Extracellular RNA in a single droplet of human serum reflects physiologic and disease states

Abstract

Extracellular RNAs (exRNAs) are present in human serum. It remains unclear to what extent these circulating exRNAs may reflect human physiologic and disease states. Here, we developed SILVER-seq (Small Input Liquid Volume Extracellular RNA Sequencing) to efficiently sequence both integral and fragmented exRNAs from a small droplet (5 μL to 7 μL) of liquid biopsy. We calibrated SILVER-seq in reference to other RNA sequencing methods based on milliliters of input serum and quantified droplet-to-droplet and donor-to-donor variations. We carried out SILVER-seq on more than 150 serum droplets from male and female donors ranging from 18 y to 48 y of age. SILVER-seq detected exRNAs from more than a quarter of the human genes, including small RNAs and fragments of mRNAs and long noncoding RNAs (lncRNAs). The detected exRNAs included those derived from genes with tissue (e.g., brain)-specific expression. The exRNA expression levels separated the male and female samples and were correlated with chronological age. Noncancer and breast cancer donors exhibited pronounced differences, whereas donors with or without cancer recurrence exhibited moderate differences in exRNA expression patterns. Even without using differentially expressed exRNAs as features, nearly all cancer and noncancer samples and a large portion of the recurrence and nonrecurrence samples could be correctly classified by exRNA expression values. These data suggest the potential of using exRNAs in a single droplet of serum for liquid biopsy-based diagnostics.

Keywords: age; biomarker; breast cancer; cancer recurrence; extracellular RNA.

Fig. 1.

SILVER-seq sequencing libraries. (A) Size distribution of SILVER-seq constructed sequencing library from each serum aliquot (column), indexed by 1 to 8 (Aliquot #). Volume (microliters) is the volume of each aliquot. (B) Percentage of uniquely mapped reads of the corresponding library (column). (C) Number of exRNAs with 5 or more TPM in each library (column).

Fig. 2.

Presence of exRNAs derived from genes with tissue-specific expression. (A–C) Number and expression levels of the exRNAs derived from (A) brain-, (B) PNS-, and (C) bone marrow-specific genes. (Upper) The number of detected exRNAs in each donor. N, the total number of genes that are specifically expressed in this tissue. (Lower) Distribution of the expression of the exRNAs derived from the corresponding tissue-specific genes. (D) Distribution of SILVER-seq reads on all of the KRAS exons (x axis). (Upper) Cumulative read counts from all serum samples. (Lower) The number of serum samples with reads mapped to respective KRAS exons.

Fig. 3.

Correlations of exRNA expression with sex and age. (A) Scatter plot of normalized SILVER-seq reads mapped to X (x axis) and Y (y axis) chromosomes of every serum sample (circle). Male and female samples are colored in blue and red, respectively. (B) Numbers of exRNAs that are positively (green) and negatively (pink) correlated with age in each RNA type (row). (C) Disease classes (rows) that are associated with age-correlated exRNA genes; x axis, adjusted P value from association tests. (D) Scatter plot of exRNA age (x axis) and chronological age (y axis) for every sample (circle).

Fig. 4.

The exRNA expression in cancer and normal serum samples. (A) Distribution of exRNA expression levels of every gene in the human genome (60,675 genes in total, hg38) in 2 representative cancer samples (C10, C86) and 2 representative normal samples (N9, N15). See SI Appendix, Fig. S9 for all other samples. (B) Volcano plot of log fold change (cancer/normal) (x axis) and FDR (y axis) for all exRNAs (dots). (C–G) Expression levels of (C) RAC2, (D) KRAS, (E) CAMK2A, and (F) AL121652.1 and (G) AC048346.1 exRNA in cancer (red) and normal serum (blue). (H–K) receiver operating characteristic (ROC) curves of classification results based on (H) scaRNA, (I) polymorphic pseudogene, (J) lincRNA, and (K) miRNA, based on 3-fold cross-validations (red, green, blue).

Fig. 5.

Classification of serum samples from patients with or without cancer recurrence. (A and B) Representative ROC curves from cross-validations, using all of the genes of each RNA type as features, including (A) unprocessed pseudogenes and (B) lincRNAs. (C and D) Classifications based on prior-association genes. (C) Average AUC of 100 cross-validations (y axis) based each number of prior-association genes used as features (x axis). (D) Representative ROC curves from 100 cross-validation based on 215 prior-association genes as features.