A Therapeutically Actionable Protumoral Axis of Cytokines Involving IL-8, TNFα, and IL-1β

IL-8 (CXCL8) was the first chemokine to be discovered on the basis of its ability to chemoattract polymorphonuclear leukocytes (1, 2). IL-8 is an important driver of inflammation, acting on its cognate receptors CXCR1 and CXCR2 (3, 4). The absence of the IL8 gene from the mouse genome has made research on the functions of this chemokine more difficult, although it is perceived as an important mediator for bacterial defense and autoimmunity (5). In cancer, IL-8 mediates several pathogenic roles (6) because it is able to recruit macrophages and granulocytes to the tumor tissue microenvironment that thereby become enabled to act as myeloid-derived suppressor cells (7, 8). In addition, IL-8 acts as a direct proangiogenic factor on tumor vasculature (9) and has been shown to mediate trophic effects on cancer cells and cancer stem cells (10). In fact, IL-8 may explain resistance to VEGF/VEGFR antagonists at least in some cases (11).

Recent reports have correlated the concentration of IL-8 in the serum of cancer patients with a profound lack of benefit from immunotherapy based on checkpoint inhibitors (12–14). The chemoattraction of myeloid leukocytes to the tumor is probably behind these observations, because tumor-associated macrophages (15) and, above all, the presence of granulocytes (16, 17) conveys a worse prognosis, in part as a result of impairing adaptive antitumor immune responses (18, 19). IL-8 not only attracts but also modulates the function of tumor-infiltrating myeloid leukocytes. For instance, IL-8 readily induces the extrusion of neutrophil extracellular traps (NET; refs. 8, 20). NETosis has been associated in mouse models with metastasis and cancer progression (7) and has recently been shown to interfere with cytotoxicity by T and natural killer (NK) cells (20).

In the tumor microenvironment, IL-8 is produced mainly by malignant cells with some contribution from myeloid leukocytes (21). An important question is what are the mechanisms that control IL8 transcription in tumors. Various factors have been described, but the main common denominator is the activation of NF-κB (22, 23). In this regard, considering inflammatory mediators present in the tumor and whose receptors act as NF-κB inducers, TNFα and IL-1β became obvious candidates (24). In fact, both TNFα and IL-1β are frequently expressed by leukocytes infiltrating human and rodent tumors, and IL-1R and TNFR1/2 are commonly expressed by tumor cells. Previous reports have shown that recombinant TNFα and IL-1β induce IL8 transcription in an NF-κB–dependent fashion (22, 23).

Blockade of TNFα and IL-1β has been shown to be advantageous for cancer treatment in various models (25, 26). IL-1β blockade with canakinumab reduced the risk of cancer-associated mortality in a large clinical trial testing if this monoclonal antibody (mAb) prevents cardiovascular events (27, 28). In mouse models of renal cancer, IL-1β is critical to elicit an immunosuppressive myeloid infiltrate (26).

TNFα blockade has been shown to exert antitumor effects in mouse models especially when combined with checkpoint inhibitors (29–31). In patients with cancer, TNFα has been blocked as a monotherapy intervention (32–34) and anti-TNFα agents have recently been used in combination with checkpoint inhibitors, showing promising results in a small cohort of patients with melanoma (35). The availability of clinical-grade inhibitory agents is explained because such cytokines can be pharmacologically blocked with therapeutic success for the treatment of rheumatoid arthritis, inflammatory bowel disease in the case of TNFα (36–39), and family periodic fever syndromes and juvenile rheumatoid arthritis in the case of IL-1β (40, 41).

In this study, we have been able to correlate IL1B and TNF transcription in tumors with the transcripts encoding IL8. In keeping with these findings, both TNFα and IL-1β induce functional IL-8 release from all tested human tumor cell lines representing a variety of tissue origins, as well as from primary organoids and fresh human cancer explants. In all these cases, including xenografted human tumors, IL-8 induction could be blunted by the pharmacologic blockade of TNFα or IL-1β.

To address whether the transcripts for the cytokines TNFα and IL-1β correlated with IL-8, The Cancer Genome Atlas (TCGA) database was interrogated in a pan-cancer fashion. As shown in Fig. 1A, TNF and IL1B showed clearly significant correlations with IL8 mRNA expression levels. Moreover, TNF and IL1B mRNA showed a good correlation with each other (Fig. 1B). As shown in Supplementary Fig. S1A–S1F, this association was observed for the most prevalent and deadly human malignant diseases. Interestingly, as can be visualized in the three-dimensional representation, the transcripts for the three cytokines correlated among them (Fig. 1C). Such an observation led to the hypothesis that the three cytokines might have a functional relationship in the tumor microenvironment.

To confirm such a strong correlative result, we accessed a database of primary colorectal cancer (Colonomics). In these series of patients, it was observed that IL8 correlated with TNF in the tumor tissue, but not in microdissected adjacent nontransformed tissue and normal colon mucosa (Fig. 1D and E).

To ascertain if IL-1β and TNFα could lead to tumor release of IL-8, we incubated a wide panel of human cancer cell lines representing colon cancer, lung cancer, renal cancer, breast cancer, and melanoma for 72 hours. It was uniformly found that TNFα induced secretion (Fig. 2A) and transcription (Fig. 2B) of IL-8. Importantly, cultures in the presence of the clinical-grade TNFα inhibitors infliximab and etanercept inhibited such effects (Fig. 2A). IL-1β also induced the biosynthesis and transcription of IL-8 at least in colon cancer and breast cancer cell lines up to comparable levels to those induced by TNFα (Fig. 2C). Of note, the effect of recombinant IL-1β was inhibited by soluble IL-1Rα (anakinra) that acts as an IL-1β trap. TNFα induced IL-8 via TNFR1 as shown by blockade with an anti-TNFR1–specific antibody, even if the tumor cells expressed detectable mRNA levels for both TNFR1 and TNFR2 (Supplementary Fig. S2A and S2B).

Next, we considered whether IL-8 could be cooperatively induced in tumor cells by a combination of IL-1β and TNFα. Results in Fig. 2D indicate that IL-8 production is additively increased by the combination of the two cytokines in a fashion that can be inhibited by the blocking agents for each respective cytokine. Furthermore, testing serial dilutions of these cytokines in combination indicated some degree of additive but not cooperative synergistic effects in three independent human cell lines derived from colon cancer, breast cancer, and non–small lung cancer (Fig. 2E).

As mentioned, IL8 is not conserved between rodents and humans, and thus we wondered whether there was TNFα and IL-1β control on other chemokines acting on CXCR1 and CXCR2, which could be the main functional orthologs in mice for IL-8. Indeed, as shown in Supplementary Fig. S3, CXCL1 and CXCL2 were induced by TNFα both at the protein (Supplementary Fig. S3A) and transcription (Supplementary Fig. S3B) levels in four different mouse tumor cell lines, indicating interspecies preservation of these functions. IL-1β also induced CXCL1 and CXCL2 in a manner that could be blocked by anakinra (Supplementary Fig. S3C).

In the tumor tissue microenvironment, activated T cells and macrophages may produce TNFα as well as malignant cells do (42). To mimic these conditions, we stimulated peripheral blood T cells with CD3⁺CD28 mAbs and collected the cell-free supernatants. Such supernatants readily induced IL-8 production from cultured HT29 colon cancer cells (Supplementary Fig. S4A and S4B). This was confirmed with supernatants from magnetically isolated CD4 and CD8 T cells before activation. In all these cases, etanercept or infliximab reduced IL-8 production, indicating that TNFα was the dominant IL-8–inducing factor present in such conditioned supernatants from activated T cells.

In the case of supernatants from monocyte-derived macrophages activated with LPS, again TNFα was the dominant mediator leading to IL-8 production by HT29 colon cancer cells (Supplementary Fig. S4A–S4C).

Because both T cells and macrophages could be responsible for IL-1β and TNFα production in the tumor microenvironment, we divided the transcriptomic analysis of TCGA across solid malignancies in hot and cold tumors with regard to T-cell infiltrates according to an established T-cell signature. Interestingly, the correlation of IL8 transcripts with IL1B and TNF was observed in both cold and hot tumors (Supplementary Fig. S5A and S5B).

In three fresh surgical specimens of ovarian and clear-cell renal carcinomas, we could detect heterogeneous but measurable quantities of TNFα, IL-1β, and IL-8 in the interstitial fluid or the malignant tissue at the protein level (Supplementary Fig. S6A). In similar samples from four cases, we sought to ascertain the identity of the stromal cells producing IL-1β and TNFα at the protein level by multicolor flow cytometry. In single-cell suspensions, T cells and myeloid populations were gated as exemplified in Supplementary Fig. S6B, and on such gated T cells, intracellular IL-1β and TNFα were analyzed. Supplementary Fig. S6C and S6D show the relative contributions of myeloid cells and T lymphocytes to the production of these cytokines, indicating heterogeneity of the contribution from each leukocyte compartment in each cancer case.

IL-8 produced by tumor cells under the influence of TNFα was functional, because it could mediate chemotaxis of human monocyte-derived macrophages and peripheral blood neutrophils (Supplementary Fig. S7A–S7C). Of note, chemotaxis could be blocked if the tumor cell cultures had been treated with infliximab or etanercept and if the resulting supernatant was blocked with the IL-8–neutralizing antibody BMS-986253 (ref. 43; Supplementary Fig. S7B and S7C).

IL-8 is also a well-known inducer of NET extrusion from granulocytes (7). In our hands, the supernatants of tumor cells stimulated with TNFα were able to induce NETosis on human neutrophils as recombinant IL-8 does. If the cultures of tumor cells with TNFα were set up in the presence of etanercept, infliximab, or the IL-8–neutralizing antibody BMS-986253, such induction of NETosis was almost completely abolished (Supplementary Fig. S7D).

Given the fact that TNFα released by T cells undergoing activation can enhance IL-8 production by tumor cells, we sought to investigate if T cells stimulated as a result of cancer immunotherapy would trigger in vivo the production of the IL-8 ortholog murine chemokines CXCL1 and CXCL2. For this purpose, mice bearing B16OVA melanomas (Fig. 3A) were treated with anti–PD-1 + anti–CTLA-4 checkpoint inhibitors with or without cotreatment with etanercept to neutralize TNFα. In such mice, there was an increase of CXCL1 and CXCL2 in their serum and correspondingly increased levels of Cxcl1 and Cxcl2 mRNA expression in their tumor nodules (Fig. 3B).

CXCL1 and CXCL2 could be counterproductive for immu­notherapy and to investigate such a possibility, experiments modeling treatment of established B16OVA melanomas with anti–PD-1 + anti–CTLA-4 mAbs were performed (Fig. 3C). In this experimental setting, we studied whether pharmacologic inhibition of CXCR1 and CXCR2 would result in improved treatment outcomes (Fig. 3C). In this model, the combined treatment regimen with checkpoint inhibitors only attains partial control of tumor progression. However, adding the CXCR1/2 inhibitor reparixin to this treatment regimen resulted in further delays of tumor progression and complete rejections in two out of six mice (Fig. 3D and E). These results provide evidence supporting that CXCR1/2 agonist chemokines induced by T cell–derived TNFα in the tumor microenvironment as a result of treatment with checkpoint inhibitors play a negative role in therapeutic outcome. Importantly, such a deleterious mechanism is druggable to increase efficacy.

To confirm and extend such data suggestive of opportunities in combinatorial immunotherapies (44), we repeated such experiments using another inhibitor of CXCR2 named AZ12376429. Combinatorial regimens shown in Supplementary Fig. S8A showed that both reparixin and AZ12376429 ameliorated the control of B16OVA tumors as achieved by the dual checkpoint inhibition with anti–CTLA-4 + anti–PD-1 mAbs (Supplementary Fig S8B and S8C). Moreover, a combination of etanercept and anakinra ameliorated control to a similar extent and added marginal additional benefit to that achieved by the CXCR pharmacologic inhibitors in multiple combination regimens, indicating that a major route of benefit upon IL-1β and TNFα blockade might be exerted via the CXCR axis, at least in this syngeneic mouse melanoma model.

To more closely reproduce the conditions of tumors in which TNFα and IL-1β elicit IL-8, we used fresh surgical explants from patients with resectable clear-cell renal carcinoma and endometrial carcinoma. The tissue was minced to fragments 1 to 2 mm³ in volume that were set up in 72-hour cultures under the influence of the added stimuli (Fig. 4A).

As Fig. 4B shows, TNFα was able to induce IL-8 secretion to the tissue culture supernatant and its mRNA transcription. The addition of etanercept to the cultures neutralized the induction of IL-8. In two additional explants of renal cell carcinoma and endometrial cancer, available tissue was sufficient to stimulate with both TNFα and IL-1β. In these fresh tumor tissue cultures, both TNFα or IL-1β induced IL-8 secretion from the cultured tumor fragments (Fig. 4C). In this setting, infliximab and etanercept neutralized IL-8 production as elicited by TNFα, while anakinra blunted IL-1β–induced IL-8 production.

To further confirm the induction of IL-8 by TNFα in more closely physiologic conditions, we challenged cultures of primary organoids derived from two cases of colon cancer. Figure 4D shows that the TNFα induced secretion and transcription of IL-8 from such three-dimensional tissue cultures in a manner that was also neutralized by etanercept and in­fliximab. IL-1β did not induce IL-8 in these cases, a finding that is probably explained by negligible levels of IL1R2 transcript expression by the organoids.

Lewis lung carcinoma cells were subcutaneously engrafted to form tumors in syngeneic mice in such a manner that the resulting tumor nodules could be repeatedly injected with TNFα delivered inside the tumor lesions. A group of mice could be intraperitoneally dosed with etanercept to systemically block TNFα (Fig. 5A). Experiments in Fig. 5B show that mRNA encoding both Cxcl1 and Cxcl2 was readily induced by intratumoral TNFα injections as compared with the cases intratumorally injected with saline buffer. Cotreatments with etanercept completely abolished the transcription of these chemokines.

Excised tumors were also stained by tissue immunofluorescence for Ly6G, and results in Fig. 5C indicate a clearly more prominent neutrophil infiltration. Figure 5D shows that more percentage area was occupied by NETs stained as citrullinated histone-3⁺ (H3-CIT⁺) structures. Both TNFα effects were blocked by etanercept treatments. Representative immunofluorescence images are provided in Fig. 5E. In line with this, IL8 mRNA expression in TCGA analyzed in a pan-cancer fashion correlates with gene signatures denoting the abundance of neutrophils and macrophages (Fig. 5F).

Rag2^−/− IL2Rγc^−/− mice were xenografted with HT29 colon cancer cells to form subcutaneous tumor nodules. As depicted in the scheme in Fig. 6A, repeated intratumoral injection of TNFα inside the nodules was feasible since day +7. Some of the mice also received two intraperitoneal doses of etanercept at the indicated time points to entrap and neutralize TNFα.

Tumors surgically harvested on day +10 were minced, and the resulting fragments were plated in 72-hour cultures. As can be seen in Fig. 6B and C, tumor fragments that had been intratumorally injected with TNFα more readily produced IL-8 to the supernatant, and an increase of IL8 mRNA transcription was observed. Etanercept in vivo treatments blocked such an effect. Furthermore, in the serum of those mice, concentrations of IL-8 were higher, and this increase was not observed if TNFα had been neutralized by etanercept cotreatment (Fig. 6D).

A similar scheme (Fig. 6E) was followed to perform experiments with intratumoral injections of recombinant IL-1β. In this case, IL-8 secretion from cultured explanted tumor tissue fragments was increased and partially reduced by intraperitoneal treatments of the mice with anakinra (Fig. 6F). This finding was also substantiated at the transcriptional level by measuring IL8 mRNA in excised tumor fragments (Fig. 6G).

We also considered whether there could be a cross-regulation of IL-1β and TNFα in these experiments at the transcriptional levels using the mRNA samples from Fig. 6A–G. As shown in Supplementary Fig. S9A and S9B, intratumoral injections of either TNFα or IL-1β were unable to cross-regulate the expression of each other using quantitative RT-PCR with specific primers for the human (Supplementary Fig S9A) or mouse (Supplementary Fig S9B) versions of these genes. Of note, TNFα intratumoral injection in the xenografted mice was able to upregulate the transcription of itself from the mouse and human components of the xenografted tumors, while IL-1β induced the transcription of itself only from cells of mouse origin (Supplementary Fig S9B).

Taken together, these data support that circumstances leading to TNFα and/or IL-1β expression in the tumors result in enhanced IL-8 production and upregulation of its local bioactivities.

To model immunotherapy of patients with checkpoint inhibitors as closely as possible, we humanized the subcutaneously HT29-xenografted mice with human peripheral blood mononuclear cells from healthy donors (Fig. 6H). Such mice were treated with the clinically approved combination of checkpoint inhibitors nivolumab and ipilimumab with or without etanercept cotreatments to neutralize TNFα. Tumors retrieved on day 14 showed clearly increased concentrations of IL-8 and showed higher content of IL8-encoding transcripts (Fig. 6I). If deleterious effects of IL-8 on checkpoint immunotherapy are taken into account, the TNF-mediated induction of IL-8 is to be considered a harmful albeit druggable counteracting mechanism.

In clinical trials performed almost two decades ago (45), patients with advanced cancers from a variety of tissue origins were recruited to a phase I clinical trial of infliximab monotherapy. We were able to retrieve an eight-case series of well-preserved plasma samples from one of these patient cohorts (Fig. 7A), drawn immediately prior to and 1 day after infliximab infusion and subsequently stored at −80°C. In this exploratory cohort, we could detect a reduction of circulating IL-8 concentrations in six of eight (75%) cases, while there were increases in the other two cases (Fig. 7B). Given this result, we then assayed a second cohort of paired plasma samples from 16 patients (Fig. 7A) that showed reductions of circulating IL-8 in 10 of the 16 cases analyzed as a validation cohort (Fig. 7C). In Fig. 7D, we provided the actual measured concentrations in the plasma samples showing statistical significance.

These results are very suggestive of a role for TNFα at regulating circulating IL-8 in human cancer patients as shown in plasma samples taken shortly after blockade of TNFα with infliximab before other confounding factors may occur and offer a proof of concept of the actionable nature of this cytokine axis in patients.

Protumor inflammation is one of the hallmarks of cancer (46). This protumor inflammation is sustained by cytokines such as IL-1β, TNFα, and IL-6 (47). This occurs in the context of the intratumoral presence of myeloid leukocytes such as macrophages and granulocytes (48, 49). Interestingly, intratumoral myeloid leukocytes, which are known to abundantly produce those cytokines, usually correlate with poor prognosis across solid malignancies (16, 50). Regarding T-cell immunity against tumors, it is already known that this is negatively affected by the presence of such inflammatory mediators (51–53), as well as by the infiltration of neutrophils (54) and tumor-associated macrophages (55, 56)—collectively referred to as myeloid-derived suppressor cells (57).

TNFα and IL-1β have in common their ability to turn on NF-κB transcription factors in cells expressing receptors for these cytokines, such as on malignant cells in many instances. Interestingly, NF-κB induces IL-8 (58), which is an ELR⁺ CXCL chemokine, which in turn can readily chemoattract macrophages and neutrophils (10). Previous reports had shown that IL-8 is induced by TNFα and IL-1β (24) although not by IL-6, at least in our hands (Supplementary Fig. S10A and S10B). Therefore, in tumors, a vicious circle may be in operation in which IL-1β– and TNFα-producing myeloid or lymphoid leukocytes induce IL-8 expression from the transformed cells. The IL-8 chemokine, in turn, may mediate further increases in the myeloid infiltrate of the malignant tissue microenvironment, creating a protumor positive feedback mechanism (8).

In this study, we provide multiple lines of evidence that TNFα and IL-1β are able to elicit functional IL-8 from malignant cells of multiple tissue origins. Our experiments tried to mimic as closely as possible the natural tumor microenvironment using short-term cultured fresh human tumor explants and primary organoids. Moreover, xenografted tumors were also inducible to produce IL-8 in response to TNFα or to IL-1β. The detectable concomitant presence of these three cytokines at the protein level in the tumor interstitial fluid was substantiated in three fresh surgical human specimens. At the bulk transcriptional level, we observe in the TCGA analyzed in a pan-cancer fashion clear correlations of the levels of IL1B and TNF transcripts with those encoding IL8. In colorectal cancer TCGA cases, such an association of IL8 with IL1B had been reported before (59), and it had also been published that there is a correlation of IL8 transcripts with the expression of IL1B and TNF by myeloid leukocytes in renal and urothelial cancer (13).

IL-8 is known to modulate protumor functions and has been associated with a lack of survival benefit in patients treated with checkpoint inhibitors (12, 13). These correlative data and the IL-8 induction results have given rise to the notion that such undesirable inflammation should be suppressed (14). In line with this, a number of approved drugs exist that can neutralize or entrap TNFα (infliximab, adalimumab, etanercept, and certolizumab; ref. 39), IL-1β (canakinumab, anakinra; ref. 60), and IL-8 (HumaxIL8 or BMS-986253; refs. 43, 61). Moreover, CXCR1 and CXCR2 can also be pharmacologically inhibited (62). In our hands, blockade of TNFα or IL-1β was able to attenuate IL-8 or its functional orthologs in mice. These pharmacologic interventions should reduce neutrophils and macrophages in the tumor microenvironment, as we have observed in LLC-derived syngeneic mouse tumors intratumorally injected with TNFα. Moreover, in human cancer, a clear association between IL8 mRNA and gene signatures denoting the presence of neutrophils and macrophages in the tumor microenvironment has been found. Thus, feasible transient cytokine blockades may break these vicious cytokine circles.

The implications of our work extend to cancer immunotherapy, because it might be interesting to act, perhaps transiently, on TNFα and IL-1β to blunt their function in a reversible manner. These sorts of treatments are prescribed routinely to patients with autoimmune or inflammatory conditions (39, 60). Some tumor-inhibitory effects of TNFα or IL-1β blockade have been reported in mice and humans (63, 64). Currently, these concepts have been revived in combinations of cytokine blockers with checkpoint inhibitors (NCT03400332; ref. 35). In turn, IL-8 and other ELR⁺ CXCL chemokines can be also inhibited in their functions by neutralizing the chemokines with antibodies or pharmacologically inhibiting the functions of the CXCR1 and CXCR2 receptors.

Reassuringly, from the perspective of translation to a clinical setting, canakinumab blockade of IL-1β is associated with decreased incidence and mortality from cancer (28), and patients with rheumatoid arthitis under treatment with TNFα blockade seem not to increase their risk of cancer incidence or progression (65) and even perhaps reduce it according to large registries in the United Kingdom (66, 67).

We found in our study that the TNFα produced by activated T cells also turns on this immunosuppressive three-pronged cytokine loop. Therefore, upon partially successful immunotherapy, there will be more TNFα production in the tumor from activated T lymphocytes. This mechanism can be counterproductive and worth inhibiting because TNFα is not involved in the antitumor immune response as previously shown in mouse models (30, 31). To support this notion of the T-cell response under immunotherapy raising the levels of TNF transcription in the tumor, a recent report showed in paired biopsies of melanoma that clinically responding patients, but not progressing patients, increased tumor TNF transcripts (68). In our hands, melanoma-bearing mice treated with checkpoint inhibitors turn on the IL-8 orthologs CXCL1 and CXCL2 in their tumor tissue. Such an effect seems to be deleterious for the immunotherapy outcome because CXCR1/2 pharmacologic inhibition gave rise to better therapeutic results. Furthermore, in humanized mouse models bearing human tumors, which were reconstituted with human peripheral blood lymphocytes, a similar mechanism could be recapitulated following treatment with checkpoint inhibitors. These findings alert us to the possibility that TNFα produced by T cells undergoing activation in patients as a result of checkpoint inhibition would turn on a counterproductive IL-8–CXCR1/2 axis. This deleterious axis of cytokines is druggable because blocking agents for each of the components are clinically available or under clinical trial development.

In fact, the TNFα-mediated impairment of cancer immunotherapy would involve not only IL-8–mediated reshaping of the myeloid immune microenvironment but also promotion of activation-induced cell death in antitumor T lymphocytes (31, 69). Moreover, TNFα seems to play a role in the pathogenesis of immune-related adverse events elicited by checkpoint inhibitors (31). Therefore, with regard to safety, neutralization of these cytokines is likely to be beneficial in a prophylactic manner as well (31).

Indeed, in a clinical trial blocking TNFα with single-agent infliximab that had taken place two decades ago, such circulating IL-8 reductions were observed (45). For this work, we were able to retrieve paired plasma samples pre– and post–infliximab treatment onset from 24 treated patients in that clinical trial. Such samples showed an early reduction of IL-8 levels in the majority of such cases, indicating that at least the TNFα–IL-8 axis seems to be functional in patients with cancer.

All things considered, immunotherapy strategies are at hand involving the repurposing of agents blocking IL-1β, TNFα, IL-8, or CXCR1/2, which may be deployed in combination with checkpoint inhibitors and other cancer immunotherapies with which they might synergize as a result of interfering with protumor pathogenic inflammatory loops (63).

Transcriptomic data from human tissues were used to assess the correlation between IL8, TNF, and IL1B. Gene expression data from a set of 98 tumor–paired normal tissues from patients with colorectal cancer and 50 mucosa from healthy donors were used (Colonomics project: https://www.colonomics.org; NCBI BioProject PRJNA188519; ref. 70). In addition, transcriptomic data from six different primary tumors in the TCGA project were downloaded from the cBioPortal repository and log₂ transformed. A total of 592 colorectal adenocarcinoma, 510 lung adenocarcinoma, 300 ovarian serous cystadenocarcinoma, 515 head and neck squamous cell carcinoma, 1,082 breast invasive carcinoma, and 177 pancreatic adenocarcinoma samples were used.

In each data set, Pearson correlation was calculated to assess if there is a correlation in expression between IL8–TNF and IL8–IL1B. Linear regression models were adjusted to calculate R-squared and P values. To calculate a pan-cancer correlation using the 3,176 samples from TCGA, an adjustment for reduction of the batch effect was performed with the ComBat function from the R package sva.

Gene expression data were also used to quantify immune cell infiltration in TCGA samples. The R package MCPcounter was used. MCPcounter (Microenvironment Cell Populations-counter) is a method for quantifying the relative abundance of immune cells in heterogeneous tissues using marker genes optimized for interrogating transcriptomic data. Eight different cell types were plotted to assess the correlation between the immune cell infiltration score and IL8 expression (neutrophil, macrophage/monocyte, T cell, CD8⁺ T cell, NK cell, B cell, myeloid dendritic cell, cancer-associated fibroblast). Also, Pearson correlation was calculated in each data set and pan-cancer.

Finally, samples were scored using genes in the T cell–inflammatory (TIS) signature (71) with the Gene Set Variation Analysis from R package GSVA. This score was used to select tumors with extreme phenotypes (<Q1 and >Q3), classifying them into “T-cell infiltrated” (n = 807) and “non–T-cell-infiltrated” (n = 795) categories. Then, a correlation between IL8/TNF and IL8/IL1B was calculated in each group of samples.

The HT29 and CAKI-1 cell lines were obtained from ATCC in 2011 and retrieved from the verified master cell bank. The UMBY human melanoma cell line was collected at the Erlangen clinical facility from primary surgical samples that were kept in early-culture passages in RPMI complete media. The H2009 cell line was a kind gift from Dr. Karmele Valencia (CIMA, Pamplona, Spain). All human cell lines were cultured in RPMI (Gibco) complete media (10% FBS Sigma, 1% penicillin/streptomycin, Gibco) with the exception of SKRB3, which was cultured in DMEM-F12 (Gibco) with 10% FBS (Sigma) and 1% penicillin/streptomycin (Gibco). The MC38 cells were a kind gift from Dr. Karl E. Hellström (University of Washington, Seattle, WA) in September 1998. B16OVA cells were provided by Dr. Lieping Chen (Yale University, New Haven, CT) in November 2001. LLC cells were a kind gift from Dr. Delia Nelson (Curtin University, Perth, Australia). The KPC cell line was provided by Dr. Silve Vicent (CIMA, Pamplona, Spain) in 2019. Mouse tumor cell lines were cultured in complete media containing RPMI 1640 medium (Gibco) supplemented with 10% FBS (Sigma-Aldrich), 100 IU/mL penicillin and 100 μg/mL streptomycin (Gibco) and 5 × 10⁻⁵ mol/L 2-mercaptoethanol (Gibco). B16OVA cells were also supplemented with 400 μg/mL geneticin (Gibco). All cell lines were grown in a humidified incubator with 5% CO₂ at 37°C for at least 7 days before inoculation into mice. All cell lines were routinely tested for Mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit (Lonza).

For in vitro induction of IL-8, 100,000 tumor cells were added to 48 multiwell plates. After 72 hours with TNFα (100 or 1,000 U/mL; PeproTech) or IL-1β (50 ng/mL; PeproTech) as stimuli and different TNFα inhibitors (4 μg/mL etanercept or infliximab) or an IL-1β blocker (4 μg/mL anakinra), supernatants and cells were collected to check IL-8 secretion using a human ELISA kit (BD) and IL8 transcription by qRT-PCR. The same scheme was used for CXCL1 and CXCL2 induction experiments in mouse tumor cell lines. CXCL1 and CXCL2 secretion was analyzed using the CXCL1 ELISA kit (R&D Systems) and CXCL2 ELISA kit (R&D Systems), while Cxcl1 and Cxcl2 mRNA transcription was assessed by qRT-PCR.

Lymphocytes, neutrophils, and monocytes were obtained from the peripheral blood of a blood donor database mainly composed of graduate students (male and female, young adults) of the Universidad de Navarra (Pamplona, Spain) following written, signed, and dated informed consent according to a protocol approved by the institutional ethics committee.

Human total lymphocytes as well as CD4⁺ and CD8⁺ T lymphocytes were isolated from total blood by Ficoll gradients following a negative magnetic selection with a Pan T-cell, CD4⁺ T-cell, or CD8⁺ T-cell isolation kit (Miltenyi Biotec), respectively. All T cells were activated separately by plate-bound anti-CD3 (OKT3, 1 μg/mL, BioLegend) and 5 μg/mL of soluble anti-CD28 (clone CD28.2, BioLegend). After 48 hours, supernatants were recovered. Before IL-8 induction experiments, we checked TNFα concentration in those supernatants using the human TNFα ELISA kit (BD) to adjust the final concentration of TNFα to 40 U/mL.

For neutrophil enrichment, peripheral blood samples drawn in heparin-containing tubes were subjected to Ficoll separation. Erythrocytes and neutrophil-containing pellets were then mixed with one volume of cold PBS and an extra volume of 6% dextran/0.9% NaCl solution. After being inverted 18 to 20 times to ensure adequate mixing, the mixture was placed upright at room temperature for 1 hour until phase separation was completed. The yellowish supernatant was recovered into a 50-mL tube and spun at 1,200 rpm for 12 minutes at 4°C. Then, the supernatant was discarded, and the pellet was resuspended in 5-mL ACK buffer for 5 minutes at room temperature. After incubation, the cells were washed with 45 mL of cold PBS. The supernatant was discarded, and the pellet was suspended in PBS. To ensure the correct purification, the pellet was immunostained with CD15-PE. Neutrophil purity was >95% (CD15 bright neutrophils).

Human monocytes were purified from total peripheral blood mononuclear cells by positive magnetic selection with a CD14⁺ isolation kit (Miltenyi). To generate GM-CSF–like macrophages, monocytes were cultured for 6 days in the presence of 50 ng/mL GM-CSF before being stimulated with 1 ng/mL LPS for 24 hours.

For migration assays, 150,000 neutrophils or macrophages were added to the top of 0.5-μm pore polycarbonate transwell inserts (5 μm for neutrophils or 8 μm/L for macrophages; Corning Costar) in 100 μL of RPMI-free media. The lower compartment was then filled with 300 μL of HT29 tumor culture supernatants after 72 hours with 1,000 U/mL TNFα and some of them with TNFα inhibitors (4 μg/mL etanercept or infliximab) or the IL-8 blocker BMS-986253 (10 μg/mL) in RPMI serum-free media. Migration was allowed for 4 hours, and then neutrophil migration was assessed by trypan blue.

Neutrophils were resuspended in HBSS (Gibco). Two hundred thousand cells were then placed in 150 μL of HBSS in 8-well uncoated IBIDI microscopy chambers (IBIDI). Supernatants from HT29 cells, after a 72-hour culture with 1,000 U/mL TNFα and some conditions with the TNFα inhibitors etanercept (4 μg/mL), infliximab (4 μg/mL), or the IL-8 blocker BMS-986253 (10 μg/mL), were added on top at the indicated concentrations.

Cells were incubated for 4 hours at 37°C and 5% CO₂, Sytox Green (25 nmol/L, Thermo Fisher) was added, and the mix was incubated for 5 minutes. Cells were PFA fixed by adding 50 μL of 16% PFA (Thermo) and kept at 4°C until confocal microscopy was performed. Triplicate wells of each condition were used.

Confocal microscopy was performed to quantify NET formation. Z-stacks (10–30 μm, 40× magnification) were taken in an LSM800 equipped with a 488 diode and a plan-apochromat 1,3 N/A Oil DIC III objective. For NET area quantification, FIJI software and the particle analyzer plugin were used. Only structures depicting NET morphology and positive for Sytox green were selected for area quantification, and intact granulocyte nuclei were excluded from the analysis.

All mouse procedures were approved by the ethics committee for animal experimentation of the regional government of Navarra under Spanish regulations (study 078-21). Mice were housed at the animal facility of CIMA (Pamplona, Spain). Female, 5- to 6-week-old C57BL/6 mice were purchased from Envigo (Barcelona, Spain). Rag2^−/−IL2Rγc^−/− mice were bred in our facilities (CIMA, Pamplona, Spain). Littermates of the same age (7 to 10 weeks old) female mice were randomly assigned to experimental groups.

For CXCL1 and CXCL2 in vivo induction by TNFα, C57BL/6 mice were injected subcutaneously with 0.5 × 10⁶ LLC cells. After 6 days, tumors were injected intratumorally with PBS or 50 ng TNFα (PeproTech) on days 7, 8, and 9. Some mice were injected intraperitoneally with 3 mg/kg etanercept on days 6 and 8. To study IL-8 induction by TNFα, the same scheme of treatment was used in Rag2^−/−IL2Rγc^−/− mice xenografted with 2 × 10⁶ HT29 human colon cells. On day 10, tumors and serum were collected to further analyze IL8 secretion and transcription.

For in vivo IL-8 induction in response to IL-1β, Rag2^−/−IL2Rγc^−/− mice were injected subcutaneously with 2 × 10⁶ HT29 human colon cells. On days 7, 8, and 9, tumors were injected intratumorally with 100 ng of IL-1β, and some animals also received anakinra (100 mg/kg) intraperitoneally from days 6 to 9. As in previous experiments, tumors were collected on day 10 to assess IL-8 secretion after ex vivo cultures.

For CXCL1/2 induction in response to checkpoint blockade agents, C57BL/6 mice were injected subcutaneously with 0.5 × 10⁶ B16OVA cells. Mice were treated with 100 μg anti–PD-1 + 100 μg anti–CTLA-4 as indicated in figure scheme. Depending on the experiment, one group of mice also received 3 mg/kg etanercept intraperitoneally, 200 μg reparixin intraperitoneally daily between days 9 and 22, or 1 mg/kg AZ12376429 by oral gavage.

Humanized mice were created by intravenous injection of 10⁷ human peripheral blood mononuclear cells on Rag2^−/−IL2Rγc^−/− mice. Those mice were xenografted with a human HT29 colon cancer cell line and received intraperitoneally 100 μg ipilimumab + 100 μg nivolumab regimen with or without 3 mg/kg etanercept as indicated on the figure schemes.

Organoids were generated from specimens from four patients with colorectal cancer who underwent surgery in the Clínica Universidad de Navarra. Written informed consent was obtained from all the patients, and all procedures were approved by the Universidad de Navarra ethics committee (Study 2019.96). These were isolated and frozen for later experiments.

For the isolation and establishment of primary tumor organoids, tumors were minced and incubated in Advanced DMEM-F12 (Thermo), 5 mg/mL Collagenase II, 1.25 mg/mL dispase, 2.5% FBS, 100× HEPES buffer (Thermo), 100× glutamax (Thermo), and 100 ng/mL primocin (Invivogen) on a rocker (Miltenyi) at 37°C for 2 to 3 hours. The sample was collected in Advanced DMEM-F12, FBS, 100× HEPES buffer, 100× glutamax, and 100 ng/mL primocin and passed through a cell strainer and centrifuged at 850 rpm for 7 minutes at 4°C. Then 10 μL of 10 mg/mL DNase I were added, and the mixture was incubated at 37°C for 5 minutes. Finally, the sample was resuspended in 100% Matrigel (Corning) and plated in a 24-well plate prewarmed and placed in the incubator at 37°C for 30 minutes. When Matrigel solidified, 500 μL of Complete Feeding Media [Advanced DMEM-F12, 100× HEPES buffer, 100× glutamax, 100 ng primocin, 20% R-Spondin conditioned medium, 500 nmol/L A83-01 (Sigma), 50 ng/mL hEGF (PeproTech), 100 ng/mL mNoggin (PeproTech), 100 ng/mL hFGF10 (PeproTech), 10 nmol/L Gastrin I (TOCRIS), 3 μmol/L SB202190 (Stem Cell), 10 nmol/L prostaglandin E2 (TOCRIS), 1.25 mmol/L N-acetylcysteine (Sigma), 10 mmol/L nicotinammide (Sigma), and 1× B-27 supplement (Thermo Fisher)] supplemented with 10.5 mmol/L Y-27632 (rho kinase inhibitor) were added on top of the Matrigel, and the 24-well plate was placed in the incubator.

Organoids were passaged approximately every week. To split the primary tumor organoids, droplets were broken by adding 1 mL of ice-cold splitting media (Advanced DMEM-F12, 100× HEPES buffer, 100× glutamax, and 100 ng/mL primocin) and pipetting up and down. Organoids were collected in a tube, centrifuged at 1,200 rpm for 5 to 10 minutes at 4°C, and the medium was removed. Organoids were broken up with a fire-polished pipette and centrifuged at 1,200 rpm for 5 to 10 minutes at 4°C. The pellet was mixed with Matrigel and plated in 50-μL droplets in 24-well plates. After solidification, 500 μl Complete Feeding Media (Advanced DMEM-F12, 100× HEPES buffer, 100× glutamax, 125 μg/mL primocin, 20% R-Spondin conditioned medium, 500 nmol/L A83-01, 50 ng/mL hEGF, 100 ng/mL mNoggin, 100 ng/mL hFGF 10, 10 nmol/L gastrin I, 3 μmol/L SB202190, 10 nmol/L prostaglandin E2, 1.25 mmol/L N-acetylcysteine, 10 mmol/L nicotinamide, and 1× B-27 supplement) were added.

Organoids were cryopreserved in 10% recovery cell culture freezing medium (Thermo Fisher) as master and working biobanks. Organoids < passage 35 were used in experiments.

For the IL-8 induction experiments, tumor spheroids were plated in 48 multiwell plates for 72 hours with 1,000 U/mL of rTNFα. TNFα inhibition was performed by etanercept (4 μg/mL) or infliximab (4 μg/mL). Each condition was made in duplicate. At the end of the experiments, supernatants and spheroids were collected separately to analyze IL-8 secretion using a human IL-8 ELISA kit (BD) and IL8 mRNA transcription by qRT-PCR.

Tumor samples were obtained from patients with renal or endometrial cancer who required standard-of-care surgery. Samples were collected between December 2020 and February 2021 at the Clinica Universidad de Navarra. The protocol was approved by the local Clinical Research Committee (Protocol ID: INTRON 2017, version 3.2). Written informed consent was received from all patients. A pathologist evaluated surgical samples and selected a fragment for research use. Tumors were collected in DMEM high glucose with glutamax (Life Technologies) supplemented with sodium pyruvate 1 mmol/L (Life Technologies), 1× MEM nonessential AA (Life Technologies), 2 mmol/L glutamine (Life Technologies), 1% penicillin–streptomycin (Life Technologies), and 10% FBS (Sigma) and manually cut in 1 × 1 mm³ pieces on ice. These patient-derived tumor fragments (PDTF) were immediately placed in 48-well culture plates with 500 μL of culture medium, with a total of four pieces per well. For IL-8 induction, 100 U/mL TNFα or 50 ng/mL IL-1β were added to the cultures. Additionally, etanercept (4 μg/mL) or infliximab (4 μg/mL) was used to neutralize TNFα and anakinra (4 μg/mL) to block IL-1β and/or its combinations and kept at 37°C for 48 hours. As a control, only culture medium was used. Triplicate wells of each condition were used. After culture time, supernatants and fragments were collected to analyze the production of IL-8 by ELISA (BD) and IL8 mRNA transcription by qRT-PCR, respectively.

For flow cytometry analysis, PDTF cultures were also supplemented with TMI-1 at 1 μmol/L (Sigma) and placed at 37°C for 8 to 12 hours. A mix of LPS (Sigma) at a final concentration of 10 ng/mL, PMA (Sigma) at 0.1 μg/mL, and ionomycin (Sigma) at 1 μg/mL was used as a positive control. After overnight culture, samples were incubated for 4 hours at 37°C with BD Golgi-Stop Protein Transporter Inhibitor and analyzed by flow cytometry. For analysis of immune infiltrates, we pooled the 16 to 20 PDTFs for each experimental condition and protease-digested them in order to obtain a single-cell suspension. For this purpose, once the supernatant was removed, we added 250 μL of collagenase type D and DNase I for 30 minutes at 37°C. PDTFs were then processed with a 60-μm cell strainer into a single-cell suspension. Cells were stained with Maleimide PromoFluor (Promocell), anti-human CD45-PB (BioLegend), anti-human CD3-PE (BioLegend), anti-human CD4-BV650 (BioLegend), anti-human CD8-BUV395 (BD Bioscience), anti-human CD11b-PCPCy5.5 (BioLegend), anti-human CD16-BV605 (BioLegend), anti-human CD14-PECy7 (BioLegend), anti-human CD56-PEDazzle594 (BioLegend), anti-human CD68-BV785 (BioLegend), and anti-human HLA-DR-APC-Cy7 (BioLegend) antibodies. Subsequently, following a step of membrane permeabilization with cytofix/cytoperm (Becton Dickinson), anti-human TNFα-AF488 (BioLegend), anti-human IL-1β-AF647 (BioLegend) were used for intracellular staining, and samples were acquired with Cytoflex-LX (Beckman Coulter) and analyzed with CytExpert and FlowJo software. Results show the percentage of the cell population among the total of TNFα- or IL-1β–producing gated cells.

Total RNA was extracted from tumor cells, organoids, or fresh human tumor explants with the Maxwell RSC simplyRNA extraction kit (Promega) according to the manufacturer's instructions and subsequently retrotranscribed into cDNA using the M-MLV enzyme kit (Invitrogen). Real-time PCR reactions were performed in the Bio-Rad CFX qPCR system with customized primers for hIL8 (FW 5′-CCAGGAAGAAACCACCGGA and RV 5′-GAAATCAGGAGGCTGCCAAG), mCxcl1 (FW 5′-AAAGATGCTAAAAGGTGTCC and RV 5′-GTATAGTGTTGTCAGAAGCC), mCxcl2 (FW 5′GGGTTGACTTCAAGAACATC and RV 5′-CCTTGCCTTTGTTCAGTATC), hTNF (FW 5′-GACACCATGAGCACTGAAAGC and RV 5′-AGCTTGAGGGTTTGCTACAAC), and hIL1B (FW 5′-TGATGGCTTATTACAGTGGCAA and RV 5′-GTCCATGGCCACAACAACTG).

For immunofluorescence of the frozen sections, LLC tumors were frozen in Optimal Cutting Temperature Compound (Thermo Fisher Scientific) and cut into 7-μm sections. Slides were fixed in 4% PFA immediately after thawing and washed three times with TBS. Slides were then blocked and incubated overnight at 4°C with antibodies to Ly6G-PE (10 μg/mL, BioLegend) and cit-H3 (10 μg/mL, Abcam). Slides were then counterstained with DAPI (Thermo Fisher) and viewed under an LSM 800 microscope. Images were captured by using ZEN software (Zeiss). The total Ly6G area and total cit-H3 area were quantified with the particle analyzer tool of FIJI.

The plasma samples came from a phase I clinical study conducted between 2003 and 2006 that assessed the tolerability and biological activity of infliximab in patients with advanced cancer. Full details of the trial can be found in Brown and colleagues (45). Patients were eligible if they had a histologically confirmed diagnosis of advanced and/or metastatic solid cancer for which no satisfactory treatment exists or against which established treatments had failed; a World Health Organization performance status of 0 or 1; and life expectancy of at least 3 months.

The study was approved by North East London and the City Health Authority Research Ethics Committee (LREC P/02/150) and Lothian Research Ethics Committee (LREC/2002/8/31) and conducted according to the declaration of Helsinki. All patients gave voluntary written informed consent. Patients received infliximab at 5 mg/kg (n = 21) or 10 mg/kg (n = 20) intravenously at 0 and 2 weeks and then every 4 weeks. A 10-mL sample of whole blood was collected pretreatment in sterile preservative-free heparinized tubes (30 U/mL) before treatment, at 24 hours, and at further intervals during the treatment cycles. All aliquots were snap-frozen in liquid nitrogen and stored at −80°C. Samples from 25 of the 41 patients were available in sufficient quantity and quality for the current study.

For the exploratory cohort, plasma IL-8 was measured with a Meso Scale Discovery V-PLEX Proinflammatory Panel 1 Human Kit (#K15049D) following the manufacturer's protocol. For the validation cohort, plasma IL-8 levels were detected using an IL-8 human ELISA Kit (Thermo Fisher Scientific, #KHC0081).

One exploratory cohort patient sample was also analyzed using the ELISA kits to validate the V-Plex results.

Data were processed using GraphPad Prism 6.0. Means and standard errors of the mean or standard deviation are presented as averages and error bars as indicated in the figure legend.

Two-tailed Student t tests or two-way ANOVA tests were used if results fitted normality, or Mann–Whitney U tests were used to analyze statistical differences between independent groups. When differences are statistically significant, the significance is represented with asterisks: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

The data generated in this study are available within the article and its supplementary data files.

M.F. Sanmamed reports grants from Roche, and personal fees from MSD and Bristol Myers Squibb outside the submitted work. M. Alvarez reports other support from Highlight Therapeutics and PharmaMar outside the submitted work. F.R. Balkwill reports personal fees from iOmx Therapeutics and GSK outside the submitted work. I. Melero reports grants and personal fees from Bristol Myers Squibb, Roche, AstraZeneca, Bioncotech, Genmab, and Pharmamar, grants from Alligator, and personal fees from F-Star, Numab, Gossamer, Pieris, Merus, Amunix, ThirdRock, and Highlight Therapeutics outside the submitted work. No disclosures were reported by the other authors.

I. Olivera: Formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. R. Sanz-Pamplona: Software, formal analysis, investigation, writing–review and editing. E. Bolanos: Data curation, validation, investigation, writing–review and editing. I. Rodriguez: Resources, methodology, writing–review and editing. I. Etxeberria: Investigation, methodology, writing–review and editing. A. Cirella: Investigation, methodology. J. Egea: Validation, investigation, methodology. S. Garasa: Validation, visualization, methodology, writing–review and editing. I. Migueliz: Resources, methodology. I. Eguren-Santamaria: Data curation, investigation, methodology, writing–review and editing. M.F. Sanmamed: Conceptualization, validation, methodology, writing–review and editing. J. Glez-Vaz: Data curation, formal analysis, validation, investigation. A. Azpilikueta: Data curation, formal analysis, visualization, methodology. M. Alvarez: Formal analysis, validation, investigation, writing–review and editing. M.C. Ochoa: Investigation, visualization, methodology, writing–review and editing. B. Malacrida: Formal analysis, investigation, methodology, writing–review and editing. D. Propper: Methodology, writing–review and editing. C.E. de Andrea: Formal analysis, investigation, writing–review and editing. P. Berraondo: Data curation, formal analysis, visualization, writing–review and editing. F.R. Balkwill: Conceptualization, data curation, supervision, funding acquisition, methodology, writing–review and editing. A. Teijeira: Conceptualization, resources, data curation, software, formal analysis, visualization, methodology, writing–review and editing. I. Melero: Conceptualization, resources, data curation, supervision, funding acquisition, visualization, methodology, writing–original draft, project administration, writing–review and editing.

We are grateful to the excellent project management support by Belen Palencia and Esther Guirado as well as to Eneko Elizalde for animal facility management, to the Unit of Microscopy, and to the Unit of Cytometry. This project has been supported by MINECO SAF2017-83267-C2-1-R and PID2020-112892RB-100 supported by MCIN/AEI/10.13039/501100011033 and by the I-ON network supported by Bristol Myers Squibb to I. Melero. This project has received funding from the European Union's Horizon 2020 research and innovation program (grant agreement no. 635122-PROCROP), Fundación de la Asociación Española Contra el Cáncer (AECC) GCB15152947MELE and salary support for M. Alvarez, Fundación La Caixa and Fundación BBVA, Gobierno de Navarra Salud, Gobierno de Navarra Proyecto LINTERNA Ref: 0011-1411, the Mark Foundation, Fundación BBVA, and Fundación Olga Torres. A. Teijeira is supported by the Ramon y Cajal program from the Spanish Ministry of Science (RyC 2019-026406-I). E. Bolanos is supported by CRUK Programme Grant C587/25714.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.