Standard machine learning approaches outperform deep representation learning on phenotype prediction from transcriptomics data.

BACKGROUND: The ability to confidently predict health outcomes from gene expression would catalyze a revolution in molecular diagnostics. Yet, the goal of developing actionable, robust, and reproducible predictive signatures of phenotypes such as clinical outcome has not been attained in almost any disease area. Here, we report a comprehensive analysis spanning prediction tasks from ulcerative colitis, atopic dermatitis, diabetes, to many cancer subtypes for a total of 24 binary and multiclass prediction problems and 26 survival analysis tasks. We systematically investigate the influence of gene subsets, normalization methods and prediction algorithms. Crucially, we also explore the novel use of deep representation learning methods on large transcriptomics compendia, such as GTEx and TCGA, to boost the performance of state-of-the-art methods. The resources and findings in this work should serve as both an up-to-date reference on attainable performance, and as a benchmarking resource for further research. RESULTS: Approaches that combine large numbers of genes outperformed single gene methods consistently and with a significant margin, but neither unsupervised nor semi-supervised representation learning techniques yielded consistent improvements in out-of-sample performance across datasets. Our findings suggest that using l2-regularized regression methods applied to centered log-ratio transformed transcript abundances provide the best predictive analyses overall. CONCLUSIONS: Transcriptomics-based phenotype prediction benefits from proper normalization techniques and state-of-the-art regularized regression approaches. In our view, breakthrough performance is likely contingent on factors which are independent of normalization and general modeling techniques; these factors might include reduction of systematic errors in sequencing data, incorporation of other data types such as single-cell sequencing and proteomics, and improved use of prior knowledge.

Evaluation of a Janus kinase 1 inhibitor, PF-04965842, in healthy subjects: A phase 1, randomized, placebo-controlled, dose-escalation study.

AIMS: To determine the safety, tolerability, pharmacokinetics and pharmacodynamics of the Janus kinase 1-selective inhibitor, PF-04965842. METHODS: This was a phase 1, first-in-human, randomized, double-blind, placebo-controlled, combination single- and multiple-dose escalation, parallel design study in healthy subjects (http://clinicaltrials.gov, NCT01835197). Subjects received a single dose of placebo or 3, 10, 30, 100, 200, 400 or 800 mg PF-04965842 (single ascending dose phase) and placebo or 30 mg once daily (QD), 100 mg QD, 200 mg QD, 400 mg QD, 100 mg twice daily (BID) or 200 mg BID PF-04965842 for 10 consecutive days (multiple ascending dose phase). The primary objective was to determine the safety and tolerability of PF-04965842. RESULTS: Seventy-nine subjects were randomized and received study treatments. There were no deaths or serious adverse events. The most frequent treatment-emergent adverse events were headache (n = 13), diarrhoea (n = 11) and nausea (n = 11). PF-04965842 was absorbed rapidly (median time at which maximum plasma concentration occurred generally ≤1 h following either single- or multiple-dose administration) and eliminated rapidly (mean t½ 2.8-5.2 h after 10 days of QD or BID administration in the multiple ascending dose phase). Increases in maximum plasma concentration and area under the concentration-time curve were dose proportional up to 200 mg (single or total daily doses) with an apparent trend towards greater than proportional increases with higher doses. Less than 4.4% of the dose was recovered unchanged in urine. Changes in pharmacodynamic biomarkers were consistent with the known effects of Janus kinase signalling inhibition. CONCLUSIONS: These results support further evaluation of PF-04965842 for clinical use in patients with inflammatory diseases.

Antagonism of chemokine receptor CCR8 is ineffective in a primate model of asthma.

BACKGROUND: Expression of the T-cell-associated chemokine receptor CCR8 and its ligand CCL1 have been demonstrated to be elevated in patients with asthma. CCR8 deficiency or inhibition in models of allergic airway disease in mice resulted in conflicting data. OBJECTIVE: To investigate the effects of a selective small molecule CCR8 inhibitor (ML604086) in a primate model of asthma. METHODS: ML604086 and vehicle were administered by intravenous infusion to 12 cynomolgus monkeys during airway challenge with Ascaris suum. Samples were collected throughout the study to measure pharmacokinetics (PK) and systemic CCR8 inhibition, as well as inflammation, T helper 2 (Th2) cytokines and mucus in bronchoalveolar lavage (BAL). Airway resistance and compliance were measured before and after allergen challenge, and in response to increasing concentrations of methacholine. RESULTS: ML604086 inhibited CCL1 binding to CCR8 on circulating T-cells>98% throughout the duration of the study. However, CCR8 inhibition had no significant effect on allergen-induced BAL eosinophilia and the induction of the Th2 cytokines IL-4, IL-5, IL-13 and mucus levels in BAL. Changes in airway resistance and compliance induced by allergen provocation and increasing concentrations of methacholine were also not affected by ML604086. CONCLUSIONS: These results clearly demonstrate a dispensable role for CCR8 in ameliorating allergic airway disease in atopic primates, and suggest that strategies other than CCR8 antagonism should be considered for the treatment of asthma.

The prostaglandin D2 receptor CRTH2 is important for allergic skin inflammation after epicutaneous antigen challenge.

BACKGROUND: Cutaneous prostaglandin (PG) D2 levels increase after scratching. Chemoattractant receptor-homologous molecule expressed on receptor on T(H)2 cells (CRTH2) mediates chemotaxis to PGD2 and is expressed on T(H)2 cells and eosinophils, which infiltrate skin lesions in patients with atopic dermatitis. OBJECTIVE: We sought to examine the role of CRTH2 in a murine model of atopic dermatitis. METHODS: CRTH2(-/-) mice and wild-type control animals were epicutaneously sensitized by means of repeated application of ovalbumin (OVA) to tape-stripped skin for 7 weeks and then challenged by means of OVA application to tape-stripped previously unsensitized skin for 1 week. Skin histology was assessed by means of hematoxylin and eosin staining and immunohistochemistry. Cytokine mRNA expression was examined by means of quantitative RT-PCR. Levels of PGD2, antibody, and cytokines were measured by means of ELISA. RESULTS: PGD2 levels significantly increased in skin 24 hours after tape stripping, although not in skin subjected to repeated sensitization with OVA. Allergic skin inflammation developed normally at sites of chronic epicutaneous sensitization with OVA in CRTH2(-/-) mice but was severely impaired in previously unsensitized skin challenged with OVA, as evidenced by significantly decreased skin infiltration with eosinophils and CD4(+) cells and impaired T(H)2 cytokine mRNA expression. Impaired skin inflammation at sites of acute OVA challenge in CRTH2(-/-) mice was not due to an impaired systemic response to epicutaneous sensitization because OVA-specific IgG1 and IgE antibody levels and OVA-driven splenocyte secretion of cytokines in these mice were comparable with those seen in wild-type control animals. CONCLUSIONS: CRTH2 promotes allergic skin inflammation in response to cutaneous exposure to antigen in previously sensitized mice.

Coordinated involvement of mast cells and T cells in allergic mucosal inflammation: critical role of the CC chemokine ligand 1:CCR8 axis.

CCL1 is the predominant chemokine secreted from IgE-activated human and mouse mast cells in vitro, colocalizes to mast cells in lung biopsies, and is elevated in asthmatic airways. CCR8, the receptor for CCL1, is expressed by approximately 70% of CD4(+) T lymphocytes recruited to the asthmatic airways, and the number of CCR8-expressing cells is increased 3-fold in the airways of asthmatic subjects compared with normal volunteers. In vivo, CCL1 expression in the lung is reduced in mast cell-deficient mice after aeroallergen provocation. Neutralization of CCL1 or CCR8 deficiency results in reduced mucosal lung inflammation, airway hyperresponsiveness, and mucus hypersecretion to a similar degree as detected in mast cell-deficient mice. Adenoviral delivery of CCL1 to the lungs of mast cell-deficient mice restores airway hyperresponsiveness, lung inflammation, and mucus hypersecretion to the degree observed in wild-type mice. The consequences of CCR8 deficiency, including a marked reduction in Th2 cytokine levels, are comparable with those observed by depletion of CD4(+) T lymphocytes. Thus, mast cell-derived CCL1- and CCR8-expressing CD4(+) effector T lymphocytes play an essential role in orchestrating lung mucosal inflammatory responses.

IL-21 receptor signaling is integral to the development of Th2 effector responses in vivo.

Interleukin 21 (IL-21) is a member of the common gamma-chain family of cytokines, which influence a broad spectrum of immunologic responses. A number of studies have examined the function of IL-21, but its specific role in Th1/Th2-cell differentiation and related effector responses remains to be clarified. Thus, we generated IL-21R-deficient mice and have investigated the role of IL-21R signaling using a series of in vivo experimentally induced disease models. We first addressed the role of IL-21R signaling in Th2 immune responses by examining allergic airway inflammation, and Nippostrongylus brasiliensis and Heligmosomoides polygyrus antihelminth responses. In each of these systems, IL-21R signaling played a clear role in the development of Th2 responses. Comparatively, IL-21R signaling was not required for the containment of Leishmania major infection or the development of experimental autoimmune myocarditis, indicative of competent Th1 and Th17 responses, respectively. Adoptive transfer of T cells and analysis of IL-21R+/+/IL-21R-/- chimera mice revealed that IL-21R-signaling was central to Th2-cell survival or migration to peripheral tissues. Overall, our data show IL-21 plays a crucial role in supporting polarized Th2 responses in vivo, while appearing superfluous for Th1 and Th17 responses.

CCR8 expression identifies CD4 memory T cells enriched for FOXP3+ regulatory and Th2 effector lymphocytes.

CD4+ Th2 cells are important regulators of allergic inflammation. CCR8 is thought to play a role in Th2-mediated responses, however, expression of CCR8 in peripheral blood has not been fully characterized. Using a fluorescent form of the ligand selective for CCR8 (F-CCL1), we identified the leukocytes expressing CCR8 in human, monkey, and mouse peripheral blood. CCR8 expression is primarily restricted to a subset of human CD4 memory T lymphocytes (15%). Approximately 40% of CCR8+CD4+ T cells express Th2 cytokines IL-4 or IL-13 while 13% express the Th1 cytokine IFN-gamma. In fact, 50% of all Th2, but only 5% of Th1, cells express CCR8. Upon anti-CD3/anti-CD28 mAb-mediated activation, CCR8+CD4+ T cells secrete 3- to 7-fold higher levels of IL-4, IL-5, IL-9, and IL-13 and 10- to 20-fold lower levels of IFN-gamma or IL-17, compared with CCR8-CD4+ memory T cells. Two-thirds of CCR8+CD4 T cells express cutaneous lymphocyte-associated Ag while the majority lack gut-homing receptors. CCR8+CD4+ cells express CCR7 and CD62L and are present in spleen and lymph nodes of mice. Approximately 25% of CCR8+CD4 T cells express CD25high while 20% of CCR8+CD4+ express the T regulatory cell transcription factor FOXP3 accounting for 60% of all FOXP3-expressing CD4+ T cells. In conclusion, CCR8 marks a diverse subset of CD4 memory T cells enriched for T regulatory and Th2 cells which have the potential for recruitment into sites of allergic inflammation where they could participate in the induction and regulation of the allergic response.

Pharmacological characterization of guinea pig chemoattractant receptor-homologous molecule expressed on Th2 cells (CRTH2).

Chemoattractant receptor-homologous molecule expressed on Th2 cells (CRTH2), a G protein-coupled receptor activated by prostaglandin D(2) (PGD(2)), has been identified as a receptor expressed on cell types critical to the pathogenesis of asthma. The cDNA encoding guinea pig CRTH2 was cloned and mRNA expression examined in selected tissues. Transcript profiling of guinea pig CRTH2 indicated relatively high levels of expression in bone marrow, intermediate levels in brain and relatively low levels in lung, spleen, thymus, lymph node, etc. Characterization of the molecular pharmacology of guinea pig CRTH2 revealed that guinea pig CRTH2 exhibited a greater affinity for Delta(12)-PGJ(2), a stable PGD(2) metabolite relative to human CRTH2. The CRTH2 selective agonists 13,14-dihydro-15-keto PGD(2) and Delta(12)-PGJ(2) induced the recruitment of eosinophils following intradermal administration of these ligands in guinea pigs. Chemotaxis of guinea pig eosinophils was elicited by either PGD(2) or Delta(12)-PGJ(2), and was abolished by a CRTH2-specific antagonist. These results indicate that PGD(2) and the stable metabolite, Delta(12)-PGJ(2), play important roles in CRTH2 activation in the guinea pig.

T follicular helper cells express a distinctive transcriptional profile, reflecting their role as non-Th1/Th2 effector cells that provide help for B cells.

Effector T cell responses have long been viewed in the context of the Th1/Th2 paradigm. Recently, a third major subset of nonpolarized effector T cells that provides help to B cells has been identified. These T cells, termed T follicular helper (T(FH)) cells, home to the B cell areas of secondary lymphoid tissue, through interactions mediated via the chemokine receptor CXCR5 and its ligand CXCL13. Affymetrix microarrays were used to identify transcription factors, cytokines, and cell surface molecules that underlie the differentiation pathways and functional properties of the T(FH) subset. The transcriptional profile of human CXCR5(+) T(FH) cells was compared with that of Th1 and Th2 cells, which enabled the identification of numerous genes expressed preferentially by T(FH) cells, over the other effector subsets. Certain T(FH) genes were also expressed by B cells and thus appear to be particularly relevant for humoral immunity. Abs were used to confirm the expression of several factors. In particular, CD84 and CD200, the cytokine IL-21, and the transcription factor BCL6 were all strongly associated with T(FH) cells. Gene microarrays reveal a highly distinctive transcriptional profile for a third subset of effector T cells that differs markedly from Th1 and Th2 cells.

FcepsilonRI-dependent gene expression in human mast cells is differentially controlled by T helper type 2 cytokines.

BACKGROUND: Mast cells (MCs) proliferate in response to T(H)2 cytokines and express genes de novo after activation. Limited information is available concerning the interplay between these events. OBJECTIVE: We explored the potential for T(H)2 cytokines to alter activation-dependent gene expression by MCs. METHODS: Cord blood-derived human (h)MCs maintained in stem cell factor (SCF) alone were compared with replicates treated with IL-4, IL-5, or IL-9, respectively, for their patterns of FcepsilonRI-dependent gene induction using microarray technology. RESULTS: Activation of SCF-treated hMCs upregulated their expression of roughly 140 transcripts at 2 hours, including genes involved in cell cycle progression and arrest. Each cytokine substantially modified this profile; approximately 800 inducible genes apiece were controlled by IL-5 or IL-9, whereas 169 inducible genes were controlled by IL-4. IL-4 favored the induction of cytokines and of genes associated with cell growth arrest (GADD34, GAS-1, CIDE-A, INK4D, and BAX) and completely abolished the enhanced proliferation observed in the other 3 groups after activation. Conversely, IL-5 priming induced preferential upregulation of genes involved in cell proliferation and did not abolish thymidine incorporation. CONCLUSIONS: T(H)2 cytokines differentially modulate gene induction in hMCs after FcepsilonRI cross-linkage. IL-4 uniquely controls cytokine gene expression by hMCs and might also limit their activation-driven proliferation.

CCR4 blockade does not inhibit allergic airways inflammation.

The CC chemokine receptor 4 (CCR4) shows selectivity for the recruitment of memory T cell subsets, including those of the T helper cell type 2 (Th2) phenotype. In humans, CCR4+ T cells are recruited to the asthmatic lung in response to allergen challenge; however, the contribution of this pathway to allergic disease remains uncertain. We therefore investigated the role of CCR4 in allergic airways inflammation in the guinea pig. Blockade of CCR4 with a specific antibody resulted in only minor changes in numbers of CCR4+ Th cells in the bronchoalveolar lavage fluid of allergen-challenged guinea pigs and failed to inhibit the generation of eotaxin/CC chemokine ligand (CCL)11 or macrophage-derived chemokine/CCL22 or the recruitment of inflammatory leukocytes to the lung. These data suggest that although CCR4 was originally proposed as a marker of Th2 status, antigen-specific Th2 cells are recruited to the lung predominantly by other pathways. This study casts doubts on the validity of CCR4 as a therapeutic target in the treatment of asthma.

Expression of the type 2 receptor for cysteinyl leukotrienes (CysLT2R) by human mast cells: Functional distinction from CysLT1R.

Cysteinyl leukotrienes (cysLTs) mediate vascular leakage and bronchoconstriction through the smooth muscle-associated CysLT type 1 receptor (CysLT1R), one of at least two loosely homologous cysLT-binding G protein-coupled receptors. We previously reported that CysLT1R is expressed by cultured human mast cells (hMCs), and that priming these cells with IL-4 enhances their sensitivity to calcium flux and cytokine generation in response to cys-LTs and the nucleotide ligand, uridine diphosphate (UDP), without increasing their surface expression of CysLT1R. We now report that hMCs express the type 2 receptor for cysLTs (CysLT2R) as well, and that the amount of surface CysLT2R protein increases in response to priming with IL-4. The selective function of CysLT2R was evident based on uninhibited IL-8 secretion by IL-4-primed hMCs stimulated with cys-LTs or UDP in the presence of the selective CysLT1R antagonist MK571. MK571 did inhibit IL-5 generation, calcium flux, and phosphorylation of extracellular signal-regulated kinase. IL-8 secretion was inhibited by pertussis toxin and a selective p38 kinase inhibitor, SB203580. The CysLT2 response may permit the cys-LTs and nucleotides generated in infection and tissue injury to elicit IL-8 generation by hMCs, potentially leading to neutrophilic infiltration, a characteristic of aerosol challenge-induced late-phase responses and of sudden death associated with asthma.

CC chemokine receptor 9 expression defines a subset of peripheral blood lymphocytes with mucosal T cell phenotype and Th1 or T-regulatory 1 cytokine profile.

The chemokine receptor CCR9 is expressed on most small intestinal lamina propria and intraepithelial lymphocytes and on a small subset of peripheral blood lymphocytes. CCR9-expressing lymphocytes may play an important role in small bowel immunity and inflammation. We studied the phenotype and functional characteristics of CCR9(+) lymphocytes in blood from normal donors. A subset of CCR9(+) T cells have a phenotype of activated cells and constitutively express the costimulatory molecules CD40L and OX-40. In contrast to CCR9(-), CCR9(+)CD4(+) peripheral blood T cells proliferate to anti-CD3 or anti-CD2 stimulation and produce high levels of IFN-gamma and IL-10. IL-10-producing cells were exclusively detected within the CCR9(+) subset of CD4(+) T cells by intracellular staining and were distinct from IL-2- and IFN-gamma-producing cells. Moreover, memory CCR9(+)CD4(+) lymphocytes respond to CD2 stimulation with proliferation and IFN-gamma/IL-10 production, whereas memory CCR9(-)CD4(+) cells were unresponsive. In addition, memory CCR9(+)CD4(+) T cells support Ig production by cocultured CD19(+) B cells in the absence of prior T cell activation or addition of exogenous cytokines. Our data show that the memory subset of circulating CCR9(+)CD4(+) T cells has characteristics of mucosal T lymphocytes and contains cells with either Th1 or T-regulatory 1 cytokine profiles. Studies on the cytokine profile and Ag specificity of this cell subset could provide important insight into small intestinal immune-mediated diseases and oral tolerance in humans.

CCR4 in human allergen-induced late responses in the skin and lung.

We studied the regulation of CCR4 expression in peripheral blood and in human models of cutaneous and pulmonary allergen challenge. CCR4 expression was detectable on freshly isolated CD4+ lymphocytes and in CD4+ and CD8+ T cell lines derived from blood of atopic donors. Numbers of CCR4+ cells were up-regulated in T cell lines expanded in the presence of IL-4. CCR4 mRNA was absent at baseline in normal subjects in lung and skin, but present at baseline in the lung of some atopics. Baseline expression of CCR4 mRNA and protein was higher in lung vs. skin, but allergen-induced increases in CCR4 mRNA+ cells were observed in both organs. CCR4 protein+ cells were present at higher levels after allergen challenge in atopics compared to normal subjects. CCR4 may be important in the recruitment of T lymphocytes at sites of allergic inflammation, in a non-organ-specific manner.

Expression of the chemokine receptors CCR4, CCR5, and CXCR3 by human tissue-infiltrating lymphocytes.

Differential expression of adhesion molecules and chemokine receptors has been useful for identification of peripheral blood memory lymphocyte subsets with distinct tissue and microenvironmental tropisms. Expression of CCR4 by circulating memory CD4(+) lymphocytes is associated with cutaneous and other systemic populations while expression of CCR9 is associated with a small intestine-homing subset. CCR5 and CXCR3 are also expressed by discrete memory CD4(+) populations in blood, as well as by tissue-infiltrating lymphocytes from a number of sites. To characterize the similarities and differences among tissue-infiltrating lymphocytes, and to shed light on the specialization of lymphocyte subsets that mediate inflammation and immune surveillance in particular tissues, we have examined the expression of CCR4, CXCR3, and CCR5 on CD4(+) lymphocytes directly isolated from a wide variety of normal and inflamed tissues. Extra-lymphoid tissues contained only memory lymphocytes, many of which were activated (CD69(+)). As predicted by classical studies, skin lymphocytes were enriched in CLA expression whereas intestinal lymphocytes were enriched in alpha(4)beta(7) expression. CCR4 was expressed at high levels by skin-infiltrating lymphocytes, at lower levels by lung and synovial fluid lymphocytes, but never by intestinal lymphocytes. Only the high CCR4 levels characteristic of skin lymphocytes were associated with robust chemotactic and adhesive responses to TARC, consistent with a selective role for CCR4 in skin lymphocyte homing. In contrast, CXCR3 and CCR5 were present on the majority of lymphocytes from each non-lymphoid tissue examined, suggesting that these receptors are unlikely to determine tissue specificity, but rather, may play a wider role in tissue inflammation.

The identification, characterization, and distribution of guinea pig CCR4 and epitope mapping of a blocking antibody.

Th2 lymphocytes play a central role in the control and maintenance of allergic inflammation. The chemokine receptor CCR4 is preferentially expressed on the surface of Th2 lymphocytes polarised in vitro. However, CCR4 is found on the surface of a significant proportion of circulating memory T lymphocytes, some of which are capable of producing the Th1-associated cytokine interferon gamma. To investigate the function of CCR4 on guinea pig (gp) T lymphocytes, we identified the open-reading frame of gpCCR4, which encodes a 361-amino acid protein with 88 and 81% amino acid identity to human and murine CCR4 sequences, respectively. Cells transfected with gpCCR4 migrated toward the human and murine orthologues of the CCR4 ligands, macrophage-derived chemokine and thymus and activation-regulated chemokine. Surface expression of CCR4, using an anti-human CCR4 monoclonal antibody, 10E4, was detected on approximately 12% of guinea pig peripheral blood T helper cells, and CCR4(+) guinea pig thymocytes were detected in low numbers. However, CCR4(+) T helper cells constituted approximately 9% of the T lymphocyte population within the normal guinea pig lung and 52% of the guinea pig bronchoalveolar lavage fluid, which is consistent with a role for CCR4 in T lymphocyte development and trafficking through normal tissues. Subsequent analysis of chimeric chemokine receptors indicated that 10E4, a functional inhibitor of gpCCR4 responses, recognized the amino terminus of CCR4.