Challenges of implementing computer-aided diagnostic models for neuroimages in a clinical setting.

Advances in artificial intelligence have cultivated a strong interest in developing and validating the clinical utilities of computer-aided diagnostic models. Machine learning for diagnostic neuroimaging has often been applied to detect psychological and neurological disorders, typically on small-scale datasets or data collected in a research setting. With the collection and collation of an ever-growing number of public datasets that researchers can freely access, much work has been done in adapting machine learning models to classify these neuroimages by diseases such as Alzheimer's, ADHD, autism, bipolar disorder, and so on. These studies often come with the promise of being implemented clinically, but despite intense interest in this topic in the laboratory, limited progress has been made in clinical implementation. In this review, we analyze challenges specific to the clinical implementation of diagnostic AI models for neuroimaging data, looking at the differences between laboratory and clinical settings, the inherent limitations of diagnostic AI, and the different incentives and skill sets between research institutions, technology companies, and hospitals. These complexities need to be recognized in the translation of diagnostic AI for neuroimaging from the laboratory to the clinic.

Adversarial confound regression and uncertainty measurements to classify heterogeneous clinical MRI in Mass General Brigham.

In this work, we introduce a novel deep learning architecture, MUCRAN (Multi-Confound Regression Adversarial Network), to train a deep learning model on clinical brain MRI while regressing demographic and technical confounding factors. We trained MUCRAN using 17,076 clinical T1 Axial brain MRIs collected from Massachusetts General Hospital before 2019 and demonstrated that MUCRAN could successfully regress major confounding factors in the vast clinical dataset. We also applied a method for quantifying uncertainty across an ensemble of these models to automatically exclude out-of-distribution data in AD detection. By combining MUCRAN and the uncertainty quantification method, we showed consistent and significant increases in the AD detection accuracy for newly collected MGH data (post-2019; 84.6% with MUCRAN vs. 72.5% without MUCRAN) and for data from other hospitals (90.3% from Brigham and Women's Hospital and 81.0% from other hospitals). MUCRAN offers a generalizable approach for deep-learning-based disease detection in heterogenous clinical data.

Label-free detection and discrimination of respiratory pathogens based on electrochemical synthesis of biomaterials-mediated plasmonic composites and machine learning analysis.

Seasonal outbreaks of respiratory viral infections remain a global concern, with increasing morbidity and mortality rates recorded annually. Timely and false responses contribute to the widespread of respiratory pathogenic diseases owing to similar symptoms at an early stage and subclinical infection. The prevention of emerging novel viruses and variants is also a big challenge. Reliable point-of-care diagnostic assays for early infection diagnosis play a critical role in the response to threats of epidemics or pandemics. We developed a facile method for specifically identifying different viruses based on surface-enhanced Raman spectroscopy (SERS) with pathogen-mediated composite materials on Au nanodimple electrodes and machine learning (ML) analyses. Virus particles were trapped in three-dimensional plasmonic concave spaces of the electrode via electrokinetic preconcentration, and Au films were simultaneously electrodeposited, leading to the acquisition of intense and in-situ SERS signals from the Au-virus composites for ultrasensitive SERS detection. The method was useful for rapid detection analysis (<15 min), and the ML analysis for specific identification of eight virus species, including human influenza A viruses (i.e., H1N1 and H3N2 strains), human rhinovirus, and human coronavirus, was conducted. The highly accurate classification was achieved using the principal component analysis-support vector machine (98.9%) and convolutional neural network (93.5%) models. This ML-associated SERS technique demonstrated high feasibility for direct multiplex detection of different virus species for on-site applications.

Construction of a confounder-free clinical MRI dataset in the Mass General Brigham system for classification of Alzheimer's disease.

Deep learning has the potential to standardize and automate diagnostics for complex medical imaging data, but real-world clinical images are plagued by a high degree of heterogeneity and confounding factors that may introduce imbalances and biases to such processes. To address this, we developed and applied a data matching algorithm to 467,464 clinical brain magnetic resonance imaging (MRI) data from the Mass General Brigham (MGB) healthcare system for Alzheimer's disease (AD) classification. We identified 18 technical and demographic confounding factors that can be readily distinguished by MRI or have significant correlations with AD status and isolated a training set free from these confounds. We then applied an ensemble of 3D ResNet-50 deep learning models to classify brain MRIs between groups of AD, mild cognitive impairment (MCI), and healthy controls. From a confounder-free matched dataset of 287,367 MRI files, we achieved an area under the receiver operating characteristic (AUROC) of 0.82 in distinguishing healthy controls from patients with AD or MCI. We also showed that confounding factors in heterogeneous clinical data could lead to artificial gains in model performance for disease classification, which our data matching approach could correct. This approach could accelerate using deep learning models for clinical diagnosis and find broad applications in medical image analysis.

Deep learning for sex classification in resting-state and task functional brain networks from the UK Biobank.

Classification of whole-brain functional connectivity MRI data with convolutional neural networks (CNNs) has shown promise, but the complexity of these models impedes understanding of which aspects of brain activity contribute to classification. While visualization techniques have been developed to interpret CNNs, bias inherent in the method of encoding abstract input data, as well as the natural variance of deep learning models, detract from the accuracy of these techniques. We introduce a stochastic encoding method in an ensemble of CNNs to classify functional connectomes by sex. We applied our method to resting-state and task data from the UK BioBank, using two visualization techniques to measure the salience of three brain networks involved in task- and resting-states, and their interaction. To regress confounding factors such as head motion, age, and intracranial volume, we introduced a multivariate balancing algorithm to ensure equal distributions of such covariates between classes in our data. We achieved a final AUROC of 0.8459. We found that resting-state data classifies more accurately than task data, with the inner salience network playing the most important role of the three networks overall in classification of resting-state data and connections to the central executive network in task data.

Single-participant structural similarity matrices lead to greater accuracy in classification of participants than function in autism in MRI.

BACKGROUND: Autism has previously been characterized by both structural and functional differences in brain connectivity. However, while the literature on single-subject derivations of functional connectivity is extensively developed, similar methods of structural connectivity or similarity derivation from T1 MRI are less studied. METHODS: We introduce a technique of deriving symmetric similarity matrices from regional histograms of grey matter volumes estimated from T1-weighted MRIs. We then validated the technique by inputting the similarity matrices into a convolutional neural network (CNN) to classify between participants with autism and age-, motion-, and intracranial-volume-matched controls from six different databases (29,288 total connectomes, mean age = 30.72, range 0.42-78.00, including 1555 subjects with autism). We compared this method to similar classifications of the same participants using fMRI connectivity matrices as well as univariate estimates of grey matter volumes. We further applied graph-theoretical metrics on output class activation maps to identify areas of the matrices that the CNN preferentially used to make the classification, focusing particularly on hubs. LIMITATIONS: While this study used a large sample size, the majority of data was from a young age group; furthermore, to make a viable machine learning study, we treated autism, a highly heterogeneous condition, as a binary label. Thus, these results are not necessarily generalizable to all subtypes and age groups in autism. RESULTS: Our models gave AUROCs of 0.7298 (69.71% accuracy) when classifying by only structural similarity, 0.6964 (67.72% accuracy) when classifying by only functional connectivity, and 0.7037 (66.43% accuracy) when classifying by univariate grey matter volumes. Combining structural similarity and functional connectivity gave an AUROC of 0.7354 (69.40% accuracy). Analysis of classification performance across age revealed the greatest accuracy in adolescents, in which most data were present. Graph analysis of class activation maps revealed no distinguishable network patterns for functional inputs, but did reveal localized differences between groups in bilateral Heschl's gyrus and upper vermis for structural similarity. CONCLUSION: This study provides a simple means of feature extraction for inputting large numbers of structural MRIs into machine learning models. Our methods revealed a unique emphasis of the deep learning model on the structure of the bilateral Heschl's gyrus when characterizing autism.

Ensemble Deep Learning on Large, Mixed-Site fMRI Datasets in Autism and Other Tasks.

Deep learning models for MRI classification face two recurring problems: they are typically limited by low sample size, and are abstracted by their own complexity (the "black box problem"). In this paper, we train a convolutional neural network (CNN) with the largest multi-source, functional MRI (fMRI) connectomic dataset ever compiled, consisting of 43,858 datapoints. We apply this model to a cross-sectional comparison of autism spectrum disorder (ASD) versus typically developing (TD) controls that has proved difficult to characterize with inferential statistics. To contextualize these findings, we additionally perform classifications of gender and task versus rest. Employing class-balancing to build a training set, we trained [Formula: see text] modified CNNs in an ensemble model to classify fMRI connectivity matrices with overall AUROCs of 0.6774, 0.7680, and 0.9222 for ASD versus TD, gender, and task versus rest, respectively. Additionally, we aim to address the black box problem in this context using two visualization methods. First, class activation maps show which functional connections of the brain our models focus on when performing classification. Second, by analyzing maximal activations of the hidden layers, we were also able to explore how the model organizes a large and mixed-center dataset, finding that it dedicates specific areas of its hidden layers to processing different covariates of data (depending on the independent variable analyzed), and other areas to mix data from different sources. Our study finds that deep learning models that distinguish ASD from TD controls focus broadly on temporal and cerebellar connections, with a particularly high focus on the right caudate nucleus and paracentral sulcus.

Normative pathways in the functional connectome.

Functional connectivity is frequently derived from fMRI data to reduce a complex image of the brain to a graph, or "functional connectome". Often shortest-path algorithms are used to characterize and compare functional connectomes. Previous work on the identification and measurement of semi-metric (shortest circuitous) pathways in the functional connectome has discovered cross-sectional differences in major depressive disorder (MDD), autism spectrum disorder (ASD), and Alzheimer's disease. However, while measurements of shortest path length have been analyzed in functional connectomes, less work has been done to investigate the composition of the pathways themselves, or whether the edges composing pathways differ between individuals. Developments in this area would help us understand how pathways might be organized in mental disorders, and if a consistent pattern can be found. Furthermore, studies in structural brain connectivity and other real-world graphs suggest that shortest pathways may not be as important in functional connectivity studies as previously assumed. In light of this, we present a novel measurement of the consistency of pathways across functional connectomes, and an algorithm for improvement by selecting the most frequently occurring "normative pathways" from the k shortest paths, instead of just the shortest path. We also look at this algorithm's effect on various graph measurements, using randomized matrix simulations to support the efficacy of this method and demonstrate our algorithm on the resting-state fMRI (rs-fMRI) of a group of 34 adolescent control participants. Additionally, a comparison of normative pathways is made with a group of 82 age-matched participants, diagnosed with MDD, and in doing so we find the normative pathways that are most disrupted. Our results, which are carried out with estimates of connectivity derived from correlation, partial correlation, and normalized mutual information connectomes, suggest disruption to the default mode, affective, and ventral attention networks. Normative pathways, especially with partial correlation, make greater use of critical anatomical pathways through the striatum, cingulum, and the cerebellum. In summary, MDD is characterized by a disruption of normative pathways of the ventral attention network, increases in alternative pathways in the frontoparietal network in MDD, and a mixture of both in the default mode network. Additionally, within- and between-groups findings depend on the estimate of connectivity.

Light-Driven Processes Control Both Rhodopsin Maturation and Recycling in Mosquito Photoreceptors.

Many invertebrates carry out a daily cycle of shedding and rebuilding of the photoreceptor's photosensitive rhabdomeric membranes. The mosquito Aedes aegypti shows a robust response, losing nearly all Aaop1 rhodopsin from the rhabdomeric membranes during the shedding process at dawn. Here, we made use of Aaop1 antibodies capable of distinguishing newly synthesized, glycosylated rhodopsin from mature nonglycosylated rhodopsin to characterize the fate of Aaop1 during the shedding and rebuilding processes. The rhabdomeric rhodopsin is moved into large cytoplasmic vesicles at dawn and is subsequently degraded during the standard 12 h daytime period. The endocytosed rhodopsin is trafficked back to the photosensitive membranes if animals are shifted back to dark conditions during the morning hours. During the daytime period, small vesicles containing newly synthesized and glycosylated Aaop1 rhodopsin accumulate within the cytoplasm. At dusk, these vesicles are lost as the newly synthesized Aaop1 is converted to the nonglycosylated form and deposited in the rhabdomeres. We demonstrate that light acts though a novel signaling pathway to block rhodopsin maturation, thus inhibiting the deglycosylation and rhabdomeric targeting of newly synthesized Aaop1 rhodopsin. Therefore, light controls two cellular processes responsible for the daily renewal of rhodopsin: rhodopsin endocytosis at dawn and inhibition of rhodopsin maturation until dusk. SIGNIFICANCE STATEMENT: Organisms use multiple strategies to maximize visual capabilities in different light conditions. Many invertebrates show a daily cycle of shedding the photoreceptor's rhabdomeric membranes at dawn and rebuilding these during the following night. We show here that the Aedes aegypti mosquito possesses two distinct light-driven cellular signaling processes for modulating rhodopsin content during this cycle. One of these, endocytosis of rhabdomeric rhodopsin, has been described previously. The second, a light-activated cellular pathway acting to inhibit the anterograde movement of newly synthesized rhodopsin, is revealed here for the first time. The discovery of this cellular signaling pathway controlling a G-protein-coupled receptor is of broad interest due to the prominent role of this receptor family across all areas of neuroscience.

A framework for incorporating DTI Atlas Builder registration into Tract-Based Spatial Statistics and a simulated comparison to standard TBSS.

Tract-based spatial statistics (TBSS)6 is a software pipeline widely employed in comparative analysis of the white matter integrity from diffusion tensor imaging (DTI) datasets. In this study, we seek to evaluate the relationship between different methods of atlas registration for use with TBSS and different measurements of DTI (fractional anisotropy, FA, axial diffusivity, AD, radial diffusivity, RD, and medial diffusivity, MD). To do so, we have developed a novel tool that builds on existing diffusion atlas building software, integrating it into an adapted version of TBSS called DAB-TBSS (DTI Atlas Builder-Tract-Based Spatial Statistics) by using the advanced registration offered in DTI Atlas Builder7. To compare the effectiveness of these two versions of TBSS, we also propose a framework for simulating population differences for diffusion tensor imaging data, providing a more substantive means of empirically comparing DTI group analysis programs such as TBSS. In this study, we used 33 diffusion tensor imaging datasets and simulated group-wise changes in this data by increasing, in three different simulations, the principal eigenvalue (directly altering AD), the second and third eigenvalues (RD), and all three eigenvalues (MD) in the genu, the right uncinate fasciculus, and the left IFO. Additionally, we assessed the benefits of comparing the tensors directly using a functional analysis of diffusion tensor tract statistics (FADTTS10). Our results indicate comparable levels of FA-based detection between DAB-TBSS and TBSS, with standard TBSS registration reporting a higher rate of false positives in other measurements of DTI. Within the simulated changes investigated here, this study suggests that the use of DTI Atlas Builder's registration enhances TBSS group-based studies.

Expression and light-triggered movement of rhodopsins in the larval visual system of mosquitoes.

During the larval stages, the visual system of the mosquito Aedes aegypti contains five stemmata, often referred to as larval ocelli, positioned laterally on each side of the larval head. Here we show that stemmata contain two photoreceptor types, distinguished by the expression of different rhodopsins. The rhodopsin Aaop3 (GPROP3) is expressed in the majority of the larval photoreceptors. There are two small clusters of photoreceptors located within the satellite and central stemmata that express the rhodopsin Aaop7 (GPROP7) instead of Aaop3. Electroretinogram analysis of transgenic Aaop7 Drosophila indicates that Aaop3 and Aaop7, both classified as long-wavelength rhodopsins, possess similar but not identical spectral properties. Light triggers an extensive translocation of Aaop3 from the photosensitive rhabdoms to the cytoplasmic compartment, whereas light-driven translocation of Aaop7 is limited. The results suggest that these photoreceptor cell types play distinct roles in larval vision. An additional component of the larval visual system is the adult compound eye, which starts to develop at the anterior face of the larval stemmata during the 1st instar stage. The photoreceptors of the developing compound eye show rhodopsin expression during the 4th larval instar stage, consistent with indications from previous reports that the adult compound eye contributes to larval and pupal visual capabilities.

A database of circadian and diel rhythmic gene expression in the yellow fever mosquito Aedes aegypti.

BACKGROUND: The mosquito species Aedes aegypti is the primary vector of many arboviral diseases, including dengue and yellow fevers, that are responsible for a large worldwide health burden. The biological rhythms of mosquitoes regulate many of the physiological processes and behaviors that influence the transmission of these diseases. For insight into the molecular basis of biological rhythms, diel and circadian gene expression profiling has been carried out for many species. To bring these resources to Aedes aegypti researchers, we used microarray technology to carry out a genome wide assessment of gene expression during the 24 hour light/dark (LD) cycle and during constant darkness (DD). The purpose of this report is to describe the methods, the validation of the results, and the organization of this database resource. DESCRIPTION: The Aedes aegypti Circadian Database is a publicly accessible database that can be searched via a text-based query to visualize 44 hour temporal expression patterns of a given gene in Ae. aegypti heads under diel (observed under a 12 hour/12 hour LD cycle) and circadian (observed under DD) conditions. Profiles of gene expression under these conditions were assayed by Nimblegen 12-plex microarrays and rhythmicity was objectively assessed by the JTK_CYCLE algorithm. The output of the search is a graphical representation of the expression data along with computed period length, the time-of-day of gene expression peaks, and statistical determination for rhythmicity. CONCLUSION: Our results show that at least 7.9% of the gene set present in the Aedes aegypti head are rhythmic under LD conditions and 6.7% can be considered circadian, oscillating under constant dark conditions. We present these results in the Aedes aegypti Circadian Database through Bioclock, a public website hosted by the University of Notre Dame at http://www.nd.edu/~bioclock/. This website allows searchable browsing of this quantitative gene expression information. The visualization allows for gene-by-gene comparison of transcript expression under both diel and circadian conditions, and the results are presented graphically in a plot profile of gene expression. The Ae. aegypti Circadian Database provides a community resource for observing diel and circadian fluctuations in gene expression across the Ae. aegypti genome.

Rhodopsin management during the light-dark cycle of Anopheles gambiae mosquitoes.

The tropical disease vector mosquito Anopheles gambiae possesses 11 rhodopsin genes. Three of these, GPROP1, GPROP3, and GPROP4, encode rhodopsins with >99% sequence identity. We created antisera against these rhodopsins and used immunohistology to show that one or more of these rhodopsins are expressed in the major R1-6 photoreceptor class of the adult A.gambiae eye. Under dark conditions, rhodopsin accumulates within the light-sensitive rhabdomere of the photoreceptor. Light treatment, however, causes extensive movement of rhodopsin to the cytoplasmic compartment. Protein electrophoresis showed that the rhodopsin is present in two different forms. The larger form is an immature species that is deglycosylated during the posttranslational maturation process to generate the smaller, mature form. The immature form is maintained at a constant level regardless of lighting conditions. These results indicate that rhodopsin biosynthesis and movement into the rhabdomere occurs at a constant rate. In contrast, the mature form increases in abundance when animals are placed in dark conditions. Light-triggered internalization and protein degradation counteracts this rhodopsin increase and keeps rhabdomeric rhodopsin levels low in light conditions. The interplay of the constant maturation rate with light-triggered degradation causes rhodopsin to accumulate within the rhabdomere only in dark conditions. Thus, Anopheles photoreceptors possess a mechanism for adjusting light sensitivity through light-dependent control of rhodopsin levels and cellular location.

Role of semaphorin-1a in the developing visual system of the disease vector mosquito Aedes aegypti.

BACKGROUND: Despite the devastating impact of mosquito-borne illnesses on human health, very little is known about mosquito developmental biology, including development of the mosquito visual system. Mosquitoes possess functional adult compound eyes as larvae, a trait that makes them an interesting model in which to study comparative developmental genetics. Here, we functionally characterize visual system development in the dengue and yellow fever vector mosquito Aedes aegypti, in which we use chitosan/siRNA nanoparticles to target the axon guidance gene semaphorin-1a (sema1a). RESULTS: Immunohistochemical analyses revealed the progression of visual sensory neuron targeting that results in generation of the retinotopic map in the mosquito optic lobe. Loss of sema1a function led to optic lobe phenotypes, including defective targeting of visual sensory neurons and failed formation of the retinotopic map. These sema1a knockdown phenotypes correlated with behavioral defects in larval photoavoidance. CONCLUSIONS: The results of this investigation indicate that Sema1a is required for optic lobe development in A. aegypti and highlight the behavioral importance of a functioning visual system in preadult mosquitoes.

Rhodopsin coexpression in UV photoreceptors of Aedes aegypti and Anopheles gambiae mosquitoes.

Differential rhodopsin gene expression within specialized R7 photoreceptor cells divides the retinas of Aedes aegypti and Anopheles gambiae mosquitoes into distinct domains. The two species express the rhodopsin orthologs Aaop8 and Agop8, respectively, in a large subset of these R7 photoreceptors that function as ultraviolet receptors. We show here that a divergent subfamily of mosquito rhodopsins, Aaop10 and Agop10, is coexpressed in these R7 photoreceptors. The properties of the A. aegypti Aaop8 and Aaop10 rhodopsins were analyzed by creating transgenic Drosophila expressing these rhodopsins. Electroretinogram recordings, and spectral analysis of head extracts, obtained from the Aaop8 strain confirmed that Aaop8 is an ultraviolet-sensitive rhodopsin. Aaop10 was poorly expressed and capable of eliciting only small and slow light responses in Drosophila photoreceptors, and electroretinogram analysis suggested that it is a long-wavelength rhodopsin with a maximal sensitivity near 500 nm. Thus, coexpression of Aaop10 rhodopsin with Aaop8 rhodopsin has the potential to modify the spectral properties of mosquito ultraviolet receptors. Retention of Op10 rhodopsin family members in the genomes of Drosophila species suggests that this rhodopsin family may play a conserved role in insect vision.

Daily rhythms in antennal protein and olfactory sensitivity in the malaria mosquito Anopheles gambiae.

We recently characterized 24-hr daily rhythmic patterns of gene expression in Anopheles gambiae mosquitoes. These include numerous odorant binding proteins (OBPs), soluble odorant carrying proteins enriched in olfactory organs. Here we demonstrate that multiple rhythmically expressed genes including OBPs and takeout proteins, involved in regulating blood feeding behavior, have corresponding rhythmic protein levels as measured by quantitative proteomics. This includes AgamOBP1, previously shown as important to An. gambiae odorant sensing. Further, electrophysiological investigations demonstrate time-of-day specific differences in olfactory sensitivity of antennae to major host-derived odorants. The pre-dusk/dusk peaks in OBPs and takeout gene expression correspond with peak protein abundance at night, and in turn coincide with the time of increased olfactory sensitivity to odorants requiring OBPs and times of increased blood-feeding behavior. This suggests an important role for OBPs in modulating temporal changes in odorant sensitivity, enabling the olfactory system to coordinate with the circadian niche of An. gambiae.

Light-mediated control of rhodopsin movement in mosquito photoreceptors.

Multiple mechanisms contribute to a photoreceptor's ability to adapt to ambient light conditions. The mosquito Aedes aegypti expresses the long-wavelength rhodopsin Aaop1 in all R1-R6 photoreceptors and most R8 photoreceptors. These photoreceptors alter the cellular location of Aaop1 and reorganize their photosensitive rhabdomeric membranes on a daily basis. During daylight periods, Aaop1 is excluded from the light-sensitive rhabdomeres and localized to multivesicular bodies (MVBs) within the photoreceptor cytoplasm. In the dark, Aaop1 accumulates in the rhabdomeres and no Aaop1-containing MVBs are present in the cytoplasm. Manipulation of light treatments shows the cellular movement of Aaop1 in and out of the rhabdomere is directly controlled by light. In a separate process, the photoreceptors reduce Aaop1 protein content during a time period spanning from late afternoon into the first 2 h of the dark period. Aaop1 levels then gradually increase through the dark period and remain high following movement of Aaop1 to the cytoplasm at dawn. These results demonstrate that mosquito photoreceptors control rhodopsin availability during the daily light-dark cycle by novel mechanisms not discerned from analysis of traditional invertebrate models. These mechanisms will maximize a photoreceptor's light sensitivity range and therefore may be common in organisms active in low light conditions.

Characterization and differentiation of equine umbilical cord-derived matrix cells.

Stem cells are being evaluated in numerous human clinical trials and are commercially used in veterinary medicine to treat horses and dogs. Stem cell differentiation, homing to disease sites, growth and cytokine factor modulation, and low antigenicity contribute to their therapeutic success. Bone marrow and adipose tissue are the two most common sources of adult-derived stem cells in animals. We report on the existence of an alternative source of primitive, multipotent stem cells from the equine umbilical cord cellular matrix (Wharton's jelly). Equine umbilical cord matrix (EUCM) cells can be cultured, cryogenically preserved, and differentiated into osteo-, adipo-, chondrogenic, and neuronal cell lineages. These results identify a source of stem cells that can be non-invasively collected at birth and stored for future use in that horse or used as donor cells for treating unrelated horses.