Complement pathway gene activation and risingcirculating immune complexes characterize earlydisease in HIV-associated tuberculosisHanif Esmaila,b,c,d,1, Rachel P. Laie, Maia Lesoskyb,c,f, Katalin A. Wilkinsonb,c,e, Christine M. Grahamg, Stuart Horswelle,Anna K. Coussensb,c,h, Clifton E. Barry IIIb,c,i,j,k, Anne O’Garrag,l, and Robert J. Wilkinsona,b,c,e,1

aDepartment of Medicine, Imperial College London, London W2 1PG, United Kingdom; bWellcome Center for Infectious Diseases Research in Africa,Institute of Infectious Disease and Molecular Medicine, University of Cape Town, Cape Town 7925, Republic of South Africa; cDepartment of Medicine,University of Cape Town, Cape Town 7925, Republic of South Africa; dRadcliffe Department of Medicine, University of Oxford, Oxford OX3 9DU, UnitedKingdom; eThe Francis Crick Institute, London NW1 2AT, United Kingdom; fDivision of Epidemiology and Biostatistics, School of Public Health and FamilyMedicine, University of Cape Town, Cape Town 7925, Republic of South Africa; gThe Francis Crick Institute, Laboratory of Immunoregulation and Infection,London NW1 2AT, United Kingdom; hDivision of Medical Microbiology, Department of Pathology, University of Cape Town, Cape Town 7925, Republic ofSouth Africa; iFaculty of Medicine and Health Sciences, Stellenbosch University, Cape Town 7505, Republic of South Africa; jTuberculosis Research Section,National Institute of Allergy and Infectious Diseases, NIH, Bethesda, MD 20892; kDepartment of Clinical Laboratory Sciences, University of Cape Town, CapeTown 7925, Republic of South Africa; and lNational Heart and Lung Institute, Imperial College London, London W2 1PJ, United Kingdom

Edited by Barry R. Bloom, Harvard T. H. Chan School of Public Health, Boston, MA, and approved December 13, 2017 (received for review July 2, 2017)

The transition between latent and active tuberculosis (TB) occursbefore symptom onset. Better understanding of the early eventsin subclinical disease will facilitate the development of diagnosticsand interventions that improve TB control. This is particularlyrelevant in the context of HIV-1 coinfection where progression ofTB is more likely. In a recent study using [18F]-fluoro-2-deoxy-D-glucose positron emission/computed tomography (FDG-PET/CT)on 35 asymptomatic, HIV-1–infected adults, we identified 10 par-ticipants with radiographic evidence of subclinical disease, signif-icantly more likely to progress than the 25 participants without. Togain insight into the biological events in early disease, we per-formed blood-based whole genome transcriptomic analysis onthese participants and 15 active patients with TB. We found tran-scripts representing the classical complement pathway and Fcγ re-ceptor 1 overabundant from subclinical stages of disease. Levels ofcirculating immune (antibody/antigen) complexes also increased insubclinical disease and were highly correlated with C1q transcriptabundance. To validate our findings, we analyzed transcriptomicdata from a publicly available dataset where samples were availablein the 2 y before TB disease presentation. Transcripts representingthe classical complement pathway and Fcγ receptor 1 were alsodifferentially expressed in the 12 mo before disease presentation.Our results indicate that levels of antibody/antigen complexes in-crease early in disease, associatedwith increased gene expression ofC1q and Fcγ receptors that bind them. Understanding the role thisplays in disease progression may facilitate development of interven-tions that prevent this, leading to a more favorable outcome andmay also be important to diagnostic development.

tuberculosis | HIV | complement | immune complex | incipient disease

Conventionally, tuberculosis (TB) is divided into stages ofasymptomatic latent infection, during which bacillary repli-cation is effectively controlled in a healthy host, and active dis-ease in which this has failed, resulting in symptomaticdeterioration. Understanding the transition between these twostates is important, although until recently this has been over-looked (1). Active disease is usually defined by a combination ofsymptoms, pathology (radiographically or histologically identi-fied), and culturable bacilli. These features do not appear si-multaneously but develop over time and may be intermittentlypresent, hence the active disease processes may begin monthsbefore symptom onset, i.e., as subclinical active disease (2). Insome of those who initially reactivate, disease progression maybe arrested and regress particularly when disease extent is lim-ited (3). A greater understanding of the early events in disease iscritical for the development of novel approaches to both identify

people in early subclinical stages of disease and interventions toprevent progression. This is of particular importance in thosewith HIV-1 coinfection where TB disease progression is morelikely. However, studying the natural history of TB in this groupis further complicated by the imperative to provide preventivetherapy, so novel approaches are needed.In macaques the failure of the granuloma has been shown to be

followed by cellular infiltration within bronchi (pneumonia) (4). Inhuman autopsy studies, pneumonic infiltration has also been ob-served as the first pathological sign of pulmonary disease (5). Dis-ease regression and self-healing of lesions is associated with fibrosis;however, it appears that disease risk following this is significantlyincreased (6). Medical imaging is a key approach to detecting evi-dence of early disease in asymptomatic persons, facilitated by acharacteristic distribution of pathology. Chest radiography (CXR)

Significance

Understanding events in early tuberculosis disease will facili-tate the development of novel tests to predict disease pro-gression and interventions to prevent it. Blood-basedtranscriptomic approaches consistently identify several path-ways relevant to clinical disease. Here we show in asymp-tomatic people with HIV infection and early subclinicaltuberculosis defined by FDG-PET/CT that transcripts relating tothe classical complement pathway and Fcγ receptor remainenriched, accompanied by rising levels of circulating antibody/antigen immune complexes. We confirm that transcripts re-lating to these pathways also rise in the 12 mo prior to diseasepresentation in HIV-uninfected people. This supports observa-tions that antigen may be present in early disease despite be-ing paucibacillary and demonstrates that modulation of theimmune response could occur via immune complex formation.

Author contributions: H.E., C.E.B., A.O., and R.J.W. designed research; H.E., R.P.L., K.A.W.,C.M.G., and A.K.C. performed research; C.E.B., A.O., and R.J.W. contributed new reagents/analytic tools; H.E., R.P.L., M.L., S.H., A.K.C., A.O., and R.J.W. analyzed data; and H.E., A.O.,and R.J.W. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Published under the PNAS license.

Data deposition: The data reported in this paper have been deposited in the Gene Ex-pression Omnibus (GEO) database, https://www.ncbi.nlm.nih.gov/geo (accession no.GSE69581).1To whom correspondence may be addressed. Email: hanif.esmail@ndcls.ox.ac.uk or r.j.wilkinson@imperial.ac.uk.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplemental.

E964–E973 | PNAS | Published online January 16, 2018 www.pnas.org/cgi/doi/10.1073/pnas.1711853115

Dow

nloa

ded

by g

uest

on

Apr

il 1,

202

1

http://crossmark.crossref.org/dialog/?doi=10.1073/pnas.1711853115&domain=pdfhttp://www.pnas.org/site/aboutpnas/licenses.xhtmlhttps://www.ncbi.nlm.nih.gov/geohttp://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE69581mailto:hanif.esmail@ndcls.ox.ac.ukmailto:r.j.wilkinson@imperial.ac.ukmailto:r.j.wilkinson@imperial.ac.ukhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalwww.pnas.org/cgi/doi/10.1073/pnas.1711853115

has long been used for this purpose; however, CXR has limitationsin both sensitivity and reproducibility (7). [18F]-fluoro-2-deoxy-D-glucose positron emission/computed tomography (FDG-PET/CT) isa highly sensitive imaging modality, which combines cross-sectionalanatomical detail from a CT scan with quantitation and localizationof metabolic activity by quantifying uptake of FDG (a glucose an-alog) by PET scan. FDG uptake in tuberculosis is related to acti-vated macrophages and an inflammatory infiltrate at the site ofdisease (8). In a recent clinical study, we utilized FDG-PET/CT toidentify evidence of early disease in asymptomatic people otherwiseconventionally considered to have latent infection [QuantiFERONGold In-Tube (QFN-GIT) positive, sputum culture negative, CXRno evidence of active disease] but at high risk of disease progressionas HIV-1–infected, antiretroviral therapy (ART) naïve. Ofthe 35 people scanned, 10 (28.6%) had evidence of subclinicaldisease by FDG-PET/CT and they were significantly more likelyto progress to clinical disease (2).Transcriptomic profiling of peripheral blood has provided critical

insight into active TB (9–12). Blankley et al. (13, 14) have recentlyanalyzed 16 such studies and shown overall findings in modularanalysis of the transcriptome to be remarkably consistent. Meta-profiling identified 380 genes differentially abundant in active TB inat least 9 of the 16 studies. Based on pathway analysis, they thendetermined the main functional groups of genes in active TB torelate to: IFN-signaling/inducible genes, the complement pathway, Igreceptors, recognition of pathogen-associated molecular patterns,and inflammasome and proinflammatory pathways. However, it isnot clear which of these processes occur at the earliest stages ofdisease. Understanding such host–pathogen interaction in early dis-ease may provide insight that could facilitate novel approaches todiagnose and manage subclinical disease. Zak et al. (15) have re-cently demonstrated in HIV-1–uninfected adolescents that tran-scriptional differences are present in blood months before clinicalpatency, suggesting that transcriptomic profiling of peripheral bloodmight be a useful approach to provide insight into subclinical disease.In the present study, we utilized the cohort of HIV-1–infected

adults described above in which presence or absence of subclinicalTB was defined by FDG-PET/CT (2) and corecruited controlparticipants with active TB. We investigated which components ofthe active TB whole blood transcriptional profile were present inthose with evidence of subclinical disease to provide insight intothe biology underlying early progressive disease.

ResultsDetails of the clinical and radiographic features of the clinical cohorthave been previously described (2). Thirty-five asymptomatic,healthy, HIV-1–infected, ART-naïve participants with medianCD4 count of 517/mm3, a positive QFN-GIT, and no previoushistory of TB, were screened for active TB by two sputum cul-tures (induced if needed) and CXR. All participants were ofAfrican ancestry and residents in Khayelitsha (township of CapeTown, South Africa). Following 42-d negative sputum cultures,participants were confirmed as remaining symptom-free, un-derwent FDG-PET/CT, and were commenced on isoniazid pre-ventive therapy. FDG-PET/CT scans were systematically analyzed,with abnormalities categorized and spatially mapped, interpreted inthe context of historical autopsy studies, and then classified as beingconsistent with subclinical active TB or not. Ten of the 35 participantswere classified as having subclinical active TB, due to either infiltratesand/or fibrotic scars (n = 9, consistent with bronchogenic spread ofdisease) or multiple active nodules (n = 1, consistent with hematog-enous spread of disease) on FDG-PET/CT (referred to as “subclinicalTB”); the remaining 25 participants had no evidence of disease ac-tivity on FDG-PET/CT scan (referred to as “latent TB”). Four of the10 participants with evidence of subclinical TB were determined tohave disease progression (defined as developing TB symptoms alongwith radiographic deterioration or culture positivity) requiring com-mencement of standard TB therapy in contrast to 0/25 participants

with no evidence of subclinical TB. Participants were followed up for6 mo; 27 of the 35 participants (6 subclinical and 21 latent) then hadrepeat FDG-PET/CT. All 6 of the subclinical participants had im-provement in baseline abnormalities in lymph nodes and lung pa-renchyma in contrast to only 1/21 of those without evidence ofsubclinical TB. Three participants (2 in the subclinical group and 1 inthe latent group) had been commenced on antiretroviral therapyduring followup as CD4 count fell below 350/mm3, in accordance withlocal standard of care at the time of recruitment.In addition, we recruited 15 participants with symptomatic,

microbiologically confirmed pulmonary TB (referred to as “ac-tive TB”) for blood sampling. There were no significant differ-ences between those with active, subclinical, and latent TB withregard to median age (P = 0.41), sex (P = 0.52), or medianCD4 count (P = 0.09). Conversely, median viral load (VL) andmedian C-reactive protein (CRP) were significantly differentbetween active and latent TB (P < 0.0001 and P = 0.0001, re-spectively), but not between subclinical and latent TB (P =0.97 and P = 0.08, respectively) (SI Appendix, Fig. S1).

Eighty-Two Transcripts Distinguish Subclinical and Active TB fromLatent TB. Transcript abundance in whole blood RNA was de-termined by Illumina HumanHT-12 v4.0 Expression BeadChipmicroarray. We initially defined a set of transcripts that weredifferentially abundant between latent and active TB to determinea transcriptional profile of active TB. From the 47,231 probes onthe microarray, 18,919 were present in at least 10% of samples(Materials and Methods). The 18,919 transcripts were then statis-tically filtered [ANOVA with Benjamini–Hochberg false discoveryrate (FDR) correction (pcorr < 0.05)] to identify 2,573 transcriptsshowing significant differential abundance between the differentstages of TB. From these transcripts, we identified 893 transcriptswith at least a 1.5-fold change (FC) difference in abundance be-tween active and latent TB. To determine which of these tran-scripts might be relevant to early disease, we filtered the 893transcripts by FC to identify those differentially abundant betweensubclinical and latent TB. As disease pathology in subclinical TB isof minimal extent by comparison with active TB, we reasoned thatthe relevant transcripts may be expressed in blood at lower levelsin comparison with active TB and hence both 1.25- and 1.5-FCcutoffs were used, resulting in a 203- and 82-transcript list, re-spectively (SI Appendix, Fig. S2 and Dataset S1).Hierarchical clustering was then performed on the 893, 203,

and 82 transcripts, which confirmed progressive improvement inability to classify those with subclinical TB from those with latentTB while clustering them with active TB (Fig. 1A). Two partic-ipants with subclinical TB did not cluster with active TB, onewith multiple active nodules consistent with hematogeous spreadof disease, and the other who, of those with abnormalities con-sistent with bronchogenic spread, had the smallest infiltrate, withamong the lowest metabolic activity in lesion or lymph node byFDG uptake (SI Appendix, Table S1). In 27 participants seen at6-mo followup, after receiving either isoniazid preventive therapy(IPT) or standard TB therapy, the subclinical and latent groupsbecame indistinguishable by transcriptional signature (Fig. 1B).To examine whether abundance of these transcripts altered as

subclinical disease activity increased, we investigated the asso-ciation of metabolic activity by FDG uptake with transcriptabundance. Participants with subclinical TB were further dividedinto two categories: five participants showing intense FDG up-take [visual score (VS) = 3] within lung parenchyma or centrallymph nodes, and five participants with less intense FDG uptake(VS = 0–2) (SI Appendix, Table S1). Hierarchical clustering us-ing the group mean expression values for the 82 transcriptsdemonstrated that the subclinical participants with greater meta-bolic activity, clustered more closely with active TB than the sub-clinical TB group with low FDG uptake, which in turn clusteredcloser to the latent TB group (Fig. 1C). This was then further

Esmail et al. PNAS | Published online January 16, 2018 | E965

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

Apr

il 1,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplemental

demonstrated by disease risk score (determined for each participantby subtracting the summed normalized expression values ofunderabundant transcripts from that of overabundant transcriptsas described in ref. 12) (Fig. 1D). There was no significant dif-ference in mean CRP (2.62 mg/L vs. 2.88 mg/L; P = 0.85, normalrange

higher serum concentration of CIC than latent TB (Fig. 3G).Furthermore serum CIC as a single marker to discriminate thefour individuals subsequently treated for TB during followup fromthe 31 individuals that were not, showed an area under the curve(AUC) of 0.754. A cutoff of 192 μg Eq/mL had sensitivity of 75%and specificity of 93.6% and correctly classified 91.4%. Of note,the only participant of the four subsequently treated for active TBwith a serum CIC less than 192 μg Eq/mL had a different radio-

graphic presentation subclinical disease, with multiple nodulessuggestive of hematogenous spread in contrast to the broncho-genic pattern of disease seen in the remainder.In addition to binding C1q, antibody/antigen immune com-

plexes also bind Fcγ receptors and Fcγ binding protein (FCGBP).Those with subclinical TB and active TB also had relative over-abundance of transcripts relating to the Fc fragment of IgG, highaffinity I receptor (CD64) (FCGR1) with FCGR1A/B/C all being

A

B

C

Fig. 2. Transcripts relating to the classical complement pathway are overabundant in subclinical and active TB. (A) Top four enriched canonical pathwayslisted by significance (P < 0.05, Fisher’s exact) for the 893-transcript signature of active TB shown. The bar chart shows the proportion of these pathwaysrepresented in the 893-, 203-, and 82-transcript signatures. (B) Schematic of the complement pathway, created using Ingenuity Pathway Analysis software,showing components of the pathway with overabundant transcripts in active (n = 15) and subclinical TB (n = 10) compared with latent TB (n = 25) (in pink).Box and whisker plots show normalized expression of complement components SERPING1 and C1QB (part of 82-transcript signature) and C2 and C5 (part of203-transcript signature) by TB disease state. (C) Box and whisker plots show normalized expression of Ig-related transcripts FCGBP and FCGR1C (part of 82-transcript signature) and FCGR1A and FCGR1B (part of 203-transcript signature) by TB disease state.

Esmail et al. PNAS | Published online January 16, 2018 | E967

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

Apr

il 1,

202

1

present in the 82- and/or 203-transcript list; however, there was areduction in abundance of FCGBP (Fig. 2C and SI Appendix,Table S2). FCGR1A, FCGR1B, and FCGBP were all present inthe 380 metasignature. FCGBP is the most consistently under-abundant transcript in active TB identified in 14 previous studies.By contrast, FCGR1B and FCGR1A are among the most consis-tently overabundant transcripts in active TB being present in16 and 14 studies, respectively. RT-PCR was also conducted usingprimers for FCGR1C, FCGBP, C1QB, and SERPING1, whichconfirmed the differences demonstrated by microarray betweenlatent, subclinical, and active TB (SI Appendix, Fig. S3).

Classical Complement and IFN-Signaling Pathways Are Increased inProgressive TB in HIV-Uninfected Persons. To validate our findingsand establish whether classical complement and Fcγ-receptorgene activation were important in early TB disease, we rean-alyzed the RNA sequencing data from the publicly availabledataset from the study reported by Zak et al. (15), in which,samples were available in the 2 y before disease presentationwith appropriate nonprogressor controls. Although no accuratesubclinical disease phenotype was ascertained, we reasoned thatthose in the 6–12 mo before disease presentation may have hadsubclinical disease. Furthermore, participants in Zak et al. (15)were all HIV uninfected, allowing us to determine whether ourfindings were generalizable. To investigate the biological pro-cesses relevant to progressing disease, we first established thenumber of transcripts significantly differentially abundant(pcorr < 0.05) and with >1.5 FC overabundance between pro-gressors and nonprogressors at

and those with TB only (HIV−TB+) had significant differences inexpression in 24/38 modules (15 increased expression and 9 reducedexpression) (SI Appendix, Table S3). Both the HIV-only group andTB-only group had significantly increased expression in all three IFNmodules. Those with HIV-associated TB (HIV+ART−TB+) hadgreater expression of the IFN modules in comparison with thosewith HIV or TB only (Fig. 5 and SI Appendix, Table 3). Those withHIV but on ART (HIV+ART+TB−) with fully suppressed viral loadshowed normalization of expression within the IFNmodules and hadno significant differences in any of the 38 modules in comparisonwith HIV−TB− controls. Given the background increased abundance

of IFN-related transcripts in HIV-infected persons not on ART, it islikely that any increased expression within these pathways in earlytuberculosis would be less discriminatory.Recognizing that interferons are well known to enhance ex-

pression of Fcγ receptor, we next sought to establish the effect ofIFN on expression of complement components. We cross-referencedthe 36 genes comprising the IPA complement canonical pathwayagainst the Interferome database, a database of IRGs from a varietyof experimental settings (www.interferome.org/interferome/home.jspx). All human data on IRGs were from in vitro ex-periments; we restricted analysis to studies on human blood

A

B

Fig. 4. Complement and Fcγ-receptor genes significantly differentially expressed between progressors and nonprogressors from Zak et al.’s analysis ofRNAseq data for HIV-uninfected persons who eventually progressed to TB in comparison with matched nonprogressors from Zak et al. (15). Genes with fewerthan one read on average across all samples were filtered out, with expression analysis therefore conducted on 25,518 genes (Materials and Methods). A totalof 892 genes were significantly differentially abundant (pcorr < 0.05) between progressors 0–180 d before diagnosis and matched nonprogressors, of which362 were also >1.5-fold overabundant. A total of 610 genes were significantly differentially abundant (pcorr < 0.05) between progressors 181–360 d beforediagnosis and matched nonprogressors, of which 154 were also >1.5-fold overabundant. IPA pathway analysis was conducted on these gene lists. (A) Top fourenriched canonical pathways listed by significance (P < 0.05, Fisher’s exact) for the 362 genes and 154 genes differentially abundant at 0–180 and 181–360 dbefore diagnosis. The bar chart shows the proportion of these pathways represented. (B) Dotplots showing gene abundance (as normalized read counts) inprogressors in 180-d blocks before disease presentation in comparison with 1:1 matched nonprogressors (labeled 1–4, see Materials and Methods and DatasetS1). All components of the complement pathway or Fcγ receptor within the group of significantly differentially abundant at either

and genes showing more than twofold increase following IFNstimulation. Of the 36 genes, 9 (25%) had evidence of increasedexpression on IFN stimulation; C1QA/B/C, C2, CFB, CR1, CR2,ITGAX, and SERPING1. While abundance of some of these po-tentially IFN-regulated complement components were enriched intuberculosis others were not; in addition, C5 did not have experi-mental evidence of IFN regulation. Furthermore overall abundanceof transcripts relating to the TB relevant complement and FCGR1components (C1QA/B/C, SERPING1, C2, C5, and FCGR1A/B/C)while significantly increased in TB, was not significantly affectedby HIV status (two-way ANOVA with TB and HIV as explana-tory variables of disease risk score; r2 0.58, TB P < 0.0001 and HIVP = 0.12) with excellent discriminatory ability of these transcriptsfor TB in both HIV-infected and -uninfected persons (disease riskscore; HIV infected, AUC 0.997; HIV uninfected, AUC 0.938) (SIAppendix, Fig. S5). It is therefore unlikely that the complementpathway components up-regulated in TB merely reflects IFN-responsive genes.To establish whether this pattern of complement and FCGR

transcript expression was seen in other diseases, we analyzed thedataset [Gene Expression Omnibus (GEO) accession no.GSE42834] of Bloom et al. (18) in which participants with activeTB, active sarcoid, nonactive sarcoid, lung cancer, pneumoniaand healthy controls had whole blood transcriptional profilingundertaken. We profiled transcript abundance of the 81 comple-ment and FCGR transcripts (62 transcripts relating to the 36 com-plement genes in the IPA database and all 19 FCGR transcripts onthe platform). A total of 19 of 81 transcripts showed greater thantwofold difference in transcript abundance in at least one of thedisease categories compared with healthy controls. Initial hierar-chical clustering of disease groups for these 19 transcripts showedactive TB, clustered with active sarcoid and community-acquiredpneumonia clustered with lung cancer (SI Appendix, Fig. S6). Wethen performed K-means clustering (3 clusters, 50 iterations, Eu-clidean distance metric) to provide an unbiased assessment of thepattern of transcript abundance across the different diseases. Onecluster containing transcripts showing greatest abundance in tuber-culosis over other diseases included C1QB, SERPING1, FCGR1A,

FCGR1B, and FCGR1C. The cluster containing transcripts showinggreatest abundance in diseases other than tuberculosis (particularlypneumonia and lung cancer) included CR1, ITGAM, ITGAX,C3AR1, FCGR2A, and FCGR3B. The final cluster containedFCGRBP as a single transcript that was underabundant in tuber-culosis, community-pneumonia, and lung cancer in comparison withcontrol. These data support our findings of complement and FCGRexpression in tuberculosis, demonstrated in other cohorts presentedin the paper, and also reemphasize that complement componentexpression is not simply a function of IFN stimulation or a genericinflammatory signal.

DiscussionWe have previously shown that in a subgroup of asymptomatic,HIV-1–infected persons, conventionally diagnosed with latentTB, evidence of subclinical disease can be identified, utilizinghigh-resolution imaging (FDG-PET/CT), which is not apparentby routine clinical screening. Those with evidence of subclinicaldisease on FDG-PET/CT were also found to be more likely toprogress clinically and to require treatment for active TB (2). Inthis study, we have identified components of the active TB wholeblood transcriptional response that are also present in subclinicaldisease and shown that circulating immune complexes rise dur-ing subclinical disease.Previous transcriptional studies have consistently indicated sev-

eral key pathways in active TB (IFN-signaling/inducing genes,inflammasome/proinflammatory pathways, recognition of PAMP,the complement pathway, and Ig receptors) but these studies havenot been able to dissect the biological processes that may occur atthe outset of disease. We determined that a subset of transcripts inthe whole blood transcriptional response to symptomatic active TBwere relevant to subclinical disease and that these transcripts wereenriched for the classical complement pathway and Ig receptors(FCγ receptor 1), suggesting that a cellular response to antibody/antigen complexes may occur early in disease. This was supportedby our finding that CIC in serum increased in relation to TB dis-ease status and was highly correlated with transcript abundance ofC1q. Furthermore, with increasing metabolic activity of subclinical

Fig. 5. Comparison of modular transcript abundance by TB and HIV status. Radar plots depicting FC in transcript abundance in comparison with medianmodular transcript abundance in control group (HIV−TB−) for 29 modules. Plot (Upper) shows FC of median modular transcript abundance for HIV+ART−TB−

(green), HIV− TB+ (red), HIV+ART−TB+ (blue), and HIV+ART+TB− (pink) in comparison with HIV−TB− controls (black) overlaid on the same graph. Plots (Lower)show FC for each individual participant in the four groups in comparison with control (HIV−TB−) with the median value in bold. Log2 scale. See also SI Ap-pendix, Table S3.

E970 | www.pnas.org/cgi/doi/10.1073/pnas.1711853115 Esmail et al.

Dow

nloa

ded

by g

uest

on

Apr

il 1,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1711853115/-/DCSupplementalwww.pnas.org/cgi/doi/10.1073/pnas.1711853115

disease, the transcriptional response aligned more closely withactive TB, and levels of CIC increased. Importantly the abundanceof these transcripts and levels of CIC did not show any significantrelationship with HIV viral load.We validated our findings utilizing the publicly available data

reported by Zak et al. (15). While subclinical disease was notcharacterized by imaging, regular samples were available up to 2 ybefore clinical disease presentation in HIV-uninfected persons. Ofnote in contrast to our study, Zak et al. also excluded participantswho developed TB within 6 mo of enrollment in the training set and3 mo of enrollment in the validation set. Our analysis confirmedthat transcripts relating to the classical complement pathway and Igreceptors increased during the 12 mo before disease presentation inHIV-uninfected persons. However, in addition, IFN-signalingpathways also appear to increase in this population. While IFN-signaling pathways were prominent in our 893-transcript signa-ture, differentiating active from latent TB in HIV-infected persons,they became less prominent as this signature was filtered to allowbetter distinction of subclinical TB from latent TB. We showed thatuntreated HIV infection itself leads to increases in IFN-relatedtranscripts in whole blood. Hence, the finding that the IFN-signaling pathway transcripts were less discriminatory in earlydisease in those with untreated HIV is likely due to the highbackground level of these transcripts in this population. We alsoshowed that the pattern of complement and FCGR expressionvaried in different pulmonary diseases. While the pattern of ex-pression in tuberculosis was also seen in active sarcoidosis, thoughat a lower level of expression, a different pattern of expression wasseen in community-acquired pneumonia and lung cancer whereexpression of C3 binding receptors was more prominent. Tuber-culosis and sarcoidosis have previously been shown to have manysimilarities in whole blood transcriptional profile (18, 19).Transcripts relating to the classical complement pathway (in

particular, C1Q and SERPING1) and Fcγ receptor 1 are amongthe most consistently overabundant in active TB, and FCGBP asthe most consistently underabundant, supporting their role inearly disease. Cliff et al. (20) also reported that transcripts relatingto the complement pathway decrease within the first week oftreatment for active TB. This suggests that the abundance oftranscripts relating to the complement pathway may mark a risingantigen load, rising early in disease and falling rapidly withtreatment. CIC has frequently been reported to be elevated inactive tuberculosis, increasing with extent of disease and reducingwith treatment, and CIC has been shown to contain secreted TBantigens [e.g., 38 kDa and 30 kDa (Ag85)] and nonsecreted an-tigens [e.g., 16 kDa (HspX)] (21, 22). In addition to opsonizationand complement activation, antibodies have an additional func-tional role in regulating the immune response through interactionwith innate cells via a variety of Fc receptors, which may directlyaffect effector mechanisms and clearance of the pathogen (23–25).Hunter and coworkers (26), Hunter et al. (27), and Hunter (28)

investigated the histological appearances of early pulmonary TB inhumans in several studies. They demonstrated that the earliestfeature of disease is a lipid-rich pneumonia in which alveoli arefilled with foamy macrophages with CD4 cells present at signifi-cantly lower frequency than in the granuloma. They also observedlarge clusters of B cells bordering this region. Very few acid-fastbacilli (AFB) are visible at this stage, but abundant mycobacterialantigen is evident on immunohistochemical staining. As necrosisdevelops within the pneumonia, there is then a marked increase innumbers of AFB. Increased concentration of both 38-kDa TBantigen and 38-kDa–specific IgG have been identified in bron-choalveolar lavage samples of patients with TB (29).Based on our findings, we propose that at this early stage of

disease, the increase in mycobacterial antigen concentration re-sults in an increased concentration of antibody/antigen complex atthe site of disease. This may lead to increased expression ofcomponents of the classical complement pathway in recruited cells

(to facilitate the targeted delivery of complement) and up-regulation of Fcγ receptor on monocytes and neutrophils to pro-mote phagocytosis, as we have observed. By contrast, expression ofFCGBP is reduced. FCGBP has been shown to be the mostconsistently underabundant gene in active compared with latentTB but little is known about its function. Kobayashi et al. (30)have shown that Fcγ binding protein is present in sputum aswell as other bodily secretions and that it inhibits activation ofclassical complement pathway. Its reduction in TB disease maytherefore facilitate C1q binding of immune complexes promotingcomplement activation.Formation of immune complexes in disease has both beneficial

and detrimental effects for the host, and the complement systemcan both promote and mitigate immunopathology. Immunecomplex aggregates that precipitate and persist within the lung arelikely to be damaging through C3a/C5a-mediated neutrophil re-cruitment increasing inflammation and tissue necrosis, which maypromote bacillary multiplication (31). It has been suggested thatsolubilization and diffusion of immune complexes away from tis-sue via extracellular fluid and the systemic circulation for clear-ance by the mononuclear phagocytic system is critical to minimizelocal inflammation and tissue damage (32, 33). This appears to befacilitated by rapid C1q initiation of the classical complementpathway and C3b deposition, which results in reduction in the sizeof the immune complex aggregates to maintain solubility; defi-ciencies in classical complement components reduce immunecomplex solubility and increase risk of immune complex-mediatedautoimmune diseases (32, 33). It is possible, therefore, that theincreased expression of classical complement components in tu-berculosis is directly in response to increased production of im-mune complexes at the site of disease to allow localized delivery ofC1q to inhibit the precipitation of immune complexes and mini-mize lung damage. Although, our insight has been informed byanalysis of blood, Cai et al. (34) demonstrated that C1q mRNAand protein levels were greater at the site of disease (in pleuralfluid and bronchoalveolar lavage fluid) compared with blood inactive pleural and pulmonary tuberculosis.We found that although abundance of classical complement

components correlated with circulating immune complexes, theydid not correlate well with serum levels of the proteins they encode.There are several possible explanations of this. For blood cellsmigrating to the disease site, expression as well as consumption ofcomplement may occur locally. In addition, complement compo-nents may also be bound to immune complexes, whereas the assaywould only measure unbound complement proteins. Furthermore,the whole blood transcriptome predominantly reflects mRNA ex-pression in leukocytes, whereas circulating complement proteinsare manufactured within the liver and by a number of other celltypes (e.g., epithelial and endothelial cells) affecting the correlationof whole blood RNA and serum protein (35).There are limitations to this study. We used a broad definition

of subclinical disease, which grouped together several distinctabnormalities found to be consistent with early disease, with ref-erence to historical autopsy studies. Our sample size was modestand participants often had a combination of abnormalities andhence we were unable to tease out transcriptional differencesbetween participants with potentially different disease pathogen-esis. For example, there may be differences between those pre-senting with multiple nodules, suggestive of hematogeneousspread of disease, in comparison with disease spreading bron-chogenically, which may be interesting to characterize. Largerstudies will be needed to explore this in greater detail and furthervalidate findings. In addition, we used peripheral blood to in-vestigate pathogenesis of disease; while this approach is widelyadopted, confirmation of our findings is needed in samples fromsite of disease. Our study was conducted in ART-naïve, HIV-infected participants with well-preserved CD4 counts with vali-dation in HIV-uninfected adolescents; confirmatory studies in

Esmail et al. PNAS | Published online January 16, 2018 | E971

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

Apr

il 1,

202

1

other populations are required. Furthermore, we demonstratedthat CIC rises in relation to disease state in tuberculosis, althoughthis may have a role as a biomarker for identification of subclinicaldisease, further studies will be needed to confirm these findings inindependent cohorts.Our results indicate that the abundance of transcripts relating to

the classical complement pathway and Fcγ receptors is increasedfrom early disease in both untreated HIV-infected adults andHIV-uninfected adolescents in contrast to transcripts related toIFN signaling, which appear less discriminatory in untreated HIV-infected people. This will be an important consideration in thedevelopment of universal transcript-based biomarkers that arepredictive of disease. Our findings support the observation that inearly disease, though paucibacillary, antigen may still be present.This may have the potential to modulate the immune response byforming antibody complexes that initiate the complement pathwayand bind Fcγ receptors up-regulated on cells. However, furtherstudies will be needed to clarify in TB, how immune complexesform, their composition, and the role they play in early disease.Developing a greater understanding of the early events in TBdisease will be important to future TB control efforts. Not allthose who initially reactivate latent infection and develop evidenceof early subclinical disease will go on to develop symptomaticclinical disease, as regression will occur in a proportion. De-termining the factors that govern progression or regression ofdisease will allow us to develop rational interventions that maylead to more favorable outcomes and prevent disease.

Materials and MethodsRecruitment of Participants. Ethical approval for this study was provided bythe research ethics committees of the University of Cape Town (013/2011) andStellenbosch University (N12/11/079). Recruitment and classification of par-ticipants whose samples were utilized for this study have been described indetail previously (2). Briefly, all participants and controls were HIV-1–infectedadults, who were ART naïve and residents in Khayelitsha, a periurban town-ship of Cape Town, South Africa, where >95% of people are of Xhosa origin.Recruitment took place between 2011 and 2013; TB incidence at the time ofthe study was >1,000/100,000. All consent documents were provided to po-tential participants in English or Xhosa and read through with a member ofthe study team to ensure full understanding and capacity to consent beforesigning. Participants recruited to undergo PET/CT were recruited in a two-stage process. At initial screening consent, they were provided with an in-formation leaflet in English or Xhosa explaining the study procedures and therisks and benefits of study involvement. At the end of the screening period,this leaflet was explained in detail with eligible participants and any queries orconcerns addressed before final consenting for study entry.

The 35 asymptomatic participants undergoing FDG-PET/CT had bloodsampling for RNA (Tempus, Thermo Fisher Scientific) and serum on the day ofscan before receiving any IPT. The 15 HIV-1–infected controls with symp-tomatic, microbiologically confirmed pulmonary tuberculosis were age, sex,and CD4 matched to the 35 asymptomatic participants undergoing FDG-PET/CT and had blood sampling for RNA and serum within 24 h of commencingTB treatment, but did not undergo FDG-PET/CT.

Additional participants were recruited as controls for modular analysis. Allwere residents in Khayelitsha. All participants were of African ancestry (al-most entirely of Xhosa origin). Symptomatic HIV-1–infected andHIV-1–uninfectedparticipants with active TB had their diagnosis confirmed by either a positivesputum culture for Mtb or a positive sputum Xpert MTB-RIF (Cepheid); sputumsmear status was also established. Blood sampling of active TB controls wascarried out before, or within 24 h of treatment commencing. HIV-1–infectedparticipants on ART were established on medication for >6 mo and had asuppressed viral load (

RT-PCR Arrays. RNA extracted from whole blood used for the microarrays wasreverse transcribed using the RT2 First Strand Kit (Qiagen), which includes agenomic DNA (gDNA) elimination step. cDNA was analyzed using 384-wellRT2 Profiler Custom PCR Arrays (Qiagen) on the Roche 480 platform. Thearrays included four housekeeping (HK) genes (HPRT1, B2M, RPLP0, andHSP90AB1), three control reactions to monitor gDNA contamination andreverse transcription efficiency, and the genes of interest (including C1QB,FCGBP, FCGR1, and SERPING1). Delta Ct (ΔCt) was calculated according tothe average Ct of the four HK genes. Delta delta Ct (ΔΔCt) was calculated asthe sample ΔCT-median ΔCT of all samples, and FC was calculated as 2(−ΔΔCt).Log2-normalized FC values (mean = 0, variance = 1) were plotted as boxplotsin Qlucore Omics Explorer.

Serum Analysis. Blood was drawn into 5-mL SST II Advance tubes (BD Diag-nostics), transported to the laboratory, and centrifuged at 1,200 × g for10 min. The serum was aliquoted and stored at −80 °C. Two of 15 active TBparticipants had no serum available. All 35 participants that underwent FDG-PET/CT had serum available. Analytes were measured by ELISA: SerpinG1(Cusabio), circulating immune complex (Bühlmann), C1q (Abcam), and C5(assay measures uncleaved C5 and C5a) (Abcam). ELISA plates were read on aBio-Rad iMark microplate reader (Bio-Rad) with standard curves generatedfor each analyte. Values that were below manufacturer-determined minimaldetectable limit or the minimum asymptote of the standard curve wereassigned a zero value.

Statistical Analysis.Apart frommicroarray data where analysis was conductedprimarily in GeneSpring GX version 12.6 (Agilent) and by Ingenuity PathwayAnalysis software (Qiagen) as described above, statistical analysis of con-

tinuous and categorical variables, correlation analysis, and data visualizationwere conducted in Stata ver. 12.1 (Stata Corp) and Prism ver. 7.0c (GraphPadsoftware). The normality of data was assessed by the Shapiro–Wilk test andvariance was compared by F test or Bartlett’s test. Nonparametric data werecompared using the Mann–Whitney u test or the Kruskal–Wallis test andparametric data were compared using t test or ANOVA. Two-stage linearstep-up procedure of Benjamini, Krieger, and Yekutieli was to control forfalse discovery rate following the Kruskal–Wallis test if needed. Correlationwas performed by Spearman’s rho or Pearson’s test, depending on distri-bution of data. Proportions were compared by χ2 test or Fisher’s exact test (ifthe contingency included a number ≤5).

ACKNOWLEDGMENTS. We thank all participants and clinic and laboratorystaff for invaluable assistance and Douglas B. Young (Francis Crick Institute)for early discussion around study design and analysis. This work was fundedby the Wellcome Trust (Awards 090170, 203135, and 104803), the Bill andMelinda Gates Foundation/Wellcome Trust Grand Challenges in GlobalHealth (Award 37822), the Intramural Research Program of NIH/NationalInstitute of Allergy and Infectious Diseases and NIH Grant R01 HL106804.R.J.W. is supported by the Francis Crick Institute that receives its core fundingfrom Cancer Research UK (Award FC00110218), the UK Medical ResearchCouncil (Award FC00110218), and the Wellcome Trust (Award FC00110218).R.J.W. also received support from the European Union (Award FP7 HEALTH F3-2012-305578), National Research Foundation of South Africa (Award 96841),and Medical Research Council of South Africa (Award SHIP-02-2013). A.O. isalso supported by the Francis Crick Institute. A.K.C. is supported by the MedicalResearch Council of South Africa (Award SHIP-02-2013) and the NIH TB Re-search Unit 1U19AI111276.

1. Esmail H, Barry CE, 3rd, Young DB, Wilkinson RJ (2014) The ongoing challenge oflatent tuberculosis. Philos Trans R Soc Lond B Biol Sci 369:20130437.

2. Esmail H, et al. (2016) Characterization of progressive HIV-associated tuberculosisusing 2-deoxy-2-[18F]fluoro-D-glucose positron emission and computed tomography.Nat Med 22:1090–1093.

3. Loveday M, et al. (2017) Drug-resistant tuberculosis in patients with minimal symp-toms: Favourable outcomes in the absence of treatment. Int J Tuberc Lung Dis 21:556–563.

4. Lin PL, et al. (2014) Sterilization of granulomas is common in active and latent tu-berculosis despite within-host variability in bacterial killing. Nat Med 20:75–79.

5. Hunter RL (2011) Pathology of post primary tuberculosis of the lung: An illustratedcritical review. Tuberculosis (Edinb) 91:497–509.

6. Steinbrück P, Dănkovã D, Edwards LB, Doster B, Livesay VT (1972) Tuberculosis risk inpersons with “fibrotic” x-ray lesions. Bull Int Union Tuberc 47:135–159.

7. Koppaka R, Bock N (2004) How reliable is chest radiography? Toman’s TuberculosisCase Detection, Treatment, and Monitoring: Questions and Answers, ed Frieden T(WHO, Geneva), pp 51–60.

8. Mamede M, et al. (2005) [18F]FDG uptake and PCNA, Glut-1, and hexokinase-II ex-pressions in cancers and inflammatory lesions of the lung. Neoplasia 7:369–379.

9. Jacobsen M, et al. (2007) Candidate biomarkers for discrimination between infectionand disease caused by Mycobacterium tuberculosis. J Mol Med (Berl) 85:613–621.

10. Maertzdorf J, et al. (2011) Human gene expression profiles of susceptibility and re-sistance in tuberculosis. Genes Immun 12:15–22.

11. Berry MP, et al. (2010) An interferon-inducible neutrophil-driven blood transcrip-tional signature in human tuberculosis. Nature 466:973–977.

12. Kaforou M, et al. (2013) Detection of tuberculosis in HIV-infected and -uninfectedAfrican adults using whole blood RNA expression signatures: A case-control study.PLoS Med 10:e1001538.

13. Blankley S, et al. (2016) A 380-gene meta-signature of active tuberculosis comparedwith healthy controls. Eur Respir J 47:1873–1876.

14. Blankley S, et al. (2016) The transcriptional signature of active tuberculosis reflectssymptom status in extra-pulmonary and pulmonary tuberculosis. PLoS One 11:e0162220.

15. Zak DE, et al.; ACS and GC6-74 cohort study groups (2016) A blood RNA signature fortuberculosis disease risk: A prospective cohort study. Lancet 387:2312–2322.

16. Walport MJ (2001) Complement. Second of two parts. N Engl J Med 344:1140–1144.17. Chaussabel D, et al. (2008) A modular analysis framework for blood genomics studies:

Application to systemic lupus erythematosus. Immunity 29:150–164.18. Bloom CI, et al. (2013) Transcriptional blood signatures distinguish pulmonary tu-

berculosis, pulmonary sarcoidosis, pneumonias and lung cancers. PLoS One 8:e70630,and erratum (2013) 8, 10.1371/annotation/7d9ec449-aee0-48fe-8111-0c110850c0c1.

19. Maertzdorf J, et al.; TBornotTB Network (2012) Common patterns and disease-relatedsignatures in tuberculosis and sarcoidosis. Proc Natl Acad Sci USA 109:7853–7858.

20. Cliff JM, et al. (2013) Distinct phases of blood gene expression pattern through tu-berculosis treatment reflect modulation of the humoral immune response. J Infect Dis207:18–29.

21. Raja A, Ranganathan UD, Ramalingam B (2006) Clinical value of specific detection ofimmune complex-bound antibodies in pulmonary tuberculosis. Diagn Microbiol InfectDis 56:281–287.

22. Raja A, Narayanan PR, Mathew R, Prabhakar R (1995) Characterization of mycobac-terial antigens and antibodies in circulating immune complexes from pulmonary tu-berculosis. J Lab Clin Med 125:581–587.

23. Lu LL, et al. (2016) A functional role for antibodies in tuberculosis. Cell 167:433–443.e14.

24. Kozakiewicz L, Phuah J, Flynn J, Chan J (2013) The role of B cells and humoral im-munity in Mycobacterium tuberculosis infection. Adv Exp Med Biol 783:225–250.

25. Nimmerjahn F, Ravetch JV (2008) Fcgamma receptors as regulators of immune re-sponses. Nat Rev Immunol 8:34–47.

26. Welsh KJ, Risin SA, Actor JK, Hunter RL (2011) Immunopathology of postprimarytuberculosis: Increased T-regulatory cells and DEC-205-positive foamy macrophages incavitary lesions. Clin Dev Immunol 2011:307631.

27. Hunter RL, Actor JK, Hwang SA, Karev V, Jagannath C (2014) Pathogenesis of postprimary tuberculosis: Immunity and hypersensitivity in the development of cavities.Ann Clin Lab Sci 44:365–387.

28. Hunter RL (2016) Tuberculosis as a three-act play: A new paradigm for the patho-genesis of pulmonary tuberculosis. Tuberculosis (Edinb) 97:8–17.

29. Raja A, Baughman RP, Daniel TM (1988) The detection by immunoassay of antibodyto mycobacterial antigens and mycobacterial antigens in bronchoalveolar lavage fluidfrom patients with tuberculosis and control subjects. Chest 94:133–137.

30. Kobayashi K, et al. (2002) Distribution and partial characterisation of IgG Fc bindingprotein in various mucin producing cells and body fluids. Gut 51:169–176.

31. Larsen GL, Mitchell BC, Henson PM (1981) The pulmonary response of C5 sufficientand deficient mice to immune complexes. Am Rev Respir Dis 123:434–439.

32. Schifferli JA, Ng YC, Peters DK (1986) The role of complement and its receptor in theelimination of immune complexes. N Engl J Med 315:488–495.

33. Schifferli JA, Peters DK (1983) Complement, the immune-complex lattice, and thepathophysiology of complement-deficiency syndromes. Lancet 2:957–959.

34. Cai Y, et al. (2014) Increased complement C1q level marks active disease in humantuberculosis. PLoS One 9:e92340.

35. Morgan BP, Gasque P (1997) Extrahepatic complement biosynthesis: Where, whenand why? Clin Exp Immunol 107:1–7.

36. Obermoser G, et al. (2013) Systems scale interactive exploration reveals quantitativeand qualitative differences in response to influenza and pneumococcal vaccines.Immunity 38:831–844.

37. Love MI, Huber W, Anders S (2014) Moderated estimation of fold change and dis-persion for RNA-seq data with DESeq2. Genome Biol 15:550.

Esmail et al. PNAS | Published online January 16, 2018 | E973

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

Apr

il 1,

202

1