Multi-dimensional classification using Bayesian networks for stationary and evolving streaming data

Hoy en día, con la evolución continua y rápida de las tecnologías de la información y los dispositivos de computación, se recogen y almacenan continuamente grandes volúmenes de datos en distintos dominios y a través de diversas aplicaciones del mundo real. La extracción de conocimiento útil de una cantidad tan enorme de datos no se puede realizar habitualmente de forma manual, y requiere el uso de técnicas adecuadas de aprendizaje automático y de minería de datos. La clasificación es una de las técnicas más importantes que ha sido aplicada con éxito a varias áreas. En general, la clasificación se compone de dos pasos principales: en primer lugar, aprender un modelo de clasificación o clasificador a partir de un conjunto de datos de entrenamiento, y en segundo lugar, clasificar las nuevas instancias de datos utilizando el clasificador aprendido. La clasificación es supervisada cuando todas las etiquetas están presentes en los datos de entrenamiento (es decir, datos completamente etiquetados), semi-supervisada cuando sólo algunas etiquetas son conocidas (es decir, datos parcialmente etiquetados), y no supervisada cuando todas las etiquetas están ausentes en los datos de entrenamiento (es decir, datos no etiquetados). Además, aparte de esta taxonomía, el problema de clasificación se puede categorizar en unidimensional o multidimensional en función del número de variables clase, una o más, respectivamente; o también puede ser categorizado en estacionario o cambiante con el tiempo en función de las características de los datos y de la tasa de cambio subyacente. A lo largo de esta tesis, tratamos el problema de clasificación desde tres perspectivas diferentes, a saber, clasificación supervisada multidimensional estacionaria, clasificación semisupervisada unidimensional cambiante con el tiempo, y clasificación supervisada multidimensional cambiante con el tiempo. Para llevar a cabo esta tarea, hemos usado básicamente los clasificadores Bayesianos como modelos. La primera contribución, dirigiéndose al problema de clasificación supervisada multidimensional estacionaria, se compone de dos nuevos métodos de aprendizaje de clasificadores Bayesianos multidimensionales a partir de datos estacionarios. Los métodos se proponen desde dos puntos de vista diferentes. El primer método, denominado CB-MBC, se basa en una estrategia de envoltura de selección de variables que es voraz y hacia delante, mientras que el segundo, denominado MB-MBC, es una estrategia de filtrado de variables con una aproximación basada en restricciones y en el manto de Markov. Ambos métodos han sido aplicados a dos problemas reales importantes, a saber, la predicción de los inhibidores de la transcriptasa inversa y de la proteasa para el problema de infección por el virus de la inmunodeficiencia humana tipo 1 (HIV-1), y la predicción del European Quality of Life-5 Dimensions (EQ-5D) a partir de los cuestionarios de la enfermedad de Parkinson con 39 ítems (PDQ-39). El estudio experimental incluye comparaciones de CB-MBC y MB-MBC con los métodos del estado del arte de la clasificación multidimensional, así como con métodos comúnmente utilizados para resolver el problema de predicción de la enfermedad de Parkinson, a saber, la regresión logística multinomial, mínimos cuadrados ordinarios, y mínimas desviaciones absolutas censuradas. En ambas aplicaciones, los resultados han sido prometedores con respecto a la precisión de la clasificación, así como en relación al análisis de las estructuras gráficas que identifican interacciones conocidas y novedosas entre las variables. La segunda contribución, referida al problema de clasificación semi-supervisada unidimensional cambiante con el tiempo, consiste en un método nuevo (CPL-DS) para clasificar flujos de datos parcialmente etiquetados. Los flujos de datos difieren de los conjuntos de datos estacionarios en su proceso de generación muy rápido y en su aspecto de cambio de concepto. Es decir, los conceptos aprendidos y/o la distribución subyacente están probablemente cambiando y evolucionando en el tiempo, lo que hace que el modelo de clasificación actual sea obsoleto y deba ser actualizado. CPL-DS utiliza la divergencia de Kullback-Leibler y el método de bootstrapping para cuantificar y detectar tres tipos posibles de cambio: en las predictoras, en la a posteriori de la clase o en ambas. Después, si se detecta cualquier cambio, un nuevo modelo de clasificación se aprende usando el algoritmo EM; si no, el modelo de clasificación actual se mantiene sin modificaciones. CPL-DS es general, ya que puede ser aplicado a varios modelos de clasificación. Usando dos modelos diferentes, el clasificador naive Bayes y la regresión logística, CPL-DS se ha probado con flujos de datos sintéticos y también se ha aplicado al problema real de la detección de código malware, en el cual los nuevos ficheros recibidos deben ser continuamente clasificados en malware o goodware. Los resultados experimentales muestran que nuestro método es efectivo para la detección de diferentes tipos de cambio a partir de los flujos de datos parcialmente etiquetados y también tiene una buena precisión de la clasificación. Finalmente, la tercera contribución, sobre el problema de clasificación supervisada multidimensional cambiante con el tiempo, consiste en dos métodos adaptativos, a saber, Locally Adpative-MB-MBC (LA-MB-MBC) y Globally Adpative-MB-MBC (GA-MB-MBC). Ambos métodos monitorizan el cambio de concepto a lo largo del tiempo utilizando la log-verosimilitud media como métrica y el test de Page-Hinkley. Luego, si se detecta un cambio de concepto, LA-MB-MBC adapta el actual clasificador Bayesiano multidimensional localmente alrededor de cada nodo cambiado, mientras que GA-MB-MBC aprende un nuevo clasificador Bayesiano multidimensional. El estudio experimental realizado usando flujos de datos sintéticos multidimensionales indica los méritos de los métodos adaptativos propuestos. ABSTRACT Nowadays, with the ongoing and rapid evolution of information technology and computing devices, large volumes of data are continuously collected and stored in different domains and through various real-world applications. Extracting useful knowledge from such a huge amount of data usually cannot be performed manually, and requires the use of adequate machine learning and data mining techniques. Classification is one of the most important techniques that has been successfully applied to several areas. Roughly speaking, classification consists of two main steps: first, learn a classification model or classifier from an available training data, and secondly, classify the new incoming unseen data instances using the learned classifier. Classification is supervised when the whole class values are present in the training data (i.e., fully labeled data), semi-supervised when only some class values are known (i.e., partially labeled data), and unsupervised when the whole class values are missing in the training data (i.e., unlabeled data). In addition, besides this taxonomy, the classification problem can be categorized into uni-dimensional or multi-dimensional depending on the number of class variables, one or more, respectively; or can be also categorized into stationary or streaming depending on the characteristics of the data and the rate of change underlying it. Through this thesis, we deal with the classification problem under three different settings, namely, supervised multi-dimensional stationary classification, semi-supervised unidimensional streaming classification, and supervised multi-dimensional streaming classification. To accomplish this task, we basically used Bayesian network classifiers as models. The first contribution, addressing the supervised multi-dimensional stationary classification problem, consists of two new methods for learning multi-dimensional Bayesian network classifiers from stationary data. They are proposed from two different points of view. The first method, named CB-MBC, is based on a wrapper greedy forward selection approach, while the second one, named MB-MBC, is a filter constraint-based approach based on Markov blankets. Both methods are applied to two important real-world problems, namely, the prediction of the human immunodeficiency virus type 1 (HIV-1) reverse transcriptase and protease inhibitors, and the prediction of the European Quality of Life-5 Dimensions (EQ-5D) from 39-item Parkinson’s Disease Questionnaire (PDQ-39). The experimental study includes comparisons of CB-MBC and MB-MBC against state-of-the-art multi-dimensional classification methods, as well as against commonly used methods for solving the Parkinson’s disease prediction problem, namely, multinomial logistic regression, ordinary least squares, and censored least absolute deviations. For both considered case studies, results are promising in terms of classification accuracy as well as regarding the analysis of the learned MBC graphical structures identifying known and novel interactions among variables. The second contribution, addressing the semi-supervised uni-dimensional streaming classification problem, consists of a novel method (CPL-DS) for classifying partially labeled data streams. Data streams differ from the stationary data sets by their highly rapid generation process and their concept-drifting aspect. That is, the learned concepts and/or the underlying distribution are likely changing and evolving over time, which makes the current classification model out-of-date requiring to be updated. CPL-DS uses the Kullback-Leibler divergence and bootstrapping method to quantify and detect three possible kinds of drift: feature, conditional or dual. Then, if any occurs, a new classification model is learned using the expectation-maximization algorithm; otherwise, the current classification model is kept unchanged. CPL-DS is general as it can be applied to several classification models. Using two different models, namely, naive Bayes classifier and logistic regression, CPL-DS is tested with synthetic data streams and applied to the real-world problem of malware detection, where the new received files should be continuously classified into malware or goodware. Experimental results show that our approach is effective for detecting different kinds of drift from partially labeled data streams, as well as having a good classification performance. Finally, the third contribution, addressing the supervised multi-dimensional streaming classification problem, consists of two adaptive methods, namely, Locally Adaptive-MB-MBC (LA-MB-MBC) and Globally Adaptive-MB-MBC (GA-MB-MBC). Both methods monitor the concept drift over time using the average log-likelihood score and the Page-Hinkley test. Then, if a drift is detected, LA-MB-MBC adapts the current multi-dimensional Bayesian network classifier locally around each changed node, whereas GA-MB-MBC learns a new multi-dimensional Bayesian network classifier from scratch. Experimental study carried out using synthetic multi-dimensional data streams shows the merits of both proposed adaptive methods.

Copresentation of natural HIV-1 agonist and antagonist ligands fails to induce the T cell receptor signaling cascade

It is not known how human immunodeficiency virus type 1 (HIV-1)-derived antagonist peptides interfere with intracellular activation of cytotoxic T lymphocytes (CTL). We identified Gag epitope variants in HIV-1-infected patients that act as antagonists of CTL responses to unmutated epitopes. We then investigated the effect that presentation of each variant has on the early events of T cell receptor (TCR) signal transduction. We found that altered peptide ligands (APL) failed to induce phosphorylation of pp36, a crucial adaptor protein involved in TCR signal transduction. We further investigated the effect that simultaneous presentation of APL and native antigen at low, physiological, peptide concentrations (1 nM) has on TCR signal transduction, and we found that the presence of APL can completely inhibit induction of the protein tyrosine phosphorylation events of the TCR signal transduction cascade.

Latent varicella–zoster virus is located predominantly in neurons in human trigeminal ganglia

Varicella–zoster virus (VZV) is a human herpesvirus that causes varicella (chicken pox) as a primary infection and, after a variable period of latency in trigeminal and dorsal root ganglia, reactivates to cause herpes zoster (shingles). Both of these conditions may be followed by a variety of neurological complications, especially in immunocompromised individuals such as those with human immunodeficiency virus (HIV) infection. There have been a number of conflicting reports regarding the cellular location of latent VZV within human ganglia. To address this controversy we examined fixed wax-embedded trigeminal ganglia from 30 individuals obtained at autopsy, including 11 with HIV infection, 2 neonates, and 17 immunocompetent individuals, for the presence of latent VZV. Polymerase chain reaction (PCR), in situ hybridization, and PCR in situ amplification techniques with oligonucleotide probes and primer sequences to VZV genes 18, 21, 29, and 63 were used. VZV DNA in ganglia was detected in 15 individuals by using PCR alone, and in 12 individuals (6 normal non-HIV and 6 positive HIV individuals, but not neonatal ganglia) by using PCR in situ amplification. When in situ hybridization alone was used, 5 HIV-positive individuals and only 1 non-HIV individual showed VZV nucleic acid signals in ganglia. In all of the VZV-positive ganglia examined, VZV nucleic acid was detected in neuronal nuclei. Only occasional nonneuronal cells contained VZV DNA. We conclude from these studies that the neuron is the predominant site of latent VZV in human trigeminal ganglia.

Social stress results in altered glucocorticoid regulation and shorter survival in simian acquired immune deficiency syndrome

From early in the AIDS epidemic, psychosocial stressors have been proposed as contributors to the variation in disease course. To test this hypothesis, rhesus macaques were assigned to stable or unstable social conditions and were inoculated with the simian immunodeficiency virus. Animals in the unstable condition displayed more agonism and less affiliation, shorter survival, and lower basal concentrations of plasma cortisol compared with stable animals. Early after inoculation, but before the emergence of group differences in cortisol levels, animals receiving social threats had higher concentrations of simian immunodeficiency virus RNA in plasma, and those engaging in affiliation had lower concentrations. The results indicate that social factors can have a significant impact on the course of immunodeficiency disease. Socially induced changes in pituitary–adrenal hormones may be one mechanism mediating this relationship.

Correlates of latent and productive HIV type-1 infection in tonsillar CD4+ T cells

Correlates of virus load and characteristics of virus-producing cells in tonsillar tissue were investigated. Our results suggest that when less than 1:100 tonsillar CD4+ T cells from individuals infected with HIV type-1 (HIV-1) contain replication competent provirus, the level of CD4+ T cells in tonsils is comparable to that observed in uninfected individuals. Virus load at or above this level was associated with low CD4 cell numbers in tonsillar tissue. Only a few percent of all infected T cells in tonsillar tissue were active virus producers, with minor differences observed between individuals. Plasma viremia was found to correlate with infectious virus load in tonsillar tissue. With less than 1:1,000 of CD4 cells in lymphoid tissues being involved in active virus production, direct cytopathic effect by HIV-1 on infected CD4 cells is unlikely to fully explain the immunodeficiency seen in AIDS.

Btk dosage determines sensitivity to B cell antigen receptor cross-linking

Mutations in Btk result in the B cell immunodeficiencies X-linked agammaglobulinemia (XLA) in humans and X-linked immunodeficiency (xid) in mice. Btk is a critical component of signaling pathways regulating B cell development and function. We used a genetic approach to determine whether Btk is also limiting for these processes. One allele of a murine Btk transgene expressed a dosage of Btk (25% of endogenous levels in splenic B cells) sufficient to restore normal numbers of phenotypically mature conventional B cells in xid mice. 2,4,6-trinitrophenyl–Ficoll response, anti-IgM-induced proliferation, B1 cell development, and serum IgM and IgG3 levels remained significantly impaired in these animals. B cells from Btk −/− transgenic mice also responded poorly to anti-IgM, indicating that the xid mutation does not create a dominant negative form of Btk. Response to 2,4,6-trinitrophenyl–Ficoll and B cell receptor cross-linking were increased 3- to 4-fold in xid mice homozygous for the transgene. These results demonstrate that Btk is a limiting component of B cell antigen receptor signaling pathways and suggest that B cell development and response to antigen may require different levels of Btk activity.

Perturbation of the T cell repertoire in rheumatoid arthritis

Aberrations in the T cell repertoire with the emergence of oligoclonal populations have been described in patients with rheumatoid arthritis (RA). However, the extent of the repertoire perturbations as well as the underlying mechanisms are not known. We now have examined the diversity of the peripheral CD4 T cell repertoire by determining the frequencies of arbitrarily selected T cell receptor (TCR) β-chain sequences. Healthy individuals displayed a highly diverse repertoire, with a median frequency of individual TCR β-chain sequences of 1 in 2.4 × 107 CD4 T cells. In RA patients, the median TCR β-chain frequency was increased 10-fold, indicating marked contraction of the repertoire (P < 0.001). The loss in TCR diversity was not limited to CD4 memory T cells but also involved the compartment of naive T cells, suggesting that it reflected an abnormality in T cell repertoire formation and not a consequence of antigen recognition in the synovium. Also, control patients with chronic inflammatory disease such as hepatitis C expressed a diverse repertoire indistinguishable from that of normals. Telomere length studies indicated an increased replicative history of peripheral CD4 T cells in RA patients, suggesting an enhanced turnover within the CD4 compartment. Compared with age-matched controls, terminal restriction fragment sizes were 1.7 kilobases shorter (P < 0.001). These data demonstrate an altered CD4 T cell homeostasis in RA that may contribute to the autoimmune response as well as to the immunodeficiency in these patients.

Molecular cloning, sequence, expression, and processing of the interleukin 16 precursor

Interleukin 16 (IL-16) has been shown to function as chemoattractant factor, as a modulator of T-cell activation, and as an inhibitor of immunodeficiency virus replication. The recent identification of inconsistencies in published IL-16 cDNA nucleotide sequences led to the proposal that IL-16 is synthesized in the form of a large precursor protein (pro-IL-16). To identify the true transcriptional start of the IL-16 mRNA rapid amplification of cDNA ends methods were applied. The complete pro-IL-16 cDNA was subsequently molecularly cloned, sequenced, and expressed in COS-7 cells. We report here that pro-IL-16 is most likely synthesized as a 67-kDa protein and is encoded from a major 2.6-kb transcript. Recombinant pro-IL-16 polypeptides are specifically cleaved in lysates of CD8(+) cells, suggesting that the naturally secreted bioactive form of IL-16 is smaller than the originally published 130 amino acids fragment. Moreover, in contrast to other interleukins such as IL-15, IL-16 mRNA expression is almost exclusively limited to lymphatic tissues underlining the potential of IL-16 as an immune regulatory molecule.

Inactivating mutations in an SH2 domain-encoding gene in X-linked lymphoproliferative syndrome

X-linked lymphoproliferative syndrome (XLP) is an inherited immunodeficiency characterized by increased susceptibility to Epstein–Barr virus (EBV). In affected males, primary EBV infection leads to the uncontrolled proliferation of virus-containing B cells and reactive cytotoxic T cells, often culminating in the development of high-grade lymphoma. The XLP gene has been mapped to chromosome band Xq25 through linkage analysis and the discovery of patients harboring large constitutional genomic deletions. We describe here the presence of small deletions and intragenic mutations that specifically disrupt a gene named DSHP in 6 of 10 unrelated patients with XLP. This gene encodes a predicted protein of 128 amino acids composing a single SH2 domain with extensive homology to the SH2 domain of SHIP, an inositol polyphosphate 5-phosphatase that functions as a negative regulator of lymphocyte activation. DSHP is expressed in transformed T cell lines and is induced following in vitro activation of peripheral blood T lymphocytes. Expression of DSHP is restricted in vivo to lymphoid tissues, and RNA in situ hybridization demonstrates DSHP expression in activated T and B cell regions of reactive lymph nodes and in both T and B cell neoplasms. These observations confirm the identity of DSHP as the gene responsible for XLP, and suggest a role in the regulation of lymphocyte activation and proliferation. Induction of DSHP may sustain the immune response by interfering with SHIP-mediated inhibition of lymphocyte activation, while its inactivation in XLP patients results in a selective immunodeficiency to EBV.

WIP, a protein associated with Wiskott–Aldrich syndrome protein, induces actin polymerization and redistribution in lymphoid cells

Wiskott–Aldrich syndrome (WAS) is an X-linked immunodeficiency caused by mutations that affect the WAS protein (WASP) and characterized by cytoskeletal abnormalities in hematopoietic cells. By using the yeast two-hybrid system we have identified a proline-rich WASP-interacting protein (WIP), which coimmunoprecipitated with WASP from lymphocytes. WIP binds to WASP at a site distinct from the Cdc42 binding site and has actin as well as profilin binding motifs. Expression of WIP in human B cells, but not of a WIP truncation mutant that lacks the actin binding motif, increased polymerized actin content and induced the appearance of actin-containing cerebriform projections on the cell surface. These results suggest that WIP plays a role in cortical actin assembly that may be important for lymphocyte function.

The DNA Helicase Activity of BLM Is Necessary for the Correction of the Genomic Instability of Bloom Syndrome Cells

Bloom syndrome (BS) is a rare autosomal recessive disorder characterized by growth deficiency, immunodeficiency, genomic instability, and the early development of cancers of many types. BLM, the protein encoded by BLM, the gene mutated in BS, is localized in nuclear foci and absent from BS cells. BLM encodes a DNA helicase, and proteins from three missense alleles lack displacement activity. BLM transfected into BS cells reduces the frequency of sister chromatid exchanges and restores BLM in the nucleus. Missense alleles fail to reduce the sister chromatid exchanges in transfected BS cells or restore the normal nuclear pattern. BLM complements a phenotype of a Saccharomyces cerevisiae sgs1 top3 strain, and the missense alleles do not. This work demonstrates the importance of the enzymatic activity of BLM for its function and nuclear localization pattern.

Cluster of Differentiation Antigen 4 (CD4) Endocytosis and Adaptor Complex Binding Require Activation of the CD4 Endocytosis Signal by Serine Phosphorylation

Cluster of differentiation antigen 4 (CD4), the T lymphocyte antigen receptor component and human immunodeficiency virus coreceptor, is down-modulated when cells are activated by antigen or phorbol esters. During down-modulation CD4 dissociates from p56lck, undergoes endocytosis through clathrin-coated pits, and is then sorted in early endosomes to late endocytic organelles where it is degraded. Previous studies have suggested that phosphorylation and a dileucine sequence are required for down-modulation. Using transfected HeLa cells, in which CD4 endocytosis can be studied in the absence of p56lck, we show that the dileucine sequence in the cytoplasmic domain is essential for clathrin-mediated CD4 endocytosis. However, this sequence is only functional as an endocytosis signal when neighboring serine residues are phosphorylated. Phosphoserine is required for rapid endocytosis because CD4 molecules in which the cytoplasmic domain serine residues are substituted with glutamic acid residues are not internalized efficiently. Using surface plasmon resonance, we show that CD4 peptides containing the dileucine sequence bind weakly to clathrin adaptor protein complexes 2 and 1. The affinity of this interaction is increased 350- to 700-fold when the peptides also contain phosphoserine residues.

Somatic Cell Mutants Resistant to Retrovirus Replication: Intracellular Blocks during the Early Stages of Infection

To identify cellular functions involved in the early phase of the retroviral life cycle, somatic cell mutants were isolated after selection for resistance to infection. Rat2 fibroblasts were treated with chemical mutagens, and individual virus-resistant clones were recovered after selection for resistance to infection. Two clones were characterized in detail. Both mutant lines were resistant to infection by both ecotropic and amphotropic murine viruses, as well as by human immunodeficiency virus type 1 pseudotypes. One clone showed a strong block to reverse transcription of the retroviral RNA, including formation of the earliest DNA products. The second clone showed normal levels of viral DNA synthesis but did not allow formation of the circular DNAs normally found in the nucleus. Cell fractionation showed that the viral preintegration complex was present in a form that could not be extracted under conditions that readily extracted the complex from wild-type cells. The results suggest that the DNA was trapped in a nonproductive state and excluded from the nucleus of the infected cell. The properties of these two mutant lines suggest that host gene products play important roles both before and after reverse transcription.

The Saccharomyces cerevisiae Homologue of Human Wiskott–Aldrich Syndrome Protein Las17p Interacts with the Arp2/3 Complex

Yeast Las17 protein is homologous to the Wiskott–Aldrich Syndrome protein, which is implicated in severe immunodeficiency. Las17p/Bee1p has been shown to be important for actin patch assembly and actin polymerization. Here we show that Las17p interacts with the Arp2/3 complex. LAS17 is an allele-specific multicopy suppressor of ARP2 and ARP3 mutations; overexpression restores both actin patch organization and endocytosis defects in ARP2 temperature-sensitive (ts) cells. Six of seven ARP2 ts mutants and at least one ARP3 ts mutant are synthetically lethal with las17Δ ts confirming functional interaction with the Arp2/3 complex. Further characterization of las17Δ cells showed that receptor-mediated internalization of α factor by the Ste2 receptor is severely defective. The polarity of normal bipolar bud site selection is lost. Las17-gfp remains localized in cortical patches in vivo independently of polymerized actin and is required for the polarized localization of Arp2/3 as well as actin. Coimmunoprecipitation of Arp2p with Las17p indicates that Las17p interacts directly with the complex. Two hybrid results also suggest that Las17p interacts with actin, verprolin, Rvs167p and several other proteins including Src homology 3 (SH3) domain proteins, suggesting that Las17p may integrate signals from different regulatory cascades destined for the Arp2/3p complex and the actin cytoskeleton.

Protective immunity to HIV-1 in SCID/beige mice reconstituted with peripheral blood lymphocytes of exposed but uninfected individuals

Immunodeficiency typically appears many years after initial HIV infection. This long, essentially asymptomatic period contributes to the transmission of HIV in human populations. In rare instances, clearance of HIV-1 infection has been observed, particularly in infants. There are also reports of individuals who have been frequently exposed to HIV-1 but remain seronegative for the virus, and it has been hypothesized that these individuals are resistant to infection by HIV-1. However, little is known about the mechanism of immune clearance or protection against HIV-1 in these high-risk individuals because it is difficult to directly demonstrate in vivo protective immunity. Although most of these high-risk individuals show an HIV-1-specific cell-mediated immune response using in vitro assays, their peripheral blood lymphocytes (PBLs) are still susceptible to HIV infection in tissue culture. To study this further in vivo, we have established a humanized SCID mouse infection model whereby T-, B-, and natural killer-cell defective SCID/beige mice that have been reconstituted with normal human PBLs can be infected with HIV-1. When the SCID/beige mice were reconstituted with PBLs from two different multiply exposed HIV-1 seronegative individuals, the mice showed resistance to infection by two strains of HIV-1 (macrophage tropic and T cell tropic), although the same PBLs were easily infected in vitro. Mice reconstituted with PBLs from non-HIV-exposed controls were readily infected. When the same reconstituted mice were depleted of human CD8 T cells, however, they became susceptible to HIV-1 infection, indicating that the in vivo protection required CD8 T cells. This provides clear experimental evidence that some multiply exposed, HIV-1-negative individuals have in vivo protective immunity that is CD8 T cell-dependent. Understanding the mechanism of such protective immunity is critical to the design and testing of effective prophylactic vaccines and immunotherapeutic regimens.