tool for molecular harvesting in vivo using electroporation. E-biopsy has the potential to reduce the risks and morbidities of excision biopsy and to provide additional information and better pro ling of the tumor and the probed environment in vivo. We demonstrate that e-biopsy enables the in vivo distinction between tumor and non-tumor samples and locations in the 4T1 mice model. Due to its minimal invasive nature, e-biopsy can potentially enable tumor sampling at multiple locations. We therefore hope that e-biopsy will potentially facilitate shedding light on the clonal subpopulation composition of tumors. This information on the tumor's heterogeneity may be vitally important for higher precision personalized therapies. We therefore believe that e-biopsy represents a useful addition to the toolbox available to scientists and practitioners in their approach to treating cancer patients.

Animals
All animal procedures were approved by the Israel National Council for Animal Experimentation (Study no. IL-19-3-114). Five 8-week-old female Balb/c female mice weighing ~20g were provided by the Science in Action, Ltd. CRO. The animals were housed in cages with access to food and water ad libitum and were maintained on a 12h light/dark cycle at a room temperature of ~21°C and a relative humidity of 30%-70%. All in vivo experiments were conducted by a professional veterinarian as per Israel National Council for Animal Experimentation guidelines and regulations.
In vivo 4T1 tumor model 4T1 cell line was purchased from the American Type Culture Collection (Manassas, VA, USA). The cells were cultured in RPMI-1640 media with L-Glutamine supplemented with 10% fetal bovine serum (FBS), 0.11 mg/ml sodium pyruvate, 100 U/ml penicillin, and 100 µg/ml streptomycin (Biological Industries, Israel) at 37° C in a humidi ed CO 2 incubator. 4T1 cells were subcutaneously injected (0.5X10 6 cells) into Balb/c female mice.

Histology
Specimens were harvested immediately after the treatment and xed in 10% formalin. Samples in plastic cassettes were dehydrated through ascending ethanol concentrations, transferred into xylene, and then para nized, by an automated machine. Next, the samples were manually embedded into para n blocks. The para n blocks were sectioned at approximately 3-5 microns thickness. Sections were placed on glass slides. Slides were stained with Hematoxylin & Eosin (H&E) and covered by an automated machine.

Immunohistochemistry
Para n blocks were sectioned at approximately 3-5 microns thickness. Sections were placed on SuperFrost Plus™ glass slides. Slides were incubated overnight at 60º C. Slides were stained using the standard procedure in Ventana BenchMark Ultra automated slide stainer in combination with Ventana UltraView Universal DAB Detection Kit (Ventana, Roche Diagnostics cat #760-500). The slides were stained with the following antibodies: monoclonal mouse anti-Human Ki-67, clone MIB-1 (Dako, cat# M7240), diluted 1:200, and monoclonal mouse anti-Human Glypican-3 (GPC3), clone 1G12 (BioCare Medical, and cat# PM396 AA), ready to use. Slides were counterstained in Mayer's Hematoxylene, dehydrated through ascending ethanol concentrations, cleared in Xylene, mounted, and covered.
Pulsed electric eld application for protein extraction in vivo E-biopsy was performed with a 23G needle at 6 positions inside each tumor: 2 in the center, 2 at the periphery, and 2 in the middle between the center and the periphery (Fig. 1). The needle was connected to a cathode. The second 23G needle, connected to the anode, was held at a 1cm distance from the rst needle. The pulsed electric eld was applied using the electric eld pulse generator (BTX830, Harvard Apparatus, MA).
Electroporation was performed using a combination of high-voltage short pulses with low-voltage long pulses as follows: 40 pulses 1000V, 40µs, 4Hz, and 40 pulses 150Vcm −1 , 15ms, delivered at 4Hz. After the PEF treatment, the liquids were extracted from the tissue to the needle applied to a vacuum with a 1.5mL syringe. The liquids were immediately transferred to 1.5 ml tubes with 100µl double distilled water (DDW).
Isolating proteins from the pulsed electric eld extracted juices

Mass spectrometry analysis
The tryptic peptides were desalted using C18 tips (Harvard Apparatus,MA), dried, and re-suspended in 0.1% formic acid. The peptides were resolved by reverse-phase chromatography on 0.075 X 180-mm fused silica capillaries (J&W) packed with Reprosil reversed-phase material (Dr. Maisch GmbH, Germany). The peptides were eluted with a linear 180-minute gradient of 5 to 28%, 15 minutes' gradient of 28 to 95%, and 25 minutes at 95% acetonitrile with 0.1% formic acid in water at ow rates of 0.15 µl/min. Mass spectrometry was performed using Q-Exactive Plus mass spectrometer (ThermoFischer Scienti c, CA) in a positive mode using a repetitively full MS scan followed by collision-induced dissociation (HCD) of the 10 most dominant ions selected from the rst MS scan.
The mass spectrometry data from all biological repeats were analyzed using MaxQuant software 1.5.2.8 vs. the mouse proteome from the UniProt database with 1% FDR. The data were quanti ed by label-free analysis using the same software, based on extracted ion currents (XICs) of peptides, enabling quantitation from each LC/MS/MS run for each peptide identi ed in any of the experiments.

Bioinformatics and statistical analysis
Data for 4,781 proteins was obtained from the mass spectrometry analysis, 4,519 of which were accompanied by valid protein and gene ids.

Inter-sample correlation analysis
Pearson and Spearman correlations were estimated between LFQ-intensity protein pro les of each sample with scipy.stats.pearsonr and scipy.stats.spearmanr functions respectively (Fig. S2, Table S2).
To count the identi ed proteins (Table S1, Table S5) by each method (e-biopsy vs Lysis), we de ned all proteins with strictly positive LFQintensity as existing within the speci c sample (any sample in case of Location_MIX). If a protein was identi ed by e-biopsy/lysis only, it was marked as uniquely captured by e-biopsy/lysis. Otherwise (protein LFQ-intensity > 0 measured by both methods) it was marked as simultaneously captured by both methods. The value for Mouse_Average was derived as an average of all of the comparisons of samples within the same mouse.
Differential expression analysis of control (healthy breast) and tumor (4T1) samples The protein representations for control were constructed as 5D vectors based on e-biopsy LFQ-intensity measurements from healthy breast tissue in 5 mice. The protein representations for tumors were constructed as 5D vectors based on the average of two LFQ-intensity measurement replicas at speci c 4T1 tumor locations from 5 mice. Paired two-tail Student t-test was performed with scipy.stats.ttest_rel function. Further, the overabundance comparison of the obtained distribution to the random model was performed 43 (Fig. 2). 13 genes with Student t-test p-values below 0.01 at each location simultaneously (73 such genes were identi ed in Peripheral, 160 in Middle and 164 in Center) were labelled as strongly overexpressed (no underexpressed genes for the same criteria were identi ed) in breast cancer tissue ( In the further analysis we call these 242 differentially expressed proteins -potential 4T1 biomarkers, speci cally for the subsequent search for intra-tumor heterogeneity markers, particularly for ltering out Table 9 from Table S4.

Intratumor differential expression analysis
The protein representations for all tumor locations were constructed as 10D vectors based on e-biopsy LFQ-intensity measurements from two replicas at speci c tumor locations in 5 mice. Paired two-tail Student t-test was performed with scipy.stats.ttest_rel function. Then, the overabundance comparison of the obtained distribution to the random model was performed 43 (Fig. 4).

Gene Ontology analysis
The proteins detected simultaneously in both tissues/locations were sorted as per their Student t-test t-statistic values (in decreasing direction for overexpression, and increasing direction for underexpression). Then cellular processes, functions, and components based on Gene Ontology (GO) were tested for signi cant (mHG p-value < 1e-06) overabundance at the top of the obtained proteins list using Gorilla tool 70,71,72 .   Overabundance plots comparing (Methods) the distribution of the protein differential expression scores between 4T1 tumor locations.