Cancer immunology data engine reveals secreted AOAH as a potential immunotherapy

Lanqi Gong; Jie Luo; Emily Yang; Beibei Ru; Ziyang Qi; Yuma Yang; Anshu Rani; Abhilasha Purohit; Yu Zhang; Grace Guan; Rohit Paul; Trang Vu; Zuojia Chen; Renyue Ji; Chi-Ping Day; Chuan Wu; Glenn Merlino; David Fitzgerald; Grégoire Altan-Bonnet; Kenneth Aldape; Jiansheng Wu; Xinyuan Guan; Peng Jiang

doi:10.1016/j.cell.2025.07.004

. Author manuscript; available in PMC: 2026 Jul 28.

Published in final edited form as: Cell. 2025 Jul 28;188(18):5062–5080.e32. doi: 10.1016/j.cell.2025.07.004

Cancer immunology data engine reveals secreted AOAH as a potential immunotherapy

Lanqi Gong ^1,^†, Jie Luo ^2,^†, Emily Yang ¹, Beibei Ru ¹, Ziyang Qi ², Yuma Yang ², Anshu Rani ¹, Abhilasha Purohit ¹, Yu Zhang ³, Grace Guan ¹, Rohit Paul ⁴, Trang Vu ¹, Zuojia Chen ⁵, Renyue Ji ⁶, Chi-Ping Day ¹, Chuan Wu ⁵, Glenn Merlino ^7,^§, David Fitzgerald ⁸, Grégoire Altan-Bonnet ⁹, Kenneth Aldape ¹⁰, Jiansheng Wu ¹¹, Xinyuan Guan ^2,^*, Peng Jiang ^1,^12,^*

PMCID: PMC12434699 NIHMSID: NIHMS2096169 PMID: 40730154

Summary

Secreted proteins are central mediators of intercellular communications and can serve as therapeutic targets in diverse diseases. The ~1903 human genes encoding secreted proteins are difficult to study through common genetic approaches. To address this hurdle and, more generally, to discover cancer therapeutics, we developed the Cancer Immunology Data Engine (CIDE, https://cide.ccr.cancer.gov), which incorporates 90 omics datasets spanning 8575 tumor profiles with immunotherapy outcomes from 17 solid tumor types. CIDE systematically identifies all genes associated with immunotherapy outcomes. Then, we focused on secreted proteins prioritized by CIDE without known cancer roles and validated regulatory effects on immune checkpoint blockade for AOAH, CR1L, COLQ, and ADAMTS7 in mouse models. The top hit, AOAH (Acyloxyacyl Hydrolase), potentiates immunotherapies in multiple tumor models by sensitizing T-cell receptors to weak antigens and protecting dendritic cells through depleting immunosuppressive arachidonoyl phosphatidylcholines and oxidized derivatives.

Graphical Abstract

graphic file with name nihms-2096169-f0001.jpg

In brief

The Cancer Immunology Data Engine, integrating 90 immuno-oncology omics datasets, identifies genes that regulate immunotherapy efficacy. AOAH, the top hit secreted protein, potentiates T-cell receptors and dendritic cells by depleting arachidonoyl phosphatidylcholines and oxidized derivatives.

Introduction

Secreted proteins such as cytokines, growth factors, and soluble enzymes mediate broad functions, including intercellular signaling and immune response. Despite a long research history, anticancer therapies based on secreted proteins have many limitations¹. For example, few patients demonstrate an objective response to engineered IL2². Similarly, anti-cytokine therapeutics often fail to bring clinical benefits, as in the case of anti-VEGF in Glioblastoma³ and anti-TGFβ/PDL1 in triggering hyperprogressions⁴.

Underlying these therapeutic limitations is a lack of systematic knowledge of secreted protein functions in cancer. The human genome contains ~1903 genes encoding secreted proteins⁵, but 61% of these genes have no reported cancer-associated function in the literature⁶. Most basic and therapeutic research on secreted proteins focuses on about 100 well-known cytokines⁷, such as IL2, IL15, VEGF, or TGFβ¹, leaving most of the secreted repertoire unexplored. This bias stems largely from limitations in the approaches used to study secreted proteins in physiological settings.

Pooled genetic screens using CRISPR, shRNA, or open reading frame technologies have been the primary approach to revealing gene functions⁸. However, such pooling strategies are ineffective for studying secreted proteins because nearby cells without the gene knockout can still secrete the protein and compensate for the perturbation of the secreted protein in a neighboring cell⁹. An alternative to pooled approaches is arrayed genetic screens, which interrogate secreted protein functions with in vitro assays^10–12. However, these approaches lack physiological context, as secreted proteins typically interact with other factors in the tissue ecosystems.

These limitations suggest that a new strategy is needed to identify secreted protein regulators and therapies in cancer. Here, we developed a data-driven solution using a large amount of omics data from immuno-oncology clinical studies¹³. The primary function of many secreted proteins, such as cytokines, is modulating immunological response⁷. Therefore, we hypothesized that omics data from clinical studies of cancer immunotherapies can provide a data-driven path to revealing secreted protein functions. We built the Cancer Immunology Data Engine (CIDE) and prioritized secreted proteins predicted to promote or inhibit immunotherapy response in multiple tumor types. Validations of CIDE predictions revealed AOAH as a potential immunotherapy.

Results

A data-integrative framework for immuno-oncology research

We created the Cancer Immunology Data Engine (CIDE, https://cide.ccr.cancer.gov). CIDE incorporates 90 omics datasets spanning 8575 bulk-tumor pretreatment profiles from 5957 patients with immunotherapy outcomes from 17 solid tumor types (Figure 1A and Table S1). While much of the data is available to the public, about half of the immunotherapy clinical datasets analyzed here are not publicly accessible, even after the publication of the associated papers (summarized in Table S1). We negotiated with pharmaceutical companies, research labs, and hospitals to release sequencing data and de-identified patient information from their published papers for our analysis. Although the raw data for these papers remains unavailable except by request from the original authors, CIDE publicly presents the associations between molecular readouts and clinical outcomes from these datasets with interactive query modules. CIDE includes gene expression, copy number alteration, somatic mutation, and DNA methylation data (Figure 1A). Our in-house framework processed all data uniformly with quality filters (Figure S1A–C). To our knowledge, CIDE harbors the most comprehensive immuno-oncology data among existing databases (Table S2).

Figure 1. — (A) Omics data from pretreatment tumors with cancer immunotherapy outcomes. From the center outward: the core shows the fraction of patients with each clinical endpoint; the middle layer shows the number of patients for each data type; the outer layer shows the fraction of patients from each cancer type. The outside table shows the dataset counts for each data type, with the column color key the same as the middle layer. RECIST: Response Evaluation Criteria in Solid Tumors.

(B) CIDE prioritization workflow. First, users input a gene set (Input 1) and select omics data cohorts (Input 2). Data types are colored as in panel A. Numbers indicate patient counts. CIDE then ranks input genes based on the median gene risk scores across selected cohorts (Output 1). Users then select candidates (Input 3) for dissecting cell-lineage expression based on single-cell (sc) RNA-seq datasets (Output 2 part 1), cell-type-specific associations between scRNA-seq gene expression and immunotherapy outcomes (Output 2 part 2), CRISPR screen data in cancer cells under immunological killing pressure (Output 3), and links to the DepMap database⁴¹ for genetic screens in cancer cells and Tres database¹⁷ for CRISPR screens in T cells. “Sub” represents a sub-component of anti-tumor immunity, such as Prf1-null T cells, supernatants from Cancer - T co-cultures, or IFNγ.

(C) Performance for binary response prediction, shown as the Area Under the receiver operating characteristic Curve (AUC). Each dot represents a pretreatment tumor cohort with counts labeled beside each box plot. The biomarker order follows the average performance rank across all endpoint types (this panel and Figure S1D). The bottom and top of the boxes are the 25th and 75th percentiles, respectively (interquartile range). Whiskers encompass 1.5 times the interquartile range. The black dotted line represents 0.5 as the random expectation. The red dotted line represents 0.7 as a good performance.

(D) CRISPR screen phenotype scores from a co-culture between B16F10 cells and pmel T cells⁴². Each gene’s phenotype score was computed as the log2 fold-changes (log2FC) between treatment and control groups. The absolute log2FC values, representing the phenotype strength, are shown with violin plots that present distributions smoothed by a kernel density estimator. The comparisons were through the two-sided Wilcoxon rank-sum tests.

(E) Histograms of Wilcoxon rank-sum z-scores comparing phenotype scores between secreted versus intracellular protein-coding genes. The Wilcoxon rank-sum tests were done as panel D across all CRISPR screen datasets. The comparisons between z-scores and zero were through the two-sided Wilcoxon signed-rank tests.

Users can input any gene, and CIDE outputs risk scores calculated as the association between gene values (expression, mutation, copy number, or promoter methylation) and clinical outcomes across omics cohorts (Table S3). Users can also input multiple genes and choose to query any subset of clinical omics cohorts, single-cell transcriptomics, and CRISPR screens in cancer models (Figure 1B and Table S4). CIDE then ranks input genes by scores from selected cohorts and prioritizes gene candidates (Figure 1B).

The CIDE data collection enabled us to evaluate previously reported immunotherapy response biomarkers systematically across datasets (Methods). When assessed across a large set of omics data, none of the 21 biomarkers we collected from 11 publications could achieve reliable predictions (Figure 1C and Figure S1D). This result indicates limited utility beyond the datasets or cancer types used in the original studies.

Risk scores predict anti- and pro-tumor effects of secreted proteins

As discussed in the introduction, genetic screening approaches may perform poorly in studying secreted proteins. We analyzed 1,190 CRISPR knockout pooled screens and indeed found that perturbation of secreted proteins resulted in significantly fewer phenotypes than intracellular proteins (Figure 1D, E, and Figure S1E). Thus, an alternative approach is necessary to reveal secreted protein functions in cancer.

For all human genes covered in CIDE, we computed risk associations between the expression levels and cancer immunotherapy outcomes (Methods, Examples in Figure 2A and Table S3) while adjusting for clinical covariates that we can curate from each publication and include in regressions (Table S5). We then focused on 1903 genes encoding secreted proteins for further analysis. 136 genes were prioritized based on their median risk score and Wilcoxon signed-rank tests across 50 genome-wide bulk transcriptomics cohorts from pretreatment tumors (Figure 2B for the primary endpoint of each cohort, Figure S2A, B for all available endpoints, Methods). 70 genes were significantly associated with favorable immunotherapy outcomes, while 66 were associated with adverse outcomes (Table S6). Among the results were well-known regulators of cancer immunotherapy response, such as IFNG¹⁴, FLT3LG¹⁵, and CXCL13¹⁶ (Figure 2B).

Figure 2. — (A) Example survival plots. The y-axis presents the fraction of patients with survival higher at each timepoint (x-axis) for tumors whose gene expression has high or low values, with cutoff determined by the best separation. Z-scores and P-values were evaluated by the two-sided Wald test in the Cox-PH regression using continuous values without any cutoffs.

(B) Gene risk scores. We collected 50 pretreatment tumor transcriptomics cohorts (columns) with diverse types of patient immunotherapy outcomes (markers on the bottom). The risk score of each gene (rows) quantifies the association between the expression level and outcome. The heatmap includes only the top 66 secreted proteins with significant median risk scores across cohorts (Methods). The complete list of 136 genes prioritized is available in Table S6. The absolute median risk is shown on the left. We annotated pro-tumor versus anti-tumor functions (solid colors) by a blinded literature search. Transparent colors represent follow-up annotations for genes not assigned labels in the blinded annotation.

(C) Confusion table of CIDE predictions on blinded annotations. The p-value was from the two-sided Fisher’s exact test.

(D) Prediction performance of secreted protein functions based on the confusion table in panel C. Accuracy_coin: baseline value by random predictions. F1_coin: baseline value by predicting all genes as anti-tumor or pro-tumor.

(E) Tumor-resilient T-cell (Tres) scores among prioritized secreted proteins. X-axis: Median CIDE risk scores across immunotherapy cohorts in panel B. Y-axis: Median scores across single-cell datasets analyzed by Tres¹⁷. P-values were from the two-sided Wilcoxon signed-rank test comparing Tres values across single-cell datasets against zero.

Different meta-analysis approaches lead to similar gene ranks (Figure S2C). Thus, we choose the simplest method, the Wilcoxon signed-rank test, which leads to a conservative estimation of significant hits compared to other methods (Figure S2D). Risk score profiles across cohorts are inconsistent, even for the same cancer type, treatment type, and endpoint (Figure S2E). For this reason, our gene prioritization method utilized group statistics across all cohorts rather than statistics from any single cohort.

Gene risk scores represent associations, not causality. To explore causality, we shuffled the 136 identified genes, removed the “favorable” or “adverse” labels, and recruited a thusly blinded colleague to annotate each gene as “pro-” or “anti-” tumor based on publications that perturbed each gene and measured effects on tumor progression (Figure 2B left color bars and Table S6). Median gene risk scores output by CIDE across immunotherapy cohorts reliably predicted pro-tumor versus anti-tumor effects of identified genes (Figure 2C, D).

Predicted secreted regulators of immunotherapy outcomes

Encouraged by the high prediction performance of gene risk scores on anti-tumor versus pro-tumor functions of secreted proteins (Figure 2C and D), we focused on genes without any known function in cancer to identify secreted regulators of tumor immunity (Stars in Figure 2B). AOAH, CR1L, and COLQ rank the top three candidates with negative risk associations with immunotherapy outcomes, while ADAMTS7 ranks first for positive risk (Figure 2B). According to our tumor-resilient T cell (Tres) model¹⁷, AOAH and COLQ have significantly positive Tres scores (Figure 2E), indicating that CD8 T cells with high AOAH or COLQ expression are more resilient in solid tumors than those with low levels.

The expression levels of these candidates are significantly associated with favorable (AOAH, CR1L, COLQ) or adverse (ADAMTS7) cancer immunotherapy outcomes in diverse solid tumor types (Figure 3A and Figure S3A). Single-cell RNA-seq data show that these four candidates are expressed in diverse cell lineages (Figure S3B). AOAH has the most significant associations with immunotherapy outcomes in multiple solid tumor types (Figure 3A), such as pancreatic (Figure 2A), melanoma, renal, and hepatocellular carcinomas (Figure S3C). Tumors with high AOAH expression have high cytotoxic T-cell infiltration levels (Figure S3D and E). Consistent with AOAH and COLQ having positive Tres scores (Figure 2E), therapeutic T cells infused into cellular immunotherapy responders^18,19 have higher AOAH and COLQ expression levels than those from non-responders (Figure 3B and Figure S3F).

Figure 3. — (A) Risk scores for prioritized secreted proteins. Each dot represents one clinical cohort (n = 50). The color represents cancer types. The shape represents endpoint types. The boxplot format is the same as in Figure 1C. P-values were computed using the two-sided Wilcoxon signed-rank test comparing group values and zero.

(B) *AOAH* and *COLQ* expression in infusion samples from anti-CD19 CAR T responders and non-responders with chronic lymphocytic leukemia¹⁸. Each dot represents one patient, shown in box plots as panel A. P-values were computed through the two-sided Wilcoxon rank-sum test.

(C) Average *AOAH* expression across detection spots in spatial transcriptomics of hepatocellular carcinoma treated with anti-PD1^20,21, shown as in panel B.

(D) Example of spatial transcriptomics in hepatocellular tumors²⁰.

(E) *In vivo* validation of prioritized secreted proteins on anti-PD1 response. The mean and standard error of the mean (sem) were shown at different time points. P-values were computed using the two-sided Wilcoxon rank-sum test at the last time point with complete tumor size measurements.

(F) Endpoint-free survival plots for all mice in panel E. P-values were computed using the two-sided log-rank test.

(G) Validation of protein secretion effects by inoculating a mixture of gene-OE and vector-OE tumor cells into mice, shown as in panel E. P-values were computed using the two-sided Wilcoxon rank-sum test, comparing the gene or gene-vector mixture against vector groups.

(H) Endpoint-free survival plots for all mice in panel G, shown as in panel F. P-values were computed using the two-sided log-rank test, comparing the gene or gene-vector mixture against vector groups.

(I) AOAH’s effects on B16F10 tumors, shown as in panel G.

(J) Endpoint-free survival plots for mice in panel I, shown as in panel H.

(K) RIL-175 orthotopic tumor luminescence and RenCa subcutaneous tumor growth, shown as in panel E.

(L) Endpoint-free survival plots for mice in panel K, shown as in panel F.

To examine the spatial distribution of AOAH in tumors, we searched for spatial transcriptomics data from tumors treated with immune checkpoint blockades and found two anti-PD1 studies covering 15 hepatocellular carcinoma tumors^20,21. AOAH expression levels are consistently higher among responders than non-responder tumors (Figure 3C), and spatial transcriptomics data show that the AOAH expression spans the entire tumor region (Figure 3D).

In vivo validation of AOAH, COLQ, CR1L, and ADAMTS7

Motivated by the associations between clinical outcomes and expression of AOAH, COLQ, CR1L, and ADAMTS7, we aimed to experimentally validate their functions in anti-tumor immune response using mouse tumor models. First, we separately overexpressed Aoah, Colq, and Cr1l (predicted as anti-tumor) in the anti-PD1 resistant B16-mhgp100 cell line and overexpressed Adamts7 (predicted as pro-tumor) in the anti-PD1 sensitive B2905-M4 cell line via lentiviral transductions. There was no significant change in cell proliferation or DNA replication in these overexpression lines (Figure S3G and H), indicating no overt effect of these secreted proteins on cell-intrinsic phenotypes.

Next, we performed subcutaneous tumor injections into C57BL/6 mice and treated them with anti-PD1 and IgG control. Aoah strongly suppressed and Adamts7 strongly stimulated tumor progression in both IgG and anti-PD1 treatment groups. In contrast, Cr1l and Colq suppressed tumor growth in both groups but achieved statistical significance only in the anti-PD1 groups (Figure 3E, F, and Table S7A). This result could indicate a mechanistic link between anti-PD1 and Cr1l and Colq, or could simply result from more substantial effects of Aoah and Adamts7 on tumor progression relative to Cr1l and Colq.

To further establish that secreted proteins mediated these effects, we inoculated a mixture of 10% Aoah-overexpression (OE), Cr1l-OE, Colq-OE, or 20% Adamts7-OE cells into 90% or 80% control vector-OE cells into mice. For each mixture, we observed an effect similar to 100% OE cells on the anti-PD1 response (Figure 3G, H, and Table S7A). These results suggest that Aoah, Colq, Cr1l, and Adamts7 expression in a minority of cells can modulate anti-PD1 efficacy by diffusively affecting immune response in the entire tumor.

AOAH potentiates immunotherapy responses in multiple tumor models

Among the four secreted proteins associated with immune checkpoint blockade (ICB) response, AOAH was the top computational candidate and had the strongest in vivo anti-tumor effects (Figure 3E–H). Further, AOAH did not induce tumor shrinkage in immunodeficient mice (Figure S3I), suggesting its function as an immune modulator. Accordingly, we decided to further characterize AOAH’s roles in anti-tumor immunity and designed four types of in vivo experiments.

The first type of experiment utilized Aoah overexpression (OE) in multiple syngeneic tumor models. The B16F10 melanoma cell line is more resistant to anti-PD1 than the B16-mhgp100 cell line used in the previous section (Figure 3E–H) because the mgp100 antigen expressed by B16F10 is much less immunogenic than the hgp100 antigen expressed by B16-mhgp100²². Aoah-OE in B16F10 tumors resulted in a remarkable increase in the efficacy of anti-PD1 and anti-PD1+CTLA4 (Figure 3I, J, and Table S7A) but did not result in tumor shrinkage in combination with IgG treatment (Figure S3J). Thus, Aoah-OE promoted immunotherapy efficacy in melanoma tumors with both strong (hgp100) and weak (mgp100) antigens. We observed the same result in an additional pair of melanoma cell lines expressing OVA antigens with strong (N4) and weak (V4) immunogenicities²² (Figure S3K and L). Consistently, inoculating 10% Aoah-OE cells with 90% Vector-OE cells was sufficient to promote ICB (Figure 3I and J). Aoah OE also potentiated anti-PD1 responses in the orthotopic RIL-175 hepatocellular carcinoma and the subcutaneous Renca renal cancer models (Figure 3K, L, and Table S7A). This result is consistent with the high correlation between AOAH expression and favorable outcomes in liver and renal cancer patients upon immunotherapy (Figure S3C).

The second type of experiment examined the effects of AOAH in spontaneous hepatocellular carcinoma (HCC) models, which induce tumors from mouse hepatocytes driven by MYC and β-catenin²³ (Figure S4A). Compared to the syngeneic models used previously, this model develops tumors spontaneously, which is more similar to how human cancers arise and therefore more physiologically relevant²⁴. High expression of Aoah inhibited tumor progression in both OVA+SIN+SIY (OS) antigen-high and antigen-null HCC (Figure 4A).

Figure 4. — (A) Tumor luminescence in the livers of C57BL/6 mice with spontaneous hepatocellular carcinoma (HCC). P-values were computed using the two-sided Wilcoxon rank-sum test, comparing values at the last measurements.

(B) Experimental design of adoptive transfer of *Aoah*-armored pmel CD8+ T cells into the tumor-bearing C57BL/6 mice.

(C) Tumor growth and endpoint-free survival of B16F10 tumors in C57BL/6 mice post T-cell transfer, shown as in Figure 3E and F. P-values were computed using the two-sided Wilcoxon rank-sum (left panel) and log-rank (right panel) tests.

(D) B16 tumor growth in the C57BL/6 mice who received intratumoral injections of rhAOAH (5mg/kg) and vehicle, shown as in Figure 3E.

(E) Endpoint-free survival plots for mice in panel D, shown as in Figure 3F.

(F) Marker-positive cell fractions determined by multiplexed immunofluorescence (multi-IF) in B16F10 tumors with *Aoah* or vector overexpressions. P-values were computed using the two-sided Wilcoxon rank-sum test.

(G) Representative multi-IF images for panel F.

(H) Marker-positive cell fractions determined by multi-IF in spontaneous HCC tumors with *Aoah* or vector overexpressions. P-values were computed using the two-sided Wilcoxon rank-sum test, and only significant values (p <= 0.05) are shown.

(I) Representative multi-IF images for panel H.

(J) Marker-positive cell fractions in the CD45+ tumor-infiltrating leukocytes isolated from B16 tumors, determined by flow analyses. P-values were computed using the two-sided Wilcoxon rank-sum test.

The third type of experiment tested whether AOAH induction in immune cells can enhance the efficacy of adoptive cell transfer in solid tumors. AOAH is mainly expressed by myeloid cells, such as dendritic cells (DCs) and macrophages, and lymphocytes, such as CD8 T cells and natural killer cells in tumors (Figure S3B). First, we overexpressed Aoah in the mouse dendritic cell line DC2.4 (Figure S4B). Aoah-OE and vector-OE DC2.4 cells were injected into B16-mhgp100-bearing mice intraperitoneally. The median tumor volume with injections of Aoah-OE DC2.4 is 12.7% of that with vector-OE DC2.4 (Figure S4C, one-sided Wilcoxon rank-sum p-value = 0.055). Second, we overexpressed Aoah in CD8+ pmel T-cell receptor (TCR) T cells that target B16F10 cells and injected TCR T cells intravenously (Figure 4B). Compared to vector-armored T cells, Aoah-armored TCR-T cells exhibited more potent B16F10 tumor suppression and prolonged mouse survival (Figure 4C, Figure S4D, and Table S7A) and did not cause noticeable weight loss or lymphoid tissue damage (Figure S4E and F).

The fourth type of experiment administered recombinant human AOAH protein (rhAOAH) to tumor-bearing mice. We produced His-tag or Fc-fusion recombinant human AOAH in HEK293 cells and isolated each protein to >99.7% purity with endotoxin levels <0.1 EU/mg (Methods). Intratumoral administration of 2–5 mg/kg rhAOAH protein in mice significantly suppressed tumor progression and synergized with anti-PD1 and anti-PD1 & CTLA4 combination treatment (Figure 4D, E, Figure S4G, and Table S7A) without causing weight loss (Figure S4H).

Finally, to understand how AOAH alters the microenvironment during tumor progression, we analyzed the composition of CD45+ tumor-infiltrating leukocytes isolated from B16-mhgp100, B16F10, and spontaneous hepatocellular tumors after treatment with ICB. We found higher infiltrations of CD8+ T cells and CD11c+ dendritic cells in Aoah-OE and rhAOAH-treated tumors (Figure 4F–J and Figure S4I), suggesting that AOAH promotes CD8+ T-cell expansion and antigen presentation.

Molecular programs induced by Aoah in vivo

To characterize Aoah-induced molecular programs, we performed RNA-seq on tumors from hepatocellular carcinoma and melanoma mouse models that overexpressed Aoah or vector control (Table S7B). As expected, across four different tumor models, differential gene expression profiles between Aoah-OE and vector-OE tumors (Figure 5A) were superficially divergent (Figure S5A). Therefore, we performed gene set enrichment analysis (GSEA)²⁵ (Figure 5B), screening for common pathways whose gene members are induced by Aoah. Top pathways enriched across models include Arachidonic acid metabolism, Antigen processing and presentation, T-cell receptor (TCR) signaling, IFNγ response, and other pathways related to these terms (Figure 5C and D).

The enrichment of antigen processing and presentation and TCR signaling (Figure 5C) corroborates positive correlations between AOAH expression and cytotoxic T-cell infiltration across human tumors (Figure S3D and E) and the Aoah-enhanced infiltrations of CD8 T cells and dendritic cells in mouse tumors (Figure 4F–J). The enrichment of IFNγ response (Figure 5C and Figure S5B) suggests IFNγ release as a consequence of TCR activation. One paradoxical enrichment is the arachidonic acid metabolism (Figure 5B and C). Arachidonic acid is often linked to phospholipids²⁶. The GSEA result, along with the known lipase activity of AOAH²⁷, suggests that AOAH may function in tumors as a phospholipase.

Single-cell analysis of immune cell characteristics induced by Aoah

To specifically study the effects of Aoah-OE on immune cells, we performed single-cell RNA-seq on CD45+ cells isolated from spontaneous hepatocellular tumors (Figure 5E) induced with and without OS antigens. Tumors overexpressing Aoah contained higher fractions of T and natural killer (NK) cells and lower fractions of polymorphonuclear myeloid-derived suppressor cells (Figure 5F and Figure S5C). The level of TCR clonal expansion was consistently elevated among cytotoxic CD8 T cells in tumors with Aoah OE compared to tumors with vector OE (Figure 5G). In contrast, B-cell receptors and immunoglobulin classes did not change consistently (Figure S5D and E).

For each immune cell subtype, we computed the log2FC of gene expression upon Aoah OE (Methods). GSEA revealed immune activation pathways, such as “antigen processing and presentation” and “cell adhesion molecules,” enriched in profiles of monocytes, macrophages, and T cells (Figure 5H). A notable enrichment is the oxidative phosphorylation pathway in T/NK cells, proliferating T cells, monocytes, and macrophages (Star 3 in Figure 5H and Figure S5F). Previous studies showed that impaired oxidative phosphorylation limits the self-renewal of T cells exposed to persistent antigens²⁸. Our data suggest that Aoah induction might protect oxidative phosphorylation (Figure 5H), thus enhancing the anti-tumor efficacy of T cells²⁸.

Molecular programs induced by CR1L, COLQ, and ADAMTS7

While we focused mainly on the top candidate AOAH, we also performed a preliminary analysis of the molecular programs associated with the anti-tumor (CR1L, COLQ) or pro-tumor (ADAMTS7) candidates validated by our initial in vivo experiments (Figure 3E–H). We generated bulk RNA-seq profiles of tumors with gene-OE and vector-OE and created differential gene expression profiles (Figure 5I and Table S7B). Then, we performed GSEA (Figure 5J).

Cr1l OE induced two programs: immune cell chemotaxis and B-cell receptor (BCR) signaling (Figure 5J). Gene sets induced by Cr1l OE in mouse tumors were also highly expressed in human tumors with high CR1L expression (Figure 5K); however, such consistencies were less evident for COLQ and ADAMTS7 (Figure S5G). Top co-enriched pathways (associated with mouse Cr1l OE and human CR1L co-expression) include BCR signaling, αβ T-cell activation, and leukocyte-mediated immunity (Figure 5L). Further, BCR clonotype counts and immunoglobulin heavy chains (IGH) clonalities are significantly higher in Cr1l-OE mouse tumors than in vector controls (Figure 5M). We presented example genes of Cr1l-induced parallels between mice and humans. CR1L expression in tumors is highly correlated with IGHM (BCR activation marker) and CCL3 (immune chemotaxis marker) in a cohort of melanoma patients treated with anti-PD1²⁹ (Figure 5N and Figure S5H).

Colq OE in mouse tumors induced four transcriptomics programs, broadly classified as oxidative phosphorylation, immune cell chemotaxis, humoral immune response, and actin-myosin filament sliding (Figure 5J). Consistent with the pro-tumor role of Adamts7 (Figure 3E–H), Adamts7 OE significantly suppressed several anti-tumor immunity programs, including IFNγ response and production, peptide antigen processing and presentation, and T-cell activation (Figure 5J). While these RNA-seq analyses suggest roles for Cr1l, Colq, and Adamts7 in anti-tumor immunity, further studies will be required to elucidate the underlying mechanisms.

AOAH enhances TCR activation and antigen recognition

Analyses from the mouse models above suggested that AOAH potentiates CD8+ T cells in tumors, particularly tumors with weak antigens. Thus, we hypothesized that AOAH promotes TCR activation and antigen recognition, and tested this hypothesis using three in vitro models.

The first is an antigen-specific tumor-killing model, pitting pmel CD8+ T cells against B16-mhgp100 and B16F10 cells. The antigen-specific T cells treated with rhAOAH killed antigen-positive tumor cells more effectively (Figure 6A and Figure S6A–C) and exhibited higher cytotoxic cytokine secretion (Figure S6D).

Figure 6. — Panels A, C-G, and J show the mean and standard deviation values. In panels B-G, J, and K, sample sizes and comparison methods are the same as in panel B for each condition.

(A) T-cell-mediated tumor killing, quantified as normalized tumor-expressing mCherry fluorescence in B16 cells co-cultured with CD8+ pmel T cells (n = 4 cell culture replicates). P-values were computed through the two-way ANOVA with time and treatment as two factors.

(B) Concentrations of cytokines released by mouse CD8+ T cells after activation. P-values were computed using the two-sided paired t-test, comparing values at each condition (n = 3 biological replicates). The line presents the mean value per condition.

(C) Concentrations of cytokines released by human CD8+ T cells after activation.

(D) Fractions of CD25+ cells among mouse CD8+ T cells after activation.

(E) Fractions of CD25+ cells and CD69+ cells among human CD8+ T cells after activation.

(F) Normalized luminescence intensity of human Jurkat NFAT reporter cells at different anti-CD3/28 beads:cells ratios.

(G) Fractions of p-CD3ζ+ and p-LCK+ cells in mouse CD8+ T cells after activation by anti-CD3/28 antibodies for 10 minutes.

(H) Western blots of TCR signaling markers in human CD8+ T cells after activation.

(I) Differentially expressed genes and enriched pathways in rhAOAH-treated mouse CD8+ T cells based on RNA-seq. The fold changes and p-values were calculated by DESeq2 (sample sizes in Table S7C), and normalized enrichment scores were computed using GSEA²⁵ and hallmark pathways⁴⁴.

(J) Fractions of mouse Granzyme A+ and Perforin+ cytotoxic CD8+ T cells after activation by hgp100 and mgp100-pulsed splenocytes.

(K) Concentrations of cytokines released by pmel CD8+ T cells after activation by hgp100 and mgp100-pulsed splenocytes. The line presents the mean value per condition.

The second model utilizes anti-CD3/CD28 antibodies to activate TCR in mouse and human CD8+ T cells. The rhAOAH treatment significantly enhanced pro-inflammatory cytokine secretion (Figure 6B, C, and Figure S6E) and T-cell activation (Figure 6D–F and Figure S6F). Treatment of mouse and human CD8+ T cells with rhAOAH resulted in phosphorylation of TCR-induced downstream factors, including CD3ζ, LCK, ZAP-70, and LAT, early in the activation timeline (Figure 6G, H, and Figure S6G). RNA-seq of rhAOAH-treated CD8+ T cells shows up-regulation of T-cell effector gene sets³⁰, including E2F targets, G2M checkpoint, and mTORC1 signaling (Figure 6I and Table S7C).

The third model is an MHCI-dependent co-culture system in which antigen-specific CD8 T cells interact with splenocytes pulsed with antigens of various immunogenicity levels. The rhAOAH treatment significantly enhanced the T-cell cytotoxicity and pro-inflammatory cytokine secretion for intermediate (hgp100), strong (OVA-N4), and weak (mgp100 and OVA-V4) antigens (Figure 6J, K, and Figure S6H). Particularly, rhAOAH consistently promoted IFNγ secretion for antigens with low immunogenicity and density (Figure 6K and Figure S6H). RNA-seq of splenocyte co-cultures treated with rhAOAH shows enrichment of antigen presentation via MHCI, interferon signaling, inflammatory responses, cell cycle, glycolysis, and mTORC1 signaling (Figure S6I and J), which was consistent with most pathways in CD8+ T cells activated by anti-CD3/CD28 (Figure 6I).

AOAH depletes arachidonoyl phosphatidylcholines

As our experiments above did not involve bacteria, the anti-tumor role of AOAH in our models should not involve its enzymatic activities on lipopolysaccharides. Alternatively, previous studies showed that AOAH can function as a phospholipase^27,31. Therefore, we hypothesized that the mechanism of AOAH-enhanced activation of T cells is related to its phospholipase activity. To test this hypothesis, we performed untargeted lipidomics on the cell culture medium and pellets for T cells treated with rhAOAH and vehicle control (Figure 7A and Figure S7A).

Figure 7. — Panels D, G, H, K, L-N show the mean and standard deviation values for each condition. In panels D-H, K-M, and O, comparisons are made through the two-sided paired t-test. Only p-values < 0.05 and passed the K-S normality test were shown.

(A) Lipid species depleted by rhAOAH. Y-axis: log2FC of MassSpec peak areas between rhAOAH and buffer treatments. Each bar represents the mean log2FC across ion adducts for the same lipid, with each ion as a point. Only lipids with log2FC < −1 are shown.

(B) Lipid class enrichment analysis. The log2FC of lipids within each class (blue) was compared with values of lipids outside the class (gray) by the two-sided Wilcoxon rank-sum test. Each value group was shown by a violin plot, smoothed by a kernel density estimator. Only significant lipid classes are shown (FDR < 0.05, Methods).

(C) Fatty acid enrichment analysis for the PC diacyl (phosphatidylcholine) class from the T-cell culture medium. The log2FC values were compared between lipids with arachidonic acid (20:4) on the sn-2 position and other lipids by the two-sided Wilcoxon rank-sum test. Values in each group are shown by violin plots as in panel B. Fatty acids on the sn-1 position were labeled for lipids with log2FC < −1.

(D) Concentrations of arachidonic acid released from mouse CD8+ T cells after rhAOAH treatment (n = 5 biological replicates).

(E) Inhibition ratios of PC(16:0–20:4) on cytokine release from activated mouse CD8+ T cells. Each dot represents the cytokine concentration ratio between the PC-treated and mock-treated groups (raw data in Figure S7B, n = 3 biological replicates). The line shows the mean value per condition.

(F) The inhibition ratios of PC(16:0–20:4) on cytokine release from activated human CD8+ T cells (raw data in Figure S7C, n = 3 human donors), shown as in panel E.

(G) Fractions of p-CD3ζ+, p-LCK+, and CD25+ cells in PC(16:0–20:4) pre-treated CD8+ mouse T cells after activation (n = 3 biological replicates).

(H) Fractions of CD25+ and CD69+ cells in PC(16:0–20:4) pre-treated CD8+ human T cells after activation (n = 3 human donors).

(I) Lipid peroxidation quantification based on the intracellular ROS level in activated mouse CD8+ T cells. The Y-axis shows the ratio between initial lipid sensor and peroxidized sensor levels in each cell. All ratios are shown as violin plots smoothed by a kernel density estimator. P-values were computed through the two-sided Wilcoxon rank-sum test, comparing median ratios between two groups (n = 3 biological replicates).

(J) Differentially-expressed genes and pathways in PC(16:0–20:4)-treated mouse CD8+ T cells based on RNA-seq. The fold changes and p-values were calculated by DESeq2 (sample sizes in Table S7C), and normalized enrichment scores were computed using GSEA²⁵ and hallmark pathways⁴⁴.

(K) The percentage inhibition of dendritic cells’ antigen-presentation and co-stimulatory markers among live CD11c+ cells (n = 3 biological replicates). Mouse splenocytes were pre-treated with oxPAPC(16:0–20:4) and co-cultured with OT-1 CD8+ T cells in the presence of 10 nM OVA antigen.

(L) The chromatogram of point mutation of Serine 263 to Alanine in the AOAH sequence⁴⁰ (top), and the diminished catalytic activity (measured as p-nitrophenyl abundance) towards p-nitrophenyl acetate (bottom). Values are compared at the last measurements (n = 3 biological replicates).

(M) Concentrations of arachidonic acid released from mouse and human CD8+ T cells treated with wild-type rhAOAH, mutant rhAOAH, and vehicle, measured by the competitive ELISA (n = 5 biological replicates for mouse CD8+ T cells, and n = 3 for human CD8+ T cells).

(N) Fractions of CD25+ and CD69+ cells in wild-type AOAH, mutant AOAH, and vehicle-treated CD8+ mouse T cells post activation. P-values were computed through the two-sided unpaired t-test (n = 3 cell culture replicates).

(O) Concentrations of cytokines released by mouse CD8+ T cells after activation in the presence of wild-type or mutant rhAOAH, shown as in panel E. Comparisons were done between wild-type values and mutant values at each concentration (n = 3 biological replicates).

To evaluate whether AOAH preferentially depletes or enriches any lipid categories, we performed a lipid class enrichment analysis that compares the log2FC values of lipids with a similar configuration against values of lipids without the configuration (Methods). Only four lipid classes in the T-cell culture medium passed the false discovery rate (FDR) threshold of less than 0.05 (Figure 7B). The depletion of diacyl glycerophosphocholine (PC), also named phosphatidylcholine, achieved the highest statistical significance (Figure 7B).

Phosphatidylcholine has two fatty acids connected on the sn-1 and sn-2 positions by ester links³². We further analyzed the fatty acids enriched at each location and found only one enrichment: the 20:4 (arachidonic acid) on the sn-2 position (Figure 7C and Methods). Higher arachidonic acid release from rhAOAH-treated mouse CD8+ T cells was independently verified by Enzyme-Linked Immunosorbent Assay (ELISA) (Figure 7D). These results were consistent with the enrichment of arachidonic acid metabolism in the transcriptomics analysis (Figure 5B and C).

AOAH rescues TCR inhibition by arachidonoyl phosphatidylcholines

Among phosphatidylcholines (PC) from human plasma, 16:0 and 18:0 are the two most abundant fatty acid species in the sn-1 position³³. To evaluate the effects of PC(16:0–20:4) and PC(18:0–20:4) on TCR activation, we pretreated mouse CD8+ T cells with both lipids in a serum-free medium to avoid interference from pre-existing lipids in the fetal bovine serum (FBS). The result demonstrated that both PC(16:0-C20:4) and PC(18:0-C20:4) can diminish T-cell activation, as shown by decreased secretion of IFNγ, TNFα, and IL-2 (Figure S7B). Consistent with the mouse results, in human CD8+ T cells, PC(16:0–20:4) significantly inhibited the secretion of IFNγ and TNFα after TCR stimulation (Figure S7C).

AOAH treatment rescued T-cell suppression from PC(16:0–20:4) at both low and high lipid concentrations (Figure 7E) and suppression from PC(18:0–20:4) at low but not at high concentrations (Figure S7D). In human CD8+ T cells, rhAOAH treatment rescued the PC(16:0–20:4) inhibitory effects on the secretion of IFNγ and TNFα after TCR stimulation (Figure 7F and Figure S7C). The AOAH-mediated rescue of PC(16:0–20:4) inhibition on TCR activation was confirmed by flow analysis in mouse and human CD8+ T cells (Figure 7G, H, Figure S7E, and F).

Previous studies showed that the arachidonyl moiety on the PC sn-2 position can trigger lipid peroxidation³⁴. In addition, reactive oxygen species (ROS) are produced at high levels during T-cell activation and can enhance lipid peroxidation via oxidation of PCs, thereby inhibiting T-cell functions³⁵. We showed that AOAH treatment reduced lipid peroxidation in CD8+ T cells (Figure 7I and Figure S7G), explaining AOAH rescue of T cells from arachidonyl PCs.

The sn-2 hydrolysis product of PC(16:0–20:4) is lysophosphatidylcholine (LPC) mono-acyl 16:0. A recent study demonstrated LPC(16:0) as an immunosuppressive lipid by suppressing T-cell motility¹⁰. Our untargeted lipidomics analysis showed that AOAH treatment depleted both PC(16:0–20:4) and its sn-2 hydrolysis product LPC(16:0) (Figure 7A, C). This result is consistent with AOAH’s capability as both phospholipase and lysophospholipase²⁷ and suggests that AOAH can completely hydrolyze the immunosuppressive PC(16:0–20:4) to promote T-cell functions.

Last, RNA-seq on mouse CD8+ T cells treated with PC(16:0–20:4) showed down-regulation of cell cycle-related genes (Figure 7J and Table S7C). In contrast, these cell cycle genes were up-regulated by AOAH treatment (Figure 6I), demonstrating that PC(16:0–20:4) had the opposite effect compared to AOAH.

AOAH protects dendritic cells from oxidized PC(16:0–20:4) inhibition

Besides CD8 T cells, our previous analysis showed that AOAH promoted dendritic cell (DC) functions (Figure S4C) and infiltration in tumors (Figure 4F, H, J). Existing studies demonstrated that oxidized PC(16:0–20:4) (oxPAPC) inhibits DC functions^36,37. Further, oxPAPC can induce the migration and invasive capacity of human cancer cells³⁸. OxPAPC levels are higher in plasma samples of human lung cancer patients compared to lung disease controls³⁹. As derivatives of oxPAPC are known AOAH substrates³¹, we hypothesized that AOAH could protect DCs from immunosuppressive oxPAPC.

To test this hypothesis, we pre-treated mouse splenocytes (as antigen-presenting cells) with oxPAPC and co-cultured them with OT-1 CD8+ T cells in the presence of OVA antigens. We found that oxPAPC significantly diminished the expression of MHC-I, MHC-II, CD40, and CD80 on CD11c+ DCs among splenocytes (Figure 7K and Figure S7H), deteriorating their antigen presentation and co-stimulatory functions. As predicted, AOAH treatment reverted the inhibitory effect of oxPAPC on CD11+ DCs (Figure 7K and Figure S7I).

Serine 263 is a catalytic site

A previous study of AOAH binding to the lipopolysaccharide identified the catalytic site as Serine 263, initiating a nucleophilic attack on the substrate’s carbonyl carbon⁴⁰. We hypothesized that AOAH also depended on the same site to promote T-cell functions via catalyzing PC hydrolysis. We introduced a point mutation, converting Serine 263 to Alanine⁴⁰, and found the Ser263-Ala mutant AOAH reduced the esterase activity of AOAH using a test substrate p-nitrophenyl acetate, but did not completely abolish its enzymatic activity (Figure 7L). Consistent with this result, Arachidonic acid release from CD8+ T cells treated with mutant AOAH was significantly lower than that of wild-type AOAH treatment, but not as low as vehicle-treated cells (Figure 7M). Similarly, mutant AOAH treatment had a significantly lower impact on the T-cell activation and cytotoxic cytokine release than the wild-type protein (Figure 7N, O, and Figure S7J). These results confirmed that S263 is a vital catalytic site for enhancing TCR activation.

Discussion

In this study, we used CIDE to study secreted proteins, which are often difficult to study through genetic methods but are amenable to data-centric approaches. However, the utility of CIDE is not restricted to studying secreted proteins and applies broadly to the characterization of all human genes in cancer. In addition to the functionalities utilized in this study, CIDE includes additional modules, such as analysis of somatic DNA alterations, and incorporates additional datasets from non-immunotherapy cohorts.

We uncovered the function of AOAH in promoting TCR activation and dendritic cell functions by depleting immunosuppressive phospholipids. However, the underlying mechanisms of other validated candidates (ADAMTS7, COLQ, and CR1L) in modulating anti-tumor immune response are still unclear and may hold promise for future studies.

Limitations of the Study

Hurdles still exist for the clinical application of AOAH. We showed that rhAOAH proteins are effective in shrinking tumors when delivered by intratumoral injections, a method unsuitable for widespread clinical use. Adoptive therapy using AOAH-armed TCR T cells in humans would currently demand isolation of cells from individuals, transduction of patient T cells in a Good Manufacturing Practice-regulated facility, and clinical re-introduction of the cells to the patient, a much more complex and expensive process than we faced in mice. Future studies are necessary to engineer AOAH as a broadly applicable immunotherapy through systemic administration.

Resource Availability

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Peng Jiang (peng.jiang@nih.gov).

Materials Availability

Correspondence and requests for materials should be addressed to the lead contact. The unique identifiers of all biological materials are listed in the key resources table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
InVivoMAb anti-mouse PD-1 (CD279, Clone RMP1-14)	Bio X Cell	BE0146; RRID: AB_10949053
InVivoMAb anti-mouse CTLA-4 (CD152, Clone 9H10)	Bio X Cell	BE0131; RRID: AB_10950184
InVivoMAb rat IgG2a isotype control, anti-trinitrophenol	Bio X Cell	BE0089; RRID: AB_1107769
LAT Rabbit pAb	ABclonal	A5650; RRID: AB_2766410
LCK Rabbit pAb	ABclonal	A2177; RRID: AB_2764195
ZAP70 Rabbit pAb	ABclonal	A2195; RRID: AB_2764212
Phospho-LAT-Y171 Rabbit pAb	ABclonal	AP0809; RRID: AB_2771257
Phospho-LCK-Y394 Rabbit pAb	ABclonal	AP0182; RRID: AB_2771262
Phospho-ZAP70-Y319 Rabbit pAb	ABclonal	AP0467; RRID: AB_2771650
Monoclonal Anti-Vinculin antibody	Sigma	V9131; RRID: AB_477629
Amersham ECL Mouse IgG, HRP-linked whole Ab	Cytiva	NA931-1ML; RRID: AB_772210
Amersham ECL Rabbit IgG, HRP-linked whole Ab	Cytiva	NA934-1ML; RRID: AB_772206
BD Horizon^™ BUV737 Rat Anti-Mouse CD45	BD Biosciences	748371; RRID: AB_2872790
APC/Cyanine7 anti-mouse CD8a	Biolegend	100714; RRID: AB_312753
FITC anti-mouse CD4	Biolegend	100405; RRID: AB_312690
Brilliant Violet 605^™ anti-mouse CD19	Biolegend	115540; RRID: AB_2563067
BD OptiBuild^™ BUV395 Rat Anti-Mouse CD138	BD Biosciences	740240; RRID: AB_2739987
APC anti-mouse CD14	Biolegend	123311; RRID: AB_940574
PE anti-mouse CD11c	Biolegend	117308; RRID: AB_313777
PE anti-mouse CD3	Biolegend	100205; RRID: AB_312662
FITC Mouse Anti-Mouse Vβ 13 TCR	BD Biosciences	553204; RRID: AB_394706
PerCP/Cyanine5.5 anti-mouse CD25	Biolegend	102030; RRID: AB_893288
PE anti-mouse Granzyme A	Biolegend	149704; RRID: AB_2565310
FITC anti-mouse Perforin	Biolegend	154310; RRID: AB_2910315
Phospho-CD247 (CD3 zeta) (Tyr142) Monoclonal Antibody (3ZBR4S), PE	eBioscience	12-2478-42; RRID: AB_2744700
Phospho-LCK (Tyr505) Monoclonal Antibody (SRRCHA), PerCP-eFluor^™ 710	eBioscience	46-9076-42; RRID: AB_2573868
Anti-mouse CD4	Servicebio	GB15064; RRID: AB_3095557
Anti-mouse CD8	Servicebio	GB15068; RRID: AB_3246431
Anti-mouse CD20	Servicebio	GB11540; RRID: AB_3698721
Anti-mouse CD11c	Servicebio	GB11059; RRID: AB_2905514
Anti-mouse CD14	Servicebio	GB11254; RRID: AB_3065044
Anti-mouse pan-CK	Bioss	bs-1712R; RRID: AB_10855057
Rabbit Anti-Goat IgG Antibody, HRP conjugate	Servicebio	GB23204; RRID: AB_2938981
APC/Cyanine7 anti-human CD8	BioLegend	344714; RRID: AB_2044006
APC anti-human CD25	BioLegend	302610; RRID: AB_314280
PE anti-human CD69	BioLegend	310906; RRID: AB_314841
BD Pharmingen^™ APC-Cy^™7 Rat Anti-Mouse CD45	BD Biosciences	561037; RRID: AB_10563075
FITC anti-mouse CD11c	BioLegend	117306; RRID: AB_313775
APC anti-mouse CD40	BioLegend	124612; RRID: AB_1134072
PE anti-mouse CD80	BioLegend	104708; RRID: AB_313129
APC anti-mouse I-A/I-E	BioLegend	107614; RRID: AB_313329
PE anti-mouse H-2Kb	BioLegend	116508; RRID: AB_313735
Bacteria and virus strains
One Shot^™ Stbl3^™ Chemically Competent E. coli	Invitrogen	C737303
Biological samples
Human peripheral blood	NIH blood center	N/A
Chemicals, peptides, and recombinant proteins
DAPI (4',6-Diamidino-2-Phenylindole, Dilactate)	BioLegend	422801
iF440-Tyramide	Servicebio	G1250
iF488-Tyramide	Servicebio	G1231
iF546-Tyramide	Servicebio	G1251
iF594-Tyramide	Servicebio	G1242
iF647-Tyramide	Servicebio	G1232
iF700-Tyramide	Servicebio	G1252
Recombinant human AOAH protein (His-tag)	Wuxi Biologics	WBP70579_1
Recombinant human AOAH protein (Fc-fusion)	Wuxi Biologics	WBP70579_4
Recombinant human AOAH protein (His-tag, S263-mutant)	Wuxi Biologics	WBP71306_1
Recombinant human IL-2 Protein	R&D Systems	202-IL
TGF beta 1/TGFB1 Protein, Human	MedChemExpress	HY-P70543
Human gp100 peptide (25–33)	GenScript	RP20344
Mouse gp100 peptide (25–33)	AnaSpec	AS-64752
OVA-N4 peptide	Grégoire Altan-Bonnet’s Lab	N/A
OVA-V4 peptide	Grégoire Altan-Bonnet’s Lab	N/A
16:0–20:4 phosphatidylcholine	Avanti Polar Lipids	850459
18:0–20:4 phosphatidylcholine	Avanti Polar Lipids	850469
Critical commercial assays
Mouse Acyloxyacyl hydrolase (AOAH) ELISA Kit	AFG Scientific	EK732092
Mouse IFN gamma ELISA Kit	Invitrogen	KMC4021
Mouse TNF alpha ELISA Kit	Invitrogen	BMS607-3
Mouse IL-2 ELISA Kit	Invitrogen	BMS601
Human IFN gamma ELISA Kit	Invitrogen	KHC4021
Human TNF alpha ELISA Kit	Invitrogen	KHC3011
Arachidonic Acid ELISA Kit	abcam	ab287798
Deposited data
Source data of figures	Zenodo	https://doi.org/10.5281/zenodo.15012968
The Cancer Genome Atlas	TCGA	https://xenabrowser.net/datapages/?cohort=TCGA%20Pan-Cancer%20(PANCAN)&removeHub=https%3A%2F%2Fxena.treehouse.gi.ucsc.edu%3A443
International Cancer Genome Consortium	ICGC	https://dcc.icgc.org
Clinical Proteomic Tumor Analysis Consortium	CPTAC	https://proteomic.datacommons.cancer.gov/pdc/cptac-pancancer
Therapeutically Applicable Research to Generate Effective Treatments	TARGET	https://portal.gdc.cancer.gov
PREdiction of Clinical Outcomes from Genomic profiles	PRECOG	https://precog.stanford.edu
Cancer transcriptomics data from studies without immunotherapies at the TIDE database	TIDE others	http://tide.dfci.harvard.edu
Single-cell RNA-seq data for lineage-specific gene expression analysis	This paper	Listed in Table S1
Bulk-tumor omics data from cancer immunotherapy studies	This paper	Listed in Table S1
Single-cell RNA-seq of MYC-Luc tumors with Aoah overexpression	This paper	GEO: GSE291594
Bulk RNA-seq of MYC-Luc tumors with Aoah overexpression	This paper	GEO: GSE291586
Bulk RNA-seq of B16 tumors with Aoah overexpression	This paper	GEO: GSE291587
Bulk RNA-seq of B16 tumors with Cr1l overexpression	This paper	GEO: GSE291584
Bulk RNA-seq of B16 tumors with Colq overexpression	This paper	GEO: GSE291582
Bulk RNA-seq of M4 tumors with Adamts7 overexpression	This paper	GEO: GSE291354
Bulk RNA-seq of mouse CD8 T cells treated with AOAH or PC(16:0_20:4)	This paper	GEO: GSE291590
Bulk RNA-seq of co-cultures between hgp100-pulsed splenocytes with Pmel mouse CD8 T cells treated with human recombinant AOAH protein or vehicle (PBS + 10% glycerol).	This paper	GEO: GSE291589
Experimental models: Cell lines
B16F10 cell line	ATCC	CRL-6475
B16-mhgp100 cell line	Nicholas P. Restifo’s Lab	N/A
RIL-175 cell line	Stephanie Ma’s Lab	N/A
RenCa cell line	ATCC	CRL-2947
Jurkat-Lucia^™ NFAT-CD28 cell line	InvivoGen	jktl-nfat-cd28
Experimental models: Organisms/strains
C57BL/6J mouse	Jackson Laboratory	000664; RRID: IMSR_JAX:000664
Balb/c mouse	Charles River Laboratories	028; RRID: MGI:2161072
Pmel mouse	Glenn Merlino’s Lab	RRID: IMSR_JAX:005023
OT-1 mouse	Grégoire Altan-Bonnet’s Lab	RRID: IMSR_JAX:003831
NSG mouse	Jackson Laboratory	005557; RRID: IMSR_JAX:005557
Oligonucleotides
Aoah-forward	Quintarabio	TGAAGACCGCCTGCTATTTCT
Aoah-reverse	Quintarabio	GGGGAAGTGGGTAGAGATGAC
Colq-forward	Quintarabio	TGATGATGGTAATCCCGATGTGA
Colq-reverse	Quintarabio	CTCGCATGTCAGGTAGCCAA
Cr1l-forward	Quintarabio	AGGTCTCTTCTCGGAGTTCAG
Cr1l-reverse	Quintarabio	GCAGCAAAACTTCTAGCTTGACT
Adamts7-forward	Quintarabio	ATTGGGAACCCCATTCACATC
Adamts7-reverse	Quintarabio	TCTGCCACCTACAGAAGTTCTT
Gzma-forward	Quintarabio	TTTGGCAATAAGTCAGCTCCC
Gzma-reverse	Quintarabio	GGTGATGCCTCGCAAAATACC
Gzmb-forward	Quintarabio	TCTCGACCCTACATGGCCTTA
Gzmb-reverse	Quintarabio	TCCTGTTCTTTGATGTTGTGGG
Ifng-forward	Quintarabio	ACAGCAAGGCGAAAAAGGATG
Ifng-reverse	Quintarabio	TGGTGGACCACTCGGATGA
Cd25-forward	Quintarabio	CAAGAACGGCACCATCCTAAA
Cd25-reverse	Quintarabio	TCCTAAGCAACGCATATAGACCA
Zeb2-forward	Quintarabio	GCTACACGTTCGCCTACCG
Zeb2-reverse	Quintarabio	CCTTGGGTTAGCATTTGGTGC
Rpl13a-forward	Quintarabio	AGCCTACCAGAAAGTTTGCTTAC
Rpl13a-reverse	Quintarabio	GCTTCTTCTTCCGATAGTGCATC
Actb-forward	Quintarabio	GGCTGTATTCCCCTCCATCG
Actb-reverse	Quintarabio	CCAGTTGGTAACAATGCCATGT
Recombinant DNA
pReceiver-Lv156-EF1-Vector	Genecopoeia	EX-Lv156
pReceiver-Lv156-EF1-Aoah	Genecopoeia	EX-Mm06544-Lv156
pReceiver-Lv156-EF1-Cr1l	Genecopoeia	EX-Mm01941-Lv156
pReceiver-Lv156-EF1-Colq	Genecopoeia	EX-Mm30120-Lv156
pReceiver-Lv156-EF1-Adamts7	Genecopoeia	EX-Mm24248-Lv156
pLV-mCherry	Addgene	Addgene #36084
Luciferase-expressing plasmid	In-house made
MSCV-IRES-Thy1.1	Addgene	Addgene #17442
MSCV-IRES-Thy1.1-Aoah	In-house made	Addgene # 233141
EF1A-MYC-IRES-lucOS	Addgene	Addgene #129776
EF1A-MYC-IRES-luc	Addgene	Addgene #129775
pCMV(CAT)T7-SB100	Addgene	Addgene #34879
pT3-N90-CTNNB1	Addgene	Addgene #31785
PKT2/CLP-Vector	In-house made	Addgene # 239611
PKT2/CLP-Aoah	In-house made	Addgene # 239612
Software and algorithms
Logistic regression with Firth correction	This paper	https://github.com/data2intelligence/logis_batch
Python 3.9.13	Rossum and Drake⁵⁴	https://www.anaconda.com
Scipy 1.11.4	SciPy 1.0 Contributors⁵⁵	https://scipy.org
Django 4.1.4	Django Software Foundation	https://www.djangoproject.com/download
MySQL 8.0.32	Oracle	https://dev.mysql.com/downloads/mysql
R 4.3.0	R Core Team⁵⁶	https://www.r-project.org
Fastp 0.23.2	Chen et al.⁵⁷	https://github.com/OpenGene/fastp
Sratoolkit 3.0.5	SRA Toolkit Development Team	https://github.com/ncbi/sra-tools
Samtools 1.17	Li et al.⁵⁸	https://github.com/samtools
Bedtools 2.30.0	Quinlan and Hall⁵⁹	https://github.com/arq5x/bedtools2
Salmon 1.10.0	Patro et al.⁶⁰	https://combine-lab.github.io/salmon
STAR 2.7.10b	Dobin et al.⁶¹	https://github.com/alexdobin/STAR
DESeq2	Love et al.⁶²	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
combine_pvalues	GitHub	https://github.com/kylessmith/combine_pvalues
TrimGalore 0.6.7	Babraham Bioinformatics	https://github.com/FelixKrueger/TrimGalore
RSEM 1.3.3	Li and Dewey⁶³	https://github.com/deweylab/RSEM
MiXCR 4.3.2	Bolotin et al.⁶⁴	https://mixcr.com
BWA-mem2 2.2.1	Vasimuddin et al.⁶⁵	https://github.com/bwa-mem2
Picard 3.1.0	Broad Institute	https://github.com/broadinstitute/picard
VarScan 2.4.6	Koboldt et al.⁶⁶	https://varscan.sourceforge.net
VEP 106	McLaren et al.⁶⁷	https://github.com/Ensembl/ensembl-vep
vcf2maf 1.6.21	Kandoth et al.⁶⁸	https://github.com/mskcc/vcf2maf
GATK 4.4.0.0	McKenna et al.⁶⁹	https://github.com/broadinstitute/gatk
Strelka 2.9.10	Kim et al.⁷⁰	https://github.com/Illumina/strelka
OncoKB 3.3.2	Chakravarty et al.⁷¹	https://www.oncokb.org
DNAcopy 1.74.1	Seshan and Olshen	https://bioconductor.org/packages/DNAcopy
CutAdapt 4.4	Martin⁷²	https://cutadapt.readthedocs.io
SortMeRNA 4.3.6	Kopylova et al.⁷³	https://github.com/sortmerna
SpaCET 1.2.0	Ru et al.⁷⁴	https://github.com/data2intelligence/SpaCET
cellranger-7.2.0	10X Genomics	https://www.10xgenomics.com/support/software/cell-ranger
Seurat 5.0.1	Hao et al.⁷⁵	https://github.com/satijalab/seurat
Scrublet 0.2.2	Wolock et al.⁷⁶	https://github.com/swolock/scrublet
GSEA 4.3.2	Subramanian et al.²⁵	https://www.gsea-msigdb.org
mosdepth 0.3.3	Pedersen and Quinla⁴⁵	https://github.com/brentp/mosdepth
fastq_screen 0.16.0	Wingett and Andrews⁴⁶	https://www.bioinformatics.babraham.ac.uk/projects/fastq_screen/
ascatngs 4.5.0	Raine et al.⁵⁰	https://github.com/cancerit/ascatNgs
tidyestimate 1.1.1	Yoshihara et al.⁴⁹	https://github.com/KaiAragaki/tidyestimate
Collapsible_tree 1.0	Gao et al.⁷⁷	https://github.com/data2intelligence/collapsible_tree
FlowJo v11	BD Biosciences	https://www.flowjo.com/download
GraphPad Prism v10	Dotmatics	https://www.graphpad.com
HALO AI 4.0.5107.445	Indica Labs	https://indicalab.com/halo-ai/

Open in a new tab

Data and Code Availability

All analysis results are publicly available at https://cide.ccr.cancer.gov. Figure source data and processed data from open-access studies are available at https://doi.org/10.5281/zenodo.15012968. For in-house processed data from restricted access studies, we listed the application links at https://cide.ccr.cancer.gov/download/ and will email users our data upon receiving the access approval. The TCGA data are available through https://xenabrowser.net/datapages/?cohort=TCGA%20Pan-Cancer%20(PANCAN)&removeHub=https%3A%2F%2Fxena.treehouse.gi.ucsc.edu%3A443. The ICGC data are available through https://dcc.icgc.org. The CPTAC data are available through https://proteomic.datacommons.cancer.gov/pdc/cptac-pancancer. The TARGET data are available through https://portal.gdc.cancer.gov. The PRECOG data are available through https://precog.stanford.edu. The tumor transcriptomics data from studies without immunotherapies at the TIDE database are available through http://tide.dfci.harvard.edu. For RNA-seq data generated in this study, we deposited all of them into the NCBI GEO with accessions in the Key Resources Table. The code for clinical association analysis is available at https://github.com/data2intelligence/CIDE_main. The code for gene prioritization is available at https://github.com/data2intelligence/CIDE_Prioritization.

STAR Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell lines and cell culture

All cell lines were routinely tested for mycoplasma via a PCR-based method using the Universal Mycoplasma Detection Kit (ATCC) every month. We confirmed with commercial and collaborative sources from which the cells were obtained that the cell lines were authentic and free of cross-contaminations.

B16F10 (sex of cell: male), B16-mhgp100 (sex of cell: male), and RenCa (sex of cell: male) cells were maintained in RPMI-1640 with GlutaMAX^™ (Thermofisher), 10% FBS (Gibco), 20 mM HEPES (Gibco), 100 IU ml⁻¹ penicillin/streptomycin (Gibco), and 1/500 Plasmocin^® prophylactic (Invivogen). RIL-175 cells (sex of cell: not reported) were maintained in DMEM (Thermofisher) with 10% FBS (Gibco), 100 IU ml⁻¹ penicillin/streptomycin (Gibco), and 1/500 Plasmocin^® prophylactic (Invivogen). Jurkat-Lucia^™ NFAT-CD28 Cells (Invivogen, sex of cell: male) were maintained in IMDM (Thermofisher), 2 mM L-glutamine (Gibco), 25 mM HEPES (Gibco), 10% HI-FBS, 100 IU ml⁻¹ penicillin/streptomycin (Gibco), 100 μg/ml Normocin^™, as per manufacturer’s instructions. All cell lines were cultured in a humidified, 5% CO2, 37°C incubator.

Human primary CD8+ T cell isolation

The collection of peripheral blood from healthy donors was approved and performed at the NIH Blood Center. The donor (male, 68 years old) consented to blood being used in the study. Blood samples were mixed with Ficoll-Paque PLUS (Cytiva), and this mixture was centrifuged at 400× g with the lowest acceleration and deceleration speed for 25 minutes at room temperature. The peripheral blood mononuclear cells (PBMC) were carefully collected using a Pasteur pipette, centrifuged at 300×g for 10 minutes, and washed with PBS containing 1% FBS (Gibco) and 0.5% EDTA. After centrifugation, the PBMCs were incubated in ACK lysing buffer (Gibco) for 5 minutes at room temperature to lyse RBCs and then neutralized by RPMI-1640 with 10% FBS and centrifuged at 300×g for 5 minutes. The PBMCs were then resuspended in EasySep buffer (STEMCELL Technologies) at 5E7 cells/mL. Human CD8+ T cells were isolated using the EasySep^™ Human CD8+ T Cell Isolation Kit (STEMCELL Technologies), as per manufacturer’s instructions.

Mouse primary CD8+ T cell isolation

Spleens from 8–10-week-old C57BL/6 mice were minced on a 70-μm cell strainer (Falcon) to dissociate spleenocytes. After centrifugation, the splenocytes were incubated in ACK lysing buffer (Gibco) for 2 minutes on ice to lyse the red blood cells (RBCs), neutralized by RPMI-1640 with 10% FBS, and centrifuged at 300×g for 5 minutes. The splenocytes were then resuspended in an EasySep buffer (STEMCELL Technologies) at 1E8 cells/mL. Mouse CD8+ T cells were isolated using the EasySep^™ Mouse CD8+ T Cell Isolation Kit (STEMCELL Technologies), as per manufacturer’s instructions.

Mouse and human CD8+ primary T cell culture

Mouse and human CD8+ T cells were cultured in TexMACS^™ medium (Miltenyi) supplemented with 10% HI-FBS and 50 μM β-mercaptoethanol (Gibco) at 1–1.5E6 cells/1–1.5mL, and incubated at 37 °C, 5% CO2. Prior to TCR activation, 1.5E6 cells/mL CD8+ T cells were pre-treated with 10 μg/mL rhAOAH or Vehicle in the complete T-cell medium without rhIL-2 for 20–22 hours.

Animal models

NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice, pmel-1 transgenic mice, and OT-1 transgenic mice were bred at the National Cancer Institute, National Institutes of Health (Maryland, USA). C57BL/6J and BALB/c mice were purchased from Charles River Laboratories (Massachusetts, USA). Pmel-1 mice were from Dr. Glenn Merlino’s lab, and OT-1 mice were from Dr. Grégoire Altan-Bonnet’s lab. Female mice between 6 and 12 weeks old with an average weight from 16–20g were used in the study and randomized into treatment groups. The maximal tumor size permitted by the ethics committee is 2 cm in diameter, and no tumor burden exceeded that limit in our experiment.

All procedures involving animals were compiled according to the ethical regulations of animal studies at NIH, according to the animal protocol CDSL-001 (approved by the NCI Animal Ethics Committee of NIH), or conducted under protocol 23–480 (approved by the Committee on the Use of Live Animals in Teaching and Research at HKU). Each cage contained 5 mice. Mice were housed on 12 h light–dark cycles with the ambient temperature at 20–25 °C and 40–60% humidity.

Tumor in vivo model selections

To validate the roles of secreted proteins predicted by CIDE, we selected mouse tumor models (Table S7A) with one criterion that the anti-PD1 sensitivity of the model should be opposite to the CIDE-predicted role (i.e., anti-tumor or pro-tumor) of the candidate gene. Our model selection has three parts:

The universal model selection for anti-tumor secreted proteins (AOAH, CR1L, and COLQ) utilized the B16-mhgp100 model, which is anti-PD1 resistant (Figure 3E, F), because we want to test whether the induction of these proteins can sensitize B16-mhgp100 tumors to anti-PD1.
The universal model selection for pro-tumor secreted proteins (ADAMTS7) utilized the B2905-M4 model, which is anti-PD1 sensitive (Figure 3E, F), because we want to test whether induction of ADAMTS7 can de-sensitize B2905-M4 tumors to anti-PD1.
As our in-depth study focuses on AOAH, we included additional tumor models, such as B16F10 (melanoma), Renca (renal), RIL-175 (hepatocellular), and MYC-Luc, MYC-LucOS (hepatocellular). These models are all resistant to immune checkpoint blockade, as we want to test whether the AOAH induction can sensitize therapy-resistant models to anti-PD1/CTLA4.

Human immunotherapy clinical datasets

We systematically collected cancer immunotherapy datasets from published studies to our best awareness. More than half of the publications required additional data access applications. We submitted application forms and kept following up with authors once per month until we got approvals, rejections, or no replies for six months. All applications to the dbGaP database⁷⁸ were approved. For the EGA database⁷⁹, four applications were rejected, and all other applications were approved. All publicly available and access-approved datasets are listed in Table S1.

METHOD DETAILS

Resolve clinical data inconsistency

The key analysis of this study is computing the association between gene molecular status and clinical outcomes upon cancer immunotherapies. However, association scores from clinical data tend to have considerable inconsistencies, even with the same cancer type, the same type of treatment, and the same endpoint (examples in Figure S2E). The inconsistency of results among similar studies may derive from the following reasons.

Noise: The associations between tumor molecular profiles and clinical outcomes are inherently noisy. First, many cohorts have low sample sizes (e.g., 20 patients for the Melanoma anti-CTLA4 to anti-PD1 Campbell 2023⁸⁰; 25 for the Melanoma Hugo 2016⁸¹). Second, the biopsy procedures, such as needle biopsy (e.g., Breast Pusztai 2021⁸² and Wolf 2022⁸³ cohorts), take a sub-portion of one tumor out of a patient’s many tumors in metastatic settings; thus, the biopsy cannot represent tumor microenvironments in the whole patient. Third, the patient outcomes do not entirely depend on tumors. Other factors, such as supportive care and life qualities, may also determine therapy outcomes.
Biological variances: The anti-PD1/PDL1 antibodies differed across studies (Nivolumab, Pembrolizumab, Atezolizumab, Avelumab, etc). Moreover, different clinical centers may have distinct treatment protocols, such as immunotherapy cycles, stop criteria, chemotherapy regimens, radiation doses, and timing.
Demographic differences: As different studies are from various locations, variations in age, gender, ethnicity, and local environment (e.g., diet, microbiome, pollution, local infections) can influence both gene expression and outcomes.
Study designs: Different studies may utilize formalin-fixed paraffin-embedded (FFPE) or fresh frozen samples. For FFPE samples, different studies may or may not enrich tumor regions, thus including distinct portions of immune and stroma across cohorts.

All factors above can induce random variations of risk score profiles, which lead to opposing correlation values and insufficient false discovery rates in most cohorts. To mediate all the challenges above, we have utilized the following solutions:

Meta-analysis: We prioritized secreted protein-coding genes by pooling data from multiple cohorts to increase statistical power and identify robust associations. In other words, even if noise may exist on risk scores from individual profiles, the overall trend of certain genes can still pass rigorous statistical thresholds based on false discovery rates. Thus, we only focus on the top candidates from the meta-analysis (Method sections “Gene prioritizations for secreted proteins” and “Alternative meta-analysis methods”).
Harmonization: We used standardized protocols for processing and normalizing RNA and whole-exome sequencing data to minimize technical variabilities due to data analysis procedures (Figure S1A, B, and Method sections “Omics data processing” and “Quality controls”).
Covariate control: When testing the associations between molecular status and clinical outcomes, we account for subgroup effects (e.g., molecular subtypes, demographic factors, batches) that might influence results as covariates in regressions (Table S5).
Experimental validation: We validate data-driven findings through orthogonal wet lab experiments, including in vitro cell models and in vivo mouse models. The robust wet-lab phenotypes related to AOAH, CR1L, COLQ, and ADAMTS7 independently supported the robustness of conclusions from our data analysis (Figure 3 and Figure 4).

Omics data processing

For studies with raw data available, we uniformly processed their RNA-seq and whole-exome sequencing data (WES) with our in-house omics analysis framework (Figure S1A). The RNA-seq output contains gene and isoform levels in transcript per million (TPM) with the log2(TPM+1) transformation. The WES output contains copy number alterations (CNA) and weighted somatic mutations at the gene level.

The somatic mutation weights are computed as follows. For MuTect2, the TLOD scores are first trimmed between 0 and 200 and linearly scaled to 0 and 1. For VarScan, the SSC scores are trimmed between 0 and 255 and linearly scaled to 0 and 1. For Strelka, the QSS_NT or QSI_NT (if QSS_NT is unavailable in the output) scores are trimmed between 0 and 3070 and then linearly scaled to 0 and 1. Finally, the mutation weight is the combined confidence among three mutation calls using the noisy-or $1 - \prod_{i = 1}^{3} (1 - confidence_i)$ .

For studies without raw data released, we downloaded the processed data from either supplementary materials of the paper or source code repositories (Table S1). We listed download sources of non-immunotherapy omics cohorts, including TCGA, ICGC, CPTAC, TARGET, PRECOG, and other clinical cohorts from the TIDE frameworks, in the “Deposited data” section of the Key Resources Table.

Quality controls

To filter out omics profiles with low quality, we implemented quality control (QC) metrics: 1, coverage; 2, species/lane mix-ups; 3, tumor purity; and one additional filter of 4, sample capture method.

Coverage

We focused on read coverage on protein-coding exons using the package mosdepth⁴⁵ with bam files as the input. We only kept samples with at least 10X exon coverage (i.e., on average 10 reads per position in protein-coding exons). For the RNA-seq datasets, there are two datasets with median exon coverage lower than 10 (Figure S1B). Thus, we eliminated these two datasets. The median coverage in all other datasets is higher than 10. Therefore, we only filtered individual samples with less than 10X coverage. For the WES, one dataset’s median coverage is very close to 10X (Figure S1B). Thus, we removed this dataset and only filtered individual samples in other datasets using the 10X threshold.
Species/lane mix-ups

We computed the fraction of uniquely mapped reads per genome using the package fastq_screen⁴⁶ with fastq files (adapter trimmed and ribosomal RNA filtered) as the input. The reference genomes included Human, Mouse, Rat, Drosophila, Worm, Yeast, Arabidopsis, Ecoli, along with PhiX, vectors, or other contaminants commonly seen in sequencing experiments. To detect potential species/lane mix-ups, we computed the “Top-one genome fold” as “fractions of reads uniquely mapped to human” / “fractions of reads uniquely mapped to the 2nd most frequent genome”. For both RNA-seq and WES data, all top-one genome fold values are higher than 50 (Figure S1B). Thus, we have not detected obvious species/lane mix-ups.
Tumor purity

Two datasets^47,48 provided tumor purity estimations. Thus, we utilized the authors’ original values. For other datasets, we estimated the tumor purities as follows: For WES data, we utilized the package ASCAT ascatngs⁵⁰ with input bam files from the tumor and normal samples. For gene expression data, we utilized the R package ESTIMATE⁴⁹ tidyestimate with gene expression profile as the input. We did not use tumor purities estimated by WES data and ASCAT for transcriptomics data because only 18 out of 50 transcriptomics cohorts have matched WES data. Further, only 8 out of 18 datasets clearly stated that the DNA and RNA samples were co-purified from the same sample. Other studies either purified DNA and RNA from separate samples or did not provide the details. Previous studies^80,84 typically use purity < 10% as the cut-off. No RNA-seq or WES samples failed this cut-off (Figure S1C). Thus, we included the tumor purity as a background covariate to adjust the regression analysis for the top candidates.
Sample capture method

One RNA-seq cohort has both frozen and FFPE samples profiled for the same set of patients⁸⁵. We only used frozen samples in the analysis, as frozen samples generally have much better RNA quality than FFPE samples⁸⁶. Other cohorts have the same sample capture methods for all samples to our best interpretation of each paper’s Methods sections.

Batch effects and clinical covariates

The omics data cohorts contain many heterogeneities, such as batch effects and diverse clinical covariates, which may affect the associations between molecular data and clinical outcomes. Thus, we added batch effects and clinical covariates in our regressions instead of directly correcting the transcriptomics data using software like combat⁸⁷. Previous studies demonstrated that correcting for batch directly in the data could bias group differences, and including batch as a covariate in statistical models is a better alternative⁸⁸. The criteria for preparing covariates are listed as follows:

Batch

Several studies have samples collected separately from distinct batches, sites, institutions, or countries. We treat this information as covariates in the multivariate regression, testing the association between molecular status and clinical endpoints. Please see the column “Batch” in Table S5.
Disease types
- Option 1: Stratification: If a cohort contains multiple subtypes, we split it into several datasets, each covering only one subtype with a tumor count >= 15. The disease type split is shown in the column “Disease” in Table S5. Two melanoma studies have patients naive to Ipilimumab or progressed after Ipilimumab. The other breast cancer and non-small lung cancer studies included subtypes.
- Option 2: Regression: If the stratification of disease types leads to a sample size < 15 in a subtype, we encode the subtype information as covariates in the multivariate regression testing the association between molecular status and clinical endpoints. In particular, two pan-cancer studies^89,90 included patients from diverse disease types. Please see the column “Subtype” in Table S5.
Other covariates
Besides batch and disease types, many other factors may affect the clinical outcome and the association between molecular status and clinical outcomes. Thus, we aim to extract covariates from the clinical information file of each study for factors satisfying the following criteria:
1. For string factors, the information can be encoded as numerical values;
2. The factor does not contain null values for > 5% of cases;
3. Covariates encoded from the factor do not contain only one dominant value for an individual covariate (i.e., the same value for over 95% of cases).

Threshold selection in survival plots

In survival analyses of human clinical data, Z-scores and P-values were computed by the two-sided Wald test in the Cox-PH regression using continuous values without any cutoffs. However, the kaplan meier (KM) plot requires a threshold to cut target gene expression into high versus low values for visualization purposes. We used the best separation procedure. First, we rank all patients by the molecular level of the gene in the test. Second, at each threshold position, we cut the gene value into 1 (higher than the current threshold) and 0 (otherwise) and run the Cox-PH regression on binary gene values to compute the z-score associated with the target gene covariate. Finally, we select the threshold where the z-score had the same sign (+/−) as the z-score from continuous regression without a cutoff and achieved the largest absolute value.

Interactive CIDE server

We created an interactive web server presenting all analysis results and source data of open-access cohorts using the Django framework and MySQL database. This server is hosted on the Amazon Web Services cloud by NCI, NIH at https://cide.ccr.cancer.gov. The CIDE server presents three basic functionalities:

Search: Query whether a gene is associated with cancer clinical outcome.
Prioritize: Input a gene set and prioritize top candidates for follow-ups.
Evaluate: Compare the input biomarker and other published biomarkers based on their predictive power for immunotherapy clinical outcomes.

Immunotherapy response biomarker analysis

We included the following biomarkers from previous literature that aim to predict immune checkpoint blockade outcomes.

T effector	Median expression of the gene set: CD8A, GZMA, GZMB, IFNG, EOMES, CXCL9, CXCL10, TBX21 from the POPLAR trial⁹¹.
TIDE.exclusion^*	Correlation between the input gene expression profile and T-cell exclusion signature⁵².
Cytotoxic T	Median expression of the gene set: CD8A, CD8B, GZMA, GZMB, PRF1⁵².
CytoSig.IFNG	The IFNG activity score computed by CytoSig⁵³.
CytoSig.TGFB1	The TGFB1 activity score computed by CytoSig⁵³.
Cytolytic	Median expression of the gene set: GZMA, GZMB, PRF1⁹²
IMPRES	A prediction score using 15 gene pairs⁹³.
CD8	Median expression of CD8A and CD8B.
IFNG6	Median expression of the gene set: IDO1, CXCL10, CXCL9, HLA-DRA, STAT1, IFNG defined in a previous study¹⁴
CXCL13	Gene expression of CXCL13, one top predictor of immunotherapy response in a previous meta-analysis⁹⁴.
CD38	Gene expression of CD38, one top predictor of immunotherapy response in a previous meta-analysis⁹⁴.
PDL1	Gene expression of CD274. PDL1 is an immunohistochemistry biomarker for anti-PD1/PDL1 treatment.
IFNG	Gene expression of IFNG.
TIDE.geneset	A geneset version of TIDE biomarker⁵² provided on the TIDE website, computed as the median expression of CD274 and IFNG minus the median expression of SERPINB9, TGFB1, FAP, VEGFA, and ANGPT2.
B.Clonality	Clonality of the BCR IGH sequences called from bulk RNA-seq data⁴³.
mutation_burden	log2(Sum of nonsynonymous mutations in coding regions + 1)
BATF3_DC	Median expression of the gene set: BATF3, IRF8, THBD, CLEC9A, XCR1, representing the level of tumor-residing BATF3 dendritic cells⁹⁵.
TIDE^*	Tumor immune dysfunction and exclusion score, defined in a previous study⁵².
T.Clonality	Clonality of the T-cell receptor beta CDR3 sequences called from bulk RNA-seq⁴³.
Aneuploidy^*	Sum of copy number alteration absolute values divided by the total chromosomal size, adapted from a previous study⁹⁶.
TGFb^*	Median expression of TGFB1 and TGFBR2 as in a previous study⁹⁷.

Open in a new tab

Negative biomarkers. We reversed their value signs when computing the association between biomarker values and clinical outcomes.

For each endpoint type, we tested the prediction metrics as associations between biomarker values computed using pretreatment transcriptomics and clinical outcomes with all possible clinical covariates in regressions (Table S5). For the binary outcome, performance metrics include the Area Under the receiver operating characteristic Curve (AUC) and z-scores from the two-sided Wald test in the logistic regression with Firth correction. For the Response Evaluation Criteria in Solid Tumors (RECIST) outcomes, the performance metric is computed as the t-value from the two-sided t-test in the linear regression. For the progression-free and overall survival outcomes, performance metrics are z-scores from the two-sided Wald test in Cox-PH regressions.

Spatial transcriptomics data analysis

We downloaded two datasets of Visium spatial transcriptomics (ST) from hepatocellular carcinoma patients treated with anti-PD1^20,21. To compare the expression of AOAH between responders and non-responders, we aggregated the counts of all spatial spots to generate a pseudo-bulk for each ST sample, normalizing the data to log2(TPM+1). The expression level of AOAH in ST samples was visualized using the R package SpaCET⁷⁴.

Blinded literature annotation

The computational personnel first prioritized secreted protein-coding genes involved in cancer immunotherapy response (Methods section Gene prioritizations for secreted proteins). Then, we randomized the order of genes and masked the predicted anti-tumor or pro-tumor functions. For each secreted protein-coding gene prioritized by CIDE, the immunologist only got the shuffled gene list and only considered previous studies with genetic perturbation data to annotate potential cancer functions in five categories:

Anti-tumor;
Pro-tumor;
Conflict: previous literature showed both anti-tumor and pro-tumor functions;
No effect: previous studies showed that the genetic perturbation of the candidate gene did not alter tumor growth.
Unknown: cannot find any previous literature.

The blinded annotation was used to compute the prioritization performance metrics in Figure 2C, D. After the blinded annotation, we revealed the predicted anti-tumor or pro-tumor directions and further searched the literature, focusing on genes in the Unknown category. Figure 2B shows both the blinded (solid color) and unblinded (transparent color) annotations.

According to our literature annotations, only 10 out of 136 prioritized genes (7.4%) have no known cancer functions (Table S6), while 61.1% of human genes coding secreted proteins do not have known functions in cancer by literature mining⁶. This gap is likely explained by the prioritization process itself, which selected for genes that are robustly expressed in cancer cohorts and are therefore more likely to have roles in cancer and be well-studied. Further, the 92.6% (126/136) genes that were annotated to have a role in cancer may be an overestimate of the actual number because many studies used for annotation lack rigorous experimental controls, such as perturbation rescues, independent models, and in vivo validations.

Single-cell RNA-seq data analysis

Cell lineage classification

We used the R package Seurat⁷⁵ to process and merge eight scRNA-seq samples from hepatocellular carcinoma (HCC) tumors. Quality control was conducted to filter out the cells with unique feature counts > 4000, <500, or >15% mitochondrial counts. We then normalized the total counts in each cell to 10,000, followed by log2 transformation to generate the normalized data. We applied Uniform Manifold Approximation and Projection (UMAP) analysis to the first 30 principal components (PCs) for dimensional reduction and data visualization. We clustered the cells using FindClusters from Seurat, which is a shared nearest neighbor (SNN) modularity optimization-based clustering algorithm. We used Scrublet⁷⁶ to calculate doublet scores. The marker genes of major lineage and sub-lineage are listed as follows (adapted from https://www.cellsignal.com/pathways/immune-cell-markers-mouse).

Cell type	Marker gene
Major lineage
Plasma	Jchain
B	Cd19
T	Cd3e
NK	Nkx1-1, Ncr1, Klrk1
MoMΦ	Cd68, Cd14, Adgre1
PMN-MDSC	Ly6g
pDC	Siglech, Bst2
cDC	Xcr1, Clec9a
T/NK cluster
Tprolif	Mki67
CD4	Cd4
Th1	Stat4, Tbx21, Cxcr3, Ifngr1
Tfh	Bcl6, Cxcr5, Il21
Treg	Foxp3
CD8	Cd8a, Cd8b1
CTL	Ifng, Gzma, Gzmb, Prf1
CD8 Tex	Pdcd1, Tigit, Lag3, Havcr2

Open in a new tab

We cannot assign any known lineages in our cell marker table for four T-cell clusters with the following cluster IDs: 2-Tunknown, 4-Tunknown, 8-Tunknown, and 11-Tunknown. To identify the potential sub-lineages for these T cells, we performed differential analyses of gene markers and signatures.

First, within each tumor, we computed the average gene expression profiles across all cells for each T-cell cluster. To analyze the molecular characteristics of each T-cell cluster, we further computed the log2 fold-change between each Tunknown cluster and the average across all T cells. Then, we compared the log2FC values across 8 tumors and zero using the two-sided Wilcoxon signed-rank test to identify markers with significant positive or negative values. Then, we converted the p-values to false discovery rates (FDR) using the Benjamini-Hochberg correction. By studying up-regulated or down-regulated genes, we can assign three Tunknown clusters into lineages with existing literature support.

11-Tunknown: Cd4 and Cd8 double negative, high Bcl2 as the top-one up-regulated gene. The transcriptomic characteristics of this cluster are similar to a population of premature thermocytes⁹⁸.

Gene	mean	std	p	FDR
*Bcl2*	4.59	0.65	0.00781	0.021
*Cd4*	−1.23	0.23	0.00781	0.021
*Cd8a*	−1.62	0.32	0.00781	0.021
*Cd8b1*	−2.02	0.44	0.00781	0.021

Open in a new tab

2-Tunknown: Double negative, high Themis, as the top one upregulated gene in this cluster, but not in other Tunknown clusters. Themis can promote T cell development and maintenance⁹⁹.

Gene	mean	std	p	FDR
*Themis*	2.54	0.15	0.00781	0.019
*Cd4*	−0.15	0.11	0.0156	0.033
*Cd8a*	−0.39	0.29	0.0156	0.033
*Cd8b1*	−0.80	0.44	0.0156	0.033

Open in a new tab

8-Tunknown: TGFβ anergic. T cells in this cluster express high Cblb, a marker of anergic T cells, and high Itgav and Itgb8, which activate TGFβ in the extracellular space¹⁰⁰. Previous studies show that avb8-high T cells exist in tumors¹⁰¹.

Gene	mean	std	p	FDR
*Itgav*	2.44	0.63	0.00781	0.021
*Itgb8*	1.76	0.33	0.00781	0.021
*Cblb*	2.41	0.56	0.00781	0.021
*Cd4*	0.19	0.32	0.109	0.181
*Cd8a*	−1.34	0.26	0.00781	0.021
*Cd8b1*	−2.00	0.47	0.00781	0.021

Open in a new tab

Consistent with this, CytoSig revealed that T cells in this cluster have high TGFB3 and TGFB1 signaling activities.

Signal	Coef	StdErr	Z-score	P-value
TGFB3	0.041	0.004	9.948	< 0.001
TGFB1	0.019	0.004	4.684	< 0.001

Open in a new tab

4-Tunknown: We cannot explain any top markers of this T-cell cluster, thus label this cluster as Cmss1, the top-one up-regulated gene of this cluster.

Differential expression profiles

For each HCC tumor, we first computed the mean log2(CPM/100+1) across all cells within each lineage to get a lineage-specific transcriptomics profile. Then, for each cell type in HCC tumors with or without the OS antigen, we computed the log2 fold change of expression between tumors with Aoah vs vector overexpression.

V(D)J receptor clonotype analysis

We utilized the 10X CellRanger software to compute T-cell receptor (TCR) and B-cell receptor (BCR) clonotypes. During the clonotype grouping stage, cell barcodes are placed in groups called clonotypes. Each clonotype consists of all descendants of a single, fully rearranged common ancestor, as approximated computationally by the 10x enclone framework¹⁰². For T cells, the lack of somatic hypermutation in TCRs yields clonotypes with identical V(D)J transcripts. For B cells, fully rearranged BCRs can undergo somatic hypermutations, which can increase antigen affinity. Thus, for BCRs, V(D)J transcripts in a clonotype can differ in any position.

To quantify the level of TCR and BCR clonal expansion, we computed the clonality, a metric commonly used in TCR analysis¹⁰³. The range of clonality is within [0, 1].

Clonality = 1 - normalized Entropy = 1 - Entropy / ln(n), where $Entropy = - \sum_{i = 1}^{n} p_{i} l n (p_{i})$

n: total number of cells

p_i: fraction for a specific clonotype i.

Lipid class enrichment analysis

For cell culture medium or pellets, the log2FC values of a lipid species were computed as the log2 fold change of MassSpec areas between rhAOAH and Vehicle (PBS + 10% glycerol) treatments. Then, we defined a lipid configuration as the joint of class (e.g., glycerophosphocholines, etc.) and subclass (e.g., diacyl, etc.) keys, and grouped log2FC values by the lipid configuration. For a configuration with at least three lipid species, we compared the log2FC difference between lipids within and outside the configuration by the two-sided Wilcoxon rank-sum test. The p-values were converted to false discovery rates (FDR) by the Benjamini-Hochberg correction, and FDR < 0.05 was the threshold to select significant results.

For phosphatidylcholine, we also analyzed the fatty acid enrichment at the sn-1 and sn-2 positions. For each configuration pair (fatty acid, sn position), we compared log2FC values of lipids with this configuration against values without this configuration using the two-sided Wilcoxon rank-sum test. The p-values were converted to FDRs by the Benjamini-Hochberg correction, and FDR < 0.05 was the threshold to select significant results.

Hydrodynamic tail vein injection (HDTVi)

We used two murine hepatocellular carcinoma models driven by MYC and beta-catenin (CTNNB1) overexpression with (LucOS) or without (Luc) highly immunogenic OVA+SIN+SIY (OS) antigens²³. Both models show resistance to anti-PD1 immunotherapy. A sterile 0.9% NaCl solution/plasmid mixture was prepared as follows: We prepared 10 μg of pT3-EF1a-MYC-IRES-luciferase (MYC-Luc), 10 μg of pT3-EF1a-MYC-IRES-luciferase-OS (MYC-LucOS), 10 μg of pT3-N90-CTNNB1 (CTNNB1), 10 μg of PKT2/CLP-Vector, 10 μg of PKT2/CLP-Aoah, and a 3:1 ratio of transposon to SB100 transposase–encoding plasmid dissolved in 2 mL of 0.9% NaCl solution and tail-vein injected 10% of the weight of each C57BL/6 mouse in volume in 5–7 seconds. For the anti-PD1 experiment, treatments were initiated 8–9 days after hydrodynamic delivery of plasmids, a time point that already presents malignant tumor cells. Anti-PD1 (BioXCell, #BE0146, 200 μg/dose) was given twice per week for a total of 4 doses. Bioluminescence imaging was taken twice a week. Vectors for hydrodynamic delivery were produced using the EndoFree Maxi Plasmid kit (TIANGEN).

Cell Proliferation Assay

B16-mhgp100 cells and B2905-M4 were seeded at 1,000 cells/well in a 96-well plate and incubated in a 37 °C humidified CO2 incubator. After overnight incubation, the number of cells was determined by the Cell Proliferation Kit II (XTT, Roche) every day for 5 days (B16-mhgp100) or 4 days (B2905-M4), as per the manufacturer’s protocol. The absorbance was measured at 460 nm with a reference wavelength of 620 nm using a microplate reader.

Mouse and human CD8+ T cell activation

For antibody-based activation assays, 1 μg/mL, 0.5 μg/mL, and 0.1 μg/mL anti-mouse/human CD3 antibody (Tonbo) in PBS was coated in a 96-well plate at 50 μL per well and incubated at 37°C for 2 hours. After incubation, the coated wells were washed with sterile PBS twice, and 3E5 CD8+ T cells in 200 μL complete T-cell medium supplemented with 100 IU/mL recombinant human IL-2 (R&D Systems), and 0.5 μg/mL anti-mouse/human CD28 antibody were seeded into each well.

The CD8+ T cells were activated for a short period of time (5, 10, and 20 minutes) to evaluate TCR signaling cascades and for a longer period of time (24 hours) to quantify the fraction of activated T cells and the release of cytokines.

For the antigen-specific activation assays, splenocytes isolated from 8–10-week-old C57BL/6 mice were pulsed¹⁰⁴ with hgp100 (Genescript), mgp100 (Genescript), OVA-N4, OVA-V4 (from Grégoire Altan-Bonnet lab) antigenic peptides at varied concentrations (1 μM, 100 nM, 10 nM, 1 nM) in complete RPMI-1640 at 37°C for 2 hours. After incubating, the pulsed splenocytes were washed with complete RPMI-1640 twice and centrifuged at 300×g for 5 minutes. The pulsed splenocytes were seeded into a round-bottom 96-well plate at 3E5/100 μL per well, containing complete T-cell medium with 100 IU/mL recombinant human IL-2. The pmel and OT-1 mouse CD8+ T cells were mixed with the pulsed splenocytes at 1E5/100 μL per well. The antigen-specific CD8+ T cells were activated for 24 and 48 hours to quantify the fraction of cytotoxic T cells and the release of cytokines.

Lentivirus transfection and transduction

Full-length cDNA of luciferase and murine Aoah were constructed into pLenti6.3/V5-TOPO (Invitrogen, K5310–00) and pReceiver-Lv156-EF1-Vector (GeneCopoeia, EX-Lv156) by homologous recombination, respectively. The plasmid DNA was transformed into One Shot^™ Stbl3^™ Chemically Competent E. coli (Invitrogen) via heat shock, and was then purified with QIAprep Spin Miniprep Kit (QIAGEN), as per manufacturer’s instructions. 1E6 per well Lenti-X^™ 293T cells (Takara) were seeded onto a 6-well plate. 24 hours after cell seeding, two packaging plasmids psPAX2 (Addgene, 12260) and pMD2.G (Addgene, 12259), together with the pReceiver-Lv156-EF1-Vector or pReceiver-Lv156-EF1-Aoah/Cr1l/Colq/Adamts7 plasmid, were transfected into the 293T cells using the Lipofectamine 3000 transfection reagent (Invitrogen), according to the manufacturer’s protocol.

The culture medium was replaced with Opti-MEM^™ with GlutaMAX^™ (Gibco), supplemented with 1:500 ViralBoost Reagent (Alstem) 6 hours after transfection. The lentivirus was harvested by spinning down the viral supernatant at 1,000× g, 4°C for 15 minutes to remove the cell debris at 48 and 72 hours post-transfection. For lentivirus transduction, B16-mhgp100, B16F10, B2095-M4, RIL-175, RenCa, and DC2.4 cells were mixed with lentivirus at a 1:1 dilution with culture medium, 10 μg/mL polybrene (Sigma-Aldrich) was added to the mixture for 24–48 hours before refreshing the medium. Two days after infection, puromycin (Sigma-Aldrich) was added for the selection and maintenance of cells with gene overexpression. RIL-175-Aoah cells co-infected with lentivirus harboring luciferase were further selected using blasticidin (Sigma-Aldrich). For all cell models, qRT-PCR confirmed the successful gene overexpression. In addition to qRT-PCR, for B16-mhgp100 and DC2.4, ELISA confirmed the successful Aoah transduction (source data file “ELISA_qPCR.xlsx” in the Zenodo (https://doi.org/10.5281/zenodo.15012968)).

AOAH recombinant human protein production

The recombinant human AOAH proteins (rhAOAH) with His-tag or Fc-fusion were purchased from a Contract Research Development company (WuXi Biologics). Briefly, human recombinant proteins were purified from the 1L cell-culture supernatant of HEK293 cells with AOAH construct overexpression. The final buffer is PBS pH7.4 with 10% glycerol. Protein sequences are available below. Quality control reports are available in the Zenodo (https://doi.org/10.5281/zenodo.15012968).

1)
His-tag protein sequence:

MGWSCIILFLVATATGVHS (Signal peptide) HHHHHH (His-tag) DDDDK (Enterokinase cleavage site) LSNGHTCVGCVLVVSVIEQLAQVHNSTVQASMERLCSYLPEKLFLKTTCYLVIDKFGSDIIKLLSADMNADVVCHTLEFCKQNTGQPLCHLYPLPKETWKFTLQKARQIVKKSPILKYSRSGSDICSLPVLAKICQKIKLAMEQSVPFKDVDSDKYSVFPTLRGYHWRGRDCNDSDESVYPGRRPNNWDVHQDSNCNGIWGVDPKDGVPYEKKFCEGSQPRGIILLGDSAGAHFHISPEWITASQMSLNSFINLPTALTNELDWPQLSGATGFLDSTVGIKEKSIYLRLWKRNHCNHRDYQNISRNGASSRNLKKFIESLSRNKVLDYPAIVIYAMIGNDVCSGKSDPVPAMTTPEKLYSNVMQTLKHLNSHLPNGSHVILYGLPDGTFLWDNLHNRYHPLGQLNKDMTYAQLYSFLNCLQVSPCHGWMSSNKTLRTLTSERAEQLSNTLKKIAASEKFTNFNLFYMDFAFHEIIQEWQKRGGQPWQLIEPVDGFHPNEVALLLLADHFWKKVQLQWPQILGKENPFNPQIKQVFGDQGGH
2)
His-tag protein with S263A mutation:

MGWSCIILFLVATATGVHS (Signal peptide) HHHHHH (His-tag) DDDDK (Enterokinase cleavage site) LSNGHTCVGCVLVVSVIEQLAQVHNSTVQASMERLCSYLPEKLFLKTTCYLVIDKFGSDIIKLLSADMNADVVCHTLEFCKQNTGQPLCHLYPLPKETWKFTLQKARQIVKKSPILKYSRSGSDICSLPVLAKICQKIKLAMEQSVPFKDVDSDKYSVFPTLRGYHWRGRDCNDSDESVYPGRRPNNWDVHQDSNCNGIWGVDPKDGVPYEKKFCEGSQPRGIILLGDAAGAHFHISPEWITASQMSLNSFINLPTALTNELDWPQLSGATGFLDSTVGIKEKSIYLRLWKRNHCNHRDYQNISRNGASSRNLKKFIESLSRNKVLDYPAIVIYAMIGNDVCSGKSDPVPAMTTPEKLYSNVMQTLKHLNSHLPNGSHVILYGLPDGTFLWDNLHNRYHPLGQLNKDMTYAQLYSFLNCLQVSPCHGWMSSNKTLRTLTSERAEQLSNTLKKIAASEKFTNFNLFYMDFAFHEIIQEWQKRGGQPWQLIEPVDGFHPNEVALLLLADHFWKKVQLQWPQILGKENPFNPQIKQVFGDQGGH
3.1)
Fc-fusion protein part 1:

MGWSCIILFLVATATGVHS (Signal peptide) DKTHTCPPCPAPELLGGPSVFLFPPKPKDTLMISRTPEVTCVVVDVSHEDPEVKFNWYVDGVEVHNAKTKPREEQYNSTYRVVSVLTVLHQDWLNGKEYKCKVSNKALPAPIEKTISKAKGQPREPQVYTLPPSRDELTKNQVSLWCLVKGFYPSDIAVEWESNGQPENNYKTTPPVLDSDGSFFLYSKLTVDKSRWQQGNVFSCSVMHEALHNHYTQKSLSLSPG (human IgG1) GGGGSGGGGSGGGGS (3 repeats of G4S linker) LSNGHTCVGCVLVVSVIEQLAQVHNSTVQASMERLCSYLPEKLFLKTTCYLVIDKFGSDIIKLLSADMNADVVCHTLEFCKQNTGQPLCHLYPLPKETWKFTLQKARQIVKKSPILKYSRSGSDICSLPVLAKICQKIKLAMEQSVPFKDVDSDKYSVFPTLRGYHWRGRDCNDSDESVYPGRRPNNWDVHQDSNCNGIWGVDPKDGVPYEKKFCEGSQPRGIILLGDSAGAHFHISPEWITASQMSLNSFINLPTALTNELDWPQLSGATGFLDSTVGIKEKSIYLRLWKRNHCNHRDYQNISRNGASSRNLKKFIESLSRNKVLDYPAIVIYAMIGNDVCSGKSDPVPAMTTPEKLYSNVMQTLKHLNSHLPNGSHVILYGLPDGTFLWDNLHNRYHPLGQLNKDMTYAQLYSFLNCLQVSPCHGWMSSNKTLRTLTSERAEQLSNTLKKIAASEKFTNFNLFYMDFAFHEIIQEWQKRGGQPWQLIEPVDGFHPNEVALLLLADHFWKKVQLQWPQILGKENPFNPQIKQVFGDQGGH
3.2)
Fc-fusion protein part 2

MGWSCIILFLVATATGVHS (Signal peptide) DKTHTCPPCPAPELLGGPSVFLFPPKPKDTLMISRTPEVTCVVVDVSHEDPEVKFNWYVDGVEVHNAKTKPREEQYNSTYRVVSVLTVLHQDWLNGKEYKCKVSNKALPAPIEKTISKAKGQPREPQVYTLPPSRDELTKNQVSLSCAVKGFYPSDIAVEWESNGQPENNYKTTPPVLDSDGSFFLVSKLTVDKSRWQQGNVFSCSVMHEALHNHYTQKSLSLSPG (asymmetric IgG1 part) HHHHHH (His-tag)

RT-qPCR

Total RNA was extracted by using the PureLink^™ RNA Mini Kit (Invitrogen), and cDNA was synthesized through reverse transcription using the SuperScript^™ III First-Strand Synthesis SuperMix for qRT-PCR (Invitrogen). RT-qPCR was performed using the SYBR^™ Green PCR Master Mix (Applied Biosystems) and corresponding primers, and Ct values were detected by the StepOne Plus Real-time PCR system. Raw data were processed using SDS 19.1 software, and the relative mRNA expression level was normalized to the internal ribosomal reference gene Rpl13a or Actb. The primer sequences (5’-3’) are listed in the Key Resources Table.

Tumor inoculation and ICB treatment

For subcutaneous tumor models, 1E5 B16F10/B16mhgp100, 3E6 B2905-M4, or 5E5 Renca cells were injected subcutaneously into 7–8-week-old female C57BL/6J and BALB/c mice, respectively. For orthotopic tumor models, a skin incision of ~2.0 cm was made on the abdomen in the liver region of anesthetized mice. 1.2E5 luciferase-labeled RIL-175 cells were injected into the subcapsular space of the liver of 7–8-week-old female C57BL/6J mice.

Anti-PD1 antibody (200ug per mouse, equivalent to 10 mg/kg with 20g per mouse) was given intraperitoneally to tumor-bearing mice twice a week for a total of 4 doses, starting on day 5 (RIL-175), day 6 (B16F10), day 7 (RenCa), or day 10 (B16-mhgp100) post tumor inoculation. Anti-CTLA4 antibody (200ug per mouse, equivalent to 10mg/kg with 20g per mouse) was given to B16F10-bearing mice twice weekly for four doses, starting the next day following the 1st anti-PD1 treatment. Alternatively, control groups of mice received IgG treatment. For AOAH treatment models, human recombinant AOAH (WuXi Biologics) stored in 10% glycerol PBS (100ug per mouse, equivalent to 5 mg/kg) was given intratumorally to B16F10 and B16-mhgp100-bearing mice once every two days for a total of 10 doses, starting on day 5 post tumor inoculation. Alternatively, the control group of mice received an intratumoral injection of vehicle (10% glycerol in PBS).

Tumor volume was measured based on the formula: (Length×Width²)/2. Progression of orthotopic tumor burden and treatment effects in mice were monitored in vivo by bioluminescence imaging twice per week. Mice were anesthetized with vaporized isoflurane followed by intraperitoneal injection of D-luciferin (150 mg/kg, GoldBio), and luciferase activities were measured with the IVIS system (PerkinElmer).

Mice were euthanized according to a predetermined survival endpoint defined as tumor volume ≥ 2000 mm³ or the tumor length (longer diameter) ≥ 20 mm. For tumor growth measurement by luminescence, the endpoint has to meet the following criteria: the end point measurement of each mouse starts at Day 19 (3 days after the last PD-1 treatment); the mouse total flux has to be >1E8; or the mouse has to show a severe weakness as judged by the professional veterinarian at the University of Hong Kong (HKU) Animal Facility. These endpoints are compiled with ethical regulations of animal studies at NIH, according to the animal protocol CDSL-001, or conducted under protocol 23–480, approved by the Committee on the Use of Live Animals in Teaching and Research at HKU.

DC2.4 injection

Parental DC2.4 cell lines were purchased from Sigma (Cat. # SCC142) and cultured in RPMI-1640 with 10% FBS, 1x MEM-NEAA (Gibco), 20 mM HEPES Buffer (Gibco), and 50 μM 2-ME (Gibco), as per manufacturer’s instructions. After Aoah transduction and puromycin selection, vector and Aoah-OE DC2.4 cells were first cultured in a puromycin-free medium for 1–2 passages. 1E5 B16-mhgp100 cells were inoculated subcutaneously into 7–8-week-old C57BL/6J mice. On day 3 post-tumor inoculation, vector and Aoah-OE DC2.4 cells were first treated with 50 μg/mL mitomycin C for 30 minutes to inhibit their permanent proliferation in vivo without interfering with their antigen presentation and stimulatory functions¹⁰⁵. Subsequently, 2E6 mitomycin C-treated vector and Aoah-OE DC2.4 cells were intraperitoneally injected into the tumor-bearing mice. The additional DC2.4 injection was performed once a week on days 10 and 17. Anti-PD1 antibodies (200 μg per mouse) were given intraperitoneally to the tumor-bearing mice twice a week, starting on day 10 for four doses. The tumor growth monitoring, volume calculation, and endpoint determination are consistent with the above procedures.

Flow cytometry

B16-mhgp100 and B16F10 tumors were first dissociated by 250 μg/mL Liberase^™ TM Research Grade (Sigma-Aldrich) and 100 μg/mL DNAse I (Gibco) at 37°C for 30 minutes with vortexing every 10 minutes. The dissociated cells were filtered through a 70-μm cell strainer, centrifuged at 300×g for 10 minutes, and resuspended in EasySep Buffer. The CD45+ tumor-infiltrating leukocytes (TILs) were isolated from the cell mixture by using EasySep^™ Mouse TIL (CD45) Positive Selection Kit (STEMCELL Technologies), as per manufacturer’s instructions. The TILs were washed with Cell Staining Buffer (BioLegend) prior to antibody staining. Mouse and human CD8+ T cells were collected from culture plates after TCR activation, centrifuged at 300×g for 5 minutes, and washed with Cell Staining Buffer. TILs or CD8+ T cells were incubated with a fluorochrome-conjugated antibody cocktail at ≤1E6 cells/100 μL on ice for 20 minutes. The cells were further stained with DAPI at 0.2 μg/mL in 1 mL Cell Staining Buffer at room temperature for 5 minutes. The stained cells were washed twice before flow cytometry analysis with Cell Staining Buffer. All the flow cytometry experiments were performed on FACSymphony^™ A5 (BD Biosciences), and results were analyzed by FlowJo^™ Software v11.

Multiplex immunofluorescence staining

Spontaneous hepatocellular (HCC) nodules and B16F10 tumors were carefully dissected from mouse livers and skin, respectively. They were then fixed in 10% Neutral buffered formalin at room temperature overnight, dehydrated, and embedded into paraffin blocks.

For HCC tumors: Dewaxing, antigen retrieval, blocking, serial antibody incubation, and DAPI staining were performed on the HCC paraffin sections using the 7-color mIHC staining kit (Servicebio Technology), as per manufacturer’s instructions. Panoramic scanning of mIHC sections was done using a Nikon Eclipse C1 fluorescence microscope. The fluorochromes used included SpAqua, SpGreen, SpGold, SpOrange, Spred, Cy5, Cy5.5, and DAPI. The antibodies used for mIHC staining were listed in the Key Resources table.

For B16F10 tumors: 4-plex fluorescent staining was accomplished using a sequence of triple staining on an autostainer, followed by a combination of manual and automated staining for the 4th target. Triple fluorescent staining was performed on the Leica Biosystems Bond RX autostainer using the Bond Polymer Refine Kit (Leica Biosystems DS9800), with omission of the PostPrimary reagent, DAB, and Hematoxylin. After antigen retrieval with Citrate (Bond Epitope Retrieval 1), sections were incubated for 30’ with CD8a (eBioscience #14-0195-82, 1:25), followed by ImmPRESS^® HRP-conjugated Goat anti-Rat (Vector Labs) and OPAL Fluorophore 570 (AKOYA). The CD8a antibody complex was stripped by heating with Bond Epitope Retrieval 1. Sections were then incubated for 30’ with PEP8h (non-commercial, 1:500), followed by the Bond Polymer reagent and OPAL Fluorophore 480. The PEP8h antibody complex was stripped by heating with Bond Epitope Retrieval 1. Sections were then incubated for 60’ with CD3 (Bio-Rad #MCA1477, 1:50), followed by ImmPRESS^® HRP-conjugated Goat anti-Rat (Vector Labs) and OPAL Fluorophore 620. The CD3 antibody complex was stripped by heating with Bond Epitope Retrieval 1. Sections were then incubated overnight at 4C with CD11c (Cell Signaling Technology #97585, 1:350), followed by the Bond Polymer reagent and OPAL Fluorophore 520. Sections were stained with DAPI and coverslipped with Prolong Gold AntiFade Reagent (Invitrogen). Images were captured using the AKOYA PhenoImager whole slide scanner.

The cell segmentation and marker-positive cell fraction quantifications were done using HALO AI v4.0 with the algorithm HighPlex FL v4.0.4.

Retrovirus transduction on pmel CD8+ T cells

Full-length cDNA of murine Aoah was constructed into MSCV-IRES-Thy1.1. The plasmid DNA transformation, purification, and Lenti-X^™ 293T cell seeding procedures were the same as those in lentivirus preparation. 24 hours after cell seeding, pLC and Gag, together with the MSCV-IRES-Thy1.1 empty vector or MSCV-IRES-Thy1.1-Aoah plasmid, were transfected into the 293T cells using the Lipofectamine 3000 transfection reagent (Invitrogen), as per manufacturer’s instructions. The culture medium was replaced with Opti-MEM^™ with GlutaMAX^™ (Gibco), supplemented with 1:500 ViralBoost Reagent (Alstem) 6 hours after transfection. The retrovirus was harvested by spinning down the viral supernatant at 3000 rpm, 4°C for 30 minutes to remove the cell debris at 48 and 72 hours post-transfection. The retrovirus was concentrated 100× by using the Retrovirus Precipitation Reagent (Alstem).

Fresh Aoah-low (verified by qRT-PCR, Ct > 30) CD8+ T cells from 8–10-week-old pmel mice were activated by Dynabeads^™ Mouse T-Activator CD3/CD28 (Gibco) at a 1:1 beads-to-cells ratio in the presence of 100 IU/mL rhIL-2 for 24 hours. The activated CD8+ T cells were transduced by the 10× fresh retrovirus in warm T-cell medium supplemented with 100 IU/mL rhIL-2 and 10 μg/mL polybrene, by spin infection at 2000×g, 32 °C for 2 hours. The transduced T cells were incubated at 37 °C overnight. Subsequently, the Thy1.1+ population was isolated by using mouse CD90.1 MicroBeads (Miltenyi), as per manufacturer’s instructions. The expression and secretion of Aoah in mouse pmel CD8+ T cells were validated by qRT-PCR and ELISA.

Adoptive T-cell transfer

Pmel CD8+ T cells were expanded for 7–10 days in a complete T-cell medium with 100 IU/mL rhIL-2 after retrovirus transduction. 1E5 B16F10 cells were subcutaneously inoculated into 7–8-week-old C57BL/6 mice. One day before T-cell transfer (Day 8 or 9 post-tumor inoculation), B16F10-bearing C57BL/6 mice were irradiated with a dose of 600 cGy and randomized into two groups. The next day, 2E6 Vector and Aoah-armored pmel CD8+ T cells were transferred intravenously into the mice, and 10000 IU/0.5 mL rhIL-2 was administered intraperitoneally into the mice twice a day for 3 consecutive days.

Incucyte tumor-killing assay

Pmel CD8+ T cells were briefly activated by 1μg/mL anti-CD3+ 0.5 μg/mL anti-CD28 antibodies in the presence of 10 μg/mL rhAOAH or buffer vehicle for 24 hours. The activated CD8+ T cells were expanded for 3–4 days prior to co-culture. mCherry-labeled B16-mhgp100 and B16F10 cells were seeded into a 96-well plate at a density of 2.5E3 cells/well. Following a 24-hour culture, the expanded CD8+ T cells were seeded into each well at different effector-to-target ratios in a complete T-cell medium with 100 IU/mL rhIL-2 and 10 μg/mL rhAOAH or Vehicle. The plates were imaged, and the red fluorescence of tumor cells was quantified every 6 hours for 72 hours using the Incucyte^® S3 Live-Cell Analysis System (Essen Bioscience).

Western blot

Human CD8+ T cells were pre-treated with 10 μg/mL rhAOAH and an equal volume of buffer vehicle in a complete T-cell medium at a density of 2E6 cells/mL. Following a 20-hour pretreatment, the CD8+ T cells were resuspended in the complete T-cell medium supplemented with 50 IU/mL rhIL-2. The cells were treated again with 10 μg/mL rhAOAH and Vehicle as previously described and further activated with a 1:500 dilution of human T Cell TransAct^™ (Miltenyi) for 5, 10, and 20 minutes at 37°C. The T cells were collected and centrifuged at 300 rpm for 5 minutes at 4°C, followed by one wash with 1× ice-cold PBS. Subsequently, the T cells were lysed in Pierce RIPA buffer (ThermoFisher), supplemented with a protease inhibitor cocktail (MCE) and a phosphatase inhibitor cocktail (MCE). The protein lysates were quantified using the Bradford protein assay, separated by SDS-PAGE, and transferred to an activated PVDF membrane (Millipore). The membrane was incubated with the primary antibody for 16–18 hours at 4°C, followed by an HRP-conjugated secondary antibody for 1 hour at room temperature. Each antibody incubation was followed by 3 times of 10-minute washes with 1× TBST. Blots were visualized using enhanced chemiluminescence (Bio-Rad) and X-ray film. The antibodies used are listed in the Key Resources Table.

NFAT luciferase assay

Jurkat-Lucia^™ NFAT-CD28 cells were purchased from Invivogen (jktl-nfat-cd28) and cultured in the recommended growth medium with selective antibiotics. The Jurkat cells were pre-treated with 10 μg/mL rhAOAH or buffer vehicle in the complete IMDM without antibiotics for 20–22 hours. The NFAT induction was followed by the manufacturer’s instructions: In a 96-well plate, 3E5 Jurkat cells were activated with Immunocult (Stem Cell Technologies) at 12.5 μl/mL (1:2), 6.25 μl/mL (1:4), 3.125 μl/mL (1:8), and 1.5625 μl/mL (1:16) or mock in 200 μL IMDM with 10 μg/mL rhAOAH or Vehicle (PBS + 10% glycerol) per well. After 24-hour activation, the supernatants were collected and centrifuged to remove cell debris. The luminescence of the supernatants was quantified by SpectraMax iD5 (Molecular Devices) and normalized by the baseline luminescence of the supernatants from unactivated Jurkat cells.

Untargeted lipidomics using mass spectrometry

The untargeted lipidomics was performed at Proteomics& Metabolomics Core at the Center for PanorOmics Sciences, Li Ka Shing Faculty of Medicine, the University of Hong Kong. Mouse CD8+ T cells were pre-treated with 10 μg/mL rhAOAH or buffer vehicle at 2E6 cells/mL in a complete T-cell medium without rhIL-2 for 22–24 hours. After pretreatment, the cell culture supernatants were collected, centrifuged, and snap-frozen in liquid nitrogen. T cells were collected, centrifuged, washed with ice-cold saline, and snap-frozen in liquid nitrogen prior to sample submission.

For cell culture supernatants, the sample was extracted with chloroform:methanol (2:1, v/v). The sample was then centrifuged at 3000×g for 5 minutes. The non-polar phase was aliquoted out and dried under nitrogen. For cell pellets, 5 mL chloroform:methanol (2:1, v/v) was added to 5 million cell pellets. The sample was then sonicated under an ice-chilled probe sonicator for 20 seconds, cooled down for 10 seconds, and another sonication for 20 seconds. The sample was then centrifuged at 3000×g for 5 minutes. 1.5 mL supernatant was aliquot out and dried under nitrogen. The sample was then reconstituted with 50 μL IPA:methanol:chloroform (1:1:0.2, v/v). Then, 3 μL was injected into the LC-MS/MS system.

The chromatographic separation was carried out on a Vanquish UPLC (Thermo Fisher, Waltham, MA, USA). The mobile phases used were 10mM ammonium formate with 0.1% formic acid in acetonitrile and water, v/v 6:4 (A), and 10mM ammonium formate with 0.1% formic acid in acetonitrile and IPA 1:9 (B). The column was a ThermoFisher Accucore C30 (2.1×150 mm, 2.6 μm). An injection volume of 3 μL and a flow rate of 0.26 mL per minute were used. The column oven temperature was set at 45°C. The gradient started at 30% B and was increased to 43% B in 2 min, then increased to 55% B in 2.1 min, 65 % B in 12 min, 85% B in 18 min, and 100 % B in 20 min, then held for 5min and decreased linearly to 30% B for re-equilibration time at starting conditions.

The mass spectrometry analysis was processed using an Orbitrap Exploris 120 mass spectrometer Thermo Fisher (Waltham, MA, USA) equipped with a HESI II probe in polar switching mode with source parameters set as follows: sheath gas flow rate, 60; auxiliary gas flow rate, 17; sweep gas flow rate, 1; spray voltage, +3.5/−3.0 kV; capillary temperature, 275 °C; S-lens RF level, 70; and heater temperature, 325 °C. Data was collected at dd-MS2 mode. Data analysis was performed using Lipidsearch (Thermofisher Scientific/Mitsui Knowledge Industries) with the default parameters for Orbitrap MS Product Search and Alignment. After alignment, raw peak areas for all identified lipids were exported to Excel files.

Phosphatidylcholine treatment assay

16:0–20:4 phosphatidylcholine (PC) and 18:0–20:4 PC were purchased from Avanti Polar Lipids. The PCs in chloroform were dried under a tissue culture hood for 1–2 hours. The dried PCs were first readily dissolved by 5 μL 70% ethanol and diluted to a 500 μM stock using a T-cell medium (without FBS). Mouse and human CD8+ T cells were pre-treated with different concentrations of PCs (30 μM, 60 μM, 120 μM, and 200 μM) in the presence of 10 μg/mL rhAOAH or Buffer matched with ethanol concentrations for 20–22 hours. After pretreatment, the CD8+ T cells were activated by anti-CD3/CD28 antibodies following the same procedure as described in “Mouse and human CD8+ T cell activation”.

Immunophenotyping of dendritic cells

PC(16:0–20:4) was dried and oxidized (ox) under air in the sterile tissue culture hood for 3 days³⁶. On day 3, 1.5E6 splenocytes isolated from C57BL/6 mice were pre-treated with 30 μM oxPC(16:0–20:4) re-dissolved in 70% ethanol in 1.5 mL complete T-cell medium. 4 hours or 12 hours later, 1E5 OT-1 CD8+ T cells were mixed with 3E5 in a 96-well plate in 200 μL complete T-cell medium with 100 IU/mL rhIL-2 and 10 nM OVA peptide. 24 hours after T-cell activation, all cells were collected, washed, and stained with the desired antibodies and DAPI. Flow cytometry analysis was performed to quantify MHC-I, MHC-II, CD40, and CD80 expression on live CD11c+ dendritic cells.

QUANTIFICATION AND STATISTICAL ANALYSIS

Gene and clinical outcome associations

Among cohorts with gene values (expression, mutation, copy number alteration, promoter methylation) from pretreatment bulk tumors, we computed the risk scores according to the type of endpoints.

Survival outcomes: We utilized the Cox proportional hazard (PH) regressions. Gene risk scores are the z-scores (Coef/ StdErr) from the two-sided Wald test through R programming.
RECIST outcomes: We utilized the ordinary least squares (OLS). Gene risk scores are negative t-values (Coef/ StdErr) from the two-sided t-test through the Python scipy package.
Binary outcomes: If only binary response information is available, we utilized the two-sided Wilcoxon rank-sum test through the scipy package. Gene risk scores are negative z-scores. If other clinical covariates (e.g., Age, Gender, and Stage) are available, we utilized logistic regression with Firth correlation¹⁰⁶ through our logis_batch package in C programming language. Gene risk scores are the negative z-scores (Coef / StdErr).

If multiple endpoints are available for each cohort, we prioritized the endpoints in this order: Overall survival > Progression-free survival > RECIST > Binary response, which assigns only one score profile per cohort in later prioritization analysis. The analysis methods for each cohort are available in tabs “Analysis_*” of Table S1. The source code for this analysis is available at https://github.com/data2intelligence/CIDE_main. All computational results are publicly available at https://cide.ccr.cancer.gov.

Gene prioritizations for secreted proteins

We utilized 50 non-redundant genome-wide transcriptomics cohorts from pre-immunotherapy bulk tumors. The list of 1903 genes coding secreted protein was from the Human Protein Atlas¹⁰⁷ on August 10, 2022 (https://www.proteinatlas.org/humanproteome/tissue/secretome). We excluded genes coding cell surface receptors because the current study focuses on secreted protein functions. We collected 1638 human genes encoding cell surface receptors by joining the set of genes from Gene Ontology¹⁰⁸ term “GO0038023_signaling_receptor_activity” (release 2025-03-16) and receptor sets from CellPhoneDB¹⁰⁹ interactions (v5.0.0). Besides excluding cell surface receptor genes, we restricted our secreted protein sets to those with known secretome locations annotated by the Human Protein Atlas¹⁰⁷. Finally, 1645 genes are candidates for further analysis.

Then, among 1645 genes, we removed genes with no risk scores computed in over 5% of datasets due to the absence of gene coverage in the original study or regression failures. Then, for each gene, we computed the two-sided Wilcoxon signed-rank test comparing risk scores across 50 cohorts and zero. Then, we converted the p-values to false discovery rates (FDR) by the Benjamini-Hochberg correction¹¹⁰.

We also filtered genes with outlier datasets whose gene risk scores are in the reverse direction compared to the overall median risk scores. We counted the datasets for each gene with risk scores > 2 and < −2 as pos_count and neg_count, respectively. For genes with positive median risk scores, we only kept genes where pos_count - neg_count >= 3. For genes with negative median risk scores, we only kept genes where neg_count - pos_count >= 3.

Among genes associated with positive risks, we further filtered genes whose protein levels are not high in the CPTAC tumor proteomics data v1¹¹¹. Seven tumor types have more than 10 tumors with proteomics MassSpec data paired between the tumor and adjacent normal tissue. We performed the two-sided Wilcoxon signed-rank test comparing the protein levels and converted the p-values to false-discovery rates (FDR) using the Benjamini-Hochberg correction. We only kept positive-risk genes with FDR < 0.05 in at least one tumor type.

For visualization simplicity, in Figure 2B, we only selected the top 66 genes ranked by absolute values of median risk scores across 50 cohorts, and include the full list in Table S6.

Besides using transcriptomics data, we tested the utility of somatic mutation and copy number alteration (CNA) data in finding secreted proteins modulating tumor progression. However, we did not find any secreted protein-coding genes among genes with somatic mutations called in any CIDE cohorts. Further, using the same prioritization thresholds described above, we found zero secreted protein-coding genes whose CNA is significantly associated with clinical outcomes across 28 cohorts with CNA data.

Previous studies showed autocrine signaling of pro-tumor secreted proteins sent by cancer cells¹¹². However, these studies did not necessarily show that cancer cell secretion is due to copy number alterations or gain-of-function mutations. For example, the MDA-MB-231 breast tumor cell line secreted TGFβ1¹¹². According to the DepMap⁴¹ (v24Q4), the TGFβ1 copy number log2(relative to ploidy + 1) of MDA-MB-231 = 1.02, indicating a neutral TGFβ1 copy in MDA-MB-231. Corroborating with this result, TGFB1 copy number alterations are rare events in TCGA cohorts, with about a 5% event rate in only one out of 33 cancer types profiled (cBioportal¹¹³ query on June 05, 2025). Thus, the lack of significant targets from mutation and CNA data in our pan-cancer analysis is due to the insufficient mutation and CNA events of secreted protein-coding genes in tumor cells.

The source code for this analysis is available at https://github.com/data2intelligence/CIDE_Prioritization. The interactive prioritization module is also available at https://cide.ccr.cancer.gov.

Alternative meta-analysis methods

A core step of our meta-analysis is to derive the statistical significance for each gene across cohorts (Figure 2). We used the two-sided Wilcoxon signed-rank test to compare risk scores across 50 cohorts to zero. However, alternative methods, such as Fisher’s p-value combination, are common approaches to combining results across cohorts. Thus, we also evaluated five additional meta-analysis approaches that combine p-values from statistical tests in individual cohorts: 1, Fisher’s method; 2, Fisher’s method with p-value decorrelation; 3, truncated product method; 4, truncated product method with p-value decorrelation; and 5, permutation test.

P-value decorrelation and truncated product method

Zaykin and colleagues¹¹⁴ introduced two methods: 1, the truncated product method for combining p-values; 2, the p-value decorrelation methods for merging p-values from non-independent tests. The rationale behind this study is twofold. First, compared to Fisher’s combination method, the truncated product method¹¹⁴ provides more significant p-values as the authors said: “Experience shows that the ordinary Fisher product test loses power in cases where there are a few large P-values. This can happen when tests are one sided, with non-centrality in the “wrong” direction, or when there is a predominance of near-null effects. By truncating, these large components are removed, thereby providing more power.” We utilized the implementation of the Python package combine_pvalues. Second, Fisher’s p-value combination method only applies to independent tests¹¹⁵; thus, it will overstate the statistical significance of the merged p-value from dependent tests. The decorrelation of p-values is essential before applying p-value combinations. As we cannot find any existing decorrelation packages, we implemented the procedure in the previous study¹¹⁴:

Σ: non-degenerate correlation matrix for the vector of p-values R.
Φ(x): the probability distribution function for the standard normal distribution.
If Σ is positive definite (true for our data), then there exists a matrix C (Cholesky factor), such that Σ = CC^T. Decorrelated p-values vector R* = 1 - Φ{C⁻¹Φ⁻¹(1-R)}

Permutation test

Similar to our initial Wilcoxon signed-rank test, we compare medians of risk scores against the random values by permutating gene risk scores across genes within each cohort. The p-value is computed as Probability(|random median| >= |real median|) after 10000 randomizations.

Combining one-sided p-values

Before applying p-value combination methods, we converted two-sided p-values Probability(|random risk| >= |real risk|), reported by regressions, to one-sided p-values on the positive-tailed Probability(random risk >= real risk) and negative-tailed Probability(random risk <= real risk). The reason is that a two-sided p-value does not indicate the direction (positive or negative) of risk scores. However, for each gene, when we merge p-values from regressions across diverse cohorts, we care about whether the gene level is associated with higher or lower risks. The Wilcoxon signed-rank test and permutation test, which are all based on risk scores, can directly tell the risk directions, while Fisher’s or truncated product methods cannot unless we explicitly utilize one-sided p-values.

Method comparisons and result similarities

We computed the Spearman rank correlation among merged p-values from different approaches and found that our Wilcoxon signed-rank and permutation tests led to similar results (Figure S2C). Then, for each merging method, we converted p-values across genes to false discovery rates (FDR) by the Benjamini-Hochberg correction and counted genes with FDR < 0.05. As we applied Fisher and truncated product methods on one-sided p-values on both positive and negative tails, we sum the significant gene counts (#genes with median risk > 0 and positive-tailed FDR < 0.05 + #genes with median risk < 0 and negative-tailed FDR < 0.05). We found that the Fisher and truncated product methods led to more than two-fold more significant genes than the permutation and rank-sum tests (Figure S2D). After p-value decorrelation, the number of significant genes dropped to the lowest among all methods for Fisher’s method. Finally, we selected our initial approach based on the Wilcoxon signed-rank test, which is the most straightforward approach.

Experimental data group comparisons

Statistical tests used in comparing data from distinct experimental groups are indicated in figure legends and summarized as follows:

Fishers’ exact test compares the consistency between CIDE predictions and blinded literature annotations, done through the scipy package.
The differential gene expression analysis of RNA-seq data is done through the two-sided Wald test implemented in the DESeq2 package.
The gene set enrichment analysis is done through the permutation test implemented in the GSEA package.
For wet lab data with more than 10 samples per group or complete separations between two groups, we utilized the two-sided Wilcoxon rank-sum test for unpaired data or signed-rank test for paired data, implemented in the scipy package.
For wet lab data with less than 10 samples or incomplete separations between two groups, we utilized the two-sided t-test with the K-S normality test through the GraphPad Prism. We used the unpaired t-test for unpaired data and the paired t-test for paired data between two groups. Only results that have p-values < 0.05 and passed the K-S normality test were shown.
For the T-cell-mediated tumor killing curves measured by IncuCyte, we used the two-way ANOVA test with time and treatment as two factors through the GraphPad Prism.

Supplementary Material

Figure S1. Computational frameworks and analyses, related to Figure 1.

(A) RNA and Whole-exome sequencing data processing pipelines.

(B) Quality control metrics for sequencing data. The Exon Coverage represents the average read count coverage across all protein-coding exons, computed by the mosdepth package⁴⁵. The One Genome Fold represents the ratio of reads mapped between the human genome and the second-best model organism genomes included in the fastq_screen package⁴⁶. The metrics are shown in box plots, as in Figure 1C.

(C) Tumor purity estimations. Two datasets^47,48 provided tumor purity estimations, so we utilized the authors’ original values. For other datasets, we estimated the tumor purities using the ESTIMATE⁴⁹ package for transcriptomics data and ASCAT⁵⁰ package for WES data. The metrics are shown in box plots, as in Figure 1C. The sample counts are labeled for each dataset beneath each box plot.

(D) Performance metrics for predicting different endpoints. Each dot represents a transcriptomics study of pretreatment tumors, with the immunotherapy endpoint type labeled on the y-axis. For binary response, progression-free survival (PFS), and overall survival (OS), the metric is the Wald test z-score from regressions with all available covariates. For the Response Evaluation Criteria in Solid Tumors (RECIST) outcome, the metric is the t-test t-value from regressions with all available covariates. The order of biomarkers (x-axis) follows the average performance rank across all endpoint types. The metrics are shown in box plots, as in Figure 1C. The patient count for each dataset is labeled beneath each box plot. The black dotted line represents 0 as the random expectation. The red dotted line represents a good performance (z = 2 or t = 2).

(E) Histograms of Wilcoxon rank-sum z-scores comparing phenotype scores between secreted versus intracellular protein-coding genes. The Wilcoxon rank-sum tests were done as in Figure 1E across all CRISPR screen datasets. The comparisons between Wilcoxon rank-sum z-scores and zero were through the two-sided Wilcoxon signed-rank tests, with p-values shown together with the median and the number of datasets. The Tres¹⁷ and DepMap⁴¹ cohorts are shown here.

NIHMS2096169-supplement-1.pdf^{(21.1MB, pdf)}

Figure S2. Gene risk scores from immunotherapy transcriptomics cohorts, related to Figure 2.

(A) Risk scores by endpoint types. The heatmap of risk scores is shown as in Figure 2B, except for the following difference. For each cohort, we included risk scores computed for every endpoint type available in this panel, while Figure 2B only shows the risk scores computed from the primary endpoint of each cohort.

(B) Statistical significance of risk scores by endpoint types. For each gene in panel A, we compared the risk scores computed for each endpoint against zero using the two-sided Wilcoxon signed-rank test and converted the p-values to false discovery rates (FDR). The y-axis presents the median risk scores across all cohorts with each endpoint. The bar color presents the FDR levels.

(C) Similarities of gene ranks by p-values from different meta-analysis methods combining statistical test results from 50 cohorts. Besides the rank-sum methods utilized in the main manuscript, we compared p-values from diverse meta-analysis methods (Methods). For each method, we ranked genes based on the −log10(positive-tailed p-value) for genes with positive median risk scores across cohorts or log10(negative-tailed p-value) for genes with negative median risk scores across cohorts. For Wilcoxon and permutation test methods, we utilized the two-sided p-values for both positive-tailed and negative-tailed p-values. We show similarities in gene ranks using hierarchical clustering with the Spearman rank correlation distance. The conclusion is that different meta-analysis approaches lead to similar results (rank correlations > 0.8 across all clusters).

(D) Number of genes with FDR < 0.05 for diverse methods. The Wilcoxon approach that we selected is among the more conservative approaches.

(E) Pearson correlations of risk scores from the anti-PD1/PDL1 studies from the same cancer type (melanoma or renal). RCC: renal cell carcinoma; CCRCC: clear cell RCC; mRCC: metastatic RCC; ICB: immune checkpoint blockade; OS: overall survival; PFS: progression-free survival.

NIHMS2096169-supplement-2.pdf^{(4.5MB, pdf)}

Figure S3. Predicted secreted regulators of cancer immunotherapy outcomes, related to Figure 3.

(A) Risk scores with tumor purity correction, shown as in Figure 3A. The risk scores for top candidates were re-computed by adding the tumor purity estimation as a covariate in the regression model.

(B) Gene expression in diverse cell lineages. The edge color represents the gene expression level of the cell type connected on the right. (NK: natural killer; CAF: cancer-associated fibroblast; DC: dendritic cell; pDC: plasmacytoid DC; cDC: conventional DC.)

(D) Correlation between AOAH expression and cytotoxic T lymphocyte (CTL) infiltration among pancreatic tumors treated with nivolumab plus chemotherapy⁵¹. The CTL infiltration was estimated as the median expression of CD8A, CD8B, GZMA, GZMB, and PRF1⁵².

(E) Correlations with CTL infiltrations for prioritized secreted proteins, shown as in Figure 3A.

(F) AOAH expression in infusion samples from adoptive T-cell transfer responders and non-responders with melanoma¹⁹. Each dot represents the infusion product for each patient. The gene expression values are shown in box plots as in Figure 3C. P-values were computed through the two-sided Wilcoxon rank-sum test, comparing values between responders and non-responders.

(G) In vitro growth of cancer cells in culture, measured by XTT assay (n = 3 cell culture replicates). The metabolic activity is measured as optical density at 492nm (read) divided by the value at 620nm (reference).

(H) EdU staining quantification of newly synthesized DNA from B16-mhgp100 (Aoah, Cr1l, and Colq overexpression) and B2905-M4 (Adamts7 overexpression) cells. The fraction of EdU-positive among DAPI-positive cells is shown. The two-sided Wilcoxon rank-sum tests were performed to compare values between the target and vector overexpression groups. Although positive fractions for Adamts7 overexpressed cells achieved a statistical difference, the fold ratio of median values between the two groups is only 1.12. Thus, the ratio is not much different from no change (fold ratio = 1).

(I) Growth of subcutaneous B16-mhgp100 tumors with Aoah overexpression in immunodeficient NSG mice, shown as in Figure 3E.

(J) Growth of subcutaneous B16F10 tumors with Aoah overexpression in C57BL/6 mice treated with IgG controls, shown as in Figure 3E.

(K) Tumor growth curves for B16 tumor models with strong (N4) and weak (V4) antigens, shown as in Figure 3E.

(L) Endpoint-free survival plots for panel K, shown as in Figure 3F.

NIHMS2096169-supplement-3.pdf^{(10.6MB, pdf)}

Figure S4. In vivo validation of AOAH, related to Figure 4.

(A) The design of plasmids used for the hydrodynamic tail vein injection (HDTVi) to induce spontaneous hepatocellular carcinoma (HCC) in mouse livers.

(B) ELISA quantification of AOAH levels secreted by DC2.4 with Aoah or vector overexpression, with the p-value computed using the two-sided Wilcoxon rank-sum test (n = 3 cell culture replicates, RT-qPCR median fold = 17.4).

(C) Tumor growth and endpoint-free survival of B16-mhgp100 tumors in C57BL/6 mice with intraperitoneal injections of DC2.4 with Aoah or vector overexpression, shown as in Figure 4C.

(D) Flow cytometry analysis of CD3, CD8, and TCR expression in CD45+ tumor-infiltrating leukocytes (TIL) isolated from B16F10 tumors at day 7 post TCR-T adoptive therapy (n = 4 mice). p values were computed using the two-sided unpaired t test, comparing group values. Only p values < 0.05 are shown.

(E) The body weight change of B16F10-bearing mice after TCR-T transfer, shown as in Figure 4D.

(F) Representative Hematoxylin and Eosin images from spleens, lymph nodes, and bone marrows of tumor-free C57BL/6 mice and B16F10 tumor-bearing C57BL/6 mice after TCR-T therapy. Scale bar = 200 μM (spleen and lymph node) and 20 μm (bone marrow).

(G) The tumor growth and endpoint-free survival curves of C57BL/6J mice bearing subcutaneous B16F10 tumors treated with serial concentrations of rhAOAH. In the left panel shown as in Figure 4D, P-values were computed using the two-sided Wilcoxon rank-sum test comparing values between protein treatment and vehicle groups at the last time point with complete tumor measurements. In the right panel shown as in Figure 4E, P-values were computed using the two-sided log-rank test comparing survival time between protein treatment and vehicle groups. Only p-values < 0.05 are shown.

(H) The body weight change of B16F10 tumor-bearing mice during rhAOAH (5mg/kg) + immune checkpoint blockade treatments, shown as in Figure 4D.

(I) Additional representative multi-IF images of major TIL subtypes in the MYC β-catenin induced spontaneous hepatocellular carcinoma (HCC) sections.

NIHMS2096169-supplement-4.pdf^{(8.1MB, pdf)}

Figure S5. In vivo molecular landscape induced by Aoah, related to Figure 5.

(A) Hierarchical clustering of log2FC profiles, using Pearson correlation as the distance.

(B) IFNγ activities predicted by the CytoSig package⁵³. The p-value is computed through the two-sided permutation test with 1000 randomizations.

(D) B-cell receptor (BCR) immunoglobulin heavy chain (IGH) clonality for B cells, compared as in Figure 5G.

(E) IGH class fractions.

(F) Top 20 enriched terms from the gene ontology biological processes (GO_BP), shown as in Figure 5H that utilized KEGG terms.

(G) Co-enriched pathways for Aoah, Colq, and Adamts7 overexpressions, shown as in Figure 5K.

(H) The association between the CR1L expression level and overall survival in a melanoma cohort with anti-PD1 treatment²⁹, shown as in Figure 2A.

NIHMS2096169-supplement-5.pdf^{(2MB, pdf)}

Figure S6. Validations of TCR activation by AOAH, related to Figure 6.

(A) Representative images of tumor cells expressing mCherry fluorescence in co-cultures. As our Tres model¹⁷ predicted that AOAH-positive T cells were resilient to immunosuppressive signals, such as TGFβ (Figure 2E), we also tested the T-cell mediated tumor killing in the presence of TGFβ (5 ng/mL).

(B) The T-cell cytotoxicity measured by the LDH assay in the co-culture. P-values were computed through the two-sided unpaired t-test (n = 5 cell culture replicates). Mean and standard deviation values are shown.

(C) Normalized mCherry fluorescence in B16-mhgp100 and B16F10 cells co-cultured with CD8+ pmel T cells in the presence of 5 ng/mL TGFβ. Mean and standard deviation values are shown (n = 5 cell culture replicates). P-values were computed through the two-way ANOVA with time and treatment as two factors.

(D) Concentrations of cytokines in the co-culture after 48 (B16-mhgp100) and 72 hours (B16F10). P-values were computed through the two-sided unpaired t-test (n = 3 cell culture replicates). Mean and standard deviation values are shown.

(E) RhAOAH pretreatment prior to T-cell activation and rhAOAH treatment during T-cell activation both promoted IFNγ secretion. P-values were computed through the two-sided paired t-test (n = 3 biological replicates). Mean and standard deviation values are shown.

(F) Flow cytometry gating corresponding to Figure 6D, E, and J.

(G) Flow cytometry gating corresponding to Figure 6G.

(H) Concentrations of cytokines released by OT-1 CD8+ T cells after activation by OVA-N4 and OVA-V4 pulsed splenocytes as antigen-presenting cells. P-values were computed through the two-sided paired t-test, comparing values at each concentration (n = 3 biological replicates). The line shows the mean value at each concentration.

(I) Differentially expressed genes calculated by DESeq2 and significantly enriched pathways with normalized enrichment scores in rhAOAH-treated pmel CD8+ T cells co-cultured with hgp100-pulsed splenocytes computed by GSEA. Study designs, sample sizes, and statistics are available in Table S7C.

(J) Validation of key genes associated with cytotoxicity and T-cell activation using qRT-PCR.

NIHMS2096169-supplement-6.pdf^{(2.7MB, pdf)}

Figure S7. Lipidomics studies of AOAH-mediated TCR activation, related to Figure 7.

(A) Enriched lipid species by rhAOAH, shown as in Figure 7A. Only enriched lipids with log2FC > 1 are shown.

(B) Concentrations of cytokines released from mouse CD8+ T cells after PC(16:0–20:4) and PC(18:0–20:4) pretreatment and TCR activation. Mean and standard deviation values are shown for each condition (n = 3 biological replicates).

(C) Concentrations of cytokines released from human CD8+ T cells after PC(16:0–20:4) pretreatment and TCR activation. Mean and standard deviation values are shown for each condition (n = 3 human donors).

(D) The inhibition effect of PC(18:0–20:4) on cytokine release from activated mouse CD8+ T cells (Methods). Each dot represents the cytokine concentration ratio between the PC-treated group and mock-treated group (raw data in panel B). P-values were computed through the two-sided paired t-test, comparing values at each concentration (n = 3 biological replicates). The line shows the mean value per condition.

(E) Flow cytometry analysis of p-CD3ζ, p-LCK, and CD25 in mouse CD8+ T cells after TCR activation, corresponding to Figure 7G.

(F) Flow cytometry analysis of CD25 and CD69 in human CD8+ T cells after TCR activation, corresponding to Figure 7H.

(G) Flow cytometry analysis of the fluorescent shift of the lipid peroxidation sensor in mouse CD8+ T cells after TCR activation, corresponding to Figure 7I.

(H) The expressions of MHC-I, MHC-II, CD40, and CD80 on CD11+ DCs from mouse splenocytes after OVA-specific activation of OT-1 CD8+ T cells. Mean and standard deviation values are shown for each condition (n = 3 biological replicates), corresponding to the percentage of inhibition calculated in Figure 7K.

(I) Flow cytometry analysis of MHC-I, MHC-II, CD40, and CD80 expressions on CD11+ DCs from mouse splenocytes, corresponding to Figure 7K. For MHC-I and II, the Mean Fluorescence Intensities are labeled for each condition. For CD40 and CD80, the percentages of marker-positive cells are shown for each condition.

(J) Flow cytometry analysis of CD25 and CD69 in mouse CD8+ T cells after TCR activation using wild-type rhAOAH, mutant rhAOAH, and vehicle treatment, corresponding to Figure 7N.

NIHMS2096169-supplement-7.pdf^{(873.5KB, pdf)}

Table S1. Cancer immunotherapy data from clinical studies, related to Figure 1.

This table has five tabs, enumerated below. The meaning of most columns is evident from their names, with exceptions explained.

Dataset_Bulk: Information about bulk-tumor omics datasets from cancer immunotherapy clinical studies. The Category column presents the following dataset types:
- Expression: gene expression data
- Mutation: gene somatic mutations by whole-exome sequencing (WES).
- CNA: copy number alterations computed from WES data
The “Raw data” column indicates whether the raw sequencing data is available for unified pre-processing with the CIDE framework. The “Source” column shows the original places or accession numbers for data download.
Analysis_Bulk: analyses performed on the datasets in Dataset_Bulk. IDs in the first column match those in the Dataset_Bulk ID column. The Method column presents analysis approaches for different clinical outcome types:
- Cox-PH: Cox proportional hazard (PH) regression with the two-sided Wald test for survival outcomes.
- OLS: Ordinary Least Squares regression with the two-sided t-test for the Response Evaluation Criteria in Solid Tumors (RECIST) outcomes.
- Rank-sum: Two-sided Wilcoxon rank-sum test for the binary response without clinical covariates to adjust.
- Logit: Logistic regression with the two-sided Wald test and Firth correction [S1] for the binary response with clinical covariates to adjust.
The column “Result file” presents the file names of the statistical analysis results, which are included in “gene_scores.zip” in Zenodo (https://doi.org/10.5281/zenodo.15012968).
Dataset_Single_Cell: information about single-cell RNA-seq datasets from cancer immunotherapy clinical studies.
Analysis_Single_Cell: analyses performed on the datasets in Dataset_Single_Cell, itemized by cell types. IDs in the first column match those in the Dataset_Single_Cell ID column. The Method column is the same as “Analysis_Bulk”. The column “Result file” presents the file names of the statistical analysis results, which are included in “gene_scores.zip” in Zenodo (https://doi.org/10.5281/zenodo.15012968).
Dataset_Lineage: Single-cell RNA-seq data showing cell lineage expression of genes, adapted from a previous study [S2].

NIHMS2096169-supplement-9.xlsx^{(34.5KB, xlsx)}

NIHMS2096169-supplement-8.pdf^{(330.4KB, pdf)}

Table S5. Clinical endpoints and covariates in datasets, related to Figure 2.

The table includes details of clinical information contained in our analysis. Each row presents a study providing one or more cohorts.

The column group “Stratification” includes the columns for stratifying omics data from one study into sub-cohorts for analysis. In each cell, distinct sub-cohorts were separated by “|”.
The column “Endpoints” presents clinical outcome types available for each study, separated by “|”. Each endpoint type has its own statistical analysis.
The column group “Regression Covariates” includes the covariates adjusted in the regression analysis. In each cell, distinct covariates were separated by “&”.

NIHMS2096169-supplement-10.xlsx^{(21.5KB, xlsx)}

Table S6. Gene function annotation, related to Figure 2.

The table included blinded annotations (colored in black) for statistical evaluations and follow-up annotations (colored in grey) to identify missing literature in the first-round blinded annotation. However, unblinded follow-up annotations are not involved in evaluating prediction accuracy (Figure 2C and D).

TME: tumor microenvironment; DC: dendritic cell; ICB: immune checkpoint blockade; Treg: T-regulatory cells; NK: natural killer; EMT: epithelial-mesenchymal transition; ECM: extracellular matrix; HCC: hepatocellular carcinoma; CRC: colorectal cancer; TNBC: triple-negative breast cancer; HNSCC: Head and Neck Squamous Cell Carcinoma; ESCC: esophageal squamous cell carcinoma; GBM: Glioblastoma.

NIHMS2096169-supplement-11.xlsx^{(35.3KB, xlsx)}

Table S7. Statistical significance and sample counts, related to Figures 3 – 7.

(A) Survival analyses for mouse in vivo experiments, related to Figures 3 and 4.

(B) RNA-seq analyses for mouse in vivo experiments, related to Figure 5.

(C) RNA-seq analyses for in vitro cell line experiments, related to Figures 6 and 7. Different from panel B, all experiments in panel C follow a paired design, meaning biological replicates are all matched pairs between treatment and control conditions. Thus, when performing the DESeq2 analysis, we added a Group covariate column to the design matrix to indicate the paired design.

NIHMS2096169-supplement-12.xlsx^{(26.4KB, xlsx)}

Highlights.

CIDE integrates 8575 omics profiles from 5957 patients with immunotherapy outcomes
AOAH, CR1L, COLQ, and ADAMTS7 are secreted proteins modulating immunotherapy response
AOAH potentiates immunotherapies by potentiating T-cell receptors and dendritic cells
Arachidonoyl phosphatidylcholines, depleted by AOAH, inhibit antitumor immunity

Acknowledgments

This work was supported by the NIH intramural budget (ZIA BC 011889), NCI Flex Synergy Award, and CRI Technology Impact Award (CRI4239) to P.J.. Also, this work was partially supported by the intramural Research Program (ZIA BC 011801 to C.W.) and Theme-based Research Scheme Funds (T12-703/22-R and T12-703/23-N) to X.G. from the Research Grant Council of Hong Kong. X.G. is the Sophie YM Chan Professor in Cancer Research at HKU. L.G., and T.V. were supported by the NCI T2I Postdoc Fellowship. We thank Alexei Gorelik, Bhushan Nagar, Li Peng, Liyuan Fu, Julia Su, Feng Li, Chunxiao Zhang, Baktiar Karim, and Noemi Kedei for their discussions. We thank Kevin Chang for managing our data access applications and agreements. This work utilized the computational resources of the NIH HPC Biowulf cluster.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Declaration of Interests

NCI has filed a provisional patent based on this work, with P.J. and L.G. as inventors.

References

1.Leonard WJ, and Lin J-X (2023). Strategies to therapeutically modulate cytokine action. Nat. Rev. Drug Discov 22, 827–854. [DOI] [PubMed] [Google Scholar]
2.Vaishampayan UN, Tomczak P, Muzaffar J, Winer IS, Rosen SD, Hoimes CJ, Chauhan A, Spreafico A, Lewis KD, Bruno DS, et al. (2022). Nemvaleukin alfa monotherapy and in combination with pembrolizumab in patients (pts) with advanced solid tumors: ARTISTRY-1. J. Clin. Oncol 40, 2500–2500. [Google Scholar]
3.Gilbert MR, Dignam JJ, Armstrong TS, Wefel JS, Blumenthal DT, Vogelbaum MA, Colman H, Chakravarti A, Pugh S, Won M, et al. (2014). A randomized trial of bevacizumab for newly diagnosed glioblastoma. N. Engl. J. Med 370. 10.1056/NEJMoa1308573. [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Chiang CL, Lam TC, Li JCB, Chan KSK, El Helali A, Lee YYP, Law LHT, Zheng D, Lo AWI, Kam NW, et al. (2023). Efficacy, safety, and correlative biomarkers of bintrafusp alfa in recurrent or metastatic nasopharyngeal cancer patients: a phase II clinical trial. Lancet Reg Health West Pac 40, 100898. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Uhlén M, Karlsson MJ, Hober A, Svensson AS, Scheffel J, Kotol D, Zhong W, Tebani A, Strandberg L, Edfors F, et al. (2019). The human secretome. Sci. Signal 12. 10.1126/scisignal.aaz0274. [DOI] [PubMed] [Google Scholar]
6.Lever J, Zhao EY, Grewal J, Jones MR, and Jones SJM (2019). CancerMine: a literature-mined resource for drivers, oncogenes and tumor suppressors in cancer. Nat. Methods 16, 505–507. [DOI] [PubMed] [Google Scholar]
7.Murphy K, and Weaver C (2016). Janeway’s Immunobiology (Garland Science; ). [Google Scholar]
8.Shalem O, Sanjana NE, Hartenian E, Shi X, Scott DA, Mikkelson T, Heckl D, Ebert BL, Root DE, Doench JG, et al. (2014). Genome-scale CRISPR-Cas9 knockout screening in human cells. Science 343. 10.1126/science.1247005. [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Dhainaut M, Rose SA, Akturk G, Wroblewska A, Nielsen SR, Park ES, Buckup M, Roudko V, Pia L, Sweeney R, et al. (2022). Spatial CRISPR genomics identifies regulators of the tumor microenvironment. Cell 185, 1223–1239.e20. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Zhang T, Yu W, Cheng X, Yeung J, Ahumada V, Norris PC, Pearson MJ, Yang X, van Deursen W, Halcovich C, et al. (2024). Up-regulated PLA2G10 in cancer impairs T cell infiltration to dampen immunity. Sci Immunol 9, eadh2334. [DOI] [PubMed] [Google Scholar]
11.Lin H, Lee E, Hestir K, Leo C, Huang M, Bosch E, Halenbeck R, Wu G, Zhou A, Behrens D, et al. (2008). Discovery of a cytokine and its receptor by functional screening of the extracellular proteome. Science 320, 807–811. [DOI] [PubMed] [Google Scholar]
12.Gonzalez R, Jennings LL, Knuth M, Orth AP, Klock HE, Ou W, Feuerhelm J, Hull MV, Koesema E, Wang Y, et al. (2010). Screening the mammalian extracellular proteome for regulators of embryonic human stem cell pluripotency. Proc. Natl. Acad. Sci. U. S. A 107, 3552–3557. [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Jiang P, Sinha S, Aldape K, Hannenhalli S, Sahinalp C, and Ruppin E (2022). Big data in basic and translational cancer research. 22, 625–639. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Ayers M, Lunceford J, Nebozhyn M, Murphy E, Loboda A, Kaufman DR, Albright A, Cheng JD, Kang SP, Shankaran V, et al. (2017). IFN-γ-related mRNA profile predicts clinical response to PD-1 blockade. J. Clin. Invest 127. 10.1172/JCI91190. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Lai J, Mardiana S, House IG, Sek K, Henderson MA, Giuffrida L, Chen AXY, Todd KL, Petley EV, Chan JD, et al. (2020). Adoptive cellular therapy with T cells expressing the dendritic cell growth factor Flt3L drives epitope spreading and antitumor immunity. Nat. Immunol 21. 10.1038/s41590-020-0676-7. [DOI] [PubMed] [Google Scholar]
16.Ng KW, Boumelha J, Enfield KSS, Almagro J, Cha H, Pich O, Karasaki T, Moore DA, Salgado R, Sivakumar M, et al. (2023). Antibodies against endogenous retroviruses promote lung cancer immunotherapy. Nature 616, 563–573. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
17.Zhang Y, Vu T, Palmer DC, Kishton RJ, Gong L, Huang J, Nguyen T, Chen Z, Smith C, Livák F, et al. (2022). A T cell resilience model associated with response to immunotherapy in multiple tumor types. Nat. Med 28. 10.1038/s41591-022-01799-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Fraietta JA, Lacey SF, Orlando EJ, Pruteanu-Malinici I, Gohil M, Lundh S, Boesteanu AC, Wang Y, O’Connor RS, Hwang W-T, et al. (2018). Determinants of response and resistance to CD19 chimeric antigen receptor (CAR) T cell therapy of chronic lymphocytic leukemia. Nat. Med 24, 563–571. [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Krishna S, Lowery FJ, Copeland AR, Bahadiroglu E, Mukherjee R, Jia L, Anibal JT, Sachs A, Adebola SO, Gurusamy D, et al. (2020). Stem-like CD8 T cells mediate response of adoptive cell immunotherapy against human cancer. Science 370. 10.1126/science.abb9847. [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Liu Y, Xun Z, Ma K, Liang S, Li X, Zhou S, Sun L, Liu Y, Du Y, Guo X, et al. (2023). Identification of a tumour immune barrier in the HCC microenvironment that determines the efficacy of immunotherapy. J. Hepatol 78, 770–782. [DOI] [PubMed] [Google Scholar]
21.Zhang S, Yuan L, Danilova L, Mo G, Zhu Q, Deshpande A, Bell ATF, Elisseeff J, Popel AS, Anders RA, et al. (2023). Spatial transcriptomics analysis of neoadjuvant cabozantinib and nivolumab in advanced hepatocellular carcinoma identifies independent mechanisms of resistance and recurrence. Genome Med. 15, 72. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Achar SR, Bourassa FXP, Rademaker TJ, Lee A, Kondo T, Salazar-Cavazos E, Davies JS, Taylor N, François P, and Altan-Bonnet G (2022). Universal antigen encoding of T cell activation from high-dimensional cytokine dynamics. Science 376, 880–884. [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Ruiz de Galarreta M, Bresnahan E, Molina-Sánchez P, Lindblad KE, Maier B, Sia D, Puigvehi M, Miguela V, Casanova-Acebes M, Dhainaut M, et al. (2019). β-Catenin Activation Promotes Immune Escape and Resistance to Anti-PD-1 Therapy in Hepatocellular Carcinoma. Cancer Discov. 9, 1124–1141. [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Sharpless NE, and Depinho RA (2006). The mighty mouse: genetically engineered mouse models in cancer drug development. Nat. Rev. Drug Discov 5. 10.1038/nrd2110. [DOI] [PubMed] [Google Scholar]
25.Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, et al. (2005). Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U. S. A 102. 10.1073/pnas.0506580102. [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Koeberle A, Shindou H, Koeberle SC, Laufer SA, Shimizu T, and Werz O (2013). Arachidonoyl-phosphatidylcholine oscillates during the cell cycle and counteracts proliferation by suppressing Akt membrane binding. Proc. Natl. Acad. Sci. U. S. A 110. 10.1073/pnas.1216182110. [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Munford RS, and Hunter JP (1992). Acyloxyacyl hydrolase, a leukocyte enzyme that deacylates bacterial lipopolysaccharides, has phospholipase, lysophospholipase, diacylglycerollipase, and acyltransferase activities in vitro. J. Biol. Chem 267. [PubMed] [Google Scholar]
28.Vardhana SA, Hwee MA, Berisa M, Wells DK, Yost KE, King B, Smith M, Herrera PS, Chang HY, Satpathy AT, et al. (2020). Impaired mitochondrial oxidative phosphorylation limits the self-renewal of T cells exposed to persistent antigen. Nat. Immunol 21. 10.1038/s41590-020-0725-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Gide TN, Quek C, Menzies AM, Tasker AT, Shang P, Holst J, Madore J, Lim SY, Velickovic R, Wongchenko M, et al. (2019). Distinct Immune Cell Populations Define Response to Anti-PD-1 Monotherapy and Anti-PD-1/Anti-CTLA-4 Combined Therapy. Cancer cell 35. 10.1016/j.ccell.2019.01.003. [DOI] [PubMed] [Google Scholar]
30.Carnevale J, Shifrut E, Kale N, Nyberg WA, Blaeschke F, Chen YY, Li Z, Bapat SP, Diolaiti ME, O’Leary P, et al. (2022). RASA2 ablation in T cells boosts antigen sensitivity and long-term function. Nature 609. 10.1038/s41586-022-05126-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Zou B, Goodwin M, Saleem D, Jiang W, Tang J, Chu Y, Munford RS, and Lu M (2021). A highly conserved host lipase deacylates oxidized phospholipids and ameliorates acute lung injury in mice. Elife 10. 10.7554/eLife.70938. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Lehninger AL, Nelson DL, and Cox MM (2005). Lehninger Principles of Biochemistry (Macmillan; ). [Google Scholar]
33.Marai L, and Kuksis A (1969). Molecular species of lecithins from erythrocytes and plasma of man. J. Lipid Res 10. [PubMed] [Google Scholar]
34.Ni Z, Sousa BC, Colombo S, Afonso CB, Melo T, Pitt AR, Spickett CM, Domingues P, Domingues MR, Fedorova M, et al. (2019). Evaluation of air oxidized PAPC: A multi laboratory study by LC-MS/MS. Free Radic Biol Med 144, 156–166. [DOI] [PubMed] [Google Scholar]
35.Chen X, Song M, Zhang B, and Zhang Y (2016). Reactive Oxygen Species Regulate T Cell Immune Response in the Tumor Microenvironment. Oxidative medicine and cellular longevity 2016. 10.1155/2016/1580967. [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Blüml S, Kirchberger S, Bochkov VN, Krönke G, Stuhlmeier K, Majdic O, Zlabinger GJ, Knapp W, Binder BR, Stöckl J, et al. (2005). Oxidized phospholipids negatively regulate dendritic cell maturation induced by TLRs and CD40. J Immunol 175, 501–508. [DOI] [PubMed] [Google Scholar]
37.Ramakrishnan R, Tyurin VA, Veglia F, Condamine T, Amoscato A, Mohammadyani D, Johnson JJ, Zhang LM, Klein-Seetharaman J, Celis E, et al. (2014). Oxidized lipids block antigen cross-presentation by dendritic cells in cancer. J Immunol 192, 2920–2931. [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Seok JK, Hong E-H, Yang G, Lee HE, Kim S-E, Liu K-H, Kang HC, Cho Y-Y, Lee HS, and Lee JY (2021). Oxidized Phospholipids in Tumor Microenvironment Stimulate Tumor Metastasis via Regulation of Autophagy. Cells 10. 10.3390/cells10030558. [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Godzien J, Lopez-Lopez A, Sieminska J, Jablonowski K, Pietrowska K, Kisluk J, Mojsak M, Dzieciol-Anikiej Z, Barbas C, Reszec J, et al. (2023). Exploration of oxidized phosphocholine profile in non-small-cell lung cancer. Front Mol Biosci 10, 1279645. [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Gorelik A, Illes K, and Nagar B (2018). Crystal structure of the mammalian lipopolysaccharide detoxifier. Proc. Natl. Acad. Sci. U. S. A 115, E896–E905. [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Ghandi M, Huang FW, Jané-Valbuena J, Kryukov GV, Lo CC, McDonald ER 3rd, Barretina J, Gelfand ET, Bielski CM, Li H, et al. (2019). Next-generation characterization of the Cancer Cell Line Encyclopedia. Nature 569, 503–508. [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Pan D, Kobayashi A, Jiang P, de Andrade LF, Tay RE, Luoma AM, Tsoucas D, Qiu X, Lim K, Rao P, et al. (2018). A major chromatin regulator determines resistance of tumor cells to T cell-mediated killing. Science 359. 10.1126/science.aao1710. [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Bolotin DA, Poslavsky S, Davydov AN, Frenkel FE, Fanchi L, Zolotareva OI, Hemmers S, Putintseva EV, Obraztsova AS, Shugay M, et al. (2017). Antigen receptor repertoire profiling from RNA-seq data. Nature biotechnology 35. 10.1038/nbt.3979. [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Liberzon A, Birger C, Thorvaldsdóttir H, Ghandi M, Mesirov JP, and Tamayo P (2015). The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell systems 1. 10.1016/j.cels.2015.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Pedersen BS, and Quinlan AR (2018). Mosdepth: quick coverage calculation for genomes and exomes. Bioinformatics 34, 867–868. [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Wingett SW, and Andrews S (2018). FastQ Screen: A tool for multi-genome mapping and quality control. F1000Res 7, 1338. [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Liu D, Schilling B, Liu D, Sucker A, Livingstone E, Jerby-Arnon L, Zimmer L, Gutzmer R, Satzger I, Loquai C, et al. (2019). Integrative molecular and clinical modeling of clinical outcomes to PD1 blockade in patients with metastatic melanoma. Nature medicine 25. 10.1038/s41591-019-0654-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Roper N, Velez MJ, Chiappori A, Kim YS, Wei JS, Sindiri S, Takahashi N, Mulford D, Kumar S, Ylaya K, et al. (2021). Notch signaling and efficacy of PD-1/PD-L1 blockade in relapsed small cell lung cancer. Nat Commun 12, 3880. [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Yoshihara K, Shahmoradgoli M, Martínez E, Vegesna R, Kim H, Torres-Garcia W, Treviño V, Shen H, Laird PW, Levine DA, et al. (2013). Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun 4, 2612. [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Raine KM, Van Loo P, Wedge DC, Jones D, Menzies A, Butler AP, Teague JW, Tarpey P, Nik-Zainal S, and Campbell PJ (2016). ascatNgs: Identifying Somatically Acquired Copy-Number Alterations from Whole-Genome Sequencing Data. Curr Protoc Bioinformatics 56, 15.9.1–15.9.17. [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Padrón LJ, Maurer DM, O’Hara MH, O’Reilly EM, Wolff RA, Wainberg ZA, Ko AH, Fisher G, Rahma O, Lyman JP, et al. (2022). Sotigalimab and/or nivolumab with chemotherapy in first-line metastatic pancreatic cancer: clinical and immunologic analyses from the randomized phase 2 PRINCE trial. Nat. Med 28, 1167–1177. [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Jiang P, Gu S, Pan D, Fu J, Sahu A, Hu X, Li Z, Traugh N, Bu X, Li B, et al. (2018). Signatures of T cell dysfunction and exclusion predict cancer immunotherapy response. Nat. Med 24. 10.1038/s41591-018-0136-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Jiang P, Zhang Y, Ru B, Yang Y, Vu T, Paul R, Mirza A, Altan-Bonnet G, Liu L, Ruppin E, et al. (2021). Systematic investigation of cytokine signaling activity at the tissue and single-cell levels. Nat. Methods 18. 10.1038/s41592-021-01274-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Van Rossum G, and Drake FL (2009). Python 3 Reference Manual (CreateSpace; ). [Google Scholar]
55.Virtanen P, Gommers R, Oliphant TE, Haberland M, Reddy T, Cournapeau D, Burovski E, Peterson P, Weckesser W, Bright J, et al. (2020). SciPy 1.0: fundamental algorithms for scientific computing in Python. Nat Methods 17, 261–272. [DOI] [PMC free article] [PubMed] [Google Scholar]
56.R Core Team (2021). R: A Language and Environment for Statistical Computing. Preprint at R Foundation for Statistical Computing. [Google Scholar]
57.Chen S, Zhou Y, Chen Y, and Gu J (2018). fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 34, i884–i890. [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, Marth G, Abecasis G, Durbin R, and 1000 Genome Project Data Processing Subgroup (2009). The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079. [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Quinlan AR, and Hall IM (2010). BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841–842. [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Patro R, Duggal G, Love MI, Irizarry RA, and Kingsford C (2017). Salmon provides fast and bias-aware quantification of transcript expression. Nat. Methods 14, 417–419. [DOI] [PMC free article] [PubMed] [Google Scholar]
61.Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, and Gingeras TR (2013). STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29. 10.1093/bioinformatics/bts635. [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Love MI, Huber W, and Anders S (2014). Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550. [DOI] [PMC free article] [PubMed] [Google Scholar]
63.Li B, and Dewey CN (2011). RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics 12. 10.1186/1471-2105-12-323. [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Bolotin DA, Poslavsky S, Mitrophanov I, Shugay M, Mamedov IZ, Putintseva EV, and Chudakov DM (2015). MiXCR: software for comprehensive adaptive immunity profiling. Nat. Methods 12. 10.1038/nmeth.3364. [DOI] [PubMed] [Google Scholar]
65.Vasimuddin M, Misra S, Li H, and Aluru S (2019). Efficient Architecture-Aware Acceleration of BWA-MEM for Multicore Systems. In 2019 IEEE International Parallel and Distributed Processing Symposium (IPDPS), pp. 314–324. [Google Scholar]
66.Koboldt DC, Chen K, Wylie T, Larson DE, McLellan, Mardis ER, Weinstock GM, Wilson RK, and Ding L (2009). VarScan: variant detection in massively parallel sequencing of individual and pooled samples. Bioinformatics 25. 10.1093/bioinformatics/btp373. [DOI] [PMC free article] [PubMed] [Google Scholar]
67.McLaren W, Gil L, Hunt SE, Riat HS, Ritchie GR, Thormann A, Flicek P, and Cunningham F (2016). The Ensembl Variant Effect Predictor. Genome Biol. 17. 10.1186/s13059-016-0974-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Kandoth C, Gao J, qwangmsk, Mattioni M, Struck A, Boursin Y, Penson A, and Chavan S (2018). mskcc/vcf2maf: vcf2maf v1.6.16 (Zenodo) 10.5281/ZENODO.593251. [DOI] [Google Scholar]
69.McKenna A, Hanna M, Banks E, Sivachenko A, Cibulskis K, Kernytsky A, Garimella K, Altshuler D, Gabriel S, Daly M, et al. (2010). The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 20, 1297–1303. [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Kim S, Scheffler K, Halpern AL, Bekritsky MA, Noh E, Källberg M, Chen X, Kim Y, Beyter D, Krusche P, et al. (2018). Strelka2: fast and accurate calling of germline and somatic variants. Nat. Methods 15, 591–594. [DOI] [PubMed] [Google Scholar]
71.Chakravarty D, Gao J, Phillips SM, Kundra R, Zhang H, Wang J, Rudolph JE, Yaeger R, Soumerai T, Nissan MH, et al. (2017). OncoKB: A Precision Oncology Knowledge Base. JCO Precis Oncol 2017. 10.1200/PO.17.00011. [DOI] [PMC free article] [PubMed] [Google Scholar]
72.Martin M (2011). Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet. journal 17, 10–12. [Google Scholar]
73.Kopylova E, Noé L, and Touzet H (2012). SortMeRNA: fast and accurate filtering of ribosomal RNAs in metatranscriptomic data. Bioinformatics 28, 3211–3217. [DOI] [PubMed] [Google Scholar]
74.Ru B, Huang J, Zhang Y, Aldape K, and Jiang P (2023). Estimation of cell lineages in tumors from spatial transcriptomics data. Nat. Commun 14, 568. [DOI] [PMC free article] [PubMed] [Google Scholar]
75.Hao Y, Stuart T, Kowalski MH, Choudhary S, Hoffman P, Hartman A, Srivastava A, Molla G, Madad S, Fernandez-Granda C, et al. (2024). Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat. Biotechnol 42, 293–304. [DOI] [PMC free article] [PubMed] [Google Scholar]
76.Wolock SL, Lopez R, and Klein AM (2019). Scrublet: Computational Identification of Cell Doublets in Single-Cell Transcriptomic Data. Cell systems 8. 10.1016/j.cels.2018.11.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
77.Gao Y, Patro R, and Jiang P (2024). Collapsible tree: interactive web app to present collapsible hierarchies. Bioinformatics 40. 10.1093/bioinformatics/btae645. [DOI] [PMC free article] [PubMed] [Google Scholar]
78.Mailman, Feolo M, Jin Y, Kimura M, Tryka K, Bagoutdinov R, Hao L, Kiang A, Paschall J, Phan L, et al. (2007). The NCBI dbGaP database of genotypes and phenotypes. Nat. Genet 39. 10.1038/ng1007-1181. [DOI] [PMC free article] [PubMed] [Google Scholar]
79.Freeberg MA, Fromont LA, D’Altri T, Romero AF, Ciges JI, Jene A, Kerry G, Moldes M, Ariosa R, Bahena S, et al. (2022). The European Genome-phenome Archive in 2021. Nucleic Acids Res. 50. 10.1093/nar/gkab1059. [DOI] [PMC free article] [PubMed] [Google Scholar]
80.Campbell KM, Amouzgar M, Pfeiffer SM, Howes TR, Medina E, Travers M, Steiner G, Weber JS, Wolchok JD, Larkin J, et al. (2023). Prior anti-CTLA-4 therapy impacts molecular characteristics associated with anti-PD-1 response in advanced melanoma. Cancer Cell 41, 791–806.e4. [DOI] [PMC free article] [PubMed] [Google Scholar]
81.Hugo W, Zaretsky JM, Sun L, Song C, Moreno BH, Hu-Lieskovan S, Berent-Maoz B, Pang J, Chmielowski B, Cherry G, et al. (2016). Genomic and Transcriptomic Features of Response to Anti-PD-1 Therapy in Metastatic Melanoma. Cell 165. 10.1016/j.cell.2016.02.065. [DOI] [PMC free article] [PubMed] [Google Scholar]
82.Pusztai L, Yau C, Wolf DM, Han HS, Du L, Wallace AM, String-Reasor E, Boughey JC, Chien AJ, Elias AD, et al. (2021). Durvalumab with olaparib and paclitaxel for high-risk HER2-negative stage II/III breast cancer: Results from the adaptively randomized I-SPY2 trial. Cancer cell 39. 10.1016/j.ccell.2021.05.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
83.Wolf DM, Yau C, Wulfkuhle J, Brown-Swigart L, Gallagher RI, Lee PRE, Zhu Z, Magbanua MJ, Sayaman R, O’Grady N, et al. (2022). Redefining breast cancer subtypes to guide treatment prioritization and maximize response: Predictive biomarkers across 10 cancer therapies. Cancer cell 40. 10.1016/j.ccell.2022.05.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
84.Anagnostou V, Niknafs N, Marrone K, Bruhm DC, White JR, Naidoo J, Hummelink K, Monkhorst K, Lalezari F, Lanis M, et al. (2020). Multimodal genomic features predict outcome of immune checkpoint blockade in non-small-cell lung cancer. Nat Cancer 1, 99–111. [DOI] [PMC free article] [PubMed] [Google Scholar]
85.Budhu A, Pehrsson EC, He A, Goyal L, Kelley RK, Dang H, Xie C, Monge C, Tandon M, Ma L, et al. (2023). Tumor biology and immune infiltration define primary liver cancer subsets linked to overall survival after immunotherapy. Cell Rep Med 4, 101052. [DOI] [PMC free article] [PubMed] [Google Scholar]
86.Jacobsen SB, Tfelt-Hansen J, Smerup MH, Andersen JD, and Morling N (2023). Comparison of whole transcriptome sequencing of fresh, frozen, and formalin-fixed, paraffin-embedded cardiac tissue. PLoS One 18, e0283159. [DOI] [PMC free article] [PubMed] [Google Scholar]
87.Johnson WE, Li C, and Rabinovic A (2007). Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics (Oxford, England) 8. 10.1093/biostatistics/kxj037. [DOI] [PubMed] [Google Scholar]
88.Nygaard V, Rødland EA, and Hovig E (2016). Methods that remove batch effects while retaining group differences may lead to exaggerated confidence in downstream analyses. Biostatistics 17, 29–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
89.Sy CY, Lien SC, Wang BX, Clouthier DL, Hanna Y, Cirlan I, Zhu K, Bruce JP, El Ghamrasni S, Iafolla MAJ, et al. (2021). Pan-cancer analysis of longitudinal metastatic tumors reveals genomic alterations and immune landscape dynamics associated with pembrolizumab sensitivity. Nat. Commun 12. 10.1038/s41467-021-25432-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
90.Li G, Choi JE, Kryczek I, Sun Y, Liao P, Li S, Wei S, Grove S, Vatan L, Nelson R, et al. (2023). Intersection of immune and oncometabolic pathways drives cancer hyperprogression during immunotherapy. Cancer Cell 41, 304–322.e7. [DOI] [PMC free article] [PubMed] [Google Scholar]
91.Fehrenbacher L, Spira A, Ballinger M, Kowanetz M, Vansteenkiste J, Mazieres J, Park K, Smith D, Artal-Cortes A, Lewanski C, et al. (2016). Atezolizumab versus docetaxel for patients with previously treated non-small-cell lung cancer (POPLAR): a multicentre, open-label, phase 2 randomised controlled trial. Lancet (London, England) 387. 10.1016/S0140-6736(16)00587-0. [DOI] [PubMed] [Google Scholar]
92.Rooney MS, Shukla SA, Wu CJ, Getz G, and Hacohen N (2015). Molecular and genetic properties of tumors associated with local immune cytolytic activity. Cell 160. 10.1016/j.cell.2014.12.033. [DOI] [PMC free article] [PubMed] [Google Scholar]
93.Auslander N, Zhang G, Lee JS, Frederick DT, Miao B, Moll T, Tian T, Wei Z, Madan S, Sullivan RJ, et al. (2018). Robust prediction of response to immune checkpoint blockade therapy in metastatic melanoma. Nat Med 24, 1545–1549. [DOI] [PMC free article] [PubMed] [Google Scholar]
94.Litchfield K, Reading JL, Puttick C, Thakkar K, Abbosh C, Bentham R, Watkins TBK, Rosenthal R, Biswas D, Rowan A, et al. (2021). Meta-analysis of tumor- and T cell-intrinsic mechanisms of sensitization to checkpoint inhibition. Cell 184. 10.1016/j.cell.2021.01.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
95.Spranger S, Dai D, Horton B, and Gajewski TF (2017). Tumor-Residing Batf3 Dendritic Cells Are Required for Effector T Cell Trafficking and Adoptive T Cell Therapy. Cancer cell 31. 10.1016/j.ccell.2017.04.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
96.Davoli T, Uno H, Wooten EC, and Elledge SJ (2017). Tumor aneuploidy correlates with markers of immune evasion and with reduced response to immunotherapy. Science 355. 10.1126/science.aaf8399. [DOI] [PMC free article] [PubMed] [Google Scholar]
97.Mariathasan S, Turley SJ, Nickles D, Castiglioni A, Yuen K, Wang Y, Kadel EE, Koeppen H, Astarita JL, Cubas R, et al. (2018). TGFβ attenuates tumour response to PD-L1 blockade by contributing to exclusion of T cells. Nature 554. 10.1038/nature25501. [DOI] [PMC free article] [PubMed] [Google Scholar]
98.Veis DJ, Sentman CL, Bach EA, and Korsmeyer SJ (1993). Expression of the Bcl-2 protein in murine and human thymocytes and in peripheral T lymphocytes. J Immunol 151, 2546–2554. [PubMed] [Google Scholar]
99.Mélique S, Vadel A, Rouquié N, Yang C, Bories C, Cotineau C, Saoudi A, Fazilleau N, and Lesourne R (2024). THEMIS promotes T cell development and maintenance by rising the signaling threshold of the inhibitory receptor BTLA. Proc Natl Acad Sci U S A 121, e2318773121. [DOI] [PMC free article] [PubMed] [Google Scholar]
100.Nolte M, and Margadant C (2020). Controlling Immunity and Inflammation through Integrin-Dependent Regulation of TGF-β. Trends in cell biology 30. 10.1016/j.tcb.2019.10.002. [DOI] [PubMed] [Google Scholar]
101.Dodagatta-Marri E, Ma HY, Liang B, Li J, Meyer DS, Chen SY, Sun KH, Ren X, Zivak B, Rosenblum, et al. (2021). Integrin αvβ8 on T cells suppresses anti-tumor immunity in multiple models and is a promising target for tumor immunotherapy. Cell reports 36. 10.1016/j.celrep.2021.109309. [DOI] [PMC free article] [PubMed] [Google Scholar]
102.Jaffe DB, Shahi P, Adams BA, Chrisman AM, Finnegan PM, Raman N, Royall AE, Tsai F, Vollbrecht T, Reyes DS, et al. (2022). enclone: precision clonotyping and analysis of immune receptors. bioRxiv, 2022.04.21.489084. 10.1101/2022.04.21.489084. [DOI] [Google Scholar]
103.Tumeh PC, Yearley JH, Shintaku IP, Taylor EJ, Robert L, Chmielowski B, Spasic M, Henry G, Ciobanu V, West AN, et al. (2014). PD-1 blockade induces responses by inhibiting adaptive immune resistance. Nature 515. 10.1038/nature13954. [DOI] [PMC free article] [PubMed] [Google Scholar]
104.Sornasse T, Flamand V, De Becker G, Bazin H, Tielemans F, Thielemans K, Urbain J, Leo O, and Moser M (1992). Antigen-pulsed dendritic cells can efficiently induce an antibody response in vivo. The Journal of experimental medicine 175. 10.1084/jem.175.1.15. [DOI] [PMC free article] [PubMed] [Google Scholar]
105.Okada N, Tsujino M, Hagiwara Y, Tada A, Tamura Y, Mori K, Saito T, Nakagawa S, Mayumi T, Fujita T, et al. (2001). Administration route-dependent vaccine efficiency of murine dendritic cells pulsed with antigens. Br J Cancer 84, 1564–1570. [DOI] [PMC free article] [PubMed] [Google Scholar]
106.Firth D (March 1993). Bias reduction of maximum likelihood estimates. Biometrika 80, 27–38. [Google Scholar]
107.Uhlén M, Fagerberg L, Hallström BM, Lindskog C, Oksvold P, Mardinoglu A, Sivertsson Å, Kampf C, Sjöstedt E, Asplund A, et al. (2015). Proteomics. Tissue-based map of the human proteome. Science 347. 10.1126/science.1260419. [DOI] [PubMed] [Google Scholar]
108.Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, Dolinski K, Dwight SS, Eppig JT, et al. (2000). Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25, 25–29. [DOI] [PMC free article] [PubMed] [Google Scholar]
109.Efremova M, Vento-Tormo M, Teichmann SA, and Vento-Tormo R (2020). CellPhoneDB: inferring cell-cell communication from combined expression of multi-subunit ligand-receptor complexes. Nat Protoc 15, 1484–1506. [DOI] [PubMed] [Google Scholar]
110.Benjamini Y, and Hochberg Y (1995). Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. J. R. Stat. Soc. Series B Stat. Methodol 57, 289–300. [Google Scholar]
111.Li Y, Dou Y, Da Veiga Leprevost F, Geffen Y, Calinawan AP, Aguet F, Akiyama Y, Anand S, Birger C, Cao S, et al. (2023). Proteogenomic data and resources for pan-cancer analysis. Cancer Cell 41, 1397–1406. [DOI] [PMC free article] [PubMed] [Google Scholar]
112.Martínez-Carpio PA, Mur C, Fernández-Montolí ME, Ramon JM, Rosel P, and Navarro MA (1999). Secretion and dual regulation between epidermal growth factor and transforming growth factor-beta1 in MDA-MB-231 cell line in 42-hour-long cultures. Cancer letters 147. 10.1016/s0304-3835(99)00261-x. [DOI] [PubMed] [Google Scholar]
113.Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, Jacobsen A, Byrne CJ, Heuer ML, Larsson E, et al. (2012). The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer discovery 2. 10.1158/2159-8290.CD-12-0095. [DOI] [PMC free article] [PubMed] [Google Scholar]
114.Zaykin DV, Zhivotovsky LA, Westfall PH, and Weir BS (2002). Truncated product method for combining P-values. Genetic epidemiology 22. 10.1002/gepi.0042. [DOI] [PubMed] [Google Scholar]
115.Fisher RA (1970). Statistical methods for research workers. In Breakthroughs in statistics: Methodology and distribution (Springer; ), pp. 66–70. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Figure S1. Computational frameworks and analyses, related to Figure 1.

(A) RNA and Whole-exome sequencing data processing pipelines.

NIHMS2096169-supplement-1.pdf^{(21.1MB, pdf)}

Figure S2. Gene risk scores from immunotherapy transcriptomics cohorts, related to Figure 2.

(D) Number of genes with FDR < 0.05 for diverse methods. The Wilcoxon approach that we selected is among the more conservative approaches.

NIHMS2096169-supplement-2.pdf^{(4.5MB, pdf)}

Figure S3. Predicted secreted regulators of cancer immunotherapy outcomes, related to Figure 3.

(A) Risk scores with tumor purity correction, shown as in Figure 3A. The risk scores for top candidates were re-computed by adding the tumor purity estimation as a covariate in the regression model.

(E) Correlations with CTL infiltrations for prioritized secreted proteins, shown as in Figure 3A.

(I) Growth of subcutaneous B16-mhgp100 tumors with Aoah overexpression in immunodeficient NSG mice, shown as in Figure 3E.

(J) Growth of subcutaneous B16F10 tumors with Aoah overexpression in C57BL/6 mice treated with IgG controls, shown as in Figure 3E.

(K) Tumor growth curves for B16 tumor models with strong (N4) and weak (V4) antigens, shown as in Figure 3E.

(L) Endpoint-free survival plots for panel K, shown as in Figure 3F.

NIHMS2096169-supplement-3.pdf^{(10.6MB, pdf)}

Figure S4. In vivo validation of AOAH, related to Figure 4.

(A) The design of plasmids used for the hydrodynamic tail vein injection (HDTVi) to induce spontaneous hepatocellular carcinoma (HCC) in mouse livers.

(C) Tumor growth and endpoint-free survival of B16-mhgp100 tumors in C57BL/6 mice with intraperitoneal injections of DC2.4 with Aoah or vector overexpression, shown as in Figure 4C.

(E) The body weight change of B16F10-bearing mice after TCR-T transfer, shown as in Figure 4D.

(H) The body weight change of B16F10 tumor-bearing mice during rhAOAH (5mg/kg) + immune checkpoint blockade treatments, shown as in Figure 4D.

(I) Additional representative multi-IF images of major TIL subtypes in the MYC β-catenin induced spontaneous hepatocellular carcinoma (HCC) sections.

NIHMS2096169-supplement-4.pdf^{(8.1MB, pdf)}

Figure S5. In vivo molecular landscape induced by Aoah, related to Figure 5.

(A) Hierarchical clustering of log2FC profiles, using Pearson correlation as the distance.

(B) IFNγ activities predicted by the CytoSig package⁵³. The p-value is computed through the two-sided permutation test with 1000 randomizations.

(D) B-cell receptor (BCR) immunoglobulin heavy chain (IGH) clonality for B cells, compared as in Figure 5G.

(E) IGH class fractions.

(F) Top 20 enriched terms from the gene ontology biological processes (GO_BP), shown as in Figure 5H that utilized KEGG terms.

(G) Co-enriched pathways for Aoah, Colq, and Adamts7 overexpressions, shown as in Figure 5K.

(H) The association between the CR1L expression level and overall survival in a melanoma cohort with anti-PD1 treatment²⁹, shown as in Figure 2A.

NIHMS2096169-supplement-5.pdf^{(2MB, pdf)}

Figure S6. Validations of TCR activation by AOAH, related to Figure 6.

(F) Flow cytometry gating corresponding to Figure 6D, E, and J.

(G) Flow cytometry gating corresponding to Figure 6G.

(J) Validation of key genes associated with cytotoxicity and T-cell activation using qRT-PCR.

NIHMS2096169-supplement-6.pdf^{(2.7MB, pdf)}

Figure S7. Lipidomics studies of AOAH-mediated TCR activation, related to Figure 7.

(A) Enriched lipid species by rhAOAH, shown as in Figure 7A. Only enriched lipids with log2FC > 1 are shown.

(E) Flow cytometry analysis of p-CD3ζ, p-LCK, and CD25 in mouse CD8+ T cells after TCR activation, corresponding to Figure 7G.

(F) Flow cytometry analysis of CD25 and CD69 in human CD8+ T cells after TCR activation, corresponding to Figure 7H.

(G) Flow cytometry analysis of the fluorescent shift of the lipid peroxidation sensor in mouse CD8+ T cells after TCR activation, corresponding to Figure 7I.

(J) Flow cytometry analysis of CD25 and CD69 in mouse CD8+ T cells after TCR activation using wild-type rhAOAH, mutant rhAOAH, and vehicle treatment, corresponding to Figure 7N.

NIHMS2096169-supplement-7.pdf^{(873.5KB, pdf)}

Table S1. Cancer immunotherapy data from clinical studies, related to Figure 1.

This table has five tabs, enumerated below. The meaning of most columns is evident from their names, with exceptions explained.

Dataset_Bulk: Information about bulk-tumor omics datasets from cancer immunotherapy clinical studies. The Category column presents the following dataset types:
- Expression: gene expression data
- Mutation: gene somatic mutations by whole-exome sequencing (WES).
- CNA: copy number alterations computed from WES data
The “Raw data” column indicates whether the raw sequencing data is available for unified pre-processing with the CIDE framework. The “Source” column shows the original places or accession numbers for data download.
Analysis_Bulk: analyses performed on the datasets in Dataset_Bulk. IDs in the first column match those in the Dataset_Bulk ID column. The Method column presents analysis approaches for different clinical outcome types:
- Cox-PH: Cox proportional hazard (PH) regression with the two-sided Wald test for survival outcomes.
- OLS: Ordinary Least Squares regression with the two-sided t-test for the Response Evaluation Criteria in Solid Tumors (RECIST) outcomes.
- Rank-sum: Two-sided Wilcoxon rank-sum test for the binary response without clinical covariates to adjust.
- Logit: Logistic regression with the two-sided Wald test and Firth correction [S1] for the binary response with clinical covariates to adjust.
The column “Result file” presents the file names of the statistical analysis results, which are included in “gene_scores.zip” in Zenodo (https://doi.org/10.5281/zenodo.15012968).
Dataset_Single_Cell: information about single-cell RNA-seq datasets from cancer immunotherapy clinical studies.
Analysis_Single_Cell: analyses performed on the datasets in Dataset_Single_Cell, itemized by cell types. IDs in the first column match those in the Dataset_Single_Cell ID column. The Method column is the same as “Analysis_Bulk”. The column “Result file” presents the file names of the statistical analysis results, which are included in “gene_scores.zip” in Zenodo (https://doi.org/10.5281/zenodo.15012968).
Dataset_Lineage: Single-cell RNA-seq data showing cell lineage expression of genes, adapted from a previous study [S2].

NIHMS2096169-supplement-9.xlsx^{(34.5KB, xlsx)}

NIHMS2096169-supplement-8.pdf^{(330.4KB, pdf)}

Table S5. Clinical endpoints and covariates in datasets, related to Figure 2.

The table includes details of clinical information contained in our analysis. Each row presents a study providing one or more cohorts.

The column group “Stratification” includes the columns for stratifying omics data from one study into sub-cohorts for analysis. In each cell, distinct sub-cohorts were separated by “|”.
The column “Endpoints” presents clinical outcome types available for each study, separated by “|”. Each endpoint type has its own statistical analysis.
The column group “Regression Covariates” includes the covariates adjusted in the regression analysis. In each cell, distinct covariates were separated by “&”.

NIHMS2096169-supplement-10.xlsx^{(21.5KB, xlsx)}

Table S6. Gene function annotation, related to Figure 2.

NIHMS2096169-supplement-11.xlsx^{(35.3KB, xlsx)}

Table S7. Statistical significance and sample counts, related to Figures 3 – 7.

(A) Survival analyses for mouse in vivo experiments, related to Figures 3 and 4.

(B) RNA-seq analyses for mouse in vivo experiments, related to Figure 5.

NIHMS2096169-supplement-12.xlsx^{(26.4KB, xlsx)}

Data Availability Statement

[R1] 1.Leonard WJ, and Lin J-X (2023). Strategies to therapeutically modulate cytokine action. Nat. Rev. Drug Discov 22, 827–854. [DOI] [PubMed] [Google Scholar]

[R2] 2.Vaishampayan UN, Tomczak P, Muzaffar J, Winer IS, Rosen SD, Hoimes CJ, Chauhan A, Spreafico A, Lewis KD, Bruno DS, et al. (2022). Nemvaleukin alfa monotherapy and in combination with pembrolizumab in patients (pts) with advanced solid tumors: ARTISTRY-1. J. Clin. Oncol 40, 2500–2500. [Google Scholar]

[R3] 3.Gilbert MR, Dignam JJ, Armstrong TS, Wefel JS, Blumenthal DT, Vogelbaum MA, Colman H, Chakravarti A, Pugh S, Won M, et al. (2014). A randomized trial of bevacizumab for newly diagnosed glioblastoma. N. Engl. J. Med 370. 10.1056/NEJMoa1308573. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R4] 4.Chiang CL, Lam TC, Li JCB, Chan KSK, El Helali A, Lee YYP, Law LHT, Zheng D, Lo AWI, Kam NW, et al. (2023). Efficacy, safety, and correlative biomarkers of bintrafusp alfa in recurrent or metastatic nasopharyngeal cancer patients: a phase II clinical trial. Lancet Reg Health West Pac 40, 100898. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R5] 5.Uhlén M, Karlsson MJ, Hober A, Svensson AS, Scheffel J, Kotol D, Zhong W, Tebani A, Strandberg L, Edfors F, et al. (2019). The human secretome. Sci. Signal 12. 10.1126/scisignal.aaz0274. [DOI] [PubMed] [Google Scholar]

[R6] 6.Lever J, Zhao EY, Grewal J, Jones MR, and Jones SJM (2019). CancerMine: a literature-mined resource for drivers, oncogenes and tumor suppressors in cancer. Nat. Methods 16, 505–507. [DOI] [PubMed] [Google Scholar]

[R7] 7.Murphy K, and Weaver C (2016). Janeway’s Immunobiology (Garland Science; ). [Google Scholar]

[R8] 8.Shalem O, Sanjana NE, Hartenian E, Shi X, Scott DA, Mikkelson T, Heckl D, Ebert BL, Root DE, Doench JG, et al. (2014). Genome-scale CRISPR-Cas9 knockout screening in human cells. Science 343. 10.1126/science.1247005. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R9] 9.Dhainaut M, Rose SA, Akturk G, Wroblewska A, Nielsen SR, Park ES, Buckup M, Roudko V, Pia L, Sweeney R, et al. (2022). Spatial CRISPR genomics identifies regulators of the tumor microenvironment. Cell 185, 1223–1239.e20. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R10] 10.Zhang T, Yu W, Cheng X, Yeung J, Ahumada V, Norris PC, Pearson MJ, Yang X, van Deursen W, Halcovich C, et al. (2024). Up-regulated PLA2G10 in cancer impairs T cell infiltration to dampen immunity. Sci Immunol 9, eadh2334. [DOI] [PubMed] [Google Scholar]

[R11] 11.Lin H, Lee E, Hestir K, Leo C, Huang M, Bosch E, Halenbeck R, Wu G, Zhou A, Behrens D, et al. (2008). Discovery of a cytokine and its receptor by functional screening of the extracellular proteome. Science 320, 807–811. [DOI] [PubMed] [Google Scholar]

[R12] 12.Gonzalez R, Jennings LL, Knuth M, Orth AP, Klock HE, Ou W, Feuerhelm J, Hull MV, Koesema E, Wang Y, et al. (2010). Screening the mammalian extracellular proteome for regulators of embryonic human stem cell pluripotency. Proc. Natl. Acad. Sci. U. S. A 107, 3552–3557. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R13] 13.Jiang P, Sinha S, Aldape K, Hannenhalli S, Sahinalp C, and Ruppin E (2022). Big data in basic and translational cancer research. 22, 625–639. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R14] 14.Ayers M, Lunceford J, Nebozhyn M, Murphy E, Loboda A, Kaufman DR, Albright A, Cheng JD, Kang SP, Shankaran V, et al. (2017). IFN-γ-related mRNA profile predicts clinical response to PD-1 blockade. J. Clin. Invest 127. 10.1172/JCI91190. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R15] 15.Lai J, Mardiana S, House IG, Sek K, Henderson MA, Giuffrida L, Chen AXY, Todd KL, Petley EV, Chan JD, et al. (2020). Adoptive cellular therapy with T cells expressing the dendritic cell growth factor Flt3L drives epitope spreading and antitumor immunity. Nat. Immunol 21. 10.1038/s41590-020-0676-7. [DOI] [PubMed] [Google Scholar]

[R16] 16.Ng KW, Boumelha J, Enfield KSS, Almagro J, Cha H, Pich O, Karasaki T, Moore DA, Salgado R, Sivakumar M, et al. (2023). Antibodies against endogenous retroviruses promote lung cancer immunotherapy. Nature 616, 563–573. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]

[R17] 17.Zhang Y, Vu T, Palmer DC, Kishton RJ, Gong L, Huang J, Nguyen T, Chen Z, Smith C, Livák F, et al. (2022). A T cell resilience model associated with response to immunotherapy in multiple tumor types. Nat. Med 28. 10.1038/s41591-022-01799-y. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R18] 18.Fraietta JA, Lacey SF, Orlando EJ, Pruteanu-Malinici I, Gohil M, Lundh S, Boesteanu AC, Wang Y, O’Connor RS, Hwang W-T, et al. (2018). Determinants of response and resistance to CD19 chimeric antigen receptor (CAR) T cell therapy of chronic lymphocytic leukemia. Nat. Med 24, 563–571. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R19] 19.Krishna S, Lowery FJ, Copeland AR, Bahadiroglu E, Mukherjee R, Jia L, Anibal JT, Sachs A, Adebola SO, Gurusamy D, et al. (2020). Stem-like CD8 T cells mediate response of adoptive cell immunotherapy against human cancer. Science 370. 10.1126/science.abb9847. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R20] 20.Liu Y, Xun Z, Ma K, Liang S, Li X, Zhou S, Sun L, Liu Y, Du Y, Guo X, et al. (2023). Identification of a tumour immune barrier in the HCC microenvironment that determines the efficacy of immunotherapy. J. Hepatol 78, 770–782. [DOI] [PubMed] [Google Scholar]

[R21] 21.Zhang S, Yuan L, Danilova L, Mo G, Zhu Q, Deshpande A, Bell ATF, Elisseeff J, Popel AS, Anders RA, et al. (2023). Spatial transcriptomics analysis of neoadjuvant cabozantinib and nivolumab in advanced hepatocellular carcinoma identifies independent mechanisms of resistance and recurrence. Genome Med. 15, 72. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R22] 22.Achar SR, Bourassa FXP, Rademaker TJ, Lee A, Kondo T, Salazar-Cavazos E, Davies JS, Taylor N, François P, and Altan-Bonnet G (2022). Universal antigen encoding of T cell activation from high-dimensional cytokine dynamics. Science 376, 880–884. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R23] 23.Ruiz de Galarreta M, Bresnahan E, Molina-Sánchez P, Lindblad KE, Maier B, Sia D, Puigvehi M, Miguela V, Casanova-Acebes M, Dhainaut M, et al. (2019). β-Catenin Activation Promotes Immune Escape and Resistance to Anti-PD-1 Therapy in Hepatocellular Carcinoma. Cancer Discov. 9, 1124–1141. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R24] 24.Sharpless NE, and Depinho RA (2006). The mighty mouse: genetically engineered mouse models in cancer drug development. Nat. Rev. Drug Discov 5. 10.1038/nrd2110. [DOI] [PubMed] [Google Scholar]

[R25] 25.Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, et al. (2005). Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U. S. A 102. 10.1073/pnas.0506580102. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R26] 26.Koeberle A, Shindou H, Koeberle SC, Laufer SA, Shimizu T, and Werz O (2013). Arachidonoyl-phosphatidylcholine oscillates during the cell cycle and counteracts proliferation by suppressing Akt membrane binding. Proc. Natl. Acad. Sci. U. S. A 110. 10.1073/pnas.1216182110. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R27] 27.Munford RS, and Hunter JP (1992). Acyloxyacyl hydrolase, a leukocyte enzyme that deacylates bacterial lipopolysaccharides, has phospholipase, lysophospholipase, diacylglycerollipase, and acyltransferase activities in vitro. J. Biol. Chem 267. [PubMed] [Google Scholar]

[R28] 28.Vardhana SA, Hwee MA, Berisa M, Wells DK, Yost KE, King B, Smith M, Herrera PS, Chang HY, Satpathy AT, et al. (2020). Impaired mitochondrial oxidative phosphorylation limits the self-renewal of T cells exposed to persistent antigen. Nat. Immunol 21. 10.1038/s41590-020-0725-2. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R29] 29.Gide TN, Quek C, Menzies AM, Tasker AT, Shang P, Holst J, Madore J, Lim SY, Velickovic R, Wongchenko M, et al. (2019). Distinct Immune Cell Populations Define Response to Anti-PD-1 Monotherapy and Anti-PD-1/Anti-CTLA-4 Combined Therapy. Cancer cell 35. 10.1016/j.ccell.2019.01.003. [DOI] [PubMed] [Google Scholar]

[R30] 30.Carnevale J, Shifrut E, Kale N, Nyberg WA, Blaeschke F, Chen YY, Li Z, Bapat SP, Diolaiti ME, O’Leary P, et al. (2022). RASA2 ablation in T cells boosts antigen sensitivity and long-term function. Nature 609. 10.1038/s41586-022-05126-w. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R31] 31.Zou B, Goodwin M, Saleem D, Jiang W, Tang J, Chu Y, Munford RS, and Lu M (2021). A highly conserved host lipase deacylates oxidized phospholipids and ameliorates acute lung injury in mice. Elife 10. 10.7554/eLife.70938. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R32] 32.Lehninger AL, Nelson DL, and Cox MM (2005). Lehninger Principles of Biochemistry (Macmillan; ). [Google Scholar]

[R33] 33.Marai L, and Kuksis A (1969). Molecular species of lecithins from erythrocytes and plasma of man. J. Lipid Res 10. [PubMed] [Google Scholar]

[R34] 34.Ni Z, Sousa BC, Colombo S, Afonso CB, Melo T, Pitt AR, Spickett CM, Domingues P, Domingues MR, Fedorova M, et al. (2019). Evaluation of air oxidized PAPC: A multi laboratory study by LC-MS/MS. Free Radic Biol Med 144, 156–166. [DOI] [PubMed] [Google Scholar]

[R35] 35.Chen X, Song M, Zhang B, and Zhang Y (2016). Reactive Oxygen Species Regulate T Cell Immune Response in the Tumor Microenvironment. Oxidative medicine and cellular longevity 2016. 10.1155/2016/1580967. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R36] 36.Blüml S, Kirchberger S, Bochkov VN, Krönke G, Stuhlmeier K, Majdic O, Zlabinger GJ, Knapp W, Binder BR, Stöckl J, et al. (2005). Oxidized phospholipids negatively regulate dendritic cell maturation induced by TLRs and CD40. J Immunol 175, 501–508. [DOI] [PubMed] [Google Scholar]

[R37] 37.Ramakrishnan R, Tyurin VA, Veglia F, Condamine T, Amoscato A, Mohammadyani D, Johnson JJ, Zhang LM, Klein-Seetharaman J, Celis E, et al. (2014). Oxidized lipids block antigen cross-presentation by dendritic cells in cancer. J Immunol 192, 2920–2931. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R38] 38.Seok JK, Hong E-H, Yang G, Lee HE, Kim S-E, Liu K-H, Kang HC, Cho Y-Y, Lee HS, and Lee JY (2021). Oxidized Phospholipids in Tumor Microenvironment Stimulate Tumor Metastasis via Regulation of Autophagy. Cells 10. 10.3390/cells10030558. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R39] 39.Godzien J, Lopez-Lopez A, Sieminska J, Jablonowski K, Pietrowska K, Kisluk J, Mojsak M, Dzieciol-Anikiej Z, Barbas C, Reszec J, et al. (2023). Exploration of oxidized phosphocholine profile in non-small-cell lung cancer. Front Mol Biosci 10, 1279645. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R40] 40.Gorelik A, Illes K, and Nagar B (2018). Crystal structure of the mammalian lipopolysaccharide detoxifier. Proc. Natl. Acad. Sci. U. S. A 115, E896–E905. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R41] 41.Ghandi M, Huang FW, Jané-Valbuena J, Kryukov GV, Lo CC, McDonald ER 3rd, Barretina J, Gelfand ET, Bielski CM, Li H, et al. (2019). Next-generation characterization of the Cancer Cell Line Encyclopedia. Nature 569, 503–508. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R42] 42.Pan D, Kobayashi A, Jiang P, de Andrade LF, Tay RE, Luoma AM, Tsoucas D, Qiu X, Lim K, Rao P, et al. (2018). A major chromatin regulator determines resistance of tumor cells to T cell-mediated killing. Science 359. 10.1126/science.aao1710. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R43] 43.Bolotin DA, Poslavsky S, Davydov AN, Frenkel FE, Fanchi L, Zolotareva OI, Hemmers S, Putintseva EV, Obraztsova AS, Shugay M, et al. (2017). Antigen receptor repertoire profiling from RNA-seq data. Nature biotechnology 35. 10.1038/nbt.3979. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R44] 44.Liberzon A, Birger C, Thorvaldsdóttir H, Ghandi M, Mesirov JP, and Tamayo P (2015). The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell systems 1. 10.1016/j.cels.2015.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R45] 45.Pedersen BS, and Quinlan AR (2018). Mosdepth: quick coverage calculation for genomes and exomes. Bioinformatics 34, 867–868. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R46] 46.Wingett SW, and Andrews S (2018). FastQ Screen: A tool for multi-genome mapping and quality control. F1000Res 7, 1338. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R47] 47.Liu D, Schilling B, Liu D, Sucker A, Livingstone E, Jerby-Arnon L, Zimmer L, Gutzmer R, Satzger I, Loquai C, et al. (2019). Integrative molecular and clinical modeling of clinical outcomes to PD1 blockade in patients with metastatic melanoma. Nature medicine 25. 10.1038/s41591-019-0654-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R48] 48.Roper N, Velez MJ, Chiappori A, Kim YS, Wei JS, Sindiri S, Takahashi N, Mulford D, Kumar S, Ylaya K, et al. (2021). Notch signaling and efficacy of PD-1/PD-L1 blockade in relapsed small cell lung cancer. Nat Commun 12, 3880. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R49] 49.Yoshihara K, Shahmoradgoli M, Martínez E, Vegesna R, Kim H, Torres-Garcia W, Treviño V, Shen H, Laird PW, Levine DA, et al. (2013). Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun 4, 2612. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R50] 50.Raine KM, Van Loo P, Wedge DC, Jones D, Menzies A, Butler AP, Teague JW, Tarpey P, Nik-Zainal S, and Campbell PJ (2016). ascatNgs: Identifying Somatically Acquired Copy-Number Alterations from Whole-Genome Sequencing Data. Curr Protoc Bioinformatics 56, 15.9.1–15.9.17. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R51] 51.Padrón LJ, Maurer DM, O’Hara MH, O’Reilly EM, Wolff RA, Wainberg ZA, Ko AH, Fisher G, Rahma O, Lyman JP, et al. (2022). Sotigalimab and/or nivolumab with chemotherapy in first-line metastatic pancreatic cancer: clinical and immunologic analyses from the randomized phase 2 PRINCE trial. Nat. Med 28, 1167–1177. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R52] 52.Jiang P, Gu S, Pan D, Fu J, Sahu A, Hu X, Li Z, Traugh N, Bu X, Li B, et al. (2018). Signatures of T cell dysfunction and exclusion predict cancer immunotherapy response. Nat. Med 24. 10.1038/s41591-018-0136-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R53] 53.Jiang P, Zhang Y, Ru B, Yang Y, Vu T, Paul R, Mirza A, Altan-Bonnet G, Liu L, Ruppin E, et al. (2021). Systematic investigation of cytokine signaling activity at the tissue and single-cell levels. Nat. Methods 18. 10.1038/s41592-021-01274-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R54] 54.Van Rossum G, and Drake FL (2009). Python 3 Reference Manual (CreateSpace; ). [Google Scholar]

[R55] 55.Virtanen P, Gommers R, Oliphant TE, Haberland M, Reddy T, Cournapeau D, Burovski E, Peterson P, Weckesser W, Bright J, et al. (2020). SciPy 1.0: fundamental algorithms for scientific computing in Python. Nat Methods 17, 261–272. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R56] 56.R Core Team (2021). R: A Language and Environment for Statistical Computing. Preprint at R Foundation for Statistical Computing. [Google Scholar]

[R57] 57.Chen S, Zhou Y, Chen Y, and Gu J (2018). fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 34, i884–i890. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R58] 58.Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, Marth G, Abecasis G, Durbin R, and 1000 Genome Project Data Processing Subgroup (2009). The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R59] 59.Quinlan AR, and Hall IM (2010). BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841–842. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R60] 60.Patro R, Duggal G, Love MI, Irizarry RA, and Kingsford C (2017). Salmon provides fast and bias-aware quantification of transcript expression. Nat. Methods 14, 417–419. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R61] 61.Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, and Gingeras TR (2013). STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29. 10.1093/bioinformatics/bts635. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R62] 62.Love MI, Huber W, and Anders S (2014). Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R63] 63.Li B, and Dewey CN (2011). RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics 12. 10.1186/1471-2105-12-323. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R64] 64.Bolotin DA, Poslavsky S, Mitrophanov I, Shugay M, Mamedov IZ, Putintseva EV, and Chudakov DM (2015). MiXCR: software for comprehensive adaptive immunity profiling. Nat. Methods 12. 10.1038/nmeth.3364. [DOI] [PubMed] [Google Scholar]

[R65] 65.Vasimuddin M, Misra S, Li H, and Aluru S (2019). Efficient Architecture-Aware Acceleration of BWA-MEM for Multicore Systems. In 2019 IEEE International Parallel and Distributed Processing Symposium (IPDPS), pp. 314–324. [Google Scholar]

[R66] 66.Koboldt DC, Chen K, Wylie T, Larson DE, McLellan, Mardis ER, Weinstock GM, Wilson RK, and Ding L (2009). VarScan: variant detection in massively parallel sequencing of individual and pooled samples. Bioinformatics 25. 10.1093/bioinformatics/btp373. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R67] 67.McLaren W, Gil L, Hunt SE, Riat HS, Ritchie GR, Thormann A, Flicek P, and Cunningham F (2016). The Ensembl Variant Effect Predictor. Genome Biol. 17. 10.1186/s13059-016-0974-4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R68] 68.Kandoth C, Gao J, qwangmsk, Mattioni M, Struck A, Boursin Y, Penson A, and Chavan S (2018). mskcc/vcf2maf: vcf2maf v1.6.16 (Zenodo) 10.5281/ZENODO.593251. [DOI] [Google Scholar]

[R69] 69.McKenna A, Hanna M, Banks E, Sivachenko A, Cibulskis K, Kernytsky A, Garimella K, Altshuler D, Gabriel S, Daly M, et al. (2010). The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 20, 1297–1303. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R70] 70.Kim S, Scheffler K, Halpern AL, Bekritsky MA, Noh E, Källberg M, Chen X, Kim Y, Beyter D, Krusche P, et al. (2018). Strelka2: fast and accurate calling of germline and somatic variants. Nat. Methods 15, 591–594. [DOI] [PubMed] [Google Scholar]

[R71] 71.Chakravarty D, Gao J, Phillips SM, Kundra R, Zhang H, Wang J, Rudolph JE, Yaeger R, Soumerai T, Nissan MH, et al. (2017). OncoKB: A Precision Oncology Knowledge Base. JCO Precis Oncol 2017. 10.1200/PO.17.00011. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R72] 72.Martin M (2011). Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet. journal 17, 10–12. [Google Scholar]

[R73] 73.Kopylova E, Noé L, and Touzet H (2012). SortMeRNA: fast and accurate filtering of ribosomal RNAs in metatranscriptomic data. Bioinformatics 28, 3211–3217. [DOI] [PubMed] [Google Scholar]

[R74] 74.Ru B, Huang J, Zhang Y, Aldape K, and Jiang P (2023). Estimation of cell lineages in tumors from spatial transcriptomics data. Nat. Commun 14, 568. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R75] 75.Hao Y, Stuart T, Kowalski MH, Choudhary S, Hoffman P, Hartman A, Srivastava A, Molla G, Madad S, Fernandez-Granda C, et al. (2024). Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat. Biotechnol 42, 293–304. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R76] 76.Wolock SL, Lopez R, and Klein AM (2019). Scrublet: Computational Identification of Cell Doublets in Single-Cell Transcriptomic Data. Cell systems 8. 10.1016/j.cels.2018.11.005. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R77] 77.Gao Y, Patro R, and Jiang P (2024). Collapsible tree: interactive web app to present collapsible hierarchies. Bioinformatics 40. 10.1093/bioinformatics/btae645. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R78] 78.Mailman, Feolo M, Jin Y, Kimura M, Tryka K, Bagoutdinov R, Hao L, Kiang A, Paschall J, Phan L, et al. (2007). The NCBI dbGaP database of genotypes and phenotypes. Nat. Genet 39. 10.1038/ng1007-1181. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R79] 79.Freeberg MA, Fromont LA, D’Altri T, Romero AF, Ciges JI, Jene A, Kerry G, Moldes M, Ariosa R, Bahena S, et al. (2022). The European Genome-phenome Archive in 2021. Nucleic Acids Res. 50. 10.1093/nar/gkab1059. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R80] 80.Campbell KM, Amouzgar M, Pfeiffer SM, Howes TR, Medina E, Travers M, Steiner G, Weber JS, Wolchok JD, Larkin J, et al. (2023). Prior anti-CTLA-4 therapy impacts molecular characteristics associated with anti-PD-1 response in advanced melanoma. Cancer Cell 41, 791–806.e4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R81] 81.Hugo W, Zaretsky JM, Sun L, Song C, Moreno BH, Hu-Lieskovan S, Berent-Maoz B, Pang J, Chmielowski B, Cherry G, et al. (2016). Genomic and Transcriptomic Features of Response to Anti-PD-1 Therapy in Metastatic Melanoma. Cell 165. 10.1016/j.cell.2016.02.065. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R82] 82.Pusztai L, Yau C, Wolf DM, Han HS, Du L, Wallace AM, String-Reasor E, Boughey JC, Chien AJ, Elias AD, et al. (2021). Durvalumab with olaparib and paclitaxel for high-risk HER2-negative stage II/III breast cancer: Results from the adaptively randomized I-SPY2 trial. Cancer cell 39. 10.1016/j.ccell.2021.05.009. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R83] 83.Wolf DM, Yau C, Wulfkuhle J, Brown-Swigart L, Gallagher RI, Lee PRE, Zhu Z, Magbanua MJ, Sayaman R, O’Grady N, et al. (2022). Redefining breast cancer subtypes to guide treatment prioritization and maximize response: Predictive biomarkers across 10 cancer therapies. Cancer cell 40. 10.1016/j.ccell.2022.05.005. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R84] 84.Anagnostou V, Niknafs N, Marrone K, Bruhm DC, White JR, Naidoo J, Hummelink K, Monkhorst K, Lalezari F, Lanis M, et al. (2020). Multimodal genomic features predict outcome of immune checkpoint blockade in non-small-cell lung cancer. Nat Cancer 1, 99–111. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R85] 85.Budhu A, Pehrsson EC, He A, Goyal L, Kelley RK, Dang H, Xie C, Monge C, Tandon M, Ma L, et al. (2023). Tumor biology and immune infiltration define primary liver cancer subsets linked to overall survival after immunotherapy. Cell Rep Med 4, 101052. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R86] 86.Jacobsen SB, Tfelt-Hansen J, Smerup MH, Andersen JD, and Morling N (2023). Comparison of whole transcriptome sequencing of fresh, frozen, and formalin-fixed, paraffin-embedded cardiac tissue. PLoS One 18, e0283159. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R87] 87.Johnson WE, Li C, and Rabinovic A (2007). Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics (Oxford, England) 8. 10.1093/biostatistics/kxj037. [DOI] [PubMed] [Google Scholar]

[R88] 88.Nygaard V, Rødland EA, and Hovig E (2016). Methods that remove batch effects while retaining group differences may lead to exaggerated confidence in downstream analyses. Biostatistics 17, 29–39. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R89] 89.Sy CY, Lien SC, Wang BX, Clouthier DL, Hanna Y, Cirlan I, Zhu K, Bruce JP, El Ghamrasni S, Iafolla MAJ, et al. (2021). Pan-cancer analysis of longitudinal metastatic tumors reveals genomic alterations and immune landscape dynamics associated with pembrolizumab sensitivity. Nat. Commun 12. 10.1038/s41467-021-25432-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R90] 90.Li G, Choi JE, Kryczek I, Sun Y, Liao P, Li S, Wei S, Grove S, Vatan L, Nelson R, et al. (2023). Intersection of immune and oncometabolic pathways drives cancer hyperprogression during immunotherapy. Cancer Cell 41, 304–322.e7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R91] 91.Fehrenbacher L, Spira A, Ballinger M, Kowanetz M, Vansteenkiste J, Mazieres J, Park K, Smith D, Artal-Cortes A, Lewanski C, et al. (2016). Atezolizumab versus docetaxel for patients with previously treated non-small-cell lung cancer (POPLAR): a multicentre, open-label, phase 2 randomised controlled trial. Lancet (London, England) 387. 10.1016/S0140-6736(16)00587-0. [DOI] [PubMed] [Google Scholar]

[R92] 92.Rooney MS, Shukla SA, Wu CJ, Getz G, and Hacohen N (2015). Molecular and genetic properties of tumors associated with local immune cytolytic activity. Cell 160. 10.1016/j.cell.2014.12.033. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R93] 93.Auslander N, Zhang G, Lee JS, Frederick DT, Miao B, Moll T, Tian T, Wei Z, Madan S, Sullivan RJ, et al. (2018). Robust prediction of response to immune checkpoint blockade therapy in metastatic melanoma. Nat Med 24, 1545–1549. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R94] 94.Litchfield K, Reading JL, Puttick C, Thakkar K, Abbosh C, Bentham R, Watkins TBK, Rosenthal R, Biswas D, Rowan A, et al. (2021). Meta-analysis of tumor- and T cell-intrinsic mechanisms of sensitization to checkpoint inhibition. Cell 184. 10.1016/j.cell.2021.01.002. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R95] 95.Spranger S, Dai D, Horton B, and Gajewski TF (2017). Tumor-Residing Batf3 Dendritic Cells Are Required for Effector T Cell Trafficking and Adoptive T Cell Therapy. Cancer cell 31. 10.1016/j.ccell.2017.04.003. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R96] 96.Davoli T, Uno H, Wooten EC, and Elledge SJ (2017). Tumor aneuploidy correlates with markers of immune evasion and with reduced response to immunotherapy. Science 355. 10.1126/science.aaf8399. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R97] 97.Mariathasan S, Turley SJ, Nickles D, Castiglioni A, Yuen K, Wang Y, Kadel EE, Koeppen H, Astarita JL, Cubas R, et al. (2018). TGFβ attenuates tumour response to PD-L1 blockade by contributing to exclusion of T cells. Nature 554. 10.1038/nature25501. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R98] 98.Veis DJ, Sentman CL, Bach EA, and Korsmeyer SJ (1993). Expression of the Bcl-2 protein in murine and human thymocytes and in peripheral T lymphocytes. J Immunol 151, 2546–2554. [PubMed] [Google Scholar]

[R99] 99.Mélique S, Vadel A, Rouquié N, Yang C, Bories C, Cotineau C, Saoudi A, Fazilleau N, and Lesourne R (2024). THEMIS promotes T cell development and maintenance by rising the signaling threshold of the inhibitory receptor BTLA. Proc Natl Acad Sci U S A 121, e2318773121. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R100] 100.Nolte M, and Margadant C (2020). Controlling Immunity and Inflammation through Integrin-Dependent Regulation of TGF-β. Trends in cell biology 30. 10.1016/j.tcb.2019.10.002. [DOI] [PubMed] [Google Scholar]

[R101] 101.Dodagatta-Marri E, Ma HY, Liang B, Li J, Meyer DS, Chen SY, Sun KH, Ren X, Zivak B, Rosenblum, et al. (2021). Integrin αvβ8 on T cells suppresses anti-tumor immunity in multiple models and is a promising target for tumor immunotherapy. Cell reports 36. 10.1016/j.celrep.2021.109309. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R102] 102.Jaffe DB, Shahi P, Adams BA, Chrisman AM, Finnegan PM, Raman N, Royall AE, Tsai F, Vollbrecht T, Reyes DS, et al. (2022). enclone: precision clonotyping and analysis of immune receptors. bioRxiv, 2022.04.21.489084. 10.1101/2022.04.21.489084. [DOI] [Google Scholar]

[R103] 103.Tumeh PC, Yearley JH, Shintaku IP, Taylor EJ, Robert L, Chmielowski B, Spasic M, Henry G, Ciobanu V, West AN, et al. (2014). PD-1 blockade induces responses by inhibiting adaptive immune resistance. Nature 515. 10.1038/nature13954. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R104] 104.Sornasse T, Flamand V, De Becker G, Bazin H, Tielemans F, Thielemans K, Urbain J, Leo O, and Moser M (1992). Antigen-pulsed dendritic cells can efficiently induce an antibody response in vivo. The Journal of experimental medicine 175. 10.1084/jem.175.1.15. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R105] 105.Okada N, Tsujino M, Hagiwara Y, Tada A, Tamura Y, Mori K, Saito T, Nakagawa S, Mayumi T, Fujita T, et al. (2001). Administration route-dependent vaccine efficiency of murine dendritic cells pulsed with antigens. Br J Cancer 84, 1564–1570. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R106] 106.Firth D (March 1993). Bias reduction of maximum likelihood estimates. Biometrika 80, 27–38. [Google Scholar]

[R107] 107.Uhlén M, Fagerberg L, Hallström BM, Lindskog C, Oksvold P, Mardinoglu A, Sivertsson Å, Kampf C, Sjöstedt E, Asplund A, et al. (2015). Proteomics. Tissue-based map of the human proteome. Science 347. 10.1126/science.1260419. [DOI] [PubMed] [Google Scholar]

[R108] 108.Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, Dolinski K, Dwight SS, Eppig JT, et al. (2000). Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25, 25–29. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R109] 109.Efremova M, Vento-Tormo M, Teichmann SA, and Vento-Tormo R (2020). CellPhoneDB: inferring cell-cell communication from combined expression of multi-subunit ligand-receptor complexes. Nat Protoc 15, 1484–1506. [DOI] [PubMed] [Google Scholar]

[R110] 110.Benjamini Y, and Hochberg Y (1995). Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. J. R. Stat. Soc. Series B Stat. Methodol 57, 289–300. [Google Scholar]

[R111] 111.Li Y, Dou Y, Da Veiga Leprevost F, Geffen Y, Calinawan AP, Aguet F, Akiyama Y, Anand S, Birger C, Cao S, et al. (2023). Proteogenomic data and resources for pan-cancer analysis. Cancer Cell 41, 1397–1406. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R112] 112.Martínez-Carpio PA, Mur C, Fernández-Montolí ME, Ramon JM, Rosel P, and Navarro MA (1999). Secretion and dual regulation between epidermal growth factor and transforming growth factor-beta1 in MDA-MB-231 cell line in 42-hour-long cultures. Cancer letters 147. 10.1016/s0304-3835(99)00261-x. [DOI] [PubMed] [Google Scholar]

[R113] 113.Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, Jacobsen A, Byrne CJ, Heuer ML, Larsson E, et al. (2012). The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer discovery 2. 10.1158/2159-8290.CD-12-0095. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R114] 114.Zaykin DV, Zhivotovsky LA, Westfall PH, and Weir BS (2002). Truncated product method for combining P-values. Genetic epidemiology 22. 10.1002/gepi.0042. [DOI] [PubMed] [Google Scholar]

[R115] 115.Fisher RA (1970). Statistical methods for research workers. In Breakthroughs in statistics: Methodology and distribution (Springer; ), pp. 66–70. [Google Scholar]

PERMALINK

Cancer immunology data engine reveals secreted AOAH as a potential immunotherapy

Lanqi Gong

Jie Luo

Emily Yang

Beibei Ru

Ziyang Qi

Yuma Yang

Anshu Rani

Abhilasha Purohit

Yu Zhang

Grace Guan

Rohit Paul

Trang Vu

Zuojia Chen

Renyue Ji

Chi-Ping Day

Chuan Wu

Glenn Merlino

David Fitzgerald

Grégoire Altan-Bonnet

Kenneth Aldape

Jiansheng Wu

Xinyuan Guan

Peng Jiang

Summary

Graphical Abstract

In brief

Introduction

Results

A data-integrative framework for immuno-oncology research

Figure 1. Cancer Immunology Data Engine (CIDE).

Risk scores predict anti- and pro-tumor effects of secreted proteins

Figure 2. Gene risk scores from immunotherapy data predict secreted protein cancer roles.

Predicted secreted regulators of immunotherapy outcomes

Figure 3. Previously unidentified secreted regulators of cancer immunotherapy response.

In vivo validation of AOAH, COLQ, CR1L, and ADAMTS7

AOAH potentiates immunotherapy responses in multiple tumor models

Figure 4. AOAH potentiates in vivo immunotherapy efficacies.

Molecular programs induced by Aoah in vivo

Figure 5. Tumor molecular landscape induced by Aoah overexpression (OE).

Single-cell analysis of immune cell characteristics induced by Aoah

Molecular programs induced by CR1L, COLQ, and ADAMTS7

AOAH enhances TCR activation and antigen recognition

Figure 6. AOAH enhances CD8+ T-cell activation.

AOAH depletes arachidonoyl phosphatidylcholines

Figure 7. AOAH depletes immunosuppressive arachidonoyl phosphatidylcholines.

AOAH rescues TCR inhibition by arachidonoyl phosphatidylcholines

AOAH protects dendritic cells from oxidized PC(16:0–20:4) inhibition

Serine 263 is a catalytic site

Discussion

Limitations of the Study

Resource Availability

Lead Contact

Materials Availability

Data and Code Availability

STAR Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell lines and cell culture

Human primary CD8+ T cell isolation

Mouse primary CD8+ T cell isolation

Mouse and human CD8+ primary T cell culture

Animal models

Tumor in vivo model selections

Human immunotherapy clinical datasets

METHOD DETAILS

Resolve clinical data inconsistency

Omics data processing

Quality controls

Batch effects and clinical covariates

Threshold selection in survival plots

Interactive CIDE server

Immunotherapy response biomarker analysis

Spatial transcriptomics data analysis

Blinded literature annotation

Single-cell RNA-seq data analysis

Cell lineage classification

Differential expression profiles

V(D)J receptor clonotype analysis

Lipid class enrichment analysis