p16Ink4a and p21Cip1/Waf1 promote tumour growth by enhancing myeloid-derived suppressor cells chemotaxis

p16Ink4a and p21Cip1/Waf1 act as tumour suppressors through induction of cellular senescence. However, senescence-independent roles of these CDK inhibitors are not well understood. Here, we report an unexpected function of p16Ink4 and p21Cip1/Waf1, namely, tumour promotion through chemotaxis. In monocytic myeloid-derived suppressor cells (Mo-MDSCs), p16Ink4 and p21Cip1/Waf1 are highly expressed and stimulate CX3CR1 chemokine receptor expression by preventing CDK-mediated phosphorylation and inactivation of SMAD3. Thus, deletion of p16 Ink4 and p21 Cip1/Waf1 reduces CX3CR1 expression, thereby inhibiting Mo-MDSC accumulation in tumours expressing CX3CL1 and suppressing the tumour progression in mice. Notably, blockade of the CX3CL1/CX3CR1 axis suppresses tumour growth, whereas inactivation of CDKs elicits the opposite effect. These findings reveal an unexpected function of p16 Ink4a and p21 Waf1/Cip1 and indicate that regulation of Mo-MDSCs chemotaxis is a valuable potential strategy for control of tumour development.

T he p16 Ink4a and p21 Waf1/Cip1 genes, which encode cyclindependent kinase (CDK) inhibitors, are upregulated in cultured mammalian primary cells upon detection of various potentially oncogenic stimuli 1,2 . This unique feature of p16 Ink4a and p21 Waf1/Cip1 , together with their ability to induce irreversible cell cycle arrest (termed cellular senescence), suggests that these genes act as a safeguard against neoplasia [3][4][5] . Indeed, mice lacking p16 Ink4a and/or p21 Waf1/Cip1 exhibit early onset of cancer [6][7][8][9] , illustrating the importance of p16 Ink4a and p21 Waf1/Cip1 in tumour suppression in vivo. To observe the physiological roles of p16 Ink4a and p21 Waf1/Cip1 during tumour formation, we previously generated transgenic mice lines expressing firefly luciferase under the control of the p16 Ink4a or p21 Waf1/Cip1 gene promoters; these were termed p16-luc or p21-luc mice, respectively, and we revealed the timing, and hence, the likely roles of p16 Ink4a and/or p21 Waf1/Cip1 expression in de novo tumorigenesis in vivo 10,11 . Similar to our work, Burd et al. reported the generation of another strain of p16 Ink4a reporter mice (p16 Luc mice), in which the p16 Ink4a coding sequence was replaced with cDNA encoding firefly luciferase 12 . Notably, in addition to ageing and de novo tumorigenesis, p16 Ink4a expression was strikingly induced in the stroma of developing neoplasia. Lethal irradiation coupled with bone marrow (BM) transplantation from syngeneic p16 Luc mice indicated the presence of p16 Ink4a -expressing BM-derived cells in the tumour microenvironment 12 . However, it is not known which cell types express p16 Ink4a in the stroma of developing neoplasia. Moreover, the biological role of p16 Ink4a expression in these cancer stromal cells remains unclear.
In this study, we found that both p16 Ink4a and p21 Waf1/Cip1 are highly expressed in myeloid-derived suppressor cells (MDSCs) without any noticeable senescence-associated phenotypes. MDSCs are known to negatively regulate immune responses and facilitate tumour progression 13 . The systemic expansion of MDSCs in BM and the MDSC migration to peripheral are the hallmarks of tumour-bearing condition 14 . In mice, MDSCs are defined as Gr-1 + CD11b + cells and categorised into 2 subtypes: polymorphonuclear (PMN-) and monocytic (Mo-) MDSCs 15,16 . Mo-MDSCs are a CD11b + Ly6C high Ly6G − subset that is highly immunosuppressive and exerts its effects in both antigen-specific and antigen-nonspecific manners 15,16 . In contrast, CD11b + Ly6C int Ly6G + PMN-MDSCs are moderately immunosuppressive and act via antigenspecific mechanisms 15,16 . Mo-and PMN-MDSCs differ markedly in terms of the molecular mechanisms involved in their migration/ chemotaxis. Mo-MDSCs express C-C chemokine receptor type (CCR)2, CCR5 and CX3C chemokine receptor (CX3CR)1 17,18 , whereas PMN-MDSCs express CXCR2 19 . The accumulation of intratumoural MDSCs dependent on these chemokine receptors is well correlated with tumour growth 17,19 .
In the present study, we reveal that p16 Ink4a and p21 Waf1/Cip1 upregulate CX3CR1 expression by preventing CDK-mediated phosphorylation and inactivation of SMAD3 in Mo-MDSCs. Thus, deletion of p16 Ink4a and p21 Waf1/Cip1 in mice results in a substantial decrease in infiltration of Mo-MDSCs into tumours and causes slower growth of tumour allografts. Conversely, inactivation of CDKs by chemical inhibitors increases the expression of CX3CR1 in Mo-MDSCs, resulting in accumulation of Mo-MDSCs in tumours and consequent acceleration of tumour growth in allograft mouse models. These results uncover a novel function of p16 Ink4a and p21 Waf1/Cip1 in MDSC chemotaxis, and provide valuable new insight into how to bypass this undesirable side effect of CDK inhibitors.

Results
p16 and p21 are expressed in MDSCs in tumour-bearing mice. We previously performed in vivo imaging of p16 Ink4a or p21 Cip1/ Waf1 expression in mice and elucidated the dynamics of their expression during the development of skin cancer, using p16-luc or p21-luc mice [9][10][11] . This approach, together with the analysis of p16 Ink4a and/or p21 Cip1/Waf1 -knockout mice, suggested that p16 Ink4a and p21 Cip1/Waf1 are likely to be expressed in premalignant benign tumour cells, thereby having a protective role against malignant transformation. However, we were unable to exclude the possibility of p16 Ink4a and p21 Cip1/Waf1 expression in non-tumour cells within the tumour microenvironment in these experimental settings. Thus, to ascertain this possibility, we inoculated p16-and p21-luc mice with syngeneic cancer cell lines, such as Lewis lung carcinoma (LLC) or spindle cell tumour (SCT) cells, neither of which expresses luciferase. Notably, inoculation of these cancer cell lines induced significant levels of luciferase activity in the area of tumour formation, but not in contralateral Matrigel-injected areas, in both p16-luc and p21-luc mice (Fig. 1a,  b). This is consistent with a previous report describing the presence of BM-derived p16 Ink4a -expressing stroma cells in the tumour microenvironment 12 . Collectively, these results lead us to an idea that the expression of p16 Ink4a and/or p21 Cip1/Waf1 may have important roles in establishing the tumour microenvironment during cancer development.
To explore this possibility further, we next attempted to identify the cell types expressing p16 Ink4a and/or p21 Cip1/Waf1 in tumour stroma. In addition to cells in the tumour stroma, spleen cells in tumour-bearing (TB) mice expressed both p16 Ink4a and p21 Cip1/Waf1 and were CD11b + and Gr-1 + , as indicated by immunohistochemistry ( Fig. 1c-f). Peripheral CD11b + Gr-1 + cells are known to be MDSCs 20 and MDSCs are often induced by the presence of tumours in mice and humans 21,22 . Accordingly, PMN-MDSCs and Mo-MDSCs were purified from the spleens of tumour-bearing mice, and p16 Ink4a and p21 Cip1/Waf1 mRNA levels were examined by quantitative real-time reverse transcription (qRT-) PCR (Fig. 1g, h). Interestingly, although p16 Ink4a was expressed in both PMN-MDSCs and Mo-MDSCs, p21 Cip1/Waf1 was only expressed in Mo-MDSCs.
As p16 Ink4a and p21 Cip1/Waf1 CDK inhibitors have established roles in cellular senescence, we tested if p16 Ink4a -and/or p21 Cip1/ Waf1 -expressing MDSCs exhibit senescence-like phenotypes. Consistent with a previous report 23 , BM Mo-MDSCs are proliferative and the percentage of Mo-MDSCs in the S phase increases in mice lacking both p16 Ink4a and p21 Cip1/Waf1 (p16/ p21-DKO mice), compared to in wild-type (WT) mice (Supplementary Fig. 1a). On the other hand, in either splenic or intratumoural MDSCs, there is no difference in cell cycle phase distribution between MDSCs from WT mice and those from p16/ p21-DKO mice ( Supplementary Fig. 1a). Notably, although proliferation of MDSCs isolated from spleen was rarely detected by a 5-ethynyl-2′-deoxyuridine (EdU) incorporation assay in vivo ( Supplementary Fig. 1b), a carboxyfluoroscein diacetate succinimidyl ester (CFSE) dilution analysis indicated that a substantial amount of these MDSCs (Mo-MDSCs >20%, PMN-MDSCs >60%) resumed proliferation upon stimulation with GM-CSF in vitro ( Supplementary Fig. 1c). Moreover, other senescenceassociated phenotypic characteristics, such as accumulation of γH2AX foci and 53BP1 foci (signs of DNA damage), reduction of lamin B1 expression 24 , and induction of IL-6 expression 25 , were not observed in these MDSCs (Supplementary Fig. 1d-g). These results, together with the observations that these MDSCs were resistant to ABT-263, a senolytic drug that specifically kills senescent cells 26 , in both in vitro and in vivo ( Supplementary  Fig. 1h, i), indicate that these MDSCs are very unlikely to be in a state of cellular senescence despite their high expression of p16 Ink4a and p21 Cip1/Waf1 . These findings then raise questions about the roles of p16 Ink4a and p21 Cip1/Waf1 expression in MDSCs.
p16 and p21 in Mo-MDSCs promote tumorigenesis in vivo. MDSCs have been reported to exert immunosuppressive effects and promote tumour development 14 . To verify the tumourpromoting effect of MDSCs expressing p16 Ink4a and p21 Cip1/Waf1 , WT and p16/p21-DKO mice were subcutaneously inoculated with orthotopic SCT cells. Surprisingly, SCT cells grew more slowly in p16/p21-DKO mice than in sex-and age-matched WT mice (Fig. 2a, b; Supplementary Fig. 2), suggesting that p16 Ink4a and p21 Cip1/Waf1 have a tumour-promoting role, at least in the present experimental setting. The primary function of tumourderived MDSCs is reportedly immunosuppression, especially inhibition of T cell activation 14 . We therefore investigated whether p16 Ink4a and p21 Cip1/Waf1 are involved in immunosuppressive activity of MDSCs by assessing the activation of T cells cocultured with MDSCs from WT or p16/p21-DKO mice. Purified PMN-and Mo-MDSCs were added to OVA-pulsed DC2.4 cells, a dendritic cell line 27 , and to RF33.70 cells, an OVA (SIINFEKL)specific T-T hybridoma 28 . T cell activation was evaluated by measuring interleukin (IL)-2 production by RF33.70 cells 29 . Unexpectedly, however, there was no significant difference in immunosuppressive capacity between WT and p16/p21-DKO MDSCs ( Supplementary Fig. 3a). Similar results were also obtained by performing a co-culture assay of splenic Mo-MDSCs or intratumoural Mo-MDSCs with purified splenic T cells stimulated with anti-CD3/CD28 antibody-coated beads (Supplementary Fig. 3b, c). In agreement with these observations, expression of the MDSC-derived immunosuppressive factors arginase (Arg)1 30 and nitric oxide synthase (Nos)2 31 was not downregulated in Mo-MDSCs lacking p16 Ink4a and p21 Cip1/Waf1 (Supplementary Fig. 3d). The expression levels of Il1rn, which encodes IL-1RA and promotes tumour growth by inhibiting an inflammatory cytokine IL-1α 32 , and Cd274, which encodes programmed death ligand (PD-L)1 and kills T cells 33 , were similar between WT MDSCs and p16/p21-DKO MDSCs ( Supplementary  Fig. 3d). Therefore, it is unlikely that the expression of p16 Ink4a and p21 Cip1/Waf1 in MDSCs is related to T cell suppression. Furthermore, although MDSCs reportedly promote angiogenesis 34 and the epithelial-mesenchymal transition (EMT) 35 to assist tumour growth, RNA-sequencing analysis showed that no angiogenesis-or EMT-inducing factors were downregulated in Mo-MDSCs from p16/p21-DKO mice compared with those from WT mice (Supplementary Data 1 and 2).   We thus next investigated whether the antitumoural phenotype in p16/p21-DKO mice is attributable to MDSC-mediated effects. Indeed, adoptive transfer experiments indicated that Mo-MDSCs purified from p16/p21-DKO mice were less effective in promoting tumour growth than those from WT mice (Fig. 2c). Consistent with a previous report 36 , transferred PMN-MDSCs did not promote tumour growth regardless of the presence or absence of p16 Ink4a and p21 Cip1/Waf1 expression (Fig. 2d). Collectively, these results strongly indicate that Mo-MDSCs exhibit differences in terms of tumour growth between WT and p16/p21-DKO mice. It should also be noted that p16 Ink4a -and p21 Cip1/Waf1 -dependent facilitation of tumour growth was observed when Mo-MDSCs were intravenously transferred but not intratumorally transferred (Fig. 2c, e). These results suggest that the failure of Mo-MDSC infiltration into tumour allografts is likely to underlie defects in tumour-promoting activity in p16/ p21-DKO mice. Certainly, more intravenously injected WT Mo-MDSCs were observed in the tumour than p16/p21-DKO Mo-MDSCs, but there was no difference in Mo-MDSC infiltration upon intratumoural injection (Fig. 2f). We additionally compared MDSC populations in tumour-bearing WT and p16/p21-DKO mice by flow cytometry. There were no differences between PMN-and Mo-MDSC populations in either the BM or spleen; however, fewer intratumoural Mo-MDSCs were obtained from p16/p21-DKO mice than from WT mice (Fig. 2g, h). Furthermore, more activated (CD44 high CD62L low ) CD8 + T cells (Supplementary Fig. 4a, b), but fewer regulatory T cells, which are peripherally expanded by MDSCs 37,38 , were present in the tumour-draining lymph nodes of p16/p21-DKO mice (Supplementary Fig. 4c, d). In addition, expression of another activation marker, CD69, was increased, and the population of cells positive for PD-1 39 , an exhaustion marker, was decreased in intratumour CD8 T cells of p16/p21-DKO mice (Supplementary Fig. 4e-h). Together, these data demonstrate that Mo-MDSC localisation is important for local immunosuppression and consequent tumour development 17,40 .
CX3CR1 expression is reduced in Mo-MDSCs lacking p16 and p21. We next sought to identify the mechanism underlying the inability of Mo-MDSCs to infiltrate tumour allografts in p16/p21-DKO mice. Chemotaxis is important for immune cell localisation; MDSC migration into tumours, which is dependent on chemokine receptors, affects tumour progression. We therefore attempted to identify chemokine receptors that were highly expressed in Mo-MDSCs but not in PMN-MDSCs, and that were additionally downregulated in p16/p21-DKO mice relative to WT mice. Ccr2, Ccr5 and Cx3cr1 were identified by RNA-sequencing analysis, and the differences in their expression levels were validated by qRT-PCR ( CX3CR1 was also downregulated in the BM Mo-MDSCs lacking both p16 Ink4a and p21 Cip1/Waf1 , where progenitor cells differentiate into MDSCs and MDSCs undergo expansion (Fig. 3c, d). However, a significant difference in CX3CR1 expression in intratumour Mo-MDSCs between WT and p16/p21-DKO was not observed (Fig. 3c, d), although the number of intratumoural Mo-MDSCs in p16/p21-DKO mice was much smaller compared to WT mice, suggesting that CX3CR1 highly expressing Mo-MDSCs are concentrated in tumour site, regardless of the p16 Ink4a and p21 Cip1/Waf1 gene status. This finding suggests that the inefficiency of Mo-MDSCs to infiltrate tumour allografts in p16/p21-DKO mice is likely attributable to the reduction of CX3CR1 expression in these mice. CX3CL1 + tumours exhibit p16 and p21-dependent tumour growth. To confirm the role of reduced CX3CR1 expression in the failure of Mo-MDSC infiltration into tumour allografts, we next assessed whether our findings were reproducible in other syngeneic tumour cell lines. Unexpectedly, LLC cells showed no differences in tumour growth between WT and p16/p21-DKO mice ( Supplementary Fig. 6a). Notably, however, qRT-PCR analysis revealed that the expression level of CX3CL1 (also known as fractalkine), the sole CX3CR1 ligand, was much smaller in LLC cells than in SCTs ( Fig. 4a; Supplementary Fig. 6b). On the other hand, the CCR2 ligands CCL2 and CCL7 were more highly expressed in LLC cells than in SCTs ( Supplementary  Fig. 6c, d). These results suggest that the tumour-suppressive effect of p16 Ink4a and p21 Cip1/Waf1 is likely dependent on the levels of CX3CL1 expression in tumour cells.
Furthermore, slower growth of LLC tumour allografts in p16/ p21-DKO mice was observed when CX3CL1 was ectopically expressed in LLC cells (Fig. 4a, b). Conversely, shRNA-mediated depletion of Cx3cl1 in SCT cells significantly reduced tumour growth in WT syngeneic immunocompetent mice (Fig. 4a, c), but not in immunodeficient mice (Fig. 4d). As expected, the population of intratumoural Mo-MDSCs was reduced, whereas that of intratumoural CD8 + T cells was increased by Cx3cl1 knockdown relative to control SCTs (Fig. 4e, f). Consistently, CX3CL1 neutralisation antibody suppressed tumour progression (Fig. 4g), indicating that the CX3CL1-CX3CR1 pathway represents a potential therapeutic target. These data indicate that tumour growth acceleration induced by p16 Ink4a and p21 Cip1/ Waf1 is dependent on the CX3CL1-CX3CR1 pathway in tumours. It was previously reported that CX3CR1-CX3CR1 pathway regulates myeloid cell survival by inhibition of pro-apoptotic signals 41 ; therefore, we analysed the apoptosis of Mo-MDSCs. However, there was no difference in the expression levels of the apoptotic inhibitory factor Bcl2 between WT and DKO Mo-MDSCs ( Supplementary Fig. 7a). The ratio of apoptotic Mo-MDSCs in spleen or tumour was similar between WT and DKO mice ( Supplementary Fig. 7b). Thus attraction rather than IFN-γ enhances Mo-MDSC differentiation from BM cells. We next examined the mechanisms underlying the effect of p16 Ink4a and p21 Cip1/Waf1 on Cx3cr1 expression in Mo-MDSCs. Purified Mo-MDSCs are unstable and rapidly differentiate into various myeloid cell types 16 ; therefore, a stable Mo-MDSC culture was required for this step. MDSCs reportedly differentiate from BM cells in vitro by culturing in medium containing IL-6 and GM-CSF 42 ; however, this method mainly induces differentiation into PMN-MDSCs but not Mo-MDSCs. It has been shown that the IFN-γ-STAT1 axis has an important role in the immunosuppressive ability of Mo-MDSCs 15,43 . Therefore, we tested whether IFN-γ is involved in the differentiation of MDSCs, especially into Mo-MDSCs. Flow cytometry analysis revealed a greater number of Mo-MDSCs in cultures containing IFN-γ than in those containing GM-CSF and IL-6 ( Supplementary Fig. 8a). There were no differences in T cell suppression, the primary function of MDCS, between BM-derived cells cultured in the presence of GM-CSF, IL-6 and IFN-γ, and BM-derived MDSCs induced by GM-CSF with IL-6 ( Supplementary Fig. 8b). In addition to Arg1 and Nos2, which are important for the immunosuppressive function of MDSCs, the MDSC marker genes S100a8 and S100a9 44,45 were strongly induced when cells were cultured in GM-CSF, IL-6 and IFN-γ ( Supplementary Fig. 8c). On the basis of these results, we concluded that BM cells stimulated with GM-CSF, IL-6 and IFN-γ may be used as BM-derived Mo-MDSCs (BM-Mo-MDSCs).
We next attempted to identify the transcription factors that regulate Cx3cr1 expression under the control of CDK activity in Mo-MDSCs. Substantial reduction of tumour growth was observed in p16/p21-DKO mice but not in mice lacking either p16 Ink4a or p21 Cip1/Waf1 (single KO mice) (Fig. 2a, b; Supplementary Fig. 2), suggesting that both CDK2 and CDK4 associate  . e, f Representative flow cytometry plots of CD45 + CD11b + -gated (e) and CD45 + -gated (f) cells in tumours from WT mice 2 weeks after inoculation with Cx3cl1-knockdown or control SCT cells. Cells were labelled with CD45, CD11b, Ly6C and Ly6G (e) and CD45 and CD8 (f) antibodies. Graphs indicate CD11b + Ly6C high Ly6G − (Mo-MDSC) (e) and CD3 + CD8 + (CD8 + T cell) populations (f) among CD45 + cells. g SCT growth curve in WT mice intraperitoneally injected every other day with anti-CX3CL1 antibody (4 µg per mouse) or isotype control rat IgG (4 µg per mouse) from day 10 to 23; data are presented as the mean ± SEM. The statistical significance was determined by Student's t-test; **p < 0.01 and ***p < 0.001; NS not significant NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02281-x ARTICLE with this transcription factor in a complementary fashion. On the basis of this assumption, we identified SMAD3 as a candidate transcription factor using the STRING database (http://string-db. org). It has been reported that phosphorylation of the SMAD3 linker region by CDK2 and CDK4 inhibits SMAD3 transcriptional activity [53][54][55] . Indeed, the linker phosphorylation (Ser-213) in SMAD3 was increased in BM and splenic Mo-MDSCs from p16/p21-DKO mice compared with those from WT mice (Fig. 5e).
Notably, the levels of CX3CR1 expression were inversely correlated with those of phosphorylated SMAD3 (Figs. 3c and 5e). These data suggest that p16 Ink4a   We next confirmed whether the change of linker phosphorylation level in SMAD3 affects Cx3cr1 expression directly. Two SMAD-binding elements, SBE1 and SBE2, were present in the Cx3cr1 promoter region. A chromatin immunoprecipitation (ChIP) assay showed that SMAD3 binding to SBE2 was downregulated in p16/p21-DKO and upregulated by treatment with a CDK inhibitor (Fig. 5f). SMAD3 activation then stimulated Cx3cr1 promoter activity via binding to SBE2 (Fig. 5g) and ectopic expression of CDK2 inhibited SBE-mediated Cx3cr1 promoter activity (Fig. 5h). Together, these data strongly suggest that inactivation of CDK2 induce Cx3cr1 expression via SMAD3 activation.
Flavopiridol enhances SCT growth in vivo. CDK inhibitors such as flavopiridol, which is used to treat leukaemia 46 , have attracted recent attention as a result of their potential as anticancer agents. However, the present results suggest that CDK inhibition may accelerate tumour growth by enhancing Mo-MDSC infiltration into tumours expressing CX3CL1 in vivo. To explore this possibility, we generated SCT cells expressing SV40 large T antigen (SCT-LT), in which the retinoblastoma tumour suppressor (Rb)pathway is inactivated, to elicit resistance against CDK inhibitors ( Supplementary Fig. 9a, b). The growth of SCT-LT was substantially increased in mice treated with flavopiridol, accompanied by enhanced CX3CR1 expression in Mo-MDSCs and increased accumulation of Mo-MDSCs in tumours (Fig. 6a-e). These data, in conjunction with the notion that a certain proportion of human cancers contains disruption of the Rb-pathway, strongly suggest that treatment with CDK inhibition may promote tumour development, depending on the level of expression of CX3CL1 and the Rb status in the tumour.

Discussion
To our knowledge, the present work is the first to report that p16 Ink4a and p21 Cip1/Waf1 CDK inhibitors, which are considered cell-intrinsic tumour suppressors, in fact promote tumour development, as demonstrated in the present allograft mouse model. Both p16 Ink4a and p21 Cip1/Waf1 are expressed in Mo-MDSCs and facilitate infiltration of Mo-MDSC into the tumour microenvironment by upregulating CX3CR1 expression and conferring resistance to anti-tumour immune mechanisms (Figs. 1-5; Supplementary Fig. 10a). Notable, moreover, administration of flavopiridol was found to accelerate the development of orthotopically transplanted SCT-LT cells in syngeneic mice ( Fig. 6; Supplementary Fig. 10b). These results, together with previous reports showing that high p16 INK4A expression in the tumour stroma was strongly associated with high risk of recurrence 56 or poor survival 57 in some types of cancer, suggest that there might be the negative side effects of CDK inhibitor drugs used for cancer treatment.
Regarding the role of p16 Ink4a in myeloid cells, it is known to promote M1 polarisation, which reflects an inflammatory state for macrophages 58 . Indeed, the present study also showed Arg1 upregulation in p16/p21-DKO Mo-MDSCs ( Supplementary  Fig. 3d). It is generally believed that M2 skewing promotes tumour growth. Therefore, it is most likely that the decreased tumour growth in p16/p21-DKO mice is caused by MDSC alteration rather than defective macrophage polarisation.
Although the role of CDK2 and CDK4 (CDK2/4) in cell cycle control is well established, there is increasing evidence that CDK2/4 also have cell-cycle independent roles in various nonproliferating cells 55 . In the present study, we revealed a novel function of CDK2 in regulating CX3CR1 expression via phosphorylation and inactivation of SMAD3 in Mo-MDSCs (Fig. 5). On the basis of our comparative data of MDSCs in bone marrow, in spleen and in tumour, we propose a possible scenario whereby p16/p21 regulates Mo-MDSCs function during their maturation. Once Mo-MDSCs leave BM, they start to mature, terminate its proliferation and exert their own character, for example chemotaxis and immunosuppression. During this maturation process, CDK2 activity is sustained, yet not enough to drive cell cycle, and tuned by p16 Ink4a and p21 Cip1/Waf1 to lead the expression of CX3CR1 for chemotaxis ( Supplementary Fig. 1a; Figs. 3c, d and 5d, e).
The chemotaxis of immune cells contributes to important aspects of the tumour microenvironment 59 ; in particular, the use of immune checkpoint inhibitors as therapeutic agents, which has achieved significant results in cancer therapy, has revealed the importance of tumour-infiltrating immune cells for predicting the response to checkpoint blockade 60,61 . Note that the absence of CX3CL1 expression in pre-treatment tumours 60 is reportedly correlated with the response to the anti-PD-L1 antibody. We speculate that high CX3CL1 expression might recruit Mo-MDSCs to tumour, thereby increasing PD-L1-independent immunosuppression. However, the role of the CX3CL1-CX3CR1 axis in antitumour immunity is highly controversial. For instance, the correlation between CX3CL1 expression and the poor prognosis of cancer patients varies depending on the type of cancer: in colorectal cancer and gastric adenocarcinomas, high CX3CL1 expression is associated with better prognosis 62,63 ; however, the opposite outcome is observed in breast cancer and hepatocellular carcinoma 64,65 . Using SCT cells, which are derived from the skin of p16 Ink4a knockout mice and are highly expressing CX3CL1, we showed that treatment with anti-CX3CL1 exerted significant suppressive effects against tumour progression, which has not been previously demonstrated in vivo (Fig. 4g). We therefore suggest that CX3CL1 represents a novel molecular target for cancer therapy, especially in combination with CDK inhibitors (Supplementary Fig. 10b).
In tumour microenvironment, cancer-associated fibroblasts (CAFs) and pre-cancerous cells are shown to exhibit senescencelike phenotypes including SASPs. For example, it was recently reported that some CAFs and pre-cancerous hepatocytes recruit MDSCs via IL-6 and CCL2-mediated pathway, respectively, thereby protecting cancer cells from immune cells 66,67 . These results, together with our findings that p16 Ink4a and p21 Cip1/Waf1 enhance Mo-MDSC chemotaxis, led us to speculate that the formation of the tumour microenvironment may have a synergistic effect on protumoural MDSC recruitment The role of p16 Ink4a in the development of cellular senescence is well known; furthermore, p16 Ink4a has been the most widely accepted marker for senescent cells for over two decades 1,11,12,68 . Notably, moreover, a recent study reported that elimination of p16 Ink4a -expressing senescent cells from mice prolonged lifespan and alleviated ageing related diseases including cancer 69 , indicating that the elimination of senescent cells would be beneficial for homoeostasis, especially in aged animals. However, because MDSCs express high levels of p16 Ink4a without any noticeable senescent phenotype, the results of eliminating p16 Ink4a expressing cells must be interpreted with caution, as previously suggested by Baker et al.
In summary, our results reveal a novel function of p16 Ink4a and p21 Cip1/Waf1 CDK inhibitors in MDSCs, thus expanding our understanding of the mechanisms by which the chemotaxis of MDSCs is regulated. A greater understanding of the molecular mechanisms linking these CDK inhibitors to the chemotaxis of immune cells should therefore provide valuable novel strategies to control tumorigenesis.

Methods
Mice. WT C57BL/6J mice were purchased from Charles River Laboratories Japan and CLEA Japan. SCID-Beige (CB17.Cg-Prkdc scid Lyst bg-J /CrlCrlj) mice were purchased from Charles River Laboratories Japan. p16/p21-DKO mice were generated as previously reported 9 . p16-luc mice 11 and p21-luc mice 10 were generated as previously reported and backcrossed with C57BL/6 mice for at least 8 generations. All animals were maintained according to protocols approved by the Committee for the Use and Care of Experimental Animals of the Japanese Foundation for Cancer Research (Tokyo, Japan) and the Research Institute for Microbial Diseases, Osaka University (Osaka, Japan).
MDSC isolation. PMN-MDSCs (Ly6G + ) and Mo-MDSCs (Ly6G − Gr-1 + ) were isolated from the BMs, spleens, or tumours of SCT-bearing mice 3 weeks after injection, using the Myeloid-Derived Suppressor Cell Isolation kit for mice (Miltenyi Biotec) according to the manufacturer's instructions, with all steps performed at 4°C. The purity of MDSC subfractions was typically higher than 95%.
Quantitative real-time RT-PCR. For analysis of gene expression, total RNA was extracted using the mirVana RNA Isolation kit (Ambion/Thermo Fisher Scientific) and cDNA was synthesised using the PrimeScript RT Reagent kit (Takara Bio). Quantitative real-time RT-PCR (qRT-PCR) was performed using SYBR Premix Ex Taq (Takara Bio) on a Prism 7900HT system (Applied Biosystems). Primers used are listed in Supplementary Data 3.
Flow cytometry analysis. Cell suspensions were prepared from spleen or lymph node tissue by sieving and gentle pipetting. To prepare intratumoural cell suspensions, minced tumours were agitated in Hank's balanced salt solution containing 1 mg/ml collagenase IV, 100 µg/ml hyaluronidase and 2 U/ml DNAse IV (Sigma-Aldrich) at 20-25°C for 2 h, and strained through a 70-µm mesh (Greiner Bio-One). Non-specific labelling of Fc receptors was prevented by incubating the cells with anti-CD16/CD32 antibody (1:500, 93; BioLegend). Intracellular staining of FoxP3 and intranuclear staining were performed with the FOXP3 Fix/Perm Buffer Set (BioLegend) and True-Nuclear Transcription Factor Buffer Set (BioLegend), respectively, according to the manufacturer's protocol. The antibodies used are listed in Supplementary Data 4. Flow cytometry was performed on a FACSAria instrument with FACSDiva software (BD Biosciences) or Attune NxT (Thermo Fisher Scientific), and data were analysed with FlowJo v9 or v10 software (Tree Star).
Cell cycle analysis. For flow cytometry-based cell cycle analysis, cells were labelled with antibodies against cell surface antigens, fixed with 4% paraformaldehyde (Nacalai Tesque) for 15 min, and incubated in 10 µg/ml 4′,6-diamidino-2-phenylindole (Dojindo)/0.1% Triton X-100 (Nacalai Tesque) in phosphate-buffered saline (PBS) for 30 min at 20-25°C. Antigen-independent T cell receptor stimulation was performed as follows: T cells were isolated from the spleens of WT C57/BL6J mice 3 weeks after injection using the Pan-T Cell Isolation Kit II for mouse (Miltenyi Biotec). Isolated T cells were stained with 5 µM CFSE. CFSElabelled T cells were stimulated with Dynabeads Mouse T-Activator CD3/CD28 (Veritas Life Science) and co-cultured with purified splenic Mo-MDSCs in 96-well round-bottomed plates (BM Equipment). After 4 days, cells were labelled with antibodies against CD4 or CD8 (BioLegend), and the CFSE signal from live CD4 +or CD8 + -gated cells was analysed by flow cytometry. Dividing cells were identified as those showing diluted CFSE signals relative to unstimulated T cells.
RNA-sequencing. PMN-and Mo-MDSCs were purified from the spleens of SCTbearing WT and p16/p21-DKO mice. RNA isolation was performed using the mirVana RNA Isolation kit. Library preparation was performed using a TruSeq stranded mRNA sample prep kit (Illumina) according to the manufacturer's instructions. Sequencing was performed on an Illumina HiSeq 2500 platform in a 75-base single-end mode. Illumina Casava1.8.2 software was used for base calling. Sequenced reads were mapped to the mouse reference genome sequences (mm10) using TopHat v2.0.13 in combination with Bowtie2 ver. 2.2.3 and SAMtools ver. 0.1.19. The fragments per kilobase of exon per million mapped fragments (FPKMs) were calculated using Cuffnorm version 2.2.1.
Viral infection and establishment of stable cell lines. LLC and SCT cells were infected with retrovirus encoding mouse Cx3cl1 (in pMarX-puro) and mouse Cx3cl1 shRNA (in pRetrosuper-puro) or SV40LT antigen (in pMarX-puro), respectively. Infected cells were selected in 5 µg/ml puromycin-containing medium for 7 days, and the selected cells were maintained in medium containing 1 µg/ml puromycin (Wako Pure Chemical Industries).
BM-Mo-MDSC preparation. The protocol for BM-Mo-MDSC preparation was modified from that used in a previous report. Femurs were removed from C57BL/6 mice and the BM was flushed. Red blood cells were lysed with ammonium chloride. BM cells were adjusted to a concentration of 1 × 10 6 /ml in 10% FBS-containing RPMI1640 medium supplemented with 20 ng/ml mouse GM-CSF, 40 ng/ml mouse IL-6 and 2 ng/ml mouse IFN-γ (Peprotech). Cells were maintained at 37°C in a 5% CO 2 -and 3% O 2 -humidified atmosphere for 4-5 days. Non-adherent and loosely adherent cells were collected by gentle pipetting. In some experiments, BM-Mo-MDSCs were treated with flavopiridol HCl (Selleckchem) or HLM (Tocris Bioscience) along with GM-CSF, IL-6, and IFN-γ on day 4, and collected 24 h later.
In vitro kinase assay. One million Mo-MDSCs purified from bone marrow, spleen and tumour were lysed by cell lysis buffer (Cell Signaling Technology), and immunoprecipitated by anti-CDK2, and Protein A Dynabeads (Thermo Fisher Scientific). The antibodies are listed in Supplementary Data 4. For kinase reaction, precipitated CDK2 was incubated in 20 mM Tris-HCl (pH7.5) containing 5 ng/µl of recombinant SMAD3 (Sigma-Aldrich), 100 µM ATP (Promega), 0.1 mg/ml BSA, 1 mM DTT and 40 mM MgCl 2 at 30°C for 2 h. ATP to ADP conversion during kinase reaction was measured by ADP-Glo Kinase Assay (Promega) and Powerscan HT (DS Pharma Biomedical).
ChIP analysis. ChIP analysis was performed using EZ-ChIP (Millipore), Dynabeads Protein G (Thermo Fisher Scientific) and anti-Smad3 antibody (Abcam) or normal rabbit IgG (Cell Signaling Technology) as a negative control. Information on the antibodies used may be found in Supplementary Data 4. Immunoprecipitated DNA was quantified by qRT-PCR. For information on primers used, see Supplementary Data 3.
Luciferase reporter assay. The mouse Cx3cr1 gene promoter region (from −700 to +397) was amplified from the genome of a C57BL/6J mouse. PCR mutagenesis for 2 SBEs was performed using the primers listed in Supplementary Data 3. The amplified fragments were inserted into the pGL3basic vector (Promega). cDNAs for human CDK2 and CDK4 were obtained by PCR and subcloned into the pcDNA3. THP-1 (JCRB0112.1) cells were provided by RIKEN BioResource Center through the National BioResource Project of MEXT, Japan and cultured in RPMI1640 medium with 10% FBS, 10 mM HEPES buffer, 100 µM sodium pyruvate (Gibco/Thermo Fisher Scientific), 100 U/ml penicillin, 100 µg/ml streptomycin and 50 mM 2-mercaptoethanol (Wako Pure Chemical Industries). THP-1 cells were transfected with plasmids, using Amaxa Cell Line Nucleofector Kit V and Nucleofector I Device (Lonza); 48 h later, luciferase activity was measured with the Dual-Luciferase Reporter Assay System (Promega) according to the manufacturer's instructions. Reporter activity was normalised to that of Renilla luciferase derived from the co-transfected pCMV-Renilla vector. Data are representative of three independent experiments.
Western blotting. Cells were lysed in Laemmli's sample buffer. The samples were subjected to SDS-polyacrylamide gel electrophoresis followed by immunoblot analysis with the indicated antibodies and appropriate secondary antibodies conjugated with horseradish-peroxidase. Bound antibodies were visualised by chemiluminescence following incubation with Chemi-Lumi One L HRP Substrate (Nacalai Tesque) or Amersham ECL Prime (GE Healthcare). For information on the antibodies used, see Supplementary Data 4.
Statistical analysis. Student's t-test (two-tailed) was performed to evaluate the significance of differences between two groups.
Data availability. RNA-sequencing data have been deposited in the NCBI's Gene Expression Omnibus database (GEO GSE93359). All other remaining data are available within the article and supplementary files, or available from the authors upon request.