Fibrinogen binding to human blood platelets: effect of gamma chain carboxyterminal structure and length

Recent evidence suggests that fibrinogen binding to platelets is mediated by the 12 carboxyterminal amino acid residues of the gamma chain. Because human plasma fibrinogen gamma chains differ in mol wt and carboxyterminal amino acid sequence, we examined the effect of such gamma chain heterogeneity on platelet-fibrinogen interactions, using two fibrinogens of distinct composition, separated by ion exchange chromatography. One fibrinogen possessed only gamma chains of mol wt 50,000 (F gamma 50), the predominant gamma chain species found in plasma. The other fibrinogen possessed equal amounts of gamma chains with mol wt 50,000 and 57,500 (F gamma 50,57.5), with the longer gamma chain (gamma 57.5) possessing an amino acid extension at the carboxyterminal end. The latter fibrinogen was 50% less effective than F gamma 50 in supporting ADP-induced platelet aggregation at concentrations of .01 to 2 mg/mL. Scatchard analysis revealed no difference in the binding affinities of the two fibrinogens to ADP- treated platelets, but the amount of F gamma 50,57.5 that was bound to platelets at saturation was only 50% that of F gamma 50. Fibrinogen receptors that remained unoccupied in the presence of saturating concentrations of F gamma 50,57.5, however, could be occupied by fresh F gamma 50. Excess unlabeled F gamma 50 displaced both radiolabeled fibrinogens from activated platelets, and both fibrinogens bound to the same platelet receptor, as judged by the inhibition of binding to stimulated platelets by a monoclonal antibody directed against the glycoprotein (GP) IIb/IIIa complex. Furthermore, an intact GPIIb/IIIa complex was required for these reactions, since platelets incubated with EDTA at 37 degrees C at alkaline pH failed to aggregate and bound neither fibrinogen in response to ADP following recalcification. Approximately 50% of each fibrinogen bound irreversibly to platelets after one hour and failed to dissociate in the presence of 10 mmol/L of EDTA or excess unlabeled F gamma 50. The data demonstrate that heterodimeric F gamma 50,57.5 binds less well to platelets and supports platelet aggregation only half as well as homodimeric F gamma 50. These results support prior conclusions that the carboxyterminal portion of the gamma chain is important in platelet-fibrinogen interactions, and suggest that the 20 amino acid, hydrophobic gamma chain carboxyterminal extension of F gamma 50,57.5 may sterically hinder the interaction of this fibrinogen with platelet receptors.     

Inter-interpreter agreement for ABR tracings. Results from 100 patients under otoneurologic investigation

The auditory brainstem response (ABR) tracings of 100 patients under investigation for a neural lesion were examined independently by two audiologists. Each tracing for the ear under question was classified as normal or abnormal. There was 94% agreement. Suggestions are made as to how to establish quality control in audiology clinics.     

Network hubs cease to be influential in the presence of low levels of advertising

Attempts to find central “influencers,” “opinion leaders,” “hubs,” “optimal seeds,” or other important people who can hasten or slow diffusion or social contagion has long been a major research question in network science. We demonstrate that opinion leadership occurs only under conventional but implausible scope conditions. We demonstrate that a highly central node is a more effective seed for diffusion than a random node if nodes can only learn via the network. However, actors are also subject to external influences such as mass media and advertising. We find that diffusion is noticeably faster when it begins with a high centrality node, but that this advantage only occurs in the region of parameter space where external influence is constrained to zero and collapses catastrophically even at minimal levels of external influence. Importantly, nearly all prior agent-based research on choosing a seed or seeds implicitly occurs in the network influence only region of parameter space. We demonstrate this effect using preferential attachment, small world, and several empirical networks. These networks vary in how large the baseline opinion leadership effect is, but in all of them it collapses with the introduction of external influence. This implies that, in marketing and public health, advertising broadly may be underrated as a strategy for promoting network-based diffusion.

Among the central theoretical and practical attractions of social network analysis is the promise that key nodes, known as “opinion leaders” or “influentials,” hold structural power to change the ideas and behaviors of entire social systems (1–3). An extensive literature in sociology, physics, and network science centers on how best to measure network centrality. From the beginning, much of this literature takes as its motivation identifying a node or nodes that are optimal seeds for diffusion (4–8).^* For instance, a seminal study of how doctors prescribe new drugs ascribed this behavior to key doctors in the advice network (11). In such applied contexts as “viral” marketing and public health outreach, opinion leadership suggests the promise that a structurally important node (and, by extension, the social network analyst who can identify that node) is the key to controlling the spread of a product, health behavior, or other idea or behavior (2, 3, 8, 12–14).The influentials literature focuses on network sources of information, but in most realistic scenarios people have sources of information that transcend the network (15–17). Introducing these nonnetwork sources of information may qualitatively change the nature of diffusion, and specifically the role of a highly central hub or hubs. In many theories and simulations, agents are constrained to only observe information through a social graph, but real people are not so myopic. Even if we are most attentive to word of mouth from our social ties, we also learn about new ideas and behaviors from mass media, advertising, government mandates, and even direct observation of events. If it begins raining and everyone opens her umbrella, the proximate cause of this behavior is a response to nature rather than information spreading through a social network (18). Some diffusion models meaningfully incorporate roles for external sources of information (15, 19, 20), but other models effectively assume an entirely word-of-mouth process even if their narrative theory allows for external influence (3).The computational experiment we present in this article contributes to a large body of social networks literature on influentials and opinion leadership (7, 8), but takes as its microfoundations a diffusion model from marketing that involves both network-based diffusion and external influence from sources like advertising (15). We conduct a large-scale computer simulation in which we seed diffusion with either the most central node or a node chosen at random in various empirical and algorithmically generated networks.^† We test the opinion leadership hypothesis for various points in parameter space where one axis is the strength of network-based diffusion (e.g., “word of mouth”) and the other axis is the strength of an external force (e.g., advertising and mass media). We measure the strength of opinion leadership for each point in parameter space by how much faster diffusion occurs when the initial node is highly central versus chosen at random.The experiment adapts a mixed-influence model outlined by Bass (15, 21–23) to test whether the effect of central nodes on diffusion is robust to the presence of external influence. In the Bass model, people are exposed to information about the innovation from two sources: interpersonal imitation (with a density-dependent hazard) and external influence (with a constant hazard). Interpersonal influence represents the effect of word of mouth (or closely analogous processes like local network externalities or person-to-person spread) (24, 25). External influence represents the effect of advertising, mass media, internet search, or government mandates (15, 17, 23, 26). Traditionally, the Bass model is represented as a differential equation that measures diffusion in aggregate over time. The aggregate approach has the advantage of simplicity but makes it impossible to integrate network structure. We therefore adapt the Bass model to an agent-based model, which allows for potential emergent properties of unequal influence between nodes based on their structural positions.The Bass model defines the rate of new adoptions in aggregate as follows:ΔNt=a+bNtNmax−Nt,where Nt is the cumulative number of people who have adopted as of time t, a is the coefficient of external influence, b is the coefficient of interpersonal influence, and Nmax is the asymptotic number of people who will ever adopt. To include the effect of network structure on individual adoption, we adapt this equation to an agent-based model. In the agent-based model, for each agent i at time t:piadoptsattimet|ihasnotadoptedbeforet=α+βfractionofi''sneighborswhoadoptedbeforetimet.α is a constant hazard of adoption, representing the weight given to advertising and other external influences on diffusion, and β is the weight given to social or network influence.^‡ To ensure that α and β are on comparable scales, we allow them to range between 0 and a maximum value that saturates the network with a consistent probability. We identify these maxima with a separate set of simulations, which identify the values at which α and β saturate the network in 100 ticks or less in 50% of trials. We refer to these α and β maxima as “LD50,” as a metaphor for the standard “lethal dose 50%” metric in toxicology. Full details of estimating the LD50 values appear in SI Appendix, Determining parameter range.^§ To highlight changes at the lowest end of parameter space, we explore both dimensions of the parameter space on a log scale. In both the aggregate and agent-based Bass models, once a person adopts, she cannot abandon the innovation, meaning the number of adopters increases monotonically.Our experimental setup varies the seed, meaning the initial innovator in the simulation. In the simulation, innovations start at one person, the seed, and spread outward from that person.^¶ Our control condition seeds the innovation with a randomly chosen person in the network. Our treatment condition seeds the innovation with the most central person in the network, as measured by betweenness. In most networks, betweenness is right-skewed so in our networks the most central node is anywhere from six to several hundred SDs above the mean.^# We test the effects on preferential attachment networks (shown in Fig. 1) and small world networks generated in igraph (27) as well as the giant components of the Democratic National Committee email network (548 nodes and 2,442 edges), Enron email network (33,696 nodes and 180,811 edges), and a network of retweets and mentions on Twitter (532,325 nodes and 694,606 edges). We focus on preferential attachment networks in Figs. 1 and ​and22 but show robustness of our key finding to all these networks in Fig. 3 and SI Appendix.Open in a separate windowFig. 1.Example of a preferential attachment network generated with the Barabási–Albert algorithm (30) with 1,000 nodes, one edge per node, and an exponent of 1. We focus on this network as relatively favorable to opinion leadership but in Fig. 3 and SI Appendix show other networks. The yellow node is the highest betweenness node used to test the effect of influentials.Open in a separate windowFig. 2.Cumulative number of adopters, denoted CDF, in simulations assuming only network diffusion (α = 0, β = LD50) in a preferential attachment network (1,000 nodes, one edge per node). The plot shows the confidence interval around the mean of both experimental conditions: simulations seeded with the highest betweenness node and simulations seeded with a randomly selected node. Seeding with the highest betweenness person saturates half the network (indicated by the red horizontal line) over twice as fast.Open in a separate windowFig. 3.Ratio of mean time to midsaturation in simulations targeting a randomly chosen node in the network versus targeting the highest betweenness node. A shows the full parameter space for randomly generated preferential attachment networks (1,000 nodes, one edge per node). The gray cells represent right censored cases. Targeting a highly central person results in adoption that is over twice as fast, but only when there is no effect of advertising (α=0). B shows a summary across several algorithmically generated and empirical networks as we assume high levels of network diffusion (β = LD50) but vary external influence (α) as a percentage of each LD50 value, plotted on a logarithmic scale. This is the equivalent to the top row of cells in A, but substituting a y axis for the heat dimension and showing more networks. Across all these networks, targeting a highly central person results in faster adoption, but only when there is no effect of advertising (α=0). The impact of highly central seeds approaches parity with random seeds at even very low positive levels of advertising.Fig. 2 shows the central tendencies of the cumulative distribution functions by random versus highest betweenness seed node given the assumption of peak social influence (β = 0, β = LD50).^|| Under those conditions, innovations that start with the most central person spread to half of the people in the network over twice as fast.As Fig. 2 indicates, the gap between the conditions is approximately widest at time to 50% adoption (cumulative distribution function [CDF] = 500, displayed as a red horizontal line), making it the metric most favorable to opinion leadership. In addition, time to 50% adoption is much less vulnerable to right censorship than time to saturation. We use this metric, average time to 50% adoption, to summarize the full parameter space. In Fig. 3A, we demonstrate how diffusion speed on a preferential attachment network responds to varying the α and β parameters separately for random seeds and seeding at the highest betweenness node. The heat dimension shows the ratio of the mean time to 50% saturation for a random seed over that for a high centrality seed. (SI Appendix, Fig. S1 shows how this ratio is derived from Fig. 2.) Seeding with a highly central node has an advantage but only when α = 0. This advantage disappears quickly for all points in parameter space where α > 0, dropping precipitously at the next interval (α = 0.26% of LD50), and the advantage of a highly central seed node almost completely vanishes for points in parameter space where α > 3% of LD50.The heat map in Fig. 3A only illustrates results for preferential attachment networks, but in Fig. 3B we provide sparklines summarizing several networks for the plane of parameter space where β = LD50 (i.e., the equivalent of the top row of the heat map). When α = 0, the effect of a highly central seed node varies substantially by the type of network, being trivial in a small world, but substantial in the three empirical networks. However, the finding from preferential attachment networks that the advantage of seeding with the peak betweenness node collapses rapidly when α > 0 replicates in all other networks, no matter how strong the highly central seed node effect is when α=0. Targeting the central node materially speeds adoption only in the region of parameter space where there is no external influence (α = 0). In all networks, there is a precipitous drop in the effect of highly central seeding as α goes from zero to 0.26% of the LD50 and the highly central seed effect is essentially absent when α reaches even a few percentage points of its LD50 value. SI Appendix, Figs. S3–S6 contains full heat maps for all networks listed in the sparklines plot of Fig. 3.These findings indicate that the positive effect of targeting the most central node as opinion leader is subject to a highly restrictive scope condition. Previous research has shown that opinion leadership requires substantial inequality in centrality (28), but many phenomena of interest meet that scope condition. Here, we show the much more demanding scope condition of the absence of advertising or other forms of external influence. When no external influences are present, targeting a highly central person results in diffusion that can spread to half of the network faster than if a person were chosen at random, with the advantage being trivial for small world networks and an order of magnitude for the email networks. However, in the presence of external influences, even extremely weak external influences, identifying and seeding with an opinion leader do not lead to appreciably faster adoption of an innovation. This suggests that the simulation literature on optimal seeding to opinion leaders only applies under restrictive scope conditions that likely apply to few empirical scenarios. When diffusion follows the network strictly, as in the spread of a sexually transmitted disease (29) or clandestine communication with a cell structure, then centrality can have appreciable effects. However, the diffusion of a product, behavior, or belief, will normally involve some level of external influence, and even if that external influence is dwarfed by network influence, there should be no effect of the seed node’s network position so long as external influence exists at all.Adding in even weak advertising effects nullifies the impact of seeding with the most central node. Advertising creates a nonzero probability that people can adopt without exposure from other adopters, conceptually similar to increasing the number of seeds. Our findings thus suggest that advertisers or public health officials who are planning a campaign should consider that advertising can also promote network-based spread and may do so more efficiently than identifying and recruiting a highly central seed node. This implies a return to the early “two-step flow” model, in which most people adopt based on influence from numerous minor opinion leaders of purely local influence, who in turn got information from mass media (19, 20).There is substantial evidence that ideas and behaviors spread via interpersonal influence, but this is neither the same thing as an emergent property of critical importance for a highly central node nor a practical upshot that seeding with a central node is important under realistic circumstances. While social connections remain important for the spread of ideas, products, and behaviors, our simulations highlight the importance of the context in which those networks are embedded. Our results imply that in studies of diffusion the effect of mass media and advertising on the spread of a trend changes the nature of network-based diffusion, even if mass media and advertising have a weak role in and of themselves. To understand the drivers behind a trend, it is not sufficient to understand how well positioned the initial adopter is to spread the trend. We must also understand whether advertising or other broad forces like mass media, government mandates, or search engines seed the trend widely, and thereby render the choice of the initial adopter, no matter how central to the network, irrelevant.     

Tubeufiales,ord. nov., integrating sexual and asexual generic names

Tubeufiaceae is based on the generic type Tubeufia, which is characterized by superficial, oval and bright ascomata, bitunicate asci, mostly long fusiform to filiform, transeptate ascospores and hyphomycetous asexual states with helicosporous conidia. Most species in this family are saprobic on terrestrial woody substrates and some are aquatic. Their distinct morphology as well as combined LSU, SSU and TEF1 sequence analysis show that Tubeufiaceae should be accommodated in a new order Tubeufiales, which is introduced in this paper. Phylogenetic analyses of combined LSU and ITS sequences were used to resolve genera and species within the family Tubeufiaceae. In this study, we examine and incorporate sexual and asexual states of genera in Tubeufiales to provide a modern treatment, based on single names. An epitype for Tubeufia javanica, the type species of Tubeufia, is designated and represents Tubeufia sensu stricto. The genera Acanthophiobolus, Acanthostigma, Boerlagiomyces, Chlamydotubeufia, Kamalomyces, Podonectria, Thaxteriella and Thaxteriellopsis are accepted, Acanthostigmina is reinstated, and the asexual genera Aquaphila, Helicoma, Helicomyces, Helicosporium and Tamhinispora are accepted in Tubeufiaceae. Three new genera Acanthohelicospora, Helicangiospora and Neoacanthostigma are introduced. The genus Bifrontia is added to the family based on morphological similarity. The incongruous morphological genera Acanthostigmella, Amphinectria, Chaetocrea, Chaetosphaerulina, Glaxoa, Malacaria, Melioliphila, Paranectriella, Puttemansia, Rebentischia and Uredinophila are excluded from Tubeufiaceae despite having characteristic ascomata with setae and multiseptate long spores. A key to genera accepted in Tubeufiaceae is provided.     

Inhibition of lung fibroblast growth by human lung mononuclear cells

Mononuclear cell-fibroblast interactions in the normal human lung are poorly understood. To investigate these interactions, we characterized the effect on lung fibroblast growth of supernatants from nonadherent lung mononuclear cells cultured with or without phytohemagglutinin (PHA). Supernatants from unstimulated cells had no effect on fibroblast growth. Supernatants from lung mononuclear cells stimulated by PHA inhibited fibroblast growth in a dose-dependent fashion. This inhibition was mediated by a nondialyzable, noncytotoxic, soluble factor(s) that was partially heat labile and sensitive to digestion with trypsin. Production of this factor(s) was dependent on T-lymphocytes and occurred within 4 h of the initiation of lung mononuclear cell-PHA cultures. Because prostaglandins can mediate mononuclear cell effects on fibroblast growth, we examined the relationship between the inhibition of fibroblast growth caused by PHA-stimulated nonadherent lung mononuclear cells and lung fibroblast prostaglandin production. The nonadherent lung mononuclear cell supernatants stimulated fibroblast PGE2 production, and exogenous PGE2 was capable of inhibiting fibroblast growth. However, the inhibition of fibroblast growth caused by nonadherent lung mononuclear cells stimulated by PHA was not mediated by fibroblast prostaglandin production. Inhibition of lung fibroblast growth by stimulated lung mononuclear cells may be important in regulating fibrosis in the human lung.     

Reduced sarcoplasmic reticulum Ca(2+) load mediates impaired contractile reserve in right ventricular pressure overload

Myocardial contractile reserve is significantly attenuated in patients with advanced heart failure. The aim of this study was to identify mechanisms of impaired contractile reserve in a large animal model that closely mimics human myocardial failure. Progressive right ventricular hypertrophy and failure were induced by banding the pulmonary artery in kittens. Isometric contractile force was measured in right ventricular trabeculae (n=115) from age-matched Control and Banded feline hearts. Rapid cooling contractures (RCC) were used to determine sarcoplasmic reticulum (SR) Ca(2+) load while assessing the ability of changes in rate, adrenergic stimulation and bath Ca(2+) to augment contractility. The positive force-frequency relationship and robust pre- and post-receptor adrenergic responses observed in Control trabeculae were closely paralleled by increases in RCC amplitude and the RCC2/RCC1 ratio. Conversely, the severely blunted force-frequency and adrenergic responses in Banded trabeculae were paralleled by an unchanged RCC amplitude and RCC2/RCC1 ratio. Likewise, supraphysiologic levels of bath Ca(2+) were associated with severely reduced contractility and RCC amplitude in Banded trabeculae compared to Controls. There were no differences in myofilament Ca(2+) sensitivity or length-dependent increases in contractility between Control and Banded trabeculae. There was a significant decrease in SR Ca(2+)-ATPase pump abundance and phosphorylation of phospholamban and ryanodine receptor in Banded trabeculae compared with Controls. A reduced ability to increase SR Ca(2+) load is the primary mechanism of reduced contractile reserve in failing feline myocardium. The similarity of impaired contractile reserve phenomenology in this feline model and transplanted hearts suggests mechanistic relevance to human myocardial failure.     

Applying the CiPA approach to evaluate cardiac proarrhythmia risk of some antimalarials used off‐label in the first wave of COVID‐19

We applied a set of in silico and in vitro assays, compliant with the Comprehensive In Vitro Proarrhythmia Assay (CiPA) paradigm, to assess the risk of chloroquine (CLQ) or hydroxychloroquine (OH‐CLQ)‐mediated QT prolongation and Torsades de Pointes (TdP), alone and combined with erythromycin (ERT) and azithromycin (AZI), drugs repurposed during the first wave of coronavirus disease 2019 (COVID‐19). Each drug or drug combination was tested in patch clamp assays on seven cardiac ion channels, in in silico models of human ventricular electrophysiology (Virtual Assay) using control (healthy) or high‐risk cell populations, and in human‐induced pluripotent stem cell (hiPSC)‐derived cardiomyocytes. In each assay, concentration‐response curves encompassing and exceeding therapeutic free plasma levels were generated. Both CLQ and OH‐CLQ showed blocking activity against some potassium, sodium, and calcium currents. CLQ and OH‐CLQ inhibited I _Kr (half‐maximal inhibitory concentration [IC₅₀]: 1 µM and 3–7 µM, respectively) and I _K1 currents (IC₅₀: 5 and 44 µM, respectively). When combining OH‐CLQ with AZI, no synergistic effects were observed. The two macrolides had no or very weak effects on the ion currents (IC₅₀ > 300–1000 µM). Using Virtual Assay, both antimalarials affected several TdP indicators, CLQ being more potent than OH‐CLQ. Effects were more pronounced in the high‐risk cell population. In hiPSC‐derived cardiomyocytes, all drugs showed early after‐depolarizations, except AZI. Combining CLQ or OH‐CLQ with a macrolide did not aggravate their effects. In conclusion, our integrated nonclinical CiPA dataset confirmed that, at therapeutic plasma concentrations relevant for malaria or off‐label use in COVID‐19, CLQ and OH‐CLQ use is associated with a proarrhythmia risk, which is higher in populations carrying predisposing factors but not worsened with macrolide combination.

Study Highlights

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

The antimalarials chloroquine (CLQ) or hydroxychloroquine (OH‐CLQ) used off‐label in combination with a macrolide antibiotic (erythromycin [ERT] or azithromycin [AZI]) during the first wave of coronavirus disease 2019 (COVID‐19) have been long associated with cardiovascular side effects, namely QT prolongation. There is a paucity of nonclinical data, meeting the new Comprehensive In Vitro Proarrhythmia Assay (CiPA) paradigm, to properly assess the proarrhythmic risk of these drugs, used alone as antimalarials or combined with a macrolide, such as in some COVID‐19 clinical trials.

• WHAT QUESTION DID THIS STUDY ADDRESS?

What is the propensity of CLQ and OH‐CLQ to affect cardiac ion currents and to delay ventricular repolarization using the set of in vitro and in silico assays proposed by the CiPA approach? What is the associated safety margin compared with therapeutic plasma levels? Does the combination with AZI or ERT aggravate this risk?

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

The risk of QT prolongation and Torsades de Pointes (TdP) liability was consistently demonstrated in all CiPA assays for both CLQ and OH‐CLQ at exposure levels relevant for malaria treatment or off‐label use in COVID‐19 and is primarily due to I _Kr current blockade. This risk is higher in populations carrying predisposing factors. Combination with AZI or ERT does not worsen the CLQ/OH‐CLQ TdP risk in virtual ventricular myocyte models or human‐induced pluripotent stem cell‐derived cardiomyocytes.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

Our integrated nonclinical dataset supports the US Food and Drug Administration (FDA)/European Medicines Agency (EMA) recommendations to apply strong caution in CLQ and OH‐CLQ use, in particular with high‐dose regimens and/or in patients with additional risk factors, such as in COVID‐19. The combination of these drugs with a macrolide does not worsen the QT prolongation/TdP risk.     

Reactivity to Factor-VIII Concentrates of Lymphocytes from Patients with Haemophilia and von Willebrand''s Disease

S ummary . T o tcst the antigenicity of two materials used in the management of factor-VIII deficiency, lymphocytes from normal volunteers and from patients with haemophilia and with von Willebrand's disease (vWd) were exposed in vitro to cryoprecipitate and to glycine-precipitated factor VIII (Hemofil). As measured by incorporation of ¹⁴C-thymidine, the factor-VIII concentrates stimulated lymphocytes from some haemophiliacs and nontransfused normal donors, whereas lymphocytes from patients with vWd were unreactive. The data provide further immunological support for the concept that haemophilia and vWd are two separate and distinct disorders.     

Relationship of environmental exposures to the clinical phenotype of sarcoidosis

STUDY OBJECTIVES: Sarcoidosis is a granulomatous disorder with heterogeneous clinical manifestations, which are potentially reflective of a syndrome with different etiologies leading to similar histologic findings. We examined the relationship between environmental and occupational exposures, and the clinical phenotype of sarcoidosis. DESIGN: We performed a cross-sectional study of incident sarcoidosis cases that had been identified by A Case Control Etiologic Study of Sarcoidosis. Subjects were categorized into the following two groups: (1) pulmonary-only disease; and (2) systemic disease (with or without pulmonary involvement). Logistic regression was used to examine the associations of candidate exposures with clinical phenotype. SETTING: Ten academic medical centers across the United States. PATIENTS: The current study included 718 subjects in whom sarcoidosis had been diagnosed within 6 months of study enrollment. Patients met the following criteria prior to enrollment: (1) tissue confirmation of noncaseating granulomas on tissue biopsy on one or more organs within 6 months of study enrollment with negative stains for acid-fast bacilli and fungus; (2) clinical signs or symptoms that were consistent with sarcoidosis; (3) no other obvious explanation for the granulomatous disease; and (4) age > 18 years. MEASUREMENTS AND RESULTS: Several exposures were associated with significantly less likelihood of having extrapulmonary disease in multivariate analysis, including agricultural organic dusts and wood burning. The effects of many of these exposures were significantly different in patients of different self-defined race. CONCLUSIONS: The differentiation of sarcoidosis subjects on the basis of clinical phenotypes suggests that these subgroups may have unique environmental exposure associations. Self-defined race may play a role in the determination of the effect of certain exposures on disease phenotypes.