Intrathyroidal MHC class II antigen expression and thyroid autoimmunity

The association between MHC polymorphisms and autoimmune thyroid disease has been complicated by the observation of MHC class II antigen expression by the human thyroid gland. It is possible that the MHC associations observed in animal and human population studies may have mechanistic explanations at the level of the thyroid cell. There is evidence for expression of HLA-DR allospecific antigen in both normal and abnormal human thyroid cells, with enhanced expression in patients with autoimmune thyroid disease. Such MHC class II expression appears to be mediated primarily by lymphokine secretion from intrathyroidal T lymphocytes. Thyroid cell HLA-DR antigen participates in activation and amplification of T cells and is likely to be involved in presentation of thyroid antigen to the immune system. The relationship between these immune interactions and the initial events leading to the development of autoimmune thyroid disease remains to be understood.     

A role for UDP-glucose glycoprotein glucosyltransferase in expression and quality control of MHC class I molecules

UDP-glucose:glycoprotein glucosyltransferase 1 (UGT1) serves as a folding sensor in the calnexin/calreticulin glycoprotein quality control cycle. UGT1 recognizes disordered or hydrophobic patches near asparagine-linked nonglucosylated glycans in partially misfolded glycoproteins and reglucosylates them, returning folding intermediates to the cycle. In this study, we examine the contribution of the UGT1-regulated quality control mechanism to MHC I antigen presentation. Using UGT1-deficient mouse embryonic fibroblasts reconstituted or not with UGT1, we show that, although formation of the peptide loading complex is unaffected by the absence of UGT1, the surface level of MHC class I molecules is reduced, MHC class I maturation and assembly are delayed, and peptide selection is impaired. Most strikingly, we show using purified soluble components that UGT1 preferentially recognizes and reglucosylates MHC class I molecules associated with a suboptimal peptide. Our data suggest that, in addition to the extensively studied tapasin-mediated quality control mechanism, UGT1 adds a new level of control in the MHC class I antigen presentation pathway.     

Prevalent expression of MHC class I chain-related molecule A in human osteosarcoma

MHC class I chain-related molecule A (MICA) is one of the major ligands for activating immune-receptor NKG2D which is expressed on NK cells and cytotoxic T lymphocytes. The release and sustained expression of MICA protein can impair NKG2D-mediated cytotoxic activity by reducing NKG2D receptor on immune effector cells. The aim of the study was to investigate the expression and release of MICA in human osteosarcoma. RT-PCR, immunohistochemistry, western blotting and flow cytometry were used in analyzing the expression of MICA. Serum level of soluble MICA was quantitated by ELISA. Our data showed that MICA is prevalently expressed in osteosarcoma both in mRNA and protein level. Upregulation of MICA was found in osteosarcoma compared with benign tumors and normal bone tissues. Higher level of soluble MICA in serum can be detected in osteosarcoma patients. In conclusion, prevalent expression of MICA and higher serum level of soluble MICA may suggest a deficiency of MICA-NKG2D mediated immunosurveillance in osteosarcoma patients. Restoring the expression of NKG2D receptor on immune effector cells may contribute to a therapeutic strategy for human osteosarcoma.     

Highly diverged MHC class I mismatches are acceptable for haematopoietic stem cell transplantation

A fully major histocompatilbility complex (MHC) matched donor is not available for the majority of patients in need of a haematopoietic stem cell transplantation (SCT), which illustrates the need for a tool to define acceptable MHC disparities. Previously, we noticed that a variety of single MHC class I mismatched allogeneic donor-recipient pairs did not elicit an allogeneic cytotoxic-lymphocyte (CTL) response in vitro if the MHC amino-acid sequences had five or more differences in the alpha-helices plus five or more differences in the beta-sheet (> or =5alpha5beta) (7). To address the clinical relevance of this observation, we analysed CTL precursor (CTLp) assay outcome and SCT outcome in 53 Dutch recipients of a single MHC class I mismatched graft from an unrelated donor. Overall patient survival was 44% after 4 years. In multivariate analysis, recipients of a > or =5alpha5beta mismatched graft with negative CTLp frequencies in vitro before transplantation demonstrated superior survival: survival at 4 years was 80% as compared to 47% in recipients of other mismatched grafts with negative CTLp frequencies (hazard ratio=0.131; 95% CI=(0.03-0.61); P=0.009). This option of acceptable mismatches may enlarge the pool of potentially acceptable stem cell donors.     

A role for MHC class I molecules in synaptic plasticity and regeneration of neurons after axotomy

Recently, MHC class I molecules have been shown to be important for the retraction of synaptic connections that normally occurs during development [Huh, G.S., Boulanger, L. M., Du, H., Riquelme, P. A., Brotz, T. M. & Shatz, C. J. (2000) Science 290, 2155-2158]. In the adult CNS, a classical response of neurons to axon lesion is the detachment of synapses from the cell body and dendrites. We have investigated whether MHC I molecules are involved also in this type of synaptic detachment by studying the synaptic input to sciatic motoneurons at 1 week after peripheral nerve transection in beta2-microglobulin or transporter associated with antigen processing 1-null mutant mice, in which cell surface MHC I expression is impaired. Surprisingly, lesioned motoneurons in mutant mice showed more extensive synaptic detachments than those in wild-type animals. This surplus removal of synapses was entirely directed toward inhibitory synapses assembled in clusters. In parallel, a significantly smaller population of motoneurons reinnervated the distal stump of the transected sciatic nerve in mutants. MHC I molecules, which traditionally have been linked with immunological mechanisms, are thus crucial for a selective maintenance of synapses during the synaptic removal process in neurons after lesion, and the lack of MHC I expression may impede the ability of neurons to regenerate axons.     

Nicotinamide decreases MHC class II but not MHC class I expression and increases intercellular adhesion molecule-1 structures in non-obese diabetic mouse pancreas

Pancreases of untreated and nicotinamide (NIC)-treated pre-diabetic (10-week-old) and overtly diabetic (25-week-old) female NOD (non-obese diabetic) mice and of NON (non-obese non-diabetic) control mice were studied, with the following results. (1) Islets and ducts of overtly diabetic untreated NOD mice (25-week-old) were found to express low levels of MHC class I and II molecules, like NON controls, and high levels of adhesive molecules. (2) NIC was able to slightly affect glycaemia and insulitis, slowing down diabetes progression. Moreover it significantly decreased MHC class II expression (but not class I) in vivo by week 10, and significantly enhanced intercellular adhesion molecule-1 (ICAM-1) expression, mainly by week 25, within the pancreas, where 5-bromo-2'-deoxyuridine positive nuclei and insulin positive cells were present, demonstrating that a stimulation of endocrine cell proliferation occurs. (3) In addition, NIC partly counteracted the fall of superoxide dismutase levels, observed in untreated diabetic NOD animals. (4) In vitro studies demonstrated that NIC: (i) was able to significantly reduce nitrite accumulation and to increase NAD+NADH content significantly, and (ii) was able to increase the levels of interleukin-4, a T helper 2 lymphocyte (Th2) protective cytokine, and of interferon-alpha (IFN-alpha), which is known to be able to induce MHC class I and ICAM-1 but not MHC class II expression, as well as IFN-gamma, which is also known to be able to induce MHC class I and ICAM-1 expression. The latter, although known to be a proinflammatory Th1 cytokine, has also recently been found to exert an anti-diabetogenic role. This study therefore clearly shows that adhesive mechanisms are ongoing during the later periods of diabetes in pancreatic ducts of NOD mice, and suggests they may be involved in a persistence of the immune mechanisms of recognition, adhesion and cytolysis and/or endocrine regeneration or differentiation processes, as both NIC-increased ICAM-1 expression and 5-bromo-2'-deoxyuridine positivity imply. The effects of NIC on MHC class II (i.e. a reduction) but not class I, and, mainly, on ICAM-1 expression (i.e. an increase), together with the increase in Th2 protective cytokine levels are very interesting, and could help to explain its mechanism of action and the reasons for alternate success or failure in protecting against type 1 diabetes development.     

Natural killer cell licensing in mice with inducible expression of MHC class I

Mouse natural killer (NK) cells acquire effector function by an education process termed “licensing” mediated by inhibitory Ly49 receptors which recognize self-MHC class I. Ly49 receptors can bind to MHC class I on targets (in trans) and also to MHC class I on the NK-cell surface (in cis). Which of these interactions regulates NK-cell licensing is not yet clear. Moreover, there are no clear phenotypic differences between licensed and unlicensed NK cells, perhaps because of the previously limited ability to study NK cells with synchronized licensing. Here, we produced MHC class I-deficient mice with inducible MHC class I consisting of a single-chain trimer (SCT), ovalbumin peptide-β2 microgloblin-H2K^b (SCT-K^b). Only NK cells with a Ly49 receptor with specificity for SCT-K^b were licensed after MHC class I induction. NK cells were localized consistently in red pulp of the spleen during induced NK-cell licensing, and there were no differences in maturation or activation markers on recently licensed NK cells. Although MHC class I-deficient NK cells were licensed in hosts following SCT-K^b induction, NK cells were not licensed after induced SCT-K^b expression on NK cells themselves in MHC class I-deficient hosts. Furthermore, hematopoietic cells with induced SCT-K^b licensed NK cells more efficiently than stromal cells. These data indicate that trans interaction with MHC class I on hematopoietic cells regulates NK-cell licensing, which is not associated with other obvious phenotypic changes.The ability of natural killer (NK) cells to be activated by their targets is controlled by their activation and inhibitory receptors (1, 2). Activation receptors can recognize ligands expressed on their targets, triggering NK cells if their inhibitory receptors are not engaged. In mice, most Ly49 molecules are inhibitory receptors that recognize MHC class I molecules on target cells (3). Engagement of Ly49 receptors by MHC class I leads to signals that block NK-cell triggering during effector responses. These receptors explain the “missing-self” hypothesis, which postulates that NK cells survey tissues for normal levels of the ubiquitously expressed MHC class I molecule (4). In the absence of MHC class I, inhibition is released, and NK cells then can be activated through their activation receptors to kill and produce cytokines. Thus, the NK-cell inhibitory receptors control NK-cell responses at the effector level.NK cells in MHC class I-deficient mice, such as β₂-microglobulin (β2m)^−/− and TAP^−/− mice, do not receive inhibitory signals and should be overactive (5–7). Instead, however, NK cells from these animals are hyporesponsive to activation receptor cross-linking. Moreover, NK cells apparently lacking known MHC-specific receptors also are hyporesponsive (8). Previously, we reported that the NK-cell inhibitory receptors have a second function that can explain these paradoxical findings (7). Through recognition of self-MHC class I via their inhibitory MHC class I-specific receptors, NK cells become licensed (or educated), resulting in the functional competence of their activation receptors. For example, in C57BL/6 (H2^b) mice, Ly49C⁺ NK cells are licensed by recognition of self-H2K^b and secrete more IFN-γ after NK1.1 stimulation than do Ly49C⁻ NK cells. In contrast, IFN-γ production is comparable in Ly49A⁺ and Ly49A⁻ NK cells, indicating that Ly49A⁺ NK cells are not licensed in C57BL/6 mice because Ly49A does not recognize an H2^b class I allele. On the other hand, Ly49A⁺ NK cells are specifically licensed in mice expressing its MHC class I ligand, H2D^d, as self-MHC. Thus, the recognition of self-MHC class I leads to enhanced function of NK-cell activation receptors, and this enhancement is influenced by self-MHC–specific Ly49 receptors in mice.Other studies suggest that a similar process occurs for human NK cells that preferentially express killer Ig-like receptors (KIRs) for HLA class I recognition and effector inhibition. Although the mouse lectin-like Ly49 receptors and human KIRs are structurally distinct, they share many other features, supporting the concept that each species evolved a different genetic solution for the MHC class I-specific inhibitory receptors; that is, Ly49 receptors and KIRs now are considered to be outstanding examples of convergent evolution (9, 10). As with NK cells bearing self-MHC–specific Ly49 receptors in mice, human NK cells with self-HLA–specific KIRs display enhanced function (11–13). In contrast, NK cells from the same donor but without expression of self-HLA–specific KIRs show decreased function. Moreover, NK cells bearing the same KIR but without its cognate ligand in other donors do not display enhanced function. Thus, licensing (i.e., the education effects of self-MHC class I recognition by relevant NK-cell receptors) is now accepted as being operational in mice and humans (13).Other than expression of self-MHC–specific Ly49 receptors or KIRs and enhanced capacity to signal through their activation receptors, licensed mouse and human NK cells are not clearly distinguishable from unlicensed NK cells (7, 11, 12). However, this apparent similarity could result from a continuous process in which small numbers of NK cells are educated and then join “older” NK cells, and major phenotypic changes during licensing may be transient. Therefore previous phenotypic studies of licensed NK cells may have been limited by the inability to investigate NK cells in which licensing is synchronized.Crystallographic studies revealed two potential interaction sites on MHC class I at which interactions with Ly49 receptors might take place: Site 1 is on the left side of the peptide-binding cleft (when viewed from above with the α1 helix shown on top); site 2 is below the peptide-binding cleft (14). Mutational studies demonstrated that the Ly49 receptors engage site 2 in trans to inhibit effector functions, because point mutations at site 2 abrogated the capacity of transfected MHC class I molecules to inhibit NK cells in an Ly49-dependent manner (15–17). Furthermore, we recently showed that these same residues are involved in conferring Ly49-dependent licensing effects because transgenic (Tg) expression of site 2-mutant MHC class I molecules failed to induce licensed NK cells, whereas wild-type MHC class I alleles allowed licensing (18). Thus, Ly49 receptors interact with the same site on their MHC class I ligands for effector inhibition and licensing, providing another link of these two processes.Because NK cells also express MHC class I molecules, prior studies suggested that Ly49 receptors also can interact with their ligands on the same NK-cell surface, i.e., in cis as well as in trans with their target cell (19). Site 2 on MHC class I molecules is involved in both cis and trans interactions of Ly49 receptors with their MHC class I ligands. Studies of Tg mice expressing a mutated Ly49A receptor that cannot engage in in cis interactions but retains the ability to engage in in trans interactions showed that NK cells expressing mutated Ly49A were not licensed, suggesting that NK cells are licensed via cis interactions (20). However, the possibility that the mutated Ly49A disrupted other unknown signals or interactions required for NK-cell licensing cannot be ruled out completely. Moreover, we and others recently identified the importance of trans interactions in NK-cell licensing because unlicensed splenic β2m^−/− NK cells acquired the licensed phenotype when adoptively transferred to wild-type hosts (21, 22). On the other hand, non–H2D^d-expressing Ly49A⁺ NK cells can acquire H2D^d through contact with H2D^d-expressing cells in an Ly49A-dependent manner in vitro and in vivo (23, 24), suggesting that the transferred NK cells could have acquired host MHC class I, allowing potential interactions with MHC class I in cis as well as in trans. Thus, it remains unclear whether cis or trans interactions are required for NK-cell licensing.Previously, we used mice in which an MHC class I single-chain trimer (SCT) consisting of ovalbumin (OVA) peptide (SIINFEKL), β2m, and H2K^b heavy chain (SCT-K^b) is expressed in an MHC class I-deficient environment (7). The SCT-K^b molecule is specifically and solely recognized by Ly49C when stably expressed as a transgene in C57BL/6 mice, conferring the licensed phenotype to Ly49C⁺ NK cells. Here we were able to monitor Ly49C⁺ NK cells after acute induction of SCT-K^b and found that unlicensed Ly49C⁺ NK cells became licensed without changes in NK-cell maturation or activation markers or localization in spleen. Furthermore, our data showed trans interaction with hematopoietic cells is required for NK-cell licensing, but cis interaction is not.     

Rheumatoid T lymphocyte MHC class II expression: in vitro stimulation produces normal MHC class II expression, independent of proliferation

MHC Class II mediated immune responses, such as tuberculin PPD, are often subnormal in rheumatoid arthritis (RA). We investigated peripheral blood T lymphocyte proliferation and Class II expression after stimulation with phytohemagglutinin, OKT3 and tuberculin PPD, using double label immunofluorescence and flow cytometry. No difference was found in the percentage of Class II positive T lymphocytes or intensity of fluorescence of Class II between matched controls and patients with RA. A disparity between Class II expression and proliferation was noted, particularly in the 50% of patients with RA with poor proliferation to PPD, who showed normal Class II expression, indicating that these 2 functions are under separate control.     

Inheritance of MHC class II genes in IDDM studied in population-based affected and control families

Summary The transmission of HLA-DR and DQ was compared between 46 families with at least one child affected by insulin dependent diabetes mellitis (IDDM) and 43 healthy control families. In the patient families, there was an increased transmission of DR4 (p<0.025) and DQB1*0302 (p<0.01) from both parents to the index patient. There was an increased transmission of DQB1*0302 (p<0.03) from the mothers only. The non-inherited maternal haplotypes showed a significantly decreased frequency (p<0.01) of positively associated haplotypes (DR4-DQA1* 0301-DQB1*0302, DR3-DQA1*0501-DQB1*0201) compared to all parental haplotypes in the control families. In the control families neither transmission rates nor frequencies of non-inherited haplotypes differed from those expected in the control families. In conclusion, the observed reduction of IDDM-positively associated haplotypes in patient non-inherited maternal haplotypes, but not in non-inherited paternal haplotypes, suggests that tolerance during fetal life to maternal non-inherited HLA molecules may be important to diabetes development.Abbreviations HLA Human Leucocyte antigen - NIMH non-inherited maternal haplotype - NIPH non-inherited paternal haplotype - MHC major histocompatibility complex - RPE relative predispositional effect - RFLP restriction fragment length polymorphism - w/v weight volume     

Relevance of complotyping and subtyping of MHC class I gene products in haplotype definition for allogeneic bone marrow transplantation

In preparation for a bone marrow transplantation 217 patients and their families were complotyped for Bf, C4A and C4B in addition to the routinely performed HLA-A,B,C,DR and HLA-D typing. In 147 families uncertainties in haplotype definition occurred which could be solved in 37 cases (25%) by complotyping. Additionally, patients and their relatives were subtyped for class I gene products by one-dimensional isoelectric focusing, a method by which serologically identical HLA-A, B, or C antigens could be split in five out of 22 cases tested. The results obtained clearly show the relevance of both methodologies for finding the best match of donor/recipient pairs to help to prevent MHC-induced graft-versus-host disease after bone marrow transplantation.     

MHC class II expression through a hitherto unknown pathway supports T helper cell-dependent immune responses: implications for MHC class II deficiency

MHC class II (MHCII) deficiency or bare lymphocyte syndrome (BLS) is a severe immunodeficiency characterized by deficient T helper (Th)-cell-dependent immunity. The disease is caused by defects of the MHCII promoter complex resulting in low or absent MHCII expression. We demonstrate in a murine model of MHCII deficiency (RFX5- or CIITA-deficient mice) that residual MHCII expression by professional antigen-presenting cells (APCs) is sufficient to support activation of adoptively transferred Th cells. Furthermore, upon transplantation of WT thymic epithelium, we observed development of endogenous Th cells with restoration of Th-cell-dependent antibody responses and immunity to cytomegalovirus infection, thus opening the possibility of an alternative treatment regimen for BLS. Residual MHCII expression was further induced by the presence of Th cells and also other stimuli. Analysis of CIITA/RFX5 double-deficient animals revealed that this inducible MHCII expression is genetically independent of the known promoter complex and thus constitutes an alternative MHCII expression pathway. In these experiments, we also detected a novel repressive function of the RFX complex in the absence of CIITA.     

Beta cell MHC class I is a late requirement for diabetes

Type 1 diabetes occurs as a result of an autoimmune attack on the insulin-producing beta cells. Although CD8 T cells have been implicated both early and late in this process, the requirement for direct interaction between these cells and MHC class I on the beta cells has not been demonstrated. By using nonobese diabetic mice lacking beta cell class I expression, we show that both initiation and progression of insulitis proceeds unperturbed. However, without beta cell class I expression, the vast majority of these mice do not develop hyperglycemia. These findings demonstrate that a direct interaction between CD8 T cells and beta cells is not required for initiation or early disease progression. The requirement for class I on beta cells is a relatively late checkpoint in the development of diabetes.     

Direct peptide-regulatable interactions between MHC class I molecules and tapasin

Tapasin (Tpn) has been implicated in multiple steps of the MHC class I assembly pathway, but the mechanisms of function remain incompletely understood. Using purified proteins, we could demonstrate direct binding of Tpn to peptide-deficient forms of MHC class I molecules at physiological temperatures. Tpn also bound to M10.5, a pheromone receptor-associated MHC molecule that has an open and empty groove and that shares significant sequence identity with class I sequences. Two types of MHC class I-Tpn complexes were detectable in vitro depending on the input proteins; those depleted in beta(2)m, and those containing beta(2)m. Both were competent for subsequent assembly with peptides, but the latter complexes assembled more rapidly. Thus, the assembly rate of Tpn-associated class I was determined by the conditions under which Tpn-MHC class I complexes were induced. Peptide loading of class I inhibited Tpn-class I-binding interactions, and peptide-depletion enhanced binding. In combination with beta(2)m, certain peptides induced efficient dissociation of preformed Tpn-class I complexes. Together, these studies demonstrate direct Tpn-MHC class I interactions and preferential binding of empty MHC class I by Tpn, and that the Tpn-class I interaction is regulated by both beta(2)m and peptide. In cells, Tpn is likely to be a direct mediator of peptide-regulated binding and release of MHC class I from the TAP complex.     

Recycling MHC class I molecules and endosomal peptide loading.

MHC class I molecules usually present peptides derived from endogenous antigens that are bound in the endoplasmic reticulum. Loading of exogenous antigens on class I molecules, e.g., in cross-priming, sometimes occurs, but the intracellular location where interaction between the antigenic fragment and class I takes place is unclear. Here we show that measles virus F protein can be presented by class I in transporters associated with antigen processing-independent, NH(4)Cl-sensitive manner, suggesting that class I molecules are able to interact and bind antigen in acidic compartments, like class II molecules. Studies on intracellular transport of green fluorescent protein-tagged class I molecules in living cells confirmed that a small fraction of class I molecules indeed enters classical MHC class II compartments (MIICs) and is transported in MIICs back to the plasma membrane. Fractionation studies show that class I complexes in MIICs contain peptides. The pH in MIIC (around 5.0) is such that efficient peptide exchange can occur. We thus present evidence for a pathway for class I loading that is shared with class II molecules.     

Essential glycan-dependent interactions optimize MHC class I peptide loading

In this study we sought to better understand the role of the glycoprotein quality control machinery in the assembly of MHC class I molecules with high-affinity peptides. The lectin-like chaperone calreticulin (CRT) and the thiol oxidoreductase ERp57 participate in the final step of this process as part of the peptide-loading complex (PLC). We provide evidence for an MHC class I/CRT intermediate before PLC engagement and examine the nature of that chaperone interaction in detail. To investigate the mechanism of peptide loading and roles of individual components, we reconstituted a PLC subcomplex, excluding the Transporter Associated with Antigen Processing, from purified, recombinant proteins. ERp57 disulfide linked to the class I-specific chaperone tapasin and CRT were the minimal PLC components required for MHC class I association and peptide loading. Mutations disrupting the interaction of CRT with ERp57 or the class I glycan completely eliminated PLC activity in vitro. By using the purified system, we also provide direct evidence for a role for UDP-glucose:glycoprotein glucosyltransferase 1 in MHC class I assembly. The recombinant Drosophila enzyme reglucosylated MHC class I molecules associated with suboptimal ligands and allowed PLC reengagement and high-affinity peptide exchange. Collectively, the data indicate that CRT in the PLC enhances weak tapasin/class I interactions in a manner that is glycan-dependent and regulated by UDP-glucose:glycoprotein glucosyltransferase 1.     

Regulation of CNS synapses by neuronal MHC class I

Until recently, neurons in the healthy brain were considered immune-privileged because they did not appear to express MHC class I (MHCI). However, MHCI mRNA was found to be regulated by neural activity in the developing visual system and has been detected in other regions of the uninjured brain. Here we show that MHCI regulates aspects of synaptic function in response to activity. MHCI protein is colocalized postsynaptically with PSD-95 in dendrites of hippocampal neurons. In vitro, whole-cell recordings of hippocampal neurons from beta2m/TAP1 knockout (KO) mice, which have reduced MHCI surface levels, indicate a 40% increase in mini-EPSC (mEPSC) frequency. mEPSC frequency is also increased 100% in layer 4 cortical neurons. Similarly, in KO hippocampal cultures, there is a modest increase in the size of presynaptic boutons relative to WT, whereas postsynaptic parameters (PSD-95 puncta size and mEPSC amplitude) are normal. In EM of intact hippocampus, KO synapses show a corresponding increase in vesicles number. Finally, KO neurons in vitro fail to respond normally to TTX treatment by scaling up synaptic parameters. Together, these results suggest that postsynaptically localized MHCl acts in homeostatic regulation of synaptic function and morphology during development and in response to activity blockade. The results also imply that MHCI acts retrogradely across the synapse to translate activity into lasting change in structure.