Mortality in Severe Human Immunodeficiency Virus-Tuberculosis Associates With Innate Immune Activation and Dysfunction of Monocytes

Summary Mortality remains unacceptably high in patients hospitalized with human immunodeficiency virus–associated tuberculosis. Among these patients, half had mycobacteremia. Death was associated with higher concentrations of immune activation and anti-inflammatory markers, and impaired ex vivo innate responses to bacterial antigens.

We hypothesized that hospitalized HIV-tuberculosis patients, compared with HIV-infected outpatients without tuberculosis (controls), would have elevation of sepsis biomarkers and proinflammatory cytokines as well as evidence of increased antiinflammatory signalling and decreased innate immune responses to bacterial stimuli and that these features would be associated with mortality. Second, we hypothesized that tuberculosis mycobacteremia would be associated with these immunological features and mortality. Additionally, we tested whether ex vivo treatment with recombinant interferon (IFN)-ɣ (a potential host-directed therapy) could restore monocyte responses to LPS.

Study Design and Population
We conducted a prospective cohort study in Khayelitsha, Cape Town, South Africa. Khayelitsha has a tuberculosis notification rate of 1065/100 000 person-years (City of Cape Town, unpublished data) and antenatal HIV seroprevalence of 34% [15].
Nonpregnant HIV-infected patients with CD4 counts <350 cells/µL and diagnosed with tuberculosis on admission to Khayelitsha Hospital were recruited between June 2014 and October 2014. Patients were excluded if they were transfused or had received more than 1 dose of tuberculosis treatment within the preceding month. Patients with microbiologically proven rifampicin-susceptible tuberculosis were included. Selection bias was minimized by using a random selection procedure.
HIV-infected outpatients with CD4 counts <350 cells/µL without active tuberculosis were recruited at Úbuntu Clinic, Khayelitsha, Cape Town (controls). To ensure more appropriate matching, only control patients with CD4 counts <150 cells/µL were included in the final analyses.

Outcomes and Definitions
We aimed to determine immunologic changes associated with the primary outcome of 12-week mortality. Second, we assessed associations of tuberculosis mycobacteremia, defined as at least 1 MycoFlytic blood culture growing Mtb, with immunologic profile and outcome. Sepsis definitions were adapted from published criteria (Supplementary Table 1) [9].

Procedures
A detailed description of ethical aspects and data collection is provided in the Supplementary Methods. All participants had the following tests performed: Xpert MTB/RIF (Mycobacterium tuberculosis/rifampin) assay on sputum and urine, tuberculosis culture on sputum, chest X ray, full blood count and differential, chemistry, HIV viral load, and CD4 count. MycoFlytic blood cultures were performed for hospitalized patients. Control patients were excluded if a tuberculosis symptom screen or any tuberculosis diagnostic test was positive. Hospitalized patients were contacted by telephone at 4 weeks and clinically reviewed at 12 weeks.
Whole blood was stimulated for 6 hours. Anti-(myco-)bacterial responses were tested using Escherichia coli-derived LPS, heat-killed Streptococcus pneumoniae, and heat-killed Mtb strain H37Rv. The ex vivo effect of IFN-ɣ was assessed in a costimulation assay with LPS.
Samples were stained with surface and intracellular markers (Supplementary Table 2) and acquired on a BD LSR Fortessa Flow Cytometer. Data were analyzed in FlowJo, version 10 (Ashland, Oregon). Gating strategies are illustrated in Supplementary Figure 1. Concentrations of 12 cytokines (Supplementary Table 2) were measured in culture supernatants using a Luminex multiplex assay.
Medians were compared among groups using the appropriate statistical test. Variables were investigated for associations with mortality in a Cox proportional hazards model. A prioridefined potential confounders were age, sex, ART status, HIV viral load and CD4 count, and Luminex plate number; a variable was retained in the model if introduction led to a >10% change in the effect measure.
All reported q values were calculated using Benjamini-Hochberg procedures for multiple-testing correction [16]. P values < .05 and q values <.10 were regarded as significant.
Patients for whom data were available for all stimulation conditions were included (n = 37) for principal component analysis (PCA); 90 variables were included per patient. Using a Shapiro test, all variables with P < .05 were log-transformed, and scaling was done. Data missing due to technical difficulties (112/3300 values; 3.4%) were imputed using the K-nearest-neighbor technique.

Clinical Presentation and Outcomes
Compared with controls, HIV-tuberculosis patients had higher HIV viral loads and were more often anaemic (Table 1). Also, 35/41 (85%) HIV-tuberculosis patients who could produce sputum had a positive sputum Xpert MTB/RIF or culture for Mtb; 31/60 (52%) patients grew Mtb on blood culture; 15/35 (43%) had a positive urine Xpert MTB/RIF; and 16 patients had positive cultures or Xpert MTB/RIF from other sites. In addition, 28/60 (47%) patients had standard bacterial blood cultures performed prior to receipt of broad-spectrum antibiotics; none grew pathologic bacteria. By 12 weeks, 16 HIV-tuberculosis patients had died (26.7%) a median of 12 (IQR, 0-24) days after enrollment (Supplementary Table 3); none were lost to follow-up. All deaths occurred while the patients were in the hospital. Four patients had suspected bacterial sepsis when they died, and 11 deaths were attributed to disseminated tuberculosis; 1 of them had disseminated tuberculosis and features that suggested bacterial sepsis.
Compared with HIV-tuberculosis patients who survived, patients who died were older and presented more often with severe sepsis or septic shock (  Abbreviations: ART, antiretroviral therapy; HIV, human immunodeficiency virus; ND, not determined; VL, viral load. a q value of comparisons between the HIV-infected control group and HIV-associated tuberculosis patients. Significant differences (q < .10) are in bold. b q value of comparisons between the group of patients who died and the group who survived. Significant differences (q < .10) are in bold.
HIV-tuberculosis patients who died had significantly higher concentrations of procalcitonin compared with survivors (Table 1).

Immune Phenotype in HIV-Tuberculosis Versus Controls
In unstimulated samples, patients with HIV-tuberculosis had higher percentages of tumor necrosis factor (TNF)-α+ monocytes compared with controls ( Figure 1) and higher supernatant concentrations of colony-stimulating factor (CSF)-3 and IFN-ɣ, whereas supernatant concentrations of IL-1β and IL-8 were lower (Table 2).

Pro-and Antiinflammatory Changes in Hospitalized HIV-Tuberculosis Patients Who Died Versus Survivors
Immune activation was associated with mortality. Patients who died had higher percentages of CD16+CD14+ monocytes and higher supernatant concentrations of CSF-3 in unstimulated samples compared with patients who survived ( Figure 1, Table 2). Higher percentages of activated CD16+CD14+ monocytes ( Figure 2A) were independently associated with mortality, as were supernatant concentrations of CSF-3, IL-1RA, IL-6, and TNF-α ( Figure 2).
Compared with survivors, patients who died had lower absolute counts of IL-6+ and TNF-α+ producing monocytes in response to LPS and S. pneumoniae (Figure 1). Lower supernatant IL-1β and IL-6 concentrations and higher CSF-3 concentrations were measured in S. pneumoniae stimulations of patients who died (Table 2). Impaired monocyte and TNF-α and IL-6 production in response to S. pneumoniae and lower percentages of IL-6+ neutrophils in response to LPS were independently associated with mortality ( Figure 2A, Figure 3), as were increased supernatant concentrations of CSF-3 and lower concentrations of IL-1β and TNF-ɑ in response to S. pneumoniae ( Figure 2B).
In response to heat-killed Mtb, patients who died had lower absolute counts of IL-6+ and TNF-ɑ+ neutrophils. No differences were seen in Mtb culture supernatants ( Table 2). Mtb responses were not associated with mortality ( Figure 2).
In PCA, the first principal component (PC1) was significantly associated with mortality (P = .003; Figure 4A-B). After all samples were arranged by their value on PC1, an immunological signature associated with mortality was revealed ( Figure 4C), characterized by increased production of cytokines in unstimulated conditions and impaired production of proinflammatory cytokines in response to all antigen stimuli.

Mtb Mycobacteremia Is Associated With Cytopenia but Not With Sepsis Severity
Mycobacteremic patients had lower platelet (median 203 vs 311 × 10 9 /L), lymphocyte (median 0.42 vs 0.66 × 10 9 /L), and monocyte counts (median 0.25 vs 0.42 × 10 9 /L) compared with non-mycobacteremic patients (P = .05, P = .0003, and P = .03, respectively). There were no differences in plasma concentrations of procalcitonin or lactate, percentages of CD16+CD14+ monocytes in unstimulated samples, or percentages of IL-6+ or TNF-α+ monocytes in response to stimulations (Supplementary Figure 3). Mycobacteremic patients had lower absolute counts of IL-6+ monocytes compared with non-mycobacteremic tuberculosis patients in unstimulated samples and lower absolute counts of both IL-6+ and TNF-α+ monocytes in response to LPS (Supplementary Figure 3). There were no differences in supernatant cytokine concentrations (Supplementary Table 4). None of the variables were associated with mycobacterial load, expressed in days to blood culture positivity (results not shown).

No Reversal of Immune Deficits With IFN-ɣ Costimulation
There were no differences in neutrophil and monocyte capacity to produce IL-6 or TNF-ɑ, nor in the concentration of any of the extracellular cytokines, when the LPS with IFN-ɣ costimulation was compared with LPS only.

DISCUSSION
In HIV-infected patients diagnosed while in the hospital with microbiologically proven drug-susceptible tuberculosis, 12-week mortality was 27%. More than half of patients had mycobacteremia; however, this was not associated with mortality. Patients who died had an immune phenotype characterized by higher concentrations of proinflammatory cytokines (CSF-3, TNF-ɑ, and IL-6) and antiinflammatory IL-1RA, an increased proportion of proinflammatory CD14+CD16+ monocytes, and impaired capacity of innate cells to produce proinflammatory cytokines in response to bacterial antigen stimuli.
Increased concentrations of proinflammatory cytokines CSF-3, IL-6, and TNF-ɑ in unstimulated samples were associated with mortality. These cytokines are mainly produced by innate cells, suggesting a more activated state of the innate immune system in patients who die. This is consistent with results from other studies of mortality in HIV-associated tuberculosis [14], ARTnaive patients who start ART [17], and bacterial sepsis in HIVinfected patients [18,19]. There are several plausible explanations for this association. It is possible that immune activation reflects more disseminated tuberculosis. Another possible explanation is a causal relation: higher concentrations of proinflammatory cytokines led to increased tissue damage, organ failure, and immune exhaustion, impairing host defense to other pathogens. Increased percentages of CD16+CD14+ monocytes were associated with mortality. This activated monocyte subset generally produces higher amounts of proinflammatory cytokines and is more phagocytic compared with the classic CD16 subset [20]. However, previous studies have shown that monocyte functionality may be impaired in active tuberculosis, with CD16+CD14+ monocytes refractory to dendritic cell maturation, leading to impaired antigen presentation and decreased secretion of IL-1β and IL-12 [21,22]. Impaired phagocytic and antigen-presenting capacity of CD14+CD16+ cells has also been described in bacterial sepsis. Monocyte deactivation, with reduced production of TNF-ɑ in response to LPS was associated with fatal outcome [10,23]. Our findings indicate that expansion of the CD16+ CD14+ monocyte population is observed together with impaired total monocyte and neutrophil responses to bacterial antigens in patients who die, potentially contributing to mortality.
The immunological phenotype associated with mortality is similar to what has been described in bacterial sepsis. In bacterial sepsis, time to intravenous broad-spectrum antibiotic treatment is associated inversely with survival [9]. We did not find such an association for time to tuberculosis treatment. Although a potential association might have been masked by the lack of exact data on time of tuberculosis treatment initiation in hours, a more likely explanation is the fact that tuberculosis treatment is administered orally, and drug absorption might be hampered by intestinal tissue damage by tuberculosis or HIV.
Although invasive pneumococcal disease is one of the most frequent and lethal bacterial infections in HIV-infected patients, most studies focus on LPS responses. Interestingly, we found that impaired responses to S. pneumoniae were also strongly associated with mortality, indicating defects in host defense to this pathogen.
Reduced proinflammatory responses of innate cells to LPS and S. pneumoniae in HIV-tuberculosis patients, compared with controls, suggest that tuberculosis has an immunosuppressive effect that is additive to HIV. CD4 count and HIV viral load were not associated with mortality, whereas the innate immune features were independently associated.
Tuberculosis mycobacteremia was neither associated with mortality nor with more severe derangement of sepsis biomarkers. Mycobacteremia was associated with cytopenia, but there were no functional differences of innate cells. Previous studies have shown that in hospitalized febrile patients, mycobacteremia was associated with mortality [5,6,24], whereas in other studies among HIV-tuberculosis patients with mycobacteremia, it was not [7,8]. Our findings illustrate that patients with severe HIV-tuberculosis can develop features of a sepsis syndrome even when mycobacterial blood cultures are negative.
There is increasing interest in host-directed immunotherapies for tuberculosis [25]. Recombinant IFN-ɣ has been shown to be beneficial in the treatment of cryptococcal meningitis [26] and other fungal infections [27], and a trial to investigate its application for bacterial sepsis is ongoing [28]. We found no effect of ex vivo costimulation with recombinant IFN-ɣ in restoring monocyte responses to LPS. Although ex vivo data cannot be directly extrapolated to in vivo conditions, our findings do not support investigation of recombinant IFN-ɣ as a potential immunotherapy in this patient subset. Our data are supported by 2 clinical trials [25,29] that showed no beneficial effects of IFN-ɣ on sputum culture conversion in drug-sensitive or drug-resistant pulmonary tuberculosis. CSF-3 was investigated as adjunctive immune therapy for bacterial sepsis, but there was no significant survival benefit [30]. Our data of increased, rather than decreased, concentrations of CSF-3 in patients who die do not support investigation of CSF-3 for host-directed therapy in HIV-tuberculosis.
The early mortality [2] and prevalence of mycobacteremia [6,7,24] reported here are similar to that found in studies from Africa and suggest our immunological findings are generalizable to other settings with a high tuberculosis and HIV burden. Limitations of our study include a limited number of bacterial blood cultures and lack of post-mortem examinations. Due to the strategy of treating patients with sepsis syndrome with broad-spectrum antibiotics at primary care referral centers upon referral to the hospital, only 47% of patients had bacterial blood cultures performed prior to receipt of antibiotics. Postmortem examinations were planned; however, in this study, none of the families agreed to this. Ex vivo markers of immune exhaustion were not measured, and this should be the subject of future studies.
Major strengths of our study are the variety of antigens/ organisms used for stimulations, enabling in vitro simulation of gram-negative, gram-positive, and mycobacterial infections. The inclusion of an HIV-infected control group without active tuberculosis and the exclusion of patients without microbiologically confirmed tuberculosis facilitate conclusions on the associations of findings with severe tuberculosis specifically, minimizing misclassification bias of other diagnoses or advanced HIV alone.
Our study has several implications for clinical care and future research. The immune profile observed in HIVtuberculosis patients, particularly those who died, suggests that disseminated tuberculosis in the context of advanced HIV infection can significantly impair host innate responses to bacteria, possibly resulting in an immunological predisposition to bacterial superinfections [3,4]. This study adds to our understanding of immunopathology in HIV-tuberculosis. By focusing on patients who required hospital admission and innate immune changes, we confirmed that the high mortality in this patient subset is associated with an immunological phenotype similar to bacterial sepsis. Immune activation, with higher concentrations of proinflammatory cytokines and expansion of the CD16+CD14+ monocyte population, is potentially leading to increased tissue damage. Simultaneously, there is impairment of innate immune functional responses, with reduced production of proinflammatory cytokines in response to antigens of bacterial pathogens. In the future, immunomodulatory interventions that have proven beneficial in bacterial sepsis should also be evaluated for patients with severe HIV-tuberculosis.