Development and Validation of a Deep Learning Model to Quantify Glomerulosclerosis in Kidney Biopsy Specimens.

Importance: A chronic shortage of donor kidneys is compounded by a high discard rate, and this rate is directly associated with biopsy specimen evaluation, which shows poor reproducibility among pathologists. A deep learning algorithm for measuring percent global glomerulosclerosis (an important predictor of outcome) on images of kidney biopsy specimens could enable pathologists to more reproducibly and accurately quantify percent global glomerulosclerosis, potentially saving organs that would have been discarded. Objective: To compare the performances of pathologists with a deep learning model on quantification of percent global glomerulosclerosis in whole-slide images of donor kidney biopsy specimens, and to determine the potential benefit of a deep learning model on organ discard rates. Design, Setting, and Participants: This prognostic study used whole-slide images acquired from 98 hematoxylin-eosin-stained frozen and 51 permanent donor biopsy specimen sections retrieved from 83 kidneys. Serial annotation by 3 board-certified pathologists served as ground truth for model training and for evaluation. Images of kidney biopsy specimens were obtained from the Washington University database (retrieved between June 2015 and June 2017). Cases were selected randomly from a database of more than 1000 cases to include biopsy specimens representing an equitable distribution within 0% to 5%, 6% to 10%, 11% to 15%, 16% to 20%, and more than 20% global glomerulosclerosis. Main Outcomes and Measures: Correlation coefficient (r) and root-mean-square error (RMSE) with respect to annotations were computed for cross-validated model predictions and on-call pathologists' estimates of percent global glomerulosclerosis when using individual and pooled slide results. Data were analyzed from March 2018 to August 2020. Results: The cross-validated model results of section images retrieved from 83 donor kidneys showed higher correlation with annotations (r = 0.916; 95% CI, 0.886-0.939) than on-call pathologists (r = 0.884; 95% CI, 0.825-0.923) that was enhanced when pooling glomeruli counts from multiple levels (r = 0.933; 95% CI, 0.898-0.956). Model prediction error for single levels (RMSE, 5.631; 95% CI, 4.735-6.517) was 14% lower than on-call pathologists (RMSE, 6.523; 95% CI, 5.191-7.783), improving to 22% with multiple levels (RMSE, 5.094; 95% CI, 3.972-6.301). The model decreased the likelihood of unnecessary organ discard by 37% compared with pathologists. Conclusions and Relevance: The findings of this prognostic study suggest that this deep learning model provided a scalable and robust method to quantify percent global glomerulosclerosis in whole-slide images of donor kidneys. The model performance improved by analyzing multiple levels of a section, surpassing the capacity of pathologists in the time-sensitive setting of examining donor biopsy specimens. The results indicate the potential of a deep learning model to prevent erroneous donor organ discard.

Deep learning quantification of percent steatosis in donor liver biopsy frozen sections.

BACKGROUND: Pathologist evaluation of donor liver biopsies provides information for accepting or discarding potential donor livers. Due to the urgent nature of the decision process, this is regularly performed using frozen sectioning at the time of biopsy. The percent steatosis in a donor liver biopsy correlates with transplant outcome, however there is significant inter- and intra-observer variability in quantifying steatosis, compounded by frozen section artifact. We hypothesized that a deep learning model could identify and quantify steatosis in donor liver biopsies. METHODS: We developed a deep learning convolutional neural network that generates a steatosis probability map from an input whole slide image (WSI) of a hematoxylin and eosin-stained frozen section, and subsequently calculates the percent steatosis. Ninety-six WSI of frozen donor liver sections from our transplant pathology service were annotated for steatosis and used to train (n = 30 WSI) and test (n = 66 WSI) the deep learning model. FINDINGS: The model had good correlation and agreement with the annotation in both the training set (r of 0.88, intraclass correlation coefficient [ICC] of 0.88) and novel input test sets (r = 0.85 and ICC=0.85). These measurements were superior to the estimates of the on-service pathologist at the time of initial evaluation (r = 0.52 and ICC=0.52 for the training set, and r = 0.74 and ICC=0.72 for the test set). INTERPRETATION: Use of this deep learning algorithm could be incorporated into routine pathology workflows for fast, accurate, and reproducible donor liver evaluation. FUNDING: Mid-America Transplant Society.

Deep Learning Global Glomerulosclerosis in Transplant Kidney Frozen Sections.

Transplantable kidneys are in very limited supply. Accurate viability assessment prior to transplantation could minimize organ discard. Rapid and accurate evaluation of intra-operative donor kidney biopsies is essential for determining which kidneys are eligible for transplantation. The criterion for accepting or rejecting donor kidneys relies heavily on pathologist determination of the percent of glomeruli (determined from a frozen section) that are normal and sclerotic. This percentage is a critical measurement that correlates with transplant outcome. Inter- and intra-observer variability in donor biopsy evaluation is, however, significant. An automated method for determination of percent global glomerulosclerosis could prove useful in decreasing evaluation variability, increasing throughput, and easing the burden on pathologists. Here, we describe the development of a deep learning model that identifies and classifies non-sclerosed and sclerosed glomeruli in whole-slide images of donor kidney frozen section biopsies. This model extends a convolutional neural network (CNN) pre-trained on a large database of digital images. The extended model, when trained on just 48 whole slide images, exhibits slide-level evaluation performance on par with expert renal pathologists. Encouragingly, the model's performance is robust to slide preparation artifacts associated with frozen section preparation. The model substantially outperforms a model trained on image patches of isolated glomeruli, in terms of both accuracy and speed. The methodology overcomes the technical challenge of applying a pretrained CNN bottleneck model to whole-slide image classification. The traditional patch-based approach, while exhibiting deceptively good performance classifying isolated patches, does not translate successfully to whole-slide image segmentation in this setting. As the first model reported that identifies and classifies normal and sclerotic glomeruli in frozen kidney sections, and thus the first model reported in the literature relevant to kidney transplantation, it may become an essential part of donor kidney biopsy evaluation in the clinical setting.

Antithrombin nanoparticles inhibit stent thrombosis in ex vivo static and flow models.

OBJECTIVE: Despite significant advances in intravascular stent technology, safe prevention of stent thrombosis over prolonged periods after initial deployment persists as a medical need to decrease device failure. The objective of this project was to assess the potential of perfluorocarbon nanoparticles (NP) conjugated with the direct thrombin inhibitor D-phenylalanyl-L-prolyl-L-arginyl chloromethylketone (PPACK-NP) to inhibit stent thrombosis. METHODS: In a static model of stent thrombosis, 3 × 3-mm pieces of stainless steel coronary stents were cut and adsorbed with thrombin to create a procoagulant surface that would facilitate thrombus development. After treatment with PPACK-NP or control NP, stents were exposed to platelet-poor plasma (PPP) or platelet-rich plasma (PRP) for set time points up to 60 minutes. Measurements of final clot weight in grams were used for assessing the effect of NP treatment on limiting thrombosis. Additionally, groups of stents were exposed to flowing plasma containing various treatments (saline, free PPACK, control NP, and PPACK-NP) and generated thrombi were stained and imaged to investigate the treatment effects of PPACK-NP under flow conditions. RESULTS: The static model of stent thrombosis used in this study indicated a significant reduction in thrombus deposition with PPACK-NP treatment (0.00067 ± 0.00026 g; n = 3) compared with control NP (0.0098 ± 0.0015 g; n = 3; P = .026) in PPP. Exposure to PRP demonstrated similar effects with PPACK-NP treatment (0.00033 ± 0.00012 g; n = 3) vs control NP treatment (0.0045 ± 0.00012 g; n = 3; P = .000017). In additional studies, stents were exposed to both PRP pretreated with vorapaxar and PPACK-NP, which illustrated adjunctive benefit to oral platelet inhibitors for prevention of stent thrombosis. Additionally, an in vitro model of stent thrombosis under flow conditions established that PPACK-NP treatment inhibited thrombus deposition on stents significantly. CONCLUSIONS: This study demonstrates that antithrombin perfluorocarbon NPs exert marked focal antithrombin activity to prevent intravascular stent thrombosis and occlusion.

Delivery of a Protease-Activated Cytolytic Peptide Prodrug by Perfluorocarbon Nanoparticles.

Melittin is a cytolytic peptide derived from bee venom that inserts into lipid membranes and oligomerizes to form membrane pores. Although this peptide is an attractive candidate for treatment of cancers and infectious processes, its nonspecific cytotoxicity and hemolytic activity have limited its therapeutic applications. Several groups have reported the development of cytolytic peptide prodrugs that only exhibit cytotoxicity following activation by site-specific proteases. However, systemic administration of these constructs has proven difficult because of their poor pharmacokinetic properties. Here, we present a platform for the design of protease-activated melittin derivatives that may be used in conjunction with a perfluorocarbon nanoparticle delivery system. Although native melittin was substantially hemolytic (HD50: 1.9 µM) and cytotoxic (IC50: 2.4 µM), the prodrug exhibited 2 orders of magnitude less hemolytic activity (HD50: > 100 µM) and cytotoxicity (IC50: > 100 µM). Incubation with matrix metalloproteinase-9 (MMP-9) led to cleavage of the prodrug at the expected site and restoration of hemolytic activity (HD50: 3.4 µM) and cytotoxicity (IC50: 8.1 µM). Incubation of the prodrug with perfluorocarbon nanoparticles led to stable loading of 10,250 peptides per nanoparticle. Nanoparticle-bound prodrug was also cleaved and activated by MMP-9, albeit at a fourfold slower rate. Intravenous administration of prodrug-loaded nanoparticles in a mouse model of melanoma significantly decreased tumor growth rate (p = 0.01). Because MMPs and other proteases play a key role in cancer invasion and metastasis, this platform holds promise for the development of personalized cancer therapies directed toward a patient's individual protease expression profile.

Programmable nanoparticle functionalization for in vivo targeting.

The emerging demand for programmable functionalization of existing base nanocarriers necessitates development of an efficient approach for cargo loading that avoids nanoparticle redesign for each individual application. Herein, we demonstrate in vivo a postformulation strategy for lipidic nanocarrier functionalization with the use of a linker peptide, which rapidly and stably integrates cargos into lipidic membranes of nanocarriers after simple mixing through a self-assembling process. We exemplified this strategy by generating a VCAM-1-targeted perfluorocarbon nanoparticle for in vivo targeting in atherosclerosis (ApoE-deficient) and breast cancer (STAT-1-deficient) models. In the atherosclerotic model, a 4.1-fold augmentation in binding to affected aortas was observed for targeted vs. nontargeted nanoparticles (P<0.0298). Likewise, in the breast cancer model, a 4.9-fold increase in the nanoparticle signal from tumor vasculature was observed for targeted vs. nontargeted nanoparticles (P<0.0216). In each case, the nanoparticle was registered with fluorine ((19)F) magnetic resonance spectroscopy of the nanoparticle perfluorocarbon core, yielding a quantitative estimate of the number of tissue-bound nanoparticles. Because other common nanocarriers with lipid coatings (e.g., liposomes, micelles, etc.) can employ this strategy, this peptide linker postformulation approach is applicable to more than half of the available nanosystems currently in clinical trials or clinical uses.

Postformulation peptide drug loading of nanostructures.

Cytolytic peptides have commanded attention for their anticancer potential because the membrane-disrupting function that produces cell death is less likely to be overcome by resistant mutations. Congruently, peptides that are involved in molecular recognition and biological activities become attractive therapeutic candidates because of their high specificity, better affinity, reduced immunogenicity, and reduced off target toxicity. However, problems of inadequate delivery, rapid deactivation in vivo, and poor bioavailability have limited clinical application. Therefore, peptide drug development for clinical use requires an appropriate combination of an effective therapeutic peptide and a robust delivery methodology. In this chapter, we describe methods for the postformulation insertion of peptide drugs into lipidic nanostructures, the physical characterization of peptide-nanostructure complexes, and the evaluation of their therapeutic effectiveness both in vitro and in vivo.

Application of a real-time, calculable limiting form of the Renyi entropy for molecular imaging of tumors.

Previously, we reported new methods for ultrasound signal characterization using entropy, H(f); a generalized entropy, the Renyi entropy, I(f)(r); and a limiting form of Renyi entropy suitable for real-time calculation, I(f),(infinity). All of these quantities demonstrated significantly more sensitivity to subtle changes in scattering architecture than energy-based methods in certain settings. In this study, the real-time calculable limit of the Renyi entropy, I(f),(infinity), is applied for the imaging of angiogenic murine neovasculature in a breast cancer xenograft using a targeted contrast agent. It is shown that this approach may be used to reliably detect the accumulation of targeted nanoparticles at five minutes post-injection in this in vivo model.

Lipid membrane editing with peptide cargo linkers in cells and synthetic nanostructures.

Current strategies for deploying synthetic nanocarriers involve the creation of agents that incorporate targeting ligands, imaging agents, and/or therapeutic drugs into particles as an integral part of the formulation process. Here we report the development of an amphipathic peptide linker that enables postformulation editing of payloads without the need for reformulation to achieve multiplexing capability for lipidic nanocarriers. To exemplify the flexibility of this peptide linker strategy, 3 applications were demonstrated: converting nontargeted nanoparticles into targeting vehicles; adding cargo to preformulated targeted nanoparticles for in vivo site-specific delivery; and labeling living cells for in vivo tracking. This strategy is expected to enhance the clinical application of molecular imaging and/or targeted therapeutic agents by offering extended flexibility for multiplexing targeting ligands and/or drug payloads that can be selected after base nanocarrier formulation.

Molecularly targeted nanocarriers deliver the cytolytic peptide melittin specifically to tumor cells in mice, reducing tumor growth.

The in vivo application of cytolytic peptides for cancer therapeutics is hampered by toxicity, nonspecificity, and degradation. We previously developed a specific strategy to synthesize a nanoscale delivery vehicle for cytolytic peptides by incorporating the nonspecific amphipathic cytolytic peptide melittin into the outer lipid monolayer of a perfluorocarbon nanoparticle. Here, we have demonstrated that the favorable pharmacokinetics of this nanocarrier allows accumulation of melittin in murine tumors in vivo and a dramatic reduction in tumor growth without any apparent signs of toxicity. Furthermore, direct assays demonstrated that molecularly targeted nanocarriers selectively delivered melittin to multiple tumor targets, including endothelial and cancer cells, through a hemifusion mechanism. In cells, this hemifusion and transfer process did not disrupt the surface membrane but did trigger apoptosis and in animals caused regression of precancerous dysplastic lesions. Collectively, these data suggest that the ability to restrain the wide-spectrum lytic potential of a potent cytolytic peptide in a nanovehicle, combined with the flexibility of passive or active molecular targeting, represents an innovative molecular design for chemotherapy with broad-spectrum cytolytic peptides for the treatment of cancer at multiple stages.

Sensitive ultrasonic delineation of steroid treatment in living dystrophic mice with energy-based and entropy-based radio frequency signal processing.

Duchenne muscular dystrophy is a severe wasting disease, involving replacement of necrotic muscle tissue by fibrous material and fatty infiltrates. One primary animal model of this human disease is the X chromosome-linked mdx strain of mice. The goals of the present work were to validate and quantify the capability of both energy and entropy metrics of radio-frequency ultrasonic backscatter to differentiate among normal, dystrophic, and steroid-treated skeletal muscle in the mdx model. Thirteen 12-month-old mice were blocked into three groups: 4 treated mdx-dystrophic that received daily subcutaneous steroid (prednisolone) treatment for 14 days, 4 positive-control mdx-dystrophic that received saline injections for 14 days, and 5 negative-control animals. Biceps muscle of each animal was imaged in vivo using a 40-MHz center frequency transducer in conjunction with a Vevo-660 ultrasound system. Radio-frequency data were acquired (1 GHz, 8 bits) corresponding to a sequence of transverse images, advancing the transducer from "shoulder" to "elbow" in 100-micron steps. Data were processed to generate both "integrated backscatter" (log energy), and "entropy" (information theoretic receiver, H(f)) representations. Analyses of the integrated-backscatter values delineated both treated-and untreated-mdx biceps from normal controls (p < 0.01). Complementary analyses of the entropy images differentiated the steroid-treated and positive-control mdx groups (p < 0.01). To our knowledge, this study represents the first reported use of quantitative ultrasonic characterization of skeletal muscle in mdx mice. Successful differentiation among dystrophic, steroid-treated, and normal tissues suggests the potential for local noninvasive monitoring of disease severity and therapeutic effects.

Molecular imaging with targeted perfluorocarbon nanoparticles: quantification of the concentration dependence of contrast enhancement for binding to sparse cellular epitopes.

Targeted, liquid perfluorocarbon nanoparticles are effective agents for acoustic contrast enhancement of abundant cellular epitopes (e.g., fibrin in thrombi) and for lower prevalence binding sites, such as integrins associated with tumor neovasculature. In this study, we sought to delineate the quantitative relationship between the extent of contrast enhancement of targeted surfaces and the density (and concentration) of bound perfluorocarbon (PFC) nanoparticles. Two dramatically different substrates were utilized for targeting. In one set of experiments, the surfaces of smooth, flat, avidin-coated agar disks were exposed to biotinylated nanoparticles to yield a thin layer of targeted contrast. For the second set of measurements, we targeted PFC nanoparticles applied in thicker layers to cultured smooth muscle cells expressing the transmembrane glycoprotein "tissue factor" at the cell surface. An acoustic microscope was used to characterize reflectivity for all samples as a function of bound PFC (determined via gas chromatography). We utilized a formulation of low-scattering nanoparticles having oil-based cores to compete against high-scattering PFC nanoparticles for binding, to elucidate the dependence of contrast enhancement on PFC concentration. The relationship between reflectivity enhancement and bound PFC content varied in a curvilinear fashion and exhibited an apparent asymptote (approximately 16 dB and 9 dB enhancement for agar and cell samples, respectively) at the maximum concentrations (approximately 150 microg and approximately 1000 microg PFOB for agar and cell samples, respectively). Samples targeted with only oil-based nanoparticles exhibited mean backscatter values that were nearly identical to untreated samples (<1 dB difference), confirming the oil particles' low-scattering behavior. The results of this study indicate that substantial contrast enhancement with liquid perfluorocarbon nanoparticles can be realized even in cases of partial surface coverage (as might be encountered when targeting sparsely populated epitopes) or when targeting surfaces with locally irregular topography. Furthermore, it may be possible to assess the quantity of bound cellular epitopes through acoustic means.

Characterization of digital waveforms using thermodynamic analogs: detection of contrast-targeted tissue in vivo.

We describe characterization of backscatter from tumor tissue targeted with a nanoparticle-based ultrasound contrast agent in vivo using analogs of thermodynamic quantities. We apply these waveform characteristics to detection of tumor neovasculature in tumors implanted in athymic nude mice, which were imaged using a research ultrasound scanner over a 2-hour period after injection of alpha upsilon beta3-targeted perfluorocarbon nanoparticles. Images were constructed from backscattered ultrasound using two different approaches: fundamental B-mode imaging and a signal receiver based on a thermodynamic analog (H(C)). The study shows that the thermodynamic analog is capable of detecting differences in backscattered signals that are not apparent with the B-mode approach.

Acoustic activation of targeted liquid perfluorocarbon nanoparticles does not compromise endothelial integrity.

Perfluorocarbon nanoparticles consisting essentially of liquid perfluoro-octyl bromide (PFOB) core surrounded by a lipid monolayer can serve as highly specific site-targeted contrast and therapeutic agents after binding to cellular biomarkers. Based on previous findings that ultrasound applied at 2 MHz and 1.9 mechanical index (MI) for a 5-min duration dramatically enhances the cellular interaction of targeted PFOB nanoparticles with melanoma cells in vitro without inducing apoptosis or other harmful effects to cells that are targeted, we sought to define mechanisms of interaction and the safety profile of ultrasound used in conjunction with liquid perfluorocarbon nanoparticles for targeted drug delivery, as compared with conventional microbubble ultrasound contrast agents under identical insonification conditions. Cell-culture inserts were used to grow a confluent monolayer of human umbilical vein endothelial cells. Definity in conjunction with continuous wave ultrasound (2.25 MHz for 1 and 5 min) increased the permeability of monolayer by four to six times above the normal, decreased transendothelial electrical resistance (a sign of reduced membrane integrity), and decreased cell viability by approximately 50%. Histological evaluation demonstrated extensive disruptions of cell monolayers. Nanoparticles (both nontargeted and targeted) elicited no changes in these different measures under similar insonification conditions and did not disrupt cell monolayers. We hypothesize that ultrasound facilitates drug transport from the perfluorocarbon nanoparticles not by cavitation-induced effects on cell membrane but rather by direct interaction with the nanoparticles that stimulate lipid exchange and drug delivery.

Sonic activation of molecularly-targeted nanoparticles accelerates transmembrane lipid delivery to cancer cells through contact-mediated mechanisms: implications for enhanced local drug delivery.

Liquid perfluorocarbon nanoparticles serve as sensitive and specific targeted contrast and drug delivery vehicles by binding to specific cell surface markers. We hypothesized that application of acoustic energy at diagnostic power levels could promote nanoparticle-associated drug delivery by stimulating increased interaction between the nanoparticle's lipid layer and the targeted cell's plasma membrane. Ultrasound (mechanical index = 1.9) applied with a conventional ultrasound imaging system to nanoparticles targeted to alpha(v)beta3-integrins on C32 melanoma cancer cells in vitro produced no untoward effects. Within 5 min, lipid delivery from nanoparticles into cell cytoplasm was dramatically augmented. We also demonstrate the operation of a potential physical mechanism for this effect, the acoustic radiation force on the nanoparticles, which may contribute to the enhanced lipid delivery. Accordingly, we propose that local delivery of lipophilic substances (e.g., drugs) from targeted nanoparticles directly into cell cytoplasm can be augmented rapidly and safely with conventional ultrasound imaging devices through nondestructive mechanisms.