Memory Reduction Research Articles

Quantization of Deep Neural Networks for Accurate Edge Computing

Deep neural networks have demonstrated their great potential in recent years, exceeding the performance of human experts in a wide range of applications. Due to their large sizes, however, compression techniques such as weight quantization and pruning are usually applied before they can be accommodated on the edge. It is generally believed that quantization leads to performance degradation, and plenty of existing works have explored quantization strategies aiming at minimum accuracy loss. In this paper, we argue that quantization, which essentially imposes regularization on weight representations, can sometimes help to improve accuracy. We conduct comprehensive experiments on three widely used applications: fully connected network for biomedical image segmentation, convolutional neural network for image classification on ImageNet, and recurrent neural network for automatic speech recognition, and experimental results show that quantization can improve the accuracy by 1%, 1.95%, 4.23% on the three applications respectively with 3.5x-6.4x memory reduction.

COMPLEXITY AND MEMORY-BASED COMPARISON OF THE BRAIN ACTIVITY BETWEEN ADHD AND HEALTHY SUBJECTS WHILE PLAYING A SERIOUS GAME

Attention deficit hyperactivity disorder (ADHD) is a mental health disorder that is very common among children and may last into their adulthood. It is known that ADHD affects the attention of patients due to problems with short-term memory. Therefore, analysis of attention and memory of these patients should come into consideration. In this study, the complexity and memory of Electroencephalogram (EEG) signals are analyzed to investigate the reduction in attention and memory of patients with ADHD compared to normal subjects while playing a serious game. To achieve this, the fractal dimension and sample entropy of EEG signals are analyzed to evaluate the alterations in the complexity of EEG signals. Moreover, the Hurst exponent of EEG signals for ADHD and non-ADHD subjects is calculated to discuss the memory of EEG signals. The results showed a smaller fractal dimension and sample entropy of EEG signals for patients with ADHD that reflects their lower attention. Besides, the Hurst exponent of EEG signals for these patients reflects their lower memory than normal subjects. Therefore, it can be concluded that the reductions of attention and memory in ADHD subjects are mapped on the reduction of complexity and memory of their EEG signals.

Efficient One-Shot Technique for Adjoint-Based Unsteady Optimization

A computationally efficient one-shot approach with a low memory footprint is presented for unsteady optimization. The proposed technique is based on a novel and unique approach that combines local-in-time and fixed-point iteration methods to advance the unconverged primal and adjoint solutions forward and backward in time to evaluate the sensitivity of the globally time-integrated objective function. This is in some ways similar to the piggyback iterations in which primal and adjoint solutions are evaluated at a frozen design. During each cycle, the primal, adjoint, and design update problems are solved to advance the optimization problem. This new coupled approach is shown to provide significant savings in the memory footprint while reducing the computational cost of primal and adjoint evaluations per design cycle. The method is first applied to an inverse design problem for the unsteady lid-driven cavity. Following this, vortex suppression and mean drag reduction for a circular cylinder in crossflow is considered. Both of these objectives are achieved by optimizing the rotational speeds for steady or periodically oscillating excitations. For all cases presented in this work, the proposed technique is shown to provide significant reductions in memory as well as computational time. It is also shown that the unsteady optimization problem converges to the same optimal solution obtained using a conventional approach.

Impact of age and gender on lymphocyte subset counts in patients with COVID-19.

In symptomatic patients with acute Coronavirus disease 2019 (COVID‐19), lymphocytopenia is one of the most prominent laboratory findings. However, to date age and gender have not been considered in assessment of COVID‐19‐related cell count alterations. In this study, the impact of COVID‐19 as well as age and gender on a large variety of lymphocyte subsets was analyzed in 33 COVID‐19 patients and compared with cell counts in 50 healthy humans. We confirm that cell counts of total lymphocytes, B, NK, cytotoxic and helper T cells are reduced in patients with severe COVID‐19, and this tendency was observed in patients with moderate COVID‐19. Decreased cell counts were also found in all subsets of these cell types, except for CD4+ and CD8+ effector memory RA+ (EMRA) and terminal effector CD8+ cells. In multivariate analysis however, we show that in addition to COVID‐19, there is an age‐dependent reduction of total, central memory (CM), and early CD8+ cell subsets, as well as naïve, CM, and regulatory CD4+ cell subsets. Remarkably, reduced naïve CD8+ cell counts could be attributed to age alone, and not to COVID‐19. By contrast, decreases in other subsets could be largely attributed to COVID‐19, and only partly to age. In addition to COVID‐19, male gender was a major factor influencing lower counts of CD3+ and CD4+ lymphocyte numbers. Our study confirms that cell counts of lymphocytes and their subsets are reduced in patients with COVID‐19, but that age and gender must be considered when interpreting the altered cell counts.

Marijuana and β-estradiol interactions on spatial learning and memory in young female rats: Lack of role of the G protein-coupled estrogen receptor (GPR30)

It has been shown that 17β-estradiol (E2) hormone is an essential biological factor for increasing the sensitivity of women to drug abuse. Recent studies have shown a potential overlap between the molecular pathways of cannabinoids and ovarian hormones. The current study evaluated the interference between the marijuana and E2 effect on spatial learning and memory and the role of the G protein-coupled estrogen receptor (GPR30) in young female rats. The animals were separated into two main groups: intact-ovary and ovariectomized (OVX) rats. The latter group received intraperitoneal injections of E2, G-1 (GPR30 agonist), G-15 (GPR30 antagonist), marijuana, and different combinations of these substances for 28 days. Spatial learning and memory were evaluated by the Morris water maze (MWM) test. We also assessed the BDNF (brain-derived neurotrophic factor) concentration and the hippocampal level of GPR30. The results showed a significant reduction of spatial learning and memory in OVX rats compared to intact-ovary rats, which were restored by E2 replacement. Moreover, treatment with G-1 mimicked E2 effects on spatial learning and memory. Marijuana impaired spatial learning and memory in intact-ovary rats, while improved in OVX rats. We also found that treatment with M + E2 induced significant impairment in spatial learning and memory; however, treatment with M + G1 and M + G15 + E2 showed no significant difference. No significant differences in BDNF expression were observed in experimental groups. These results suggest that marijuana and E2 interact in their effect on spatial learning and memory in young female rats, but GPR30 seems to play no role in this interaction.

AdaMM: Adaptive Object Movement and Motion Tracking in Hierarchical Edge Computing System.

This paper presents a novel adaptive object movement and motion tracking (AdaMM) framework in a hierarchical edge computing system for achieving GPU memory footprint reduction of deep learning (DL)-based video surveillance services. DL-based object movement and motion tracking requires a significant amount of resources, such as (1) GPU processing power for the inference phase and (2) GPU memory for model loading. Despite the absence of an object in the video, if the DL model is loaded, the GPU memory must be kept allocated for the loaded model. Moreover, in several cases, video surveillance tries to capture events that rarely occur (e.g., abnormal object behaviors); therefore, such standby GPU memory might be easily wasted. To alleviate this problem, the proposed AdaMM framework categorizes the tasks used for the object movement and motion tracking procedure in an increasing order of the required processing and memory resources as task (1) frame difference calculation, task (2) object detection, and task (3) object motion and movement tracking. The proposed framework aims to adaptively release the unnecessary standby object motion and movement tracking model to save GPU memory by utilizing light tasks, such as frame difference calculation and object detection in a hierarchical manner. Consequently, object movement and motion tracking are adaptively triggered if the object is detected within the specified threshold time; otherwise, the GPU memory for the model of task (3) can be released. Moreover, object detection is also adaptively performed if the frame difference over time is greater than the specified threshold. We implemented the proposed AdaMM framework using commercial edge devices by considering a three-tier system, such as the 1st edge node for both tasks (1) and (2), the 2nd edge node for task (3), and the cloud for sending a push alarm. A measurement-based experiment reveals that the proposed framework achieves a maximum GPU memory reduction of 76.8% compared to the baseline system, while requiring a 2680 ms delay for loading the model for object movement and motion tracking.

Androgen deprivation therapy and cognitive decline-associations with brain connectomes, endocrine status, and risk genotypes.

Evidence suggests that prostate cancer (PC) patients undergoing androgen deprivation therapy (ADT) are at risk for cognitive decline (CD), but the underlying mechanisms are less clear. In the present study, changes in cognitive performance and structural brain connectomes in PC patients undergoing ADT were assessed, and associations of cognitive changes with endocrine status and risk genotypes were explored. Thirty-seven PC patients underwent cognitive assessment, structural MRI, and provided blood samples prior to ADT and after 6 months of treatment. Twenty-seven age- and education-matched healthy controls (HCs) underwent the same assessments. CD was determined using a standardized regression-based approach and defined as z-scores ≤ -1.64. Changes in brain connectomes were evaluated using graph theory. Associations of CD with testosterone levels and genotypes (APOE, COMT, BDNF) were explored. Compared with HCs, PC patients demonstrated reduced testosterone levels (p < 0.01) and higher rates of decline for 13 out of 15 cognitive outcomes, with three outcomes related to two cognitive domains, i.e., verbal memory and visuospatial learning and memory, reaching statistical significance (p ≤ 0.01-0.04). Testosterone level changes did not predict CD. COMT Met homozygote PC patients evidenced larger reductions in visuospatial memory compared with Val carriers (p = 0.02). No between-group differences were observed in brain connectomes across time, and no effects were found of APOE and BDNF. Our results indicate that PC patients undergoing ADT may evidence CD, and that COMT Met homozygotes may be at increased risk of CD. The results did not reveal changes in brain connectomes or testosterone levels as underlying mechanisms. More research evaluating the role of ADT-related disruption of the dynamics of the hypothalamic-pituitary-gonadal axis is needed.

Sparse grid-based adaptive noise reduction strategy for particle-in-cell schemes

We propose a sparse grid-based adaptive noise reduction strategy for electrostatic particle-in-cell (PIC) simulations. By projecting the charge density onto sparse grids we reduce the high-frequency particle noise. Thus, we exploit the ability of sparse grids to act as a multidimensional low-pass filter in our approach. Thanks to the truncated combination technique [1–3], we can reduce the larger grid-based error of the standard sparse grid approach for non-aligned and non-smooth functions. The truncated approach also provides a natural framework for minimizing the sum of grid-based and particle-based errors in the charge density. We show that our approach is, in fact, a filtering perspective for the noise reduction obtained with the sparse PIC schemes first introduced in [4]. This enables us to propose a heuristic based on the formal error analysis in [4] for selecting the optimal truncation parameter that minimizes the total error in charge density at each time step. Hence, unlike the physical and Fourier domain filters typically used in PIC codes for noise reduction, our approach automatically adapts to the mesh size, number of particles per cell, smoothness of the density profile and the initial sampling technique. It can also be easily integrated into high performance large-scale PIC code bases, because we only use sparse grids for filtering the charge density. All other operations remain on the regular grid, as in typical PIC codes. We demonstrate the efficiency and performance of our approach with two test cases: the diocotron instability in two dimensions and the three-dimensional electron dynamics in a Penning trap. Our run-time performance studies indicate that our approach can provide significant speedup and memory reduction to PIC simulations for achieving comparable accuracy in the charge density.

Approximate Packet Classifiers With Controlled Accuracy

Performing exact computations can require significant resources. Approximate computing allows to alleviate resource constraints, sacrificing the accuracy of results. In this work, we consider a generalization of the classical packet classification problem. Our major contribution is to introduce representations of approximate packet classifiers with controlled accuracy and optimization techniques to reduce classifier sizes exploiting this new level of flexibility. In this work, we propose methods constructing efficient approximate representations for both LPM (longest prefix match) classifiers and classifiers with general ternary-bit filters. We validate our theoretical results with a comprehensive evaluation study showing that a small error in the actions of a classifier can lead to significant memory reductions, often comparable to the best possible theoretical reduction in the trivial case when all rules have the same action.

MO257BELIMUMAB ADD-ON THERAPY MOBILIZES MEMORY B CELLS INTO CIRCULATION OF SLE PATIENTS

Abstract Background and Aims Belimumab (BLM), a recombinant human IgG-1λ monoclonal antibody directed against B-cell activating factor (BAFF), is the first approved biological agent for patients with active severe systemic lupus erythematosus (SLE) and lupus nephritis (LN). There is clinical evidence that combining belimumab with B cell depleting therapy can ameliorate disease activity in severe, refractory SLE patients. Although BLM is a B cell directed therapy and has been shown to significantly decrease total B cells, flow cytometry observations suggest a rapid increase of circulating memory B cells. The present study investigated the dynamics of B cell subsets in patients treated with or without BLM, with a special focus on the memory compartment in order to assess the characteristics of these memory B cells. Method We first performed extensive B cell subset phenotyping by high sensitivity flowcytometry according to the Euroflow protocol on whole blood from active lupus nephritis or SLE patients with other major organ involvement treated with standard of care (SOC) consisting of high dose steroids and mycophenolate mofetil combined with or without the addition of BLM. Next we characterized memory B cell gene expression profiles with single-cell RNA and V(D)J sequencing (ScRNA-SEQ). Results By employing high sensitivity flowcytometry, we established that the absolute increase in circulating memory B cells in SLE and LN patients was significant for patients who initiated BLM but not for patients treated with standard of care (Figure 1). The increase was observed in a broad range of memory B cell subsets (Unswitched, IgG1+, IgG2+, IgA1+, IgA2+) at 2 and 4 weeks following initiation of BLM treatment. This rise in memory B cells could hypothetically be attributed to either proliferation or homeostatic modulation of tissue-resident memory B cells causing a release into the circulation. ScRNA-SEQ cell cycle gene-expression was performed and established in both groups a non-proliferating phenotype [in approximately ∼94%] of memory B cells post-treatment, including absence of MKI67 as active proliferation marker. Additionally, BLM treatment did not result in an increased, and even in a severe reduction of the largest memory B cells clones after two weeks. In contrast, no change in clonality was seen after treatment with SOC. Together these data indicate that proliferation is not likely to be responsible for the observed increase in memory B cells by BLM. Furthermore, a clear difference was found in gene-expression levels between both treatment groups: BLM was responsible for the upregulation of 72 vs 10 genes in SOC, likewise 162 vs 32 genes were downregulated. Most importantly, a significant downregulation of the migration genes SELL (CD62L), CCR7, ITGB1, RAC2 and ICAM2, were specifically seen in BLM treated patients, possibly reflecting disrupted lymphocyte trafficking preventing memory B cells to either transmigrate into tissue or be retained at the tissue level. Conclusion The addition of BLM compared to standard of care in SLE patients leads to an increase of memory B cells in the circulation which appeared independent of proliferation. The process was accompanied with a strong modulation of gene-expression levels including a reduced expression of migration markers pointing towards disrupted lymphocyte trafficking. Altogether, these data could have important implications to further improve treatment strategies in severe SLE or lupus nephritis patients, for instance by establishing a deeper depletion of (autoreactive) memory B cells by the addition of rituximab after the initiation of belimumab.

Autism Spectrum Disorder

Involvement of Fibrinogen(Fg) and fibrin during various pathologies associated with neuroinflammatory diseases associated with memory reduction are well known. Elevated level of Fg, called hyperfibrinogenemia (HFg) (e.g. ≥ 4±0.1 mg/ml of plasma during inflammation vs.~2±0.1 mg/ml of normal plasma [1]), accompanies inflammatory diseases such as stroke [2,3], hypertension [1,4], diabetes [5] and Traumatic Brain Injury (TBI) [6-9]. Blood level of Fg increases during inflammation in general [10].

Targeted therapy for GBM using CAR T-Cells and CMV

Involvement of Fibrinogen(Fg) and fibrin during various pathologies associated with neuroinflammatory diseases associated with memory reduction are well known. Elevated level of Fg, called hyperfibrinogenemia (HFg) (e.g. ≥ 4±0.1 mg/ml of plasma during inflammation vs.~2±0.1 mg/ml of normal plasma [1]), accompanies inflammatory diseases such as stroke [2,3], hypertension [1,4], diabetes [5] and Traumatic Brain Injury (TBI) [6-9]. Blood level of Fg increases during inflammation in general [10].

Correlation Between Indicators of Blood Cells AggregationLevel and the Number of Lymphoblasts and Monocytes inPatients with Glioblastomas

Involvement of Fibrinogen(Fg) and fibrin during various pathologies associated with neuroinflammatory diseases associated with memory reduction are well known. Elevated level of Fg, called hyperfibrinogenemia (HFg) (e.g. ≥ 4±0.1 mg/ml of plasma during inflammation vs.~2±0.1 mg/ml of normal plasma [1]), accompanies inflammatory diseases such as stroke [2,3], hypertension [1,4], diabetes [5] and Traumatic Brain Injury (TBI) [6-9]. Blood level of Fg increases during inflammation in general [10].

GKS and UGKS for High-Speed Flows

The gas-kinetic scheme (GKS) and the unified gas-kinetic scheme (UGKS) are numerical methods based on the gas-kinetic theory, which have been widely used in the numerical simulations of high-speed and non-equilibrium flows. Both methods employ a multiscale flux function constructed from the integral solutions of kinetic equations to describe the local evolution process of particles’ free transport and collision. The accumulating effect of particles’ collision during transport process within a time step is used in the construction of the schemes, and the intrinsic simulating flow physics in the schemes depends on the ratio of the particle collision time and the time step, i.e., the so-called cell’s Knudsen number. With the initial distribution function reconstructed from the Chapman–Enskog expansion, the GKS can recover the Navier–Stokes solutions in the continuum regime at a small Knudsen number, and gain multi-dimensional properties by taking into account both normal and tangential flow variations in the flux function. By employing a discrete velocity distribution function, the UGKS can capture highly non-equilibrium physics, and is capable of simulating continuum and rarefied flow in all Knudsen number regimes. For high-speed non-equilibrium flow simulation, the real gas effects should be considered, and the computational efficiency and robustness of the schemes are the great challenges. Therefore, many efforts have been made to improve the validity and reliability of the GKS and UGKS in both the physical modeling and numerical techniques. In this paper, we give a review of the development of the GKS and UGKS in the past decades, such as physical modeling of a diatomic gas with molecular rotation and vibration at high temperature, plasma physics, computational techniques including implicit and multigrid acceleration, memory reduction methods, and wave–particle adaptation.

CSF-Mediated Damage in Multiple Sclerosis: More than a Hypothesis

Involvement of Fibrinogen(Fg) and fibrin during various pathologies associated with neuroinflammatory diseases associated with memory reduction are well known. Elevated level of Fg, called hyperfibrinogenemia (HFg) (e.g. ≥ 4±0.1 mg/ml of plasma during inflammation vs.~2±0.1 mg/ml of normal plasma [1]), accompanies inflammatory diseases such as stroke [2,3], hypertension [1,4], diabetes [5] and Traumatic Brain Injury (TBI) [6-9]. Blood level of Fg increases during inflammation in general [10].

Astroglia in ageing

Astroglia are neural cells of ectodermal, neuroepithelial origin responsible for homoeostasis and defence of the central nervous system (CNS). Ageing reduces the functional capacity of all organs, so does that of the nervous system, the latter is evident in the reduction of cognitive abilities, learning and memory. At the same time the progression of these deficits is very much individual and lifestyle dependent, indicating operation of mechanisms counterbalancing age-dependent decline. In physiological ageing astrocytes undergo morphological atrophy and functional asthenia; astrocytic paralysis facilitates progression of age-dependent neurodegenerative disorders. Astroglial status and homoeostatic capabilities are influenced by lifestyle including intellectual engagement, social interactions, physical exercise, and healthy diet. Maintenance of healthy lifestyle is paramount for cognitive longevity.

SWM: A High-Performance Sparse-Winograd Matrix Multiplication CNN Accelerator

Many convolutional neural network (CNN) accelerators are proposed to exploit the sparsity of the networks recently to enjoy the benefits of both computation and memory reduction. However, most accelerators cannot exploit the sparsity of both activations and weights. For those works that exploit both sparsity opportunities, they cannot achieve the stable load balance through a static scheduling (SS) strategy, which is vulnerable to the sparsity distribution. In this work, a balanced compressed sparse row format and a dynamic scheduling strategy are proposed to improve the load balance. A set-associate structure is also presented to tradeoff the load balance and hardware resource overhead. We propose SWM to accelerate the CNN inference, which supports both sparse convolution and sparse fully connected (FC) layers. SWM provides Winograd adaptability for large convolution kernels and supports both 16-bit and 8-bit quantized CNNs. Due to the activation sharing, 8-bit processing can achieve theoretically twice the performance of the 16-bit processing with the same sparsity. The architecture is evaluated with VGG16 and ResNet50, which achieves: at most 7.6 TOP/s for sparse-Winograd convolution and three TOP/s for sparse matrix multiplication with 16-bit quantization on Xilinx VCU1525 platform. SWM can process 310/725 images per second for VGG16/ResNet50 with 16-bit quantization. Compared with the state-of-the-art works, our design can achieve at least 1.53 × speedup and 1.8 × energy efficiency improvement.

Fibrinogen activates IκBα signaling in neurons via binding to its receptors, ICAM‐1 and PrP C

Mild traumatic brain injury (TBI) is the most common type of brain injury that is a major health issue. The goal of the current study was to extend our previous observations that during TBI, extravasated fibrinogen (Fg) deposited in vasculo/astrocyte interfaces and was associated with neurodegeneration and short-term memory (STM) reduction. We have previously shown that an elevated level of Fg, called hyperfibrinogenemia, causes astrocyte activation via interactions with intercellular adhesion molecule-1 (ICAM-1) and cellular prion protein (PrPC). However, the direct effect of Fg on neurons is not known. Primary mouse brain cortical neurons from C57BL/6 mice (Lonza) were grown in Primary Neural Basal Medium and PNGMTM Single QuotsTM. Twenty-four well plates, with #1 glass-bottomed coverslips coated with Poly-D-Lysine (30 µg/ml) and laminin (200 µg/ml) were used for seeding neurons at the density of 200,000 cells/well. After the required 7-day growth in complete medium, neurons were treated with various doses of Fg for 24 hours unless otherwise indicated. Each experimental group contained hirudin (1 U/ml) to inhibit the conversion of Fg to fibrin via thrombin. In parallel, experiments to test neuronal viability using a live/dead assay were performed without the presence of hirudin. Interactions of Fg with neuronal ICAM-1 and PrPC were tested using a proximity ligation assay (PLA), which allows highly specific and sensitive immunofluorescent detection to be visualized and consequently protein-protein interactions in situ to be objectively quantified. Real time PCR was performed to investigate the expression of mRNAs for interleukin 6 (IL-6) and inhibitor of kappa Bα (IκBα). The levels of IκBα mRNA reflect stimulation of nuclear factor-kappa B (NF-κB) transcription. PLA confirmed Fg-ICAM-1 and Fg-PrPC interactions on the surface of neurons. An increased expression of neuronal IκBα and IL-6 was induced after treatment with Fg for 2hr or 24 hr, respectively. Data showed that Fg, in the presence or absence of hirudin, caused increased neuronal death. Thus, Fg binding to its neuronal receptors, ICAM-1 and PrPC may activate NF-κB, triggered by IκBα signaling, that leads to neuroinflammation and increased neuronal death. This study is a first attempt to define the molecular mechanisms by which Fg causes neuroinflammation and neurodegeneration that resulted in the cognitive decline previously seen during mild-to-moderate TBI in our studies.

Food insecurity in older female mice affects food consumption, coping behaviors, and memory.

Food insecurity correlates with poor physical and mental health in older individuals, but has not been studied in a laboratory animal model. This explorative study developed a laboratory mouse model for analyzing the impact of food insecurity on food consumption, stress coping mechanisms, exploratory behavior, and memory. 18-month-old CD-1 female mice were assigned to either the food insecurity exposure condition (31 mice, 8 cages) or the control condition (34 mice, 8 cages) by cage. Over four weeks, the mice that were exposed to food insecurity received varied, unpredictable portions of their baseline food consumption (50%, 75%, 125%, 150% of baseline) for four days, followed by ad libitum access for three days, to approximate the inconsistent access to food observed in households experiencing food insecurity. Behavioral tasks were conducted before and after food insecurity exposure. Mice in the food insecurity exposure condition ate less compared to control mice during food insecurity (two-way ANOVA: group x time interaction: F7,93 = 10.95, P < 0.01) but ate more when given access to high fat food (two-way ANOVA, group x time interaction: F1,14 = 11.14, P < 0.01). Mice exposed to food insecurity increased active escaping behaviors in the forced swim test (repeated measures two-way ANOVA, group x time interaction: F1,63 = 5.40, P = 0.023). Exploratory behaviors were unaffected by food insecurity. Mice exposed to food insecurity showed a reduction in memory (repeated measures two-way ANOVA, group x time interaction: F1,61 = 4.81, P = 0.037). These results suggest that exposure to food insecurity is associated with differences in food consumption patterns, active coping mechanisms, and memory. The behavioral changes associated with food insecurity may inform research on food insecurity’s impact on health in elderly humans.

Immunotherapeutic strategy based on anti-OX40L and low dose of IL-2 to prolong graft survival in sensitized mice by inducing the generation of CD4+ and CD8+ Tregs

Alloreactive memory cells play a critical role after a second transplant and are difficult to suppress. This study investigated the effect of an immunotherapeutic strategy that combines anti-OX40L, rapamycin (Rapa), and a low dose of IL-2 in a memory cell-based adoptive model. In this model, the median survival time (MST) of the grafts of the combined treatment group was significantly extended compared to that of the control group and other treatment groups. A similar effect was observed regarding a reduction in memory T cells (Tm) and inflammatory cytokines production. Also, the percentages of Foxp3+ regulatory T cells (Tregs) increased in our model. In addition, mounting evidence has shown CD8+CD122+ T cells are also Tregs. We found that the group of CD8+CD122+PD1+ T cells was markedly increased in the combined treatment group, especially in the graft. We further demonstrated that CD8+CD122+PD1+ T cells could suppress activated T cells. Our data suggest that anti-OX40L combined with Rapa and a low dose of IL-2 can suppress Tm, modulate CD4 and CD8 Tregs, and induce long-term heart allograft survival in sensitized mice.