Lung Microbial and Host Genomic Signatures as Predictors of Prognosis in Early-Stage Adenocarcinoma

Abstract

Background: Risk of early-stage lung adenocarcinoma recurrence after surgical resection is significant, and the postrecurrence median survival is approximately 2 years. Currently, there are no commercially available biomarkers that predict recurrence. In this study, we investigated whether microbial and host genomic signatures in the lung can predict recurrence.

Methods: In 91 patients with early-stage (stage IA/IB) lung adenocarcinoma with extensive follow-up, we used 16s rRNA gene sequencing and host RNA sequencing to map the microbial and host transcriptomic landscape in tumor and adjacent unaffected lung samples.

Results: Of 91 subjects, 23 had tumor recurrence over 5-year period. In tumor samples, lung adenocarcinoma recurrence was associated with enrichment in Dialister and Prevotella, whereas in unaffected lung samples, recurrence was associated with enrichment in Sphingomonas and Alloiococcus. The strengths of the associations between microbial and host genomic signatures with lung adenocarcinoma recurrence were greater in adjacent unaffected lung samples than in the primary tumor. Among microbial-host features in the unaffected lung samples associated with recurrence, enrichment in Stenotrophomonas geniculata and Chryseobacterium was positively correlated with upregulation of IL2, IL3, IL17, EGFR, and HIF1 signaling pathways among the host transcriptome. In tumor samples, enrichment in Veillonellaceae (Dialister), Ruminococcaceae, Haemophilus influenzae, and Neisseria was positively correlated with upregulation of IL1, IL6, IL17, IFN, and tryptophan metabolism pathways.

Conclusions: Overall, modeling suggested that a combined microbial/transcriptome approach using unaffected lung samples had the best biomarker performance (AUC = 0.83).

Impact: This study suggests that lung adenocarcinoma recurrence is associated with distinct pathophysiologic mechanisms of microbial-host interactions in the unaffected lung rather than those present in the resected tumor.

Figure 1:. Microbiota in tumor and unaffected lung associated with cancer recurrence.

A) Kaplan Meier curve for subjects with lung recurrence. B) Beta diversity (Bray-Curtis) showed differences based on sample type: tumor sample, unaffected lung sample, and background (BKG) samples. p-value by PERMANOVA. C) Bacterial load (copies/uL) by ddPCR showing differences between recurrence vs. no-recurrence groups in tumor and lung samples. p-values by Wilcoxon rank sum test. D) Principal Coordinate Analysis (PCoA) of tumor samples showing difference in beta diversity between subjects with and without recurrence (p=0.12, PERMANOVA). E) Bubble plot of differentially enriched taxa (FDR<0.2) between recurrence and no-recurrence groups in tumor samples. F) PCoA of unaffected lung samples showing difference in beta diversity between subjects with recurrence and without recurrence (p=0.001, PERMANOVA). G) Bubble plot of differentially enriched taxa (FDR<0.2) between recurrence and no-recurrence groups in unaffected lung samples. Taxa identified by decontam as potential contaminants are noted in Red.

Figure 2:. Random Forest classifiers as predictors of cancer recurrence.

A) Area under the curve (AUC) of receiver operator curves (ROC) based on random forest identification of taxonomic classifiers predicting recurrence in tumor and unaffected lung samples using the top 10–100% of total discriminant taxa based on Gini index values (n=309 for tumor samples, n=257 for lung samples). B) Taxa with the greatest Gini index from the random forest classifier in tumor samples (10% of taxa, with highest AUC). C) Forest plot showing median hazard ratio and 95% confidence interval of COX Proportional Hazards Model for taxa associated with time-to-recurrence in tumor samples. D) Taxa with greatest Gini index from the random forest classifier in unaffected lung samples (100% of taxa, with highest AUC), only the top 30 taxa are shown. E) Forest plot showing median hazard ratio and 95% confidence interval of COX Proportional Hazards Model for taxa associated with time-to-recurrence in unaffected lung samples, only the top 40 taxa are shown. Potential contaminants are labeled in red.

Figure 3:. Transcriptome in tumor and unaffected lung.

A) Beta diversity (Bray-Curtis) showed differences based on sample type. p-value by PERMANOVA. B) Volcano plot of differentially expressed genes (FDR < 0.2) between tumor and unaffected lung samples. C) PCoA of tumor samples showing no difference in beta diversity between subjects with and without recurrence (p=ns, PERMANOVA). D) Volcano plot of differentially expressed genes (FDR<0.2) between subjects with and without recurrence in tumor samples. E) PCoA of unaffected lung samples showing the difference in beta diversity between subjects with recurrence and without recurrence (p=0.005, PERMANOVA). F) Volcano plot of differentially expressed genes (FDR<0.2) between subjects with recurrence and without recurrence in unaffected lung samples. G) Heatmap of canonical pathway analysis based on Ingenuity Pathway Analysis comparing recurrence vs no-recurrence groups in tumor and lung samples. Purple shows upregulation of the pathway, and yellow shows downregulation of the pathway.

Figure 4:. Prediction of recurrence and Kaplan-Meier survival analyses.

Heatmap of sparse canonical correlation between the top taxa associated with recurrence and pathways from tissue transcriptome in A) adjacent unaffected lung sample and B) tumor. C) Kaplan-Meier curve/log-rank comparison between high and low risk groups defined by the mean of correlation scores calculated from tumor based on the canonical variables identified from tumor sample recurrent group. D) Kaplan-Meier curve/log-rank comparison between high and low risk groups defined by the mean of correlation scores calculated from unaffected lung sample based on the canonical variables identified from unaffected lung sample recurrent group. E) Area under the curve median and 95% confidence interval for receiver operating characteristic curve analyses calculated for host transcriptome, microbiome, and combination datasets as predictors and cancer recurrence as outcome in tumor sample based on log counts. F) Area under the curve median and 95% confidence interval for receiver operating characteristic curve analyses calculated for host transcriptome, microbiome, and combination datasets as predictors and cancer recurrence as outcome in unaffected lung sample based on log counts.