Single-cell profiling of myeloid cells in glioblastoma across species and disease stage reveals macrophage competition and specialization.

Glioblastomas are aggressive primary brain cancers that recur as therapy-resistant tumors. Myeloid cells control glioblastoma malignancy, but their dynamics during disease progression remain poorly understood. Here, we employed single-cell RNA sequencing and CITE-seq to map the glioblastoma immune landscape in mouse tumors and in patients with newly diagnosed disease or recurrence. This revealed a large and diverse myeloid compartment, with dendritic cell and macrophage populations that were conserved across species and dynamic across disease stages. Tumor-associated macrophages (TAMs) consisted of microglia- or monocyte-derived populations, with both exhibiting additional heterogeneity, including subsets with conserved lipid and hypoxic signatures. Microglia- and monocyte-derived TAMs were self-renewing populations that competed for space and could be depleted via CSF1R blockade. Microglia-derived TAMs were predominant in newly diagnosed tumors, but were outnumbered by monocyte-derived TAMs following recurrence, especially in hypoxic tumor environments. Our results unravel the glioblastoma myeloid landscape and provide a framework for future therapeutic interventions.

Deep immune profiling of patients treated with lenalidomide and dexamethasone with or without daratumumab.

CD38-targeted antibody, daratumumab, is approved for the treatment of multiple myeloma (MM). Phase 1/2 studies GEN501/SIRIUS revealed a novel immunomodulatory mechanism of action (MOA) of daratumumab that enhanced the immune response, reducing natural killer (NK) cells without affecting efficacy or safety. We further evaluated daratumumab's effects on immune cells in whole blood samples of relapsed/refractory MM patients from both treatment arms of the phase 3 POLLUX study (lenalidomide/dexamethasone [Rd] or daratumumab plus Rd [D-Rd]) at baseline (D-Rd, 40; Rd, 45) and after 2 months on treatment (D-Rd, 31; Rd, 33) using cytometry by time-of-flight. We confirmed previous reports of NK cell reduction with D-Rd. Persisting NK cells were phenotypically distinct, with increased expression of HLA-DR, CD69, CD127, and CD27. The proportion of T cells increased preferentially in deep responders to D-Rd, with a higher proportion of CD8+ versus CD4+ T cells. The expansion of CD8+ T cells correlated with clonality, indicating generation of adaptive immune response with D-Rd. D-Rd resulted in a higher proportion of effector memory T cells versus Rd. D-Rd reduced immunosuppressive CD38+ regulatory T cells. This study confirms daratumumab's immunomodulatory MOA in combination with immunomodulatory drugs and provides further insight into immune cell changes and activation status following daratumumab-based therapy.

High-Parameter Mass Cytometry Evaluation of Relapsed/Refractory Multiple Myeloma Patients Treated with Daratumumab Demonstrates Immune Modulation as a Novel Mechanism of Action.

Daratumumab is a CD38-targeted human monoclonal antibody with direct anti-myeloma cell mechanisms of action. Flow cytometry in relapsed and/or refractory multiple myeloma (RRMM) patients treated with daratumumab revealed cytotoxic T-cell expansion and reduction of immune-suppressive populations, suggesting immune modulation as an additional mechanism of action. Here, we performed an in-depth analysis of the effects of daratumumab on immune-cell subpopulations using high-dimensional mass cytometry. Whole-blood and bone-marrow baseline and on-treatment samples from RRMM patients who participated in daratumumab monotherapy studies (SIRIUS and GEN501) were evaluated with high-throughput immunophenotyping. In daratumumab-treated patients, the intensity of CD38 marker expression decreased on many immune cells in SIRIUS whole-blood samples. Natural killer (NK) cells were depleted with daratumumab, with remaining NK cells showing increased CD69 and CD127, decreased CD45RA, and trends for increased CD25, CD27, and CD137 and decreased granzyme B. Immune-suppressive population depletion paralleled previous findings, and a newly observed reduction in CD38+ basophils was seen in patients who received monotherapy. After 2 months of daratumumab, the T-cell population in whole-blood samples from responders shifted to a CD8 prevalence with higher granzyme B positivity (P = 0.017), suggesting increased killing capacity and supporting monotherapy-induced CD8+ T-cell activation. High-throughput cytometry immune profiling confirms and builds upon previous flow cytometry data, including comparable CD38 marker intensity on plasma cells, NK cells, monocytes, and B/T cells. Interestingly, a shift toward cytolytic granzyme B+ T cells was also observed and supports adaptive responses in patients that may contribute to depth of response. © 2018 The Authors. Cytometry Part A published by Wiley Periodicals, Inc. on behalf of International Society for Advancement of Cytometry.

BIGL: Biochemically Intuitive Generalized Loewe null model for prediction of the expected combined effect compatible with partial agonism and antagonism.

Clinical efficacy regularly requires the combination of drugs. For an early estimation of the clinical value of (potentially many) combinations of pharmacologic compounds during discovery, the observed combination effect is typically compared to that expected under a null model. Mechanistic accuracy of that null model is not aspired to; to the contrary, combinations that deviate favorably from the model (and thereby disprove its accuracy) are prioritized. Arguably the most popular null model is the Loewe Additivity model, which conceptually maps any assay under study to a (virtual) single-step enzymatic reaction. It is easy-to-interpret and requires no other information than the concentration-response curves of the individual compounds. However, the original Loewe model cannot accommodate concentration-response curves with different maximal responses and, by consequence, combinations of an agonist with a partial or inverse agonist. We propose an extension, named Biochemically Intuitive Generalized Loewe (BIGL), that can address different maximal responses, while preserving the biochemical underpinning and interpretability of the original Loewe model. In addition, we formulate statistical tests for detecting synergy and antagonism, which allow for detecting statistically significant greater/lesser observed combined effects than expected from the null model. Finally, we demonstrate the novel method through application to several publicly available datasets.

QQ-SNV: single nucleotide variant detection at low frequency by comparing the quality quantiles.

BACKGROUND: Next generation sequencing enables studying heterogeneous populations of viral infections. When the sequencing is done at high coverage depth ("deep sequencing"), low frequency variants can be detected. Here we present QQ-SNV (http://sourceforge.net/projects/qqsnv), a logistic regression classifier model developed for the Illumina sequencing platforms that uses the quantiles of the quality scores, to distinguish true single nucleotide variants from sequencing errors based on the estimated SNV probability. To train the model, we created a dataset of an in silico mixture of five HIV-1 plasmids. Testing of our method in comparison to the existing methods LoFreq, ShoRAH, and V-Phaser 2 was performed on two HIV and four HCV plasmid mixture datasets and one influenza H1N1 clinical dataset. RESULTS: For default application of QQ-SNV, variants were called using a SNV probability cutoff of 0.5 (QQ-SNV(D)). To improve the sensitivity we used a SNV probability cutoff of 0.0001 (QQ-SNV(HS)). To also increase specificity, SNVs called were overruled when their frequency was below the 80(th) percentile calculated on the distribution of error frequencies (QQ-SNV(HS-P80)). When comparing QQ-SNV versus the other methods on the plasmid mixture test sets, QQ-SNV(D) performed similarly to the existing approaches. QQ-SNV(HS) was more sensitive on all test sets but with more false positives. QQ-SNV(HS-P80) was found to be the most accurate method over all test sets by balancing sensitivity and specificity. When applied to a paired-end HCV sequencing study, with lowest spiked-in true frequency of 0.5%, QQ-SNV(HS-P80) revealed a sensitivity of 100% (vs. 40-60% for the existing methods) and a specificity of 100% (vs. 98.0-99.7% for the existing methods). In addition, QQ-SNV required the least overall computation time to process the test sets. Finally, when testing on a clinical sample, four putative true variants with frequency below 0.5% were consistently detected by QQ-SNV(HS-P80) from different generations of Illumina sequencers. CONCLUSIONS: We developed and successfully evaluated a novel method, called QQ-SNV, for highly efficient single nucleotide variant calling on Illumina deep sequencing virology data.

VirVarSeq: a low-frequency virus variant detection pipeline for Illumina sequencing using adaptive base-calling accuracy filtering.

MOTIVATION: In virology, massively parallel sequencing (MPS) opens many opportunities for studying viral quasi-species, e.g. in HIV-1- and HCV-infected patients. This is essential for understanding pathways to resistance, which can substantially improve treatment. Although MPS platforms allow in-depth characterization of sequence variation, their measurements still involve substantial technical noise. For Illumina sequencing, single base substitutions are the main error source and impede powerful assessment of low-frequency mutations. Fortunately, base calls are complemented with quality scores (Qs) that are useful for differentiating errors from the real low-frequency mutations. RESULTS: A variant calling tool, Q-cpileup, is proposed, which exploits the Qs of nucleotides in a filtering strategy to increase specificity. The tool is imbedded in an open-source pipeline, VirVarSeq, which allows variant calling starting from fastq files. Using both plasmid mixtures and clinical samples, we show that Q-cpileup is able to reduce the number of false-positive findings. The filtering strategy is adaptive and provides an optimized threshold for individual samples in each sequencing run. Additionally, linkage information is kept between single-nucleotide polymorphisms as variants are called at the codon level. This enables virologists to have an immediate biological interpretation of the reported variants with respect to their antiviral drug responses. A comparison with existing SNP caller tools reveals that calling variants at the codon level with Q-cpileup results in an outstanding sensitivity while maintaining a good specificity for variants with frequencies down to 0.5%. AVAILABILITY: The VirVarSeq is available, together with a user's guide and test data, at sourceforge: http://sourceforge.net/projects/virtools/?source=directory.

Multi-model inference using mixed effects from a linear regression based genetic algorithm.

BACKGROUND: Different high-dimensional regression methodologies exist for the selection of variables to predict a continuous variable. To improve the variable selection in case clustered observations are present in the training data, an extension towards mixed-effects modeling (MM) is requested, but may not always be straightforward to implement.In this article, we developed such a MM extension (GA-MM-MMI) for the automated variable selection by a linear regression based genetic algorithm (GA) using multi-model inference (MMI). We exemplify our approach by training a linear regression model for prediction of resistance to the integrase inhibitor Raltegravir (RAL) on a genotype-phenotype database, with many integrase mutations as candidate covariates. The genotype-phenotype pairs in this database were derived from a limited number of subjects, with presence of multiple data points from the same subject, and with an intra-class correlation of 0.92. RESULTS: In generation of the RAL model, we took computational efficiency into account by optimizing the GA parameters one by one, and by using tournament selection. To derive the main GA parameters we used 3 times 5-fold cross-validation. The number of integrase mutations to be used as covariates in the mixed effects models was 25 (chrom.size). A GA solution was found when R2MM > 0.95 (goal.fitness). We tested three different MMI approaches to combine the results of 100 GA solutions into one GA-MM-MMI model. When evaluating the GA-MM-MMI performance on two unseen data sets, a more parsimonious and interpretable model was found (GA-MM-MMI TOP18: mixed-effects model containing the 18 most prevalent mutations in the GA solutions, refitted on the training data) with better predictive accuracy (R2) in comparison to GA-ordinary least squares (GA-OLS) and Least Absolute Shrinkage and Selection Operator (LASSO). CONCLUSIONS: We have demonstrated improved performance when using GA-MM-MMI for selection of mutations on a genotype-phenotype data set. As we largely automated setting the GA parameters, the method should be applicable on similar datasets with clustered observations.

Quantitative prediction of integrase inhibitor resistance from genotype through consensus linear regression modeling.

BACKGROUND: Integrase inhibitors (INI) form a new drug class in the treatment of HIV-1 patients. We developed a linear regression modeling approach to make a quantitative raltegravir (RAL) resistance phenotype prediction, as Fold Change in IC50 against a wild type virus, from mutations in the integrase genotype. METHODS: We developed a clonal genotype-phenotype database with 991 clones from 153 clinical isolates of INI naïve and RAL treated patients, and 28 site-directed mutants.We did the development of the RAL linear regression model in two stages, employing a genetic algorithm (GA) to select integrase mutations by consensus. First, we ran multiple GAs to generate first order linear regression models (GA models) that were stochastically optimized to reach a goal R2 accuracy, and consisted of a fixed-length subset of integrase mutations to estimate INI resistance. Secondly, we derived a consensus linear regression model in a forward stepwise regression procedure, considering integrase mutations or mutation pairs by descending prevalence in the GA models. RESULTS: The most frequently occurring mutations in the GA models were 92Q, 97A, 143R and 155H (all 100%), 143G (90%), 148H/R (89%), 148K (88%), 151I (81%), 121Y (75%), 143C (72%), and 74M (69%). The RAL second order model contained 30 single mutations and five mutation pairs (p < 0.01): 143C/R&97A, 155H&97A/151I and 74M&151I. The R2 performance of this model on the clonal training data was 0.97, and 0.78 on an unseen population genotype-phenotype dataset of 171 clinical isolates from RAL treated and INI naïve patients. CONCLUSIONS: We describe a systematic approach to derive a model for predicting INI resistance from a limited amount of clonal samples. Our RAL second order model is made available as an Additional file for calculating a resistance phenotype as the sum of integrase mutations and mutation pairs.

Cross-validated stepwise regression for identification of novel non-nucleoside reverse transcriptase inhibitor resistance associated mutations.

BACKGROUND: Linear regression models are used to quantitatively predict drug resistance, the phenotype, from the HIV-1 viral genotype. As new antiretroviral drugs become available, new resistance pathways emerge and the number of resistance associated mutations continues to increase. To accurately identify which drug options are left, the main goal of the modeling has been to maximize predictivity and not interpretability. However, we originally selected linear regression as the preferred method for its transparency as opposed to other techniques such as neural networks. Here, we apply a method to lower the complexity of these phenotype prediction models using a 3-fold cross-validated selection of mutations. RESULTS: Compared to standard stepwise regression we were able to reduce the number of mutations in the reverse transcriptase (RT) inhibitor models as well as the number of interaction terms accounting for synergistic and antagonistic effects. This reduction in complexity was most significant for the non-nucleoside reverse transcriptase inhibitor (NNRTI) models, while maintaining prediction accuracy and retaining virtually all known resistance associated mutations as first order terms in the models. Furthermore, for etravirine (ETR) a better performance was seen on two years of unseen data. By analyzing the phenotype prediction models we identified a list of forty novel NNRTI mutations, putatively associated with resistance. The resistance association of novel variants at known NNRTI resistance positions: 100, 101, 181, 190, 221 and of mutations at positions not previously linked with NNRTI resistance: 102, 139, 219, 241, 376 and 382 was confirmed by phenotyping site-directed mutants. CONCLUSIONS: We successfully identified and validated novel NNRTI resistance associated mutations by developing parsimonious resistance prediction models in which repeated cross-validation within the stepwise regression was applied. Our model selection technique is computationally feasible for large data sets and provides an approach to the continued identification of resistance-causing mutations.

Resistance to raltegravir highlights integrase mutations at codon 148 in conferring cross-resistance to a second-generation HIV-1 integrase inhibitor.

Raltegravir is the first integrase strand-transfer inhibitor (INSTI) approved for use in highly active antiretroviral therapy (HAART) for the management of HIV infection. Resistance to antiretrovirals can compromise the efficacy of HAART regimens. Therefore it is important to understand the emergence of resistance to RAL and cross-resistance to other INSTIs including potential second-generation INSTIs such as MK-2048. We have now studied the question of whether in vitro resistance selection (IVRS) with RAL initiated with viruses derived from clinical isolates would result in selection of resistance mutations consistent with those arising during treatment regimens with HAART containing RAL. Some correlation was observed between the primary mutations selected in vitro and during therapy, initiated with viruses with identical IN sequences. Additionally, phenotypic cross-resistance conferred by specific mutations to RAL and MK-2048 was quantified. N155H, a RAL-associated primary resistance mutation, was selected after IVRS with MK-2048, suggesting similar mechanisms of resistance to RAL and MK-2048. This was confirmed by phenotypic analysis of 766 clonal viruses harboring IN sequences isolated at the point of virological failure from 106 patients on HAART (including RAL), where mutation Q148H/K/R together with additional secondary mutations conferred reduced susceptibility to both RAL and MK-2048. A homology model of full length HIV-1 integrase complexed with viral DNA and RAL or MK-2048, based on an X-ray structure of the prototype foamy virus integrase-DNA complex, was used to explain resistance to RAL and cross-resistance to MK-2048. These findings will be important for the further discovery and profiling of next-generation INSTIs.

A comparative analysis of HIV drug resistance interpretation based on short reverse transcriptase sequences versus full sequences.

BACKGROUND: As second-line antiretroviral treatment (ART) becomes more accessible in resource-limited settings (RLS), the need for more affordable monitoring tools such as point-of-care viral load assays and simplified genotypic HIV drug resistance (HIVDR) tests increases substantially. The prohibitive expenses of genotypic HIVDR assays could partly be addressed by focusing on a smaller region of the HIV reverse transcriptase gene (RT) that encompasses the majority of HIVDR mutations for people on ART in RLS. In this study, an in silico analysis of 125,329 RT sequences was performed to investigate the effect of submitting short RT sequences (codon 41 to 238) to the commonly used virco®TYPE and Stanford genotype interpretation tools. RESULTS: Pair-wise comparisons between full-length and short RT sequences were performed. Additionally, a non-inferiority approach with a concordance limit of 95% and two-sided 95% confidence intervals was used to demonstrate concordance between HIVDR calls based on full-length and short RT sequences.The results of this analysis showed that HIVDR interpretations based on full-length versus short RT sequences, using the Stanford algorithms, had concordance significantly above 95%. When using the virco®TYPE algorithm, similar concordance was demonstrated (>95%), but some differences were observed for d4T, AZT and TDF, where predictions were affected in more than 5% of the sequences. Most differences in interpretation, however, were due to shifts from fully susceptible to reduced susceptibility (d4T) or from reduced response to minimal response (AZT, TDF) or vice versa, as compared to the predicted full RT sequence. The virco®TYPE prediction uses many more mutations outside the RT 41-238 amino acid domain, which significantly contribute to the HIVDR prediction for these 3 antiretroviral agents. CONCLUSIONS: This study illustrates the acceptability of using a shortened RT sequences (codon 41-238) to obtain reliable genotype interpretations by virco®TYPE and Stanford algorithms. Implementation of this simplified protocol could significantly reduce the cost of both resistance testing and ARV treatment monitoring in RLS.

A comparison of HIV-1 drug susceptibility as provided by conventional phenotyping and by a phenotype prediction tool based on viral genotype.

Concordance between the conventional HIV-1 phenotypic drug resistance assay, PhenoSense (PS), and vircoTYPE HIV-1 (vT), a drug resistance assay based on prediction of the phenotype, was investigated in a data set from the Stanford HIV Resistance database (hivdb). Depending on the drug, between 287 and 902 genotype-phenotype data pairs were available for comparisons. Test results (fold-change values) in the two assays were highly correlated, with an overall mean correlation coefficient of 0.90 using single PS measurements. This coefficient rose to 0.94 when the vT results were compared to the mean of repeat PS measurements. These results are comparable with the corresponding correlation coefficients of 0.87 and 0.95, calculated using single measurements, and the mean of repeat measurements, respectively, as obtained in the Antivirogram assay, the conventional HIV-1 phenotypic drug resistance test on which vT is based. The proportion of resistance calls resulting in a "major" discordance (fully susceptible or maximal response by one assay but fully resistant or minimal response by the other) ranged from 0% to 8.1% for drugs for which two clinical test cut-offs were available in both assays (didanosine, abacavir, tenofovir, saquinavir/r, fosamprenavir/r, and lopinavir/r), from 2.4% to 8.1% for the drugs for which two clinical test cut-offs were available in the vT assay and one clinical test cut-off in the PS assay (lamivudine, stavudine, indinavir/r, and atazanavir/r) and from 3.1% to 10.3% for drugs for which biological test cut-offs were used (zidovudine, nevirapine, delavirdine, efavirenz, indinavir, ritonavir, nelfinavir, saquinavir, and fosamprenavir). Our analyses suggest that these assays provide comparable resistance information, which will be of value to physicians who may be presented with either or both types of test report in their practice.

Clinical cut-offs for HIV-1 phenotypic resistance estimates: update based on recent pivotal clinical trial data and a revised approach to viral mixtures.

The clinical utility of HIV-1 resistance testing is dependent upon accurate interpretation and application of results. The development of clinical cut-offs (CCOs) for most HIV antiretroviral drugs assessed by the vircoTYPE HIV-1 resistance test has been described previously. Updated CCOs based on new methodology and new data from clinical cohorts and pivotal clinical studies are presented in this communication. Data for analysis included the original records for CCO derivation from eight clinical trials and two cohort studies plus new records from the clinical cohorts and from the TITAN, POWER, and DUET clinical studies. Drug-specific linear regression models were developed to describe the relationship between baseline characteristics (phenotypic resistance as estimated by virtualPhenotype-LM using methods revised recently for handling mixed viral sequences; viral load; and treatment history), new treatment regimen, and 8-week virologic outcome. The clinical cut-offs were defined as the estimated phenotypic resistance levels (fold change, FC) associated with a 20% and 80% loss of drug activity. The development dataset included 6550 records with an additional 2299 reserved for validation. The updated, v.4.2 CCOs were generally close to the v4.1 values, with a trend observed toward marginally higher cut-offs for the NRTIs. These results suggest that the updated CCOs provide a relevant tool for estimating the contribution to virological response of individual antiviral drugs in antiretroviral drug combinations as used currently in clinical practice.

Advantages of predicted phenotypes and statistical learning models in inferring virological response to antiretroviral therapy from HIV genotype.

BACKGROUND: Inferring response to antiretroviral therapy from the viral genotype alone is challenging. The utility of an intermediate step of predicting in vitro drug susceptibility is currently controversial. Here, we provide a retrospective comparison of approaches using either genotype or predicted phenotypes alone, or in combination. METHODS: Treatment change episodes were extracted from two large databases from the USA (Stanford-California) and Europe (EuResistDB) comprising data from 6,706 and 13,811 patients, respectively. Response to antiretroviral treatment was dichotomized according to two definitions. Using the viral sequence and the treatment regimen as input, three expert algorithms (ANRS, Rega and HIVdb) were used to generate genotype-based encodings and VircoTYPE() 4.0 (Virco BVBA, Mechelen, Belgium) was used to generate a predicted -phenotype-based encoding. Single drug classifications were combined into a treatment score via simple summation and statistical learning using random forests. Classification performance was studied on Stanford-California data using cross-validation and, in addition, on the independent EuResistDB data. RESULTS: In all experiments, predicted phenotype was among the most sensitive approaches. Combining single drug classifications by statistical learning was significantly superior to unweighted summation (P<2.2x10(-16)). Classification performance could be increased further by combining predicted phenotypes and expert encodings but not by combinations of expert encodings alone. These results were confirmed on an independent test set comprising data solely from EuResistDB. CONCLUSIONS: This study demonstrates consistent performance advantages in utilizing predicted phenotype in most scenarios over methods based on genotype alone in inferring virological response. Moreover, all approaches under study benefit significantly from statistical learning for merging single drug classifications into treatment scores.