APTAK organoids were kindly gifted by Florian Greten, Georg Speyer Haus, Frankfurt. These cells are murine colorectal carcinoma cells holding the following mutational features: Deletions in the Apc-, Trp53- and Tgfbr2-gene, a myristoylated Akt-1 with constitutively activated signaling and a Kras ^G12D-mutation (16). The APTAK organoids were maintained in Advanced DMEM-F12 (Gibco/Thermo Fisher Scientific, Waltham, MA, USA) supplemented with penicillin (100 U/ml)/streptomycin (100 µg/ml) (Gibco/Thermo Fisher Scientific), HEPES 10 mmol/L (Invitrogen/Thermo Fisher Scientific), Glutamax 2 mM (Invitrogen/Thermo Fisher Scientific), N2 1× (Gibco/Thermo Fisher Scientific), B27 1× (Gibco/Thermo Fisher Scientific), and N-acetylcysteine 1 mmol/L (Sigma-Aldrich, St. Louis, MO, USA) and supplemented with hygromycin (200 µg/ml, Invitrogen/Thermo Fisher Scientific) and puromycin (2 µg/ml, InvivoGen, San Diego, CA, USA) (hereafter referred to as basal medium). The organoids were split every 3–5 days at a 1:10 ratio and cryoconserved for long-term storage.

All animal studies were performed using 8–12 weeks old C57Bl6/J-mice (purchased from Charles River, Wilmington, MA, USA) at the pathogen-free barrier facility of the CEMT, University Medical Center Freiburg. The experiments were planned and carried out in accordance with the institutional guidelines and all experiments were approved by the regional board (Regierungspräsidium Freiburg, G21/001).

We adapted the subserosal orthotopic organoid implantation technique described by Fumagalli et al. (17). APTAK organoids cultured in Cultrex RGF basement membrane extract (BME) Type 2 (R&D Systems, Minneapolis, MN, USA) were dissociated into single cells by mechanical disruption and washed with cold PBS. After determining the cell number with the cell counter TC20 (BioRad, Hercules, CA, USA), the appropriate cell number was resuspended in collagen I/5x neutralization buffer (vol/vol-ratio 1:4), and collagen domes of 100.000 cells/10 µl were plated on a 6-well plate. After 30 min polymerization at 37°C, basal medium was added and cells were incubated over night until implantation. Mice analgesia was performed with intraperitoneal (i.p.) injection of 0.1 mg/kg bodyweight (BW) buprenorphine (Temgesic, Schering-Plough, Kenilworth, NJ, USA). Anesthesia was induced by 2.5 Vol% Isoflurane (Isoflurane Piramal, Piramal, Mumbai, India) with the SomnoSuite^® Low-Flow Anesthesia System (Kent Scientific Corporation, Torrington, CT, USA). The unconscious mice were positioned on a heating pad with all legs fixed and the abdomen was shaved while anesthesia was maintained by continuous application of isoflurane via mouthpiece. After disinfection of the skin, a median laparotomy of approximately 1 cm was performed on the lower abdomen. The cecum was mobilized using two cotton swabs and placed on a wet gauze. Under the microscope, a 2 mm incision of the serosal layer was made and a deep subserosal pocket was created in which the organoid-collagen-graft was fully placed. After covering the graft with the serosal layer, an anti-adhesion layer (Seprafilm, Baxter, Deerfield, IL, USA) was placed on top. The cecum was then replaced in the abdominal cavity and the abdominal wall and skin were closed in two layers using 6–0 Vicryl (Ethicon/Johnson&Johnson, New Brunswick, NJ, USA) suture. The duration of the surgical procedure was approximately 10–15 minutes.

MR imaging was performed using a 9.4 tesla small bore animal scanner (BioSpec 94/21, Bruker Biospin, Karlsruhe Germany) and a dedicated mouse quadrature-resonator (Bruker). The mice were anesthetized under spontaneous breathing conditions using Isoflurane. Heart and respiration rates were continuously monitored and maintained at a constant level. Gating was used to reduce motion and blood flow artifacts during the scan. The MRI protocol consisted of a localizer and a T2-weighted spin echo RARE (Rapid Acquisition with Relaxation Enhancement) sequence. This sequence was performed to delineate the tumor and eventual metastasis from the surrounding healthy tissue. The RARE sequence in axial orientation featured a field-of-view (FOV) of 28 mm (1), a matrix size of 280 x 280 pixels, and an in-plane resolution of 100 x 100 µm (1). The slice thickness was 0.75 mm (TR/TEeff/FA: 300 ms/36 ms/180°). The number of slices was adjusted to the measured volume (on average 30) to ensure complete coverage of the abdomen. The Bruker ParaVision 6.0.1 software was used for imaging and images were analyzed using the NORA software supplied by the Medical Center of the University of Freiburg. The volumes of all hepatic lobes and metastases were assessed individually by manually marking the respective lobes/LMs as regions of interest (ROIs) in each section.

During isoflurane anesthesia mice were placed on a heating pad and the abdomen was shaved. Prewarmed ultrasound gel was applied to the abdomen and a VEVO 3100 (Fujifilm/VisualSonics, Toronto, Canada) ultrasound together with a small animal transducer was used for ultrasound imaging.

For distribution among the control-, SHAM-, and pHx-groups, we assessed the preoperative hepatic tumor load using MRI. Animals displaying an approximately similar number and volume of LM in the imaging were allocated equally among the different protocols to reduce the influence of cage-specific phenotypes. After general analgesia and anesthesia with buprenorphine and isoflurane as described above, mice were shaved and placed on a heating pad. After disinfection, a median laparotomy of approximately 2 cm, starting at the xiphoid, was performed and retractors were placed on both sides of the abdominal wall and under the xiphoid. The falciforme ligament and the membrane between the medial and left hepatic lobes were dissected. A 3–0 Vicryl (Ethicon/Johnson&Johnson) suture was placed around the base of the medial lobe and tightened with 3 knots, after which the ischemic lobe was cut approximately 0.1 cm distal to the suture ( Supplementary Figure 1A ). The resected lobe was placed in cooled PBS Buffer on ice until further analysis. The abdominal cavity was flushed with prewarmed saline, and the abdominal wall and skin were closed using 6–0 Vicryl (Ethicon/Johnson&Johnson) stitches. The pHx procedure took approximately 15 min. SHAM mice underwent laparotomy, including dissection of the falciforme ligament and the membrane between the medial and left lobes without resection of the lobe. Control animals received general anesthesia without any surgical procedures.

Animals were euthanized by cervical dislocation. Subsequently, the abdomen was opened and each lobe of the liver was resected directly at the vena cava. The lobes were weighed and all visible liver metastases were dissected while the organ was sliced. The entire cecum was resected and the tumor tissue was separated from the healthy intestine. We collected samples for RNA isolation (stored in RNAprotect buffer (Qiagen, Netherlands)) and for protein isolation from all tissues mentioned. Healthy liver tissue was minced and dissociated using the Liver Dissociation Kit, mouse (Miltenyi Biotec, Bergisch Gladbach, Germany) according to the manufacturer’s protocol. The primary tumor and LM were minced into <1 mm pieces and dissected using the Tumor Dissociation Kit, mouse (Miltenyi Biotec) according to the manufacturer’s instructions. The single cell suspension was filtered through a 100 µm cell strainer and ACK lysis buffer (Gibco/Thermo Fisher Scientific) was added to lyse erythrocytes during a 1.5 minutes incubation.

Cells were transferred to 100 µl FACS buffer (PBS + 0,5% BSA), incubated with TruStain FcX PLUS (anti-mouse CD16/32) antibody (BioLegend, San Diego, CA, USA) for 15 min and then stained with the respective fluorescent antibodies for 60 min. Cells were then washed with FACS buffer (for extracellular staining) and fixed with 1x Fixation/Permeabilization buffer for 60 min. For intracellular staining, we used the eBioscience™ Foxp3/Transcription Factor Staining Buffer Set (Invitrogen/Thermo Fisher Scientific) according to the manufacturer’s instructions. All FACS antibodies used were supplied by BioLegend and used undiluted with 1 µl/100 µl cell suspension. FACS was performed using Beckman Coulter Gallios (Beckman Coulter, Brea, CA, USA) or BD LSRFortessa (Becton Dickinson, Franklin Lakes, NJ, USA). Data were analyzed using Kaluza Software (Beckman Coulter). Cellular aggregates were excluded using SSC-W.

Equal amounts of single cells from the tumor/liver dissociation were seeded in RPMI-Medium (+ 1% Pen/Strep + 1% HEPES + 5 µl beta-mercaptoethanol) on a 12-well plate overnight. The cells were then stimulated by adding Cell Activation Cocktail (BioLegend) for 3 hours, followed by addition of Brefeldin A (BioLegend) for 2 hours. The cells were then harvested and stained for FACS analysis, as described above.

Tissue samples for RNA isolation were stored in RNAprotect Buffer. The tissue was shredded in 350 µl RLT buffer (Qiagen, Venlo, Netherlands) + 1% beta-mercaptoethanol. Further RNA isolation was carried out using the RNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. RNA concentrations were measured using a NanoDrop™ spectrophotometer (Thermo Fisher Scientific). For cDNA synthesis RNA was transcribed using ProtoScript II First Strand cDNA Synthesis Kit (New England Biolabs, Ipswich, MA USA) according to the manufacturer’s instructions. cDNA was finally diluted to 3 ng/µl for qPCR experiments.

For quantification of bacterial DNA, we isolated DNA from tissue samples using the DNeasy Blood & Tissue Kit (Qiagen) according to the manufacturer’s instructions, and concentrations were assessed using a NanoDrop™ spectrophotometer (Thermo Fisher Scientific).

Quantitative Real-Time PCR was performed using QuantiTect SYBR Green RT PCR mastermix (Qiagen) with 7.5 ng template cDNA and 5 pmol of each primer (purchased from Eurofins Genomics, Ebersberg, Germany; primer sequences supplied the supplements). qPCR was carried out on a Roche LightCycler 480 (Roche, Basel, Switzerland) with each sample-gene combination in triplicates. The protocol consisted of a 10 min activation step at 95°C followed by 50 cycles of 45 s amplification at 60°C and inactivation at 95°C for 15 s. 18S was used as a housekeeping gene, Cp values were calculated using the LightCycler480 software (Roche), and the relative expression of target genes was calculated by comparative method after normalization to 18S-expression.

In order to quantify bacterial infiltration, we performed qPCR with amplification of the bacterial V6 region using five different primers (see Supplements). Each sample was measured in triplicates using the QuantiFast SYBR^® Green PCR Kit (Qiagen). 100 ng DNA was mixed with 500 nM primers. By diluting the DNA of Bifidobacterium actinocoloniiforme (DSMZ-Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH, DSM22766) in 10-fold series we generated a standard curve using the Threshold cycle (Ct) values. To quantify the sample bacterial DNA, we correlated the average Ct-values to the standard curve of log values from the Bifidobacterium actinocoloniforme.

RNA sequencing was performed at Genewiz, Germany, using the standard Illumina protocol to create raw sequence files (.fastq files). Poly(A) mRNA was used to prepare cDNA libraries, which were then sequenced as 2x 150 bp paired end using the Illumina NovaSeq instrument (Illumina, San Diego, CA, USA).

Raw sequencing files were uploaded to the galaxy web platform (https://usegalaxy.eu) (18) and quality was assessed using FASTQC. Owing to good quality and no adaptor content, trimming was not necessary. Data were mapped to the murine reference genome (GRCm39) provided by GENCODE (www.gencodegenes.org, downloaded September 2021) using the STAR aligner (19) in standard settings for paired sequencing data. Reads were assigned to counts using featureCounts software (20) and the gene annotations also provided by GENCODEs (downloaded September 2021). Paired-end settings were used for featureCounts software, and a quality score of >10 was required for counting. Data were imported in R 4.0.2 (https://www.r-project.org), and differentially expressed gene analyses were conducted using DESeq2 (21). A batch effect was removed using the limma::removeBatchEffect() function (22). Differentially expressed genes were defined as padj <0.05 and abs(log2 FC) >1. The Ensembl gene ID was converted to MGI symbols using the biomaRt package (23). The fgsea package (24) conducted Gene Set Enrichment Analysis (GSEA) by ranking all genes with at least >10 counts according to the log2 FC, which were previously shrunken by the apeglm logarithm (25) before. Gene sets were accessed from MsigDB (26) and downloaded on the 26^th of May 2023. For visualization, ggplot2 (27), ggraph and ComplexHeatmaps (28) were used.

Tissue samples were shredded in RIPA buffer containing phosphatase and proteinase inhibitors (Phosstop (Roche), protease inhibitor cocktail, sodium orthovanadate, and PMSF (Santa Cruz, Santa Cruz, CA, USA) on ice. Three cycles of freezing and thawing were applied using liquid nitrogen, and the suspension was centrifuged at maximum speed for 10 min. The supernatant containing the protein lysate was transferred to a new tube. Protein concentration was measured using a Pierce BCA-Kit (Thermo Fisher Scientific) and an Azure Ao Microplate Reader (Azure Biosystems, Dublin, CA; USA).

Protein samples were aliquoted with 4x Laemmli Buffer (BioRad) and denatured at 95°C for 3 min. An equal amount of protein was loaded on a 12–20% Mini-Protean TGX precast gel (BioRad) together with a color-coded prestained broad range protein ladder (Cell Signaling Technology, Danvers, MA, USA). After SDS-electrophoresis, the proteins were blotted onto a PVDF membrane using the trans-blot turbo transfer system (BioRad) according to the manufacturer’s instructions. The membrane was then blocked with 5% BSA/TBST for 60 min at RT and incubated with the respective primary (Anti-S100A8, Anti-MMP9, Anti-Claudin-2, Anti-Occludin, and Anti-ZO-1) and secondary (anti-rabbit IgG) antibodies according to the manufacturer’s instructions (a detailed description of the antibodies used is supplied in the Supplements). For detection, the Immobilon Crescendo Western HRP substrate (Merck Millipore, Burlington, MA, USA) was added to the membrane and imaging was carried out using a ChemiDoc BioRad imager (BioRad).

Tissue samples were fixed in 4% PFA for 24 hours, dehydrated and embedded in paraffin. Sections of 3 µm were cut, rehydrated, and heat induced epitope retrieval was carried out with citrate or TE-Buffer (Zytomed Systems, Berlin, Germany). Sections were then blocked with 2.5% BSA and avidin-biotin-block (Zytomed Systems) according to the manufacturer’s instructions. Sections were then incubated with the primary antibody overnight and biotinylated secondary antibody for 1 h according to the manufacturer’s instructions (see a detailed description of the antibodies used in Supplements). Positive staining was visualized using DAB (Zytomed Systems), and the sections were counterstained with hematoxylin before dehydration and mounting with ROTI^® Histokitt (Carl Roth, Karlsruhe, Germany). Slides were digitalized using a Zeiss Axio Scan Z.1 Microscope Slide Scanner (Zeiss, Jena, Germany).

Slides were treated until secondary antibody incubation as described in the Immunohistochemistry section, and then incubated with the Vector^® TrueVIEW^® Autofluorescence Quenching Kit (Biozol, Eching, Germany) to reduce autofluorescence. After washing, nuclear staining was performed using NucBlue™ Fixed Cell Stain ReadyProbes™ (Thermo Fisher Scientific). Sections were then mounted with VECTASHIELD Vibrance Antifade Mounting Medium (Biozol) and scanned using a Zeiss Axio Scan Z.1 Microscope Slide Scanner (Zeiss).

For statistical analysis, we used a one-way analysis of variance (ANOVA) test with Tukey’s post-hoc analysis when comparing multiple samples or a two-tailed Students-t-test for two variables. Statistical significance was set at p-values <0.05, and significance levels are represented in the graphic as follows: *** = p<0.001, ** = p<0.01, * = p<0.05, n.s. = not significant. All statistical analyses were performed using GraphPad Prism.

To establish a mouse model of orthograde CRC metastasis we used murine CRC tumor organoids deficient for Apc, Trp53 and Tgfbr2 and that display a Kras gain of function mutation (G12D mutant) together with a constitutively activated Akt1 (hereafter referred to as APTAK organoids) (29). We combined this APTAK mouse model with a partial hepatectomy model to monitor the effect of pHx on the background of orthograde colorectal liver metastasis ( Figure 1A ). APTAK organoids were surgically implanted under the serosal layer of the distal cecum, leading to carcinomas with luminal contact after 2 weeks ( Figures 1B–D ). These tumors caused cellular aggregates that appeared in the liver after 3–4 weeks (microscopically detectable in hematoxylin/eosin (HE)-sections) ( Figure 1C ) and macroscopically visible metastases after 5–6 weeks ( Figures 1C, D ). To further characterize the early-appearing cellular accumulations in the liver, we performed an immunohistochemistry (IHC) of liver sections. While the liver metastases and primary tumors displayed plenty of EpCam-positive cells, we did not find any EpCam-positive epithelial cells in the microscopically visible perivascular cellular aggregates ( Supplementary Figure 1A ). Hence, we adapted the term premetastatic lesions.

To objectify the metastatic load in the liver, magnetic resonance imaging (MRI) was performed prior to liver resection. Detectable liver metastases in MRI appeared after 4–5 weeks as T2-hypointense lesions ( Figure 1D ), with the smallest distinguishable measuring 0.2 mm in diameter. To measure the time to metastasis and general tumor burden, mice were screened weekly for detectable liver metastases with ultrasound starting 3 weeks after implantation. In comparison to MRI, detection of LM in ultrasound was delayed with visible lesions starting at >1 mm in diameter ( Figure 1D ). After resection of the medial lobe ( Supplementary Figure 1B ), mice were screened again for metastases on postoperative day (POD) 7 and 14 in MRI. The animals were then euthanized and their organs were harvested for further analysis. Therefore, we set up a valuable orthograde CRLM mouse model suitable for analyzing pHx, which resembles the human situation.

In order to verify the hypertrophic effect on the residual liver and to evaluate the effect of pHx on the metastatic load we performed volumetric analysis of the MR images prior to surgery and 7 and 14 days postoperatively ( Figure 2A ). Following pHx, the mice developed significant hypertrophy of ~50% of the residual right and left lobes compared to control and SHAM-operated mice both in weight ( Figure 2B ) and volume measured in MRI (each p<0.001) ( Figure 2C , Supplementary Figure 1C ). The hypertrophy was already seen in MRI on POD7 but further increased until POD14. The total volume of healthy liver tissue in pHx mice on POD14 was comparable to that in both control groups ( Supplementary Figure 1D ). We could conclude that 1/3 partial hepatectomy in our model is sufficient to generate significant hypertrophy, consistent with the effects of a major hepatectomy in humans.

As pHx has been linked to a potentially prometastatic effect in injection models of colorectal liver metastasis before (12–14), we were curious about the impact of resection in our model of orthograde metastasis. Regarding the metastatic burden in the animals, we found no difference in the total metastatic volume within the liver on POD7 between control, SHAM, and pHx mice. On POD14, on the other hand, an increase in the metastatic volume in the residual right and left lobes was observed in the hepatectomy group compared to that in both control groups ( Figure 2D , Supplementary Figure 1E ). The same effect was observed when the metastases in the medial lobe of the control groups were included in the analysis ( Supplementary Figure 1G ). The increase in total metastatic volume was explained by an increase in the total number of metastases rather than by accelerated growth of the present metastases, and the number of metastases was elevated after pHx compared to the two other groups, analogous to the total metastatic volume ( Figure 2E , Supplementary Figure 1F ). Regarding the size of each liver metastasis individually over the postoperative period, we found no difference in the volume increase between pHx, SHAM, and control mice ( Figure 2F ). Consistently, at transcriptional level, we found a significant induction of cell proliferation represented by Ki67-expression within the healthy liver after pHx compared to SHAM and control mice, both in quantitative real-time polymerase chain reaction (qPCR) and immunohistochemistry (IHC) with p<0.05 ( Figures 2G, H ). However, proliferation in liver metastases was not altered following hepatectomy ( Figures 2I, J ). The detected general decrease in Ki67-expression in all three models ( Figure 2J ) compared to preoperative levels can be explained by the increase in necrotic tissue in larger metastases over time. In summary, pHx increased the number of de novo metastases without affecting the proliferation and growth of each metastasis. Therefore, we assumed that one of the major implications of pHx is priming of the liver for de novo metastatic implantation.

It is well established that inflammation and the tumor microenvironment (TME) heavily influence CRC carcinogenesis and metastasis (30), therefore we analyzed the effect of pHX on the TME composition of primary tumors (PT) and liver metastases (LM), and immune cell compartments in healthy liver tissues. After generating single-cell suspensions of all three tissues, immune cell compositions were analyzed using fluorescence-activated cell sorting (FACS). Regarding innate immune cells, we found an increased number of neutrophils both in healthy livers and metastases after pHx, as well as a moderate increase in monocyte count ( Figure 3A , Supplementary Figure 2A ). Comparing the abundance of these cell types at different time points after resection revealed that the accumulation in the liver did not occur before POD5 ( Figure 3B ). These changes were not accompanied by a general increase in neutrophil or monocyte abundance in the peripheral blood ( Supplementary Figure 2A ). As surgical stress has been linked to an increase in myeloid-derived suppressor cells (MDSC) in tumors, we characterized the MDSC subsets in the liver and peripheral blood in depth. We did not observe a significant difference in the abundance of both polymorphonuclear (PMN-MDSCs) and monocytic MDSCs (M-MDSCs) in either tissue after pHx. However, a change in the PMN-MDSC subtypes towards a monocytic phenotype [described by Veglia et al. (31)] was observed: pHx caused an increase in tumor-associated CD14^high PMN-MDSCs in the liver, while decreasing CD14^low cells ( Supplementary Figure 2B ). As the microbiome is an important constituent of the TME of CRC and affects CRC pathogenesis (30), we analyzed bacterial infiltration in the liver after pHx. Along with the increased infiltration of myeloid cells, we found an increased presence of bacteria within the liver following pHx ( Figure 3C ) as analyzed by qPCR, whereas bacterial infiltration of the LM was not altered by pHx ( Supplementary Figure 2C ). Taken together, pHx induces an inflammatory composition with an elevated abundance of innate immune cells in the residual liver as a delayed effect, which may be supported by increased bacterial infiltration.

Having shown that pHx surgery alters innate immune cell responses, we wondered whether pHx also affects adaptive immunity in CRLM. In our pHx mouse model, we could show that the numbers of cells of the adaptive immune system (T cells, B cells and NK cells) in the liver or LM did not change significantly after the respective surgery. However, a trend towards a decrease in CD4⁺ and CD8⁺ T cell abundance in the liver ( Figure 3D ) and especially in metastases ( Supplementary Figure 2D ) following pHx was observed in FACS analysis. To further investigate the role of T cells in this metastasis model, we performed IHC staining for CD4⁺ and CD8⁺ T cells in liver sections. We found that both CD4⁺ and CD8⁺ T cells were scarce in healthy liver tissue, accumulated around premetastatic lesions, and were completely depleted in macrometastases ( Figure 3E ). However, not only the absolute abundance of adaptive immune cells impacts tumorigenesis and metastasis, but also the function of the adaptive immune cells heavily influences tumor progression. Therefore, we functionally characterized the CD8⁺ T cells for their functionality in depth. First, we analyzed T cell cytotoxicity, exhaustion, and activation of CD8⁺ T cells in the liver in the different intervention groups using FACS. While we observed a general increase in exhaustion markers (PD-1, Tcf1) on CD8⁺ T cells along tumor progression in accordance with the present literature, pHx did not affect the expression of exhaustion markers compared to SHAM or control samples ( Supplementary Figure 2E ). Furthermore, we do not detect any changes in T cell cytotoxicity (granzyme B) and activation (CD38) after pHx ( Supplementary Figure 2E ).

As the livers of mice undergoing pHx shows veritable changes in the immune cell compartments, we wondered whether these changes lead to an impairment of the initiation of the metastatic process in the livers. To analyze the underlying transcriptional alterations in the healthy livers, we performed bulk RNA sequencing from healthy livers of four PO- and four pHx-mice. In total, 1083 upregulated and 289 downregulated differentially expressed genes (DEGs) were identified comparing the livers of PO and pHx groups after batch effect removal ( Figures 4A, B ). In order to define metastasis-associated pathways regulated after pHx, we performed Gene Set Enrichment Analysis (GSEA) with Hallmark gene sets, revealing numerous enriched pathways impacting metastasis ( Figure 4C ), such as the Hallmarks E2F targets and G2M checkpoint. Both gene sets were enriched in the pHx group and encompass important transcripts that regulate cell-cycle progression ( Figure 4D ). In addition to an increased proliferative activity, we observed an enrichment of genes involved in epithelial-mesenchymal transition ( Figure 4C ). The most striking enriched gene sets regarding a potential influence on metastatic seeding were the Hallmark terms angiogenesis, hypoxia, and inflammatory response ( Figures 4E–G ), which are interconnected pathways that potentially mediate the prometastatic effect of the surgical procedure ( Figure 4H ). Therefore, we showed that pHx may facilitate the efficient seeding of LMs due to a prearrangement of prometastatic factors in the liver.

As we have shown that several genes affecting premetastatic niche formation are increased in the transcriptomic analysis of pHx livers, we analyzed the formation of this special metastasis-favoring niche in pHx treated animals on transcriptomic and proteomic level in depth The residual liver following pHx is defined by an increased abundance of inflammatory cells and increased proliferation. Inflammatory tissue has been described to be more susceptible to metastatic seeding (32) and inflammation is an essential component of the premetastatic niche (10). To further investigate the effects of pHx, we focused on pathways that facilitate engraftment of circulating tumor cells in the organ. Several genes that regulate hepatic hypertrophy after liver resection have also been described in the context of the formation of a premetastatic niche in the liver for metastatic tumors (7, 8, 10). By analyzing the transcriptional profile of the healthy liver tissue via qPCR, we found a significant induction of several genes associated with the premetastatic niche in the hepatectomy mice compared to preoperative levels, as well as to control mice (e.g., the inflammatory genes S100A9 or IL-6, growth factors such as VEGF and TGF-β or MMPs as markers of extracellular remodeling) ( Figure 5A ). These changes in gene expression are specific for partial hepatectomy and not a general effect of surgical trauma, as they were not observed after SHAM surgery ( Figure 5A ). Several of the described genes can be directly linked to neutrophil invasion, either as they attract neutrophils [e.g. IL-6 (33) or TIMP1 (34)], or as a result of neutrophil activity [e.g. S100A9 (35), Fibronectin (36), LOX (37), HIF-1α (38) or MMP2/9 (39)]. Transcriptional regulation was confirmed at protein level via western blot, where we could also find an upregulation of representative proteins involved in the premetastatic niche (e.g., S100A8 (interaction partner of S100A9, together forming the heterodimer calprotectin) and MMP9) ( Figures 5B, C ). Furthermore, the intrahepatic localization of MMP9 and S100A8 was detected by IHC of liver sections, showing an increased presence of both proteins around perivascular premetastatic lesions ( Figures 5D, E ). Hence, we conclude that prometastatic transcriptional changes along with the inflammatory cellular composition occur in defined sectors within the macroscopically healthy liver and lead together to the formation of premetastatic niches in the liver.

One effect of hepatectomy during liver regeneration is an increased formation of tight junctions (TJ) (40). As some members of the zonula occludens proteins are thought to enable intrahepatic extravasation of circulating tumor cells, we analyzed the expression of the genes Claudin-2 and -5 as well as Occludin and ZO-1 in liver tissues in this mouse model. We found that Claudin-2, Claudin-5, Occludin and ZO-1 gene expression was increased following pHx compared to PO, Ctrl or SHAM samples in qPCR ( Figure 6A ) and on proteomic level the expression of the tight junction proteins Claudin-2, Occludin und ZO-1 was elevated following pHx in western blot ( Figure 6B ). This finding was supported by RNA sequencing data that showed enrichment of the apical junction gene set ( Figure 6C ). Additionally, we analyzed the expression of several TJ genes in primary tumors and healthy intestines, as they are linked to a worsened prognosis in several tumors and have been shown to drive colorectal carcinogenesis and EMT (41). We found that a similar upregulation of Claudin-5, Occludin and ZO-1 was observed in the healthy intestine both on transcriptomic and proteomic level ( Figures 6D, E ). Furthermore, an upregulation of Occludin in primary tumors was detected ( Figure 6F ). As tight junctions are formed in the interactions of several cell types, we stained proteins using immunofluorescence of sections of primary tumors and attached healthy ceca for intestinal localization. In the healthy intestine, TJ proteins Claudin-2, E-cadherin and ZO-1 were located at the membrane of epithelial cells with emphasis on the luminal side of the crypts ( Figure 6G ). In tumor tissue, we found TJ proteins with no spatial orientation and partially within the cytoplasm of the cells, indicating depolarization of the cells ( Figure 6G ), in accordance with the literature.

Since we observed an increase in de novo metastases in our model, we were curious if pHx has an impact on the metastatic potential in the primary tumor. While we detected a trend towards increased Ki67-expression in primary tumors ( Supplementary Figures 3A, B ), the expression of genes involved in EMT revealed no significant changes in the expression of Ctnnb1, Vimentin, Snai1 and 2 or Zeb1 ( Supplementary Figure 3C ). Additionally, the expression of EMT genes was measured in liver metastases. The expression levels of Ctnnb1, Snai1 and 2 or Cdh1 were not influenced by pHx ( Supplementary Figure 3D ).

In summary, we have demonstrated that pHx causes various changes in the cellular and transcriptional composition of the residual liver in a mouse model of orthograde colorectal liver metastasis that can be recapitulated as inflammation, extracellular remodeling, and tight junction formation. The transformation in the residual liver favors further metastatic seeding, while the impact on the primary tumor seems to be insignificant, and ultimately results in an elevated metastatic burden following pHx (see Graphical Abstract).