Amino acid preferences of retroviral proteases for amino-terminal positions in a type 1 cleavage site

Different Mutation Tolerance of Lentiviral (HIV-1) and Deltaretroviral (BLV and HTLV) Protease Precursors

The bovine leukemia virus (BLV) and the human T-lymphothropic viruses (HTLVs) are members of the deltaretrovirus genus of Retroviridae family. An essential event of the retroviral life cycle is the processing of the polyproteins by the viral protease (PR); consequently, these enzymes became important therapeutic targets of the anti-retroviral drugs. As compared to human immunodeficiency viruses (HIVs), the deltaretroviruses have a different replication strategy, as they replicate predominantly in the DNA form, by forcing the infected cell to divide, unlike HIV-1, which replicates mainly by producing a vast number of progeny virions and by reinfection. Due to bypassing the error-prone reverse transcription step of replication, the PRs of deltaretroviruses did not undergo such extensive evolution as HIV PRs and remained more highly conserved. In this work, we studied the abilities of wild-type and modified BLV, HTLV (type 1, 2 and 3), and HIV-1 PRs (fused to an N-terminal MBP tag) for self-processing. We designed a cleavage site mutant MBP-fused BLV PR precursor as well, this recombinant enzyme was unable for self-proteolysis, the MBP fusion tag decreased its catalytic efficiency but showed an unusually low K_i for the IB-268 protease inhibitor. Our results show that the HTLV and BLV deltaretrovirus PRs exhibit lower mutation tolerance as compared to HIV-1 PR, and are less likely to retain their activity upon point mutations at various positions, indicating a higher flexibility of HIV-1 PR in tolerating mutations under selective pressure.

Specificity of the HIV-1 Protease on Substrates Representing the Cleavage Site in the Proximal Zinc-Finger of HIV-1 Nucleocapsid Protein

To explore the sequence context-dependent nature of the human immunodeficiency virus type 1 (HIV-1) protease’s specificity and to provide a rationale for viral mutagenesis to study the potential role of the nucleocapsid (NC) processing in HIV-1 replication, synthetic oligopeptide substrates representing the wild-type and modified versions of the proximal cleavage site of HIV-1 NC were assayed as substrates of the HIV-1 protease (PR). The S1′ substrate binding site of HIV-1 PR was studied by an in vitro assay using KIVKCF↓NCGK decapeptides having amino acid substitutions of N17 residue of the cleavage site of the first zinc-finger domain, and in silico calculations were also performed to investigate amino acid preferences of S1′ site. Second site substitutions have also been designed to produce “revertant” substrates and convert a non-hydrolysable sequence (having glycine in place of N17) to a substrate. The specificity constants obtained for peptides containing non-charged P1′ substitutions correlated well with the residue volume, while the correlation with the calculated interaction energies showed the importance of hydrophobicity: interaction energies with polar residues were related to substantially lower specificity constants. Cleavable “revertants” showed one residue shift of cleavage position due to an alternative productive binding mode, and surprisingly, a double cleavage of a substrate was also observed. The results revealed the importance of alternative binding possibilities of substrates into the HIV-1 PR. The introduction of the “revertant” mutations into infectious virus clones may provide further insights into the potential role of NC processing in the early phase of the viral life-cycle.

Biochemical Characterization, Specificity and Inhibition Studies of HTLV-1, HTLV-2, and HTLV-3 Proteases

The human T-lymphotropic viruses (HTLVs) are causative agents of severe diseases including adult T-cell leukemia. Similar to human immunodeficiency viruses (HIVs), the viral protease (PR) plays a crucial role in the viral life-cycle via the processing of the viral polyproteins. Thus, it is a potential target of anti-retroviral therapies. In this study, we performed in vitro comparative analysis of human T-cell leukemia virus type 1, 2, and 3 (HTLV-1, -2, and -3) proteases. Amino acid preferences of S4 to S1′ subsites were studied by using a series of synthetic oligopeptide substrates representing the natural and modified cleavage site sequences of the proteases. Biochemical characteristics of the different PRs were also determined, including catalytic efficiencies and dependence of activity on pH, temperature, and ionic strength. We investigated the effects of different HIV-1 PR inhibitors (atazanavir, darunavir, DMP-323, indinavir, ritonavir, and saquinavir) on enzyme activities, and inhibitory potentials of IB-268 and IB-269 inhibitors that were previously designed against HTLV-1 PR. Comparative biochemical analysis of HTLV-1, -2, and -3 PRs may help understand the characteristic similarities and differences between these enzymes in order to estimate the potential of the appearance of drug-resistance against specific HTLV-1 PR inhibitors.

Biochemical Characterization of Human Retroviral-Like Aspartic Protease 1 (ASPRV1)

The human retroviral-like aspartic protease 1 (ASPRV1) is a mammalian retroviral-like enzyme that catalyzes a critical proteolytic step during epidermal differentiation; therefore, it is also referred to as skin-specific aspartic protease (SASPase). Neutrophil granulocytes were also found recently to express ASPRV1 that is involved in the progression of acute chronic inflammation of the central nervous system, especially in autoimmune encephalomyelitis. Thus, investigation of ASPRV1 is important due to its therapeutic or diagnostic potential. We investigated the structural characteristics of ASPRV1 by homology modeling; analysis of the proposed structure was used for interpretation of in vitro specificity studies. For in-vitro characterization, activities of SASP28 and SASP14 enzyme forms were measured using synthetic oligopeptide substrates. We demonstrated that self-processing of SASP28 precursor causes autoactivation of the protease. The highest activity was measured for GST-SASP14 at neutral pH and at high ionic strength, and we proved that pepstatin A and acetyl-pepstatin can also inhibit the protease. In agreement with the structural characteristics, the relatively lower urea dissociation constant implied lower dimer stability of SASP14 compared to that of HIV-1 protease. The obtained structural and biochemical characteristics support better understanding of ASPRV1 function in the skin and central nervous system.

Functional Study of the Retrotransposon-Derived Human PEG10 Protease

Paternally expressed gene 10 (PEG10) is a human retrotransposon-derived imprinted gene. The mRNA of PEG10 encodes two protein isoforms: the Gag-like protein (RF1_PEG10) is coded by reading frame 1, while the Gag-Pol-like polyprotein (RF1/RF2_PEG10) is coded by reading frames 1 and 2. The proteins are translated by a typical retroviral frameshift mechanism. The protease (PR) domain of RF2_PEG10 contains an -Asp-Ser-Gly- sequence, which corresponds to the consensus -Asp-Ser/Thr-Gly- active-site motif of retroviral aspartic proteases. The function of the aspartic protease domain of RF2_PEG10 remains unclear. To elucidate the function of PEG10 protease (PR_PEG10), we designed a frameshift mutant (_fsRF1/RF2_PEG10) for comparison with the RF1/RF2_PEG10 form. To study the effects of PR_PEG10 on cellular proliferation and viability, mammalian HEK293T and HaCaT cells were transfected with plasmids coding for either RF1/RF2_PEG10, the frameshift mutant (_fsRF1/RF2_PEG10), or a PR active-site (D370A) mutant _fsRF1/RF2_PEG10. Our results indicate that _fsRF1/RF2_PEG10 overexpression results in increased cellular proliferation. Remarkably, transfection with _fsRF1/RF2_PEG10 had a detrimental effect on cell viability. We hypothesize that PR_PEG10 plays an important role in the function of this retroviral remnant, mediating the proliferation of cells and possibly implicating it in the inhibition of apoptosis.

Dimer Interface Organization is a Main Determinant of Intermonomeric Interactions and Correlates with Evolutionary Relationships of Retroviral and Retroviral-Like Ddi1 and Ddi2 Proteases

The life cycles of retroviruses rely on the limited proteolysis catalyzed by the viral protease. Numerous eukaryotic organisms also express endogenously such proteases, which originate from retrotransposons or retroviruses, including DNA damage-inducible 1 and 2 (Ddi1 and Ddi2, respectively) proteins. In this study, we performed a comparative analysis based on the structural data currently available in Protein Data Bank (PDB) and Structural summaries of PDB entries (PDBsum) databases, with a special emphasis on the regions involved in dimerization of retroviral and retroviral-like Ddi proteases. In addition to Ddi1 and Ddi2, at least one member of all seven genera of the Retroviridae family was included in this comparison. We found that the studied retroviral and non-viral proteases show differences in the mode of dimerization and density of intermonomeric contacts, and distribution of the structural characteristics is in agreement with their evolutionary relationships. Multiple sequence and structure alignments revealed that the interactions between the subunits depend mainly on the overall organization of the dimer interface. We think that better understanding of the general and specific features of proteases may support the characterization of retroviral-like proteases.

Biochemical characterization of Ty1 retrotransposon protease

Ty1 is one of the many transposons in the budding yeast Saccharomyces cerevisiae. The life-cycle of Ty1 shows numerous similarities with that of retroviruses, e.g. the initially synthesized polyprotein precursor undergoes proteolytic processing by the protease. The retroviral proteases have become important targets of current antiretroviral therapies due to the critical role of the limited proteolysis of Gag-Pol polyprotein in the replication cycle and they therefore belong to the most well-studied enzymes. Comparative analyses of retroviral and retroviral-like proteases can help to explore the key similarities and differences which may help understanding how resistance is developed against protease inhibitors, but the available information about the structural and biochemical characteristics of retroviral-like, and especially retrotransposon, proteases is limited. To investigate the main characteristics of Ty1 retrotransposon protease of Saccharomyces cerevisiae, untagged and His₆-tagged forms of Ty1 protease were expressed in E. coli. After purification of the recombinant proteins, activity measurements were performed using synthetic oligopeptide and fluorescent recombinant protein substrates, which represented the wild-type and the modified forms of naturally occurring cleavage sites of the protease. We investigated the dependence of enzyme activity on different reaction conditions (pH, temperature, ionic strength, and urea concentration), and determined enzyme kinetic parameters for the studied substrates. Inhibitory potentials of 10 different protease inhibitors were also tested. Ty1 protease was not inhibited by the inhibitors which have been designed against human immunodeficiency virus type 1 protease and are approved as antiretroviral therapeutics. A quaternary structure of homodimeric Ty1 protease was proposed based on homology modeling, and this structure was used to support interpretation of experimental results and to correlate some structural and biochemical characteristics with that of other retroviral proteases.

HIV-1 Protease Uses Bi-Specific S2/S2' Subsites to Optimize Cleavage of Two Classes of Target Sites

Retroviral proteases (PRs) have a unique specificity that allows cleavage of sites with or without a P1' proline. A P1' proline is required at the MA/CA cleavage site due to its role in a post-cleavage conformational change in the capsid protein. However, the HIV-1 PR prefers to have large hydrophobic amino acids flanking the scissile bond, suggesting that PR recognizes two different classes of substrate sequences. We analyzed the cleavage rate of over 150 combinations of six different HIV-1 cleavage sites to explore rate determinants of cleavage. We found that cleavage rates are strongly influenced by the two amino acids flanking the amino acids at the scissile bond (P2-P1/P1'-P2'), with two complementary sets of rules. When P1' is proline, the P2 side chain interacts with a polar region in the S2 subsite of the PR, while the P2' amino acid interacts with a hydrophobic region of the S2' subsite. When P1' is not proline, the orientations of the P2 and P2' side chains with respect to the scissile bond are reversed; P2 residues interact with a hydrophobic face of the S2 subsite, while the P2' amino acid usually engages hydrophilic amino acids in the S2' subsite. These results reveal that the HIV-1 PR has evolved bi-functional S2 and S2' subsites to accommodate the steric effects imposed by a P1' proline on the orientation of P2 and P2' substrate side chains. These results also suggest a new strategy for inhibitor design to engage the multiple specificities in these subsites.

A recombinant fusion protein-based, fluorescent protease assay for high throughput-compatible substrate screening

In connection with the intensive investigation of proteases, several methods have been developed for analysis of the substrate specificity. Due to the great number of proteases and the expected target molecules to be analyzed, time- and cost-efficient high-throughput screening (HTS) methods are preferred. Here we describe the development and application of a separation-based HTS-compatible fluorescent protease assay, which is based on the use of recombinant fusion proteins as substrates of proteases. The protein substrates used in this assay consists of N-terminal (hexahistidine and maltose binding protein) fusion tags, cleavage sequences of the tobacco etch virus (TEV) and HIV-1 proteases, and a C-terminal fluorescent protein (mApple or mTurquoise2). The assay is based on the fluorimetric detection of the fluorescent proteins, which are released from the magnetic bead-attached substrates by the proteolytic cleavage. The protease assay has been applied for activity measurements of TEV and HIV-1 proteases to test the suitability of the system for enzyme kinetic measurements, inhibition studies, and determination of pH optimum. We also found that denatured fluorescent proteins can be renatured after SDS-PAGE of denaturing conditions, but showed differences in their renaturation abilities. After in-gel renaturation both substrates and cleavage products can be identified by in-gel UV detection.

Research applications of proteolytic enzymes in molecular biology

Proteolytic enzymes (also termed peptidases, proteases and proteinases) are capable of hydrolyzing peptide bonds in proteins. They can be found in all living organisms, from viruses to animals and humans. Proteolytic enzymes have great medical and pharmaceutical importance due to their key role in biological processes and in the life-cycle of many pathogens. Proteases are extensively applied enzymes in several sectors of industry and biotechnology, furthermore, numerous research applications require their use, including production of Klenow fragments, peptide synthesis, digestion of unwanted proteins during nucleic acid purification, cell culturing and tissue dissociation, preparation of recombinant antibody fragments for research, diagnostics and therapy, exploration of the structure-function relationships by structural studies, removal of affinity tags from fusion proteins in recombinant protein techniques, peptide sequencing and proteolytic digestion of proteins in proteomics. The aim of this paper is to review the molecular biological aspects of proteolytic enzymes and summarize their applications in the life sciences.

Genetic characterization by composite sequence analysis of a new pathogenic field strain of equine infectious anemia virus from the 2006 outbreak in Ireland

Equine infectious anemia virus (EIAV), the causative agent of equine infectious anaemia (EIA), possesses the least-complex genomic organization of any known extant lentivirus. Despite this relative genetic simplicity, all of the complete genomic sequences published to date are derived from just two viruses, namely the North American EIAVWYOMING (EIAVWY) and Chinese EIAVLIAONING (EIAVLIA) strains. In 2006, an outbreak of EIA occurred in Ireland, apparently as a result of the importation of contaminated horse plasma from Italy and subsequent iatrogenic transmission to foals. This EIA outbreak was characterized by cases of severe, sometimes fatal, disease. To begin to understand the molecular mechanisms underlying this pathogenic phenotype, complete proviral genomic sequences in the form of 12 overlapping PCR-generated fragments were obtained from four of the EIAV-infected animals, including two of the index cases. Sequence analysis of multiple molecular clones produced from each fragment demonstrated the extent of diversity within individual viral genes and permitted construction of consensus whole-genome sequences for each of the four viral isolates. In addition, complete env gene sequences were obtained from 11 animals with differing clinical profiles, despite exposure to a common EIAV source. Although the overall genomic organization of the Irish EIAV isolates was typical of that seen in all other strains, the European viruses possessed ≤80 % nucleotide sequence identity with either EIAVWY or EIAVLIA. Furthermore, phylogenetic analysis suggested that the Irish EIAV isolates developed independently of the North American and Chinese viruses and that they constitute a separate monophyletic group.

Inhibition of XMRV and HIV-1 proteases by pepstatin A and acetyl-pepstatin

The kinetic properties of two classical inhibitors of aspartic proteases (PRs), pepstatin A and acetyl-pepstatin, were compared in their interactions with HIV-1 and xenotropic murine leukemia virus related virus (XMRV) PRs. Both compounds are substantially weaker inhibitors of XMRV PR than of HIV-1 PR. Previous kinetic and structural studies characterized HIV-1 PR-acetyl-pepstatin and XMRV PR-pepstatin A complexes and suggested dramatically different binding modes. Interaction energies were calculated for the possible binding modes and suggested a strong preference for the one-inhibitor binding mode for HIV-1 PR-acetyl-pepstatin and the two-inhibitor binding mode for XMRV PR-pepstatin A interactions. Comparison of the molecular models suggested that in the case of XMRV PR the relatively unfavorable interactions at S3' and the favorable interactions at S4 and S4' sites with the statine residues may shift the ground state binding towards the two-inhibitor binding mode, whereas the single molecule ground state binding of statines to the HIV-1 PR appear to be more favorable. The preferred single molecular binding to HIV-1 PR allows the formation of the transition state complex, represented by substantially better binding constants. Intriguingly, the crystal structure of the complex of acetyl-pepstatin with XMRV PR has shown a mixed type of binding: the unusual binding mode of two molecules of the inhibitor to the enzyme, in a mode very similar to the previously determined complex with pepstatin A, together with the classical binding mode found for HIV-1 PR. The structure is thus in good agreement with the very similar interaction energies calculated for the two types of binding.

Critical differences in HIV-1 and HIV-2 protease specificity for clinical inhibitors

Clinical inhibitor amprenavir (APV) is less effective on HIV-2 protease (PR₂) than on HIV-1 protease (PR₁). We solved the crystal structure of PR₂ with APV at 1.5 Å resolution to identify structural changes associated with the lowered inhibition. Furthermore, we analyzed the PR₁ mutant (PR(1M) ) with substitutions V32I, I47V, and V82I that mimic the inhibitor binding site of PR₂. PR(1M) more closely resembled PR₂ than PR₁ in catalytic efficiency on four substrate peptides and inhibition by APV, whereas few differences were seen for two other substrates and inhibition by saquinavir (SQV) and darunavir (DRV). High resolution crystal structures of PR(1M) with APV, DRV, and SQV were compared with available PR₁ and PR₂ complexes. Val/Ile32 and Ile/Val47 showed compensating interactions with SQV in PR(1M) and PR₁, however, Ile82 interacted with a second SQV bound in an extension of the active site cavity of PR(1M). Residues 32 and 82 maintained similar interactions with DRV and APV in all the enzymes, whereas Val47 and Ile47 had opposing effects in the two subunits. Significantly diminished interactions were seen for the aniline of APV bound in PR₁ (M) and PR₂ relative to the strong hydrogen bonds observed in PR₁, consistent with 15- and 19-fold weaker inhibition, respectively. Overall, PR(1M) partially replicates the specificity of PR₂ and gives insight into drug resistant mutations at residues 32, 47, and 82. Moreover, this analysis provides a structural explanation for the weaker antiviral effects of APV on HIV-2.

Identification of the protease cleavage sites in a reconstituted Gag polyprotein of an HERV-K(HML-2) element

BACKGROUND

The human genome harbors several largely preserved HERV-K(HML-2) elements. Although this retroviral family comes closest of all known HERVs to producing replication competent virions, mutations acquired during their chromosomal residence have rendered them incapable of expressing infectious particles. This also holds true for the HERV-K113 element that has conserved open reading frames (ORFs) for all its proteins in addition to a functional LTR promoter. Uncertainty concerning the localization and impact of post-insertional mutations has greatly hampered the functional characterization of these ancient retroviruses and their proteins. However, analogous to other betaretroviruses, it is known that HERV-K(HML-2) virions undergo a maturation process during or shortly after release from the host cell. During this process, the subdomains of the Gag polyproteins are released by proteolytic cleavage, although the nature of the mature HERV-K(HML-2) Gag proteins and the exact position of the cleavage sites have until now remained unknown.

RESULTS

By aligning the amino acid sequences encoded by the gag-pro-pol ORFs of HERV-K113 with the corresponding segments from 10 other well-preserved human specific elements we identified non-synonymous post-insertional mutations that have occurred in this region of the provirus. Reversion of these mutations and a partial codon optimization facilitated the large-scale production of maturation-competent HERV-K113 virus-like particles (VLPs). The Gag subdomains of purified mature VLPs were separated by reversed-phase high-pressure liquid chromatography and initially characterized using specific antibodies. Cleavage sites were identified by mass spectrometry and N-terminal sequencing and confirmed by mutagenesis. Our results indicate that the gag gene product Pr74Gag of HERV-K(HML-2) is processed to yield p15-MA (matrix), SP1 (spacer peptide of 14 amino acids), p15, p27-CA (capsid), p10-NC (nucleocapsid) and two C-terminally encoded glutamine- and proline-rich peptides, QP1 and QP2, spanning 23 and 19 amino acids, respectively.

CONCLUSIONS

Expression of reconstituted sequences of original HERV elements is an important tool for studying fundamental aspects of the biology of these ancient viruses. The analysis of HERV-K(HML-2) Gag processing and the nature of the mature Gag proteins presented here will facilitate further studies of the discrete functions of these proteins and of their potential impact on the human host.

Identification of structural mechanisms of HIV-1 protease specificity using computational peptide docking: implications for drug resistance

Drug-resistant mutations (DRMs) in HIV-1 protease are a major challenge to antiretroviral therapy. Protease-substrate interactions that are determined to be critical for native selectivity could serve as robust targets for drug design that are immune to DRMs. In order to identify the structural mechanisms of selectivity, we developed a peptide-docking algorithm to predict the atomic structure of protease-substrate complexes and applied it to a large and diverse set of cleavable and noncleavable peptides. Cleavable peptides showed significantly lower energies of interaction than noncleavable peptides with six protease active-site residues playing the most significant role in discrimination. Surprisingly, all six residues correspond to sequence positions associated with drug resistance mutations, demonstrating that the very residues that are responsible for native substrate specificity in HIV-1 protease are altered during its evolution to drug resistance, suggesting that drug resistance and substrate selectivity may share common mechanisms.

Cysteine 95 and other residues influence the regulatory effects of Histidine 69 mutations on Human Immunodeficiency Virus Type 1 protease autoprocessing

Background

Regulated autoprocessing of HIV Gag-Pol precursor is required for the production of mature and fully active protease. We previously reported that H69E mutation in a pseudo wild type protease sequence significantly (>20-fold) impedes protease maturation in an in vitro autoprocessing assay and in transfected mammalian cells.

Results

Interestingly, H69E mutation in the context of a laboratory adapted NL4-3 protease showed only moderate inhibition (~4-fold) on protease maturation. There are six point mutations (Q7K, L33I, N37S, L63I, C67A, and C95A) between the NL4-3 and the pseudo wild type proteases suggesting that the H69E effect is influenced by other residues. Mutagenesis analyses identified C95 as the primary determinant that dampened the inhibitory effect of H69E. L63 and C67 also demonstrated rescue effect to a less extent. However, the rescue was completely abolished when H69 was replaced by aspartic acid in the NL4-3 backbone. Charge substitutions of surface residues (E21, D30, E34, E35, and F99) to neutral or positively charged amino acids failed to restore protease autoprocessing in the context of H69E mutation.

Conclusions

Taken together, we suggest that residue 69 along with other amino acids such as C95 plus L63 and C67 to a less extent modulate precursor structures for the regulation of protease autoprocessing in the infected cell.

Comparative studies on retroviral proteases: substrate specificity

Exogenous retroviruses are subclassified into seven genera and include viruses that cause diseases in humans. The viral Gag and Gag-Pro-Pol polyproteins are processed by the retroviral protease in the last stage of replication and inhibitors of the HIV-1 protease are widely used in AIDS therapy. Resistant mutations occur in response to the drug therapy introducing residues that are frequently found in the equivalent position of other retroviral proteases. Therefore, besides helping to understand the general and specific features of these enzymes, comparative studies of retroviral proteases may help to understand the mutational capacity of the HIV-1 protease.

How to find simple and accurate rules for viral protease cleavage specificities

Background

Proteases of human pathogens are becoming increasingly important drug targets, hence it is necessary to understand their substrate specificity and to interpret this knowledge in practically useful ways. New methods are being developed that produce large amounts of cleavage information for individual proteases and some have been applied to extract cleavage rules from data. However, the hitherto proposed methods for extracting rules have been neither easy to understand nor very accurate. To be practically useful, cleavage rules should be accurate, compact, and expressed in an easily understandable way.

Results

A new method is presented for producing cleavage rules for viral proteases with seemingly complex cleavage profiles. The method is based on orthogonal search-based rule extraction (OSRE) combined with spectral clustering. It is demonstrated on substrate data sets for human immunodeficiency virus type 1 (HIV-1) protease and hepatitis C (HCV) NS3/4A protease, showing excellent prediction performance for both HIV-1 cleavage and HCV NS3/4A cleavage, agreeing with observed HCV genotype differences. New cleavage rules (consensus sequences) are suggested for HIV-1 and HCV NS3/4A cleavages. The practical usability of the method is also demonstrated by using it to predict the location of an internal cleavage site in the HCV NS3 protease and to correct the location of a previously reported internal cleavage site in the HCV NS3 protease. The method is fast to converge and yields accurate rules, on par with previous results for HIV-1 protease and better than previous state-of-the-art for HCV NS3/4A protease. Moreover, the rules are fewer and simpler than previously obtained with rule extraction methods.

Conclusion

A rule extraction methodology by searching for multivariate low-order predicates yields results that significantly outperform existing rule bases on out-of-sample data, but are more transparent to expert users. The approach yields rules that are easy to use and useful for interpreting experimental data.