Articles Plasma proteomic profiling identifies CD33 as a marker of HIV control in natural infection and after therapeutic vaccination Clara Duran-Castells,a,b Anna Prats,c Bruna Oriol-Tordera,a,b Anuska Llano,a Cristina Galvez,a Javier Martinez-Picado,a,d,f ,g Ester Ballana,a Edurne Garcia-Vidal,a Bonaventura Clotet,a,c,d Jose A. Muñoz-Moreno,c,e Thomas Hanke,h,i José Moltó,c,g Beatriz Mothe,a,c,d,g Christian Brander,a,b,d,f ,g and Marta Ruiz-Riola,g,∗ a IrsiCaixa AIDS Research Institute Badalona, Hospital Universitari Germans Trias i Pujol, Badalona, Spain b Universitat Autònoma de Barcelona, Spain c Fight Infections Foundation and Infectious Diseases Department, Hospital Universitari Germans Trias i Pujol, Badalona, Spain d University of Vic-Central University of Catalonia (UVic-UCC), Vic, Spain e Faculty of Psychology and Education Sciences, Universitat Oberta de Catalunya (UOC), Barcelona, Spain f Catalan Institution for Research Advanced Studies (ICREA), Barcelona, Spain g CIBER de Enfermedades Infecciosas (CIBERINFEC), Madrid, Spain h The Jenner Institute, University of Oxford, Oxford OX3 7DQ, UK i Joint Research Center for Human Retrovirus Infection, Kumamoto University, Kumamoto, Japan Summary eBioMedicine 2023;95: 104732 Background Biomarkers predicting the outcome of HIV-1 virus control in natural infection and after therapeutic interventions in HIV-1 cure trials remain poorly defined. The BCN02 trial (NCT02616874), combined a T-cell Published Online xxx https://doi.org/10. vaccine with romidepsin (RMD), a cancer-drug that was used to promote HIV-1 latency reversal and which has 1016/j.ebiom.2023. also been shown to have beneficial effects on neurofunction. We conducted longitudinal plasma proteomics 104732 analyses in trial participants to define biomarkers associated with virus control during monitored antiretroviral pause (MAP) and to identify novel therapeutic targets that can improve future cure strategies. Methods BCN02 was a phase I, open-label, single-arm clinical trial in early-treated, HIV infected individuals. Longitudinal plasma proteomes were analyzed in 11 BCN02 participants, including 8 participants that showed a rapid HIV-1 plasma rebound during a monitored antiretroviral pause (MAP-NC, ‘non-controllers’) and 3 that remained off ART with sustained plasma viremia <2000 copies/ml (MAP-C, ‘controllers’). Inflammatory and neurological proteomes in plasma were evaluated and integration data analysis (viral and neurocognitive parameters) was performed. Validation studies were conducted in a cohort of untreated HIV-1+ individuals (n = 96) and in vitro viral replication assays using an anti-CD33 antibody were used for functional validation. Findings Inflammatory plasma proteomes in BCN02 participants showed marked longitudinal alterations. Strong proteome differences were also observed between MAP-C and MAP-NC, including in baseline timepoints. CD33/ Siglec-3 was the unique plasma marker with the ability to discriminate between MAPC-C and MAP-NC at all study timepoints and showed positive correlations with viral parameters. Analyses in an untreated cohort of PLWH confirmed the positive correlation between viral parameters and CD33 plasma levels, as well as PBMC gene expression. Finally, adding an anti-CD33 antibody to in vitro virus cultures significantly reduced HIV-1 replication and proviral levels in T cells and macrophages. Interpretation This study indicates that CD33/Siglec-3 may serve as a predictor of HIV-1 control and as potential therapeutic tool to improve future cure strategies. Funding Spanish Science and Innovation Ministry (SAF2017-89726-R and PID2020-119710RB-I00), NIH (P01- AI131568), European Commission (GA101057548) and a Grifols research agreement. Copyright © 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Keywords: Control of HIV-1 infection; Plasma proteomics; Inflammation; Neurological function; Siglec-3/CD33; Kick and kill HIV cure strategy *Corresponding author. IrsiCaixa AIDS Research Institute Badalona, Hospital Universitari Germans Trias i Pujol, Badalona, Spain. E-mail address: mruiz@irsicaixa.es (M. Ruiz-Riol). www.thelancet.com Vol 95 September, 2023 1 Articles Research in context Evidence before this study soluble plasma marker that can discriminate MAP-C from Strategies for an effective HIV cure are complicated by the MAP-NC already at baseline time points before RMD and ability of the virus to establish latent reservoirs in infected vaccine administration. Validation analyses in an unrelated cells. The BCN02 clinical study (NCT02616874) used a “kick cohort of chronically HIV-1+ individuals and in vitro and kill” strategy, combining the latency reversing agent experiments suggest that CD33/Siglec-3 may play an (RMD) and a therapeutic HIV-1 vaccine with the aim to important role in the HIV-1 life cycle and in vivo viral control reactivate the viral reservoir and trigger effective antiviral and could be employed therapeutically. immune responses. During the subsequent treatment Implications of all the available evidence interruption phase, 32% of the participants were able to Our study identifies for the first time CD33/Siglec-3 as a control viral replication to levels below 2000 viral copies/ml. biological marker of HIV-1 control in the absence of For further refinement of cure strategies and to increase the antiretroviral treatment and points towards its critical safety of such trials, there is an urgent need to identify non- involvement HIV-1 infection. Other Siglec family members invasive biomarkers that can predict viral control before ART have been reported to be responsible for the capture and interruption. The definition of such markers may also identify storage of pathogens by mature dendritic cells and for trans- novel therapeutic targets that can improve future cure infection between infected dendritic cells and target cells strategies. In this clinical setting, CD33/Siglec-3 emerged as a (e.g., Siglec-1 and Siglec-7). Here we identify plasma CD33/ biomarker for predicting whether individuals will control or Siglec-3 levels as a biomarker associated with HIV-1 control not control viremia during ART interruption, even when during the clinical BCN02 trial but also in a large cohort of assessed in samples drawn at the baseline timepoint. Aside individuals with chronic, untreated HIV-1 infection. Moreover, from defining CD33/Siglec-3 as an early predictor of virus addition of anti-CD33 monoclonal antibodies to in vitro virus control post treatment stop, our in vitro targeting cultures reduced HIV-1 replication, suggesting that CD33 may experiments also support the use specific antibodies directed be required for HIV-1 replication. The use of CD33 monoclonal against CD33/Siglec-3 in future therapeutic trials. Since CD33 antibodies was previously studied in acute myeloid leukemia targeting has also found entry in the treatment of and in hepatitis B infection. Interestingly, several neurocognitive disorders, such an intervention in HIV infected interventions targeting CD33 are also tested as potential individual may also help to stem against the cognitive therapies for Alzheimer’s disease (AD), as reduced cell surface impairments observed in HIV infection. CD33 levels allowed more efficient phagocytic clearance of Added value of this study pathogenic Amyloid beta (Aβ) and provide protection against In the present study, we observed that the administration of disease. Our data thus suggest that further explorations and an HIV-1 therapeutic vaccine together with the latency studies of the CD33/Siglec-3 protein could have significant reversing agent RMD has an impact on the plasma proteomic implications for HIV-1 cure and also for the understanding profiles, including inflammatory and neurological markers. In and potential treatment of HIV-1 associated neurocognitive particular, we identify the CD33 protein, also known as sialic disorders. acid binding immunoglobulin-like lectin 3 (Siglec-3), as a Introduction Reversal Agent (LRA) romidepsin (RMD), has previ- Since its identification in the 1980’s, the human im- ously been associated with beneficial effects on the munodeficiency virus HIV type 1 (HIV-1) represents a central nervous system through the regulation of the major global public health issue. Although combination synaptic plasticity.2 antiretroviral therapy (ART) effectively suppresses HIV- BCN02 was a single-arm, open-label clinical trial 1 replication and slows disease progression, there is no (NCT02616874) that enrolled 15 ART-suppressed early- effective cure for HIV-1 infection. Moreover, ART is treated, HIV-1+ individuals to receive a combination of associated with viral resistance and treatment-related 3 series of the LRA RMD and MVA.HIVconsv vac- side effects and, even more importantly, is not acces- cines.3,4 The HIVconsv immunogen is a chimeric pro- sible worldwide, reinforcing the urgent need for a HIV tein sequence assembled from 14 highly conserved cure. During the last decade, different cure strategies domains derived in HIV-1 Gag, Pol, Vif, and Env.3–5 On have been explored, some of them based on latency the other hand, RMD has shown increased histone-3 reactivators, such as histone deacetylase inhibitors, acetylation levels causing a relaxation of the host chro- which make host-cell chromatin more accessible to matin and consequently favoring reactivation of latent transcription factors. However their use can also impact HIV-1 virus in ex vivo experiments.6 Importantly, the host gene expression,1 which in the case of the Latency BCN02 clinical trial also included a monitored 2 www.thelancet.com Vol 95 September, 2023 Articles antiretroviral pause (MAP) for a period of 32 weeks to These longitudinal inflammatory/neurological profiles evaluate the clinical efficacy of the intervention. Of 11 uniquely identified CD33/sialic-acid-binding immuno- individuals evaluated during MAP, 8 individuals globulin-like lectin 3 (Siglec-3), whose elevated plasma reached premature ART-restart criteria (MAP non con- levels were associated with MAP non-control. Higher trollers, MAP-NC) while 3 participants were able to CD33/Siglec-3 plasma levels were also observed in in- control the virus throughout the entire 32 weeks MAP dividuals with high viral loads in an unrelated cohort period to levels below 2000 HIV-1 copies/ml (MAP with untreated HIV-1 infection. The clinical relevance of controllers, MAP-C).3 Interestingly, the outcome in these findings is further supported by our observation BCN02 has also been linked to specific signatures in the that anti-CD33 monoclonal antibody in vitro blocked host microbiome and the host gene epigenetic land- virus replication and integration, thus identifying CD33 scape, suggesting that possibly pre-existing particular- as a potentially therapeutic target in HIV-1 infection. ities among HIV-1+ individuals may condition the success of the such therapeutic interventions.7,8 Although the BCN02 clinical trial included early Methods treated individuals who initiated ART less than 6 Patients and samples months after infection and who therefore may present The BCN02 clinical trial (NCT02616874) was a phase I, with smaller and less-diverse viral reservoirs, the virus open-label, single-arm, multicenter study conducted in systematically rebounds from this reservoir and is the Barcelona (Spain).3 Fifteen HIV.consv vaccine recipients major challenge for cure strategies. While the contri- were rolled-over from the previous BCN01 trial,16 and bution of different anatomical compartments of the viral were immunized with MVA.HIVconsv followed by reservoirs to rebound remains unclear, it is known that three weekly-doses of RMD (5 mg/m2) and a second after acquisition, HIV-1 also infects the central nervous MVA.HIVconsv vaccination, before undergoing MAP system (CNS) and represents one of the anatomical for a maximum of 32 weeks (Supplementary Figure S1). compartments for latently infected cells.9 Infection of During MAP, the participants restarted ART treatment the CNS has also been related to the occurrence of HIV- if they reached the threshold of 2000 HIV-1 RNA 1 associated neurological disorders (HAND)10 and copies/ml. Available plasma samples from 11 partici- eradication of this reservoir may pose specific chal- pants were used for the longitudinal proteomic profiling lenges for LRA drug penetrability to the CNS. Interest- analysis, including 8 MAP-NC and 3 MAP-C who ingly, in animal studies the LRA RMD protected CNS completed the 32 weeks of MAP (Supplementary through remodeling chromatin and regulating synaptic Table S1). Chronic untreated HIV-1+ individuals plasticity.11 RMD administration was also explored in (n = 96, Supplementary Table S2), enrolled at the Alzheimer’s disease (AD), where it increased the gene IMPACTA clinics (Peru) and at Hospital Germans Trias expression of neurotrophic factors and reduced AD- i Pujol (Spain), were classified according to their degree associated biochemical and cellular changes.2 There- of control of viral replication; HIV-high n = 47, >50,000 fore, the inclusion of RMD in the BCN02 trial offered HIV-1 RNA copies/ml and HIV-low n = 49, <10,000 the opportunity to assess its effect on virus reactivation HIV-1 RNA copies/ml. The HIV high had a median and neurological impact, including cognitive status, CD4 count of 303 cells/mm3 (range 11–729 cells/mm3) functional outcomes, and brain imaging evaluations in while the HIV low had a median of 712 cells/mm3 the trial participants.12 Although no alterations in neu- (range 289–1343 cells/mm3). Blood samples from HIV rocognitive or functional status and brain imaging have uninfected donors from the Banc de Sang i Teixits in previously been shown in this study cohort,12 subjacent Barcelona were used for in vitro studies. brain injury in asymptomatic forms cannot be dis- carded. Relevant to the present study, such subclinical Plasma proteomic profiling events may be detectable in plasma or CSF fluids as BCN02 plasma samples were used for evaluation of shown in proteomics studies in HIV-1+ individuals inflammation, neurology and neuro-exploratory cyto- suffering HAND, which have identified aberrations in kine panels applying PEA (Proximity Extension Assays) several inflammatory and neurological damage by Olink® (https://www.olink.com). Briefly, a pair of markers.13–15 oligonucleotide-labeled with specific antibodies binds to To further explore the impact of the therapeutic the protein present in sample, bringing the two oligo- vaccination and LRA RMD on plasma markers of virus nucleotide probes in close proximity and allowing for a control and CNS integrity, longitudinally plasma prote- specific sequence to be formed by DNA polymerization. ome profiling was conducted in the BCN02 study sam- The resulting sequence is detected and quantified using ples. Overall, RMD administration and ART standard real-time PCR. Protein concentration was interruption strongly increased the protein levels of host expressed as relative expression levels, using an arbi- factors involved in inflammatory processes but also that trary unit (normalized protein expression, NPX) on a of markers of neurological disorders that all correlated Log2 scale. A high NPX value corresponds to a high with clinical, viral, as well as cognitive parameters. protein concentration (Supplementary Table S3). For www.thelancet.com Vol 95 September, 2023 3 Articles validation of relative CD33/Siglec-3 plasma levels in Sigma-Aldrich) and 10 U/ml IL-2 (Roche). After 3 days, untreated HIV-1+ individuals, proteomic analysis were PHA-blasts were infected with the HIVNL4-3 strain performed using customized arrays as previously (Multiplicity of Infection (MOI): 0.01). For monocyte- described.17 derived macrophages (MDMs), PBMCs were depleted using the EasySep™ Human Monocyte Enrichment Kit GEO access of BCN02 transcriptomics (Stem Cell). Monocytes were then incubated with CD33/Siglec-3 gene expression levels in total PBMC macrophage colony-stimulating factor (M-CSF, R&D were obtained from previously published work.8 The systems S.L. #216-MC) at 1 μg/ml for 4 days before used dataset is available in Gene Expression Omnibus infection with the HIVBaL strain (MOI: 0.01). HIV-1 (GEO), under the accession number GSE184653. The infection in PHA-blasts and MDMs was evaluated un- specific values for CD33/Siglec-3 were obtained after der different conditions: Zidovudine (AZT, 200 μg/ml) log2 transformation of the TMM normalized counts. was used as positive control of viral replication inhibition. The anti-CD33 mAb (Rabbit monoclonal [EPR4423], Evaluation of CNS functioning Abcam #ab134115) was used at concentrations ranging Neuropsychological evaluation covered 6 cognitive do- from 0.004 μg/μl, to 0.01 μg/μl. After 3 (PHA stimulated mains to provide a global composite score (NPZ-6), T cell blast) and 4 days (MDM), p24 in the culture su- which was recorded as recently published.12 In brief, pernatant was quantified by ELISA (INNOTEST HIV p24 NPZ6 included: Digit Test of the Wechsler Adult In- Antigen mAb, Fujirebio, #80563). Total HIV-1 DNA telligence Scale (WAIS-IV); the Trail Making Test (TMT- quantification was done on total DNA isolated from A); Grooved Pegboard Test (GPT); California Verbal cultured cell using the RNA/DNA purification Micro kit Learning Test (CVLT-II); Controlled Oral Word Test (Norgen Biotek Corp., #48700) as recommended by the (COWAT); and the Trail Making Test (TMT-B). manufacturer. Quantification of total HIV-1 DNA was determined by a quantitative PCR assay, using TaqMan Determination of integrated HIV-1 proviral DNA Universal Master Mix II (Applied Biosystems) in a 7500 HIV-1 proviral DNA was quantified in total PBMC real-time PCR system (Applied Biosystems). We used samples of untreated HIV-1+ individuals and in isolated the following set of primers and probe; forward primer: CD4+ T cells of BCN02 clinical trial participants by 5′ -GACGCAGGACTCGGCTTG-3′ and reverse primer: droplet digital PCR (ddPCR) in duplicates, as previously 5′ -ACTGACGCTCTCGCACCC-3′ and probe: 5′ -fluores- reported.18 Briefly, two different primer/probe sets cein amidite (FAM)-TTTGGCGTACTCACCAGTCGCC annealing to the 5′ long terminal repeat and Gag re- G-6-carboxytetramethylrhodamine (TAMRA)-3′ . GAPDH gions, respectively, were used to circumvent sequence gene (Hs02786624_g1) was used as the housekeeping mismatch in the patient proviruses, and the RPP30 gene. Gene amplification was performed in an Applied housekeeping gene was quantified in parallel to Biosystems 7500 Fast Real-Time PCR System thermo- normalize sample input. Raw ddPCR data were cycler, and the relative expression quantification was analyzed using the QX100™ droplet reader and Quan- calculated as 2−ΔCT (where CT is the median threshold taSoft v.1.6 software (Bio-Rad). cycle from 2 replicates). Real-time PCR CD33 gene STRING pathways RNA was isolated from available PBMC dry-pellets For Protein-Protein Interaction Networks, functional from participants in the untreated HIV-1 infection cohort enrichment analysis was used to analyze interaction and was retrotranscribed. TaqMan gene expression between proteins that showed significantly different assay (Applied Biosystems) was used for CD33 concentrations between time points and between MAP- (Hs01076281_m1) detection. TBP gene (Hs99999910_m1) C and MAP-NC groups (https://string-db.org/cgi/input? was used as the housekeeping gene. Gene amplification sessionId=bcCYfoHsbNV0&input_page_active_form= was performed in Applied Biosystems 7500 Fast Real- multiple_identifiers). Specifically, for GO categories Time PCR System thermocycler, and the relative expres- classification, counts in the network indicate how many sion quantification was calculated as 2−ΔCT (where CT is proteins in the analysis are annotated in a particular GO the median threshold cycle from 3 replicates). CD33 gene term and are added to the graph. expression was corrected by CD4 T cells count since this cohort of HIV-high and HIV-low individuals showed sig- Statistical analysis nificant differences in the group-specific median CD4 T Univariate and multiple comparisons statistical analyses cell counts (Supplementary Table S2). were performed using GraphPad Prism, version 8. For comparisons between patient groups, the Mann– HIV-1 infection of PHA-blasts and monocyte- Whitney and Wilcoxon signed rank tests were applied derived macrophages and for multiple group comparisons One-Way ANOVA Isolated PBMCs from HIV-1 uninfected donors were test corrected for multiple comparisons with original stimulated with 5 μg/ml Phytohaemagglutinin (PHA, False Discovery Rate (FDR) method using Benjamini 4 www.thelancet.com Vol 95 September, 2023 Articles and Hochberg method, were used. The Spearman and When comparing baseline vs MAP timepoints, 47 Pearson test were applied for the correlation analysis proteins showed different relative protein levels, with 37 with non-parametric and parametric data, respectively. showing and increase (Fig. 1b Profile III. in red, and For all analyses, p-values <0.05 were considered statis- Supplementary Table S4) and being mostly related with tically significant. Principal Component Analysis “cell growth”, “proliferation” or “cell structure” such as (PCAs), Heatmaps, Venn diagrams and correlograms TBCB, PFDN2, 4E-BP1, NPM1, FGF23, CD63, among were performed with RStudio version 1.2.5042. others. The 10 proteins that decreased their relative levels over this time frame (Fig. 1b Profile III. in green Ethics and Supplementary Table S4) were related to “response to The study was approved by the Comité Ètic d’Investi- cytokine” category like IL20, IL22RA1, LIF, GBP2, gació Clínica of Hospital Germans Trias i Pujol (CEIC: SCGB1A1 and NEP and “regulation of epithelial cell dif- EO-12-042 and PI-18-183) and all participants provided ferentiation” such as GDNF, LIF and IL20. their written informed consent. All the research In addition, the 29 proteins that showed significant involving human research participants was performed changes between both, baseline vs post-RMD and in accordance with the Declaration of Helsinki. baseline vs MAP comparisons (Supplementary Table S4), were progressively and significantly Role of the funding source increased (24 molecules) or decreased (5 proteins) over- The funders of the study did not have any role in the time. Most of the increased 24 proteins (Fig. 1b Profile design, data collection, data analyses, interpretation, or II. In red, Supplementary Table S4) were involved in writing of the study. “neuronal and axonal functions” like NEFL, MANF and CLSTN1, while the other 5 molecules that decreased included MCP3, IL5, TNF, CEACAM3 and TDGF1, Results Fig. 1b Profile II. in green, Supplementary Table S4). Plasma proteomes are impacted upon RMD To identify potential drivers of the observed changes administration in protein levels from baseline to post-RMD and/or Focused inflammatory and neurological-tailored pro- MAP timepoints, a correlation analysis was performed teomes were defined using PEA (Proximity Extension between relative protein levels and virological (plasma Assay) in plasmas samples from 11 BCN02 participants viral load (pVL) and proviral levels). Correlation anal- at baseline (BSL, week 0), 1 week after 3 infusions of ysis also include neurological parameters (NPZ6 RMD (post-RMD, week 6) and MAP timepoints (after (Fig. 1c and d) since RMD treatment has been reported MVA vaccination and while off ART; Supplementary to affect these parameters. The analyses showed that Tables S1 and S3 and Supplementary Figure S1), to after RMD infusions, 34% of plasma proteins corre- identify soluble markers modulated by RMD and ther- lated positively and 66% molecules negatively corre- apeutic vaccination. Principal Component Analysis lated with pVL. During MAP timepoint these (PCA) was based on 276 soluble factors involved in percentages changed as now 72% of the proteins inflammation and neurological processes and which positively associated and 28% negatively associated were assessed across the three timepoints of the clinical with pVL (Fig. 1c and d and p = 0.0001, chi squared trial. As shown in Fig. 1a, marked changes were test). In contrast, at post-RMD timepoint the 72% observed after RMD treatment and were intensified percent of the plasma proteins that were differentially during MAP. When compared to BSL, the administra- detected between timepoints were positively associated tion of RMD modulated the plasma levels of 49 proteins, (72%) and only the 28% negatively associated with while 76 molecules were modulated during MAP levels proviral levels, while during MAP phase these per- (Fig. 1b, Supplementary Table S4). Of the 49 proteins centages were partially inverted to 31% positively and that changed significantly upon RMD treatment, 29 69% negatively associated with proviral levels (Fig. 1c were also modulated during MAP. Proteome profiling and d p = 0.0001, chi squared test). analysis of the 20 proteins differentially detected only These data suggest that some proteome profiles at between BSL and RMD administration timepoints, post-RMD are likely reflecting the levels of integrated showed 17 molecules with an increase in relative protein virus as the participants are on cART treatment, while levels (Fig. 1b Profile I. in red and Supplementary during MAP, the observed changes mostly reflect the Table S4). Most of these proteins were related to “cyto- increased viral replication and the activation of the kine release by the interaction with viral proteins” such as “interferon signaling pathway”, with proteins such as CCL25, CXCL5, IL20RA and IL15 and “innate and IL15, CXCL10, IFI30, TNF, IL3, IL12, IL20 being more cellular adaptive immune response markers” like CD8A abundant. and CD38. Three proteins showed decreased levels after Interestingly, correlation analysis with neurological RMD administration, including CXCL10, CCL23 and parameters (NPZ6) showed that 71% of the molecules GDNF (Fig. 1b Profile I in green and Supplementary correlated positively with NPZ6 neurocognitive test Table S4). after RMD treatment and during MAP this proportion www.thelancet.com Vol 95 September, 2023 5 Articles Fig. 1: Changes in plasma proteomes during the BCN02 clinical trial. Soluble proteomes from 11 participants in the BCN02 trial at baseline (BSL), 1 week after 3 infusions of RMD (post-RMD) and during monitored antiretroviral pause (MAP) timepoints were determined using the proximity extension assay. a) Principal Component Analysis (PCA) reflecting the changes of 276 inflammatory and neurological-related plasma proteins during the BCN02 clinical trial at each tested timepoint. The BSL timepoint are is indicated in green, post-RMD timepoint in blue and 6 www.thelancet.com Vol 95 September, 2023 Articles rose to 85% (p = 0.0083, chi squared test), reflecting the the involvement of “NF-KB and MAPK cascades” were higher number of proteins that was associated with the major ontologies differentially represented between neurological parameters (such as MANF, GDNF, NEFL both groups at the MAP timepoint (Fig. 2f). Although GPNMB (Fig. 1c) at the completion of the intervention some proteins involved in neurological processes were and MAP. significantly altered by the intervention, the two groups (MAP-C and MAP-NC) did not show any distinguish- Plasma proteomes at baseline and after RMD able features in their neurological protein profile at any differentiate eventual MAP-controllers and MAP- of the tested time points. non-controllers As plasma proteomes were associated with viral loads CD33/Siglec-3 is the unique discriminative plasma and proviral levels, we next assessed whether plasma factor between MAP-C and MAP-NC across BCN02 proteome signatures at each timepoint (baseline, post- trial RMD and MAP) would have the power to differentiate When analyzing the proteome profiles discriminating individuals who can or cannot control viral replication MAP-C and MAP-NC, we observed that CD33/Siglec-3 during MAP. To that end, we compared the plasma was the unique protein consistently differentially detec- proteomes of MAP-NC individuals (>2000 HIV-1 RNA ted between the two groups across all three timepoints copies/ml and thus restarting ART) and MAP-C (>32 tested (Fig. 3a). Moreover, the plasma levels of CD33 weeks <2000 HIV-1 RNA copies/ml) at all three time- were significantly changed during the intervention points tested (baseline, post-RMD and MAP). PCA plots (Fig. 3b, BSL vs post-RMD p = 0.001 and BSL vs MAP showed that proteome profiles at all timepoints, p = 0.002, Wilcoxon test) and, despite the small number including baseline, segregated the two groups of participants in the trial, were significantly elevated in (Supplementary Figure S2a–S2c). Specifically, the MAP-NC compared with MAP-C in all three timepoints comparative analysis between MAP-C and MAP-NC by (Fig. 3c, BSL: p = 0.024, post-RMD: p = 0.049 and MAP: timepoints showed that differences in the proteome of p = 0.033, Mann–Whitney test). In addition, CD33 pro- both groups were observed at baseline (18 proteins, tein plasma levels positively correlated with proviral copy Fig. 2a), post-RMD (16 molecules, Fig. 2b) and, with numbers at baseline and after RMD administration broader profiles, during MAP (33 molecules, Fig. 2c). (Fig. 3d and e, BSL rho = 0.646 p = 0.037, post-RMD To gain insights into the differences at biological rho = 0.647 p = 0.035, Spearman rank test). A trend to- level between the observed signals at each timepoint, wards a positive association with viral reservoir was also Gene Ontology (GO) classification was performed observed at MAP timepoint (Fig. 3f, MAP rho = 0.9 but (Fig. 2d–f). Aside from the most represented with p = 0.083 (Spearman rank test) not reaching sig- prostaglandin-related ontology, the analysis at baseline nificance, likely since proviral assessments were missing indicated a differential representation of genes involved for more than half of the participants at the MAP time- in “T cell differentiation”, “lymphocyte immunity”, “inter- point. Still, MAP samples showed also a positive corre- feron-gamma production” and “myeloid cell development” lation between CD33 protein levels and pVL (Fig. 3g, ontologies (Fig. 2d). After RMD administration, the rho = 0.782 p = 0.011, Spearman rank test). To investigate major gene ontologies differentially represented be- whether cells in the peripheral blood are the major source tween MAP-C and MAP-NC included “interferon path- of CD33 proteins detected in the plasma, we assessed ways” and “negative regulation of IL5 and IL13 production” CD33 gene expression levels in PBMCs. The positive ontologies (Fig. 2e). “Innate cellular (Mast cells, Neutro- correlation between CD33 expression and plasma viral phils and Monocytes)”, “T cell immunity” categories and load levels (Supplementary Figure S3b, Spearman rank MAP timepoint in red, each point representing one participant. b) Venn diagram representing differentially detected proteins when comparing BSL vs post-RMD (n = 49) and BSL vs MAP phase timepoints (n = 76). The different profiles indicate relative abundance of proteins that were up- or down-regulated: Profile I: Significant changes only detected when comparing BSL vs post-RMD infusions (n = 20); Profile II: Proteins that show significant changes when comparing BSL vs post-RMD as well as BSL vs MAP timepoints (n = 29); and Profile III: Significant changes only detected when comparing BSL vs MAP phase (n = 47). The x-axis marks the timepoints of the clinical trial and on the y-axis, the relative levels of the proteins, represented as the mean of NPX values for each protein. Longitudinal changes between BSL vs post-RMD and BSL vs MAP phase timepoints were assessed using Wilcoxon signed rank test. c) Heatmap showing the relative protein levels (NPX) of all significantly modulated proteins from (b) (p < 0.05, n = 96) per patient and timepoint (orange-yellow scale). Each column represents a patient measured at BSL, post- RMD and MAP. Additional columns (blue-pink scale, right of the heatmap) indicate the value of correlations (Spearman’s rho) between in- dividual proteins and either plasma viral load, proviral levels or neurotest score (NPZ6). The Spearman’s rank test was used for correlation analyses. d) Pie charts representing the percentages of number of proteins that are negatively or positively correlated with viral parameters (viral load and proviral levels, respectively) and with neurological parameters (NPZ6 evaluation matrix). Blue (post-RMD), purple (MAP), dark color indicates positive correlations and light color negative correlations. The Chi-square test was applied for categorizing positive or negative correlation. Statistical significance was set at p < 0.05. www.thelancet.com Vol 95 September, 2023 7 Articles Fig. 2: Differential plasma proteomes between MAP-C and MAP-NC. a–c) Heatmap of relative protein plasma levels (NPX) of proteins differentially detected between MAP-C and MAP-NC at (a) baseline, (b) post-RMD and (c) the MAP phase of the BCN02 clinical trial. Orange- yellow scale indicates NPX values. Green represents MAP-C and blue MAP-NC participants. d–f) Gene Ontology (GO) classification of proteins that differed significantly between MAP-C and MAP-NC at each timepoint. On the y-axis, gene ontology categories are indicated, on the upper x-axis, counts in network are noted to indicate how many proteins in our network are annotated with a particular term (GO, represented as percentage in histogram) and on the lower x-axis, as continuous line, the false discovery rate (FDR) representation. test), is indeed suggestive of PBMCs be a major source of changed during the intervention. After RMD adminis- the CD33 protein in plasma (Supplementary Figure S3a, tration, CD33 protein levels were strongly associated with Wilcoxon rank test). 17 proteins, most pronouncedly with ADAM15, GGT5 To gain further inside into the mechanisms by which and NEFL (Fig. 3h, ADAM15 rho = 0.806 p = 0.0027, CD33 could influence viral control, we performed a cor- GGT5 rho = 0.7413 p = 0.009 and NEFL rho = 0.7064 relation analysis between CD33 and other proteins that p = 0.0151, Spearman rank test). These associations were were included in the inflammatory, neurology or neuro- not only maintained but statistically even more pro- exploratory cytokine panels and which significantly nounced at the MAP timepoint (Fig. 3i, ADAM15 8 www.thelancet.com Vol 95 September, 2023 Articles Fig. 3: CD33 plasma protein levels differentiate MAP-C and MAP-NC across the clinical trial. a) Venn diagram showing differentially detected proteins between MAP-C and MAP-NC in samples from each timepoint of the clinical trial (BSL, post-RMD and MAP). b) Longitudinal representation of the relative plasma CD33 levels (NPX, y-axis) at each timepoint (x-axis). Values are shown as mean and standard error. Longitudinal changes over time were assessed using Wilcoxon signed rank test. c) Cross-sectional analysis of the relative CD33 plasma levels in MAP-C (green color) and MAP-NC (blue color) individuals over the duration of the clinical trial. Clinical trial timepoints are shown on the x-axis, and relative plasma levels of CD33 shown on the y-axis. Values are shown as median and max and min. Group changes per timepoint were assessed using the Mann–Whitney U test. d–f) Correlation analysis between relative CD33 protein levels (NPX, x-axis) vs HIV proviral DNA levels (HIV DNA copies/106 PBMCs, y-axis) are shown on the y-axis in each timepoint of the study including (d) baseline, (e) post-RMD and (f) MAP phase. The Spearman’s rank test was used for correlation analyses. g) Correlation analysis showing the association between relative CD33 protein levels (NPX, x-axis) vs plasma viral load (HIV RNA copies/ml, x-axis) in MAP phase. Green dots indicate MAP-C and blue dots indicate MAP-NC. The Spearman’s rank test was used for correlation analyses. h and i) Correlogram plots of the significantly changed proteins detected between BSL vs post-RMD (h) and BSL vs MAP phase (i) that correlate with CD33 relative plasma levels. Colored dots indicate the level of correlation (blue: positive correlation and red: negative correlation). Upper graphs indicate the strongest associations observed between CD33 and 3 proteins (NEFL, GGT5 and ADAM15) in the neurology panel. The Spearman’s rank test was used for correlation analyses. Statistical significance was set at p < 0.05. www.thelancet.com Vol 95 September, 2023 9 Articles rho = 0.9696 p < 0.0001, GGT5 rho = 0.9195 p = 0.0002 In vitro CD33 targeting reduces HIV-1 replication and NEFL rho = 0.9223 p = 0.0001, Spearman rank test). and virus reactivation Interestingly, the NEFL protein, is a well-established To investigate more directly the potential involvement of marker for neurological damage in several diseases,19,20 CD33 in HIV-1 infection and viral reservoir, we tested and tend to increase the levels during MAP phase. whether targeting CD33 in vitro on T cells and These changes did not reach statistical significant dif- monocyte-derived macrophages (MDM) would impact ference between MAP-C and MAP-NC groups viral infection and replication. Indeed, adding an anti- (Supplementary Figure S3d and S3e, Wilcoxon rank test CD33 mAb to HIV-1NL4-3–infected PHA-stimulated T and Mann–Whitney test, respectively). cell cells reduced virus production in a dose-dependent Overall, these data indicate that CD33 protein levels manner (Fig. 5a One-Way ANOVA test) without differ between MAP-C and MAP-NC, even before the affecting the cell viability (Supplementary Figure S5a, BCN02 intervention and that this difference is further One-Way ANOVA test). In parallel, the amount total pronounced after RMD treatment and showed strong HIV-1 DNA was also decreased in the presence of an positive correlations with viral parameters and markers anti-CD33 mAb (Fig. 5b, One-Way ANOVA test). The of neurological damage. same was observed when HIV-1BaL-infected MDMs were cultured in presence of anti-CD33, with HIV-1 CD33/Siglec-3 validation in chronic untreated HIV-1 replication being reduced in a dose-dependent manner infected cohort (Fig. 5c p = 0.0263 and Supplementary Figure S5b, One- To validate the relationship between CD33 plasma levels Way ANOVA test). The reduction in proviral copy and virus control, we tested samples from a cohort of numbers in monocytes reached statistical significance untreated HIV-1+ individuals with different levels of vi- as well (Fig. 5d, One-Way ANOVA test). Together, these rus control by determining plasma protein and gene data suggest that targeting CD33 can reduce HIV-1 expression levels of CD33. This ART-naïve, chronically replication and/or infection, indicating that CD33 at HIV-1 infection cohort included HIV-1+ individuals with some point of the HIV-1 viral life cycle is required for high (defined as “HIV-high”, n = 47, pVL >50,000 HIV-1 effective viral propagation. RNA copies/ml) or low (defined as “HIV-low”, n = 49, pVL <10,000 HIV-1 RNA copies/ml) plasma viral loads. Significantly higher CD33 plasma protein levels were Discussion detected in HIV-high compared to HIV-low individuals The BCN02 clinical trial was a proof of concept of a (Fig. 4a, p = 0.0016, Mann–Whitney test). Moreover, “kick-and-kill” strategy employing the RMD and HIV- applying the BCN02 trial criteria for ART restart (2000 consv vaccination.3,4 A series of previously published HIV RNA copies/ml) to this comparison showed that data of BCN02 trial results have demonstrated that this untreated HIV-infected individuals with <2000 HIV RNA intervention caused marked effects on the organism at copies/ml and <50 HIV RNA copies/ml to have signifi- immunological,4 microbiome,7 epigenetic8 and neuro- cantly reduced levels of CD33 compared to individuals logical levels.12 Here, we explored the impact of the with >2000 HIV RNA copies/ml (p = 0.0340 and administration of RMD and therapeutic vaccination on p = 0.0187, respectively, ANOVA test corrected for mul- the peripheral blood proteome to identify soluble bio- tiple comparisons) (Supplementary Figure S4). As in the markers that can predict the outcome of this interven- BCN02 participants, CD33 gene expression in PBMCs tion. Our longitudinal proteome analysis revealed that from this cohort, showed higher CD33 gene expression the administration of RMD increased the abundance of levels in HIV-high individuals compared with HIV-low several plasma proteins involved in the inflammatory (Fig. 4b, p < 0.0001, Mann–Whitney test). CD33 proteo- immune response and affecting neurology processes, mic and gene expression levels at PBMCs were not sig- and that most of these changes were carried into the nificant correlated (Fig. 4c, rho = 0.319 p = 0.051, ART interruption phase of the study. As a histone Spearman rank test). As observed in the MAP phase of deacetylation inhibitor, RMD causes relaxation of the the BCN02 study, CD33 plasma levels and gene expres- host cell chromatin and enhances gene expression, of sion levels also correlated positively with plasma viral both, the integrated virus as well as host genes.8 The loads (Fig. 4d, e and h, CD33 protein levels vs pVL latter can translate into higher secretion of proteins into rho = 0.2771 p = 0.0063, and CD33 gene expression levels peripheral blood plasma, and can at least partly be re- vs pVL rho = 0.457 p = 0.004, Spearman rank test) and flected by changes in different organs, including the HIV-1 proviral copy numbers (Fig. 4f–h, CD33 protein CNS. In addition, “kick-and-kill” interventions could levels vs proviral rho = 0.2933 p = 0.0297, and CD33 gene also generate increased cell death by induction of expression levels vs proviral rho = 0.381 and p = 0.019, apoptotic proteins including CASP-8 or TNF. In line Spearman rank test). These data further indicate that with these data, previous analyses in the BCN02 trial CD33/Siglec-3 is closely related with in vivo viral repli- showed that activated T cells were sensitive to RMD cation and may serve as a predictor of the outcome after exposure as increased levels of apoptosis in CD4+ and treatment interruption and/or therapeutic interventions. CD8+ were detected.4 This was observed especially 10 www.thelancet.com Vol 95 September, 2023 Articles Fig. 4: CD33 levels in natural untreated HIV infection. Proteomic array using plasma samples from ART-naïve, chronically HIV-1 infected individuals with different degrees of HIV control (n = 96), categorized as HIV-high and HIV-low according to their viral load. a) Scatter plot showing the plasma CD33 protein levels (y-axis) in both groups (x-axis); HIV-low (n = 49, green dots) and HIV-high (n = 47, red dots). Mann– Whitney U test was applied between HIV-high and HIV-low. b) CD33 gene expression (y-axis, relative CD33 gene expression corrected for CD4 counts) measured in PBMCs of HIV-high (n = 12, red dots) and HIV-low (n = 25 green dots) (x-axis) individuals. Mann–Whitney U test was applied between HIV-high and HIV-low. c) Correlation analysis between CD33 gene expression (corrected for CD4 counts, y-axis) and CD33 protein levels in plasma considering all individuals. The Spearman’s rank test was used for correlation analyses. d and e) Correlation analysis between (d) CD33 plasma protein levels and (e) CD33 gene expression levels (x-axis) and viral load levels (HIV RNA copies/ml, y-axis). HIV-high, n = 47, red dots and HIV-low, n = 49, green dots. The Spearman’s rank test was used for correlation analyses. f and g) Correlation analysis between (f) CD33 plasma protein levels and (g) CD33 gene expression levels (x-axis) and HIV proviral DNA levels in total PBMC (HIV DNA copies/ 106 PBMCs, y-axis). Considering all the individuals (HIV-high, n = 47, red dots and HIV-low, n = 49, green dots. The Spearman’s rank test was used for correlation analyses. h) Correlogram plot between CD33 levels in plasma and PBMC expression and viral parameters. CD33 differences between groups were analyzed using the Mann–Whitney test. The Spearman’s rank test was used for correlation analyses. Statistical significance was set at p < 0.05. www.thelancet.com Vol 95 September, 2023 11 Articles Fig. 5: Targeting CD33 reduces HIV replication and provirus levels. a and c) Inhibition of HIV replication in the presence of anti-CD33 mAb, tested in (a) PHA-stimulated T cells infected with the HIVNL4-3 strain; (c) HIV infected monocyte-derived macrophages infected with the HIVBaL strain. Experimental conditions are shown on the x-axis and quantification of absolute p24 supernatant (pg/ml) is shown on the y-axis. b and d) Total HIV-1 DNA quantification in cells from the same T cell and MDMs experiments as in (a) and (c). Experimental conditions are shown on the x-axis and quantification of total HIV-1 DNA is shown on the y-axis. All the plots show the median and standard deviation of six (PHA- stimulated T cells) and five (monocyte-derived macrophages) independent experiments in duplicates for each condition. For HIV infected T cells and MDMs, One-Way ANOVA test corrected for multiple comparisons. Original FDR method of Benjamini and Hochberg was used to analyze differences between conditions. For all comparisons, p < 0.05 was considered statistically significant. during the MAP phase of the BCN02 trial, suggesting during MAP. The existence of biomarkers predictive that part of the soluble plasma proteome detected may of post-treatment virus control is supported by a stem from cell death caused by RMD toxicity, reactivated recent study describing specific plasma metabolic viral replication and/or cell lysis due to boosted cytolytic signatures associated with virus control after HIV-1 immune mechanisms.4 These different effects may also treatment interruption27 and by reports demon- help explain the results of our correlation analyses, strating increased expression of some restriction fac- where the majority of the differentially expressed pro- tors genes such as APOBEC3G and SLFN1128 or teins after RMD administration showed positive corre- immunological features such as PD-1, Tim-3 and Lag- lations with proviral levels, while during MAP, most 3 prior to ART initiation that strongly predicted time positive associations were observed for plasma viral to viral rebound.29 Furthermore, our recent studies loads.21 have shown that BCN02 MAP-C have a host-gene As RMD has been reported to have beneficial effects epigenetic landscapes8 as well as microbiome compo- in some neurological diseases2 and one of the anatom- sitions30 prior to the start of the trial that were pre- ical compartments where the latent HIV-1 reservoir is dictive of viral control during MAP. However, plasma established is the CNS,22 the participants in the BCN02 biomarkers associated with control of viremia in a trial also underwent various cognitive evaluations.12 No context of a “kick-and-kill” strategy have not been harmful effects on cognitive status, functional outcomes identified. The present data show that prior to the or brain imaging parameters were observed after RMD. BCN02 intervention, baseline plasma proteomes were In addition, CNS safety has also been confirmed previ- different between MAP-C and MAP-NC, with proteins ously after completion of the MAP.12 Despite RMD related to adaptive and innate immunity being apparently being well-tolerated, the plasma proteomics elevated in MAP-NC. Of note, the prostaglandin- revealed the levels of two neuroprotective proteins related gene ontology was the most represented one (MANF and GDNF23,24) to be increased by RMD, and two for the discrimination between MAP-C and MAP-NC. neurological damage markers (NEFL and GPNMB25,26) Particularly, MAP-C showed higher levels of plasma being increased during the MAP phase. These obser- PLA2G10 protein than MAP-NC. It is tempting to vations suggest that RMD did indeed affect the CNS, speculate that MAP-C have higher antiviral capabil- either directly or through the induction of pro- ities due to the higher levels of PLA2G10, as it has inflammatory responses and/or the reactivation of the been previously shown that PLA2G10 has the ability to viral reservoir. neutralize HIV-1 through the degradation of the viral Aside from longitudinally evaluating the effects of membrane and to inhibit HIV-1 replication in CD4 T RMD and MAP on the plasma proteomes and markers cells.31 It will be important to explore how the markers of neurodamage, we also tested whether these plasma involved in prostaglandin-related ontology, particu- signature could inform on eventual virus control larly PLA2G10, determine the response to cure 12 www.thelancet.com Vol 95 September, 2023 Articles strategies and how they could can be leveraged to one of these DMPs mapped to a Single Nucleotide achieve improved control rates in future trials. Polymorphism (SNP) associated with AD risk,37,38 it will Interestingly, CD33/Siglec-3 protein was the unique be interesting to assess to what degree CD33 influences plasma marker, whose levels were increased upon RMD neurological disease progression in HIV-1+ individuals administration and maintained when ART was stopped. suffering from HAND. Of note nowadays, therapeutic In addition, CD33 plasma levels allowed for the antibodies targeting the V-set domain or sialic acid- discrimination of MAP-C and MAP-NC individuals, binding domain of CD33 have been approved to treat already at the baseline timepoint. Intriguingly, CD33/ acute myeloid leukemia.39 Also, in the context of hepa- Siglec-3 can modulate immune responses by inhibit- titis B virus (HBV) infection, the use of anti-CD33 was ing monocytes,32 but can be also expressed on dendritic able to reverse hHBV-mediated immunosuppression in cells and in a lesser extent on activated T cells and pre-clinical in vitro studies.40 Finally, a recent study in natural killer cells. The binding with its ligands; C1q Alzheimer’s Disease has ranked lintuzumab (SGN-33) and sialylated glycoproteins (the latter also highly as one of the top repurposed drug candidates for treat- expressed in brain microglia), leads to the recruitment ing AD.41 Given this evidence, it is tempting to propose of inhibitory proteins initiating signaling cascades that that anti-CD33 mAb could be employed in HIV-1 inhibit cell activation and functions associated with infection, with a potential dual beneficial effects of cytokine release or virus phagocytosis.32 Previous reports reducing HIV-1 replication and improving HIV-1- have shown that homeostatic basal levels of plasma associated neurocognitive deficits. CD33 are low but are strongly increased under patho- To validate our observations in the small number of genic conditions, including viral infections.32 As cell BCN02 participants, a large cohort of ART naïve, surface CD33 is an interferon inducible factor,33 higher chronically HIV-1+ individuals with a wide range of viral viral burden could induce its production, and in more loads was analyzed. These results confirmed that HIV-1 extreme situation, shedding of CD33 into the plasma, as infected individuals with poor virus control (HIV-high) has been described for another member of the Siglec had elevated plasma levels of CD33, which were also family (Siglec-7) in uncontrolled HIV-1 infection. Of positively correlated with both, viral load and proviral note, pre-incubation of HIV-1 particles with soluble DNA copy numbers. This is in line with a report by Siglec-7 increased the infection rate of CD4 T cells, Rempel et al., showing that several other members of which do not constitutively express Siglec-7 at the sur- the Siglec family can facilitate infection and viral repli- face.33 Moreover, PLWH showed higher plasma levels of cation (Siglec-5, -9, and as shown above, Siglec-7).42,43 It Siglec-7 compared to seronegative individuals and these has also been well established that the direct interaction plasma levels correlated positively with HIV-1 viral between sialylactose-containing gangliosides in the HIV loads.33 It is thus plausible to think that increased levels viral membrane and the cellular lectin Siglec-1 is critical of CD33 detected in MAP-NC and HIV-high partici- for HIV-1 capture and storage by mature dendritic cells pants could accelerate HIV propagation. This would be and for trans-infection between infected dendritic cells in line with our findings of higher CD33 protein plasma and T cells.43,44 In addition, the HIV-1 gp120 surface levels in MAP-NC than MAP-C, even if total PBMC protein has been shown to serve as viral ligand for transcriptomics analysis in this small set of BCN02 Siglec-3,34 which is thought to facilitate infection of participants did not reach in all cases statistical signifi- macrophages and T cells.33,45 cance. Also, similarly to Siglec-7 study, CD33 plasma Finally, based on our observation that plasma CD33 levels were positively associated with pVL and proviral predicts post-treatment control in BCN02 and the levels in the BCN02 trial, especially during MAP when described physical associations between CD33/Siglec-3 levels of circulating virus are elevated. and HIV-1 particles, we explored the possibility to In addition to its effects in viral infections, CD33 has inhibit HIV-1 replication by targeting CD33 in in vitro been implicated in neuroinflammation, where its acti- cultures using an anti-CD33 mAb. Indeed, in T cells and vation reduces phagocytic activity of microglia cells, MDMs viral replication was strongly suppressed in a causing an accumulation of pathogenic Aβ plaques.32,34 dose-dependent manner. While the effects of such In our present study, we indeed observed very strong treatment on acute infection of CNS cells needs to be associations between CD33 plasma levels and the further explored, these therapeutic CD33 targeting an- gamma-glutamyltransferase 5 (GGT5), ADAM metal- tibodies have been used for Hepatitis B virus, AML lopeptidase domain 15 (ADAM15) and neurofilament treatment and Alzheimer Disease,39–41 moving them light protein (NEFL). In particular, NEFL has been re- closer to be tested in a clinical setting in HIV infection ported to be a biomarker of Alzheimer disease,35,36 where and with more extensive CNS assessment than what was CD33 may drive impaired phagocytic capacity of possible in the BCN02 trial. microglia and pose a risk for Alzheimer’s disease.32 The present study has a number of limitations, aside Furthermore, our previous BCN02 epigenetic analyses from the small size of the BCN02 clinical trial. In revealed differential DNA methylation positions (DMPs) particular, larger studies that include placebo controls in the CD33 gene comparing MAP-C and MAP-NC. As will be needed to confirm the ability of CD33 levels to www.thelancet.com Vol 95 September, 2023 13 Articles predict virus control in a treatment interruption phase References 1 Cillo AR, Mellors JW. Which therapeutic strategy will achieve a either assessed prior to treatment initiation or at ART cure for HIV-1? Curr Opin Virol. 2016;18:14–19. https://doi.org/10. cessation. Of note, such studies should also include fe- 1016/j.coviro.2016.02.001. male participants, as the present analyses were based 2 Zeng H, Xu L, Zou Y, Wang S. Romidepsin and metformin nanomaterials delivery on streptozocin for the treatment of Alz- largely on male study participants. Sample availability heimer’s disease in animal model. Biomed Pharmacother. 2021;141: for the current BCN02 sub-study was limited, and it did 111864. https://doi.org/10.1016/j.biopha.2021.111864. not include tissue samples such as lymph nodes, MALT 3 Mothe B, Rosás-Umbert M, Coll P, et al. HIVconsv vaccines and romidepsin in early-treated HIV-1-infected individuals: safety, and/or CSF, preventing us from conducting further immunogenicity and effect on the viral reservoir (Study BCN02). tissue-based analyses that could have helped to espe- Front Immunol. 2020;11:823. 4 Rosás-Umbert M, Ruiz-Riol M, Fernández MA, et al. In vivo ef- cially further validating the CNS findings. Despite these fects of romidepsin on T-cell activation, apoptosis and function in limitations though, our analyses document that the the BCN02 HIV-1 kick&Kill clinical trial. Front Immunol. “kick and kill” strategy used in BCN02 left specific 2020;11:418. 5 Létourneau S, Im EJ, Mashishi T, et al. Design and pre-clinical marks on inflammatory and neurological plasma pro- evaluation of a universal HIV-1 vaccine. PLoS One. 2007;2(10):e984. teomes and allowed us to identify soluble CD33 as a 6 Wei DG, Chiang V, Fyne E, et al. Histone deacetylase inhibitor romidepsin induces HIV expression in CD4 T cells from patients novel biomarker of poor virus control (both, during the on suppressive antiretroviral therapy at concentrations achieved by ART interruption phase of BCN02 and in natural HIV clinical dosing. PLoS Pathog. 2014;10(4):e1004071. infection). Our studies fall short of defining the precise 7 Borgognone A, Noguera-Julian M, Oriol B, et al. Gut microbiome signatures linked to HIV-1 reservoir size and viremia control. bio- mechanism of how soluble and/or cellular CD33 may Rxiv; 2021, 2021.10.03.462590. Available from: http://biorxiv.org/ influence viral loads, although similarities to the content/early/2021/10/04/2021.10.03.462590.abstract. mechanism of action of other Siglec family members 8 Oriol-Tordera B, Esteve-Codina A, Berdasco M, et al. Epigenetic landscape in the kick-and-kill therapeutic vaccine BCN02 clinical (i.e., Siglec-7) lend importance to the soluble form of trial is associated with antiretroviral treatment interruption (ATI) CD33. The fact that the CD33 plasma levels were outcome. EBioMedicine. 2022;78:103956. elevated already in pre-intervention samples in line with 9 Marban C, Forouzanfar F, Ait-Ammar A, et al. Targeting the brain reservoirs: toward an HIV cure. Front Immunol. 2016;7:397. this and suggest that the CD33 protein is critically 10 Eggers C, Arendt G, Hahn K, et al. HIV-1-associated neuro- involved in viral replication in vivo and that it could be cognitive disorder: epidemiology, pathogenesis, diagnosis, and treatment. J Neurol. 2017;264(8):1715–1727. therapeutically targeted in future cure approaches. 11 Chan P, Ananworanich J. Perspective on potential impact of HIV central nervous system latency on eradication. AIDS. Contributors 2019;33:S123–S133. MRR and CB designed the experimental plan. BM, TH, and JM 12 Munoz-Moreno JA, Carrillo-Molina S, Martinez-Zalacain I, et al. contributed to the generation of clinical trial samples, participant data Preserved central nervous system functioning after use of romi- and sample management. AP and JAMM performed neuropsycho- depsin as a latency-reversing agent in an HIV cure strategy. AIDS. logical testing. BOT, CDC, and ALL performed the sample processing. 2022;36(3):363–372 [cited 2022 Aug 19]. Available from: https:// JMP and CG determined viral parameters. CG, JS contributed in the journals.lww.com/aidsonline/Fulltext/2022/03010/Preserved_central_ nervous_system_functioning_after.5.aspx. recruitment of untreated HIV-1+ individuals in Lima, Peru. CDC and 13 Li M. Proteomics in the investigation of HIV-1 interactions with MRR analyzed plasma proteomics data and integrated them with host proteins. Proteomics Clin Appl. 2015;9(1–2):221–234. clinical and virological parameters. CDC performed RT-PCR and 14 Calcagno A, Atzori C, Romito A, et al. Blood brain barrier conducted the in vitro experiments. CDC and BOT run statistical, impairment is associated with cerebrospinal fluid markers of bioinformatics and integrative analyses using R program. EB and EGV neuronal damage in HIV-positive patients. J Neurovirol. helped with monocyte-derived macrophages (MDMs) infection and 2016;22(1):88–92. proviral PCR quantification procedures. CDC, CB, and MRR drafted 15 Brew B, Carroll A. HIV-associated neurocognitive disorders: recent and edited the manuscript. CB and BC procured resources and advances in pathogenesis, biomarkers, and treatment. F1000Res. 2017;6:312. funding for the execution of the study. All authors reviewed and 16 Mothe B, Manzardo C, Sanchez-Bernabeu A, et al. Therapeutic approved the final version of the manuscript. vaccination refocuses T-cell responses towards conserved regions of HIV-1 in early treated individuals (BCN 01 study). EClinicalMedi- Data sharing statement cine. 2019;11:65–80. Additional information can be found in the Supplementary Material of 17 Oriol-Tordera B, Olvera A, Duran-Castells C, et al. TL1A-DR3 this article. plasma levels are predictive of HIV-1 disease control, and DR3 co- stimulation boosts HIV-1-specific T-cell responses. J Immunol. 2020;205:3348–3357. Declaration of interests 18 Martínez-Bonet M, Puertas MC, Fortuny C, et al. Establishment BM is a consultant of AELIX THERAPEUTICS, SL outside the sub- and replenishment of the viral reservoir in perinatally HIV-1- mitted work. CB is co-founder, chief science officer and shareholder of infected children initiating very early antiretroviral therapy. Clin AELIX THERAPEUTICS. TH is a coinventor of the HIVconsv immu- Infect Dis. 2015;61(7):1169–1178. nogen. All other authors declare that they have no competing interests. 19 Liu Y, Dou K, Xue L, Li X, Xie A. Neurofilament light as a biomarker for motor decline in Parkinson’s disease. Front Neurosci. Acknowledgments 2022;16:959261. We thank all the participants involved in the study. We also give thanks 20 Ning L, Wang B. Neurofilament light chain in blood as a diagnostic and predictive biomarker for multiple sclerosis: a systematic review to PERIS (SLT017/20/000090) for the research fellowship to EGV and and meta-analysis. Sipilä JOT PLoS One. 2022;17(9):e0274565 [cited ISCIII-FIS (CPII19/00012) for support to EB. 2022 Sep 26]. Available from: https://dx.plos.org/10.1371/journal. pone.0274565. Appendix A. Supplementary data 21 Conway JM, Perelson AS, Li JZ. Predictions of time to HIV viral Supplementary data related to this article can be found at https://doi. rebound following ART suspension that incorporate personal bio- org/10.1016/j.ebiom.2023.104732. markers. PLoS Comput Biol. 2019;15(7):1–26. 14 www.thelancet.com Vol 95 September, 2023 Articles 22 Ash MK, Al-Harthi L, Schneider JR. HIV in the brain: identifying 34 Chang YC, Nizet V. Siglecs at the host–pathogen interface. viral reservoirs and addressing the challenges of an HIV cure. Springer Singapore Adv Exp Med Biol. 2020;1204:197–214. https:// Vaccines. 2021;9(8):867. doi.org/10.1007/978-981-15-1580-4_8. 23 Tseng KY, Danilova T, Domanskyi A, Saarma M, Lindahl M, 35 Brickman AM, Manly JJ, Honig LS, et al. Plasma p-tau181, p- Airavaara M. MANF is essential for neurite extension and neuronal tau217, and other blood-based Alzheimer’s disease biomarkers in a migration in the developing cortex. eNeuro. 2017;4(5):ENEURO. multi-ethnic, community study. Alzheimers Dement. 2021;17(8): 0214-17.2017. 1353–1364. 24 Cintrón-Colón AF, Almeida-Alves G, Boynton AM, Spitsbergen JM. 36 Illán-Gala I, Lleo A, Karydas A, et al. Plasma tau and neurofilament GDNF synthesis, signaling, and retrograde transport in motor light in frontotemporal lobar degeneration and Alzheimer disease. neurons. Cell Tissue Res. 2020;382(1):47–56. Neurology. 2021;96(5):e671–e683. 25 Kodidela S, Gerth K, Haque S, et al. Extracellular vesicles: a 37 Lee MN, Ye C, Villani A, et al. CD33 Alzheimer’s disease locus: possible link between HIV and Alzheimer’s disease-like pathology altered monocyte function and amyloid biology. Nat Neurosci. in HIV subjects? Cells. 2019;8(9):968. 2013;343(6175):848–850. 26 Diaz-Ortiz ME, Seo Y, Posavi M, et al. GPNMB confers risk for 38 Malik M, Chiles J, Xi HS, et al. Genetics of CD33 in Alzheimer’s Parkinson’s disease through interaction with α-synuclein. Science. disease and acute myeloid leukemia. Hum Mol Genet. 2022;377(6608):eabk0637 [cited 2022 Oct 18]. Available from: 2015;24(12):3557–3570. https://pubmed.ncbi.nlm.nih.gov/35981040/. 39 Walter RB. Investigational CD33-targeted therapeutics for acute 27 Giron LB, Palmer CS, Liu Q, et al. Non-invasive plasma glycomic myeloid leukemia. Expert Opin Investig Drugs. 2018;27(4):339–348. and metabolic biomarkers of post-treatment control of HIV. Nat https://doi.org/10.1080/13543784.2018.1452911. Commun. 2021;12(1):3922. https://doi.org/10.1038/s41467-021- 40 Tsai TY, Huang MT, Sung PS, et al. SIGLEC-3 (CD33) serves as an 24077-w. immune checkpoint receptor for HBV infection. J Clin Invest. 28 De Scheerder MA, Van Hecke C, Zetterberg H, et al. Evaluating 2021;131(11):1–15. predictive markers for viral rebound and safety assessment in blood 41 Zhang M, Schmitt-Ulms G, Sato C, et al. Drug repositioning for and lumbar fluid during HIV-1 treatment interruption. Alzheimer’s disease based on systematic “omics” data mining. J Antimicrob Chemother. 2020;75(5):1311–1320. PLoS One. 2016;11(12):e0168812. 29 Hurst J, Hoffmann M, Pace M, et al. Immunological biomarkers 42 Rempel H, Calosing C, Sun B, Pulliam L. Sialoadhesin expressed predict HIV-1 viral rebound after treatment interruption. Nat on IFN-induced monocytes binds HIV-1 and enhances infectivity. Commun. 2015;6:8495. https://doi.org/10.1038/ncomms9495. PLoS One. 2008;3(4):e1967. 30 Borgognone A, Noguera-Julian M, Oriol B, et al. Gut microbiome 43 Raïch-Regué D, Resa-Infante P, Gallemí M, et al. Role of Siglecs in signatures linked to HIV-1 reservoir size and viremia control. bioRxiv. viral infections: a double-edged sword interaction. Mol Aspects Med. 2021:2021.10.03.462590. https://doi.org/10.1101/2021.10.03.462590. 2022;90:101113. 31 Kim J, Chakrabarti B, Guha-Niyogi A, et al. Lysis of human im- 44 Izquierdo-Useros N, Lorizate M, McLaren PJ, Telenti A, munodeficiency virus type 1 by a specific secreted human phos- Kräusslich HG, Martinez-Picado J. HIV-1 capture and transmission pholipase A2. J Virol. 2007;81(3):1444–1450. by dendritic cells: the role of viral glycolipids and the cellular re- 32 Gonzalez-Gil A, Schnaar RL. Siglec ligands. Cells. 2021;10(5):1–21. ceptor Siglec-1. PLoS Pathog. 2014;10(7):e1004146. 33 Varchetta S, Lusso P, Hudspeth K, et al. Sialic acid-binding Ig-like 45 Zou Z, Chastain A, Moir S, et al. Siglecs facilitate HIV-1 infection lectin-7 interacts with HIV-1 gp120 and facilitates infection of of macrophages through adhesion with viral sialic acids. PLoS One. CD4pos T cells and macrophages. Retrovirology. 2013;10(1):154. 2011;6(9):e24559. www.thelancet.com Vol 95 September, 2023 15