Metastasis from the tumor interior and necrotic core formation are regulated by breast cancer-derived angiopoietin-like 7

Significance Aggressive tumors often die from the inside out, a process called necrotic cell death. Necrosis is associated with dissemination of cancer cells but how necrosis promotes tumor dissemination is not understood. Here, we used rats as a model organism to increase detection of dissemination events. This uncovered a sharp rise in circulating tumor cell (CTC) abundance associated with necrosis and changes in the blood vessels. Gene expression analysis revealed a tumor-derived gene program involved in shaping the tumor core ecosystem. We demonstrate that necrosis, vascular remodeling, and metastatic dissemination are dependent on a tumor-secreted factor, angiopoietin-like 7 (Angptl7). Understanding the molecular factors regulating metastatic dissemination from the necrotic core could unveil therapeutic strategies to treat and prevent metastatic cancers.


Figure S1
CTC dynamics and necrosis in patient with metastatic breast cancer * Figure S10 A B   Figure S1: Orthotopic transplantation into rats produce 3x larger tumors, 10x more CTCs, and 4x more lung metastases than into mice. S1A) Experimental schema. 4T1 mouse mammary tumor cells labeled with membrane GFP (4T1-GFP) were orthotopically transplanted into a single #4 mammary fat pad of SRG rats (n=6) or NSG mice (n=24). Animals were sacrificed at 24 days post-transplantation. S1B-C) Final tumor weight and estimated final tumor volume per animal. S1D) Blood volume collected per animal. S1E) Representative micrographs of single CTCs and CTCs clusters from 4T1-GFP transplanted SRG rats. GFP denoted in green. DAPI in blue. S1F-G) Single CTC and CTC cluster abundance per animal. Individual blood samples from 8 mice were pooled into one tube for a total of n=3 samples. Events per individual animal are reported. S1H) Percentage of CTC events that were single CTCs or CTC clusters. Mean ± SD. S1I) Representative stereomicroscope images of lung metastases from NSG mice and SRG rats.
S1J) The number of lung metastases determined by stereomicroscopy in transplanted SRG rats and NSG mice.
Mean values shown on graphs. All P-values determined by Welch's t-test. S2J-K) Correlation between lung metastases and necrotic area (J) and lung metastases and tumor size (K). Pearson R. Linear trend lines shown with 95% confidence intervals.

P-values for A-D and I determined by one-way ANOVA. P-Values for G-H determined by
Welch's t-test.
Supplemental Figure S3: Additional information on characteristics of blood vessels.
S3A) Dilated blood vessel spatial localization. Distance of VE-cadherin-positive dilated vessels to necrosis or tumor border was measured then normalized to a spatial proximity score of 0 to 1, where 0 means a spot is on the tumor border, and 1 means a spot is in a necrotic region. Spatial localization of all cells were plotted as a control. P-value determined by Kruskal-wallis test. S4F) Western blot of necrotic core and non-necrotic rim regions of 4T1 tumors to assess protein-level differential expression of Angptl7. Angptl7 is highly expressed in necrotic region of the tumor but not the non-necrotic region. GAPDH was used as loading control. 4T1 cells in vitro do not express Angptl7 and was used as the negative control. N=3 rat tumors.
S4G) Quantification of protein-level expression of Angptl7 from S3F. P-value determined by paired t-test.
S4H) Angptl7 spatial localization. Distance of Angptl7+ spots to either necrosis or tumor border was measured then normalized to a spatial proximity score of 0 to 1, where 0 means a spot is on the tumor border, and 1 means a spot is in a necrotic region. Spatial localization of all cells are plotted as a control. P-value determined by Kruskal-wallis test.
Supplemental Figure S5: Angptl7 expression increase with day post-transplantation, correlates with necrotic area, lung metastases, and dilated vessels.
S5A) Angptl7 detections from day 13 to day 27 post-transplantation based on RNA ISH. Pvalues determined by one-way ANOVA.
S5B-E) Correlation between Angptl7 detections per tumor and necrotic area, lung metastases, dilated blood vessels, or CTC events. Pearson R. Linear trend lines shown with 95% confidence intervals.
Supplemental Figure S6: Additional information on in vivo effects of Angptl7 suppression.
S6B) Estimated final tumor volume of non-targeting and Angptl7 KD tumors.
S6C) Representative images of single CTCs and CTC clusters from Angptl7 knockdown or nontargeting control transplantation into SRG rats. Cells are mCherry-positive if they express shRNAs. 4T1 cells are labeled by membrane GFP. DAPI marks nuclei.
S6D-E) GFP-positive single CTC and CTC cluster abundance in Angptl7 knockdown and nontargeting control.
All graphs reported as mean ± SD and p-values determined by one-way ANOVA.
Supplemental Figure S7: Additional information on in vivo effect of Angptl7 suppression on gene expression.
S7A) Gene sets were generated composed of tumor-derived and host-derived core and rim genes with fold change ≥ 2 and FDR ≤ 0.001. The relative enrichment of these 4 gene sets was determined for tumor core or rim from Angptl7 knockdown and non-targeting control.

3D Cell Culture:
2D cultured cells were trypsinized with 0.25% trypsin, then quenched with complete RPMI. Cell suspension was centrifuged at 400g for 5mins, then resuspended in Accumax. Accumax suspension was place in 37°C water bath for 30mins, pipetting up and down every 10 minutes. After centrifuging for 5min at 400g, pellet was resuspended in complete RPMI + 2% Corning® Matrigel® Growth Factor Reduced (GFR) Basement Membrane Matrix (354230) (Corning). Cells were plated in non-adherent 6 well plates at 150,000 cells/mL with 4mL media per well.

SRG rats:
The SRG OncoRats, SCID rats on the Sprague-Dawley background that harbors a double knockout for the Rag2 and Il2rgamma genes, were purchased from Hera Biolabs. SRG rats are a double knockout for the Rag2 and Il2rgamma genes (SD-Rag2tm2hera Il2rgtm1hera) and lack B-cells, T-cells, and NK-cells.

Orthotopic transplantation into mammary fat pad of rats and mice:
Cells to be transplanted were cultured 3D for 2 passages before being transplanted. After the second passage, cells were plated in non-adherent 6 well plates at 150k cells/mL with 4mL media per well for the cells to aggregate over 24 hours before being transplanted. Spheroids formed this way were resuspended in 1:1 Matrigel:DMEM/F12 mix, kept on ice. 600,000 cells in 20µl Matrigel DMEM/F12 mix were transplanted into the right T4 mammary fat pad of rats or mice. Estimated tumor volume was calculated based on caliper measurements with the formula: V = (W 2 × L)/2.

Blood collection from rats and mice:
For rats, peripheral blood was collected terminally though the left ventricle of the heart, into CellSearch CellSave tubes (Menarini-Silicon Biosystems) with heparin-coated syringe attached to an 18g needle. Rats were put under deep anesthesia with isoflurane to do this, and rats were euthanized immediately after blood collection. Blood tube was inverted at least 10 times immediately after blood collection, and blood was processed within 4 hours of collection. For mouse blood, peripheral blood of 8 mice were collected into one CellSave tube using a new heparin coated syringe for every mouse.

Density separation of buffy coat (nucleated cells) from blood:
Peripheral blood was diluted 1:1 with D-PBS then layered on top of 8mL of Ficoll (GE Ficoll Paque Plus GE17-1440-02 for human blood and Ficoll Paque Premium GE17-5442-02 for mouse and rat blood) in a 2.5% BSA-coated 50mL conical tube. Tube with Ficoll and diluted blood were centrifuged at 400g for 35mins with 0 acceleration and deceleration. The cloudy, white layer ("buffy coat") between the clear Ficoll layer and the top plasma layer and the plasma layer were collected into a separate BSA-coated tube. Buffy coat and plasma layer mix were centrifuged at 4°C at 3500g for 30mins to pellet nucleated cells, then resuspended with D-PBS. To visualize Buffy Coat containing CTCs, cell suspension was spun onto slides using a Cytospin (800g for 5mins). After drying slides completely, slides were fixed with 4% PFA for 5 mins, washed with D-PBS 5mins 3 times. Slides were dried completely again before being stored in -80°C. For plasma collection, plasma layer was collected from the plasma layer from the Ficoll density separation or from a separate blood tube in an AccuCyte Blood collection tube processed by Rarecyte with a 3000g 25 minutes spin at 25°C.

Immunofluorescence staining:
Slides were thawed at room temp for 5 minutes. For OCT tissue sections, slides were washed 5 mins 3x to wash OCT off. Slides were block for 1 hour with blocking solution (2.5% BSA, 5% normal goat serum or normal donkey serum, 0.3% triton in D-PBS). Primary antibody in blocking solution was placed on slides and incubated at room temp for 2 hours. We then washed the slides 5mins 3x with D-PBS before placing the secondary antibody in 5% normal goat serum or normal donkey serum in D-PBS on the slide to incubate for 1 hour. Slide was washed again 5mins 3x with D-PBS. Excess moisture was removed from the slide, and slide was mounted with 30-50µl Prolong Diamond Antifade Mountant and glass coverslip. Slide was cured overnight room temp, then stored at 4°C.
Bulk RNA-seq: 4T1 cells were transplanted into the right #4 mammary fat pad of SRG rats. At day 27 post-transplantation, primary tumors were harvested. Tumor was cut in half to expose the cross section. The inner, necrotic region and the outer, non-necrotic region were separated and snap frozen for RNA-seq and sent to Genewiz-Azenta for RNA extraction, library prep, and sequencing. Qiagen RNeasy kit (Qiagen, Inc 74104) was used for in-house RNA extraction before samples were sent to Genewiz.

Initial Bioinformatics analysis:
Image analysis and base calling were conducted by the HiSeq Control Software (HCS). Raw sequence data (.bcl files) generated from Illumina HiSeq was converted into FASTQ files and de-multiplexed using Illumina's bcl2fastq 2.17 software. One mismatch was allowed for index sequence identification.

RNA-seq computational deconvolution of mouse and rat genomes:
RNA-seq reads in xenograft samples were deconvolved for identification of the species of origin following a method similar to that described by Wingrove et al. 2 . A catenated genome was formed by catenating the Rat (Rnor_6.0) and mouse (GRCmm28) genomes and gene annotations. Prior to alignment, reads were trimmed using Trim Galore and read quality was confirmed using fastQC. Trimmed reads were identified as originating from either rat or mouse by aligning reads against the catenated genome using STAR. Paired reads with both mates aligning uniquely to either the rat or mouse genome were separated by species and counted to gene annotations using featureCounts 3 for downstream differential expression analysis. Within-species normalization of mapped reads and differential expression analysis were performed using the limma (version 3.50.0) 4 and edgeR (version 3.36.0) 5 packages in R following the procedure described in detail by Law et al. 6 . Prior to differential expression analysis, very low abundance genes were filtered using the filterByExpr function in edgeR using default parameters. Effective library sizes were normalized across conditions using the trimmed mean of M-values (TMM) normalization procedure. Differential expression analysis was performed on the filtered and normalized counts using the limma-voom linear modeling pipeline 7 . For experiments in which samples were paired by animal, animal ID was treated as an additional random effect in the linear model. Empirical Bayes smoothing was applied on both the linear model and fitted contrast coefficients using edgeR. False-discovery rates for differential expression at the gene-level were calculated using the Benjamini-Hochberg correction procedure with the toptable function in limma. Normalized transcript abundances are reported as the TMM-normalized counts-per-million obtained from the voom function in limma. Gene set enrichment analysis was performed using GSEA software (Broad Institute, v 4.2.3) 8 . Genes were ranked by log-fold change and collapsed to human orthologs using the MDSig ensembl gene id orthologs chip platform (v 7.3). GSEA was performed using a pre-ranked analysis weighted by log-fold change against the C5 gene ontology collection (v 7.5). False-discovery rates were determined using 100,000 permutations. For differential expression analysis between species (tumor and host), a modified normalization scheme was applied to account for differences in gene lengths and annotation depth between species, in addition to the standard normalization by effective library size. Our method resembles that described by Oziolor, et al 9 . First, using reads that mapped uniquely to one species of the catenated genome, an intra-species TMM normalization was performed separately for host-and tumor-mapped reads. Lengthcorrected transcript abundances were calculated as reads per kilobase million (RPKM) in edgeR, and then converted to transcripts per million (TPM). Homolog pairs were obtained by querying the Ensembl database for linked rat (rattus norvegicus Ensembl release 100) and mouse (mus musculus Ensembl release 100) datasets using the getLDS function in biomaRt (v 2.50.3). Next, both gene sets were filtered to only those with rat/mouse homolog pairs, and the rat genes were mapped onto their corresponding mouse homologs. TPM abundances originating from both species were combined into a single data set, treating species of origin as a sample identifier, for all further analysis.
To correct for differences in genome annotation depth between species, TPM values were normalized using the TMM method in edgeR. Homolog-wise differential expression analysis was performed between species using TMM-normalized, homologmapped TPM values as input into the limma-voom analysis pipeline (as above) in place of library-normalized counts. Prior to identifying top genes that were differentially expressed in the mouse tumor over the rat host, homolog pairs that exhibited zero read counts in the rat homolog for all of the samples were discarded, since these may reflect a limitation of the homolog mapping. Then, genes from mouse-core-mouse-rim were cross referenced with mouse-core-rat-core. Genes that were mouse-core vs. mouse-rim Log Fold Change ≥ 1 and FDR ≤ 0.01 and with detectable rat expression are shown in Fig 4C.

Real Time qPCR:
RNA from tumors and cell pellets were extracted using Qiagen RNeasy kit. RNA was reverse transcribed into cDNA using Superscript™ III First-Strand Synthesis System (Thermo Fisher Scientific Catalog number: 18080051) with equivalent amounts of RNA (1µg). cDNAs were mixed with indicated gene-specific primers listed and PowerUp SYBR Green Master Mix (Thermo Fisher Scientific Catalog number: A25741) and qRT-PCR was performed using an Applied Biosystems QuantStudio5 System.

QuPath quantification of RNA ISH:
RNAscope images were opened in QuPath 10 . Thresholds were created to identify tumor area and necrotic area by eye, verified to be accurate in test images, and applied to demarcate necrotic zones. Angptl7-expressing cells were identified using positive cell detection. The distance to annotation 2d feature was used to determine the distance between cells and the nearest necrotic border and cells and nearest tumor border. This process was automated in QuPath for consistent analysis across images.

Necrosis Measurements:
Tumor necrosis measurements were determined based on Hematoxylin & Eosin (H&E) or 2,3,5-triphenyltetrazolium chloride (TTC) assay. For H&E staining, tumors were sliced into 2-3 ~5mm slices by scalpel, fixed for 5 days in 10% formalin at 4°C, rocking before being sectioned and stained with hematoxylin and eosin. For the TTC assay, tumors were sliced into 2-3 ~5mm slices and stained with 1g/100ml tetrazolium salt in a 7.4pH buffer with 77.4% NaH2PO4 (0.1 M) and 22.6% Na2HPO4 (0.1M) mix, at 37°C for 20 mins, rocking. Tumor slices were then fixed with 10% formalin for 20 mins before being visualized. In the TTC assay, the TTC compound is reduced to a red TPF (1,3,5triphenylformazan) compound in live tissues due to dehydrogenase activity. White areas therefore indicate necrotic tissue, and red areas are viable regions.

Spatial enrichment localization score:
For each spot x, the spatial enrichment score (SES) was computed as Dt/(Dn+Dt) where Dn equals the distance between spot x and the nearest necrotic border and Dt equals the distance between spot x and the nearest tumor border (Dt). Accordingly, SES(x) varies from 0 to 1: equal to 0 for perfect localization to the tumor border, and 1 for perfect localization to the necrotic border. Distances calculated from QuPath were exported to R and analyzed using a custom script. A non-parametric Kruskal-Wallis test was applied to evaluate statistical significance between different SES distributions.

LC/MS plasma proteomics
Sample Preparation: Each plasma sample was depleted of high-abundance proteins by injecting 80 µL of the sample onto a Michrom Bioresources Paradigm HPLC equipped with an Agilent human MARS-6 depletion column. Unbound material eluting from the column, as observed on a UV detector, was collected and the protein concentration was determined. The volume corresponding to 100 µg of protein was subjected to the reduction of disulfide bonds by adding TCEP (tris(2-carboxyethyl)phosphine to a final concentration of 5 mM and incubating at room temperature with vortexing for 15 min.
Protein alkylation was carried out by adding 2-chloroacetamide to a final concentration of 10 mM and incubating at room temperature for 30 min. The volume of each sample was reduced to approximately 100 µL by vacuum centrifugation and methanol/chloroform precipitation was carried out. Protein pellets were washed with methanol and resuspended in 50 mM HEPES pH 8.7. Proteolytic digestion was initiated with the addition of 1 µg of Lys-C and incubating at room temperature with low vortexing for 2 hours, followed by the addition of 1 µg of trypsin and incubating overnight at 37 °C on an orbital shaker set at 700 rpm. TMT Labeling: ThermoScientific TMTpro-16plex reagent was brought to room temperature and resuspended in 20 µL of anhydrous acetonitrile and vortexed for 15 min. Labeling was carried out by adding 20 µL of each TMT reagent to its assigned sample and vortexing the samples occasionally at room temperature for 1 hour. A "label check" was performed by combining 2 µL from each labeled sample into one tube, removing the acetonitrile by vacuum centrifugation, desalting the sample on a Harvard Apparatus C18 ultra-micro spin column, and analyzing the desalted TMTlabeled peptides by LC-ESI-MS/MS. After data analysis (see below), the peptide to spectrum match (PSM) results were analyzed and labeling efficiency was determined to be greater than 98%. After this labeling check, hydroxlyamine was added to each labeled sample to a concentration of 0.5% in order to fully quench the labeling reaction. All samples were combined equally by correcting for the total protein abundances measured for each sample in the labeling check analysis. The equalized pool was subjected to vacuum centrifugation to remove acetonitrile and then desalted on a Waters SepPack C18 (3cc, 200 mg) cartridge. The desalted elution was split into equal fractions and taken to dryness. One of the fractions was injected onto a ThermoScientific Vanquish HPLC equipped with an Agilent 2.1 mm x 150 mm C18 Extend column and fractionated into a 96-well plate using basic reverse-phase conditions. The 96 fractions were concatenated into 24 pools (pool 1: fractions 1, 25, 49, 73; pool 2: fractions 2, 26, 50, 74; etc.) that were taken to dryness and each pool was analyzed by LC-ESI-MS/MS. LC-ESI-MS/MS: The generated basic reverse phase fractions were brought up in 20 µL of 2% acetonitrile in 0.1% formic acid and 5 μL was analyzed by LC/ESI MS/MS with a Thermo Scientific Easy1200 nLC (Thermo Scientific, Waltham, MA) coupled to a tribrid Orbitrap Eclipse with FAIMS (field asymmetric ion mobility spectrometry) mass spectrometer (Thermo Scientific, Waltham, MA). In-line de-salting was accomplished using a reversed-phase trap column (100 μm × 20 mm) packed with Magic C18AQ (5μm, 200 Å resin; Michrom Bioresources, Auburn, CA) followed by peptide separations on a reversed-phase column (75 μm × 270 mm) packed with ReproSil-Pur C18AQ (3μm, 120 Å resin; Dr. Maisch, Baden-Würtemburg, Germany) directly mounted on the electrospray ion source. A 120-minute gradient from 4% to 44% B (80% acetonitrile in 0.1% formic acid) at a flow rate of 300 nL/minute was used for chromatographic separations. A spray voltage of 2300 V was applied to the electrospray tip and the