The HIV Restriction Factor Profile in the Brain Is Associated with the Clinical Status and Viral Quantities

Abstract

HIV-encoded DNA, RNA and proteins persist in the brain despite effective antiretroviral therapy (ART), with undetectable plasma and cerebrospinal fluid viral RNA levels, often in association with neurocognitive impairments. Although the determinants of HIV persistence have garnered attention, the expression and regulation of antiretroviral host restriction factors (RFs) in the brain for HIV and SIV remain unknown. We investigated the transcriptomic profile of antiretroviral RF genes by RNA-sequencing with confirmation by qRT-PCR in the cerebral cortex of people who are uninfected (HIV[-]), those who are HIV-infected without pre-mortem brain disease (HIV[+]), those who are HIV-infected with neurocognitive disorders (HIV[+]/HAND) and those with neurocognitive disorders with encephalitis (HIV[+]/HIVE). We observed significant increases in RF expression in the brains of HIV[+]/HIVE in association with the brain viral load. Machine learning techniques identified MAN1B1 as a key gene that distinguished the HIV[+] group from the HIV[+] groups with HAND. Analyses of SIV-associated RFs in brains from SIV-infected Chinese rhesus macaques with different ART regimens revealed diminished RF expression among ART-exposed SIV-infected animals, although ART interruption resulted in an induced expression of several RF genes including OAS3, RNASEL, MX2 and MAN1B1. Thus, the brain displays a distinct expression profile of RFs that is associated with the neurological status as well as the brain viral burden. Moreover, ART interruption can influence the brain's RF profile, which might contribute to disease outcomes.

Keywords: ART; HIV-1; MAN1B1; RNA-seq; SIV; host restriction factors; machine learning; transcriptomics.

Figure 1

Plasma, CSF and brain viral quantities in the human cohort. Four experimental groups ((HIV[−], n  =  10), (HIV[+], n  =  10), (HIV[+]/HAND, n  =  10) and (HIV[+]/HIVE, n  =  10)) were examined for plasma and CSF HIV RNA copies/mL (A), HIV RNA (B), total DNA (C) and integrated DNA (D) copies/g of brain tissue. The horizontal lines represent the mean values (**, p < 0.01; ****, p < 0.0001).

Figure 2

Relative host restriction factors’ mRNA expression levels. RNA isolated from the post-mortem brains of the human cohort was subjected to bulk RNA-seq. Sixty known restriction factor genes with HIV restrictive capabilities were screened in the RNA-seq dataset. Differentially expressed genes are reported in different group comparisons. (A) HIV[+] compared to HIV[−], (B) HIV[+]/HAND compared to HIV[−] and (C) HIV[+]/HIVE compared to HIV[−]. The fold changes range from −2 to −1, 0, 1, 2, 3 and 4 log2 fold changes (or 0.25, 0.5, 1, 2, 4, 8 and 16-fold changes). Genes with 0 to −2 log2 fold changes are considered downregulated, and those with 0 to 4 log2 fold changes are considered upregulated. Asterisks indicate the genes that were selected for validation by qRT-PCR.

Figure 3

Heat map of upregulated host restriction factors in different experimental groups. The top upregulated genes (n = 10) were selected from each of the HIV[+] groups compared to the HIV[−] group derived from the RNA-seq database. For the comparisons within HIV[+] groups, with or without HAND, the top upregulated genes with a minimum 1.0 log2 fold change were selected.

Figure 4

Relative host restriction factors’ mRNA expression levels. Highly upregulated RFs identified in the RNA-seq database were subjected to qRT-PCR validation. The mRNA expression levels for each gene (A–R) were normalized to GAPDH and are reported as the fold change relative to the HIV[−] control group. The horizontal lines represent the mean values and the error bars represent the standard deviation (*, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001).

Figure 4

Relative host restriction factors’ mRNA expression levels. Highly upregulated RFs identified in the RNA-seq database were subjected to qRT-PCR validation. The mRNA expression levels for each gene (A–R) were normalized to GAPDH and are reported as the fold change relative to the HIV[−] control group. The horizontal lines represent the mean values and the error bars represent the standard deviation (*, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001).

Figure 5

Comparison of downregulated restriction factors in different experimental groups. Restriction factors with a minimum −0.4 log2 fold change (0.75 fold change) were selected from different group comparisons based on the RNA-seq dataset. The most downregulated genes belonged to the two HIV[+] groups with HAND, where IFITM2 and APOBEC3B had −1.36 and −1.31 log2 fold change downregulation in the HIV[+]/HAND group compared to the HIV[+]/HIVE group, respectively.

Figure 6

Relative mRNA expression of downregulated restriction factors. The top downregulated genes validated by qRT-PCR using appropriate primers showed that the SELPLG, PPIA and TREX1 mRNA expression levels were not lower in the HIV[+] and HIV[+]/HAND groups compared to HIV[−] (A–C). The MAN1B1 gene was upregulated in both HAND groups compared to the HIV[+] and HIV[−] groups, despite the RNA-seq suggesting otherwise (D). The horizontal lines represent mean values, and the error bars represent the standard deviation (*, p < 0.05; **, p < 0.01; ***, p < 0.001).

Figure 7

Analyses of variables predicting clinical group classification by machine learning. The regularized logistic regression machine learning method was used to predict variables that distinguished the HIV[+] group from the HIV[−] group (A). Shrinkage discriminant analysis predicted the variable importance for the HIV[+] and HIV[+]/HAND group classification (B). An importance value of 70 was set as a threshold. An importance value between 70 and 100 indicates a heavy reliance of machine learning prediction on a specific variable in classifying and distinguishing the compared groups. (* represents CSF or plasma viral load, and ** represents brain viral load (RNA, DNA or iDNA (integrated DNA).) Host genes were derived from qRT-PCR results.

Figure 7

Analyses of variables predicting clinical group classification by machine learning. The regularized logistic regression machine learning method was used to predict variables that distinguished the HIV[+] group from the HIV[−] group (A). Shrinkage discriminant analysis predicted the variable importance for the HIV[+] and HIV[+]/HAND group classification (B). An importance value of 70 was set as a threshold. An importance value between 70 and 100 indicates a heavy reliance of machine learning prediction on a specific variable in classifying and distinguishing the compared groups. (* represents CSF or plasma viral load, and ** represents brain viral load (RNA, DNA or iDNA (integrated DNA).) Host genes were derived from qRT-PCR results.

Figure 8

Analyses of RFs in the brains of SIV-infected non-human primates by qRT-PCR. The expression levels of 10 known RFs (A–L), as well as IFNA and IFNB1 were measured in three experimental groups of SIV-infected Chinese rhesus macaques (SIV[+], SIV[+]/ATI and SIV[+]/ART). The mRNA expression levels are normalized to GAPDH and are reported as the fold change relative to the SIV[+]/ART control group. The horizontal lines represent the mean values and the error bars represent the standard deviation (*, p < 0.05; **, p < 0.01).