Estimation of the biomechanical mammographic deformation of the breast using machine learning models.

BACKGROUND: A typical problem in the registration of MRI and X-ray mammography is the nonlinear deformation applied to the breast during mammography. We have developed a method for virtual deformation of the breast using a biomechanical model automatically constructed from MRI. The virtual deformation is applied in two steps: unloaded state estimation and compression simulation. The finite element method is used to solve the deformation process. However, the extensive computational cost prevents its usage in clinical routine. METHODS: We propose three machine learning models to overcome this problem: an extremely randomized tree (first model), extreme gradient boosting (second model), and deep learning-based bidirectional long short-term memory with an attention layer (third model) to predict the deformation of a biomechanical model. We evaluated our methods with 516 breasts with realistic compression ratios up to 76%. FINDINGS: We first applied one-fold validation, in which the second and third models performed better than the first model. We then applied ten-fold validation. For the unloaded state estimation, the median RMSE for the second and third models is 0.8 mm and 1.2 mm, respectively. For the compression, the median RMSE is 3.4 mm for both models. We evaluated correlations between model accuracy and characteristics of the clinical datasets such as compression ratio, breast volume, and tissue types. INTERPRETATION: Using the proposed models, we achieved accurate results comparable to the finite element model, with a speedup of factor 240 using the extreme gradient boosting model. These proposed models can replace the finite element model simulation, enabling clinically relevant real-time application.

Image fusion of Ultrasound Computer Tomography volumes with X-ray mammograms using a biomechanical model based 2D/3D registration.

Ultrasound Computer Tomography (USCT) is a promising breast imaging modality under development. Comparison to a standard method like mammography is essential for further development. Due to significant differences in image dimensionality and compression state of the breast, correlating USCT images and X-ray mammograms is challenging. In this paper we present a 2D/3D registration method to improve the spatial correspondence and allow direct comparison of the images. It is based on biomechanical modeling of the breast and simulation of the mammographic compression. We investigate the effect of including patient-specific material parameters estimated automatically from USCT images. The method was systematically evaluated using numerical phantoms and in-vivo data. The average registration accuracy using the automated registration was 11.9mm. Based on the registered images a method for analysis of the diagnostic value of the USCT images was developed and initially applied to analyze sound speed and attenuation images based on X-ray mammograms as ground truth. Combining sound speed and attenuation allows differentiating lesions from surrounding tissue. Overlaying this information on mammograms, combines quantitative and morphological information for multimodal diagnosis.

Automatic multimodal 2D/3D breast image registration using biomechanical FEM models and intensity-based optimization.

Due to their different physical origin, X-ray mammography and Magnetic Resonance Imaging (MRI) provide complementary diagnostic information. However, the correlation of their images is challenging due to differences in dimensionality, patient positioning and compression state of the breast. Our automated registration takes over part of the correlation task. The registration method is based on a biomechanical finite element model, which is used to simulate mammographic compression. The deformed MRI volume can be compared directly with the corresponding mammogram. The registration accuracy is determined by a number of patient-specific parameters. We optimize these parameters--e.g. breast rotation--using image similarity measures. The method was evaluated on 79 datasets from clinical routine. The mean target registration error was 13.2mm in a fully automated setting. On basis of our results, we conclude that a completely automated registration of volume images with 2D mammograms is feasible. The registration accuracy is within the clinically relevant range and thus beneficial for multimodal diagnosis.

Accelerated mammary maturation and differentiation, and delayed MMTVneu-induced tumorigenesis of K303R mutant ERalpha transgenic mice.

We identified a somatic mutation in estrogen receptor-alpha (ERalpha) in breast cancer causing a lysine to arginine transition (K303R) resulting in hypersensitivity to estrogen, altered associations with coactivators and corepressors and altered posttranslational modifications of ERalpha. We have developed a transgenic mouse expressing the K303R mutant ERalpha under control of the mouse mammary tumor virus (MMTV) promoter. At 4 months of age, K303R ERalpha transgenic animals demonstrate precocious alveolar budding compared with wild-type ERalpha transgenic mice or nontransgenic littermates. Despite these morphologic differences, K303R ERalpha transgenic mice displayed no differences in levels of ERalpha, progesterone receptor or proliferation at this time-point. Pregnancy or chronic estrogen plus progesterone exposure in K303R ERalpha transgenic mice also resulted in significantly more alveolar budding, increased beta-casein production and dilated ducts when compared with nontransgenic littermates. To examine the effects of mutant expression on tumorigenesis, mutant ERalpha mice were crossed with FVB-MMTVneu mice and significantly delayed time to neu-mediated tumorigenesis in bigenic animals. In contrast, mutant expression did not affect carcinogen-induced tumorigenesis. Collectively, these data demonstrate that aberrant estrogenic signaling through the K303R ERalpha mutation may lead to precocious alveolar budding in virgin mice, and to an expedited maturation and differentiation phenotype in the mammary glands of hormonally stimulated animals.

Direct acetylation of the estrogen receptor alpha hinge region by p300 regulates transactivation and hormone sensitivity.

Regulation of nuclear receptor gene expression involves dynamic and coordinated interactions with histone acetyl transferase (HAT) and deacetylase complexes. The estrogen receptor (ERalpha) contains two transactivation domains regulating ligand-independent and -dependent gene transcription (AF-1 and AF-2 (activation functions 1 and 2)). ERalpha-regulated gene expression involves interactions with cointegrators (e.g. p300/CBP, P/CAF) that have the capacity to modify core histone acetyl groups. Here we show that the ERalpha is acetylated in vivo. p300, but not P/CAF, selectively and directly acetylated the ERalpha at lysine residues within the ERalpha hinge/ligand binding domain. Substitution of these residues with charged or polar residues dramatically enhanced ERalpha hormone sensitivity without affecting induction by MAPK signaling, suggesting that direct ERalpha acetylation normally suppresses ligand sensitivity. These ERalpha lysine residues also regulated transcriptional activation by histone deacetylase inhibitors and p300. The conservation of the ERalpha acetylation motif in a phylogenetic subset of nuclear receptors suggests that direct acetylation of nuclear receptors may contribute to additional signaling pathways involved in metabolism and development.

A hypersensitive estrogen receptor-alpha mutation in premalignant breast lesions.

The best current model of breast cancer evolution suggests that most cancers arise from certain premalignant lesions. We have identified a common (34%) somatic mutation in the estrogen receptor (ER)-alpha gene in a series of 59 typical hyperplasias, a type of early premalignant breast lesion. The mutation, which affects the border of the hinge and hormone binding domains of ER-alpha, showed increased sensitivity to estrogen as compared with wild-type ER-alpha in stably transfected breast cancer cells, including markedly increased proliferation at subphysiological levels of estrogen. The mutated ER-alpha exhibits enhanced binding to the TIF-2 coactivator at low levels of hormone, which may partially explain its increased estrogen responsiveness. These data suggest that this mutation may promote or accelerate the development of cancer from premalignant breast lesions.

Tamoxifen-bound estrogen receptor (ER) strongly interacts with the nuclear matrix protein HET/SAF-B, a novel inhibitor of ER-mediated transactivation.

The estrogen receptor (ER) is a ligand-dependent transcription factor that acts in a cell- and promoter-specific manner. Evidence suggests that the activity of the ER can be regulated by a number of other stimuli (e.g. growth factors) and that the effects of the ER are modulated by nuclear factors termed coregulators. While the interplay among these factors may in part explain the pleiotropic effects elicited by the ER, there are several other less well described mechanisms of control, such as interactions with the nuclear matrix. Here we report that the nuclear matrix protein/scaffold attachment factor HET/SAF-B is an ER-interacting protein. ER and HET/SAF-B interact in in vitro binding assays, with HET binding to both the ER DNA-binding domain and the hinge region. Coimmunoprecipitation experiments reveal that HET/SAF-B and ER associate in cell lines in the presence or absence of estradiol, but binding is increased by the antiestrogen tamoxifen. HET/SAF-B enhances tamoxifen antagonism of estrogen-induced ER-mediated transactivation, but at high concentrations can inhibit both estrogen and tamoxifen-induced ER activity. HET/SAF-B-mediated repression of ER activity is dependent upon interaction with the ER-DBD. While the existence of high-affinity binding sites for the ER in the nuclear matrix has been known for some time, we now provide evidence of a specific nuclear matrix protein binding to the ER. Furthermore, our data showing that HET/SAF-B binds to ER particularly strongly in the presence of tamoxifen suggests that it may be important for the antagonist effect of tamoxifen.

Estrogen receptor variants.

Estrogen receptor (ER)3 gene expression in breast epithelium is an intricately regulated event. The human ER gene is transcribed from at least three different promoters which are expressed in a cell- and tissue-specific manner, and result in mRNA isoforms with unique 5'-untranslated exons. The ER is overexpressed in about two thirds of breast tumors, and even in early premalignant breast lesions compared with adjacent normal breast epithelium. Furthermore, normal breast epithelium as well as breast cancer tissue contains alternatively spliced ER mRNA variants where single or multiple exons are skipped. It is still unclear if any or all of the ER mRNA splicing variants are translated in vivo, and if a change in the balance of ER variants could effect tumor development and progression to hormone-independent growth. Although infrequent in primary breast cancer, single amino acid changes within the ER in metastatic disease which might influence cell proliferation may also contribute to neoplastic progression of the mammary epithelium.

Metal-binding properties of a calcium-dependent monoclonal antibody.

The calcium-dependent mAb, M1 (also called anti-Flag or 4E11) was studied using a newly developed metal-sensitive enzyme-linked immunosorbent assay (ELISA). This antibody, specific for a calcium complex of the peptide antigen, Asp-Tyr-Lys-Asp-Asp-Asp-Asp-Lys, has found widespread use as a mild purification reagent for Flag-epitope tagged recombinant proteins. Although M1 affinity columns release monovalent Flagged proteins in the absence of calcium, the antibody retains substantial affinity for the Flag sequence even in metal-free conditions, so that it has been impossible to use it to develop a metal-sensitive ELISA assay. This is due to the ability of the antibody to remain bound to polyvalent surface-coated antigen, for instance, when Flagged proteins are bound to ELISA plates or blotting filters. The resultant antigen polyvalence raises the avidity of the Flag antibody to a point where the reaction is essentially calcium-independent. However, when the antibody itself was made monovalent, by proteolytic cleavage to the Fab, this situation was reversed and the ELISA reaction became calcium-dependent. This new metal-dependent ELISA assay was used to explore the metal requirements of the antibody in detail. Among divalent metals, binding tapered off with increasing radius above that of calcium, or with decreasing radius below that of calcium. Several smaller metals, such as nickel, acted as inhibitors of the binding reaction. Substantial binding was demonstrated for heavy metals such as cadmium, lanthanum and samarium. Because it is of interest to use this antibody for the co-crystallization of recombinant Flag-fusion proteins, the ability to bind heavy metals was a significant finding.

Evidence from sequence information that the interleukin-1 receptor is a transmembrane GTPase.

Evidence is presented that the cytoplasmic domain of the type I interleukin-1 receptor (IL-1R) may be a GTPase. This domain conserves segments of hydrophobic amino acids that suggest a structural relatedness to the ras protooncogene protein and other members of the GTPase superfamily, despite a lack of significant detectable sequence homology. When the hydrophobic segments of the IL-1R were aligned with similar segments of the GTPases, it became apparent that the IL-1Rs possess a number of conserved amino acids that represent plausible functional residues for base-specific binding of GTP, magnesium chelation, and phosphate ester hydrolysis. Furthermore, a segment of five contiguous residues were found that is identical between ras and the IL-1R, and which is positioned to form part of the guanine base binding pocket. If this model is correct, then the IL-1Rs possess a highly conserved effector protein binding region, but one that is entirely unrelated to the effector regions of other superfamily members. Therefore, if the IL-1R is indeed a GTPase, then its activation function may be directed to as-yet unrecognized effector target proteins, as part of a unique cellular signal transduction pathway.

Mapping the detailed specificity of a calcium-dependent monoclonal antibody through the use of soluble positional scanning combinatorial libraries: identification of potent calcium-independent antigens.

The detailed specificity of monoclonal antibody M1, which has been reported to bind in a calcium-dependent manner to the 'FLAG' sequence DYKDDDDK-NH2, was examined using soluble hexa- and decapeptide positional scanning synthetic combinatorial libraries (PS-SCLs) made up of 52 x 10(6) and 4 x 10(12) different sequences, respectively. To study the influence of calcium on the specificity of this antigen-antibody interaction, each PS-SCL was screened in the presence and absence of calcium using a competitive ELISA. Overall, peptide mixtures had greater inhibitory activity against mAb M1 binding to FLAG in the absence of calcium. A total of 16 individual hexapeptides were identified, all of which contained the motif-DYK_K_(-), and were recognized by mAb M1 in the absence of calcium with 50- to 100-fold higher affinity than the FLAG octapeptide (IC50 = 273 nM). On average, the same set of peptides bound 10-fold less effectively in the presence of calcium. Upon screening the decapeptide PS-SCL in the absence of calcium, lysine was also more active in the fifth position than the original aspartic acid. Based on the screening results, 24 individual decapeptides were prepared and were found to have activities 10- to 100-fold higher than the FLAG octapeptide in the absence of calcium. The specificity of lysine at the fifth position in the antigen-antibody interaction was further examined by synthesizing and assaying substitution analogs at this position for the octapeptide and hexapeptide forms of the FLAG sequence, as well as for two hexapeptides identified from the PS-SCL. Truncation analog analysis was also carried out on the FLAG octapeptide to determine optimal antigen length for antibody binding. Overall, lysine at the fifth position could be substituted with ornithine with no significant loss in activity, and peptide length was not a critical factor for antibody binding in the absence of calcium. Also, the octapeptide having lysine at the fifth position in place of the aspartic acid had the same activity in the presence or absence of calcium. This study demonstrates the ease and effectiveness of PS-SCLs over individual peptide analogs for the examination of the degree of cross-reactivity for a given monoclonal antibody as well as for the identification of novel, high-affinity peptides.

Identification of regions in interleukin-1 alpha important for activity.

Saturation mutagenesis of the mature human interleukin-1 alpha (IL-1 alpha) gene has been performed. Following expression in Escherichia coli, the biological and receptor binding activities of the mutant proteins were examined. Most of the molecule could be altered with little effect on either function. More than 3,500 mutants were examined, and only 23 unique amino acid sequences were identified which resulted in an altered ratio of biological to binding activity when compared with wild-type IL-1 alpha. These proteins possessed mutations at 38 of the 159 amino acid residues in IL-1 alpha. Random mutagenesis at several of these positions identified further substitutions that affected activity. Examination of a model for IL-1 alpha localized most of the residues which altered activity along one face of the molecule. This region appears to be distinct from areas of IL-1 which have been postulated to make contact with IL-1 receptor.

Retrospective: 12 years of antigenic determinant predictions, and more.

The Hopp and Woods hydrophilicity method for locating antigenic determinants was published in 1981. In the years since then, the method has been used widely and has played a vital role in many antigenic structure studies. The method has been criticized occasionally and replacements for it have been proposed. However, at this time, the Hopp and Woods method remains a method of choice for identifying antigenic sites and other protein interaction sites, because it has a higher success rate than other similar methods. Key to the success of this method is its cautious approach to charge-charge interactions, giving equal weight to positively and negatively charged residues, whereas other methods tend to favor one or the other. It has become clear that sites chosen by our method tend to be highly exposed, charged regions of the protein's surface which project into the environment and therefore have ample opportunity to contact other proteins. We have been exploring new uses for the method, and have found some applications in locating sites for other types of interactions, including those with other macromolecules such as DNA and RNA. It seems likely that this simple and reliable procedure will continue to find use in predicting the locations of major antigenic epitopes, and may also find use as a general prediction method to identify interaction sites on proteins that make charge-dependent contacts with a variety of other biological macromolecules.

A calcium-dependent antibody for identification and purification of recombinant proteins.

We report a straightforward methodology for purification of recombinant proteins by incorporating a short hydrophilic peptide marker segment at their N-termini. A calcium-dependent antibody that reacts primarily with the first three amino acids of this peptide segment was used to affinity purify the fusion proteins in a single chromatographic step. The marker peptide could subsequently be removed by proteolysis with the enzyme enterokinase.

A pre-aspartate-specific protease from human leukocytes that cleaves pro-interleukin-1 beta.

Interleukin-1 beta is a 17.4-kilodalton hormone derived from a 33-kilodalton inactive precursor produced by monocytes. We used the precursor as a substrate to detect proteolytic activities in peripheral blood mono-nuclear cell-conditioned medium that might be involved in interleukin-1 beta processing. We found that the conditioned medium, following passage through DEAE-Sephacel, generates a biologically active fragment from the precursor that runs slightly higher than the mature hormone in sodium dodecyl sulfate-polyacrylamide gel electrophoresis. The responsible activity behaved as a single protein in ion exchange chromatography. It was completely inhibited by metal ion chelators and not by inhibitors of serine, cysteine, or aspartate proteases, and it was dependent on both calcium (or magnesium) and zinc. The enzyme was not inhibited by three substrate-based metalloprotease inhibitors, phosphoramidon, benzyloxycarbonyl-Gly-Leu-NH2, and N-(2-carboxy-3-phenylpropionyl)-Leu. NH2-terminal sequence analysis showed that cleavage of the precursor occurred between a histidine and an aspartate residue, and digestion of synthetic peptides indicated that the protease is specific for pre-aspartate cleavages.