Discovery of a structural class of antibiotics with explainable deep learning.

The discovery of novel structural classes of antibiotics is urgently needed to address the ongoing antibiotic resistance crisis1-9. Deep learning approaches have aided in exploring chemical spaces1,10-15; these typically use black box models and do not provide chemical insights. Here we reasoned that the chemical substructures associated with antibiotic activity learned by neural network models can be identified and used to predict structural classes of antibiotics. We tested this hypothesis by developing an explainable, substructure-based approach for the efficient, deep learning-guided exploration of chemical spaces. We determined the antibiotic activities and human cell cytotoxicity profiles of 39,312 compounds and applied ensembles of graph neural networks to predict antibiotic activity and cytotoxicity for 12,076,365 compounds. Using explainable graph algorithms, we identified substructure-based rationales for compounds with high predicted antibiotic activity and low predicted cytotoxicity. We empirically tested 283 compounds and found that compounds exhibiting antibiotic activity against Staphylococcus aureus were enriched in putative structural classes arising from rationales. Of these structural classes of compounds, one is selective against methicillin-resistant S. aureus (MRSA) and vancomycin-resistant enterococci, evades substantial resistance, and reduces bacterial titres in mouse models of MRSA skin and systemic thigh infection. Our approach enables the deep learning-guided discovery of structural classes of antibiotics and demonstrates that machine learning models in drug discovery can be explainable, providing insights into the chemical substructures that underlie selective antibiotic activity.

Lactobacillus-Deficient Cervicovaginal Bacterial Communities Are Associated with Increased HIV Acquisition in Young South African Women.

Elevated inflammation in the female genital tract is associated with increased HIV risk. Cervicovaginal bacteria modulate genital inflammation; however, their role in HIV susceptibility has not been elucidated. In a prospective cohort of young, healthy South African women, we found that individuals with diverse genital bacterial communities dominated by anaerobes other than Gardnerella were at over 4-fold higher risk of acquiring HIV and had increased numbers of activated mucosal CD4+ T cells compared to those with Lactobacillus crispatus-dominant communities. We identified specific bacterial taxa linked with reduced (L. crispatus) or elevated (Prevotella, Sneathia, and other anaerobes) inflammation and HIV infection and found that high-risk bacteria increased numbers of activated genital CD4+ T cells in a murine model. Our results suggest that highly prevalent genital bacteria increase HIV risk by inducing mucosal HIV target cells. These findings might be leveraged to reduce HIV acquisition in women living in sub-Saharan Africa.

Cervicovaginal bacteria are a major modulator of host inflammatory responses in the female genital tract.

Colonization by Lactobacillus in the female genital tract is thought to be critical for maintaining genital health. However, little is known about how genital microbiota influence host immune function and modulate disease susceptibility. We studied a cohort of asymptomatic young South African women and found that the majority of participants had genital communities with low Lactobacillus abundance and high ecological diversity. High-diversity communities strongly correlated with genital pro-inflammatory cytokine concentrations in both cross-sectional and longitudinal analyses. Transcriptional profiling suggested that genital antigen-presenting cells sense gram-negative bacterial products in situ via Toll-like receptor 4 signaling, contributing to genital inflammation through activation of the NF-κB signaling pathway and recruitment of lymphocytes by chemokine production. Our study proposes a mechanism by which cervicovaginal microbiota impact genital inflammation and thereby might affect a woman's reproductive health, including her risk of acquiring HIV.

Distinguishing Severe Acute Respiratory Syndrome Coronavirus 2 Persistence and Reinfection: A Retrospective Cohort Study.

BACKGROUND: Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) reinfection is poorly understood, partly because few studies have systematically applied genomic analysis to distinguish reinfection from persistent RNA detection related to initial infection. We aimed to evaluate the characteristics of SARS-CoV-2 reinfection and persistent RNA detection using independent genomic, clinical, and laboratory assessments. METHODS: All individuals at a large academic medical center who underwent a SARS-CoV-2 nucleic acid amplification test (NAAT) ≥45 days after an initial positive test, with both tests between 14 March and 30 December 2020, were analyzed for potential reinfection. Inclusion criteria required having ≥2 positive NAATs collected ≥45 days apart with a cycle threshold (Ct) value <35 at repeat testing. For each included subject, likelihood of reinfection was assessed by viral genomic analysis of all available specimens with a Ct value <35, structured Ct trajectory criteria, and case-by-case review by infectious diseases physicians. RESULTS: Among 1569 individuals with repeat SARS-CoV-2 testing ≥45 days after an initial positive NAAT, 65 (4%) met cohort inclusion criteria. Viral genomic analysis characterized mutations present and was successful for 14/65 (22%) subjects. Six subjects had genomically supported reinfection, and 8 subjects had genomically supported persistent RNA detection. Compared to viral genomic analysis, clinical and laboratory assessments correctly distinguished reinfection from persistent RNA detection in 12/14 (86%) subjects but missed 2/6 (33%) genomically supported reinfections. CONCLUSIONS: Despite good overall concordance with viral genomic analysis, clinical and Ct value-based assessments failed to identify 33% of genomically supported reinfections. Scaling-up genomic analysis for clinical use would improve detection of SARS-CoV-2 reinfections.

Prediction of Antimicrobial Resistance in Clinical Enterococcus faecium Isolates Using a Rules-Based Analysis of Whole-Genome Sequences.

Enterococcus faecium is a major cause of clinical infections, often due to multidrug-resistant (MDR) strains. Whole-genome sequencing (WGS) is a powerful tool to study MDR bacteria and their antimicrobial resistance (AMR) mechanisms. In this study, we used WGS to characterize E. faecium clinical isolates and test the feasibility of rules-based genotypic prediction of AMR. Clinical isolates were divided into derivation and validation sets. Phenotypic susceptibility testing for ampicillin, vancomycin, high-level gentamicin, ciprofloxacin, levofloxacin, doxycycline, tetracycline, and linezolid was performed using the Vitek 2 automated system, with confirmation and discrepancy resolution by broth microdilution, disk diffusion, or gradient diffusion when needed. WGS was performed to identify isolate lineage and AMR genotype. AMR prediction rules were derived by analyzing the genotypic-phenotypic relationship in the derivation set. Phylogenetic analysis demonstrated that 88% of isolates in the collection belonged to hospital-associated clonal complex 17. Additionally, 12% of isolates had novel sequence types. When applied to the validation set, the derived prediction rules demonstrated an overall positive predictive value of 98% and negative predictive value of 99% compared to standard phenotypic methods. Most errors were falsely resistant predictions for tetracycline and doxycycline. Further analysis of genotypic-phenotypic discrepancies revealed potentially novel pbp5 and tet(M) alleles that provide insight into ampicillin and tetracycline class resistance mechanisms. The prediction rules demonstrated generalizability when tested on an external data set. In conclusion, known AMR genes and mutations can predict E. faecium phenotypic susceptibility with high accuracy for most routinely tested antibiotics, providing opportunities for advancing molecular diagnostics.

Changes in concentrations of cervicovaginal immune mediators across the menstrual cycle: a systematic review and meta-analysis of individual patient data.

BACKGROUND: Hormonal changes during the menstrual cycle play a key role in shaping immunity in the cervicovaginal tract. Cervicovaginal fluid contains cytokines, chemokines, immunoglobulins, and other immune mediators. Many studies have shown that the concentrations of these immune mediators change throughout the menstrual cycle, but the studies have often shown inconsistent results. Our understanding of immunological correlates of the menstrual cycle remains limited and could be improved by meta-analysis of the available evidence. METHODS: We performed a systematic review and meta-analysis of cervicovaginal immune mediator concentrations throughout the menstrual cycle using individual participant data. Study eligibility included strict definitions of the cycle phase (by progesterone or days since the last menstrual period) and no use of hormonal contraception or intrauterine devices. We performed random-effects meta-analyses using inverse-variance pooling to estimate concentration differences between the follicular and luteal phases. In addition, we performed a new laboratory study, measuring select immune mediators in cervicovaginal lavage samples. RESULTS: We screened 1570 abstracts and identified 71 eligible studies. We analyzed data from 31 studies, encompassing 39,589 concentration measurements of 77 immune mediators made on 2112 samples from 871 participants. Meta-analyses were performed on 53 immune mediators. Antibodies, CC-type chemokines, MMPs, IL-6, IL-16, IL-1RA, G-CSF, GNLY, and ICAM1 were lower in the luteal phase than the follicular phase. Only IL-1α, HBD-2, and HBD-3 were elevated in the luteal phase. There was minimal change between the phases for CXCL8, 9, and 10, interferons, TNF, SLPI, elafin, lysozyme, lactoferrin, and interleukins 1ß, 2, 10, 12, 13, and 17A. The GRADE strength of evidence was moderate to high for all immune mediators listed here. CONCLUSIONS: Despite the variability of cervicovaginal immune mediator measurements, our meta-analyses show clear and consistent changes during the menstrual cycle. Many immune mediators were lower in the luteal phase, including chemokines, antibodies, matrix metalloproteinases, and several interleukins. Only interleukin-1α and beta-defensins were higher in the luteal phase. These cyclical differences may have consequences for immunity, susceptibility to infection, and fertility. Our study emphasizes the need to control for the effect of the menstrual cycle on immune mediators in future studies.

Applications of Machine Learning to the Problem of Antimicrobial Resistance: an Emerging Model for Translational Research.

Antimicrobial resistance (AMR) remains one of the most challenging phenomena of modern medicine. Machine learning (ML) is a subfield of artificial intelligence that focuses on the development of algorithms that learn how to accurately predict outcome variables using large sets of predictor variables that are typically not hand selected and are minimally curated. Models are parameterized using a training data set and then applied to a test data set on which predictive performance is evaluated. The application of ML algorithms to the problem of AMR has garnered increasing interest in the past 5 years due to the exponential growth of experimental and clinical data, heavy investment in computational capacity, improvements in algorithm performance, and increasing urgency for innovative approaches to reducing the burden of disease. Here, we review the current state of research at the intersection of ML and AMR with an emphasis on three domains of work. The first is the prediction of AMR using genomic data. The second is the use of ML to gain insight into the cellular functions disrupted by antibiotics, which forms the basis for understanding mechanisms of action and developing novel anti-infectives. The third focuses on the application of ML for antimicrobial stewardship using data extracted from the electronic health record. Although the use of ML for understanding, diagnosing, treating, and preventing AMR is still in its infancy, the continued growth of data and interest ensures it will become an important tool for future translational research programs.

Clinical sensitivity and interpretation of PCR and serological COVID-19 diagnostics for patients presenting to the hospital.

The diagnosis of COVID-19 requires integration of clinical and laboratory data. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) diagnostic assays play a central role in diagnosis and have fixed technical performance metrics. Interpretation becomes challenging because the clinical sensitivity changes as the virus clears and the immune response emerges. Our goal was to examine the clinical sensitivity of two most common SARS-CoV-2 diagnostic test modalities, polymerase chain reaction (PCR) and serology, over the disease course to provide insight into their clinical interpretation in patients presenting to the hospital. We conducted a single-center, retrospective study. To derive clinical sensitivity of PCR, we identified 209 PCR-positive SARS-CoV-2 patients with multiple PCR test results (624 total PCR tests) and calculated daily sensitivity from date of symptom onset or first positive test. Clinical sensitivity of PCR decreased with days post symptom onset with >90% clinical sensitivity during the first 5 days after symptom onset, 70%-71% from Days 9 to 11, and 30% at Day 21. To calculate daily clinical sensitivity by serology, we utilized 157 PCR-positive patients with a total of 197 specimens tested by enzyme-linked immunosorbent assay for IgM, IgG, and IgA anti-SARS-CoV-2 antibodies. In contrast to PCR, serological sensitivity increased with days post symptom onset with >50% of patients seropositive by at least one antibody isotype after Day 7, >80% after Day 12, and 100% by Day 21. Taken together, PCR and serology are complimentary modalities that require time-dependent interpretation. Superimposition of sensitivities over time indicate that serology can function as a reliable diagnostic aid indicating recent or prior infection.

Use of Routine Complete Blood Count Results to Rule Out Anaplasmosis Without the Need for Specific Diagnostic Testing.

BACKGROUND: Anaplasmosis presents with fever, headache, and laboratory abnormalities including leukopenia and thrombocytopenia. Polymerase chain reaction (PCR) is the preferred diagnostic but is overutilized. We determined if routine laboratory tests could exclude anaplasmosis, improving PCR utilization. METHODS: Anaplasma PCR results from a 3-year period, with associated complete blood count (CBC) and liver function test results, were retrospectively reviewed. PCR rejection criteria, based on white blood cell (WBC) and platelet (PLT) counts, were developed and prospectively applied in a mock stewardship program. If rejection criteria were met, a committee mock-refused PCR unless the patient was clinically unstable or immunocompromised. RESULTS: WBC and PLT counts were the most actionable routine tests for excluding anaplasmosis. Retrospective review demonstrated that rejection criteria of WBC ≥11 000 cells/µL or PLT ≥300 000 cells/µL would have led to PCR refusal in 428 of 1685 true-negative cases (25%) and 3 of 66 true-positive cases (5%) involving clinically unstable or immunocompromised patients. In the prospective phase, 155 of 663 PCR requests (23%) met rejection criteria and were reviewed by committee, which endorsed refusal in 110 of 155 cases (71%) and approval in 45 (29%), based on clinical criteria. PCR was negative in all 45 committee-approved cases. Only 1 of 110 mock-refused requests yielded a positive PCR result; this patient was already receiving doxycycline at the time of testing. CONCLUSIONS: A CBC-based stewardship algorithm would reduce unnecessary Anaplasma PCR testing, without missing active cases. Although the prospectively evaluated screening approach involved medical record review, this was unnecessary to prevent errors and could be replaced by a rejection comment specifying clinical situations that might warrant overriding the algorithm.

Evaluation of Three Commercial SARS-CoV-2 Serologic Assays and Their Performance in Two-Test Algorithms.

Sensitive and specific severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) serologic assays are needed to inform diagnostic, therapeutic, and public health decision-making. We evaluated three commercial serologic assays as stand-alone tests and as components of two-test algorithms. Two nucleocapsid antibody tests (Abbott IgG and Roche total antibody) and one spike protein antibody test (DiaSorin IgG) were included. We assessed sensitivity using 128 serum samples from symptomatic PCR-confirmed coronavirus disease 2019 (COVID-19)-infected patients and specificity using 1,204 samples submitted for routine serology prior to COVID-19's emergence, plus 64 pandemic-era samples from SARS-CoV-2 PCR-negative patients with respiratory symptoms. Assays were evaluated as stand-alone tests and as components of a two-test algorithm in which positive results obtained using one assay were verified using a second assay. The two nucleocapsid antibody tests were more sensitive than the spike protein antibody test overall (70% and 70% versus 57%; P ≤ 0.003), with pronounced differences observed using samples collected 7 to 14 days after symptom onset. All three assays were comparably sensitive (≥89%; P ≥ 0.13) using samples collected >14 days after symptom onset. Specificity was higher using the nucleocapsid antibody tests (99.3% and 99.7%) than using the spike protein antibody test (97.8%; P ≤ 0.002). When any two assays were paired in a two-test algorithm, the specificity was 99.9% (P < 0.0001 to 0.25 compared with the individual assays), and the positive predictive value (PPV) improved substantially, with a minimal effect on the negative predictive value (NPV). In conclusion, two nucleocapsid antibody tests outperformed a spike protein antibody test. Pairing two different serologic tests in a two-test algorithm improves the PPV, compared with the individual assays alone, while maintaining the NPV.

Rapid, efficient functional characterization and recovery of HIV-specific human CD8+ T cells using microengraving.

The nature of certain clinical samples (tissue biopsies, fluids) or the subjects themselves (pediatric subjects, neonates) often constrain the number of cells available to evaluate the breadth of functional T-cell responses to infections or therapeutic interventions. The methods most commonly used to assess this functional diversity ex vivo and to recover specific cells to expand in vitro usually require more than 10(6) cells. Here we present a process to identify antigen-specific responses efficiently ex vivo from 10(4)-10(5) single cells from blood or mucosal tissues using dense arrays of subnanoliter wells. The approach combines on-chip imaging cytometry with a technique for capturing secreted proteins--called "microengraving"--to enumerate antigen-specific responses by single T cells in a manner comparable to conventional assays such as ELISpot and intracellular cytokine staining. Unlike those assays, however, the individual cells identified can be recovered readily by micromanipulation for further characterization in vitro. Applying this method to assess HIV-specific T-cell responses demonstrates that it is possible to establish clonal CD8(+) T-cell lines that represent the most abundant specificities present in circulation using 100- to 1,000-fold fewer cells than traditional approaches require and without extensive genotypic analysis a priori. This rapid (<24 h), efficient, and inexpensive process should improve the comparative study of human T-cell immunology across ages and anatomic compartments.

Discovery of antibiotics that selectively kill metabolically dormant bacteria.

There is a need to discover and develop non-toxic antibiotics that are effective against metabolically dormant bacteria, which underlie chronic infections and promote antibiotic resistance. Traditional antibiotic discovery has historically favored compounds effective against actively metabolizing cells, a property that is not predictive of efficacy in metabolically inactive contexts. Here, we combine a stationary-phase screening method with deep learning-powered virtual screens and toxicity filtering to discover compounds with lethality against metabolically dormant bacteria and favorable toxicity profiles. The most potent and structurally distinct compound without any obvious mechanistic liability was semapimod, an anti-inflammatory drug effective against stationary-phase E. coli and A. baumannii. Integrating microbiological assays, biochemical measurements, and single-cell microscopy, we show that semapimod selectively disrupts and permeabilizes the bacterial outer membrane by binding lipopolysaccharide. This work illustrates the value of harnessing non-traditional screening methods and deep learning models to identify non-toxic antibacterial compounds that are effective in infection-relevant contexts.

Epithelial adhesion molecules can inhibit HIV-1-specific CD8⁺ T-cell functions.

Under persistent antigenic stimulation, virus-specific CD8⁺ T cells become increasingly dysfunctional and up-regulate several inhibitory molecules such as killer lectin-like receptor G1 (KLRG1). Here, we demonstrate that HIV-1 antigen-specific T cells from subjects with chronic-progressive HIV-1 infection have significantly elevated KLRG1 expression (P < .001); show abnormal distribution of E-cadherin, the natural ligand of KLRG1, in the intestinal mucosa; and have elevated levels of systemic soluble E-cadherin (sE-cadherin) that significantly correlate with HIV-1 viral load (R = 0.7, P = .004). We furthermore demonstrate that in the presence of sE-cadherin, KLRG1(hi) HIV-1-specific CD8⁺ T cells are impaired in their ability to respond by cytokine secretion on antigenic stimulation (P = .002) and to inhibit viral replication (P = .03) in vitro. Thus, these data suggest a critical mechanism by which the disruption of the intestinal epithelium associated with HIV-1 leads to increased systemic levels of sE-cadherin, which inhibits the effector functions of KLRG1(hi)-expressing HIV-1-specific CD8⁺ T cells systemically.

It's Not You, It's SOSA: A Case Study on Breaking Up With an FDA-Cleared Susceptibility Testing System's Oxacillin Results for Staphylococcus spp. Other Than S. aureus and S. lugdunensis.

Background: In 2021, the Clinical and Laboratory Standards Institute revised its susceptible oxacillin minimum inhibitory concentration (MIC) breakpoint for Staphylococcus spp. other than S. aureus and S. lugdunensis (SOSA) from ≤0.25 to ≤0.5 µg/mL. Here, we describe the response to this breakpoint change, which at the time of this study was not yet recognized by the US Food and Drug Administration (FDA), in our laboratory, where the primary method for antimicrobial susceptibility testing (AST) of SOSA is VITEK 2. VITEK 2 uses the Automated Expert System (AES) to integrate the results of oxacillin MIC and cefoxitin screen tests into a final interpretation; our laboratory also adjudicates discordant oxacillin and cefoxitin results using a PBP2a test. Methods: We retrospectively reviewed and assessed the yield of PBP2a testing for 189 SOSA isolates with discordant (when applying the FDA susceptible oxacillin breakpoint of ≤0.25 µg/mL) VITEK 2 oxacillin and cefoxitin results, and then prospectively incorporated PBP2a testing for isolates with oxacillin MICs of 0.5 µg/mL and positive cefoxitin screens into our algorithm. Results: Compared with accepting the VITEK 2 AES interpretation, PBP2a testing substantially improved the accuracy of mecA-mediated resistance classification in both scenarios, especially for the ∼4.7% of isolates with oxacillin MICs ≤0.5 µg/mL and positive cefoxitin screens. Conclusions: Although detection of mecA or PBP2a is the gold standard for assessment of ß-lactam resistance in staphylococci, targeting a subset of isolates for mecA or PBP2a testing based on phenotypic AST results that predict an increased risk of misclassification may be a pragmatic, labor- and cost-saving approach.

Clinical Assessment and Validation of a Rapid and Sensitive SARS-CoV-2 Test Using Reverse Transcription Loop-Mediated Isothermal Amplification Without the Need for RNA Extraction.

BACKGROUND: Amid the enduring pandemic, there is an urgent need for expanded access to rapid, sensitive, and inexpensive coronavirus disease 2019 (COVID-19) testing worldwide without specialized equipment. We developed a simple test that uses colorimetric reverse transcription loop-mediated isothermal amplification (RT-LAMP) to detect severe acute resrpiratory syndrome coronavirus 2 (SARS-CoV-2) in 40 minutes from sample collection to result. METHODS: We tested 135 nasopharyngeal specimens from patients evaluated for COVID-19 infection at Massachusetts General Hospital. Specimens were either added directly to RT-LAMP reactions, inactivated by a combined chemical and heat treatment step, or inactivated then purified with a silica particle-based concentration method. Amplification was performed with 2 SARS-CoV-2-specific primer sets and an internal specimen control; the resulting color change was visually interpreted. RESULTS: Direct RT-LAMP testing of unprocessed specimens could only reliably detect samples with abundant SARS-CoV-2 (>3 000 000 copies/mL), with sensitivities of 50% (95% CI, 28%-72%) and 59% (95% CI, 43%-73%) in samples collected in universal transport medium and saline, respectively, compared with quantitative polymerase chain reaction (qPCR). Adding an upfront RNase inactivation step markedly improved the limit of detection to at least 25 000 copies/mL, with 87.5% (95% CI, 72%-95%) sensitivity and 100% specificity (95% CI, 87%-100%). Using both inactivation and purification increased the assay sensitivity by 10-fold, achieving a limit of detection comparable to commercial real-time PCR-based diagnostics. CONCLUSIONS: By incorporating a fast and inexpensive sample preparation step, RT-LAMP accurately detects SARS-CoV-2 with limited equipment for about US$6 per sample, making this a potentially ideal assay to increase testing capacity, especially in resource-limited settings.