Repository for Publications and Research Data
News from the ETH Library
Recently Added
Magnetically powered microrobots: a medical revolution underway?
Item type: Other Journal Item
Chautems, Christophe; Zeydan, Burak; Charreyron, Samuel; et al. (2017)
Effect of supplementation of pelleted hazel (Corylus avellana) leaves on blood antioxidant activity, cellular immune response, and heart beat parameters in sheep1
Item type: Journal Article
Wang, Shaopu; Giller, Katrin; Hillmann, Edna; et al. (2019)
Hazel leaves (Corylus avellana) fed to sheep resulted in decreased methane emissions without negatively affecting feed intake and were found to have antioxidant properties in vitro. The objective of this study was to evaluate effects of hazel leaves, rich in tannins, on blood antioxidant activity, cellular immune response, and heart beat parameters in sheep. Four experimental pellets were produced by mixing alfalfa and hazel leaves in different proportions, including alfalfa alone as a control, 30% and 60% of hazel leaves, the latter also with 3.8% polyethylene glycol (PEG). Six adult, nonpregnant, nonlactating female sheep (71 ± 5.7 kg of body weight) were allocated to 4 treatments in a 6 × 4 crossover design with four 18-d periods. The diet consisted of experimental pellets and ryegrass-dominated hay (ratio 80% to 20% in dry matter), resulting in hazel leaf proportions of approximately 0%, 25%, and 50% in the total diet. Blood samples were collected at the end of each period to determine plasma total phenol concentration and markers of oxidative status as well as peripheral blood mononuclear cells (PBMC) activation and proliferation response in vitro. Heart rate (HR) and HR variability parameters were measured for 2 consecutive days in each period, during different activities (i.e., eating pellets or hay, or lying). Treatments were compared with multiple comparisons and contrast analysis was used to test for linear and quadratic relations. Compared with control, feeding a high dosage of hazel leaves enhanced (P = 0.006) the plasma total antioxidant capacity, which linearly (P = 0.016) increased with increasing level of hazel leaves in the diet. The total phenol concentration and activities of the antioxidant enzymes superoxide dismutase, catalase, and glutathione reductase in the plasma were not different (P ≥ 0.23) among the treatments; however, the latter slightly increased linearly (P = 0.047) with increasing hazel leaves proportion. No differences were observed in the activation and proliferation of PBMC among treatments. The HR decreased linearly (P ≤ 0.009) during pellet eating and lying and the root mean square of successive differences of interbeat intervals (RMSSD) increased linearly (P = 0.037) when lying with increasing level of hazel leaves in the diet. In conclusion, our findings indicate that hazel leaves are a promising supplement to improve oxidative status with no effect on cellular immune response and cardiac stress level of sheep.
Measure–Valued Perspectives on Generative Surrogates for Chaotic PDE Dynamics
Item type: Doctoral Thesis
Armegioiu, Tudor-Victor (2026)
Effectively Leveraging Data for Image and Video Restoration
Item type: Doctoral Thesis
Bernasconi, Michael (2026)
Image restoration is a fundamental task in computer vision, where the goal is reconstructing a high-quality image from a degraded version. In general this is an ill-posed task, as there are usually many possible high-quality images that can generate a given degraded image. Data-driven machine learning techniques are well suited to tackle such ill-posed problems and have become the standard approach to image restoration over the last decade. Network architectures, optimization strategies, and training data all have significant effects on the reconstruction quality and have been studied extensively. In this thesis, however, we focus on improving reconstruction quality by making sure we leverage all data that is available during inference.
In the first part, we utilize information about the degradation process to achieve higher quality single image super-resolution results. We introduce a method that is capable of performing arbitrary geometric transformations and is conditioned on the inverse downsampling transform. This conditioning prevents over-sharpening and under-sharpening artifacts in the output. It also enables user control over the level of sharpness in the restored image via adjusting the provided downsampling transform.
In the second part, we propose a method of injecting high-quality information from available reference images into existing single-image restoration methods. Prior to our work, the usage of high-quality reference images was limited to the task of super-resolution. Our method not only enables the usage of reference images in other restoration tasks, such as denoising or compression-artifact removal, but also sets a new state-of-the-art for reference-based super-resolution.
In the third part, we tackle the problem of video super-resolution. Specifically, we propose an add-on module for existing video super-resolution methods which enables propagation of temporal information over long distances. With this we are able to set a new state-of-the-art for both fixed-scale and arbitrary-scale video super-resolution. Our method also naturally allows the injection of high-quality reference images at any point in the video sequence.
In sum, this thesis presents several advances in the field of image and video restoration. These advances are made possible by carefully designing methods that are capable of leveraging all available input data.
Non-invasive Exhalomics for Assessing Rumen Fermentation in Dairy Cows: Linking the Exhaled Signal to Microbial Capacity and Expression
Item type: Doctoral Thesis
Islam, Md Zakirul (2026)
Rumen fermentation is a main determinant of how dairy cattle convert feed into milk. It sets the risk of metabolic disorders in the rumen such as ruminal acidosis, and is the primary source of enteric methane (CH4), a key determinant of livestock greenhouse gas emissions. Conventional assessment of rumen fermentation relies on invasive procedures that restrict their application in commercial settings and raise animal-welfare concerns. Fermentation is the end product of microbial activity, so assessing rumen function means reaching the microbial community itself and not only the acids and gases it releases. Sequencing surveys of the resident community do reach it, but describe what that community could do rather than what it is doing. This doctoral thesis establishes exhalomics, the metabolomics analysis of exhaled breath, as a non-invasive, high-resolution platform for characterising rumen fermentation dynamics in dairy cows, and anchors the exhaled signal to the microbial capacity and transcription that generate it. The work proceeds in three stages. The first establishes the breath measurement and tests it against invasive reference sampling and a methane-mitigating intervention (Chapters 2 to 4). The second anchors the exhaled signal in the microbial capacity and transcription that generate it (Chapter 5). The third determines what that signal predicts about fermentation state (Chapter 6).
In Chapter 2, secondary electrospray ionisation high-resolution mass spectrometry (SESI-HRMS) was used for the first time in ruminant research to identify and quantify exhaled volatile fatty acids (eVFA), acetate, propionate, and butyrate, in the exhalome of lactating dairy cows. The diurnal patterns of eVFA closely mirrored those of ruminal CH4 and H2 emissions measured by the GreenFeed system, with eVFA peaking approximately one hour before CH4. This establishes their utility as a non-invasive proxy for rumen VFA dynamics.
In Chapter 3 the approach was then validated against its invasive reference, directly comparing exhaled VFA (EX-VFA) with simultaneously collected ruminal VFA (RM-VFA) and Henry’s Law-predicted gas-phase VFA (HR-VFA). Rumen-cannulated cows fed low- and high-starch diets were employed for this purpose. No method × diet or method × time interaction emerged, a null result consistent with EX-VFA tracking the relative proportions and diurnal dynamics of RM-VFA. This supports its usefulness as a functional substitute for invasive rumen sampling.
In Chapter 4 the platform was carried into an intervention setting, investigating the combined effects of 3-nitrooxypropanol (3-NOP) and Acacia mearnsii tannin extract (TAN) on CH4 emissions and nitrogen partitioning in Holstein Friesian (HF) and Brown Swiss (BS) cows. A significant breed × 3-NOP interaction revealed greater CH4 mitigation in HF (21.7%) than in BS (13.0%), with no synergistic effect of 3-NOP and TAN. Coupled microbiome analysis of rumination bolus samples showed that HF cows harboured more propionate-associated genera (Succinivibrionaceae, Acetitomaculum) while BS retained acetate- and butyrate-linked associations. This suggests that the H2 spared from methanogenesis was redirected toward propionate pathways in HF, and offers a candidate explanation for the attenuated BS response. However, this finding needs more detailed studies to be confirmed.
Chapter 5 turned from the breath to the microbial engine beneath it. This was done by asking whether what the rumen microbial community encodes tells us what it is actually doing. In the four rumen-cannulated Braunvieh cows of Chapter 3, fed a low-starch (6.31% of DM) and a high-starch (16.2% of DM) diet in a switchback design, we separated the rumen microbiome into its genomic capacity and its expressed function by shotgun metagenomics and metatranscriptomics. Sequencing was performed on the 10:00 samples, the single diurnal timepoint at which breath, rumen-fluid metabolome, metagenome and metatranscriptome were all collected, giving 12 samples in which all four molecular layers coincide. Diet changed what the community did without changing which organisms were present. Relative abundances of 215 species-representative genomes did not separate the two diets, yet the genomic potential and the transcriptional activity of nine fermentation modules both shifted. Capacity and expression diverged: the two responded to diet in the same direction for only five of the nine modules, were essentially uncorrelated overall, and the discordance persisted under leave-one-cow-out resampling. This comparison rests on samples collected at 10:00, 1.5 h after the once-daily feed delivery, and the extent of the decoupling at other phases of the feeding cycle remains unresolved. Methanogenesis rose in genomic abundance with the high-starch diet while its transcription fell, and it was the only module whose every constituent enzyme moved upward; it is encoded by Methanobacteria present in 7 of the 215 genomes. Clostridia was the dominant class for six of the eight attributable modules, and Prevotella the leading genus for four of them. We then asked which of the four layers move together when diet shifts the fermentation system. Unsupervised statistical integration recovered a diet-associated axis shared by the rumen metabolome and the genomic potential, and the exhalome did not join it. On twelve samples this is a hypothesis rather than a settled result. I read it as two boundaries on the measurement. Of the layers measured, only the transcriptome reveals what the community is doing, and an exhaled signature captures aggregate fermentation state rather than individual metabolites.
In Chapter 6 the breath and rumen compartments were brought together at scale, to test whether a compact exhaled signature can track rumen fermentation state non-invasively, taking ruminal pH as the quantitative reference axis. We pooled 255 paired breath-rumen observations across three feeding experiments spanning ruminal pH 5.59 to 7.32, and reduced 1,694 exhaled SESI-MS features to a fourteen-feature consensus panel by cross-algorithm agreement, so that the signature was fixed before any pH model was fitted. The panel entered a generalised additive mixed model as non-linear smooth terms, added to a dietary and physiological baseline. Under leave-one-cow-out cross-validation it raised the concordance correlation coefficient for ruminal pH from 0.47 to 0.73, and it separated the lowest from the highest pH quartile with an area under the curve of 0.92 against 0.83 for the dietary covariates alone. Pooled mean bias fell from 4.58% to below 0.01%, a figure the fitted intercept largely guarantees and which I report for completeness rather than as evidence of accuracy. The remaining question was whether that predictive signature corresponds to fermentation chemistry or merely to a correlated artefact. Three lines of evidence indicate that it corresponds. Rumen-fluid metabolomics resolved the same pH extremes and correlated with the breath axis across matched samples. KEGG enrichment linked both compartments to shared central-carbon and amino-acid pathways. Genome-resolved metatranscriptomics localised the actively transcribed pathways to the dominant fermentative families, principally Lachnospiraceae. The fourteen-feature panel therefore discriminates low from high ruminal pH, and the features it uses map onto carbon and amino-acid pathways that the rumen-fluid metabolome independently implicates.
Collectively, this thesis demonstrates that SESI-HRMS-based exhalomics, combined with GreenFeed gas-flux measurements, constitutes a robust, animal-welfare-compatible platform for studying rumen H2 metabolism, characterising eVFA stoichiometry, and evaluating the efficacy of dietary CH4 mitigation strategies across diverse dairy cattle breeds. The rumen microbiome of the same animals was separated into its genomic capacity and its transcription, rather than the animals’ own genomes. Gene abundance and gene expression diverged in this cohort at the post-feeding sampling timepoint, so that functional inference required data on transcription rather than on DNA abundance alone. The exhaled signal traces to the fermentative families that actively transcribe the relevant pathways. On that basis a compact exhaled signature tracks ruminal pH under strict animal-level cross-validation, adding to rather than replacing the dietary and production covariates already recorded on farm. The measurement is integrative and within-animal rather than molecule-for-molecule. This boundary, established here rather than assumed, is what makes the prediction interpretable.