Age-dependent cytokine surge in blood precedes cancer diagnosis

Abstract

Aging is associated with increased variability and dysregulation of the immune system. We performed a system-level analysis of serum cytokines in a longitudinal cohort of 133 healthy individuals over 9 y. We found that cancer incidence is a major contributor to increased cytokine abundance variability. Circulating cytokines increase up to 4 y before a cancer diagnosis in subjects with age over 80 y. We also analyzed cytokine expression in 10 types of early-stage cancers from The Cancer Genome Atlas. We found that a similar set of cytokines is upregulated in tumor tissues, specifically after the age of 80 y. Similarly, cellular senescence activity and CDKN1A/p21 expression increase with age in cancer tissues. Finally, we demonstrated that the cytokine levels in serum can be used to predict cancers among subjects age at 80+ y. Our results suggest that latent senescent cancers contribute to age-related chronic inflammation.

Keywords: aging; cancer; cytokine surge; immunology.

Fig. 1.

Elevated systemic cytokine levels correlate with cancer incidence. The data come from Stanford-Ellison Cohort. (A) The sampling and analytical scheme for a 9-y longitudinal cohort. (B) The distribution of geometric average cytokine abundance across 557 serum samples (referred to as serum cytokine abundance hereafter) collected in a 9-y longitudinal cohort is shown. Outliers are defined as samples with a significant deviation from the central populational mean (False Discovery Rate, FDR <0.05, SI Appendix). (C) The variability of serum cytokine abundances in different age groups is shown. Each point represents the SD among samples collected in a given sampling year for a given age group. Paired Wilcoxon rank tests were performed. n.s., P > 0.05, *, P < 0.05. (D) The scheme to examine the association between serum cytokine abundance and clinical events. The cytokine abundance was quantified within the defined time window (± 2 y relative to diagnosis). It was compared with samples collected outside the diagnosis time window. (E) The relative serum cytokine abundance for different disease categories. Label N provides the number of samples within the time window for a given category of diagnosis.

Fig. 2.

The level of a broad spectrum of cytokines rises in the blood samples collected before cancer diagnosis within the aged population (80+ y). The data come from Stanford-Ellison Cohort. (A) The heatmap of serum cytokine abundance in samples collected within 2 y before cancer diagnosis (prediagnostic serum samples). We used the Euclidean distance to cluster the samples and cytokines. The two major clusters of samples (inflamed vs. noninflamed state) are highlighted by colored sidebars (red vs. blue). In addition, three colored sidebars show the cancer types, sex, and sampling age. We examined the correlation between these clinical characteristics and the inflammation state. (B) The correlation between the serum cytokine abundance and age is shown. We examined the correlation in cancer naive subjects (without a cancer history, gray) or subjects going to have a cancer diagnosis within 2 y after the serum sampling (red). (C) The prediagnostic serum samples are grouped by cancer types. The serum cytokine abundance of samples is compared to the cancer naive subjects in each age group (<80 y or 80+ y). (D) The individual cytokine whose level rises in prediagnostic samples. Two age groups (<80 y and 80+ y) were compared to their matched (sex, age, and sample year, SI Appendix) cancer naive samples. The ones reached significance for the subjects age at 80+ y (FDR <0.05) are highlighted with a bold font. No cytokine reached significance in the younger group (all of their FDR >0.2).

Fig. 3.

Long-term movement of serum cytokine baselines relative to disease diagnosis. The data come from Stanford-Ellison Cohort. (A–C) We plotted the serum cytokine abundance of samples collected around disease diagnoses (A, cardiovascular diseases; B, inflammatory conditions; C, cancers) on a time scale relative to the diagnosis. The samples are grouped by age (<80 y vs. 80+ y). The longitudinal profiles of subtypes of cancers or excluding the outliers are shown in SI Appendix, Fig. S3 A–C. The profiles for individual cytokine are shown in SI Appendix, Fig. S3D. (D and E) related to C, we plotted the two cardiovascular risk burdens (body mass index/BMI in D, and systolic blood pressure in E) against the sample collection times relative to cancer diagnosis. Upper limits of the normal values (BMI = 30, systolic blood pressure = 140 mmHg) are highlighted by the dashed red lines. The lines for smoothed average values and their 95-percentile CI in a-e were calculated by the loess function. (F) It shows the longitudinal profile of serum cytokine abundance of all subjects over 80 y with elevation outliers. All of them developed cancers, with diagnosis around the peak of cytokine abundance. Cancer type is indicated by color. The drop of cytokines at 89 y old for Subject A was verified by two independent samples in that year.

Fig. 4.

Advanced age elevates cytokine transcription in early-stage cancers along with senescence activity. The data come from TCGA. We examined the effect of advanced age (80+ y) on cytokine transcription (using the RNAseq dataset from TCGA) among early-stage cancer tissues across different cancer types (SI Appendix). (A) The sample counts for different age groups across cancer types are shown. (B) We calculated the geometric mean of transcript abundance of cytokine surge genes (Fig. 2D). This value is referred to as tissue cytokine abundance hereafter. The Upper panel reports the standardized difference/ effect sizes (defined in SI Appendix) for comparisons within individual cancer types. The length of the lines covers the 95 percentile CI, and the dot denotes the mean. The Lower panel reports the summarized standardized difference across cancer types with P-values. (C) similar to B, but we measured the correlation between the transcript abundance of CDKN1A/p21 and cytokine surge genes across early-stage cancers. (D) The standardized difference between <80 and 80+ y age groups are shown for individual cytokine surge genes. (E) We performed gene set enrichment analysis using the sets related to the cellular senescence process. SI Appendix, Table S7 provides the information for these gene sets. (F) Forest plots for cytokine genes and CDKN1A upregulated by age across cancer tissues. The presentation format is same to B.

Fig. 5.

Tissue cytokine abundance correlates with age in a nonlinear manner. The data come from TCGA. We defined patients under 60 y old as the reference group. We compared all other 5-y age range subgroups with the reference group using a similar meta-analysis procedure, as shown in Fig. 4B. (A) Tissue cytokine abundance in different age groups is compared with the reference group. (B and C) We performed gene set enrichment analysis between different age groups and the reference group. The log10 P values are shown, with a sign indicating change directions (+, upregulation; −, downregulation). (D) The relative expression level of the CDKN1A gene in different age groups is shown. n.s., P > 0.05, *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Fig. 6.

Serum cytokine abundance predicts future cancer incidence in the subjects age at 80+ y. The data come from Stanford-Ellison Cohort. The serum cytokine abundance is the geometric mean of 32 cytokines’ concentration in serum samples (same as Fig. 1B). (A) Samples from the aged group(80+ y) were grouped by the sampling time relative to the cancer diagnosis. All samples collected after the cancer diagnosis were removed. The negative samples include samples from cancer naive subjects as well as samples collected from cancer patients 5 or more years prior to the diagnosis. We performed Wilcoxon rank-sum test between the negative and other sample groups. n.s., P > 0.05, *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. (B) The ROC curve describes the accuracy of cancer prediction based on serum cytokine abundance. The comparisons were between the prediagnostic samples and the negative samples defined in A.