TIM landscape varies significantly among cancer types and treatment groupsTo begin assessing TIM heterogeneity across cancers, we conducted scRNA-seq on tumor samples from 129 patients across eight distinct cancer types: basal cell carcinoma (BCC), breast cancer, clear cell renal cell carcinoma (ccRCC), colorectal cancer (CRC), hepatocellular carcinoma (HCC), head and neck squamous cell carcinoma (HNSCC), intrahepatic cholangiocarcinoma (iCCA), and melanoma. Of these, five cancer types (BCC, ccRCC, CRC, HNSCC, and melanoma) had available clinical information regarding treatment response status (Fig. 1a and Supplementary Data 1). Quality control measures were implemented to remove potential debris, damaged cells, and doublets. Batch effect was corrected and assessed before and after the correction process using the LISI index16 (Supplementary Fig. 1a-b). A total of 47,750 myeloid cells were included for subsequent analysis (Supplementary Fig. 1c). Precise categorization of cells was achieved through manual annotation utilizing well-defined signature markers6,14,17. Within the eight cancer types, we identified 12 subtypes of macrophages and monocytes, four subtypes of cDCs, and a distinct group of mast cells (Fig. 1b-c and Supplementary Data 2). We differentiated three macrophage subtypesmainly by their expression of FOLR2 and APOE. Other markers were also enhanced, including APOC1 and TREM2 in Macro_FOLR2-APOE+ and Macro_FOLR2 + APOE + ; SELENOP in Macro_FOLR2 + APOE-; and GPNMB, CCL18, CCL13, C1QC, and C1QA in Macro_FOLR2 + APOE+ (Fig. 1c). Two clusters of cDCs were clearly separated from the plasmacytoid dendritic cell (pDC) cluster based on exclusive expression of LILRA4, GZMB, and IL3RA (Fig. 1c). HLA-DQA1 was expressed by both cDC clusters, with higher average expression in the cDC_CLEC9A cluster as previously observed6. Mast cell-specific markers were also exclusively expressed by this cell population (Fig. 1c).Fig. 1: Demographics overview of the pan-cancer myeloid atlas.a Sample numbers of each cancer type collected; numbers shown in the brackets indicated the sample number in terms of its percentage share in the whole atlas. b UMAP representation of the myeloid atlas. c Expression of the signature markers in each cell type. d The proportion of cell types across cancer types in treatment-naïve and post-treatment groups.We further characterized the functionality of these 17 identified cell types (Supplementary Fig. 2a). Macro_APOC1 + IFI27 + , Macro_ISG15, and Macro_OLFML3 exhibited functional differences despite their high expression of CXCL9 and CXCL11. Macro_APOC1 + IFI27+ primarily engaged in GTPase-activating protein binding, Macro_ISG15 was predominantly enriched in cytokine receptor binding and Macro_OLFML3 may participate in fatty acid synthase activity. Additionally, within the three macrophage types defined by FOLR2 and APOE, Macro_FOLR2 − APOE+ was mainly associated with “protein-lipid complex binding,” while Macro_FOLR2 + APOE− and Macro_FOLR2 + APOE+ were respectively linked to “protein serine/threonine/tyrosine kinase activity” and “phosphatidylinositol binding” (Supplementary Fig. 2a).Considering the signature markers for M1/M2 macrophages11,18, Macro_FOLR2 + APOE+ was mainly enriched for M1 markers and Macro_LYVE1 for M2 markers, while Macro_ISG15, Macro_NLRP3, and Macro_FOLR2 + APOE+ were enriched for both. Conversely, Macro_OLFML3, Macro_APOC1 + IFI27 + , Macro_FOLR2-APOE + , and Macro_FOLR2 + APOE- did not show enrichment for either M1 or M2 markers (Supplementary Fig. 2b).We also utilized the Role index to analyze the preferential distribution of the 17 cell types between responders and non-responders to ICB therapy, finding that Macro_ISG15 (associated with interferon), Macro_FOLR2 + APOE+ (with high CXCL9 expression), and cDC_CLEC9A (high IDO1 expression) exhibited a preference for the responder group (Supplementary Fig. 2a–c). Surprisingly, we observed a preference for the non-responder group in cDC (CD1C) and cDC3_LAMP3 (Supplementary Fig. 2c). This suggests that although DCs are involved in antigen presentation, an increase in their abundance may represent increased anti-tumor cytotoxicity and immune awakening. Overall, the above analysis highlights the limitations of the M1 and M2 classification in capturing macrophage heterogeneity, emphasizing the need for a more accurate classification system that incorporates the diversity and dynamic functional characteristics of macrophage subsets.The samples from the eight cancer types were categorized based on their treatment status, including treatment-naïve (Pre), post-treatment (Post), treatment non-responders (NR), and treatment responders (R). Subsequently, we compared the distribution of cell types between each pair of stratified groups to assess any variations (Fig. 1d, e and Supplementary Fig. 3a, b). In the Pre group, cDC (CD1C) was the predominant cell type present in iCCA and breast cancer. Macro_NLRP3 accounted for the largest proportion in BCC and ccRCC. In the Post group, pDC_LILRA4 became the main cell type in both BCC and melanoma, while Macro_LYVE1 was predominant in HCC and iCCA (Fig. 1d). Notably, mast cells remained the major cell type in CRC in both the Pre and Post groups. Mast cells were consistently the major cell type in CRC across the four stratified groups (Fig. 1d). Several cancer types exhibited significant decreases in the proportions of Macro_ISG15 and Mono_INHBA after treatment, while Macro_FOLR2 − APOE + , Macro_APOC1 + IFI27 + , Macro_LYVE1, and Macro_OLFML3 showed significant increases in the Post group (Fig. 1d, e and Supplementary Fig. 3a-b). Comparing the response groups, the proportions of Macro_NLRP3 and Mono_CD14 significantly increased in the R group across many cancer types, whereas no obvious proportional changes were observed in the NR group (Fig. 1d, e and Supplementary Fig. 3c, d). We also analyzed the expression of immunoinhibitors and immunostimulators across various cell types, finding that Macro-FOLR2-APOE+ exhibited minimal expression of immune checkpoint genes, whereas Macro-FOLR2 + APOE+ displayed high expression levels of several immunoinhibitors and immunostimulators, including CTLA4 and CD274 (PD-L1) (Supplementary Fig. 3e).This analysis collectively highlights notable variations in the composition of TIMs among different cancer types, emphasizing the heterogeneity of these cells across various cancers, treatment statuses, and response groups.Treatment and response statuses are associated with different TIM kinetic profilesThe TME is a complex milieu consisting of diverse cell types that undergo continuous dynamic shifts19,20,21. To investigate temporal changes and developmental patterns within TIMs, we conducted a comprehensive analysis of their kinetic profiles at the pan-cancer level by examining the pseudotime trajectories (a measure of differentiation progression) of each individual cell type. We compared Pre vs. Post and NR vs. R samples. Correlation analysis revealed that the TME underwent significant remodeling following immunotherapy. There were notable changes in the intercellular correlations, including a substantial increase in the correlation between Mono_CD14 and Macro_NLRP3 after immunotherapy (Supplementary Fig. 4a). We next employed Slingshot22 to determine the topological heterogeneity of the differentiation trajectories of macrophages/monocytes and DCs (Supplementary Fig. 4b). Following immunotherapy, distinct alterations were observed in the average pseudotime of specific cell types. Mono_CD14, Macro_NLRP3, Mono_CD16, and Macro_IER3 exhibited significant shortening of their average pseudotime, which would suggest a less differentiated state in the post-treatment. Conversely, pDC_LILRA4 showed a notable delay in trajectory differentiation following immunotherapy (Fig. 2a, b). Meanwhile, cDC (CD1C), cDC_LAMP3, mast cells, Macro_LYVE1, and Macro_FOLR2 + APOE+ displayed minimal changes in average pseudotime between Pre and Post samples (Fig. 2a, b). Mono_CD14, Macro_IER3, Macro_NLRP3, Macro_ISG15, and cDC_CLEC9A displayed considerable delays in pseudotime in the NR compared to the R group, while no substantial kinetic differences were detected in mast cells, Macro_IFI27, and Macro_OLFML3 in these groups (Fig. 2b, c). Interestingly, we observed more cell types with large kinetic changes in the R vs. NR groups than in the Pre vs. Post groups (Fig. 2a, b), suggesting a potential transitional difference between homeostatic and non-homeostatic differentiation in different disease statuses. Notably, Macro_FOLR2 + APOE+ displayed an expedited pseudotime in the R group, in stark contrast to Macro_FOLR2-APOE+ and Macro_FOLR2-APOE- that displayed the reverse pattern despite expressing only one of the markers (Fig. 2a, b).Fig. 2: Pseudotime transitions across TIMs.a Average pseudotime of each cell type in the treatment groupsand the response groups. b Violin plots showing the distribution of Pseudotime ordering for responders and non-responders in each cell type. Data are presented as mean values ± SEM (Standard Error of the Mean), indicated by the red dots and associated error bars. The box plots within the violin plots represent the interquartile range (IQR), with the center line indicating the median. The whiskers extend to the minimum and maximum values within 1.5 times the IQR from the 25th and 75th percentiles. Statistical significance was determined using two-sided t-test. Multiple comparison adjustment was carried out using BH FDR. P-values for each comparison between responder and non-responder groups in each cell type are labeled in the plots, shown under the names of the cell types as “P = (p-value)”. c Enriched pathways in each of the pseudotime branches for the three categories of cells, namely Macro/Mono (representing macrophages and monocytes), DC, and Mast. d Normalized enrichment scores (NES) of the enriched hallmark pathways in each pseudotime branch.Our next aim was to elucidate the prevailing biological states within stratified disease cohorts. To achieve this, we classified cells based on their kinetic states and established correlations between the differentially expressed genes in these clustered kinetic profiles and the hallmark gene sets from the Molecular Signatures Database (MSigDB)23. We observed distinct branching of pro- or anti-inflammatory signals (Fig. 2c, d and Supplementary Data 3). Within the macrophage and monocyte (Macro/Mono) populations of the R group, the majority of pseudotime states exhibited down-regulation of pro-inflammatory pathways (Fig. 2c), such as IFN-α/γresponse, TNF signaling via NF-κB, and oxidative phosphorylation24 (States 1 and 2), indicating the presence of an anti-inflammatory signaling milieu. However, State 3 exhibited an upregulation of TNF signaling, suggesting that a minority of cells still promote pro-inflammatory signaling (Fig. 2c, d, Supplementary Fig. 4b, and Supplementary Data 3). Conversely, in the NR group of Macro/Mono populations, the majority of the cells existed in a pro-inflammatory state (State 3), including cells demonstrating elevated mTORC1 signaling25.Similarly, within the DC subpopulations, two states in the R group (States 2 and 3) exhibited anti-inflammatory properties, while State 1 displayed a pro-inflammatory milieu (Fig. 2c, d). In State 1, the majority of cells were identified as cDC (CD1C), with a small subset represented by pDC_LILRA4 (Supplementary Fig. 4 and Supplementary Data 3). Conversely, in NR, pro-inflammatory oxidative phosphorylation was observed in States 2 and 3. State 4 demonstrated upregulation of the IFN-α/γ response (Fig. 2c, d), which was downregulated in States 1 and 5 (Fig. 2c, d). The pro-inflammatory State 4 primarily consisted of cDC (CD1C), alongside a small subset of pDC_LILRA4, mirroring the observations made in the R group (Supplementary Fig. 4b). Meanwhile, the kinetic states of mast cells revealed distinct clusters in the R and NR groups (Fig. 2c, d, Supplementary Fig. 4b and Supplementary Data 3). In the R group, co-upregulation of both anti- and pro-inflammatory signaling was observed within the state clusters. However, in the NR group, a clear separation between pro-inflammatory (States 1 and 4) and anti-inflammatory state clusters (States 2 and 5) was discerned (Fig. 2c, d). These varying kinetic profiles indicate the plasticity of TIMs in the TME and suggest their potential roles as either pro- or anti-pathogenic cell types in the context of immunotherapy.Distinct TIM regulatory states characterize specific cancersWe subsequently analyzed gene-regulatory networks to identify active regulatory elements (REs) within TIMs from the R and NR groups. By assessing the transcriptional activity of transcription factors (TFs) in each cell type, we measured correlations between TIMs across different cancer types and response statuses. As shown in Fig. 3a, cell types originating from the same cancer and exhibiting the same response status clustered together. We also observed a close relationship between each R and NR group with their respective cancer types, indicating the high cancer-specificity of REs in TIMs in relation to immunotherapy response. We then identified TF modules contributing to this clustering phenomenon (Fig. 3b and Supplementary Data 4). Notably, elevated YBX1 activity was observed in Macro_FOLR2 + APOE + , Macro_FOLR2 + APOE-, Macro_APOC1 + IFI27 + , cDC(CD1C), Macro_OLFML3, and cDC_CLEC9A in responders with CRC, while BCL3 activity was increased in Mono_CD14, Macro_NLRP3, Mono_INHBA, Macro_OLFML3, and Macro_ISG15 in non-responders with CRC. Intriguingly, the identified TFs were predominantly associated with macrophages. For instance, YBX1 depletion in mouse macrophages has been associated with augmented tissue damage, myofibroblast activation, and fibrosis. MAFB activity was consistently decreased in the R vs. the NR groups across various tumor-infiltrating immune cell categories, including Macro/Mono, DC, and Mast categories (Fig. 3c). MAFB was previously found to promote inflammation in classical TIMs and EMT in lipid-associated TIMs26,27. Finally, within the Macro/Mono and DC categories, the pro-inflammatory TF JUNB19 was less active in the R compared to the NR group (Fig. 3d).Fig. 3: Cell-type-specific regulon activity landscape.a Correlation of the cell types across cancer types in the response groups based on TF activity. b TF modules forming the correlation patterns in (a). c TF with common activity patterns across cancer types and across cell types in the response groups. For each cancer type and regulon, we performed two-tailed paired t-test to compare the activity scores of Post(R) and Post(NR) of each regulon across 17 different cell types, with no technical replicates involved. d TFs with common activity and expression patterns across cancer types in the response groups of Macro/Mono and DC categories.Prospective targets of cell-cell interaction in ICB responseThe intricate dynamics of cell-cell interaction within the TME involve not only TIMs but also other immune cell populations, particularly T cells. We examined both intercommunication between TIMs (Fig. 4a, b, Supplementary Figs. 5, 6, and Supplementary Data 5) and the interactions between TIMs and CD4/CD8 T cells (Fig. 4a, b, Supplementary Figs. 5, 6, and Supplementary Data 5) and the interactions between TIMs and CD4/CD8 T cells (Fig. 4c, d, Supplementary Figs. 7–9, and Supplementary Data 6), utilizing the same set of collected samples. Specifically, we analyzed 272,017 T cells from the post-treatment samples and annotated 12 subtypes of CD4 T cells and 15 subtypes of CD8 T cells based on their distinctive molecular signatures (Supplementary Fig. 7).Fig. 4: Cell-cell interaction patterns across cell types in response groups.a Cell-cell interactions between TIMs. Line thickness represents the degree of interactions. b Fold change of interaction frequencies between the R (numerator) and NR (denominator) groups. c Cell-cell interactions between TIMs and CD4 or CD8 T-cells. Line thickness represents the degree of interactions. d Fold change of TIMs and CD4 or CD8 T cells interaction frequencies, between the R (numerator) and NR (denominator) groups.We applied rigorous criteria for selecting ligand-receptor interactions, considering those with an aggregate rank <0.01 and cell-specific interaction frequencies exceeding the 75th percentile of overall frequencies for each sample condition (see Methods). As shown in Fig. 4a, b and Supplementary Fig. 5, noticeable differences emerged in the interactions between TIMs within the R and NR groups. Notably, Mono_CD16, Macro_IER3, Macro_FOLR2 + APOE + , Macro_FOLR2 + APOE-, and cDC_LAMP3 exhibited substantially higher interactions in the NR group across all cell types, whereas Macro_FOLR2-APOE+ displayed elevated fold ligand interactions with other TIMs in the R group. Additionally, we frequently observed HLA-A/B/C to LILRA1/3 interaction pairs specific to the NR group, demonstrating increased interaction fold changes across various cell types (Supplementary Fig. 6). As previously reported28,29,30,31, this interaction may indicate a myeloid deactivation and reduced immune response.Subsequently, we investigated interactions between TIMs and specific CD4/CD8 T cell subsets. Compared to TIMs-to-TIMs interactions, fewer myeloid cell types exhibited large fold changes (absolute fold change > 2) in TIMs-to-CD4/CD8 T cell interactions (Fig. 4c, d, Supplementary Data 6, and Supplementary Fig. 7). Nonetheless, we observed high interactions of pDC_LILRA4 with CD4 T cells; of Mono_CD16, Macro_IFI27, Macro_FOLR2 + APOE-, and Macro_FOLR2-APOE+ with CD8 T cells; and of Mono_INHBA, Macro_FOLR2 + APOE + , and cDC_LAMP3 with CD4/CD8 T cells (Fig. 4d and Supplementary Fig. 8). The HLA to LAG3 interaction cluster was exclusively present in the ligand interactions of pDC_LILRA4 with CD4(IFNG+ Tfh/Th1) and of Mono_CD16 with mainly CD8(GZMK+ Tex), CD8(GZMK+ Tem), and CD8(ISG + T) in the NR group (Supplementary Fig. 9). Regarding Macro_FOLR2 + APOE+ receptor interactions with CD4 T cells, we observed the unique presence of RPS19 to C5AR1 interaction in the NR group, which has previously been shown to suppress antitumor immune response through the production of immunosuppressive cytokines like TGF-β32. CD99 to PILRA/CD81 interactions were also substantially increased in the NR group, CD99 and CD81 have been associated with increased tumor migration, invasion, and metastasis33,34, while PILRA delivers inhibitory signals related to natural killer (NK) cell and DC activation35 (Supplementary Fig. 9). Although interactions between TIMs and CD4 T cells are less studied, their positive involvement in pro-tumoral functions can be inferred. We identified unique interactions among members of the TNFRSF (i.e., TNFSF10 with TNFRSF11B and LTB with CD40) in the receptor interactions of cDC_LAMP3 with CD4/CD8 T cells in the NR group (Supplementary Fig. 9).In our analysis of cell types exhibiting higher fold change in the R group, we observed significant ligand interactions involving Mono_INHBA, Macro_APOC1 + IFI27 + , and Macro_FOLR2 − APOE+ with specific subsets of T cells. Notably, HLA to LAG3 interactions were present, primarily interacting with CD4(IFNG+ Tfh/Th1), CD8(GZMK+ Tem), and CD8(Terminal Tex) cells. Additionally, RPS19 to C5AR1 interactions were found in the R group for Mono_INHBA receptor interactions with T cells (Supplementary Fig. 9). The significance of these interaction pairs in relation to response status remains unclear. However, they hold promise as potential targets for further investigation of the plasticity of TIMs and the reciprocal involvement of interactions in both response groups across myeloid-to-T cell interactions.Highly interactive TIMs exhibit cancer-specific regulatory patternsTo comprehensively understand the transcriptional regulatory profile of myeloid cells in the context of immunotherapy, we examined the regulatory patterns in cell types that exhibited differential cell-cell interactions between the R and NR groups. Our focus was on the TF signatures of highly interactive TIMs.We observed differential expression of distinct regulon modules in specific cell types. For instance, regulon module cluster c9 was uniquely expressed by pDC_LILRA4, while c11 and c3 were expressed by cDC_LAMP3. Additionally, several distinct TFs specifically regulated the associated TIM response class (Fig. 5a). In the case of pDC_LILRA4, regulons from cluster c3 and from c6 (excluding REST) were uniquely expressed by the NR group. Comparatively, ELF2, RARA, and TAF1 from c12 displayed distinctive expression patterns in the R group. Other unique signatures differentiating the R and NR groups were predominantly found in the R group, including cluster c4 for Macro_APOC1 + IFI27+ and c13 for Mono_INHBA. For other cell types, the signature differences between R and NR regulatory expressions were marked by one or more TFs that were not clustered in a regulon module. These regulatory signatures may represent key regulators responsible for the variation in cellular interactions between the R and NR groups of the same TIM type. Aside from the regulon modules, we further identified common TFs across cancer types that regulated in the same direction when comparing the R and NR groups (Fig. 5b). Notable examples include FOSB in Macro_FOLR2 + APOE+, MAFF and SPI1 in cDC_LAMP3, and MAFB in Macro_FOLR2-APOE+.Fig. 5: Targeted analysis of the highly interacting TIMs.a Standardized expression values of the TF signatures of the nine TIMs in the response groups are shown in the plot. Color gradient indicates the standardized expression values of the TFs. b TFs with common activity patterns across cancer types in the response groups of these cell types. For each cell type and regulon, we performed two-tailed paired t-test to compare the activity scores or the expression values of the regulon between Post(R) and Post(NR) across cancer types. The analysis included five different cancer types, with no technical replicates involved. c Survival analysis based on the expression of the signature genes of these cell types. Log-rank test was used to compare the survival distributions within each survival plot. No multiple testing correction was carried out. d Response index and GSEA index separating the samples of the two response groups. Each point represents a sample in the study with response information. e Multiplex immunofluorescence staining has delineated the spatial relationships between the Macro_FOLR2 + APOE+ cell subpopulation and T cells, as well as their abundance in ICB responders and non-responders among CRC patients.Different TIMs are associated with survival in different cancersTo evaluate the impact of individual stratified cell types on patient prognosis, we conducted survival analysis using post-immunotherapy cohorts including the IMvigor210 urothelial carcinoma (UC)36, ccRCC37, and skin cutaneous melanoma (SKCM)38. Patients were categorized based on the expression of signature markers for the highly interactive TIMs. Our stratification revealed that high expression of cDC_LAMP3 markers correlated with lower survival, whereas high expression of Macro_FOLR2 + APOE-, pDC_LILRA4, and Macro_APOC1 + IFI27+ markers was associated with improved survival outcomes (Fig. 5c). These markers hold potential for predicting survival outcomes following immunotherapy. However, certain cell types, including Mono_INHBA, displayed opposite survival relationships in different cancer types emphasizing the cancer-specificity and macrophage heterogeneity observed in relation to survival outcomes (Fig. 5c).After identifying the prominent interaction of HLA-A/B/C with LILRA1/3, primarily within Mono_CD16 cells of the NR cohort, we aimed to elucidate the predictive capability of these interactions. To achieve this, we evaluated the collective expression of HLA-A/B/C, LILRA1, LILRA3, LILRB2, and the signature markers of Mono_CD16 to quantify the cumulative impact of these factors on patient prognosis. The difference in survival outcomes between the two groups expressing Mono_CD16 signature markers was not significant in the UC and ccRCC cohorts (considering LILRB2 as one of the signature markers for Mono_CD16), however, when considering the additional expression of HLA-A and/or LILRA1, the higher-expression group demonstrated significantly lower survival rates in both cohorts (Supplementary Fig. 10). Conversely, higher expression of these markers was associated with better survival rates in the SKCM cohort Consequently, although HLA-A/B/C and LILRA1/LILRA3/LILRB2 were commonly observed in non-responder TIMs, their association with post- ICB patient survival appears to be cancer-specific.Construction of response index and GSEA index which accurately determined treatment responseWe subsequently conducted gene ontology (GO) pathway analysis using the top upregulated differentially expressed genes (DEGs) of each TIM cell type in each R group to unravel the underlying changes associated with treatment response. The results revealed distinct immune-related pathway differences between the R and NR groups. In the R group, Macro_FOLR2 + APOE+ was involved in the NK-mediated immune response; Mono_CD16 showed negative regulation of tumor-promoting NF-κB activity and upregulation of macrophage tolerance induction, which protects the host from chronic exposure to inflammatory mediators39; and Mono_INHBA demonstrated positive regulation of both IL-4 and IL-1αproduction, which together trigger anti-tumoral Th9 cell differentiation40 (Supplementary Fig. 11). Conversely, in the NR group, Macro_FOLR2-APOE + , Macro_FOLR2 + APOE-, and Macro_APOC1 + IFI27+ were associated with apoptotic cell clearance, and Macro_IER3 was involved in the regulation of anti-tumoral IL-15 and IL-12 production. Identifying and understanding the roles of individual TIMs under different response conditions would facilitate the targeting of deleterious factors to improve conditions within the R group.We next aimed to develop an index specifically designed to predict the response to ICB therapy. This index is based on the enriched pathways in each cell type under different response conditions and on the proportion of each cell type within these conditions for each cancer type. To construct the index, we conducted extensive gene set enrichment analyses (GSEAs) across MSigDB signature categories, encompassing all cell types across different cancers within each categorized group. This allowed us to identify cancer-specific and response-specific signatures for each cell type. By considering the cancer-specific proportions of cell types and the enrichment scores of the identified signatures, we created a treatment response index that clearly distinguished between ICB responders and non-responders at index = 0 (Fig. 5d). In certain cancer types, such as HNSCC and CRC, distinct clusters of non-responders and both responders and non-responders, respectively, were observed at negative and positive response indices. However, other cancer types displayed a wide dispersion of response indices, indicating greater heterogeneity in treatment response. Notably separation based on the response index was not possible in melanoma due to the absence of identified melanoma-specific signatures. This highlights the importance of considering cancer-specific characteristics when assessing treatment response and the variable predictive performance of the response index across different cancer types.Macro_FOLR2 + APOE+ cells are more abundant in ICB-responsive CRC patientsThe density of FOLR2+ macrophages in tumors has been correlated with survival in breast cancer and HCC patients15,41; therefore, the significant role we observed for Macro_FOLR2 + APOE+, and its interaction with T cells, is notable. To corroborate the alterations in cell type composition within the realm of immunotherapy, we conducted immunofluorescence analysis utilizing samples obtained from ICB responder and non-responder CRC patients. This revealed a higher abundance of Macro_FOLR2 + APOE+ cells in the responders compared to the non-responders (Fig. 5e). We also observed the presence of CD4 and CD8 T cells in close proximity to the Macro_FOLR2 + APOE+ cells. Conversely, we did not detect any Macro_FOLR2 + APOE+ cells in the non-responders, indicating a positive association between these cells and treatment response. Furthermore, to confirm the changes in key markers related to myeloid cells before and after neoadjuvant immunotherapy for pan cancer, we collected samples of immunotherapy responses and non-responders for HNSC and CRC. Six markers that showed increased expression levels after treatment in the sequencing results were selected for single channel immunofluorescence staining, including ISG15 (macrophages), NLRP3 (macrophages), IER3 (macrophages), and CD14 (monocytes), CD16 (monocytes), and LAMP3 (DC) (Supplementary Fig. 12). The conclusion related to these three types of myeloid cells can be verified through immunofluorescence chromosomes to confirm the results of multi omics sequencing analysis: Macro_ ISG15, Macro_ NLRP3, Macro_ IER3, Mono_ CD14, Mono_ CD16 and cDC_ The expression level of LAMP3 increases in reactive tumor tissues, and these biomarkers may be one of the response mechanisms of pan cancer immunotherapy, and may also provide a basis for clinical monitoring and efficacy prediction. We will continue to collect more types of samples for verification in the future. These findings align with our overall pan-cancer results and highlight the significance of this macrophage subtype in tumor prognosis.
