Major peripheral immunobiomarkers in oncology predominantly engage NFκB or IFN/ISG responses

Cytokines induce an array of different downstream immune pathways, for example, NFκB response, IFN-induced ISG response, AP1 signaling, SMAD signaling, TRADD/FADD signaling, FOXO signaling or p38/ERK signaling.21–23 However, to design a precise serum-screening functional assay (online supplemental figure S1), we needed to focus on immune pathways dominantly engaged by majority of cytokines/immune factors in oncological context. To enable such consolidation exercise in an unbiased manner, we pursued a two-step bioinformatic process. In step 1, we delineated the most frequently screened peripheral (oncological) immunobiomarkers using an automated MELODI literature-mining algorithm that extracts potential gene/protein identities between any two (prespecified) biomedical concepts.24 Herein, MELODI was executed to identify the most frequently occurring genes/proteins extracted from PubMed literature (1950–2017) simultaneously relevant for two concepts, that is, ‘serum of patients with cancer’ and ‘immunological factors’ (defined using precise string of keywords; see online supplemental methods for details). This exercise identified 54 immunological, inflammatory, and/or haematopoietic factors most frequently assessed as peripheral oncological biomarkers (online supplemental figure S2A). In step 2, we used these 54 major peripheral immunobiomarkers to identify the dominant downstream immune pathways and associated transcriptional programs. A computational analysis enriching for transcription factor (TF) targets via top three REACTOME biological pathways linked to the aforementioned 54 immunobiomarkers (co-delineated along with top hallmark gene sets to provide context) found that inflammation or effector immunity operating via either an NFκB response (NFκB/REL) or an IFN/ISG response (ISRE-binding TFs, IRF and IRF1/8) were the most dominantly enriched downstream pathways/TF classes (online supplemental figure S2B). Thus, the major oncological peripheral immunobiomarkers dominantly engaged the NFκB or IFN/ISG responses.

Tumor and peripheral immune cells co-enrich differential NFκB or ISG response patterns

Tumors can, directly or indirectly, influence the differential enrichment of some peripheral cytokines.5 25 Hence, we interrogated: (1) does a co-enrichment exist for NFκB or ISG responses in immune cells across both tumor and periphery; and (2) which immune cells typically capture this co-enrichment (especially in periphery)? The former was vital to prove the suitability of a data-driven framework reliant on the multiomics tumor datasets of patients with cancer, while the latter was central to delineate most suitable immune cell type. Both steps were necessary to reveal our assay’s target cancer type(s) and immune cell type.

To enable multiomics analyses, we used a validated genetic NFκB signaling signature26 and created a consensus genetic ISG response signature. The ISG response signature was a consensus between three validated genetic signatures induced downstream of IFN-α,27 IFNβ28 or IFNγ.29 We prioritized a consensus signature because a computational analysis (driven by the Interferome database of curated experimental datasets30) investigating the overlaps of downstream genes/ISGs regulated by all three IFN cytokine-types (I, II or III) showed that there was a considerably high sharing of ISGs induced downstream of all three IFN cytokine types, such that ≥85% ISGs (irrespective of which signature was used) were either overlapping between all three IFN cytokine types or between type I/II IFNs (online supplemental figure S3). Thus, to avoid unnecessary redundancy,31 we used a consensus ISG response signature.

Next, to assess the systemic inflammatory circuit on the level of NFκB or IFN/ISG responses, we analyzed an existing scRNAseq map that profiled tumor-infiltrating and (matched) blood-derived immune cells (procured from patients with renal cell or large cell neuroendocrine carcinoma),32 with the aforementioned NFκB or ISG response signatures (figure 1A,B). Across both blood and tumor immune cells, NFκB response signature was more strongly enriched than ISG response signature, such that myeloid cells particularly co-enriched these contrasting patterns in both tumors and blood (figure 1A,B and online supplemental figure S4A,B). Overall, NFκB response signature had the highest enrichment in blood myeloid cells (online supplemental figure S4A), while in tumors, only B cells, CD4⁺ T cells and Tregs had higher NFκB response signature than myeloid cells (online supplemental figure S4B). IFN/ISG response signature was relatively better enriched in T cells (online supplemental figure S4A,B). Nevertheless, for myeloid cells, it is necessary to consider that NFκB (subset 3) and ISG (subset 4) response signatures were enriched in different subsets of blood myeloid cells (figure 1A,B) such that both in blood (online supplemental figure S4C) or tumor (online supplemental figure S4D) myeloid cells, there was negative/null correlation between NFκB and ISG response signature enrichment. Overall, this implied a differential co-enrichment of NFκB and IFN/ISG response circuits between the tumor and the periphery, particularly represented by myeloid cells in both compartments.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Interrogation of the tumour derived bulk-RNAseq or tumour/blood single-cell (sc)-RNAseq data of patients with cancer. (A,B) The t-distributed stochastic neighbor embedding (tSNE) visualizations of indicated immune cell types, from scRNAseq data from patients with renal cell and large cell neuroendocrine carcinoma (n=4 patients in total) (derived from GSE139555) isolated from tumor tissue (A) or peripheral blood (B). These immune cells were further colored for NFκB and IFN/ISG response gene-signature levels. Herein, the arrows highlight main immune populations with overlaps between tumor and blood for these signatures, and the circles indicate different myeloid cell subsets. (C) PCA analyses of median expression for each gene across cancer types (from online supplemental figure S5A) and median HR values (from online supplemental figure S5B). Venn diagram represents the portion of NFκB and ISGs in each cluster. (D,E) Visualization of the hazard ratio (HR) ±95% confidence interval (CI) for the impact of expression of NFκB signaling gene signature (D) or ISG signaling gene signature (E), for the indicated TCGA cancer datasets wherein the signature expression cut-off for binary (high vs low expression) patient stratification was based on best-performing threshold principle for OS of indicated TCGA patients with cancer (LIHC, n=371; PAAD, n=177; LUSC, n=501; LUAD, n=513; HNSC, n=500; CESC, n=304; BLCA, n=405; UCEC, n=543; OV, n=374; SARC, n=259; BRCA; n=1090; KIRC; n=530) (Mantel-Cox test, *p<0.05). (F–H) tSNE visualizations of indicated immune cell-types, from scRNAseq data from patients with OV (n=5 patients) (derived from GSE146026) isolated from tumor tissue. These immune cells were furthered colored for NFκB (G) and ISG signature expression levels (H). BLCA, bladder cancer; CESC, cervical cancer; HNSC, head and neck cancer; IFN, interferon; ISG, interferon-stimulated gene; KIRC, renal cell cancer; LIHC, liver cancer; LUSC, lung cancer; LUAD, lung adenocarcinoma; NFκB, nuclear factor kappa-light-chain enhancer of activated B cells; OS, overall survival; OV, ovarian cancer; PAAD, pancreatic cancer; SARC, sarcoma; BRCA, breast cancer;TCGA, The Cancer Genome Atlas; UCEC, endometrial cancer.

Tumoral immunodynamics of NFκB or ISG responses predict survival of patients with cancer

Based on the aforementioned observations, we used The Cancer Genome Atlas (TCGA) tumor datasets for creating a data-driven framework (essential for our assay’s designing) driven by the qualitative and prognostic behavior of NFκB/ISG response patterns. Herein, we selected 12 TCGA datasets (spanning >5000 patients with cancer; see online supplemental methods for selection strategy) with diverse solid tumors typically showing either immunotherapy responsiveness (i.e., lung cancer (LUAD/LUSC), head and neck cancer (HNSC), bladder cancer, renal cell cancer, and liver cancer (LIHC)), or immunotherapy resistance (OV, endometrial cancer, sarcoma, breast cancer, pancreatic cancer, and cervical cancer (CESC)).33 34 Interestingly, nearly half of the 54 major oncological (peripheral) immunobiomarkers were not sufficiently expressed in these 12 tumor types (online supplemental figure S5A). Moreover, the 25 highly expressed immunobiomarkers did not create distinct clusters that could clearly differentiate immunotherapy responsive from non-responsive cancer types (online supplemental figure S5A). Also, these immunobiomarkers exhibited a variable cancer type-dependent prognostic impact on overall survival (OS) (online supplemental figure S5B).

Next, we attempted to understand whether the more simplified ‘dynamics’ of the binary NFκB/ISG responses may better summarize a general inflammatory status than the aforementioned heterogenous patterns. To address this, we first derived the median expression and median hazard ratios (HRs) for each of the 25 highly expressed immunobiomarkers across all 12 cancer types and applied dimensionality reduction (i.e., Principal component analysis or PCA) (online supplemental figure S5 and figure 1C). This consolidated two distinct clusters, that is, a cluster of immunobiomarkers with negative prognostic impact and those with positive prognostic impact (figure 1C). Next, we annotated the aforementioned genes as either NFκB target genes (NTGs)35 36 or ISGs based on available literature.30 This revealed that the negative prognostic cluster was associated with a disbalance in these modalities, favoring NFκB signaling (NTGs>ISGs); whereas positive prognostic cluster had a relatively better balance (NTGs≈ISGs) (figure 1C). Thus, a disbalance in tumoral immunodynamic signaling that favors NFκB responses predicted shortened patient survival, whereas more balanced signaling between ISG/NFκB responses predicted prolonged patient survival.

Ovarian tumor-linked myeloid milieu enriches for disbalanced NFκB/ISG responses

Next, we estimated the prognostic impact of tumoral NFκB/ISG response signatures on patient OS in the aforementioned 12 TCGA cancer types to delineate which cancer types might best capture the contrasting interplay between NFκB (negative prognostic) and IFN/ISG (positive prognostic) responses (figure 1D,E). Herein, predominantly CESC or OV exhibited (statistically significant) contradictory prognostic impacts for NFκB and ISG response signatures (figure 1D,E). Similar trends were also observed for LIHC/HNSC. Herein for further investigation, we selected OV since it is associated with several clinically relevant diagnostic and prognostic peripheral biomarkers like CA125, HE4, or Ova1/Overa.37

Next, we explored whether the differential NFκB/ISG response patterns observed in the aforementioned scRNAseq analyses were also applicable to OV, to ensure consistency in our data-driven framework. Henceforth, we pursued analyses of an existing scRNAseq dataset of a patient with OV (see online supplemental methods for selection strategy). Within OV tumor-derived cancer, immune and stromal cells (figure 1F),38 analyses for NFκB (figure 1G) or ISG (figure 1H) response signatures showed that myeloid cells indeed enriched for higher NFκB response signature than the ISG response signature, with the NFκB response signature being more highly enriched in myeloid cells than other immune cells (online supplemental figure S6A). Of note, while the ISG response signature was largely expressed by myeloid, endothelial or fibroblast cells, the primary source of IFN-cytokine gene (IFNG) was T/NK cells (online supplemental figure S6B). Moreover, within the myeloid cells, there was negative/null correlation between NFκB versus ISG response signatures (online supplemental figure S6C). In line with this, we observed that indeed relatively non-associated myeloid cell subsets enriched for NFκB versus ISG response signatures (online supplemental figure S7A). Interestingly, a differential REACTOME pathway enrichment analysis between NFκB response signature^HIGH versus ISG response signature^HIGH myeloid cells (online supplemental figure S7B) showed that NFκB response^HIGHmyeloid cells enriched for anti-inflammatory or immuno-regulatory pathways (e.g., prostaglandin/IL-10 signaling, and RUNX1/FOXO/CD163 pathways), whereas ISG response^HIGH myeloid cells enriched for pro-inflammatory/immunogenic pathways (e.g., antiviral/IFN signaling and antigen presentation). Altogether, this showed that OV myeloid compartment enriches for disbalanced NFκB and ISG response dynamics, wherein the former has anti-inflammatory/immunoregulatory associations, while the latter has pro-inflammatory/immunogenic associations.

sFIS assay for testing the peripheral NFκB or IFN/ISG responses of patients with cancer

The aforementioned data-driven framework informed us that the dominant peripheral immunodynamics in OV might be best captured by human myeloid cell-associated NFκB or IFN/ISG responses. However, transcriptomic or proteomic analyses of these signaling pathways’ key components within circulating human myeloid cells would not be very throughput and may not capture the overall dynamism of these signaling modalities. To address this, we decided to investigate whether co-culturing human myeloid cells in the presence of serum samples from patients with OV and then assessing the myeloid cell’s NFκB or IFN/ISG signaling as cellular assay output may capture peripheral immunodynamics in OV, in a throughput and dynamic fashion. Hence, we conceptualized an innovative in sitro sFIS assay for in vitro assessment of patient sera-induced si-NFκB and si-IFN/ISG response in human myeloid THP1 cells (figure 2A). This involved a 96-well plate-based throughput screening method using THP1 myeloid cell line stably expressing two inducible reporter constructs encoding for (1) a LUCIA luciferase gene (coding for a secreted form of coelenterazine-using luciferase enzyme) under the direct control of an ISG54 minimal promoter linked to five copies of ISRE sequences; and (2) a SEAP reporter gene (coding for the SEAP enzyme, a placental alkaline phosphatase) under the control of an IFN-β minimal promoter linked to three copies of c-REL binding site and five copies of the NFκB consensus transcriptional response element (figure 2A). Once these THP1 reporter cells sense factors/cytokines that engage the NFκB and/or IFN/ISG responses, they should secrete extracellular SEAP/LUCIA that, in presence of proper substrates, create colorimetric (SEAP) or bioluminescent (LUCIA) readouts (figure 2A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Standardization of sFIS assay. (A) Overview of the sFIS assay. THP1-dual reporter cells containing a reporter for NFκB and IFN/ISG response (see the Methods section) were stimulated with human serum of a patient with cancer, healthy donor serum (baseline) or LPS (positive control) for 24 h. Subsequently, THP1 media were checked for LUCIA and SEAP activity. (B,C) Bar graph of THP1-dual reporter cells, exposed to indicated concentrations of different agonists, reporting NFκB (B) or IFN/ISG (C) responses at 48 h post-treatment (n=3, one-way analysis of variance with Fisher’s Least Significant Difference (LSD) test; *p<0.05). (D–G) Bar graph of THP1 dual reporter cells exposed to indicated concentrations of different cytokines (D,E) or soluble immune checkpoints (F,G) reporting NFκB (D,F) or IFN/ISG (E,G) responses at 48 h post-treatment (min–max normalized, n=4, Kruskal-Wallis test; *p<0.05). cGAMP, cyclic guanosine monophosphate–adenosine monophosphate; IFN, interferon; IL, interleukin; IRF, interferon regulatory factor; ISG, interferon-stimulated gene; ISRE, interferon-stimulated response element; LPS, lipopolysaccharide; NFκB, nuclear factor kappa-light-chain enhancer of activated B cells; PD-1, programmed cell death protein 1; PDL-1, programmed cell death-ligand 1; PRR, pattern recognition receptor; TIM3, T-cell immunoglobulin and mucin domain 3; TNF, tumor necrosis factor; TLR, toll-like receptor; TRAIL, TNF-related apoptosis-inducing ligand; SEAP, secreted embryonic alkaline phosphatase; sFIS, serumfunctional immunodynamic status; -STAT, signal transducer and activator of transcription.

Accordingly, THP1 reporter cells exposed to pattern-recognition receptor agonists like LPS (a TLR4 agonist), 5′ppp-dsRNA bound to a transfection reagent LysoVec (a RIG-I agonist), or 2′3′-cGAMP (a STING agonist) differentially stimulated the NFκB (figure 2B) and/or IFN/ISG (figure 2C) responses. Similarly, treatment with diverse human recombinant cytokines induced either NFκB response (like TNF or IL1β) (figure 2D) or IFN/ISG response (like IFN-β, IFN-α, or IFN-γ) (figure 2E). One of the emerging categories of serum immunobiomarkers in immuno-oncology is the soluble immune-checkpoint (e.g., TIM3 or PD-1/PD-L1).39 Hence, we exposed these reporter cells to human recombinant PD-1, TIM3, or PD-L1 proteins and observed that whereas they were in general incapable of substantially inducing either of the signaling programs, yet recombinant TIM3/PD-1 induced threshold levels of NFκB response (figure 2F,G). Thus, we hypothesized that a THP1 NFκB and IFN/ISG reporter myeloid cell-driven sFIS assay may sufficiently capture the downstream signaling impact of peripheral cytokines/immune factors.

Peripheral immunobiomarkers hint toward disbalanced NFκB and IFN/ISG responses in patients with OV

To pursue a proof-of-concept for our sFIS assay approach, we aimed for a 'prospective retrospective' biomarker validation strategy,40 41 starting with a discovery cohort composed of clinical routine samples, that is, 98 archived serum specimens from randomly selected 32 patients with OV and random timepoints throughout the patient’s disease course at UZ Leuven, Belgium (hereafter referred to as the UZL-CSI OV cohort). For patient characteristics, see online supplemental table 1. The overall cumulative distribution of patient OS in the UZL-CSI OV cohort was not significantly different from the TCGA patients with OV, although median OS was moderately towards the lower side (online supplemental figure S8A), indicating that our cohort had more aggressive OV tumors. This point was better indicated by a comparison of tumor stages III/IV, wherein our cohort had far more patients with stage IV relative to stage III OV, as compared with the TCGA OV cohort (online supplemental figure S8B). Importantly though, for validating immunobiomarkers, the UZL-CSI cohort was highly suitable since it provided the necessary immunological dynamism, on account of being composed of a diversity of pretreatment and on/post-treatment sera samples (with chemo/targeted/immunotherapies) (online supplemental figure S8C). Moreover, we also had several patients for which multiple time-series samples were available (online supplemental figure S8D) combined with variable distribution of tumor burden, estimated via serum CA125 levels (online supplemental figure S8E). The UZL-CSI cohort was composed of all four pathological responses (pRs) to therapy (online supplemental figure S8F).

To immunologically benchmark this UZL-CSI cohort, relative to our TCGA observations, we pursued a Luminex assay-based multiplex profiling of various serum immunobiomarkers like NFκB target proteins (IL-10, IL-12p70, PGF, CCL2, CCL22, VEGF-A, CCL5, and MMP9), IFNs or ISGs-coded proteins (IFN-α, IFN-γ, CXCL10, and CCL11), and some pleiotropic factors (IL-6, latency-associated peptide/transforming growth factor-ß (LAP/TGF-β), arginase), or soluble immune checkpoints (LAG3, IDO1, PD-L1, PD-1, and TIM3) (figure 3A). Serum-CA125 was also profiled to enable direct comparison with a clinical standard biomarker (figure 3A). This multiplexed analysis showed that sera of UZL-CSI patients with OV enriched for more NFκB-target proteins or pleiotropic factors, than IFNs/ISGs-coded proteins (figure 3A). Accordingly, a gene set enrichment analysis (GSEA)-based biological pathway enrichment (REACTOME/Wikipathways) computed on the aforementioned concentrations showed a significant positive enrichment for innate immune response or myeloid inflammation and a significant negative enrichment for adaptive or lymphoid pathways (figure 3B). These observations indirectly hinted toward a peripheral disbalance between NFκB target proteins and IFN/ISGs-related proteins in patients with OV.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Immunological benchmarking and sFIS assay testing in discovery cohort of patients with OV. (A) Eligible serum samples from the UZL-CSI cohort were profiled for different immunological factors (n=74) and CA125 (n=95). Colors indicate association with categories mentioned in the legend. (B) REACTOME-Wikipathway GSEA based on concentrations of immunological biomarkers or CA125 (top 10 significance method). Violin plot (C) and sample-paired analysis (D) of UZL-CSI serum samples profiled for the si-NFκB or si-IFN/ISG response (n=96, Wilcoxon matched-pairs signed-rank test; *p<0.05). ‘§’ depicts serum samples wherein the si-IFN/ISG response values exceeded the si-NFκB response values. (E) Heatmap representation of Pearson correlation matrix of si-NFκB and si-IFN/ISG response, CA125 and immunological factors profiled from the serum of the UZL-CSI OV dataset (n=74). (F,G) Network analyses of NFκB (F) or IFN/ISG (G) response activating transcription program. Genes in blue indicate immunologically relevant factors enumerated by the PID analyses. (H,I) Bar graph of THP1-dual reporter cells exposed to indicated concentrations of TGF-β reporting IFN/ISG (H) or NFκB (I) responses at 48 h post-treatment (n=3, one-way analysis of variance corrected for false discovery rate via Benjamini, Krieger and Yekutieli test; *p<0.05). CA125, carbohydrate antigen 125; GSEA, gene set enrichment analysis; IFN, interferon; ISG, interferon-stimulated gene; NFκB, nuclear factor kappa-light-chain enhancer of activated B cells; OV, ovarian cancer; PID, Pathway Interaction Database; sFIS, serum-functional immunodynamic status; si, serum-induced; UZL-CSI, UZ Leuven-Cell Stress Immunity cohort.

sFIS assay captures peripheral disbalance in si-NFκB and si-IFN/ISG responses

We wondered whether our sFIS assay can faithfully capture the serum immunodynamics partly hinted by the aforementioned cytokines/factors. Remarkably, serum derived from these UZL-CSI patients with OV indeed induced significantly higher NFκB responses in the THP1 cells as compared with IFN/ISG responses (figure 3C). In fact, the median si-IFN/ISG responses were below our assay’s baseline (i.e., normal human serum pooled from several healthy donors), thereby suggesting a tendency of sera of patients with OV to downregulate IFN/ISG responses. In fact, there were only 11 serum specimens (out of 98) that induced a higher fold change of si-IFN/ISG responses than si-NFκB responses (figure 3D). These initial observations clearly outlined that our sFIS assay can faithfully capture the functional immunodynamics in sera of patients with OV, thereby more directly emphasizing the predominance of non-immunogenic inflammation in OV.

si-NFκB response links to wound healing-like pro-metastatic signaling, while si-IFN/ISG response is proimmunogenic

Next, we wished to gain some biological insights via integrated correlative and systems biology analyses of si-NFκB or si-IFN/ISG responses and serum immunobiomarkers. A correlation matrix between the major serum immunobiomarkers (median concentration >1 pg/mL), CA125 and sFIS assay responses showed the existence of two major clusters (figure 3E): (1) a very low concentration (median ~8 pg/mL) but very highly correlated, IFNγ-centered immunosuppressive cluster (co-recruiting soluble PD-1/PDL1); and (2) a very high concentration (median ~7000 pg/mL) but somewhat loosely correlated cluster, wherein two sub-clusters were formed by si-NFκB response and soluble TIM3/MMP9 or si-IFN/ISGs response and LAP/TGF-β. To further understand these co-clustering patterns, we carried out immunological functional network analyses using genetic representatives. This revealed that the NFκB-relevant genes were associated with MMP9 (but not TIM3) via anti-inflammatory or prometastatic signaling modules42 43 like wound healing facilitators (osteopontin-coding, SPP1 and integrinα M-coding (ITGAM))44 45 or protumorigenic lipid signaling (lysophosphatidic acid receptors-coding (LPAR1-5))46 47 (figure 3F). Interestingly, the IFN/ISG-relevant genes were associated with TGFB1 via proinflammatory factors like IL-27, IL-12, STAT3 and IRF1 (figure 3G). This implicated a contextual immunogenic role for TGF-β. In fact, it has been reported that high concentration of TGF-β can induce IFN/ISG responses.48 49 Intriguingly, we indeed saw that treatment of THP1-reporter cells with high concentrations of TGF-β (like concentrations recovered from OV-patients in UZL-CSI cohort) induced highly significant IFN/ISG (but not NFkB) responses (figure 3H,I), thereby suggesting a context-dependent pro-immunogenic role for TGF-β.

The aforementioned indications were further confirmed by correlation of these biomarkers with medium (progression-free survival (PFS)) or long-term (OS) survival of these patients with OV such that si-NFκB responses clustered with soluble-TIM3, IFNγ–PD-1–PDL1 axis, and CA125 to predict shorter PFS/OS of patients with OV (figure 4A and online supplemental figure S9), whereas si-IFN/ISG responses clustered with LAP/TGF-β, CCL5, CCL22, and CCL11 to predict prolonged PFS/OS of patients with OV (figure 4A and online supplemental figure S9). Remarkably, an immune-amplitude analysis for si-NFκB or si-IFN/ISG responses showed that patients with OV with prolonged survival (PFS/OS) (figure 4B,C) and reduced serum-CA125 (figure 4D,E) had a more pro-immunogenic orientation between si-IFN/ISG and si-NFκB responses (i.e., si-IFN/ISG resp. > si NFκB resp.), whereas patients with a clear disbalanced orientation between these two inflammatory modules (si-NFκB resp. ≥ si-IFN/ISG resp.) had much shorter survival (especially OS) and mostly increased serum-CA125. Overall, this established that sFIS assay-derived si-NFκB or si-IFN/ISG responses are robust at capturing the complex and multifactorial functional immune-signaling or prognostic modules in the periphery of patients with OV.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Immunological characteristics of si-NFκB/si-IFN/ISG responses and their prognostic impact on patients with OV. (A) PCA representation of Spearman’s correlation analyses between PFS or OS of patients with OV with si-NFκB and si-IFN/ISG response, CA125 and immunological factors profiled from the serum of the UZL-CSI OV dataset (fold change to baseline/healthy serum) (median value-based data integration for multiple serum samples per patient, n=32). (B–E) Cubic spline analyses of si-NFκB or si-IFN/ISG responses (B,C) or CA125 concentration (D,E) profiled from the serum of the UZL-CSI OV cohort dataset (median value-based data integration for multiple serum samples per patient; fold change to baseline/healthy serum) and distributed as per PFS (B,D) or OS values (C–E) (n=32). (F) Radar plot of Pearson correlation values of CA125 and si-NFκB and si-IFN/ISG response with blood cell counts (n=66). (G–I) Pearson correlation of CA125 (G), si-NFκB (fold change to baseline/healthy serum) (H) and si-IFN/ISG (fold change to baseline/healthy serum) (I) responses, to NLR (CA125, n=64; si-NFκB/si-IFN/ISG response, n=65; *p<0.05). CA125, carbohydrate antigen 125; IFN, interferon; ISG, interferon-stimulated gene; NFκB, nuclear factor kappa-light-chain enhancer of activated B cells; NLR, neutrophil-to-lymphocyte ratio; OS, overall survival; OV, ovarian cancer; PFS, progression-free survival; si, serum-induced; UZL-CSI, UZ Leuven-Cell Stress Immunity cohort.

Peripheral myeloid, but not lymphoid, cells associate with si-NFκB and si-IFN/ISG responses

The aforementioned observations suggested that si-NFκB or si-IFN/ISG responses are both linked to myeloid, rather than lymphoid, compartment. To confirm this, we correlated serum-CA125, si-NFκB, or si-IFN/ISG response and (sample-matched) absolute immune cell counts in the blood, for lymphocytes, monocytes/myeloid cells, neutrophils, and platelets. Remarkably, serum-CA125 and si-NFκB and si-IFN/ISG responses collectively showed negative correlation with lymphocytes (figure 4F). On the level of myeloid cells, serum-CA125 and si-NFκB responses positively correlated with monocytic myeloid cells and neutrophils, whereas si-IFN/ISG responses preferred to positively correlate with monocytic myeloid cells and platelets (figure 4F). This implied the presence of a contradictory interplay between platelets and neutrophils. Herein, the disjointed trends for neutrophils or lymphocytes were intriguing since neutrophil-to-lymphocyte (NLR) ratio is a well-established negative prognostic biomarker.5 Interestingly, specifically si-NFκB response (and to a certain extent, serum-CA125) positively correlated significantly to NLR (figure 4G,H), thereby reinforcing its negative prognostic role. However, si-IFN/ISG response failed to show correlation with NLR (figure 4I). Overall, this confirms that in OV, the peripheral immune macroenvironment exhibits disjointed myeloid and lymphoid crosstalk on the levels of si-NFκB or si-IFN/ISG responses.

si-NFκB or si-IFN/ISG responses associate with differential prognosis in OV

Next, we wanted to estimate the clinical biomarker associations of our sFIS assay readouts. First, we pursued differential tumor-level pR prediction for sFIS assay readouts relative to serum-CA125 (i.e., complete response (CR), partial response, stable disease, or progressive disease (PD)). Serum-CA125 was largely reduced in patients with OV showing CR, whereas it was strongly enriched in other pR subgroups, especially PD (online supplemental figure S10A). Accordingly, non-responder patients with OV (with PD) had significantly higher serum-CA125 than responder patients with OV (with CR) (online supplemental figure S10B). However, similar differentiating trends for pR were absent for sFIS assay readouts (online supplemental figure S10C–F).

Serum-CA125’s superiority in predicting differential pR is consistent with its clinical performance.25 50 However, the inability of sFIS assay readouts to not reliably predict pR is also in line with frequently published observations that most immunobiomarkers are better at predicting ‘long-term’ survival trends rather than ‘short-term’ pR.51 52 Accordingly, the patients with OV with shorter PFS (<2 years) (figure 5A) or shorter OS (<3 years) (figure 5B) possessed serum that induced significantly higher NFκB responses, whereas patients with OV with longer PFS/OS (>2/3 years) (figure 5C,D) had higher si-IFN/ISG responses. Herein, heightened serum-CA125 partially indicated shorter PFS (figure 5E) but failed to predict differential OS (figure 5F). Accordingly, si-NFκB response was significantly associated with reduced PFS/OS of patients with OV, whereas si-IFN/ISG responses were only significantly associated with OS, while serum-CA125 could only partially predict differential PFS but not OS (online supplemental figure S11A,B). Interestingly, similar to serum-CA125, NLR also did not correlate with PFS/OS in a conclusively negative prognostic manner (online supplemental figure S11C,D).

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

Prognostic impact of si-NFκB or si-IFN/ISG responses of patients with OV. (A–F) Violin plots of si-NFκB responses (A,B) or si-IFN/ISG responses (C,D) and CA125 concentration (E,F) (median value-based data integration for multiple serum samples per patient) profiled from the serum of the UZL-CSI OV cohort (fold change to baseline/healthy serum). These were subdivided as longer and shorter PFS (CA125, longer PFS n=5 vs shorter PFS n=26; si-NFκB or si-IFN/ISG resp., longer PFS n=5 vs shorter PFS n=27) (A,C,E) or OS (CA125, longer OS n=8 vs shorter OS n=23; si-NFκB or si-IFN/ISG resp., longer PFS n=9 vs shorter PFS n=23) (B,D,F) based on a cut-off of 2 years of PFS or 3 years of OS (Mann-Whitney test, two-tailed; *p<0.05). (g–j) KM plots of si-NFκB (G,I) or si-IFN/ISG (H,J) responses (median value-based data integration for multiple serum samples per patient) of the UZL-CSI OV cohort. Subdivision of PFS (G,H) or OS (I,J) as high or low were based on the 75th percentile cut-off (si-NFκB or si-IFN/ISG resp. vs PFS/OS, HIGH n=8 vs low n=24). The plots depict the HR ±95% CI (log-rank Mantel-Cox test, *p<0.05). CA125, carbohydrate antigen 125; IFN, interferon; ISG, interferon-stimulated gene; KM, Kaplan-Meier; NFκB, nuclear factor kappa-light-chain enhancer of activated B cells; OS, overall survival; OV, ovarian cancer; PFS, progression-free survival; si, serum-induced; UZL-CSI, UZ Leuven-Cell Stress Immunity cohort.

These trends were also supported by univariate Kaplan-Meier survival analysis of the PFS of patients with OV (figure 5G,H, and online supplemental figure S11E) or OS (figure 5I,J and online supplemental figure S11F), such that very high (>75th percentile cut-off) si-NFκB responses were associated with reduced survival, whereas very high si-IFN/ISG responses were associated with prolonged survival. These trends were largely substantiated by multivariate Cox Proportional Hazards (CoxPh) modeling, correcting for age, tumor stage, or intra-sFIS assay variations (online supplemental figure S12), especially for OS (online supplemental figure S12D) and differential fold-change cut-off analyses (online supplemental figure S12B–F). Overall, this highlights the biomarker efficacy of our sFIS assay (which simultaneously integrates a negative and positive prognostic factor within a single modality) in predicting medium-term/long-term survival of patients with OV.

si-NFκB responses predict malignancy risk in patients with OV

Finally, we wished to validate as well as extend our most reliable observations in a validation cohort. Serum-CA125 is routinely used for OV diagnosis; however, there is ample room for co-diagnostic modalities. Since si-NFκB response exhibited reliable negative prognostic impact and association with pro-metastatic wound healing-like signaling, we became curious about its co-diagnostic efficacy. Hence, we retrospectively analyzed serum samples from a large-scale (699 patients), multicenter European consortium biomarker study, TRANS-IOTA.19 53 TRANS-IOTA involved serum from patients recruited in Italy (Rome/Milan), Czech Republic (Prague), Belgium (Genk/Leuven), and UK (London). TRANS-IOTA consisted of serum collected at diagnosis from patients with benign ovarian lesions (n=404), borderline tumors (borderline OV, n=90), or invasive/metastatic tumors (malignant OV, n=205). Herein, PFS (but not OS) estimates were available for patients with malignant OV. For patient characteristics, please see online supplemental table 2).

Largely irrespective of sub-histological categories, serum-CA125 showed the highest increase in patients with malignant OV (figure 6A). Similarly, although si-NFκB response was observed in patients with benign lesions/borderline OV, it was more highly enriched in malignant OV (figure 6B). Overall, both serum-CA125 (figure 6C) and si-NFκB responses (figure 6D) significantly differentiated malignant OV from benign lesions or borderline OV. In line with discovery cohort, si-NFκB responses consistently exceeded si-IFN/ISG responses irrespective of histopathological categories (figure 6B and online supplemental figure S13A–C). Within the malignant OV category, serum-CA125 was strongly proficient at distinguishing the different tumor stages, showing a consistent ascending trend from stage I to stage IV malignant OVs (online supplemental figure S14A). However, si-NFκB response was mainly able to distinguish the most aggressive stage IV from the least aggressive stage I (online supplemental figure S14B). Overall, in line with the aforementioned trends, very high-serum CA125 or si-NFκB response was associated with lower PFS in patients with malignant OV, with statistical significance achieved only by si-NFκB response (figure 6E,F). The si-IFN/ISG responses did not phenocopy above trends for si-NFκB response (online supplemental figure S15A) and, in line with our discovery cohort, si-IFN/ISG responses failed to predict differential PFS (online supplemental figure S15B). Taken together, these observations validate si-NFκB response as a negative prognostic biomarker, with putative co-diagnostic efficacy along with serum-CA125.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

sFIS assay testing to estimate malignancy risk in a validation cohort of patients with ovarian cancer. (A,B) Heatmap representation of CA125 (A) or si-NFκB and si-IFN/ISG response (B) per histology category profiled from the serum of the TRANS-IOTA cohort (fold change to baseline/healthy serum, averaged values; for full patient details and numbers, see online supplemental table 2). ‘*’ indicates inclusion of patients with cysts at the ultrasound. (C,D) Violin plot of CA125 (C) or si-NFκB response (D) of patients with benign (CA125, n=302; si-NFκB response, n=404), borderline (CA125, n=85; si-NFκB response, n=90) and malignant (CA125, n=198; si-NFκB response, n=205) cancers, profiled from the serum of the TRANS-IOTA cohort (fold change to baseline/healthy serum, Kolmogorov-Smirnov test; *p<0.05). (E,F) Violin plot of PFS based on CA125 (low, n=42; high, n=14) (E) or si-NFκB response (low, n=41; high, n=15) (F) of patients with cancer (fold change to baseline/healthy serum) (Mann-Whitney test, *p<0.05). Subdivision of CA125 and si-NFκB (E,F) as high or low were based on the 75th percentile cut-off. CA125, carbohydrate antigen 125; IFN, interferon; ISG, interferon-stimulated gene; NFκB, nuclear factor kappa-light-chain enhancer of activated B cells; PFS, progression free survival; si, serum-induced; sFIS, serum-functional immunodynamic status.

Peripheral immunodynamic status can guide chemo-immunotherapy regime’s design

Patients with OV exhibiting a bad prognostic, si-NFκB response^HIGHsi-IFN/ISG response^LOW/NULL status, is consistent with OV’s immunoresistant nature. Hence, we wondered whether blunting si-NFκB response can be used as a guiding strategy for precision anti-OV immunotherapy. To address this, we pursued an in silico drug prediction relying on a computational algorithm exploiting biomedical literature-associated drug–gene relationships to predict drugs or drug–target’s associations to the NFκB response signature (figure 7A). These analyses enriched for anti-cytokine immunotherapies or inflammatory therapeutic targets, especially TNF inhibitory/blocking therapeutics like TNF-alpha inhibitors or infliximab (an anti-TNF antibody (Ab)) (figure 7). This was concurrent with the observation that, among various cytokines we screened, TNF was predictably the most robust NFκB response inducer (figure 2D).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7

sFIS assay-based prediction of chemoimmunotherapy regime’s design and in vivo testing in murine metastatic ovarian cancer model. (A) In silico drug-prediction analyses based on the NFκB response signature (cut-offs: multiple adjustment test, Bonferroni; significance level, adjusted p value=0.01). Colors indicate overlapping genes between NFκB response signature and literature-associated drug–gene sets. (B) Overview of the tumor inoculation, serum collection, and therapeutic treatment schedules for the mice experiments. (C) si-IFN/ISG and si-NFκB response of J774 dual-reporter cell lines exposed to mouse serum obtained from day 49 as a ratio to day 13 (control, n=8; anti-TNF Ab, n=8; PARPi, n=7; PARPi and anti-TNF Ab, n=8; PTX-CBP, n=7; PTX-CBP and anti-TNF Ab, n=6). (D,E) Kaplan-Meier plots of overall survival (D) or survival while considering the first drainage of ascitic fluid (E) of metastatic ID8 tumor-bearing mice treated with different therapy regimes (control, n=8; anti-TNF Ab, n=8; PARPi, n=8; PARPi and anti-TNF Ab, n=8; PTX-CBP, n=7; PTX-CBP and anti-TNF Ab, n=7) (log-rank Mantel-Cox test; *p<0.05, **p<0.01). IFN, interferon; ISG, interferon-stimulated gene; NFκB, nuclear factor kappa-light-chain enhancer of activated B cells; PARPi, PARP inhibitor; PTX-CBP, paclitaxel +carboplatin; sFIS, serum-functional immunodynamic status.

The notion of applying anti-cytokine immunotherapies like anti-TNF Ab is not new. While such anti-cytokine immunotherapies have shown preclinical success, they have often failed in clinical trials.4 It has been proposed that their success might be contingent on biomarker-driven application; however, such precision biomarkers are elusive.4 Therefore, we wondered whether si-NFκB response can better guide application of anti-TNF Ab in patients with OV. In fact, reanalyses of a small (only existing) human OV clinical trial administering infliximab, wherein pretreatment/on-treatment transcriptome for OV ascites was available,54 55 showed that TNF in OV ascites could not differentiate infliximab-responding versus non-responding patients (online supplemental figure S15C), while elevated NFκB response signature differentiated them relatively better (online supplemental figure S15D).

To validate this via sFIS assay, we used the murine ID8 cell-based orthotopic model of metastatic OV56 since these metastatic ID8 tumor-bearing mice experience a latent spike in serum TNF levels (online supplemental figure S15E). Various studies have demonstrated antitumor and immunostimulatory impact of anti-OV therapies like paclitaxel + carboplatin (PTX-CBP) and/or PARP inhibitors (PARPi),57 58 yet it is not clear which of these therapies might best potentiate the antitumor efficacy of anti-TNF Ab. Also, we were curious whether sFIS assay can predict the antitumor efficacy of the aforementioned therapeutics (alone or in combination with anti-TNF Ab). Hence, we treated the ID8 tumor-bearing mice with PTX-CBP, PARPi, or anti-TNF Ab, or combinations thereof (figure 7B). Initially, we collected the murine serum at baseline and during the aforementioned treatments (figure 7B) and screened the ability of these serum samples to induce si-NFκB or si-IFN/ISG responses in a murine version of our sFIS assay, that is, J774 murine myeloid cells stably expressing two inducible reporter constructs for NFκB or IFN/ISG response signaling. Interestingly, a ratio-metric analysis (on to pretreatment ratio) of murine sFIS assay derived si-NFκB or si-IFN/ISG responses showed that anti-TNF Ab exhibited threshold tendencies to reduce si-NFκB responses while marginally increasing si-IFN/ISG responses (figure 7C). Interestingly, combining anti-TNF Ab with PTX-CBP created a much better peripheral pro-immunogenic status (si-NFκB response^LOWsi-IFN/ISG response^HIGH) than combining with PARPi (si-NFκB response^MEDsi-IFN/ISG response^LOW), or either therapy alone (figure 7C).

To understand whether sFIS assay can predict antitumor efficacy in vivo, we estimated two long-term survival parameters, that is, OS and OS relative to first ascitic fluid’s drainage (OS-AFD). The long-term OS trends indeed followed the aforementioned sFIS assay predictions, such that the ID8 tumor-bearing mice treated with PTX-CBP and anti-TNF Ab triple-combo had the longest OS (figure 7D) and OS-AFD (figure 7E). In fact, PTX-CBP and anti-TNF Ab combo was the only condition that had significantly higher anti-OV efficacy than anti-TNF Ab alone (figure 7D,E), thereby highlighting the superiority of PTX-CBP (over PARPi) in unleashing anti-TNF Ab’s potential. Moreover, a ratio-metric analyses established that a peripheral si-IFN/ISG response^HIGHsi-NFκB response^LOW status positively correlated with prolonged OS (online supplemental figure S15F). In conclusion, the sFIS assay may help in designing anti-OV regimen, wherein si-IFN/ISG response^HIGHsi-NFkB response^LOW phenotype could guide selection of pro-immunogenic therapies.