SHP-2 and canonical PD-1: SHP-2 axis regulate myeloid cell differentiation, anti-tumor responses and innate immune memory


 PD-1 checkpoint inhibitor induces T cell inactivation by recruiting SHP-2. However, T cell-specific SHP-2-deficient mice do not have improved anti-tumor immunity. We generated mice with conditional targeting of the Ptpn11 gene (encoding for Shp-2) in T cells (Shp2f/fLckCre) or myeloid cells (Shp2f/fLysMCre), and found that Shp2f/fLysMCre mice had diminished tumor growth. As determined by RNA-seq, this was paralleled by the presence of inflammatory neutrophils and tumor-associated macrophages (TAMs) with molecular signatures of enhanced differentiation, phagocytosis and antigen-processing and presentation. SHP-2 deficient TAMs also had increased monocyte and dendritic cell (DC) specification transcription factors, chemokine and cytokine production, and expression of immunostimulatory molecules that promote T cell recruitment and activation. Monocytes from tumor-bearing Shp2f/fLysMCre mice suppressed tumor growth after transfer to naïve recipients indicating development of innate immune memory. In bone marrow myelocytes, GM-CSF, induced PD-1 expression, phosphorylation and interaction with SHP-2, the Src family kinase Lyn, and GM-CSF receptor beta chain, indicating that the PD-1:SHP-2 axis targets a key pathway of myelocyte differentiation. In contrast, SHP-2 deletion or antibody-mediated blockade of the PD-1:PD-L1 pathway enhanced phosphorylation of the transcription factors HOXA10 and IRF8 that regulate myeloid differentiation and monocytic/moDC lineage commitment, respectively. Thus, SHP-2 and the PD-1:SHP-2 axis pose a signaling restrain to myeloid differentiation and monocyte lineage commitment resulting in a myeloid landscape that suppresses anti-tumor immunity.


Introduction
The immune system plays an important protective function against tumor development and progression 1 . To escape immunosurveillance, cancer cells have developed mechanisms that mask their immunogenic features, such as expression of ligands for inhibitory receptors that directly inhibit T cell responses by engaging immune inhibitory receptors such as programmed death-1 (PD-1) 2 . Tumors also alter myeloid cells, which constitute a significant cellular fraction of the microenvironment and can employ multiple mechanisms to actively suppress anti-tumor T cell responses 3 . Recent studies have shown that tumor-infiltrating myeloid cells also include immunostimulatory subsets [4][5][6][7] .
PD-1 is a T cell inhibitor and the most extensively exploited therapeutic target of cancer immunotherapy 16 . PD-1-mediated T cell inactivation is attributed to the function of SHP-2 phosphatase, which is activated by recruitment to PD-1 cytoplasmic tail [17][18][19][20] . Based on these established data, it was expected that SHP-2 ablation would abrogate the inhibitory pathway activated downstream of PD-1 receptor in T cells, and mice with SHP-2 ablation in T cells would neither be subjected to PD-1-mediated inhibition of anti-tumor immunity nor would they be benefitted from treatment with anti-PD-1 antibody. 4 However, it was found that T cell-specific SHP-2 deletion did not improve anti-tumor immunity and did not alter responses to anti-PD-1 treatment 21 , but might rather have a detrimental effect in melanoma progression and metastasis 22 . PD-1 is also expressed in common myeloid progenitors (CMPs) and granulocyte/macrophage progenitors (GMPs), which accumulate during cancer-driven emergency myelopoiesis and give rise to immunosuppressive MDSC and TAMs. In tumor-bearing mice with myeloid-specific PD-1 ablation, accumulation of MDSCs is prevented, while output of differentiated effector myeloid cells with monocytic lineage dominance is increased 23 . It is currently unknown whether a canonical PD-1:SHP-2 axis is operative in myeloid cells.
Temporal activation of SHP-2 is critical for myeloid cell fate 24 . Gain of function SHP-2 mutations, resulting in constitutive phosphatase activation, prevent myeloid differentiation and lead to the accumulation of immature myelocytes and development of leukemia 25 . HOXA proteins, members of the HOX family of transcription factors involved in normal hematopoiesis 26 , are targets of SHP-2 27 . HOXA genes are maximally expressed in committed myeloid progenitors and are critical for normal myeloid development, whereas their aberrations are associated with leukemia 28,29 . During growth factor-induced myelopoiesis, tyrosine phosphorylation of HOXA10 in two tyrosine residues in the HOXA10 homeodomain decreases its binding affinity for target gene promoters. This abrogates HOXA10-induced transcriptional repression and differentiation proceeds 30 . HOXA10 is a substrate of activated SHP-2, which dephosphorylates HOXA and prevents derepressing of HOXA target genes and myeloid cell differentiation 27,31 . In addition, SHP-2 regulates the function of IFN regulatory factor (IRF)-8, a transcription factor that has a decisive role in myeloid progenitor differentiation toward monocytes and DCs, while limiting granulocyte development 32,33 . Failure of IRF8 expression or function leads to the generation of MDSCs 34 . IRF8 phosphorylation on Tyr-95, localized in the conserved IRF domain, is mandatory for its nuclear translocation and initiation of gene transcription 35 . The cytoplasmic-nuclear 5 shuttling of IRF8 is regulated by SHP-2, which dephosphorylates IRF8 and prevents its nuclear localization and transcriptional activation 35,36 .
In the present study we examined how SHP-2 regulates the output of cancer-driven emergency myelopoiesis and investigated whether a canonical PD-1:SHP-2 axis is operative in myeloid cells. We used mice with conditional targeting of the Ptpn11 gene (encoding for SHP-2) that leads to significant SHP-2 depletion 37 , and targeted SHP-2 in T cells (Shp2 f/f LckCre) or myeloid cells (Shp2 f/f LysMCre).
Although no difference in tumor growth was observed between Shp2 f/f LckCre and control mice, Shp2 f/f LysMCre mice had significantly reduced tumor growth that was not further decreased by anti-PD-1 treatment. At a cellular level, myeloid-specific SHP-2 targeting skewed the myeloid cell fate in tumorbearing mice from MDSCs and converted them to differentiated myeloid cells that did not induce immunosuppression. As determined by RNA-seq, this was paralleled by the presence of inflammatory neutrophils and TAMs with molecular signatures of enhanced differentiation, phagocytosis, and antigen processing and presentation. SHP-2 deficient TAMs had increased expression of the monocyte specification transcription factors IRF8, Klf4 and Zeb2, enhanced molecular features of DC-producing monocytes (moDCs), chemokine and cytokine production, and expression of immunostimulatory molecules that promoted T cell recruitment and activation. Moreover, monocytes from tumor-bearing Shp2 f/f LysMCre mice suppressed tumor growth after transfer to naïve recipients indicating development of innate anti-tumor immune memory. In bone marrow myelocytes, GM-CSF, a mediator of cancer-driven emergency myelopoiesis, induced PD-1 expression, phosphorylation and interaction with SHP-2, the Src family kinase Lyn, and GM-CSF receptor beta chain (c), indicating that the PD-1:SHP-2 axis targets a key pathway of myelocyte differentiation. In contrast, SHP-2 ablation or antibody blockade of the PD-1/PD-L1 pathway enhanced GM-CSF-mediated phosphorylation of the transcription factors HOXA10 and IRF8 that regulate leukocyte differentiation and monocytic/moDC lineage commitment, respectively.
Thus, SHP-2 and the PD-1:SHP-2 axis pose a signaling restrain in myeloid differentiation and 6 monocyte/DC lineage commitment thereby preventing the generation of innate anti-tumor immune memory and resulting in a myeloid landscape that suppresses anti-tumor immunity.

Myeloid-specific SHP-2 targeting suppresses tumor growth and modifies the immunosuppressive properties of MDSC.
To dissect the role of PD-1:SHP-2 axis in anti-tumor immune responses, we used mice with conditional targeting of the Ptpn11 gene (encoding for SHP-2), thereafter named Shp2 f/f , that significantly depletes SHP-2 expression 37 . We crossed Shp2 f/f mice with mice expressing Cre recombinase under the control of the lysozyme (LysM) promoter to induce selective depletion of the Ptpn11 gene in myeloid cells (Shp2 f/f LysMCre) or with mice expressing Cre recombinase under the control of the distal Lck promoter to induce selective depletion of the Ptpn11 gene in mature T cells (Shp2 f/f LckCre). For our studies, first, we used the murine B16-F10 melanoma tumor model because it has been informative in dissecting mechanisms of resistance to checkpoint immunotherapy 38 and also induces cancer-driven emergency myelopoiesis and output of MDSC 23 . Although no difference in tumor growth was observed between Shp2 f/f LckCre and control Shp2 f/f mice, Shp2 f/f LysMCre mice had significantly reduced tumor growth (Fig. 1A). Because CD8 + T cells are the mediators of anti-tumor responses, we assessed activation of CD8 + T cells in tumor draining lymph nodes (TDLN) that have been previously determined to be the targets of PD-1-mediated inhibition of anti-tumor T cell function 39,40 . CD8 + T cells in TDLN of Shp2 f/f LysMCre mice displayed a more activated state compared to their counterparts in Shp2 f/f LckCre and control Shp2 f/f mice (Fig. 1B, C; gating strategy, Supplementary Fig. S1). In contrast, no difference was observed in the activation of TDLN CD8 + T cells from Shp2 f/f LckCre mice compared to control Shp2 f/f tumor-bearing mice (Fig. 1B, C).
We examined the functional properties of MDSCs isolated from mice bearing B16-F10 tumors and found that MDSCs isolated from Shp2 f/f LysMCre mice had significantly diminished immunosuppressive capacity compared to their counterparts isolated from Shp2 f/f LckCre and Shp2 f/f mice 8 (Fig. 1D). Consistent with this finding, these cells also had lower expression of CD38 (Fig. 1E, F, G), a hallmark of immunosuppressive MDSC 41 . Thus, SHP-2 ablation switches the fate and function of myeloid cells away from immunosuppressive MDSC in mice bearing B16-F10 tumors.
Improved anti-tumor function induced by myeloid-specific SHP-2 deficiency is not further enhanced by PD-1-blocking immunotherapy.
Previously, we determined that PD-1 is expressed in myeloid progenitors and immature myeloid cells in tumor bearing mice, whereas PD-1 ablation switches the fate of myeloid cells away from immunosuppressive MDSC toward inflammatory monocytes and DC 23 . Based on these previous observations and our present findings, we sought to examine whether a PD-1:SHP-2 axis might be operative in myeloid cells, and whether it might be involved in the suppression of anti-tumor immunity.
To investigate this, first, we examined the therapeutic effect of PD-1 blocking immunotherapy in our experimental mice bearing B16-F10 tumors. Compared to Shp2 f/f mice receiving control IgG2a, Shp2 f/f mice treated with anti-PD-1 antibody displayed a significant reduction of tumor growth ( Fig. 2A, C). In contrast, Shp2 f/f LysMCre mice that had diminished tumor growth compared to Shp2 f/f control tumor bearing mice (Fig. 2C), did not benefit from treatment with anti-PD-1 antibody compared to treatment with control IgG2a (Fig. 2B, C). In addition, expression of CD38 in MDSC was diminished by anti-PD-1 antibody treatment in tumor-bearing Shp2 f/f control mice but not in Shp2 f/f LysMCre mice, which expressed lower levels of CD38 than Shp2 f/f tumor-bearing mice, in both IgG2a and anti-PD-1 treatment groups (Fig. 2D, E, F, G). Compared to the Shp2 f/f tumor-bearing group, Shp2 f/f LysMCre tumor-bearing mice exhibited enhanced activation and increased recruitment of T effector (TEF) CD4 + and CD8 + T cells (Fig. 2H, I, J, K). Treatment with anti-PD-1 antibody increased the activation of CD8 + T cells (Fig. 2H, I) and expansion of T effector cells (TEF) CD4 + and CD8 + T cells in Shp2 f/f but not in Shp2 f/f LysMCre tumorbearing mice, which had higher TEF cells in TDLN in both IgG2a and anti-PD-1 treatment groups (Fig. 9 2J, K). These results suggest that SHP-2 in myeloid cells has an important role in mediating effects of PD-1 blockade and, for this reason, myeloid-specific SHP-2 ablation improves anti-tumor immunity which is not further improved by PD-1 blocking immunotherapy. Moreover, SHP-2 expression in myeloid cells has dominant role over SHP-2 expression in T cells in regulating systemic anti-tumor immunity.

Myeloid-specific SHP-2 deficiency alters the differentiation fate of tumor infiltrating myeloid cells, and induces T cell recruitment and activation.
To study the effects of SHP-2 ablation in myeloid cells in more detail, we employed the MC17-51 fibrosarcoma mouse tumor model that is well-established to induce robust cancer-mediated emergency myelopoiesis, leading to significant output of bone-marrow-derived MDSCs and TAMs 42 . Consistent with our findings in mice bearing B16-F10 tumors ( Fig. 1A and Fig. 2C), we observed significantly diminished tumor growth in Shp2 f/f LysMcCe mice bearing MC17-51 tumors, compared to control Shp2 f/f mice (Fig. 3A, B, C, D, Supplementary Fig. S2), but not in Shp2 f/f LckCre mice ( Supplementary Fig. S3).
In the mouse, MDSCs consist of two major subsets, CD11b + Ly6C hi Ly6Gmonocytic (M-MDSC) and CD11b + Ly6C lo Ly6G + polymorphonuclear (PMN-MDSC) 8 . These cells have similar morphology and phenotype to normal monocytes and neutrophils but distinct genomic and biochemical profiles 43 . In humans, in addition to M-MDSC and PMN-MDSC, a small subset of poorly characterized early stage MDSC (eMDSC) has been identified 8 . We did not observe quantitative differences in tumor-infiltrating myeloid cells (Fig. 3E), but Shp2 f/f LysMCre mice had an increased fraction of M-MDSC in tumors ( Shp2 f/f LysMcre mice, which also developed smaller tumors ( Supplementary Fig. S4A-E). In the TDLN of MC17-51 tumor-bearing Shp2 f/f LysMCre mice, there was a significant increase of CD4 + and CD8 + TEF 10 cells (Fig. 3I) and higher activation of CD8 + T cells (Fig. 3J, K). Additionally, we also observed a systemic increase of TEF CD4 + cells, increase of TEF and T central memory-like (TCM) CD8 + T cells (Fig. 3L) and enhanced activation of CD4 + and CD8 + T cells (Fig. 3M, N). No differences were noted in the expression of checkpoint receptors including PD-1, PD-L1, CTLA-4, TIGIT or the inducible costimulator (ICOS) in CD4 + and CD8 + T cells ( Supplementary Fig. S5) or in the numbers and activation of Treg in TDLNs ( Supplementary Fig. S6). Thus, myeloid-specific Shp2 ablation leads to an increased tumor infiltration by Ly6C hi monocytes and concomitant recruitment and activation of TEF and TCM cells.

SHP-2 deficiency converts MDSC to differentiated and activated leukocytes.
Next, we sought to investigate further the properties of MDSC isolated from Shp2 f/f LysMcre mice. Using mice bearing MC17-51 tumors, we observed that PMN-MDSC from MC17-51 tumor-bearing    Over-representation and Gene Set Enrichment Analysis (GSEA) revealed that relative to those isolated from control tumor-bearing hosts, gene expression profiles of myeloid cells from Shp2 f/f LysMCre tumor-bearing mice were enriched for processes involved in positive regulation of cytokine production, PRC2-EZH2 targets, phagosome formation, myeloid cell differentiation, and neutrophil-mediated immunity, as well as Notch signaling and HOXA targets, and were dominated by functions of antigen 12 presentation, myeloid cell migration, mitosis and autophagy ( Fig. 4O, P, Q, and Supplementary Fig. S8).
Consistent with these, transcripts of the inflammatory mediators S100a8/9, IFN I-inducible genes Rsad2    Table 2). There was also increase in gene transcripts of DC specification such as the transcription factor Baft3 (Fig. 5B, H; Supplementary Table 2), which together with Irf8 and Klf4 is essential for DC differentiation 33,[53][54][55] . TAMs from Shp2 fl/fl LysMCre mice also had increased expression of the immunostimulatory molecules CD86 and CD83 (Fig. 5B, I;   Supplementary Table 2), which are associated with moDC differentiation 45,56 . These findings suggest that TAMs in Shp2 fl/fl LysMCre mice have enhanced gene expression programs identifying monocytes differentiated from MDPs that can give rise to Ly6C hi classical and moDC-producing monocytes 45 . This is also supported by the finding that signature transcripts of moDCs such as Csf1, Siglec1, Clec10 and  Table 2). Enhanced metabolic activity, characterized of glucose and glutamine metabolism, increased levels of amino acids, and anabolic lipid metabolism was also observed in phagocytes generated from bone marrow progenitors of Shp2 f/f LysMCre mice ( Supplementary Fig. S10). Consistent with these findings, GSEA using Gene Ontology Biological Processes Pathways gene sets showed that TAMs from Shp2 f/f LysMCre mice were characterized by highly expressed signatures of leukocyte activation, chemotaxis, migration, cytokine production, inflammatory response, and tissue remodeling ( Supplementary Fig. S11).

SHP-2 ablation induces innate immune memory and long-lasting anti-tumor properties in monocytes.
Our RNA-seq studies showed that SHP-2 deficient neutrophils in tumor-bearing mice have a gene program of effector myeloid cells enriched in gene transcripts involved in neutrophil differentiation and specification (Fig. 4; Supplementary Table 1) indicating that they are originated from GMPs 44,45 . In parallel, our RNA-seq studies of SHP-2 deficient TAMs showed enhanced expression of genes characteristic of monocytes differentiated from MDPs that can give rise to Ly6C hi classical and moDCproducing monocytes 45 . These findings suggest that altered tumor-driven emergency myelopoiesis in Shp2 f/f LysMCre mice results in increased output from two different progenitor stages, GMPs and MDPs We implanted MC17-51 tumors in Shp2 fl/fl LysMCre mice and, nine days later, collected CD45 + CD11b + Ly6C hi Ly6Gmonocytes and CD45 + CD11b + Ly6C -Ly6G + neutrophils from the bone marrow of tumor-bearing mice and adoptively transferred them into naïve WT tumor bearing hosts implanted with MC17-51 tumors (Fig 6B), as previously described 79 . Compared to control tumor-bearing mice, no difference in tumor growth was observed in mice that received neutrophils from Shp2 f/f LysMCre tumor bearers whereas the monocyte recipient group had significantly reduced tumor growth (Fig. 6C, D).
These results show that the enhanced anti-tumor immunity in Shp2 f/f LysMCre is mediated by Ly6C hi monocytes generated in the bone marrow during tumor-driven emergency myelopoiesis. Moreover, the anti-tumor properties of SHP-2 deficient Ly6C hi monocytes are long-lasting and can be transferred to naïve tumor-bearing hosts.
SHP-2 and canonical PD-1:SHP-2 signaling impede phosphorylation of HOXA10 and IRF8 transcription factors that promote myeloid cell differentiation and monocyte lineage commitment.
Next, we sought to determine the mechanistic basis of the immunological outcomes induced by myeloid-specific SHP-2 targeting. Activating SHP-2 mutations prevent myeloid cell differentiation and lead to the accumulation of immature myelocytes and development of leukemia 25 . Established SHP-2 targets in myeloid progenitors include the transcription factors HOXA and IRF8. During growth factorinduced myelopoiesis, as differentiation proceeds, tyrosine phosphorylation of HOXA10 decreases its binding affinity for target gene promoters that are repressed by HOXA10 binding. SHP-2-mediated dephosphorylation of HOXA prevents derepressing of HOXA target genes and myeloid cell differentiation 27,31 . SHP-2 also regulates the function of IRF8 because SHP-2-mediated dephosphorylation of IRF8 prevents its nuclear localization and transcriptional activation, which is required for monocyte lineage differentiation 35,36 . Our RNA-seq studies a showed that myeloid cells isolated from tumor-bearing Shp2 fl/fl LysMCre mice had enhanced neutrophil differentiation and increased HOXA targets (Fig. 4), whereas TAMs exhibited highly elevated expression of transcripts relevant to monocyte and mo/DC differentiation including Irf8, Kl4, Zeb2 and Batf3 (Fig. 5B). Based on these findings, we examined whether phosphorylation of HOXA10 and IRF8 might be altered in SHP-2 deficient bone marrow myeloid cells cultured with GM-CSF+IL-3 27,31 . Flow cytometric analysis confirmed that by 72hr of culture >95% of the Lin pos cells were myelocytes ( Supplementary Fig. S12A).
HOXA10 immunoprecipitation followed by immunoblot with anti-phosphotyrosine antibody showed enhanced HOXA10 phosphorylation in SHP-2 deficient myelocytes (Fig. 7A). Similarly, IRF8 immunoprecipitation followed by anti-phosphotyrosine immunoblot showed that SHP-2 deficient cells had increased IRF8 phosphorylation (Fig. 7B). These results provide a mechanistic basis explaining how SHP-2 ablation promotes the generation of differentiated myeloid cells that is mediated by enhanced HOXA10 phosphorylation, and enhances monocytic lineage output that is regulated by phosphorylation of IRF8, which drives the transcriptional program of monocyte and DC lineage differentiation. Because as we observed previously, after culture with GM-CSF myeloid progenitors and their progeny express PD-1 23 , we examined whether PD-1:SHP-2 interaction might be detected in myelocytes. PD-1 immunoprecipitation followed by SHP-2 immunoblot detected a robust PD-1:SHP-2 interaction in control cells that was abrogated in SHP-2 deficient myelocytes (Fig. 7C).
Our previous studies showed that GM-CSF-mediated activation of Erk and mTOR is enhanced in PD-1 deficient CMPs and GMPs during in vitro culture 23 88 . We examined if PD-1 is phosphorylated by GM-CSF at tyrosine Y248 within the conserved ITSM motif of PD-1 cytoplasmic tail, a site known to be phosphorylated by Src family kinases leading to PD-1 interaction with SHP-2 in T cells 20,86 . Immunoprecipitation with anti-PD-1 antibody followed by immunoblot with phospho-specific PD-1 antibody that recognizes pY248 20,89 , 18 showed that GM-CSF induced PD-1 phosphorylation and interaction with SHP-2 (Fig. 7M). Sequential immunoblot with Lyn-specific antibody and GM-CSFR c-specific antibody showed that both proteins were detected in PD-1 immunoprecipitations (Fig. 7M). Thus, a canonical PD-1:SHP-2 axis is activated in myeloid cells during GM-CSF-mediated signaling, resulting in the recruitment of PD-1 inhibitory machinery to a major signaling receptor involved in myeloid cell activation, proliferation and differentiation.
Because our results showed PD-1 and PD-L1 upregulation and PD-1:SHP-2 interaction in myeloid cells, we examined whether activation of PD-1:SHP-2 axis by PD-1 and PD-L1 receptor-ligand interaction during culture with GM-CSF might pose a signaling restrain to GM-CSF-mediated phosphorylation of HOXA10 and/or IRF8 and whether this might be prevented by PD-1:PD-L1 blockade, similarly to SHP-2 depletion. Addition of a PD-L1 blocking antibody during culture enhanced HOXA10 phosphorylation ( Fig. 7N) but more potently enhanced IRF8 phosphorylation (Fig. 7O). Thus, SHP-2 and, to a lesser extent, the PD-1:SHP-2 axis pose a signaling restrain in myeloid cell differentiation and lineage fate commitment to monocytes and DCs that depends on IRF8 expression and phosphorylation 33,55 .

Discussion
Our present studies reveal a previously unidentified mechanistic role of SHP-2 and the PD-1:SHP-2 axis in regulating myeloid cell differentiation, and lineage fate commitment and function in the context of cancer. We found that SHP-2 poses a restrain in GM-CSF-mediated phosphorylation of HOXA10 and IRF8 transcription factors that induce myeloid differentiation and monocyte/DC lineage commitment, respectively 90,91 . In the myeloid lineage, the c subunit of the GM-CSFR, shared also by the IL-3 and IL-5 receptors, is the major signaling subunit that is tyrosine phosphorylated by cytokine stimulation and mediates myelocyte differentiation, expansion and activation in response to growth factors 87 . Although JAK2 is the primary kinase activated downstream of GM-CSFR, the Src family member Lyn, can directly interact with GM-CSFR c subunit 88 . We found that PD-1 is phosphorylated by GM-CSF at tyrosine Y248 within the conserved ITSM motif of its cytoplasmic tail, a site known to be phosphorylated by Src family kinases leading to PD-1 interaction with SHP-2 in T cells 20,86 . Moreover, co-engagement of the PD-1:PD-L1 pathway during growth factor engagement of GM-CSFR restrains GM-CSF-mediated phosphorylation of HOXA10 and IRF8, which is reversed by a PD-L1 blocking antibody. These results explain why PD-1 23 or SHP-2 ablation in myeloid cells, as shown in the present study, lead to enhanced monocyte and moDC differentiation. T cells are responsive to chemokines whose production is controlled by type I IFN signaling produced by innate immune cells 97 . In most cases, this is initiated by viral infection of a cell and detection of pathogenic nucleic acids by intracellular DNA-and RNA-sensing systems such as cGAS/STING, AIM2, and TLR3 98 . IFN type I is also produced by DCs following stimulation of TLR7 and TLR9, which detect viral RNA and DNA molecules, respectively, that have been endocytosed or sequestered by autophagy. The production of type I IFNs in DC can be further amplified by a positive feedback loop that involves the transactivation of IRF7 and IRF8 in response to IFNα/β receptor (IFNAR) signaling 99 .
Although it was initially thought that type I IFNs exert direct anticancer effects by activating IFNAR signaling in malignant cells hence inhibiting cell cycle progression 100 , promoting terminal differentiation 101 or inducing apoptosis 102 , it is becoming increasingly clear that type I IFNs mainly function by stimulating anticancer immune responses via autocrine or paracrine signaling circuits by IFN-stimulated genes 103,104 . These have an instrumental role in the production of IFN-dependent chemokines that lead to preferential recruitment and retention of systemic CD8 + T cells in the tumor, an outcome that can be recapitulated by treatment with type I IFN or DNA damaging chemo-and radiotherapies 105 . In our work, we found that SHP-2 deficient TAMs displayed a significant increase of STING (Supplementary Table   2), TLR signaling (Fig. 5S), and type I IFN (Fig. 5T), and these were paralleled by increase of chemokines Previous studies used allosteric inhibitors of SHP-2 such as SHP099, TNO155 106-108 and RMC4550 109 with the purpose to target cancer cells, in which SHP-2 is activated downstream of RTK/Ras signaling and functions as an oncogene promoting cancer cell growth 106 , or to target both cancer and immune cells 108,109 . These studies observed upregulation of three CXCR3 ligands, specifically CXCL9, 22 CXCL10, CXCL11 in tumor cells thereby leading to T cell recruitment 108 . Other investigators reported that SHP-2 ablation in macrophages potentiated production of CXCL9 in response to IFN-, thereby facilitating tumor infiltration by T cells 110 . Combined approaches of SHP-2 allosteric inhibitors together with pharmacologic RTK inhibitors or KRAS G12C -GDP inhibitors have been employed with the main purpose of targeting signaling vulnerabilities of cancer 108,111,112 . These studies have reported that such treatments, by changing the properties of cancer, also alter immune cells of the TME, including increase of CD8 + T cells, depletion of pro-tumorigenic M2 TAMs via attenuation of Csf1 signaling 109 , increase of M1 TAMs and reduced immunosuppression function of MDSCs 107,108,112 , but the mechanisms remained elusive. Our present studies employing a genetic approach to dissect the effects of SHP-2 targeting at multiple levels, showed that myeloid-intrinsic but not T cell-intrinsic depletion of SHP-2 has a dominant anti-tumor effect by altering the myeloid lineage fate, thereby promoting the generation of differentiated granulocytes, and monocytes that became TAMs with signatures of activated antigen presenting cells. In SHP-2 deficient TAMs, we observed enhanced expression of Csf1, together with Slamf7, Siglec1, and Clec10 that form a signature of moDCs 57 . Notably, our results showed that classical M1 but also M2 genes, such as PPAR and CD206, were increased in SHP-2 deficient TAMs (Supplementary Table 2).
Because M2 macrophages are required for resolution of inflammation and tissue remodeling 12,[113][114][115] , the combined actions of M1 and M2 macrophages induced by SHP-2 depletion might be involved in the enhanced anti-tumor function by promoting proinflammatory but also tissue remodeling properties of tumor infiltrating macrophages that induce a pro-resolving-mediated suppression of tumor growth 115 .
An important finding of our studies was the SHP-2 targeting in myeloid cells resulted in longlasting anti-tumor properties of monocytes isolated from the bone marrow of Shp2 f/f LysMCre tumorbearing mice that could be transferred to naïve tumor-bearing hosts, consistent with the development of trained immunity. Until recently, it was assumed that the innate immune system cannot build immunological memory. The observations that innate immune cells can exert adaptive characteristics and 23 can build a non-specific memory, challenged this paradigm [116][117][118] . Cells of the innate immune system, such as monocytes, can build memory during their differentiation to macrophages, a process named trained (innate) immunity in which signaling, metabolic and epigenetic changes represent the molecular substrate 119 . Neutrophils may also develop trained immunity and provide long-lasting protection against cancer 79 . Emergency myelopoiesis is an integral part of trained immunity because this process develops during the differentiation of myeloid progenitors in the bone marrow 120,121 . Our results showed that SHP-2 deficiency resulted in enhanced signaling of bone marrow myeloid progenitors in response to factors of emergency myelopoiesis (Fig. 7), altered metabolic programs of TAMs and bone marrow derived phagocytes ( Fig. 5X; Supplementary Fig. S10), and led to increased expression of Ezh1 and Kdm5 epigenetic regulators in TAMs, findings consistent with signaling, metabolic and epigenetic processes associated with trained immunity 76,119,122 . By transferring individually neutrophils and monocytes from the bone marrow of Shp2 f/f LysMCre tumor-bearing mice subjected to cancer-mediated emergency myelopoiesis in vivo, we found that monocytes but not neutrophils displayed properties of long-lasting innate immune memory against tumor transferrable to naïve tumor-bearing hosts. Of note, IL-1 that drives long-term training of monocyte precursors 121 , was highly increased in SHP-2 deficient TAMs, providing evidence for an additional mechanism responsible for the development of trained immunity in monocytes. Together with our findings that SHP-2 and the PD-1:SHP-2 axis restrain the phosphorylation of HOXA10 that regulates myelocyte differentiation, but also the phosphorylation of IRF8 that guides monocyte/DC lineage fate commitment, these data indicate that SHP-2 and PD-1:SHP-2 axis might preferentially have a long-lasting functional impact in the monocytic compartment. It is tempting to speculate that the long-lasting effects of PD-1-based immunotherapy in some, but not all, patients might be related to the development of immunotherapy-mediated monocyte differentiation and trained immunity vs. generation of immunosuppressive MDSCs and TAMs during cancer-driven emergency myelopoiesis.
Further studies will investigate these new directions of central regulation of anti-tumor responses to cancer immunotherapy in patients. 24 In conclusion, our results provide multiple levels of evidence that SHP-2 and the PD-1:SHP-2 axis pose a signaling restrain to the differentiation and monocyte/DC lineage commitment of myeloid progenitors by suppressing the phosphorylation of the transcription factors HOXA10 and IRF8, resulting in a myeloid landscape that compromises anti-tumor immunity. SHP-2 depletion promotes myelocyte differentiation, activation and antigen presenting function leading to T cell recruitment and activation, and potent anti-tumor immunity.

Cell purification and processing
Single cell suspensions were made from spleens, tumor draining lymph nodes, and tumors as previously described 23 . Briefly, tumors were digested by 1 mg/ml of Collagenase I in incomplete RPMI and then

Immunoprecipitation and immunoblotting
To prepare lysates, cells were washed in PBS and lysed as previously described 123

RNA sequencing and analysis
For RNA sequencing, PMN-MDSCs (CD11b + GR1 hi Ly6G + ) were isolated from the spleens of MC17-51 fibrosarcoma bearing and Shp2 f/f LysMCre and Shp2 f/f mice by magnetic bead isolation. TAMs were isolated from the same mice by cell sorting after using the Live/Dead Fixable Far Read Dead Cell Stain kit (ThermoFisher; L34973) to identify live cells, staining with antibodies specific for CD45, CD11b, F4/80 and Ly6G and gating on CD45 + CD11b + F4/80 + Ly6Glive cells. Total RNA was isolated from the 29 cells using the Qiagen RNeasy Mini Kit (Qiagen, cat# 74104). For each sample, 400 ng of total RNA was then used in Illumina's TruSeq Stranded mRNA Library kit (Cat# 20020594) for polyA mRNA isolation and library construction. Libraries were sequenced on Illumina NextSeq 500 as paired-end 42-nt reads (Active Motif). Raw sequencing reads were quality-checked using FastQC (v0.11.5) 124 and data were pre-processed with Cutadapt (v2.5) 125 for adapter removal following best practices 126 . Gene expression quantification was performed by aligning against the GRCm38 genome using STAR (v2.7.3a) 127

Metabolite analysis
Polar metabolites were quantitatively profiled by a positive/negative ion-switching, targeted liquid chromatography tandem mass spectrometry (LC-MS/MS) based metabolomics platform using a 5500 QTRAP hybrid triple quadruple mass spectrometer (AB/SCIEX) via selected reaction monitoring (SRM) as described previously 136 . Briefly, Lin neg bone marrow cells cultured with G-CSF and GM-CSF (40 ng/ml each) for 48 hours using triplicate samples for each condition and sample type. After methanol extraction using 80% (v/v) methanol (-80 o C) was carried out, pellets were lyophilized using a SpeedVac concentrator using no heat. 20 L of LC/MS grade water were added to resuspend each sample just before LC-MS/MS analysis and 5 L of sample were injected onto the autosampler of the LC system (Shimadzu) using an amide HILIC column (Waters). Once the SRM data for ~285 metabolites were acquired, peaks 30 were integrated using a software platform for peak area integration MultiQuant 2.1 (AB/SCIEX). Data analysis was performed using online MetaboAnalyst 3.0 software.

Statistics
Statistical analysis for comparison between two groups was determined by Student's t test, while for comparison between three or more groups, ANOVA (*p value <0.05, **p value <0.01, ***p value <0.001, **** p value <0.0001) calculated with Prism software (GraphPad Software).                  Immunoprecipitation was performed with agarose-conjugated anti-PD-1 antibody followed by SDS-PAGE and immunoblot with the indicated antibodies. The abundance of GMCSFR, Lyn, SHP-2 and PD-1 phosphorylated at Y248 (pPD-1) coprecipitated with PD-1 (left panels) was normalized to immunoprecipitated PD-1 and was expressed as fold change over the value obtained in non-stimulated cells at the zero timepoint (defined as 1). Expression of indicated proteins in whole cell lysates was also examined (right panels). (N-O) Bone marrow cells from wild type C57BL/6 mice were cultured for 48h in the presence of GM-CSF (10 ng/ml) and IL-3 (5 ng/ml), alone or with anti-PD-L1 blocking antibody (MIH5) (10 µg/ml). Cells were lysed and immunoprecipitation was performed with indicated agaroseconjugated antibodies followed by SDS-PAGE and immunoblot with the indicated antibodies. (N) The abundance of phosphorylated HOXA10 (assessed by anti-pY blot) was normalized to immunoprecipitated HOXA10 and was expressed as fold change over the value obtained in cells cultured without PD-L1 37 blocking antibody (defined as 1). Expression of actin in whole cell lysates was also examined as input.

Supplementary Materials
(O) The abundance of phosphorylated IRF8 (assessed by anti-pY blot) was normalized to immunoprecipitated IRF8 and was expressed as fold change over the value obtained in cells cultured without PD-L1 blocking antibody (defined as 1). Expression of actin in whole cell lysates was also examined as input.