Discovery of Candidate HIV-1 Latency Biomarkers Using an OMICs Approach

Abstract

Infection with HIV-1 remains uncurable due to reservoirs of latently infected cells. Any potential cure for HIV will require a mechanism to identify and target these cells in vivo. We created a panel of Jurkat cell lines latently infected with the HIV DuoFlo virus to identify candidate biomarkers of latency. SWATH mass spectrometry was used to compare the membrane proteomes of one of the cell lines to parental Jurkat cells. Several candidate proteins with significantly altered expression were identified. The differential expression of several candidates was validated in multiple latently infected cell lines. Three factors (LAG-3, CD147,CD231) were altered across numerous cell lines, but the expression of most candidate biomarkers was variable. These results confirm that phenotypic differences in latently infected cells exists and identify additional novel biomarkers. The variable expression of biomarkers across different cell clones suggests universal antigen-based detection of latently infected cells may require a multiplex approach.

INTRODUCTION

Combined antiretroviral therapy (ART) can suppress HIV replication in infected individuals to levels beneath the limit of detection. Despite the efficacy of ART, HIV infection remains incurable due to the existence of reservoirs of infected cells, which include both latently infected cells, and those persistently or intermittently replicating HIV at low levels. Because of these reservoirs, cessation of ART results in a rapid rebound of virus replication. HIV persists in a small subset (<0.01%) of resting CD4+ memory T-cells (1–4), but other cell types in noncirculating tissues are also involved (5–7). A cure for HIV will require elimination of all cellular reservoirs. Strategies focused on induction of HIV gene expression to induce cytopathic effects (“shock and kill”) or immune recognition and eradication while limiting virus production with ART have thus far not succeeded. Moreover, the treatments elicit significant side-effects, and are ineffective at clearing viral reservoirs (8–10). Lack of efficacy may derive from insufficient activation of HIV-1 gene expression from all genomic integration sites, poor penetration of activators to all reservoir tissues, and/or the presence of other undefined immune- and ART-inaccessible reservoirs in infected individuals. Current knowledge suggests that a cure for HIV will likely require a synergistic combination of ART with immune-based strategies to eliminate latent reservoirs. Thus, specific recognition of varied types of latently infected cells is a prerequisite for the development of any eradication strategy.

Latency is defined by a lack of detectable viral antigen expression. Thus, targeting latently infected cells will require identification of cellular biomarkers localized on the plasma-membrane. Unique expression of host proteins in latently infected HIV cells has been demonstrated by transcriptomic and proteomic approaches (11–18). These changes appear to be driven by innate cellular mechanisms induced by HIV infection-associated dysregulation of cell homeostasis. Several candidate biomarkers of latency have been proposed, including (but not limited to) CD3, CD32a, BTK (11, 13, 19). Other factors have been associated with persistently infected cells, including PD-1, LAG-3, and TIGIT (12, 14, 15, 20). However, thus far most candidate factors appear to only be expressed in partial subsets of the latent population. Moreover, because latency occurs in multiple cell types and can be established by multiple mechanisms, enumeration/targeting of the latent population will likely require more comprehensive, flexible, and potentially individualistic methods.

Several challenges exist in studying HIV latency. Investigations of in vivo latency are hampered by the rarity of latently infected cells in infected individuals (1 in 10⁶ – 10⁷ cells). Thus, biochemical, genetic, or other -OMICs analyses that require a sufficient frequency of latently infected cells in order to detect statistically valid changes in phenotypes, are currently not tenable for clinical samples without a method to isolate or enrich the latently infected cells. Thus, in vitro models of HIV latency have been developed using two approaches- the establishment of latently infected cell lines, and ex vivo manipulation of primary CD4+ lymphocytes.

Several cell line models of HIV latency have been developed using Jurkat, ACH-2, and U937 cells (21–23). Cell lines offer the advantage of clonal isolation and consistent reactivation profiles. Several ex vivo models have been developed in recent years using cells isolated from healthy human donors ((24–30) and others). Each model is unique in terms of cell type (resting or naïve), virus type (replication competent, defective, marker-containing), whether cells are directly infected without activation or activated to facilitate infection and returned to a resting (non-dividing) state. Certain models utilize pretreatment of cells with cytokines or other factors to facilitate infection or cell survival. These manipulations likely further alter the phenotype of the cells from their in vivo state. Notably, the primary cell models of latency typically produce a population where <10% of cells are capable of reactivation (10, 11, 37). Moreover, quantification of the small fraction of the total culture that is latently infected requires pan-activation of the cells (i.e. destruction of the latent phenotype). The combination of model variation and inability to measure the extent of latency without destroying the phenotype being investigated make OMICs studies using ex vivo models challenging, if not impossible, without a method to purify or enrich latently infected cells.

The identification of a biomarker(s) of latency or another method to detect and/or enrich viable, unaltered latently infected cells would significantly advance HIV cure research. We hypothesize that identification of the diverse forms of latently infected cells in various cell and tissue types will necessitate a multiplex approach using combinatorial biomarkers. Toward that end, we sought to expand the repertoire of candidate biomarkers in latently infected cells utilizing a reductionist approach employing cell line models of latency. We constructed a panel of novel latently infected cell lines utilizing the dual-colored marker virus, HIV_DF. Importantly, our approach was to use polyclonal cell lines contain multiple HIV-1 integration sites. Proteomic analysis of the membrane fraction of one such cell line identified a number of novel candidate biomarkers. We then tested for expression of these latency associated biomarkers on a larger collection of latently infected cell lines using traditional cytometry and immunoblotting methods to define biomarker variability and consistency. The differential expression of several candidates was validated. As anticipated, the expression of many candidate biomarkers was variable across individual clones, consistent with the idea that a multiplex approach will need to be deployed for detection of the broad spectrum of cell types in latent reservoirs.

MATERIALS AND METHODS

Tissue culture and virus production.

Jurkat E6-1 cells (31), and U937 cells were obtained from the NIH AIDS Reagent Program, (Germantown, MD), and cultured in RPMI 1640 media supplemented with 10% Fetalclone III (Hyclone, Logan, UT USA), 8 mM L-glutamine, 100 U/mL penicillin, and 100 U/ml streptomycin. 293T were cultured in DMEM media with the same supplementation. All cells were cultured in humidified incubators at 37°C and 5% CO₂. VSVg-pseudotyped HIV-1 DuoFlo (HIV_DF) virus stocks were produced by transient transfection using previously described methods (32). Briefly, HIV_DF+VSVg virus was produced by transfection of DuoFlo molecular clone DNA and pMD2.G vesicular stomatitis virus glycoprotein G (VSVg) expression vector (Addgene Plasmid Repository, Cambridge, MA) using PolyJet reagent as described by the manufacturer (SignaGen, Gaithesburg, VA). Viral supernatants were collected over 48 hours, clarified by centrifugation at 4000 x g for 5 minutes, aliquoted, and stored at −80°C.

Virus infections and isolation of latently infected cell lines.

5x10⁶ Jurkat cells were infected in 24-well plates by spinoculation (1050 xg / 20°C / 2h) with HIV_DF+VSVg virus at an approximate MOI of 10. After centrifugation, cells were returned to the incubator for 2 h, then brought up to 10 ml overnight. The next day cells were diluted to 3x10⁶ cells/ml. After 14 days cells 10⁷ cells were sorted by flow cytometry at the University of Nebraska, Lincoln Flow Cytometry Core facility to select for mCherry (mCh) (+), EGFP (−) population. After sorting, cells were reanalyzed biweekly for 14 days to confirm stability of mCh+/EGFP− phenotype of sorted cells. Non-sorting flow cytometry was performed using a YETI flow cytometry. In all flow experiments, cells were stained with LIVE/DEAD stain as directed by the manufacturer (ThermoFisher) and only the live cell population analyzed. Data was analyzed using FlowJo software (version 10.6.1).

Integration site profiles of cell lines.

Genomic DNA was isolated from 2x10⁶ cells using an E.N.Z.A Tissue DNA Kit as directed by the manufacturer (Omega Bio-tek, Norcross, GA). The isolated DNA was quantified using a NanoDrop 2000/c Spectrophotometer (Thermo fisher Scientific, Waltham, MA) and then stored at −80°C. The integration sites in each cell line were mapped based on a protocol previously described (33), but modified for Sanger sequencing by using altered primers without adapter sequences: Integration site linker (5’-GTAATACGACTCACTATAGGGC-3’), first round HIV primer (5’-TGTGACTCTGGTAACTAGAGATCCCTC-3’), and second round LTR primer (5’-GAGATCCCTCAGACCCTTTTAGTCAG-3’). Final PCR products were ligated into the pGEM-T cloning vector by TA cloning as directed by the manufacturer (Promega, Madison, WI). Clones were sequenced using the SP6 universal primer (GeneWiz, South Plainfield, NJ). Sequences were analyzed using SeqBuilder v14 software (DNAStar, Madison, WI), and the location of flanking genomic DNA in the human genome was determined using UCSC BLAT genome browser (University of California Santa Cruz). The determination of whether the location was present in a gene was determined by BLASTn search using the NCBI website.

SWATH-MS.

Data-independent acquisition (DIA) SWATH-MS analyses were performed as described previously (32, 34). 20 μg of each subcellular protein fraction was trypsin digested using filter-assisted sample preparation (35). Resulting peptides were purified using ZipTips and loaded onto a C18 RRHPLC column of an Exigent Expert nanoLC 415 configured for flow rate of 40 microL/min coupled to a 6600 TripleTOF^® (Sciex) mass spectrometer. All samples were loaded using a stepwise flow rate of 10 μL/min for 8.5 min and 2 μL/min for 1 min using 100% solvent A [0.1% (v/v) formic acid in HPLC water]. Peptides were eluted at a flow rate of 0.3 μL/min using a linear gradient of 5% acetonitrile with 0.1% (v/v) formic acid to 35% over 180 min. Autocalibration of spectra occurred after acquisition of every six samples using dynamic LC–MS and MS/MS acquisitions of 25 fmol β-galactosidase. Experimental samples were processed using cyclic DIA of mass spectra using variable swaths as described in Liu et al (36). Briefly, a 50-ms survey scan (MS1) was performed and all precursors within a given swath were fragmented and analyzed (MS2). Each cycle was composed of 34 25-Da swaths which covered 400Da to 1200 Da. Total Cycle time was 3.314 s using an accumulation time of 96 ms per 25-Da swath. Ions were fragmented for each MS/MS experiment in the collision cell using rolling collision energy. Spectral alignment and targeted data extraction of DIA samples were performed using PeakView v.1.2 (Sciex) using the MDM reference spectral library generated above. All DIA files were loaded and exported in txt format in unison using an extraction window of 10 min and the following parameters: 5 peptides, 5 transitions, peptide confidence of >99%, exclude shared peptides, and XIC width set at 50 ppm. This export results in the generation of three distinct files containing the quantitative output for (1) the area under the intensity curve for individual ions, (2) the summed intensity of individual ions for a given peptide, and (3) the summed intensity of peptides for a given protein. The universal homo sapiens library from www.swathatlas.org was used to search the mass spectrometry data (37). Laboratory contaminants and reversed sequences were removed from the data set prior to statistical analysis.

Statistical Analyses.

Protein spectral features were normalized prior to statistical analyses using total sum scaling. Bayes factors (BF; (38)) were used as an indicator of statistical significance for differential protein expression. Differential protein expression between the latently infected and control cells was filtered based on a |log2(fold change (FC))| > 0.6 and a |log 10(BF)| > 1.47. These parameters correspond to a fold change > 1.5 and a BF > 30 representing a conservative false discovery rate-corrected p-value (FDR) < 0.05 (39, 40). Heatmaps of the top contrasts, to include FDR identified significant events, were generated using a Euclidean distance measurement with a Ward clustering algorithm (41). Bayesian independent samples t-tests were performed using the BEST package in R. Frequentist p-values are shown for comparative purposes. Parameters are reported as log2(FC) (Latent/Control). All statistical analyses were performed in R v.3.6.0.

Reactivation profiles of latently infected cell lines.

3x10⁶ cells/well were treated with latency reactivation agents at the following concentrations: 10 nM bryostatin, 16 nM PMA, 500 nM ionomycin, 100 nM panobinostat, 1 μg/ml PHA-M, 3 nM prostratin, 1 μM SAHA (vorinostat), and 10 ng/ml TNF-α. After 24h, the cells were pelleted, washed with PBS, stained with LIVE/DEAD viability stain for 1h incubation, repelleted and washed again with PBS, then fixed by resuspension in 3.7% formaldehyde (w/v in PBS). Cells were analyzed by flow cytometry as described above.

Expression of candidate biomarkers on new cell lines.

Expression of surface biomarkers were measured by flow cytometry as follows: 5x10⁵ cells were pelleted, washed in PBS+2% FBS, and incubated for 1 h in antibodies diluted in PBS+2% FBS as follows (fluorophore; clone, company; dilution): BTK (APC; clone 53, BD Biosciences; 1/50 dilution); CD32 (APC; clone 8.26, BD Biosciences; 1/50); CD43 (APC; clone 84-3C1, Invitrogen; 1/50); CD147 (PerCP/Cy5.5; clone HIM6, Biolegend; 1/400); CD231/TSPAN7 (APC; clone REA1191, Miltenyi Biotec; 1/100); CD317/BST/Tetherin (BV421; clone Y129, BD Biosciences; 1/800); LAG-3 (PerCP/Cy5.5; 11C3C65, Biolegend; 1/100); PD-1 (Pacific Blue; E12.2H7, Biolegend; 1/100); PD-1L (APC; 29E.2A3, Biolegend; 1/100). After incubation, cells were washed twice in PBS+2% FBS, stained for 30 min. with NearIR LIVE/DEAD viability stain (Invitrogen), washed twice in PBS, fixed in 3.7% formaldehyde (v/v in PBS), and analyzed using the YETI cytometer. Overall, triplicate cell samples were assayed at least three independent times. Cytometry data was analyzed using FlowJo software, and mean fluorescence indices were determined separately on the latent (mCh+/EGFP−) and productive (mCh+/EGFP+) populations.

Validation by immunoblot was performed essentially as previously described (32). Briefly, 1x10⁶ of each cell line were fractionated using the Qproteome cell compartment kit (Qiagen). The protein concentration of each fraction was quantified by BCA assay (ThermoFisher), and equal amounts run for immunoblots. Fractions were separated by SDS-PAGE, transferred to PVDF, and immunoblotted with the antibodies to the following: GAPDH (6C5, Santa Cruz Biotechnology); Sec62 (A303-981A, Bethyl Laboratories); CypB (A7713, Abclonal); Rab10 (A305-273, Bethyl Laboraties); and SPCS1 (A305-823A-M, Bethyl Laboratories). Species-specific HRP-conjugated secondary antibodies (Santa Cruz Biotechnology) were all diluted at 1/7500. Blots were incubated with Supersignal West Pico reagent (ThermoFisher) and imaged using an Odyssey Fc system (Li-Cor, Lincoln, NE). Images were adjusted for brightness and contrast, and cropped for figures. Blots are representative of triplicate experiments.

RESULTS

Latently Infected Cell Clones

HIV latency likely involves multiple cell types and mechanisms beyond transcriptional silencing (13, 48, 49). Moreover, integration at different genomic locations can lead to differential impacts on cellular gene expression. Only a small number of latently infected cell lines are currently available, and the majority of these cell lines are clonal. Since latency can be result from diverse mechanisms and integration locations, we set out to construct additional polyclonal cell lines latently infected with HIV. To do this, we used the DuoFlo (HIV_DF) marker virus developed by the Verdin laboratory (42). This virus expresses EGFP from the HIV LTR and mCherry (mCh) from an internal EF-1α promoter, therefore latently infected cells are EGFP−/mCh+, whereas actively expressing cells are dual positive. Thus, cells can be purified by flow cytometry and latent cells can be clearly differentiated and characterized in every experiment. Jurkat cells were single-cycle infected by spinoculation with VSVg-pseudotyped HIV_DF, cultured for 10-14 days, then purified by cell sorting (Fig. 1A). These cells were dubbed ‘JDF’ for Jurkat DuoFlo infected. Initially two bulk sorts of latent (EGFP−/mCh+) populations from two independent infections (JDF-A, JDF-B) were performed. Monitoring of the cells over time we found that the virus reactivated in a subset of cells (data not shown). After several passages of outgrowth, the latent population stabilized with ~60% and 55% latent cells in the JDF-A and JDF-B cell lines, respectively (Fig. 1B). To achieve a higher proportion of latent cells, the JDF-A cells were subjected to another round of cell sorting in which both bulk and single cell sorting was performed. The JDF-A2hi and -A2lo cell lines were derived from a bulk sort of the two distinct mCh(−)+/EGFP(−) populations in the JDF-A cells (Fig. 1C). Concurrently, we also manually cloned cells by limiting dilution in 96-well plates. A large number of clones from both single cell and limiting dilution were obtained in addition to the JDF-9, JDF-RC3, and JDF-BB1 reported here (data not shown). The mCherry and EGFP profiles of the various cells is shown in Fig. 1B and 1C.

FIG 1. Isolation of JDF cell lines.

(A) Summary of isolation method for JDF cells. Jurkat cells were transduced with HIVDF , cultured for 2-4 weeks and sorted for the mCherry (mCh) positive/EGFP negative population. Bulk cells were expanded and frozen. Cells were secondarily sorted or subcloned by limiting dilution to obtain additional cell clones. Each cell line was screened for reactivation by treatment with PMA and Ionomycin. (B) Graphic display of the average percentage of latent and active HIV expression for each cell line. Error bars denote SEM. (C) Example plots of DF infected cell lines. Represent plots of mCherry (y-axis) and EGFP (x-axis) expression in selected JDF cell lines. In all experiments, cells were stained with LIVE/DEAD reagent and plots show only cells gated as viable. Approximate location of gates used for gating of secondary sort of JDF-A cell line into ‘A2hi’ and ‘A2lo’ populations is indicated in JDF-A plot.

The integration sites of the JDF cell lines were determined as a means to investigate the clonality of the cell lines. The number of sequenced clones, number of independent integration sites, and whether integration occurred in a gene is summarized for each cell line in Table 1. The JDF-A cell line was confirmed to be polyclonal as 16 independent HIV-1 integration sites were identified, 43% of which were intragenic sites (43). The JDF-A2hi and -A2lo cell lines were also expected to both be polyclonal cell lines. Unexpectedly, only one integration site was identified in the JDF-A2lo cells, and the JDF-A2hi cells contained only two, including the location shared with the JDF-A2lo cells. The JDF-RC3, -BB1, and -9 cell lines, which were isolated by limiting dilution, were also found to have multiple sites of integration.

Table 1.

Summary of HIV-1 integration sites in cell lines.

Cell Line/Clone	# Identified Clones	Indep. sites1	# Intragenic Sites2	Genes3
JDF-A	21	16	9 (43%)	TAS2R43 (3); KMT5A; DHX9; NAA25; SMAD5; AIP; USP48
JDF-A2hi	12	2	8 (67%)	TAS2R43 (5); PACS1 (3)
JDF-A2lo	9	1	9 (100%)	TAS2R43 (9)
JDF-RC3	10	2	10 (100%)	CRKL (6); TOP1 (4)
JDF-BB1	4	4	3 (75%)	CRKL; BANK1; GLIS3
JDF-9	6	2	6 (100%)	HNRNPM (5); MED25
Total	62	27	45 (72%)

¹Number of independent sites out of total.

²Determined by NCBI BLAST search. Percentage of total sequenced clones given in parentheses.

³NCBI Gene acronyms. If multiple clones found, total number given in parentheses.

Reactivation Profiles of JDF cell clones

To serve as useful models of latency, cell clones must have the capability to reactivate HIV expression upon appropriate stimulation. Numerous latency reactivation agents (LRAs) with varying efficacies have been discovered (44, 45). Studies have established that individual models of latency respond differentially to various LRAs (46). To investigate the reactivation potential of the JDF cell lines we measured their response to each of a panel of seven LRAs (Fig. 2A). A heat-map summary of the responses of the cell lines is shown in Fig. 2B. Not all cell lines were responsive to LRA stimulation (e.g. JDF-2), but of those that were, all responded to the PKC activator PMA in combination with ionomycin. The polyclonal cell lines (JDF-A, -B, -A2hi, A2lo) were responsive to most of the LRAs, except bryostatin, and displayed similar responses. This is likely due to the heterogeneity of these cell lines, resulting in an averaging of the divergent responses from individual clones in the population. Higher variability in LRA response was seen in the subcloned lines (JDF-2, -BB1, -9, -RC3), including a loss of sensitivity to some LRAs to which the parental clones were responsive (e.g. JDF-BB1-prostratin; JDF-9-PHA-M, SAHA). Overall, these data suggest that upon subcloning the cell lines become phenotypically distinct and lose the general responsiveness of polyclonal populations.

FIG 2. Resting and reactivation profiles of DF cell lines.

(A) Example reactivation profile of JDF-A cells. Cell lines were assayed for reactivation of latent virus with a panel of known compounds. Cells were treated for 48h with indicated drugs and then analyzed by flow cytometry as described in Methods. Error bars denote SEM. (B) A heat map summary of the reactivation of each cell line listed at the right for each latency reactivation drug listed at the top. Each darker shade of green represents an approximate 20% increase in EGFP expression compared to DMSO treated control. Red squares denote no activation. Data for each cell line is representative of at least three independent experiments with triplicate replicates.

SWATH-MS Analysis of the Membrane Fraction of JDF-A Cells

A cure for HIV will require elimination or durable suppression of the latent cell reservoirs in infected individuals. Latency can occur in different cell types and lineages, therefore targeting of the latent population will likely require comprehensive and flexible methods of detection. One approach to achieve this is to utilize multiple biomarkers to increase the diversity and sensitivity of detection. Currently only a limited number of biomarkers have been proposed. In an effort to expand the palatte of latency biomarkers, we performed a proteomic screen to compare the membrane fractions of uninfected Jurkat cells to the JDF-A cell line to identify differentially expressed proteins. Membrane fractions were isolated using the Qproteome cell compartment kit (Qiagen) which isolates both the plasma membrane and organelle membranes. Lysates of membrane fractions were digested with trypsin and analyzed by SWATH-MS, resulting in the identification of 1468 and 1462 proteins from the Jurkat and JDF-A samples, respectively. A Bayesian statistical pipeline was used to identify proteins differentially expressed between the two groups at high confidence. Candidates were visualized by a volcano plot of Bayesian factor (BF) versus fold-change (FC; Fig. 3A). Candidate proteins were identified as having a BF log₁₀ ≥ 1.5 and a log₂ fold change ≥ 1.0 (Fig. 3A, dashed lines). The top candidates identified through this pipeline are listed in Table 2. A heat map of the candidate factors with the greatest fold change difference between the Jurkat and JDF-A cells is shown in Fig. 3B.

Fig 3. Identification of candidate proteins in the membrane fraction of JDF-A cells.

Volcano plot of SWATH-MS data using Baysian statistical analysis. Candidate proteins are denoted by red dots. (B) Heat map of the fold change in expression of indicated candidate proteins in JDF-A cells compared to WT Jurkat cells across quadruplicate replicates indicated at bottom.

Table 2.

Top Candidates Defined by Bayesian Statistics

Uniprot ID	Gene	Name	Log2 FC1	p-value	BF10
Q13822	ENPP2	Ectonucleotide pyrophosphatase/phosphodiesterase family member 2	5.7	9.02E-05	172.302
Q9Y6A9	SPCS1	Signal peptidase complex subunit 1	2.7	2.78E-05	446.4
Q02543	RPL18A	60S ribosomal protein L18a	0.4	3.11E-05	408.131
Q9GZT6	CCDC90B	Coiled-coil domain-containing protein 90B, mitochondrial	0.4	3.06E-05	413.708
Q99766	DMAC2L	ATP synthase subunit s, mitochondrial	−1.8	0.000169	104.578
Q56VL3	OCIAD2	OCIA domain-containing protein 2	2.59	0.000523	43.394
P55199	ELL	RNA polymerase II elongation factor ELL	2.26	0.000329	62.04
Q13586	STIM1	Stromal interaction molecule 1	1.99	0.000389	54.451
P41732	TSPAN7	Tetraspanin-7	1.25	0.000525	43.229
Q9NQE9	HINT3	Histidine triad nucleotide-binding protein 3	0.95	0.000185	97.221
Q99442	SEC62	Translocation protein SEC62	0.75	0.000847	30.081
014683	TP53I11	Tumor protein p53-inducible protein 11	−1.65	0.000781	31.977
Q9HC52	CBX8	Chromobox protein homolog 8	−1.70	0.000669	35.96
Q9UL25	RAB21	Ras-related protein Rab-21	−3.07	0.000385	54.846
Q9BUE6	ISCA1	Iron-sulfur cluster assembly 1 homolog, mitochondrial	−2.22	0.000862	29.696
P51149	RAB7A	Ras-related protein Rab-7a	−1.40	0.0009	28.751
Q96E39	RBMXL1	RNA binding motif protein, X-linked-like-1	2.33	0.000911	28.481
Q9NUI1	DECR2	Peroxisomal 2,4-dienoyl-CoA reductase	0.97	0.001289	21.984
P36639	NUDT1	7,8-dihydro-8-oxoguanine triphosphatase	2.37	0.001375	20.965
P55327	TPD52	Tumor protein D52

¹Log₂ fold change infected/control.

Validation of the Differential Expression of Candidate Biomarkers

Independent validation is essential for weeding out potential false-positive candidates in any OMICs experiment. Thus, for validation of the differential expression of proteins identified from the proteomics screen on the JDF-A cells, flow cytometric analyses were performed for the cell surface/plasma membrane proteins. Because the membrane fraction contained membrane bound organelles, immunoblotting was performed for intracellular proteins. Candidates were assessed on seven cell lines to determine the breadth of each factor as a biomarker. Antibody recognition of candidate latency biomarkers was quantified by mean fluorescence intensity (MFI) separately for the latent (mCh+/EGFP−) and productively expressing (mCh+/EGFP+) populations (Fig 4). However, it should be noted that in some cases (e.g. JDF-9, -BB1; Fig. 1C), the productive population represented a very small fraction of the total population of the cell line.

Fig 4. Expression of Candidate biomarkers on the surface of JDF cell lines.

Flow cytometry was performed on cell lines indicated on x-axis using antibodies to proteins indicated in gray box of each graph. Cells were gated into latent (light bars) and productively expressing (dark bars) populations as shown in Fig. 1 and mean fluorescence calculated using FlowJo software. Plots show the average mean fluorescence from at least three independent experiments. (*) indicates significance compared to Jurkat (−) control by one-way ANOVA with Dunnett’s multiple comparison test (p<0.05).

The expression of several previously proposed biomarker candidates, including PD-1, LAG-3, PD-1L, and CD32a, was examined to test for concordance of the JDF cell lines with other models of latency. Overall, PD-1 displayed markedly different expression between the latent and productively expressing cell populations in all the cell lines. Consistent with previous data, PD-1 exhibited significantly increased expression in the productively expressing population of all the infected cell lines. However, it was only found to have increased expression in three of the seven JDF cell lines. PD-1L showed increased expression in the productive population in four of the seven cell lines compared to control cells, and was upregulated on latent cells in only two of the seven cell lines. Notably, LAG-3 expression was significantly higher in both the latent and productive populations in all the JDF cell lines compared to the parental Jurkat cells. CD32a, which is controversial as a biomarker of latency, was found to have variable expression among the cell lines. Surprisingly, it showed lower expression in the latent population in six of the seven cell lines, albeit at a statistically significant level in only two. It was only observed to be upregulated in the latent JDF-BB1 cells. In the productive populations, CD32a was upregulated in all the cell lines, but at a statistically significant level in only two of the seven JDF lines.

The cell surface expression of several candidate factors identified in the SWATH-MS analysis was investigated next, including CD317/BST/tetherin, a known restriction factor of HIV that is downregulated by the HIV accessory factor Vpu (47). Consistent with previous data, BST expression tended to be lower in the productive populations of JDF cell lines, although it was only significantly (and substantially) reduced in three of seven cell lines. In the latent JDF cell population, BST showed significant overexpression compared to Jurkat cells in four of the seven cell lines. Notably, the other three JDF lines (JDF-B, -9, -RC3) were the cell lines that had the most significant reduction in BST expression in the productively expressing cell population. A second candidate, CD43, showed no significant change in expression across the cell lines except for the JDF-9 cells, in which it was modestly lower than Jurkat control cells. CD147, a putative receptor for cyclophilins ((48, 49) ; see below), was found to have reduced surface expression in the latent population in six of the cell lines, and the productive population of five cell lines. Similarly, the surface expression of tetraspanin 7 (TSPAN7/CD231) tended to be reduced in both the latent and productively expressing populations among the cell lines; although it was significantly reduced in the latent populations of three cell lines and three productive populations.

The method used to isolate the membrane fractions also contains membrane bound organelles and vesicles. We investigated the expression of non-plasma membrane proteins by immunoblot (Fig. 5). Notably, since the latent and productive populations of each cell line were not separated, the cell lines contained differing levels of productively expressing cells (Fig. 1B). The cytoplasmic, membrane and nuclear fractions were all analyzed for each protein (data not shown). Differential separation of the cytoplasmic and membrane fractions was confirmed by detection of GAPDH in the cytoplasmic fraction but not the membrane fraction, which was also used as a loading control (Fig. 5, top panels). Cyclophilin B, a factor we previously showed to be upregulated during HIV replication (50, 51), was also upregulated in the all JDF cell lines, including the cell lines with >80% latent populations (JDF-9, -BB1 etc., Fig 1C), compared to parental Jurkat cells.

Fig 5. Expression of Candidate biomarkers in the membrane fraction of JDF cell lines.

Subcellular fractions of cell lines indicated at top were immunoblotted with antibodies to candidate biomarkers listed at left (WCL= whole cell lysate control). Cytoplasmic GAPDH was used as loading control (top blot); all other blots are membrane fractions. Fraction integrity was determined by absence of GAPDH (2nd blot). Blots represent 3 independent experiments.

Notably, several novel candidates associated with protein trafficking in cells were identified in the SWATH-MS analysis (Table 2, supplementary data). Of these, Sec62, Rab10, and SPCS1 were found to have differential expression in a number of the JDF cell lines compared to control Jurkat cells. Sec62 is a component of the SEC61 complex that functions in protein translocation in the endoplasmic reticulum (52, 53), and interacts with HIV-1 gp41 (54). It was upregulated in a majority of the JDF cell lines. Rab10 is a member of the RAS superfamily of small GTPases involved in vesicular transport (55). Its overexpression in several of the cell lines is consistent with its presence in Staufen1-containing ribonucleoprotein complexes found in HIV-1 infected cells (56). Signal peptidase complex subunit 1 (SPCS1) showed increased expression in two the JDF cells lines (-BB1, -RC3) compared to Jurkat control cells. SPCS1 is a component of the microsomal signal peptidase complex that participates in protein processing and transport in the endoplasmic reticulum. It is known to be involved in the assembly and budding of flaviviruses (57–59), but its role in HIV-1 replication is unknown.

DISCUSSION

The fundamental question of whether latently infected HIV-1 cells contain alterations in their cellular proteome that can be utilized for selective targeting in vivo remains unresolved. Several candidate biomarkers have been proposed, including (but not limited to) CD32a, CD2, and Burton tyrosine kinase (BTK;(11, 13, 19)), but the detection of these markers may be situational at best (60). Moreover, no factor has been demonstrated to work broadly in vivo. In this study we confirm the differential expression of cellular proteins in latently infected cells in vitro using cell lines. To determine the ability of each candidate to broadly detect latency, we assessed the differential expression of each in a panel of cell lines. Overall, most candidates displayed variable expression across different cell lines. Three proteins were consistent across our panel of cell lines: LAG-3 was upregulated in the latent populations of all the cell lines; whereas CD147/Basigin and CD231/TSPAN7 were down-regulated.

Our novel biomarker screen also discovered two cellular pathways/families that appear to be dysregulated during latent HIV infection. First, was CypB and CD147. CypB is a factor we showed to be upregulated during active HIV infection (50, 51). CD147 is a receptor for cyclophilins, including CypB (49, 61), and may enhance HIV infection through its interaction with cyclophilin A (61). Among our cell lines, CypB was upregulated whereas CD147 expression was reduced on the surface of cells. Additional studies will be necessary to determine if these changes in expression are significant and what, if any, role they play in latency. The second theme that caught our attention was the upregulation of intracellular processing and translocation factors, including Sec62 and SPSC1. It is well established that HIV infection alters cellular homeostasis and usurps the vesicular secretion machinery (62, 63). It will be interesting in future studies to determine if these processes remain disrupted during latency and whether or not it can be exploited for therapeutic intervention.

Here we hypothesized that the use of cell line populations that harbor heterogeneously integrated proviruses would provide an improved in vitro model of HIV-1 latency versus single cell clones. We isolated a number of cell lines with varying levels of clonality. The cell lines also show differing latency phenotypes and sensitivity to latency reactivation agents. It was our hope that proteomic analysis of one of the heterogeneous lines (JDF-A) would identify biomarkers with broad expression across latently infected cells. Overall, most of the identified factors displayed variable expression across the cell lines. The most ubiquitous biomarkers in the cell lines were LAG-3, CD147, CD231/TSPAN7, and CypB. Notably, only LAG-3 was found to be overexpressed in the latent population of all the cell lines. CypB was also overexpressed in the majority of the cell lines, but those samples included both latent and actively expressing cells. The cell surface expression of both CD147 and CD231 were lower in the latent populations of the infected cell lines. Investigation of factors and pathways associated with these candidates may yield additional potential biomarkers of latency. Future studies will seek to do this and determine the breadth of expression of candidate biomarkers using several different primary cell models of latency, as well as in vivo in ART suppressed patients.

The utility of cell line models to identify biomarkers of latency has been philosophically questioned (64). Prior to the initiation of our studies, all of the published primary cell models of latency reported only ~0.5 – 5% of cells capable of reactivation (10, 11, 47), and no tenable method of enrichment without reactivation. A recent study reports higher recovery of latently infected cells, but involves post infection expansion, cell sorting, and induction of quiescence via cytokine treatment (65). Cell lines are advantageous as they provide reproducibly high levels of reactivation, a limitless supply, and have less genetic variability than cells derived from human donors. Reducing variability at the first stage of a proteomics project is critical for the identification of unique biomarkers that may be present in low abundance. The advantage of cells lines has been demonstrated throughout the history of HIV research as cell lines have been used as tools to identify many factors, including CD4 (66, 67), cyclinT1 (68), CXCR4 (69), CCR5 (70), TRIM5α (71), and APOBEC3G (72).

Finally, another major challenge with studying latency in vitro is that no single cell line likely mimics the diversity of latency establishment and response in vivo. Albeit, the same could be argued for latently infected primary cells as each method produces different reactivation profiles (46). Patient cells and primary cell models also do not model the multiple cell lineages in which latency is observed in vivo. Moreover, primary cell models vary in methodologies, including cell source/type, type of virus (live, or vector-based), method of infection (activated versus resting cells, spinoculation), and treatment with varying cytokines. It is telling that although a large number of primary cell models of latency have been published, but no model has become universally accepted or put into use. Our results corroborating differential expression of several known factors involved in HIV latency and persistence validates the use of cell line models as an initial step in a reductionist approach toward biomarker identification, as well as justify expanded investigation of candidates in more authentic ex vivo and in vivo models of latency.

Highlights.

• Elimination of latently infected cells is necessary to cure HIV-1 infection.

• SWATH mass spectrometry was used to identify biomarkers of HIV-1 latency using a cell line T cell model.

• Several candidates were validated in multiple latently infected cell lines.

• The expression of most biomarker candidates was variable across cell lines, indicating a multiplex method may be necessary for broad detection of latently infected cells.