Host transcriptomic signature as alternative test-of-cure in visceral leishmaniasis patients co-infected with HIV

Abstract

Background: Visceral leishmaniasis (VL) treatment in HIV patients very often fails and is followed by high relapse and case-fatality rates. Hence, treatment efficacy assessment is imperative but based on invasive organ aspiration for parasite detection. In the search of a less-invasive alternative and because the host immune response is pivotal for treatment outcome in immunocompromised VL patients, we studied changes in the whole blood transcriptional profile of VL-HIV patients during treatment.

Methods: Embedded in a clinical trial in Northwest Ethiopia, RNA-Seq was performed on whole blood samples of 28 VL-HIV patients before and after completion of a 29-day treatment regimen of AmBisome or AmBisome/miltefosine. Pathway analyses were combined with a machine learning approach to establish a clinically-useful 4-gene set.

Findings: Distinct signatures of differentially expressed genes between D0 and D29 were identified for patients who failed treatment and were successfully treated. Pathway analyses in the latter highlighted a downregulation of genes associated with host cellular activity and immunity, and upregulation of antimicrobial peptide activity in phagolysosomes. No signs of disease remission nor pathway enrichment were observed in treatment failure patients. Next, we identified a 4-gene pre-post signature (PRSS33, IL10, SLFN14, HRH4) that could accurately discriminate treatment outcome at end of treatment (D29), displaying an average area-under-the-ROC-curve of 0.95 (CI: 0.75-1.00).

Interpretation: A simple blood-based signature thus holds significant promise to facilitate treatment efficacy monitoring and provide an alternative test-of-cure to guide patient management in VL-HIV patients.

Funding: Project funding was provided by the AfricoLeish project, supported by the European Union Seventh Framework Programme (EU FP7).

Keywords: Blood signature; HIV; RNA signature; Treatment efficacy; Visceral leishmaniasis.

Fig. 1

Flow charts of study participants and study design. (a) Flow chart of inclusion and stratification by treatment outcomes of study participants (b) Lay-out of the study design and generation of 4-gene signature. LTFU: lost to follow-up, D: Day of treatment, VL-HIV: visceral leishmaniasis-HIV co-infection, DGE: differential gene expression.

Fig. 2

Distinct patterns of differentially expressed genes (DEGs) in treatment success and failure patients. (a) Overlap of differentially expressed genes (DEGs) between D0 and D29 for treatment success and failure patients. Venn diagram showing the number, uniqueness and directionality of the DEGs (>1.5 absolute fold change with FDR p-value <0.05 – see mechanistic approach in materials and methods) between D0 and D29 for treatment success (orange, n = 12) and failure cases (red, n = 16) (b) Unsupervised clustering of differentially expressed genes (DEGs) in treatment success and (c) failure patients. Heat maps showing the z-scores (bottom scale) or extent of up (red) or downregulation (blue) of the DEGs (Y-axis) at D0 (green samples) and D29 (purple samples) for all individual treatment success patients (n = 12) and all individual treatment failure patients (n = 16). Unsupervised hierarchal clustering of the samples was applied and based on Euclidean distance and complete linkage. D: Day of treatment.

Fig. 3

Enriched gene sets in the treatment success group (n = 12). Normalized enrichment scores for hallmarks, GO terms and canonical pathways from the MsigDB database and previously described blood transcriptional modules are depicted in blue (downregulated) or red (upregulated) scale. The size column represents the number of genes from the respective gene set found in the expression dataset. NES: normalized enrichment score.

Fig. 4

Functionally grouped network analyses of enriched gene sets in the treatment success group (n = 12). Functionally related clusters of enriched gene sets are represented by different colours and the label of the most significant term per cluster is shown. The node size represents the term enrichment significance and the connectivity by grey lines (kappa statistics 0.4). Similar terms were fused to reduce redundancy. Bottom tables showed the directionality of significant terms and the frequency of genes from the respective terms detected in the expression dataset. The exact number of genes detected is indicated behind each bar.

Fig. 5

Receiver operator curves (ROC) for the final random forest classifier based on the 4-gene set. (a) Showing the average AUC value of the 4-gene random forest classifier, calculated over 1000 bootstrap replicates. (b) Bar chart showing the sum of relative importancies in the classifier over 1000 bootstrap replicates.