A tissue microarray was constructed using 1 mm cores from 95 tumor blocks of stage II infiltrating large bowel carcinomas from the surgical pathology database of the Hospital Fundación Jiménez Díaz (Madrid, Spain). Two pathologists independently selected the most representative areas and reviewed histopathological features. For mRNA analyses, we used freshly frozen stages I–IV tumor samples (cohort 1) and formalin-fixed paraffin-embedded tumor samples from patients with stage III CRC (cohort 2), both from the Hospital Clínico San Carlos Tumor Bank (Madrid, Spain).21 Appropriate informed consent was obtained from all patients and no personal data were registered.

C57BL/6J, Tg(TcraTcrb)1100Mjb/J (OT-I), Tg(TcraTcrb)425Cbn/J (OT-II), and B6.Cg-Tg(Tek-cre)1Ywa/J (Tie2-Cre) mice were from The Jackson Laboratory (Bar Harbor, ME). SOD3^−/− mice were kindly provided by Tim D. Oury (University of Pittsburgh, Pittsburgh, PA). SOD3^EC-Tg mice were generated by crossing loxP-SOD3KI (SOD3^Cre−) mice21 with Tie2-Cre transgenic mice. HIF-2α^EC-KO mice were generated as described.21

The Lewis lung carcinoma (LLC; ATCC), the EG7-SOD3 thymoma, and the murine microvascular 1G11-mock and 1G11-SOD3 cell lines were cultured as described.21 27 The OVA-expressing EG7-SOD3 thymoma (and the control EG7-mock) were generated by retroviral transduction and selected by cell sorting (MoFlo XDP; Beckman Coulter) using GFP fluorescent emission. HIF-2α was overexpressed and silenced in 1G11 cells as described.21 Stable β-catenin mutant Δ90βcat overexpression was achieved by transfection with pCAG-Δ90-GFP (a gift of Anjen Chenn; Addgene no. 26645) and cells selected by cell sorting. For FoxM1 overexpression, 1G11 cells were transfected with pCMV6-Entry/FoxM1-Myc-DDK (Origene, MR210493) and clones selected by limiting dilution with neomycin (750 µg/mL; Apollo Scientific). For LAMA4 silencing, 1G11-SOD3 cells were transfected with esiRNA targeting mouse LAMA4 (esiRNA1; Sigma-Aldrich; EMU083481) or a medium GC content siRNA duplex (Stealth RNAi siRNA Negative Control Med GC, 12935300; Invitrogen). For SOD3 overexpression in vivo, high-titer stocks of adenovirus expressing mouse SOD3 (Ad-SOD3) or β-galactosidase (Ad-C) were prepared as described.21

For all tumor models, female mice 2 to 5 months old were used. Exponentially growing LLC, EG7, or EG7-SOD3 cells were implanted subcutaneously in the indicated mice. In experiments involving Doxo treatment, mice received intratumor Ad-mSOD3 or Ad-C viruses (10⁹ pfu/50 µL) injections on days 7, 9, 11, and 15 post-implantation, and Vhcl or Doxo (2.5 mg/kg, intraperitoneally; Farmitalia Carlo Erba, Italy) on days 7, 11, and 15. For in vivo WNT pathway inhibition, tumor-bearing mice were treated with XAV-939 or DMSO at days 5, 8, 11, 12, and 13. In experiments involving HIF-2α^EC-KO mice, tamoxifen (Sigma-Aldrich) was diluted in ethanol and corn oil, heated (100°C) and injected (1 mg/mouse, intraperitoneally) on days 3, 4, and 5 post-LLC or post-EG7-SOD3 inoculation. For some experiments, naïve or activated OT-I cells (10⁷ cells/mouse, intravenously) were adoptively transferred to tumor-bearing mice (see online supplementary methods). Tumors were measured blind to treatment or genotype, and tumor volume calculated using the formula V=(width²×length)/2. For Kaplan-Meier curves, mice were considered “dead” when tumor volume >700 mm³. OT-I cell number was estimated by FACS after staining with anti-Vα2-biotin, anti-Vβ5-PE, anti-CD45-BViolet570, and anti-CD8-eFluor450 antibodies, followed by streptavidin–APC. In the other tumor models, immune infiltration was determined by immunohistochemistry.

Adhesion and migration assays were performed with primary CD3⁺, CD4⁺, CD8⁺ T cells and CD11b cells isolated from the spleen of OT-II and OT-I mice by negative selection, using 10% fetal calf serum, CCL19 (250 ng/mL; naïve T cells), CXCL10 (500 ng/mL; activated T cells), CCL2 (250 ng/mL; CD11b cells), or basal medium as chemoattractants.

Total RNA was extracted from mouse tumors and cell lines using TRI-Reagent (Sigma Aldrich) or RNeasy Mini or Micro Kit (Qiagen). RNA from human samples was extracted using TRIzol (Invitrogen) or RNeasy FFPE Kit (Qiagen), as described.21 mRNA levels were quantified by qPCR in an ABI PRISM 7900HT System (Applied Biosystems), using a SYBR EvaGreen-based reaction mix (5× PyroTaq EvaGreen qPCR Mix Plus ROX, Cmb-Bioline), with the primers listed in online supplementary table S1. β-Actin or CD31 (mouse samples) and RLP10A genes (human samples) were used for normalization. ChIP assays21 were performed with the Magna ChIP G (17-611; Merck Millipore). Each data point represented is the average of a technical triplicate in an independent experiment.

RSEM-normalized RNAseq expression data from 382 CRC tumor samples were obtained from TCGA through the Firebrowse repository (http://firebrowse.org/). A T cell inflammatory signature was computed as the average of the log expression values of 1346 genes.28

Qualitative data are represented with percentages and absolute numbers, quantitative data as mean±SEM, unless otherwise indicated; individual values are also shown and the number of replicates is given in figure legends. For mouse studies, sample size was estimated using the power method and values corrected for 20% attrition. For human samples, cases were divided into negative and positive for SOD3 or LAMA4 staining using a receiver operating characteristic (ROC) curve approach. For data with a Gaussian distribution and homogeneity of variances, statistical significance was calculated with a two-tailed Student’s t-test, with the Welch correction when the number of replicas was low, for comparison of two independent groups, and one-way or two-way ANOVA with Dunnett’s, Tukey’s, or Bonferroni’s post hoc test for multiple comparisons. When these requirements were not fulfilled, data were analyzed using non-parametric tests. The potential associations between qualitative variables were analyzed with the χ² test. Correlation analyses were performed using the non-parametric Spearman’s rank correlation coefficient. The outcome measure used for the human stage II cohort study was disease-free survival, defined as the time elapsed between surgical resection of the tumor with a curative intent and recurrence of disease in months. In the univariate analysis, Kaplan-Meier survival curves were compared with log-rank tests. SOD3 expression values were compared between T cell–inflamed and non-inflamed TCGA signatures using Student’s t-test. Differences were considered significant when p value <0.05. Statistical analyses were performed using the statistical package SPSS V.20.0 for Windows, and Prism software. All statistical tests were two sided.

We analyzed the in vitro adhesion and transmigration of purified murine T (CD3⁺) and myeloid (CD11b⁺) cells through monolayers of mock- and SOD3-overexpressing 1G11 cells, a mouse microvascular EC line modified to overexpress SOD3 (for overexpression data see21). SOD3 overexpression in EC increased CD3⁺ T lymphocyte adhesion and transmigration (figure 1A,B). This SOD3 effect was T cell specific, as it was not observed for myeloid CD11b⁺ cells, and applied for both naïve and ex vivo–activated CD4⁺ and CD8⁺ T cells (figure 1C,D).

To explore this selective SOD3 effect in vivo, we injected LLC cells into syngeneic immunocompetent mice; once tumors engrafted, mice received injections of adenoviruses expressing SOD3 (Ad-SOD3) to overexpress SOD3 in the TME, or β-galactosidase (Ad-C) as control, as verified by immunohistochemistry (IHC). In these conditions, SOD3 overexpression had no effect on tumor growth or immune infiltrate (online supplementary figure S1).

We hypothesized that the absence of immunogenic stimulus in the aforementioned tumor model might limit T cell infiltration, precluding any SOD3 effect. To overcome this limitation, an immune response to LLC tumors was boosted by doxorubicin (Doxo) administration, which induces immunogenic cell death29; Ad-SOD3 or Ad-C were then injected (figure 1E). Ad-SOD3-treated tumors showed reduced growth kinetics (figure 1F), which extended mouse survival (figure 1G). This was predicted, as SOD3 increases Doxo delivery into tumors.21 Analysis of the immune infiltrate also indicated a specific increase in TIL number (CD3⁺) in Ad-SOD3-injected tumors compared with Ad-C controls (figure 1H,I), which resulted in an increased lymphoid:myeloid ratio (figure 1J). SOD3 enhanced both CD4⁺ and CD8⁺ TIL density (figure 1K,L), but not of FoxP3⁺ Treg cells (figure 1M,N), which led to a higher CD8⁺:FoxP3⁺ cell ratio in Ad-SOD3-injected than in Ad-C-injected tumors (figure 1O). Baseline T cell infiltration was comparable in Ad-C-injected, Doxo-treated (figure 1I) and untreated LLC tumors (online supplementary figure S1); this suggested that the Doxo-induced immune response is not sufficient to boost T cell extravasation in tumors in the absence of SOD3 overexpression.

As lovastatin (Lov) upregulates SOD3 in the TME,21 30 to further confirm the SOD3 effect on TIL density, we implanted LLC tumors in WT and SOD3^−/− mice, followed by Doxo and Lov treatment. As predicted,21 Lov treatment induced SOD3 expression in LLC tumors implanted in WT mice (online supplementary figure S2B) and was associated with increased T cell but not CD11b⁺ cell infiltration (online supplementary figure S2C–F). Lov increased T cell infiltration only in WT-grafted tumors, which indicated that the effect was mediated by SOD3 expression. The extremely low SOD3 levels in LLC tumors implanted in vehicle-treated WT mice might explain the lack of differences in intratumor T cell number compared with tumors grafted in SOD3^−/− mice. Genetic or pharmacological upregulation of SOD3 in the TME therefore promotes selective tumor infiltration by T cells.

Since SOD3 overexpression or Lov treatment increases Doxo absorption into the tumor parenchyma,21 30 enhancement of TIL in SOD3-overexpressing tumors might be a collateral effect of boosted anti-tumor immune responses bound to Doxo treatment. To link SOD3 directly to enhanced T cell extravasation, we performed adoptive transfer of ovalbumin (OVA)−specific, naïve CD8⁺ (OT-I) transgenic T cells to mice engrafted with mock- or SOD3-overexpressing EG7 thymomas, which also express OVA (figure 2A). EG7-SOD3 and EG7-mock cells proliferated equally in vitro (online supplementary figure S3), and tumor growth kinetics were similar in mice that did not receive OT-I cells (figure 2B). Adoptive transfer of OT-I cells restricted EG7-SOD3 growth more efficiently than that of EG7-mock tumors (figure 2C,D). Indeed, OT-I cell transfer increased survival only in mice engrafted with EG7-SOD3 but not with EG7-mock cells (figure 2D), and this was associated with an increase in intratumor OT-I cells (figure 2E).

The enhanced anti-tumor effect observed in SOD3-overexpressing tumors might be solely a result of increased OT-I infiltration, but could also due to improved cytolytic activity of OT-I TIL. To discriminate between these possibilities, we analyzed the correlation between intratumor OT-I cell number and tumor size. OT-I TIL numbers correlated inversely with tumor volume in both groups (EG7-mock: r²=0.808, p=0.006; EG7-SOD3: r²=0.773, p=0.009; figure 2F), which suggests that SOD3 overexpression did not alter OT-I effector activity, but only their arrival to the tumor.

Histochemical analyses showed higher SOD3 levels in EG7-SOD3 than in EG7-mock tumors (figure 2G); unexpectedly, this tumor-produced SOD3 accumulated near CD31⁺ structures. To determine the relevance of perivascular SOD3 accumulation for TIL density, we crossed loxP-SOD3KI mice21 with mice expressing Tie2-Cre, which drive SOD3 expression to the endothelium. Tie2-Cre⁺ (SOD3 overexpression in EC; SOD3^EC-Tg) or negative (SOD3^Cre−) littermates were inoculated with EG7 cells, after which OT-I cells were transferred to tumor-bearing mice (figure 2H). Perivascular SOD3 expression was found in ~55% of CD31⁺ structures in tumors implanted in SOD3^EC-Tg mice, whereas SOD3 was almost undetectable in tumors from SOD3^Cre− mice (figure 2I). OT-I transfer reduced tumor growth more efficiently in SOD3^EC-Tg (8/9) than in SOD3^Cre− mice (3/9; figure 2J), which increased the survival of the former (figure 2K). This anti-tumor effect was associated with enhanced OT-I cell infiltration in SOD3-expressing tumors (figure 2L). These results suggest that SOD3 abundance in the TME, and particularly in the perivascular region, directly determines TIL density and thus the efficacy of adoptively transferred tumoricidal T lymphocytes.

Analysis of selected chemokines and cytokines in cultured 1G11-mock and 1G11-SOD3 cell supernatants showed no major differences, with the exception of a slight decrease in macrophage colony-stimulating factor (M-CSF) and CCL2 levels, and a slight increase in tissue inhibitor of metalloprotease (TIMP)−1 associated to SOD3 overexpression (online supplementary figure S4A,B). M-CSF and CCL2 might be implicated in myeloid cell migration and/or expansion,31 32 whereas TIMP-1 could inhibit lymphocyte infiltration.33 34 Of the CXCR3 ligands, only CXCL10 was slightly repressed in 1G11-SOD3 compared with 1G11-mock cells (online supplementary figure S4A); CXCL9 and CXCL11 were not detected. qPCR analyses confirmed downregulation of CXCL10 and CCL2 but upregulation of CXCL1 mRNA in SOD3-expressing cells (online supplementary figure S4C). Adhesion molecules necessary for leukocyte diapedesis, such as intercellular adhesion molecule-1 (ICAM-1) and vascular cell adhesion molecule-1 (VCAM-1), showed no differential expression between 1G11-mock and 1G11-SOD3 cells (online supplementary figure S4D).

We also analyzed CXCR3 ligand expression in tumor models in which SOD3 augmented T cell infiltration in vivo. In the LLC model (figure 1E), Ad-SOD3-injected tumors showed higher CXCL9 mRNA levels than those from Ad-C-injected tumors; CXCL10 and CXCL11 did not differ (online supplementary figure S4E). In contrast, none of the CXCR3 ligands changed significantly in OT-I-transferred tumors implanted in SOD3^Cre− and SOD3^EC-Tg mice (online supplementary figure S4F), although OT-I infiltration was significantly higher in the latter (figure 2L). These results could indicate that CXCR3 ligands are not necessary mediators of the SOD3 effect on TIL infiltration.

Since SOD3 triggers HIF-2α stabilization,21 we analyzed HIF-2α involvement in selective T cell transmigration. HIF-2α overexpression in EC enhanced (figure 3A–C), whereas HIF-2α silencing restrained transmigration of CD4⁺ and CD8⁺ T cells (figure 3D–F). These findings indicate an EC-autonomous SOD3 effect on T cell transmigration mediated by HIF-2α levels. To place HIF-2α unequivocally downstream of SOD3 in the regulation of T cell transmigration in vivo, we analyzed the infiltration of adoptively transferred, antigen-activated OT-I cells into EG7-SOD3 tumors implanted in mice with endothelial deletion of HIF-2α (HIF-2α^EC-KO)21 or control (HIF-2α^f/f) littermates (figure 3G). We found reduced infiltration of activated OT-I cell in tumors grafted in HIF-2α^EC-KO mice compared with controls (figure 3H). In addition, we retrospectively analyzed TIL density in Ad-SOD3-injected, Doxo-treated LLC tumors grafted in HIF-2α^EC-KO or HIF-2α^f/f mice (figure 3I). In this model, Ad-SOD3 injection increases Doxo anti-tumor effectiveness in tumors implanted in HIF-2α^f/f mice, but not when grafted in HIF-2α^EC-KO mice.21 IHC analyses with anti-CD3, anti-CD4, and anti-CD8 antibodies showed reduced TIL density in tumors grafted in HIF-2α^EC-KO mice compared with those grafted in HIF-2α^f/f littermates (figure 3J–L). These data suggest that endothelial HIF-2α is a necessary mediator of SOD3 to increase TIL numbers.

In normoxia, SOD3 stabilizes HIF-2α but not HIF-1α.21 A search for genes whose expression has been reported specifically regulated by HIF-2α identified six potential candidates (including VEC) by which SOD3 might regulate T cell transmigration. We focused on WNT1-inducible signaling pathway protein 2 (WISP2/CCN5), a non-conventional WNT ligand of the CCN (cysteine-rich 61 (Cyr61/CCN1)/connective tissue growth factor (CTGF/CCN2)/nephroblastoma overexpressed gene (NOV/CCN3)) family, since WISP2 modifies the perivascular environment through its interaction with the integrin α_vβ₃ 35 36; WISP2 also modulates the immune cell infiltrate in breast tumors,37 although to our knowledge, specific effects on T cell migration have not been described.

WISP2 protein (figure 4A) and mRNA (figure 4B) levels were higher in 1G11-SOD3 than in 1G11-mock cells, which suggests transcriptional regulation by SOD3. HIF-2α overexpression increased (figure 4C), whereas HIF-2α silencing reduced (figure 4D) WISP2 mRNA levels in 1G11 cells. In silico analyses identified four putative hypoxia-responsive elements (HRE) in the mouse WISP2 promoter (figure 4E). Chromatin immunoprecipitation (ChIP) showed that SOD3 enhanced HIF-2α association to the WISP2 gene promoter −2000 region (figure 4F), compatible with binding to the HRE region; no enhancement was found when a distal promoter region was analyzed. Moreover, WISP2 mRNA levels were reduced in EG7-SOD3 tumors grown in HIF-2α^EC-KO mice compared with those engrafted in control mice (figure 4G); Ad-SOD3-injected LLC tumors grown in HIF-2α^EC-KO mice also showed reduced WISP2 mRNA levels than controls (figure 4H). These results suggest a HIF-2α-dependent, SOD3-induced transcriptional upregulation of WISP2 in vitro and in vivo.

In addition to WISP2, overexpression of SOD3 in 1G11 cells upregulated WNT2 and WNT2b (also known as WNT13), whereas it downregulated WNT7a mRNA levels (figure 4I). WNT2 and WNT2b mRNA levels were increased by HIF-2α overexpression and reduced in HIF-2α-silenced cells (figure 4J,K). SOD3 appears to enhance specific WNT ligand transcription in a HIF-2α-dependent manner.

Stabilization of β-catenin is an indicator of WNT pathway activation.8 Whereas CTNNB1 mRNA levels were equivalent in 1G11-mock and 1G11-SOD3 cells (not shown), SOD3 overexpression increased protein levels of β-catenin, which localized mainly to the cytosolic and juxtamembrane region in 1G11-SOD3 cells (figure 5A); only a residual fraction colocalized with DAPI in the nucleus. Time-dependent β-catenin degradation in cycloheximide (CHX)–treated 1G11-mock and 1G11-SOD3 cells confirmed the higher β-catenin levels in SOD3-expressing cells, although these differences, significant only at initial time points, were lost progressively (figure 5B). β-catenin half-life was comparable between both cell types (3.9±0.9 hours, 1G11-mock; 5.3±1.2 hours, 1G11-SOD3; p=0.49, two-tailed Student’s t-test), which indicates that SOD3 does not affect the β-catenin degradation machinery directly. With the exception of AXIN2, we found no differences in the expression of well-known prototypic WNT/β-catenin-upregulated genes such as cyclin D1 (CCND1), MYC, and vascular endothelial growth factor-A (VEGFA)38 between 1G11-mock and 1G11-SOD3 cells (figure 5D). Transcription of all these genes was nonetheless readily induced in 1G11-SOD3 cells that overexpressed the nuclear N-terminal-truncated β-catenin mutant Δ90βcat (online supplementary figure S5). The lack of induction of canonical WNT/β-catenin genes in 1G11-SOD3 cells might not be an intrinsic defect, but linked to β-catenin retention outside the nucleus.

The WNT pathway also stabilizes the forkhead box M1 (FoxM1) transcription factor,39 whose deficiency causes severe abnormalities in lung vasculature during development.40 Total FoxM1 protein levels (figure 5E) and nuclear accumulation (figure 5F,G) were higher in 1G11-SOD3 than in 1G11-mock cells. SOD3 nonetheless did not alter FOXM1 mRNA levels (not shown), which suggests a posttranscriptional regulatory mechanism. These results thus indicate that SOD3 triggers some features of the WNT/β-catenin pathway in endothelial cells.

To analyze the importance of the SOD3-induced WNT pathway on T cell transmigration, WNT pathway was inhibited in 1G11-SOD3 cells by incubation with the pan-tankyrase inhibitor XAV-939. XAV-939 treatment reduced both β-catenin and FoxM1 protein levels (figure 5H), as well as the transmigration of CD4⁺ and CD8⁺ T cells (figure 5I). In contrast, WNT pathway agonism by WISP2 stimulation of 1G11-mock cells increased β-catenin and FoxM1 protein levels (figure 5J), and recapitulated the enhanced CD4⁺ and CD8⁺ T cell transmigration observed following SOD3 overexpression (figure 5K).

To determine whether the WNT pathway affects the enhanced TIL density associated to SOD3 overexpression, we adoptively transferred activated OT-I cells in EG7-SOD3 tumor-bearing mice pre-treated intratumorally with XAV-939 (figure 5L). SOD3 overexpression did not induce the WNT/β-catenin pathway in EG7 cells (online supplementary figure S6A); moreover, the short XAV-939 treatment caused no differences in tumor weight (online supplementary figure S6B). Compared with treatment with vehicle (Vhcl), XAV-939 significantly decreased OT-I infiltration of EG7-SOD3 tumors (figure 5M). These results place WNT pathway activation as a downstream mediator of SOD3 that enhances T cell transmigration in vitro and in vivo.

To search for potential WNT-regulated genes that could affect T cell transmigration, we focused on the BM component LAMA4, a transcriptional target of FoxM140 that provides permissive signals for T cell migration.17 18 FoxM1 overexpression also upregulated LAMA4 in 1G11 cells (online supplementary figure S7). LAMA4 mRNA and protein levels were higher in SOD3-expressing than in 1G11-mock cells (figure 6A–C). HIF-2α overexpression upregulated, whereas HIF-2α silencing reduced LAMA4 mRNA levels in these cells (figure 6D,E). 1G11-SOD3 cell treatment with XAV-399 reduced LAMA4 mRNA and protein levels compared with Vhcl-treated cells (figure 6F–H). Moreover, LAMA4 mRNA levels were reduced in XAV-939-treated tumors compared with controls (figure 6I). These data suggest that SOD3 triggers LAMA4 transcription in a HIF-2α-dependent and WNT-dependent manner.

This SOD3-induced LAMA4 expression is relevant for T cell transmigration since LAMA4 silencing in 1G11-SOD3 cells reduced CD4⁺ and CD8⁺ T cell transmigration compared with siNT-transfected cells (figure 6J,K). We also found that 1G11 monolayers seeded on laminin-411-coated inserts were more permissive to T cell chemotaxis than those on laminin-511-coated transwells (figure 6L). These experiments pinpoint LAMA4 as a potential SOD3 target in enhanced transendothelial T cell migration.

We then tested whether this SOD3/HIF-2α/WNT/LAMA4 hub correlated with TIL density in vivo by retrospective analysis of LAMA4 staining in some tumor models used during the study. Endothelial SOD3 overexpression in EG7 tumors implanted in SOD3^EC-Tg mice boosted infiltration of OT-I transferred cells (figure 2L); these tumors also showed an increase in the number of CD31⁺ structures decorated with LAMA4 compared with those implanted in SOD3^Cre− mice (figure 6M,N). We also observed an association between high TIL density, SOD3 overexpression, and strong endothelial LAMA4 staining in EG7-SOD3 (figure 6O,P) and Ad-SOD3-injected LLC tumors (figure 6Q,R), two conditions that enhanced TIL density (figures 2E,I). In contrast, perivascular LAMA4 levels were low in EG7-SOD3 tumors implanted in HIF-2α^EC-KO or XAV-939-treated mice figure 6S–V), two models with decreased TIL density (figures 3H and 5M). These data not only suggest LAMA4 induction by SOD3 in vivo, but also implicate LAMA4 as a potential mediator of SOD3-enhanced T cell infiltration.

To study whether SOD3 levels influence TIL in human cancers, we used a tissue microarray (TMA) of 95 primary stage II colorectal carcinomas (online supplementary table S2) from patients diagnosed between January 2000 and December 2010. Based on ASCO and ESMO guidelines,41 the patients received neither neoadjuvant nor adjuvant chemotherapy until recurrence. Thirty patients relapsed (range 3–76 months), similar to recurrence in some studies for this tumor stage.42 We found no statistically significant association between age or gender and disease progression; although there was a tendency toward association between disease progression and tumor grade, the number of high-grade lesions was too small (8) to infer statistical significance.

Although SOD3 is normally downregulated in CRC compared with healthy colon tissue,21 26 some tumors retain some SOD3 expression in the epithelial and stromal compartments. We determined the z-score43 for SOD3 staining in the TMA to classify tumors with negative (<140) or positive (>140) SOD3 expression, according to the cut-off obtained with a ROC curve. Immunohistochemical analyses showed that SOD3 positivity was associated to enhanced endothelial WNT signaling, using β-catenin as a readout, compared with SOD3⁻ tumors (figure 7A,B). The percentage of β-catenin-stained EC was also higher in SOD3⁺ than in SOD3⁻ tumors (n=94; p=0.04, Pearson’s χ²). β-catenin staining was usually intense in the epithelium and weak in the endothelium (figure 7A,B). Most tumors with high SOD3 levels showed intense β-catenin staining in the juxtamembrane region, but not in the nucleus of the tumor cells (figure 7A,B).

Analysis of the TMA also showed increased LAMA4 staining in SOD3⁺ compared with SOD3⁻ tumors (figure 7A). The association of SOD3 and LAMA4 staining involved both neoplastic and stromal cells. We found a median z-score value for LAMA4 of 85 (range 0–300). The ROC curve indicated a cut-off value of 70 to best discriminate sample weighing sensitivity and specificity; based on this value, 48.6% of the cases were negative for LAMA4 expression. We found a significant association between LAMA4⁺ and SOD3⁺ tumors (p=0.04, Pearson’s χ²), as well as a significant correlation between the z-scores for LAMA4 and SOD3 (figure 7C). In support of this association, LAMA4 mRNA was quantified in a cohort of 106 patients with CRC (stages I–IV) for which SOD3 mRNA was previously determined.21 In contrast to published data,44 we found that LAMA4 mRNA was downregulated in CRC compared with non-tumor samples (>10 cm from the primary tumor; figure 7D). In this cohort, we found a positive correlation between LAMA4 and SOD3 mRNA levels in CRC samples (figure 7E). Both downregulation of LAMA4 mRNA and correlation with SOD3 mRNA levels were further validated in a second cohort comprising 90 CRC stage III and 24 non-tumor samples (figure 7F,G). These results support a role for SOD3 in the transcriptional regulation of LAMA4.

In the stage II CRC cohort, we observed a clear association between SOD3 positivity and CD8⁺ T cell density (figure 7H; online supplementary figure S8). The average number of intraepithelial CD8⁺ T cells differed between SOD3-positive and SOD3-negative tumors (48.57 and 26.74, respectively; p<0.0001, two-tailed Student’s t-test). There was also a significant correlation between SOD3 z-score and intraepithelial CD8⁺ T cell counts (figure 7I), but not with CD8⁺ T cell density in the colon lamina propria (figure 7J), which suggests a specific role for SOD3 in intratumor T cell infiltration. The LAMA4 z-score also correlated with intraepithelial CD8⁺ TIL counts (figure 7K), suggesting a correlation between LAMA4 induction and effector T cell infiltration. To validate this observation, we correlated SOD3 mRNA levels with a T cell inflammatory signature28 in 382 CRC tumor samples from the TCGA. Mean SOD3 expression values were increased in CRC samples with high versus low T cell infiltration (10.17 vs 9.61; p=0.23×10⁻³), with a correlation coefficient of 0.19.

In accordance with published data,45 our stage II CRC cohort also showed an association of high density of intraepithelial CD8⁺ T lymphocytes with less relapse rate (n=95; p=0.01, Pearson’s χ²; OR 0.3 for CD8 (95% CI 0.1 to 0.7), p=0.05) and with extended recurrence-free survival (figure 7L). Extended recurrence-free survival was also linked to SOD3 positivity (figure 7M), consistent with the SOD3/CD8⁺ TIL correlation (figure 7I). Of the 14 patients with SOD3⁺ tumors, only one showed recurrence during the follow-up period, whereas malignancy recurred in 29/80 patients from the SOD3⁻ group (median time to recurrence 16 months; p=0.031, Pearson’s χ²; OR 0.14 (95% CI 0.02 to 0.98), p=0.06). High LAMA4 levels tended to be associated with reduced relapse rate (OR 0.6 (95% CI 0.3 to 1)) and extended relapse-free survival (figure 7N), consistent with its association to high CD8⁺ T cell density (figure 7K).