We detect and interactively visualize occurrence, frequency, sequence, and clustering of extraintestinal manifestations (EIM) and associated immune disorders (AID) in 30,334 inflammatory bowel disease (IBD) patients (Crohn’s disease (CD) n = 15924, ulcerative colitis (UC) n = 11718, IBD unclassified, IBD-U n = 2692, 52% female, median age 40 years (IQR: 25)) with artificial intelligence (AI). 57% (CD > UC 60% vs. 54%, p < 0.00001) had one or more EIM and/or AID. Mental, musculoskeletal and genitourinary disorders were most frequently associated with IBD: 18% (CD vs. UC 19% vs. 16%, p < 0.00001), 17% (CD vs. UC 20% vs. 15%, p < 0.00001) and 11% (CD vs. UC 13% vs. 9%, p < 0.00001), respectively. AI detected 4 vs. 5 vs. 5 distinct EIM/AID communities with 420 vs. 396 vs. 467 nodes and 11,492 vs. 9116 vs. 16,807 edges (links) in CD vs. UC vs. IBD, respectively. Our newly developed interactive free web app shows previously unknown communities, relationships, and temporal patterns—the diseasome and interactome.

AbstractPurpose: Personalized medicine attempts to predict survival time for each patient, based on their individual tumor molecular profile. We investigate whether our survival learner in combination with a dimension reduction method can produce useful survival estimates for a variety of patients with cancer. Experimental Design: This article provides a method that learns a model for predicting the survival time for individual patients with cancer from the PanCancer Atlas: given the (16,335 dimensional) gene expression profiles from 10,173 patients, each having one of 33 cancers, this method uses unsupervised nonnegative matrix factorization (NMF) to reexpress the gene expression data for each patient in terms of 100 learned NMF factors. It then feeds these 100 factors into the Multi-Task Logistic Regression (MTLR) learner to produce cancer-specific models for each of 20 cancers (with >50 uncensored instances); this produces “individual survival distributions” (ISD), which provide survival probabilities at each future time for each individual patient, which provides a patient's risk score and estimated survival time. Results: Our NMF-MTLR concordance indices outperformed the VAECox benchmark by 14.9% overall. We achieved optimal survival prediction using pan-cancer NMF in combination with cancer-specific MTLR models. We provide biological interpretation of the NMF model and clinical implications of ISDs for prognosis and therapeutic response prediction. Conclusions: NMF-MTLR provides many benefits over other models: superior model discrimination, superior calibration, meaningful survival time estimates, and accurate probabilistic estimates of survival over time for each individual patient. We advocate for the adoption of these cancer survival models in clinical and research settings.

BACKGROUND: Many lung transplants fail due to chronic lung allograft dysfunction (CLAD). We recently showed that transbronchial biopsies (TBBs) from CLAD patients manifest severe parenchymal injury and dedifferentiation, distinct from time-dependent changes. The present study explored timeselective and CLAD-selective transcripts in mucosal biopsies from the third bronchial bifurcation (3BMBs), compared to those in TBBs. METHODS: We used genome-wide microarray measurements in 324 3BMBs to identify CLAD-selective changes as well as time-dependent changes and develop a CLAD classifier. CLAD-selective transcripts were identified with linear models for microarray data (limma) and were used to build an ensemble of 12 classifiers to predict CLAD. Hazard models and random forests were then used to predict the risk of graft loss using the CLAD classifier, transcript sets associated with rejection, injury, and time. RESULTS: T cell-mediated rejection and donor-specific antibody were increased in CLAD 3BMBs but most had no rejection. Like TBBs, 3BMBs showed a time-dependent increase in transcripts expressed in inflammatory cells that was not associated with CLAD or survival. Also like TBBs, the CLAD-selective transcripts in 3BMBs reflected severe parenchymal injury and dedifferentiation, not inflammation or rejection. While 3BMBs and TBBs did not overlap in their top 20 CLAD-selective transcripts, many CLAD-selective transcripts were significantly increased in both for example LOXL1, an enzyme controlling matrix remodeling. In Cox models for one-year survival, the 3BMB CLAD-selective transcripts and CLAD classifier predicted graft loss and correlated with CLAD stage. Many 3BMB CLAD-selective transcripts were also increased by injury in kidney transplants and correlated with decreased kidney survival, including LOXL1. CONCLUSIONS: Mucosal and transbronchial biopsies from CLAD patients reveal a diffuse molecular injury and dedifferentiation state that impacts prognosis and correlates with the physiologic disturbances. CLAD state in lung transplants shares features with failing kidney transplants, indicating elements shared by the injury responses of distressed organs. J Heart Lung Transplant 2022;41:1689- 1699 (c) 2022 International Society for Heart and Lung Transplantation. All rights reserved.

Transplanted lungs suffer worse outcomes than other organ transplants with many developing chronic lung allograft dysfunction (CLAD), diagnosed by physiologic changes. Histology of transbronchial biopsies (TBB) yields little insight, and the molecular basis of CLAD is not defined. We hypothesized that gene expression in TBBs would reveal the nature of CLAD and distinguish CLAD from changes due simply to time posttransplant. Whole-genome mRNA profiling was performed with microarrays in 498 prospectively collected TBBs from the INTERLUNG study, 90 diagnosed as CLAD. Time was associated with increased expression of inflammation genes, for example, CD1E and immunoglobulins. After correcting for time, CLAD manifested not as inflammation but as parenchymal response-to-wounding, with increased expression of genes such as HIF1A, SERPINE2, and IGF1 that are increased in many injury and disease states and cancers, associated with development, angiogenesis, and epithelial response-to-wounding in pathway analysis. Fibrillar collagen genes were increased in CLAD, indicating matrix changes, and normal transcripts were decreased-dedifferentiation. Gene-based classifiers predicted CLAD with AUC 0.70 (no time-correction) and 0.87 (time-corrected). CLAD related gene sets and classifiers were strongly prognostic for graft failure and correlated with CLAD stage. Thus, in TBBs, molecular changes indicate that CLAD primarily reflects severe parenchymal injury-induced changes and dedifferentiation.

Purpose The INTERHEART study previously used microarray assessment of 889 heart transplant biopsies to develop the Molecular Microscope Diagnostic System (MMDx) based on expression of rejection-associated transcripts (RATs). The present study reclassified the rejection-related states in an expanded set of 1320 prospectively collected biopsies from 645 patients from 13 centers. Methods Biopsies were classified by ensembles of classifiers and analyzed for left ventricular ejection fraction (LVEF) and survival. Results New algorithms identified 853 No rejection (NR), 179 ABMR, 76 TCMR, 13 Mixed, 161 possible ABMR (pABMR), and 38 possible TCMR (pTCMR). No rejection was subclassified as NR-Normal 462, NR-Minor 359, and NR-Early-injury 32 (Figure 1A). Compared to NR-Normal, NR-Minor biopsies had mild elevation of many inflammation transcripts (e.g. IFNG-inducible genes) and to a lesser extent parenchymal injury transcripts. NR-minor biopsies were often designated as TCMR1R by histology. In all NR biopsies, NR-Minor scores and histologic TCMR1R increased through the first year, peaking about one year, suggesting that Minor inflammation is a late response to injury, unrelated to rejection. LVEF was similar in NR-Normal and NR-Minor but depressed in TCMR and Early injury. In a 3-year post-biopsy survival analysis, NR-Minor and NR-Normal had similar survival. TCMR and Early-injury were associated with increased graft loss, but many losses were not related to rejection. Surprisingly, 76 hearts with ABMR (149 biopsies) and follow-up data had only 3 losses within 3 years post-biopsy (Figure 1B). Conclusion Many biopsies with molecular no rejection develop minor increases in rejection-related transcripts in the first year, often called TCMR1R by histology, with no apparent effects on function or survival. An unexpected finding was that molecular ABMR was associated with very few graft losses over three years. (ClinicalTrials.gov #NCT02670408).

AbstractObjectiveCOPA syndrome is a genetic disorder of retrograde cis‐Golgi vesicle transport that leads to upregulation of pro‐inflammatory cytokines (mainly IL‐1β and IL‐6) and the development of interstitial lung disease (ILD). The impact of COPA syndrome on post‐lung transplant (LTx) outcome is unknown but potentially detrimental. In this case report, we describe progressive allograft dysfunction following LTx for COPA‐ILD. Following the failure of standard immunosuppressive approaches, detailed cytokine analysis was performed with the intention of personalising therapy.MethodsMultiplexed cytokine analysis was performed on serum and bronchoalveolar lavage (BAL) fluid obtained pre‐ and post‐LTx. Peripheral blood mononuclear cells (PMBCs) obtained pre‐ and post‐LTx were stimulated with PMA, LPS and anti‐CD3/CD28 antibodies. Post‐LTx endobronchial biopsies underwent microarray‐based gene expression analysis. Results were compared to non‐COPA LTx recipients and non‐LTx healthy controls.ResultsMultiplexed cytokine analysis showed rising type I/II IFNs, and IL‐6 in BAL post‐LTx that decreased following treatment of acute rejection but rebounded with further clinical deterioration. In vitro stimulation of PMBCs suggested that myeloid cells were driving deterioration, through IL‐6 signalling pathways. Tocilizumab (IL‐6 receptor antibody) administration for 3 months (4 mg kg−1, monthly) effectively suppressed IL‐6 levels in BAL. Mucosal gene expression profile following tocilizumab suggested greater similarity to normal.ConclusionClinical effectiveness of IL‐6 receptor blockade was not observed. However, we identified IL‐6 upregulation associated with graft injury, effective IL‐6 suppression with tocilizumab and evidence of beneficial effect on molecular transcripts. This mechanistic analysis suggests a role for IL‐6 blockade in post‐LTx care that should be investigated further.

Purpose The molecular pathogenesis of chronic lung allograft dysfunction (CLAD) is poorly understood. We hypothesized that microarray analysis of TBBs and mucosal biopsies from the third bronchial bifurcation (3BMBs) would provide insight into biological pathways through gene ontology (GO) analysis. Methods We performed gene expression microarray analysis of 498 TBBs and 324 3BMBs from 10 international centers, comparing CLAD+ and CLAD- cases. CLAD was defined per 2019 ISHLT guidelines. Bayes-moderated t-tests were used to identify CLAD-associated genes. We performed GO biological process analysis with CLAD-associated genes with false discovery rate < 0.05. We also identified CLAD-associated transcripts after correcting for time post-transplant and excluding biopsies from patients diagnosed with rejection or infection. Results In TBBs CLAD-associated GO terms described lung development regulation, injury response, epithelial proliferation, epithelial-mesenchymal transition, and angiogenesis (Fig. 1A). In 3BMBs the GO terms included endocytosis and intracellular trafficking, cell-cell/cell-substrate adhesion, collagen/matrix metabolism, but unlike TBBs included immunity (Fig. 1B). Time correction did not affect transcript associations with CLAD in TBBs but diminished them in 3BMBs. Excluding biopsies with coexistent rejection or infection removed associations in TBBs but strengthened them in 3BMBs even after correcting for time post-transplant. Conclusion In our biopsy set the molecular CLAD phenotype resembles a parenchymal response to wounding that overlaps other pathologies. CLAD-associated molecular changes were independent of time in TBBs but not 3BMBs, suggesting that CLAD is primarily a time dependent airway process involving injury and either adaptive or innate immunity. Unlike TBBs, molecular changes in 3BMBs were identifiable despite coexistent rejection or infection, suggesting this biopsy format may better capture CLAD information.

Molecular assessment of paired, single piece 3BMBs achieves excellent reproducibility in its key scores. TBB scores were less reproducible likely because of tissue heterogeneity and greater anatomic distance between samples. Measurements could be stabilized by combining ≥2 pieces per microarray. Importantly, molecular assessment with 3BMBs or TBBs requires fewer pieces (2-3) than TBB histology, which requires up to ten. These data support continued study of the diagnostic and prognostic utility of 3BMBs in lung transplantation.

Parenchymal injury and late changes (atrophy-fibrosis) can be mapped in heart transplant biopsies, and their presentation correlates with low LVEF and lower 3-year survival. Injury is often, but not always, associated with rejection. Severe acute injury and the late fibrosis phenotypes are often associated with TCMR. Thus parenchymal injury is the intermediate phenotype by which rejection mediates disturbed function and survival. ClinicalTrials.gov #NCT02670408.

BACKGROUND: We previously developed molecular assessment systems for lung transplant trans-bronchial biopsies (TBBs) with high surfactant and bronchial mucosal biopsies, identifying T-cell. mediated rejection ( TCMR) on the basis of the expression of rejection-associated transcripts, but the relationship of rejection to graft loss is unknown. This study aimed to develop molecular assessments for TBBs and mucosal biopsies and to establish the impact of molecular TCMR on graft survival. METHODS: We used microarrays and machine learning to assign TCMR scores to an expanded cohort of 457 TBBs (367 high surfactant plus 90 low surfactant) and 314 mucosal biopsies. We tested the score agreement between TBB-TBB, mucosal-mucosal, and TBB-mucosal biopsy pairs in the same patient. We also assessed the association of molecular TCMR scores with graft loss (death or retransplantation) and compared it with the prognostic associations for histology and donor-specific antibodies. RESULTS: The molecular TCMR scores assigned in all the TBBs performed similarly to those in high-surfactant TBBs, indicating that variation in alveolation in TBBs does not prevent the detection of TCMR. Mucosal biopsy pieces showed less piece-to-piece variation than TBBs. TCMR scores in TBBs agreed with those in mucosal biopsies. In both TBBs and mucosal biopsies, molecular TCMR was associated with graft loss, whereas histologic rejection and donor-specific antibodies were not. CONCLUSIONS: Molecular TCMR can be detected in TBBs regardless of surfactant and in mucosal biopsies, which show less variability in the sampled tissue than TBBs. On the basis of these findings, molecular TCMR appears to be an important predictor of the risk of future graft failure. TRIAL REGISTRATION: ClinicalTrials.gov NCT02812290.(C) 2020 International Society for Heart and Lung Transplantation. All rights reserved.

BACKGROUND:Improved understanding of lung transplant disease states is essential because failure rates are high, often due to chronic lung allograft dysfunction. However, histologic assessment of lung transplant transbronchial biopsies (TBBs) is difficult and often uninterpretable even with 10 pieces. METHODS:We prospectively studied whether microarray assessment of single TBB pieces could identify disease states and reduce the amount of tissue required for diagnosis. By following strategies successful for heart transplants, we used expression of rejection-associated transcripts (annotated in kidney transplant biopsies) in unsupervised machine learning to identify disease states. RESULTS:All 242 single-piece TBBs produced reliable transcript measurements. Paired TBB pieces available from 12 patients showed significant similarity but also showed some sampling variance. Alveolar content, as estimated by surfactant transcript expression, was a source of sampling variance. To offset sampling variation, for analysis, we selected 152 single-piece TBBs with high surfactant transcripts. Unsupervised archetypal analysis identified 4 idealized phenotypes (archetypes) and scored biopsies for their similarity to each: normal; T-cell‒mediated rejection (TCMR; T-cell transcripts); antibody-mediated rejection (ABMR)-like (endothelial transcripts); and injury (macrophage transcripts). Molecular TCMR correlated with histologic TCMR. The relationship of molecular scores to histologic ABMR could not be assessed because of the paucity of ABMR in this population. CONCLUSIONS:Molecular assessment of single-piece TBBs can be used to classify lung transplant biopsies and correlated with rejection histology. Two or 3 pieces for each TBB will probably be needed to offset sampling variance.

Purpose In heart and kidney transplants, rejection is a major cause of graft loss. In kidneys, antibody-mediated rejection (ABMR) is more important than T cell-mediated rejection (TCMR), and molecules predict graft loss better than histology (JASN 26 (7):1711-1720, 2015). We examined the relative importance of ABMR vs TCMR in heart transplant endomyocardial biopsies (EMBs), and the molecules predicting graft survival. Methods The INTERHEART population includes 1219 transplant biopsies from 8 centers in Canada, USA, Australia and Europe. Gene expression was studied using microarrays, selecting the most recent biopsy per patient. Random forest classifiers were used to assess predictive accuracy and determine the importance of molecular predictors, including gene sets and scores from analyses in a reference set of 889 EMBs. Results We studied 3-year survival in 484 patients with follow-up times. Graft failure occurred in 60 patients. Median follow-up was 435 days; biopsies were mainly for indications. Surprisingly, TCMR was a greater short-term hazard than ABMR. The molecular archetype clusters for TCMR and injury (Fig. 1) had the highest risk of graft failure. Fig. 2 combines these clusters since they have similar characteristics. Conclusion Unlike kidneys, graft loss (particularly within one year) after EMB is highly associated with TCMR but not ABMR. TCMR may reflect failure of immunosuppression or non-adherence. This difference between the heart and renal transplant populations raises the possibility that TCMR is relatively more destructive, and ABMR less destructive, in heart than in kidney transplants. ClinicalTrials.gov # NCT02670408

BACKGROUND:We previously reported a microarray-based diagnostic system for heart transplant endomyocardial biopsies (EMBs), using either 3-archetype (3AA) or 4-archetype (4AA) unsupervised algorithms to estimate rejection. In the present study we examined the stability of machine-learning algorithms in new biopsies, compared 3AA vs 4AA algorithms, assessed supervised binary classifiers trained on histologic or molecular diagnoses, created a report combining many scores into an ensemble of estimates, and examined possible automated sign-outs. METHODS:We studied 889 EMBs from 454 transplant recipients at 8 centers: the initial cohort (N = 331) and a new cohort (N = 558). Published 3AA algorithms derived in Cohort 331 were tested in Cohort 558, the 3AA and 4AA models were compared, and supervised binary classifiers were created. RESULTS:A`lgorithms derived in Cohort 331 performed similarly in new biopsies despite differences in case mix. In the combined cohort, the 4AA model, including a parenchymal injury score, retained correlations with histologic rejection and DSA similar to the 3AA model. Supervised molecular classifiers predicted molecular rejection (areas under the curve [AUCs] >0.87) better than histologic rejection (AUCs <0.78), even when trained on histology diagnoses. A report incorporating many AA and binary classifier scores interpreted by 1 expert showed highly significant agreement with histology (p < 0.001), but with many discrepancies, as expected from the known noise in histology. An automated random forest score closely predicted expert diagnoses, confirming potential for automated signouts. CONCLUSIONS:Molecular algorithms are stable in new populations and can be assembled into an ensemble that combines many supervised and unsupervised estimates of the molecular disease states.