Targeting VEGF-A in myeloid cells enhances natural killer cell responses to chemotherapy and ameliorates cachexia

Chemotherapy remains a mainstay of cancer treatment but its use is often limited by the development of adverse reactions. Severe loss of body weight (cachexia) is a frequent cause of death in cancer patients and is exacerbated by chemotherapy. We show that genetic inactivation of vascular endothelial growth factor (VEGF)-A in myeloid cells prevents chemotherapy-induced cachexia by inhibiting skeletal muscle loss and the lipolysis of white adipose tissue. It also improves clearance of senescent tumour cells by natural killer cells and inhibits tumour regrowth after chemotherapy. The effects depend on the chemoattractant chemerin, which is released by the tumour endothelium in response to chemotherapy. The findings define chemerin as a critical mediator of the immune response, as well as an important inhibitor of cancer cachexia. Targeting myeloid cell-derived VEGF signalling should impede the lipolysis and weight loss that is frequently associated with chemotherapy, thereby substantially improving the therapeutic outcome.

D espite its frequent side effects, chemotherapy generally represents the first course of treatment for cancer patients. The benefits of chemotherapeutic agents stem not only from direct effects on the tumour cell but also from influences on the tumour microenvironment, resulting in a robust immune response that can be crucial to the therapeutic outcome 1 . However, drug delivery poses a significant problem as the vasculature of tumours is inefficient 2 . In most tumours, despite high vascular density, the vasculature differs from normal vascular networks and is characterized by an inefficient blood supply. Vessel abnormalities include increased permeability and tortuosity, as well as decreased pericyte coverage, which frequently cause scarce delivery of chemotherapy to the tumour and tumour hypoxia as well. Therefore, strategies to reverse this phenotype and to 'normalize' the tumour vasculature have gained increasing interest 2 . Using mouse models, we have shown that specific deletion of vascular endothelial growth factor (VEGF) in tumour-infiltrating myeloid cells leads to normalized tumour blood vessels and increased tumour cell apoptosis 3 .
Cancer-induced cachexia is the immediate cause of death in B15% of cancer patients [4][5][6] . It is characterized by involuntary weight loss that is resistant to nutritional supplementation 7 . Weight loss starts with degradation of skeletal muscle and the breakdown of white adipose tissue (WAT) mediated by the lipolytic enzymes adipose triglyceride lipase (Atgl) and hormonesensitive lipase (Hsl) 8 . Cachexia is believed to be induced by tumour-derived factors, such as tumour necrosis factor-a (TNF-a) and interleukin (IL)-6 (refs 9,10). After an initial reduction of tumour mass, treatment with chemotherapeutic agents frequently exacerbates cachexia, hampering further treatment and increasing mortality 11,12 . There is an urgent need for treatment regimens that counter the development of cachexia and thus allow continued chemotherapy.
Chemerin was initially defined as an adipokine 13 but has received considerable interest as a chemoattractant for macrophages, dendritic cells and natural killer (NK) cells [14][15][16] . NK cells and cytotoxic T cells are particularly important in the immunosurveillance and suppression of tumours 17,18 , and chemerin has been shown to improve NK cell-based tumour surveillance. Expression of the chemerin gene ((Rarres (retinoic acid receptor responder) 2) is frequently downregulated in human solid tumours, including lung cancer and melanoma. Overexpression of chemerin in melanoma cells in mouse models results in increased NK cell recruitment and tumour suppression 19 .
We now show that chemerin is a pivotal regulator of the chemotherapy-elicited immune response, as well as of therapyassociated cachexia. We demonstrate further that endothelial release of chemerin on chemotherapy can be enhanced by targeting VEGF-A in myeloid cells, leading to improved chemotherapeutic success.

Targeting of VEGF-A in myeloid cells delays tumour growth.
We have previously crossed mice with a loxP-flanked Vegfa allele to mice with the Cre recombinase under the control of the lysozyme M promoter. The VEGF-A gene is specifically deleted in the myeloid cells of the resulting mutant (Mut, LysMCre/VEGF f / f ) mice and the animals' response to chemotherapy is improved: the mice show vascular normalization and an increase in tumour cell apoptosis 3 . We subjected wild-type (WT, LysMCre À /VEGF þ / þ ) and mutant mice carrying Lewis lung carcinomas (LLCs) or B16F10 (B16) melanomas to three cycles of cisplatin treatment (cis-diamminedichloridoplatinum(II) (cisplatin, CDDP), 8 mg per kg body weight, see scheme Fig. 1a). In LLC and B16 tumours, loss of VEGF-A in myeloid cells significantly increased tumour-doubling time (Fig. 1b for LLC and Fig. 1c for B16) and was associated with significantly reduced endpoint tumour volumes (Fig. 1d,f for LLC and Fig. 1e,g for B16). In contrast, WT tumours reached endpoint volumes comparable to those of untreated tumours (Fig. 1d,e), indicative of treatment failure. Ulcerations in the mice injected with B16 melanoma cells forced termination of the control experiment ahead of schedule (Fig. 1e).
Treatment with cytotoxic agents frequently exacerbates cachexia and limits the outcome of therapy 11,12 . Untreated LLC-and B16-bearing WT and Mut mice had similarly reduced body weights at endpoint (Fig. 1h,i, respectively). On chemotherapy with cisplatin, the loss of body weight in the LLC (Fig. 1h) but not in the B16 model (Fig. 1i) depended on the presence of myeloid VEGF-A. LLC-bearing WT mice showed a significant drop in body weight that was mitigated in Mut mice by deletion of myeloid cell-derived VEGF-A (Fig. 1h).
Deletion of myeloid-derived VEGF-A improves drug delivery. Irrespective of the genotype, cisplatin treatment reduced levels of VEGF-A, lowered vascular density and increased pericyte coverage to varying degrees (  Supplementary Fig. 1D for B16) and decreased tumour hypoxia (Fig. 2e,f for LLC and Supplementary  Fig. 1E for B16). Although vascular normalization and improved oxygenation is associated with accelerated tumour growth in untreated Mut tumours, it paradoxically results in delayed tumour outgrowth after chemotherapy (Fig. 1d,f for LLC and Fig. 1e,g for B16). The apparent contradiction arises from the augmented delivery of chemotherapeutic agent to the tumour, indicated by increased levels of cisplatin-DNA adducts (Fig. 2g,h for LLC and Supplementary Fig. 1F B16). Thus, normalizing the vasculature by targeting VEGF-A exerts opposing effects on tumour growth kinetics pre-and post chemotherapy.
Therapy-induced senescence promotes immune cell recruitment. We have previously reported increased tumour cell apoptosis after chemotherapy of tumour-bearing Mut mice 3 . Not only do chemotherapeutic agents drive the apoptosis of tumour cells, they may be associated with a wide range of other outcomes including so-called premature or therapy-induced senescence 21,22 . Untreated tumours show hardly any senescence irrespective of the genotype (Fig. 3a,b, untreated for LLC and Supplementary Fig. 2A for B16 untreated). Deleting VEGF-A in myeloid cells increases tumour cell death on chemotherapy 3 and promotes therapy-induced senescence, as shown by senescenceassociated b-galactosidase (SA b-Gal) staining (Fig. 3a,b, day 14, CDDP for LLC and Supplementary Fig. 2A, day 16, CDDP for B16). The changes presumably stem at least partially from improved drug delivery. Tumours from Mut mice show a decline in SA b-Gal-positive cells at the endpoint (day 18 for LLC and day 20 for B16), whereas the number of senescent cells increases further in WT tumours (Fig. 3a,b, day 18, CDDP for LLC and Supplementary Fig. 2A, day 20, CDDP for B16). It thus seems likely to be that targeting VEGF-A in myeloid cells improves the clearance of senescent cells after chemotherapy. Senescent cells remain viable and secrete a range of inflammatory cytokines 23 . The senescence-associated secretory phenotype is believed to trigger an immune response involving macrophages and NK cells that facilitates immune cell-mediated tumour clearance [24][25][26] . We used flow cytometry to analyse tumour infiltration by immune cells following chemotherapy. There were markedly higher numbers of NK cells (NKp46) and macrophages (F4/80) in the absence of myeloid cell-derived VEGF (Fig. 3c for LLC and Supplementary Fig. 2B for B16), whereas numbers of intratumoural T-helper cells (CD4) and cytotoxic T cells (CD8) were unaffected or only slightly altered ( Fig. 3c for LLC and Supplementary Fig. 2B for B16). Cisplatin  Chemerin is released by ECs on chemotherapy. A number of cytokines and chemokines are involved in the recruitment of immune cells to malignant tumours 27,28 . The chemoattractant chemerin has a crucial role in immune cell trafficking [14][15][16] . Forced expression of chemerin in tumour cells gives rise to an NK cell-based antitumour response and restricts tumour growth, whereas low levels of chemerin are associated with tumour progression and a poor outcome 19 . Increasing the level of chemerin within the tumour may thus represent a promising therapeutic approach. We found that chemotherapy increases the level of chemerin in tumours in WT mice (Fig. 4a,b for LLC and Supplementary Fig. 4A,B for B16). Surprisingly, the effect was significantly enhanced in Mut mice, showing that the absence of VEGF-A in myeloid cells stimulates the chemotherapy-evoked expression of chemerin ( Fig. 4a,b and Supplementary Fig. 4A,B) and, in the LLC but not B16 model, drastically increases the levels of circulating chemerin ( Fig. 4c for LLC and Supplementary  Fig. 4C). Of note, chemerin messenger RNA levels are more than tenfold higher in endothelial cells (ECs) isolated from cisplatin-treated Mut LLC tumours compared with ECs isolated from cisplatin-treated Mut B16 tumours ( Supplementary Fig. 4E). This suggests that only systemically elevated chemerin levels confer protection against chemotherapy-induced cachexia. So far, the physiological source of chemerin within the tumour microenvironment has not been determined. The tumour cells themselves do not release chemerin as a consequence of chemotherapy: treatment of LLC cells in vitro with 3 mg ml À 1 cisplatin, a concentration that causes a significant DNA damage response ( Supplementary Fig. 5A), did not trigger chemerin release ( Supplementary Fig. 5B). Similarly, cisplatin treatment of B16F10 cells produced no increase in the basal level of chemerin secreted ( Supplementary Fig. 5B). Consistently, immunohistochemical analysis of tumour sections revealed only subtle chemerin reactivity in untreated LLC tumours of WT and Mut mice, as well as in tumours from cisplatin-treated WT animals (Fig. 4d). However, tumours from Mut mice showed significant chemerin immunoreactivity of the tumour vasculature on chemotherapy (Fig. 4d,e). The result indicates that tumour ECs release chemerin in response to chemotherapy, and that VEGF-A from myeloid cells suppresses the release.
To test this hypothesis, we analysed the release of chemerin by the murine EC line bEnd3. Cisplatin treatment (3 mg ml À 1 ) (Fig. 4f) caused a pronounced induction of chemerin release, accompanied by the accumulation of the transcription factor peroxisome proliferator-activated receptor-g (PPAR-g) ( Supplementary Fig. 5C,D), which stimulates chemerin expression 29 . The addition of exogenous murine VEGF-A suppresses the effect ( Supplementary Fig. 5C,D) and blocks the increased production of chemerin (Fig. 4f). Comparable results were obtained in ECs isolated from tumours of both genotypes. Chemerin and PPAR-g showed increased expression only in ECs of tumours derived from Mut mice after chemotherapy ( Fig. 4g and Supplementary Fig. 5E for LLC and Supplementary Figs 5F and 4D for B16), confirming that ablation of myeloid cell-derived VEGF-A significantly increases the expression of chemerin in response to chemotherapy. Interestingly, exogenous addition of another angiogenic factor, basic fibroblast growth factor also inhibited cisplatin-induced chemerin release from bEnd3 cells ( Supplementary Fig. 5G).
Consistently, chemotherapy with another frequently used cytotoxic agent, etoposide, resulted in improved chemotherapeutic outcome in LLC-bearing Mut mice, increased chemerin expression in tumour ECs, enhanced NK cell recruitment and decreased senescent tumour cells (Supplementary Fig. 6A-D, respectively) but failed to increase systemic chemerin levels ( Supplementary  Fig. 6E), further supporting a link between systemic chemerin levels and protection against cachexia. It is noteworthy that in contrast to cisplatin, etoposide treatment caused only a very mild exacerbation of cachexia without genotype-specific differences ( Supplementary Fig. 6F).
Interestingly, treatment of LLC-bearing mice with a VEGFneutralizing antibody at a dose that induced vascular changes reminiscent of vascular normalization ( Supplementary Fig. 7A,B) improved treatment outcome on chemotherapy with cisplatin ( Supplementary Fig. 7C) endothelial chemerin expression and NK cell recruitment ( Supplementary Fig. 7D,E). Importantly, anti-VEGF treatment under these circumstances had no impact on levels of circulating chemerin or cisplatin-exacerbated cachexia (Supplementary Fig. 7F,G). The data verify the tumour endothelium as a source of chemerin in response to chemotherapy and show that chemerin production is suppressed by VEGF-A.
Chemerin prevents therapy-associated loss of body weight. Chemerin was initially described as an adipokine with contextdependent pro-and antilipolytic effects on WAT 13,30 , as well as a regulator of skeletal muscle homeostasis 31,32 . It seemed conceivable that changes in chemerin levels could affect lipid and muscle metabolism in our LLC-bearing Mut mice, which    were protected against chemotherapy-exacerbated chemotherapy. It was initially important to assess the contribution of loss of adipose tissue and skeletal muscle to the overall weight loss associated with chemotherapy. We thus weighed gastrocnemius muscles and gonadal adipose depots in LLC-bearing mice subjected to chemotherapy. Consistent with the concept that ARTICLE cachexia involves breakdown of skeletal muscle and WAT, chemotherapy of WT mice resulted in a 430% loss of gastrocnemius weight along with a reduction in muscle fibre size (Fig. 5a,b) and in a 460% reduction in gonadal WAT (Fig. 5c) along with overall loss of body weight (Fig. 5d). The expression of the major lipolytic enzymes Atgl and Hsl in WAT isolated from cisplatin-treated WT mice was significantly upregulated (Fig. 5e,f). The increase in lipolytic enzymes and muscle degradation depended on the presence of myeloid-derived VEGF-A: chemotherapy of Mut mice caused a far smaller loss of gastrocnemius weight and WAT (Fig. 5a,c). The data suggest that differences in chemerin release underlie not only the altered tumour immune cell infiltration but also the striking difference in weight loss between WT and Mut mice following chemotherapy. To test this interpretation, we depleted chemerin by means of an anti-chemerin antibody. Remarkably, the antibody caused Mut mice to suffer the same loss of body weight (Fig. 5d), skeletal muscle (Fig. 5a,b) and WAT (Fig. 5c) as WT mice on cisplatin treatment. Furthermore, following chemotherapy the Atgl and Hsl genes were expressed at similar levels in WT mice and in Mut mice treated with the antibody (Fig. 5e,f). The differences in weight and WAT loss on chemotherapy could not be accounted for by variations in food intake, which did not depend on genotype, although chemotherapy resulted in a reduced food intake in both WT and Mut mice ( Supplementary Fig. 8A). Likewise, serum levels of the cachexia-inducing cytokines TNF-a and IL-6 were similar across genotypes and treatment regimens ( Supplementary  Fig. 8B). The protection from chemotherapy-induced cachexia in Mut mice is thus associated with the loss of myeloid cell-derived VEGF-A and the resulting increase in the level of circulating chemerin.
The cause of weight loss associated with chemotherapy is poorly understood. Our findings suggest that in addition to a proteolytic effect on skeletal muscle, cisplatin might have a strong and direct lipolytic effect that is modulated by chemerin. To investigate the possibility, gonadal WAT explants from C57Bl6/J mice were treated with cisplatin, which was found to induce Atgl expression (Fig. 5g) and to stimulate release of fatty acids (Fig. 5h). The expression of Hsl was also induced, although not significantly ( Supplementary Fig. 8C). Consistent with its context-specific role in enhancing or inhibiting lipolysis, chemerin increased Atgl expression and lipolysis in WAT explants (Fig. 5g,h) but suppressed the increased expression of Atgl and WAT lipolysis caused by addition of cisplatin (Fig. 5g,h). The experiment confirms that cisplatin directly stimulates WAT lipolysis, and that the effect is negated by chemerin, which thereby protects against therapy-associated loss of body weight.
Local and systemic effects of chemerin amend therapy outcome. Chemotherapy causes an increase in the intratumoural release of chemerin in Mut mice. Chemerin might thus be involved in the enhanced immune response in the absence of myeloid cell-derived VEGF-A, which is associated with the improved control of tumour growth. The interpretation was tested by means of an anti-chemerin antibody, which diminished chemotherapy-induced recruitment of NK cells in WT and Mut mice (Fig. 6a). The antibody completely blocked the clearance of senescent tumour cells after cytotoxic treatment in the absence of myeloid cell-derived VEGF-A, resulting in equal numbers of senescent cells in tumours from WT and Mut mice at endpoint (Fig. 6b,c). Blocking chemerin led to comparable outcomes in WT and Mut mice at endpoint (Fig. 6d). Comparable results were obtained by depleting NK cells (Fig. 6d). In the absence of NK cells, senescent cells were not cleared and remained in Mut tumours on regrowth (Fig. 6b,c) and there was no delay in tumour growth after chemotherapy (Fig. 6d). Finally, intratumoural injection of chemerin delayed tumour regrowth after chemotherapy in WT mice but had no additional effect in Mut mice (Fig. 6d). However, intratumoural chemerin injection does not further affect circulating chemerin levels in tumour-bearing and cisplatin-treated WT and Mut mice (Supplementary Fig. 8D). Moreover, neither local application of chemerin nor NK cell depletion protected against chemotherapy-induced weight loss ( Supplementary  Fig. 8E). Therefore, local and systemic chemerin effects need to be distinguished. The findings unequivocally link the improved outcome of chemotherapy in the absence of myeloid cell-derived VEGF-A to sufficient release of the chemoattractant chemerin by the endothelium, which locally activates NK cell-based antitumour defenses and prevents chemotherapy-exacerbated cachexia at the systemic level (graphical summary, Fig. 7).

Discussion
Targeting VEGF-A in myeloid cells leads to vascular normalization 3 . Here we show that targeting VEGF-A is also associated with an enhanced senescence response on chemotherapy. In addition to improved drug delivery, the reduced tumour hypoxia in Mut tumours may contribute to the effect, as hypoxia has been reported to prevent cellular senescence 33 . Although T-cell-mediated immune responses are impaired by a lack of oxygen 34 , it remains to be determined how NK cells react under hypoxic conditions. It is attractive to speculate that the reduced hypoxia in Mut mice improves NK cell-mediated cytotoxicity.
In addition to shaping the tumour vasculature, VEGF-A modulates the performance of various immune cells 35 . It may have an effect on the migration and cytotoxicity of NK cells, although findings are inconsistent 36,37 . It clearly attracts regulatory T cells to the tumour microenvironment 38 and interferes with the maturation of dendritic cells 35 . The absence of myeloid cell-derived VEGF-A from the tumour microenvironment could thus improve antitumour immune responses.
The chemotherapeutic agent cisplatin reduces vascular density and increases pericyte coverage, consistent with its known antiangiogenic properties 20 . The effect is independent of myeloid cellderived VEGF-A, although the density of blood vessels before chemotherapy is higher in tumours from WT mice than in those from mutant mice lacking VEGF-A in myeloid cells. The reduction in tumour blood vessels on chemotherapy may thus be enhanced by VEGF-A. The effect may stem from improved drug delivery and/or be related to the presumably higher number of proliferating ECs on VEGF-A-driven angiogenesis. The proliferating cells in the vasculature would be more susceptible to cytotoxic damage than quiescent cells.
Our study reveals that chemotherapy increases the level of PPAR-g within tumour ECs and stimulates them to release chemerin. However, only in the LLC model deletion of VEGF in myeloid cells resulted in increased systemic chemerin levels, whereas in the B16 model only local, intratumoural effects were observed. Local and systemic chemerin effects need to be distinguished. It is attractive to speculate that only sufficently elevated systemic (circulating) chemerin levels are able to ameliorate cisplatin-induced cachexia. These systemic and therefore cachexia-relevant effects need to be distinguished from local, intratumoural effects of chemerin, for example, clearance of senescent tumour cells and restriction of tumour growth. Therefore, local delivery by intratumoural injection of chemerin phenocopies (local) reduction of tumour size (Fig. 6d)   intratumoural chemerin injection does not further affect circulating chemerin levels in tumour-bearing and cisplatintreated WT and Mut mice ( Supplementary Fig. 8D). Likewise, deletion of VEGF in myeloid cells does not confer protection against cisplatin-induced cachexia in the B16 model (Fig. 1i). Again, the differences are in local versus systemic effects. This might again be due to the lack of increased circulating chemerin levels in cisplatin-treated Mut mice in the B16 model ( Supplementary Fig. 4C) compared with the LLC model (Fig. 4c). With the aim to reconcile the contradictory results we compared absolute chemerin mRNA expression levels in addition to n-fold expression as in the study, in isolated ECs, which we have identified as the major source of chemerin ( Fig. 4d-g) from LLC and B16 tumours across genotypes. As shown in Supplementary  Fig. 4E, chemerin mRNA levels are more than tenfold higher in ECs isolated from cisplatin-treated Mut LLC tumours compared with ECs isolated from cisplatin-treated Mut B16 tumours. In line with this, in the B16 model serum chemerin levels of cisplatin-treated Mut mice are lower than in the LLC model ( Fig. 4c and Supplementary Fig. 4C, respectively). This could explain why increased circulating chemerin levels and therefore systemic protection against chemotherapy-induced cachexia are only achieved in cisplatin-treated Mut LLC tumours, whereas local, intratumoural effects are observed in all models. Currently, we can only speculate regarding the different chemerin levels between tumour models. One reason might be that the tumour VEGF levels after cisplatin treatment in B16 tumours are generally higher (Supplementary Fig. 1A) than in LLC tumours (Fig. 2a) and, therefore, endothelial chemerin release is still repressed in B16 tumours. Alternatively, the increased expression of other angiogenic factors (for example, fibroblast growth factor; Supplementary Fig. 5G) in the B16 model may repress endothelial chemerin expression in cisplatin-treated Mut mice ( Supplementary Fig. 1E). Consistently, only increased serum levels in LLC-bearing Mut mice conferred protection against chemotherapy-induced cachexia.
The role of chemerin in skeletal muscle homeostasis is controversial 31,32 and the effect of chemerin on muscle loss in the context of cachexia is unknown. Our in vivo experiments show that chemerin prevents excessive loss of skeletal muscle on chemotherapy. Likewise, chemerin has opposing effects on lipid metabolism depending on the nutritional status and on other factors. In vitro experiments show that chemerin may have pro-or antilipolytic effects depending on the experimental conditions 13,30 . In vivo evidence is limited, although treatment of fasted mice with chemerin is known to inhibit lipolysis and release of free fatty acids 30 . Consistently, we show that lipolysis and the release of free fatty acids are downregulated by the addition of chemerin to WAT cultures after the chemotherapeutic induction of lipolysis. In contrast, chemerin treatment of WAT explants before chemotherapy induces lipolysis. We speculate that chemerin acts as a rheostat in the homeostasis of fat tissue, preventing excessive accumulation or depletion of fat reserves in the presence of powerful anti-or prolipolytic stimuli.
Tumour ECs release chemerin in response to chemotherapy but the effect is suppressed by VEGF-A derived from myeloid cells. Lowering intratumoural levels of VEGF-A after chemotherapy thus has an additional important effect: as well as normalizing the vasculature, it also fosters the endothelial production of chemerin. Consistently, elimination of myeloid cell-derived VEGF-A has a similar local effect (for example, tumour size restriction and increased NK cell infiltration as shown in Supplementary Fig. 6A-D) when etoposide, another cytotoxic agent, is used. Regarding the results in etoposide-treated LLC tumours, we would like to emphasize that etoposide treatment at the indicated dose phenocopies the intratumoural and hence local effects of cisplatin treatment in LLC-bearing Mut mice ( Supplementary Fig. 6A-D) and fails to increase systemic chemerin levels ( Supplementary Fig. 6E). Furthermore, etoposide at this dose induces only very mild cachexia ( Supplementary  Fig. 6F) compared with cisplatin treatment (Fig. 1h,i), although it still slows tumour growth (Supplementary Fig. 6A). Therefore, in this setting of overall weak chemotherapy-induced cachexia, potential protective effects against chemotherapy-induced cachexia by targeting myeloid cell-VEGF might not become apparent. Moreover, cisplatin and etoposide are non-immunogenic 39 and it will be important to investigate the effects on chemerin release of other immunogenic chemotherapeutics.
It is noteworthy that treatment with a VEGF-neutralizing antibody induced vascular normalization, improved the outcome of chemotherapy, endothelial chemerin expression and NK cell recruitment. Yet, anti-VEGF treatment under these particular conditions had no effect on cisplatin-exacerbated cachexia, presumably owing to the inability to increase systemic chemerin levels. Myeloid cell-derived VEGF has indeed been shown to play a unique role in VEGFR2-mediated signalling to the tumour endothelium that cannot be compensated for by other potential VEGF sources within the tumour microenvironment (for example, tumour cells), regardless of overall tumour VEGF levels 3 . This is attributed to the ability of myeloid cells (in particular macrophages) to generate transiently and locally very high VEGF concentrations in restricted tumour areas, which is not necessarily reflected by total VEGF levels within the tumour. Moreover, the mostly perivascular localization of tumour-associated macrophages puts them in a unique position and makes them presumably a critical and non-redundant source of VEGF directly adjacent to the abluminal side of the endothelium. This may explain why antibody-mediated general VEGF neutralization, predominantly targeting circulating VEGF, is less efficient than genetic targeting of VEGF in myeloid cells, in particular with regard to increasing endothelial chemerin release and systemic levels that are relevant for the protection against cachexia. However, general VEGF blockade in combination with cisplatin is still able to phenocopy the local effects, restricted to the tumour microenvironment (for example, tumour growth inhibition, vascular phenotype and immune cell infiltration) ( Supplementary Fig. 7).
The tumouricidal effects of many chemotherapeutic agents depends on the active contribution of immune cell effectors, especially those of the adaptive immune compartment 1 . In our tumour models, therapeutic success critically depends on NK cell-mediated tumour immune surveillance and tumour cell clearance. It is important to note that not all tumours are sensitive to NK cell-mediated tumour surveillance. Further work will be necessary to evaluate the outcome of drug-induced senescence and stromal chemerin release in tumour models that are predominantly controlled by T cells.
In summary, our study reveals that chemotherapy with cisplatin simulates tumour ECs to release chemerin. We show further that chemerin is a critical mediator of NK cell-mediated antitumour defenses and of cachexia as well (Scheme, Fig. 7). VEGF-A derived from myeloid cells suppresses the stimulation of endothelial chemerin release by chemotherapy. Hence, targeting VEGF signalling should impede the lipolysis and weight loss that is frequently associated with chemotherapy. Our study therefore offers novel therapeutic avenues to improve the overall outcome of chemotherapy.

Methods
Animals and procedures. The Animal Care and Use Committee of the Bezirksregierung Düsseldorf, Germany, approved all procedures performed on mice. Mice (C57Bl/6J) with both alleles of exon 3 of VEGF-A flanked by loxP sites (VEGF þ f / þ f ) 40 were bred with mice (C57Bl/6J) homozygous for the floxed VEGF allele expressing Cre recombinase driven by the lysozyme M promoter 41 3 . Male mice at 10-12 weeks of age (C57Bl/6J) were used. Chemotherapy was started 8 days after subcutanous injection of 10 7 LLC cells and 10 days after injection of 10 7 B16F10 cells. Three doses of cisplatin (8 mg per kg body weight, Sigma) or etoposide (15 mg kg À 1 , Sigma) were given by intraperitoneal (i.p.) injection every 2 days. Tumour size was monitored every 2 days using a caliper and the tumour volume was calculated as V ¼ p/6*A Â B2. Tumours were allowed to grow until the maximum permitted size was reached or ulcerations occured. Pimonidazole hydrochloride (Hypoxyprobe-1) was injected i.p. (60 mg per kg body weight) 30 min before tumour removal and detected by the monoclonal antibody Mab-1. Tumour doubling time was calculated as DT ¼ (T À T0) Â ln2/(lnV À lnV0), where T À T0 indicates the time between two measurements and V0 and V denote the tumour volume at these times. Mice of different genotypes were allocated randomly to the different treatment groups and analysis was carried out in a single-blinded manner. Sample size was estimated based on previous studies with the experimental models 3 .
Depletion of NK cells. After the last cytotoxic treatment, randomized cohorts of WT and Mut mice were injected i.p. with anti-NK1.1 monoclonal antibody PK136 (4 mg per kg body weight) at day 13 and 15. Control mice received i.p. injections of 100 ml PBS.
Chemerin neutralization. Randomized cohorts of WT and Mut mice received i.p. injections of 400 mg per kg body weight anti-chemerin (R&D Systems) on days 11, 13 and 15. Control mice were injected i.p. with PBS.
Assessment of skeletal muscle. Gastrocnemius muscles were dissected and used to assess muscle atrophy. These muscles were frozen and 7 mm-thick serial sections were stained histochemically for myofibrillar ATPases. Morphometric analyses were then performed on these muscles to determine the cross-sectional area of fibres for each group; n ¼ 2 mice for the gastrocnemius muscle.
Chemerin injection. Randomized cohorts of WT and Mut mice received intratumoural injections of 250 ng of recombinant, active, carrier-free murine chemerin (R&D Systems) reconstitued in PBS daily starting on day 6 until endpoint at day 18. Control mice were injected with PBS.
VEGF neutralization. WT mice received three i.p. injections at days 7, 9 and 11 of 100 mg of rat anti-mouse VEGF-A antibody (Biolegend, #512808). Control mice were injected with isotype control (#400533). RNA extraction and reverse transcriptase-quantitative PCR. Total RNA was isolated by phenol/chloroform extraction. Complementary DNA was synthesized from 3 mg of DNA-free total RNA using M-MLV Reverse Transcriptase (Promega) and oligo-dT primers (Life Technologies). Gene-specific transcription levels were determined using SYBR Green Mastermix (Promega) and an IQ5 real-time PCR machine (Bio-Rad). PCR conditions were as follows: 95°C for 10 min followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. Data were normalized to 16S or b-actin mRNA levels. The following primers were used: 16s forward primer:  (Dako). The TdT-mediated dUTP nick end labelling assay was performed in accordance with the manufacturer's instructions (Promega). Chemerin/CD31 immunofluorescence was performed after heatinduced antigen retrieval.
Senescence associated b-galactosidase staining. Ten micrometres of frozen sections of OCT-embedded tumours were stained for the senescence marker SA b-Gal according to the manufacturer's instructions (Cell Signaling) for B16F10 tumours the staining was performed with dodecanoylaminofluorescein di-b-D-galactopyranoside (C12FDG; Life Technologies) at 33 mM during 2 h at 37°C at pH 6.0.
Isolation of ECs. To isolate ECs, at experiment endpoint tumours were digested in cell lysis buffer (DMEM þ 2 mg ml À 1 collagenase type III) for 1 h at 37°C. Up to 4 Â 10 7 cells were incubated with mouse CD31 microbeads (Miltenyi Biotec) and ECs were separated to a purity of Z90% by positive selection according to the manufacturer's instruction. ECs (8 Â 10 5 ±1,3 Â 10 5 ) were isolated corresponding to 5.59% ± 1.45% of total cells.
Determination of free fatty acids of isolated WAT. Fat tissue explants were obtained from gonadal WAT of C57/Bl6J-WT mice and cut into small pieces. The pieces were maintained for 18 h in DMEM supplemented with 2% BSA at 37°C and 5% CO 2 . Aliquots of medium were analysed for free fatty acids using a commercial kit (Free Fatty Acid Quantification Kit, Abcam) following the manufacturer's instructions.
Quantitative analysis of histology markers. For quantitative analysis of blood vessels, five areas of each tumour section were randomly selected and photographed using a Nikon Eclipse E1000 microscope and the Nikon DS-Ri1 camera system. The area (number of pixels/px) marked by CD31 was measured using the ImageJ programme (National Institutes of Health) and calculated as the percentage of the area covered by 4,6-diamidino-2-phenylindole. Pericyte coverage was calculated as percentage of total number of blood vessels counted. Chemerin-positive vessels were calculated as percentage of total CD31-positive vessels. To determine cell proliferation, apoptosis and cellular senescence, cells positive for the marker in question were counted in five randomly chosen tumour areas for each section and the mean value calculated.
ELISA assay. Concentrations of VEGF-A and chemerin in tumours and aliquots of medium were determined using commercial kits (Quantikine ELISA Immunoassay, R&D Systems) and expressed in pg ml À 1 per mg of whole tissue protein. Serum levels of TNF-a and IL-6 were measured using mouse TNF-a and IL-6 quantikine ELISA kit (R&D Systems) and normalized to serum protein levels.
Cell culture. Cell lines were obtained from ATCC. Cells were cultured in DMEM high-glucose medium supplemented with 10% FCS, 50 U ml À 1 penicillin and 100 mg ml À 1 streptomycin at 37°C in a humidified atmosphere of 5% CO 2 in air and were checked for Mycoplasma contamination.
Western blotting. Protein samples were separated using a 10% polyacrylamide gel under reducing and denaturating conditions and transferred onto a polyvinylidene difluoride membrane followed by enhanced chemiluminescence (ECL) detection of the antibody. For quantitative analysis, the membranes were scanned with the ImageQuant LAS 4000mini (GE Healtcare Life Sciences) and the integrated density was measured using the software ImageJ (National Institutes of Health). Images have been cropped for presentation. Full-size images are presented in Supplementary Fig. 10.
Data availability. The data that support the findings of this study are available from the corresponding author upon request.