High resolution <Superscript>13</Superscript>C-detected solid-state NMR spectroscopy of a deuterated protein

High resolution ¹³C-detected solid-state NMR spectra of the deuterated beta-1 immunoglobulin binding domain of the protein G (GB1) have been collected to show that all ¹⁵N, ¹³C′, ¹³Cα and ¹³Cβ sites are resolved in ¹³C–¹³C and ¹⁵N–¹³C spectra, with significant improvement in T ₂ relaxation times and resolution at high magnetic field (750 MHz). The comparison of echo T ₂ values between deuterated and protonated GB1 at various spinning rates and under different decoupling schemes indicates that ¹³Cα T ₂′ times increase by almost a factor of two upon deuteration at all spinning rates and under moderate decoupling strength, and thus the deuteration enables application of scalar-based correlation experiments that are challenging from the standpoint of transverse relaxation, with moderate proton decoupling. Additionally, deuteration in large proteins is a useful strategy to selectively detect polar residues that are often important for protein function and protein–protein interactions.     

An intraresidual i(HCA)CO(CA)NH experiment for the assignment of main-chain resonances in <Superscript>15</Superscript>N, <Superscript>13</Superscript>C labeled proteins

An improved pulse sequence, intraresidual i(HCA)CO(CA)NH, is described for establishing solely ¹³C′(i), ¹⁵N(i), ¹H^N(i) connectivities in uniformly ¹⁵N/¹³C-labeled proteins. In comparison to the “out-and-back” style intra-HN(CA)CO experiment, the new pulse sequence offers at least two-fold higher experimental resolution in the ¹³C′ dimension and on average 1.6 times higher sensitivity especially for residues in α-helices. Performance of the new experiment was tested on a small globular protein ubiquitin and an intrinsically unfolded 110-residue cancer/testis antigen CT16/PAGE5. Use of intraresidual i(HCA)CO(CA)NH experiment in combination with the established HNCO experiment was crucial for the assignment of highly disordered CT16.     

Residual backbone and side-chain <Superscript>13</Superscript>C and <Superscript>15</Superscript>N resonance assignments of the intrinsic transmembrane light-harvesting 2 protein complex by solid-state Magic Angle Spinning NMR spectroscopy

This study reports the sequence specific chemical shifts assignments for 76 residues of the 94 residues containing monomeric unit of the photosynthetic light-harvesting 2 transmembrane protein complex from Rhodopseudomonas acidophila strain 10050, using Magic Angle Spinning (MAS) NMR in combination with extensive and selective biosynthetic isotope labeling methods. The sequence specific chemical shifts assignment is an essential step for structure determination by MAS NMR. Assignments have been performed on the basis of 2-dimensional proton-driven spin diffusion ¹³C–¹³C correlation experiments with mixing times of 20 and 500 ms and band selective ¹³C–¹⁵N correlation spectroscopy on a series of site-specific biosynthetically labeled samples. The decreased line width and the reduced number of correlation signals of the selectively labeled samples with respect to the uniformly labeled samples enable to resolve the narrowly distributed correlation signals of the backbone carbons and nitrogens involved in the long -helical transmembrane segments. Inter-space correlations between nearby residues and between residues and the labeled BChl a cofactors, provided by the ¹³C–¹³C correlation experiments using a 500 ms spin diffusion period, are used to arrive at sequence specific chemical shift assignments for many residues in the protein complex. In this way it is demonstrated that MAS NMR methods combined with site-specific biosynthetic isotope labeling can be used for sequence specific assignment of the NMR response of transmembrane proteins.     

Structure determination of proteins in <Superscript>2</Superscript>H<Subscript>2</Subscript>O solution aided by a deuterium-decoupled 3D HCA(N)CO experiment

We developed an NMR pulse sequence, 3D HCA(N)CO, to correlate the chemical shifts of protein backbone ¹Hα and ¹³Cα to those of ¹³C′ in the preceding residue. By applying ²H decoupling, the experiment was accomplished with high sensitivity comparable to that of HCA(CO)N. When combined with HCACO, HCAN and HCA(CO)N, the HCA(N)CO sequence allows the sequential assignment using backbone ¹³C′ and amide ¹⁵N chemical shifts without resort to backbone amide protons. This assignment strategy was demonstrated for ¹³C/¹⁵N-labeled GB1 dissolved in ²H₂O. The quality of the GB1 structure determined in ²H₂O was similar to that determined in H₂O in spite of significantly smaller number of NOE correlations. Thus this strategy enables the determination of protein structures in ²H₂O or H₂O at high pH values.     

Selective <Superscript>1</Superscript>H-<Superscript>13</Superscript>C NMR spectroscopy of methyl groups in residually protonated samples of large proteins

Methyl ¹³CHD₂ isotopomers of all methyl-containing amino-acids can be observed in residually protonated samples of large proteins obtained from [U-¹³C,¹H]-glucose/D₂O-based bacterial media, with sensitivity sufficient for a number of NMR applications. Selective detection of some subsets of methyl groups (Ala^β, Thr^γ2) is possible using simple ‘out-and-back’ NMR methodology. Such selective methyl-detected ‘out-and-back’ NMR experiments allow complete assignments of threonine γ2 methyls in residually protonated, [U-¹³C,¹H]-glucose/D₂O-derived samples of an 82-kDa enzyme Malate Synthase G. [U-¹³C,¹H]-glucose/D₂O-derived protein samples are relatively inexpensive and are usually available at very early stages of any NMR study of high-molecular-weight systems.     

Biosynthetic site-specific <Superscript>13</Superscript> C labeling of the light-harvesting 2 protein complex: A model for solid state NMR structure determination of transmembrane proteins

Partly biosynthetic site-directed isotopically ¹³C enriched photosynthetic light-harvesting 2(LH2) complexes have been prepared from Rhodopseudomonas acidophila strain 10050 by using chemically labeled [1,2,3,4–¹³C], [1,4–¹³C] and [2,3–¹³C] succinic acid as a precursor in the growth medium. Two-dimensional proton driven spin diffusion (PDSD) solid state NMR correlation spectroscopy has been used to trace each individual ¹³C isotope from the labeled succinic acid precursor to its destination into the protein and into the embedded major light-absorbing bacteriochlorophyll cofactors. For both the residues of the protein and for the cofactors distinct labeling patterns have been deduced, for protein complexes prepared from [1,4–¹³C]-succinic acid or [2,3–¹³C]-succinic labeled media. All residues, except isoleucine and leucine, have been labeled almost homogeneously by the succinic acid precursor. Carbonyl carbons in the protein backbone were labeled by [1,4–¹³C]-succinic acid, while the C and C carbons of the residues were labeled by [2,3 ¹³C]-succinic acid. Leucine and isoleucine residues were labeled using a uniformly labeled amino acid mixture in the medium. The pattern labeling yields an increase of the resolution and less spectral crowding. The partial labeling technique in combination with conventional solid state NMR methods at ultra high magnetic fields provides an attractive route to resolve chemical shifts for -helical transmembrane protein structures.     

Simultaneous <Superscript>1</Superscript>H- or <Superscript>2</Superscript>H-, <Superscript>15</Superscript>N- and multiple-band-selective <Superscript>13</Superscript>C-decoupling during acquisition in <Superscript>13</Superscript>C-detected experiments with proteins and oligonucleotides

Significant resolution improvement in ¹³C,¹³C-TOCSY spectra of uniformly deuterated and ¹³C, ¹⁵N-labeled protein and ¹³C,¹⁵N-labeled RNA samples is achieved by introduction of multiple-band-selective ¹³C-homodecoupling applied simultaneously with ¹H- or ²H- and ¹⁵N-decoupling at all stages of multidimensional experiments including signal acquisition period. The application of single, double or triple band-selective ¹³C-decoupling in 2D-[¹³C,¹³C]-TOCSY experiments during acquisition strongly simplifies the homonuclear splitting pattern. The technical aspects of complex multiple-band homonuclear decoupling and hardware requirements are discussed. The use of this technique (i) facilitates the resonance assignment process as it reduces signal overlap in homonuclear ¹³C-spectra and (ii) possibly improves the signal-to-noise ratio through multiplet collapse. It can be applied in any ¹³C-detected experiment.     

1-<Superscript>13</Superscript>C amino acid selective labeling in a <Superscript>2</Superscript>H<Superscript>15</Superscript>N background for NMR studies of large proteins

Isotope labeling by residue type (LBRT) has long been an important tool for resonance assignments at the limit where other approaches, such as triple-resonance experiments or NOESY methods do not succeed in yielding complete assignments. While LBRT has become less important for small proteins it can be the method of last resort for completing assignments of the most challenging protein systems. Here we present an approach where LBRT is achieved by adding protonated ¹⁴N amino acids that are ¹³C labeled at the carbonyl position to a medium for uniform deuteration and ¹⁵N labeling. This has three important benefits over conventional ¹⁵N LBRT in a deuterated back ground: (1) selective TROSY-HNCO cross peaks can be observed with high sensitivity for amino-acid pairs connected by the labeling, and the amide proton of the residue following the ¹³C labeled amino acid is very sharp since its alpha position is deuterated, (2) the ¹³C label at the carbonyl position is less prone to scrambling than the ¹⁵N at the α-amino position, and (3) the peaks for the 1-¹³C labeled amino acids can be identified easily from the large intensity reduction in the ¹H-¹⁵N TROSY-HSQC spectrum for some residues that do not significantly scramble nitrogens, such as alanine and tyrosine. This approach is cost effective and has been successfully applied to proteins larger than 40 kDa. Electronic Supplementary Material The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Optimization of amino acid type-specific <Superscript>13</Superscript>C and <Superscript>15</Superscript>N labeling for the backbone assignment of membrane proteins by solution- and solid-state NMR with the UPLABEL algorithm

We present a computational method for finding optimal labeling patterns for the backbone assignment of membrane proteins and other large proteins that cannot be assigned by conventional strategies. Following the approach of Kainosho and Tsuji (Biochemistry 21:6273–6279 (1982)), types of amino acids are labeled with ¹³C or/and ¹⁵N such that cross peaks between ¹³CO(i – 1) and ¹⁵NH(i) result only for pairs of sequentially adjacent amino acids of which the first is labeled with ¹³C and the second with ¹⁵N. In this way, unambiguous sequence-specific assignments can be obtained for unique pairs of amino acids that occur exactly once in the sequence of the protein. To be practical, it is crucial to limit the number of differently labeled protein samples that have to be prepared while obtaining an optimal extent of labeled unique amino acid pairs. Our computer algorithm UPLABEL for optimal unique pair labeling, implemented in the program CYANA and in a standalone program, and also available through a web portal, uses combinatorial optimization to find for a given amino acid sequence labeling patterns that maximize the number of unique pair assignments with a minimal number of differently labeled protein samples. Various auxiliary conditions, including labeled amino acid availability and price, previously known partial assignments, and sequence regions of particular interest can be taken into account when determining optimal amino acid type-specific labeling patterns. The method is illustrated for the assignment of the human G-protein coupled receptor bradykinin B2 (B₂R) and applied as a starting point for the backbone assignment of the membrane protein proteorhodopsin.     

Conformational dependence of <Superscript>13</Superscript>C shielding and coupling constants for methionine methyl groups

Methionine residues fulfill a broad range of roles in protein function related to conformational plasticity, ligand binding, and sensing/mediating the effects of oxidative stress. A high degree of internal mobility, intrinsic detection sensitivity of the methyl group, and low copy number have made methionine labeling a popular approach for NMR investigation of selectively labeled protein macromolecules. However, selective labeling approaches are subject to more limited information content. In order to optimize the information available from such studies, we have performed DFT calculations on model systems to evaluate the conformational dependence of ³ J _CSCC, ³ J _CSCH, and the isotropic shielding, σ_iso. Results have been compared with experimental data reported in the literature, as well as data obtained on [methyl-¹³C]methionine and on model compounds. These studies indicate that relative to oxygen, the presence of the sulfur atom in the coupling pathway results in a significantly smaller coupling constant, ³ J _CSCC/³ J _COCC ~ 0.7. It is further demonstrated that the ³ J _CSCH coupling constant depends primarily on the subtended CSCH dihedral angle, and secondarily on the CSCC dihedral angle. Comparison of theoretical shielding calculations with the experimental shift range of the methyl group for methionine residues in proteins supports the conclusion that the intra-residue conformationally-dependent shift perturbation is the dominant determinant of δ¹³Cε. Analysis of calmodulin data based on these calculations indicates that several residues adopt non-standard rotamers characterized by very large ~100° χ³ values. The utility of the δ¹³Cε as a basis for estimating the gauche/trans ratio for χ³ is evaluated, and physical and technical factors that limit the accuracy of both the NMR and crystallographic analyses are discussed.     

Improved pulse sequences for sequence specific assignment of aromatic proton resonances in proteins

Aromatic proton resonances of proteins are notoriously difficult to assign. Through-bond correlation experiments are preferable over experiments that rely on through-space interactions because they permit aromatic chemical shift assignments to be established independently of the structure determination process. Known experimental schemes involving a magnetization transfer across the C^β–C^γ bond in aromatic side chains either suffer from low efficiency for the relay beyond the C^δ position, use sophisticated ¹³C mixing schemes, require probe heads suitable for application of high ¹³C radio-frequency fields or rely on specialized isotopic labelling patterns. Novel methods are proposed that result in sequential assignment of all aromatic protons in uniformly ¹³C/¹⁵N labelled proteins using standard spectrometer hardware. Pulse sequences consist of routinely used building blocks and are therefore reasonably simple to implement. Ring protons may be correlated with β-carbons and, alternatively, with amide protons (and nitrogens) or carbonyls in order to take advantage of the superior dispersion of backbone resonances. It is possible to record spectra in a non-selective manner, yielding signals of all aromatic residues, or as amino-acid type selective versions to further reduce ambiguities. The new experiments are demonstrated with four different proteins with molecular weights ranging from 11 kDa to 23 kDa. Their performance is compared with that of (Hβ)Cβ(CγCδ)Hδ and (Hβ)Cβ(CγCδCɛ)Hɛ pulse sequences [Yamazaki et al. (1993) J Am Chem Soc 115:11054–11055]. Electronic Supplementary Material The online version of this article (doi: ) contains supplementary material, which is available to authorized users.     

<Superscript>1</Superscript>H, <Superscript>15</Superscript>N and <Superscript>13</Superscript>C sequence specific backbone assignment of the vanadate inhibited hematopoietic tyrosine phosphatase

The sequence-specific backbone assignment of hematopoietic protein tyrosine phosphatase (HePTP; PTPN7) in presence of vanadate has been determined, based on triple-resonance experiments using uniformly [¹³C,¹⁵N]-labeled protein. These assignments facilitate further studies of HePTP in the presence of inhibitors to target leukemia and provide further insights into the function of protein tyrosine phosphatases.     

<Superscript>13</Superscript>C-detected NMR experiments for measuring chemical shifts and coupling constants in nucleic acid bases

The paper presents a set of two-dimensional experiments that utilize direct ¹³C detection to provide proton–carbon, carbon–carbon and carbon–nitrogen correlations in the bases of nucleic acids. The set includes a ¹³C-detected proton–carbon correlation experiment for the measurement of ¹³C–¹³C couplings, the CaCb experiment for correlating two quaternary carbons, the HCaCb experiment for the ¹³C–¹³C correlations in cases where one of the carbons has a proton attached, the HCC-TOCSY experiment for correlating a proton with a network of coupled carbons, and a ¹³C-detected ¹³C–¹⁵N correlation experiment for detecting the nitrogen nuclei that cannot be detected via protons. The IPAP procedure is used for extracting the carbon–carbon couplings and/or carbon decoupling in the direct dimension, while the S³E procedure is preferred in the indirect dimension of the carbon–nitrogen experiment to obtain the value of the coupling constant. The experiments supply accurate values of ¹³C and ¹⁵N chemical shifts and carbon–carbon and carbon–nitrogen coupling constants. These values can help to reveal structural features of nucleic acids either directly or via induced changes when the sample is dissolved in oriented media. Electronic supplementary material The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

CSSI-PRO: a method for secondary structure type editing,assignment and estimation in proteins using linear combination of backbone chemical shifts

Estimation of secondary structure in polypeptides is important for studying their structure, folding and dynamics. In NMR spectroscopy, such information is generally obtained after sequence specific resonance assignments are completed. We present here a new methodology for assignment of secondary structure type to spin systems in proteins directly from NMR spectra, without prior knowledge of resonance assignments. The methodology, named Combination of Shifts for Secondary Structure Identification in Proteins (CSSI-PRO), involves detection of specific linear combination of backbone ¹H^α and ¹³C′ chemical shifts in a two-dimensional (2D) NMR experiment based on G-matrix Fourier transform (GFT) NMR spectroscopy. Such linear combinations of shifts facilitate editing of residues belonging to α-helical/β-strand regions into distinct spectral regions nearly independent of the amino acid type, thereby allowing the estimation of overall secondary structure content of the protein. Comparison of the predicted secondary structure content with those estimated based on their respective 3D structures and/or the method of Chemical Shift Index for 237 proteins gives a correlation of more than 90% and an overall rmsd of 7.0%, which is comparable to other biophysical techniques used for structural characterization of proteins. Taken together, this methodology has a wide range of applications in NMR spectroscopy such as rapid protein structure determination, monitoring conformational changes in protein-folding/ligand-binding studies and automated resonance assignment. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Glutarate and <Emphasis Type="Italic">N</Emphasis>-acetyl-<Emphasis Type="SmallCaps">l</Emphasis>-glutamate buffers for cell-free synthesis of selectively <Superscript>15</Superscript>N-labelled proteins

Cell-free protein synthesis provides rapid and economical access to selectively ¹⁵N-labelled proteins, greatly facilitating the assignment of ¹⁵N-HSQC spectra. While the best yields are usually obtained with buffers containing high concentrations of potassium l-glutamate, preparation of selectively ¹⁵N-Glu labelled samples requires non-standard conditions. Among many compounds tested to replace the l-Glu buffer, potassium N-acetyl-l-glutamate and potassium glutarate were found to perform best, delivering high yields for all proteins tested, with preserved selectivity of ¹⁵N-Glu labelling. Assessment of amino-transferase activity by combinatorial ¹⁵N-labelling revealed that glutarate and N-acetyl-l-glutamate suppress the transfer of the ¹⁵N-α-amino groups between amino acids less well than the conventional l-Glu buffer. On balance, the glutarate buffer appears most suitable for the preparation of samples containing ¹⁵N-l-Glu while the conventional l-Glu buffer is advantageous for all other samples. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Refinement of protein structure against non-redundant carbonyl <Superscript>13</Superscript>C NMR relaxation

Carbonyl ¹³C′ relaxation is dominated by the contribution from the ¹³C′ chemical shift anisotropy (CSA). The relaxation rates provide useful and non-redundant structural information in addition to dynamic parameters. It is straightforward to acquire, and offers complimentary structural information to the ¹⁵N relaxation data. Furthermore, the non-axial nature of the ¹³C′ CSA tensor results in a T₁/T₂ value that depends on an additional angular variable even when the diffusion tensor of the protein molecule is axially symmetric. This dependence on an extra degree of freedom provides new geometrical information that is not available from the NH dipolar relaxation. A protocol that incorporates such structural restraints into NMR structure calculation was developed within the program Xplor-NIH. Its application was illustrated with the yeast Fis1 NMR structure. Refinement against the ¹³C′ T₁/T₂ improved the overall quality of the structure, as evaluated by cross-validation against the residual dipolar coupling as well as the ¹⁵N relaxation data. In addition, possible variations of the CSA tensor were addressed. Electronic Supplementary Material The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

13C, 15N Resonance Assignment of Parts of the HET-s Prion Protein in its Amyloid Form

The partial ¹⁵N and ¹³C solid-state NMR resonance assignment of the HET-s prion protein fragment 218–289 in its amyloid form is presented. It is based on experiments measured at MAS frequencies in the range of 20–40 kHz using exclusively adiabatic polarization-transfer schemes. The resonance assignment within each residue is based on two-dimensional ¹³C––¹³C correlation spectra utilizing the DREAM mixing scheme. The sequential linking of the assigned residues used a set of two- and three-dimensional ¹⁵N––¹³C correlation experiments. Almost all cross peaks visible in the spectra are assigned, but only resonances from 43 of the 78 amino-acid residues could be detected. The missing residues are thought to be highly disordered and/or highly dynamic giving rise to broad resonance lines that escaped detection in the experiments applied. The line widths of the observed resonances are narrow and comparable to line widths observed in micro-crystalline samples. The 43 assigned residues are located in two fragments of about 20 residues. Electronic Supplementary Material Supplementary material is available for this article at and is accessible for authorized users.     

Empirical correlation between protein backbone <Superscript>15</Superscript>N and <Superscript>13</Superscript>C secondary chemical shifts and its application to nitrogen chemical shift re-referencing

The linear analysis of chemical shifts (LACS) has provided a robust method for identifying and correcting ¹³C chemical shift referencing problems in data from protein NMR spectroscopy. Unlike other approaches, LACS does not require prior knowledge of the three-dimensional structure or inference of the secondary structure of the protein. It also does not require extensive assignment of the NMR data. We report here a way of extending the LACS approach to ¹⁵N NMR data from proteins, so as to enable the detection and correction of inconsistencies in chemical shift referencing for this nucleus. The approach is based on our finding that the secondary ¹⁵N chemical shift of the backbone nitrogen atom of residue i is strongly correlated with the secondary chemical shift difference (experimental minus random coil) between the alpha and beta carbons of residue i − 1. Thus once alpha and beta ¹³C chemical shifts are available (their difference is referencing error-free), the ¹⁵N referencing can be validated, and an appropriate offset correction can be derived. This approach can be implemented prior to a structure determination and can be used to analyze potential referencing problems in database data not associated with three-dimensional structure. Application of the LACS algorithm to the current BMRB protein chemical shift database, revealed that nearly 35% of the BMRB entries have δ ¹⁵N values mis-referenced by over 0.7 ppm and over 25% of them have δ ¹H^N values mis-referenced by over 0.12 ppm. One implication of the findings reported here is that a backbone ¹⁵N chemical shift provides a better indicator of the conformation of the preceding residue than of the residue itself. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

HA-detected experiments for the backbone assignment of intrinsically disordered proteins

We propose a new alpha proton detection based approach for the sequential assignment of natively unfolded proteins. The proposed protocol superimposes on following features: HA-detection (1) enables assignment of natively unfolded proteins at any pH, i.e., it is not sensitive to rapid chemical exchange undergoing in natively unfolded proteins even at moderately high pH. (2) It allows straightforward assignment of proline-rich polypeptides without additional proline-customized experiments. (3) It offers more streamlined and less ambiguous assignment based on solely intraresidual ¹⁵N(i)-¹³C′(i)-H^α(i) (or ¹⁵N(i)-¹³C^α(i)-H^α(i)) and sequential ¹⁵N(i + 1)-¹³C′(i)-H^α(i) (or ¹⁵N(i + 1)-¹³C^α(i)-H^α(i)) correlation experiments together with efficient use of chemical shifts of ¹⁵N and ¹³C′ nuclei, which show smaller dependence on residue type. We have tested the proposed protocol on two proteins, small globular 56-residue GB1, and highly disordered, proline-rich 47-residue fifth repeat of EspF_U. Using the proposed approach, we were able to assign 90% of ¹H^α, ¹³C^α, ¹³C′, ¹⁵N chemical shifts in EspF_U. We reckon that the HA-detection based strategy will be very useful in the assignment of natively unfolded proline-rich proteins or polypeptide chains.     

CACA-TOCSY with alternate <Superscript>13</Superscript>C–<Superscript>12</Superscript>C labeling: a <Superscript>13</Superscript>C<Superscript>α</Superscript> direct detection experiment for mainchain resonance assignment,dihedral angle information,and amino acid type identification

We present a ¹³C direct detection CACA-TOCSY experiment for samples with alternate ¹³C–¹²C labeling. It provides inter-residue correlations between ¹³C^α resonances of residue i and adjacent C^αs at positions i − 1 and i + 1. Furthermore, longer mixing times yield correlations to C^α nuclei separated by more than one residue. The experiment also provides C^α-to-sidechain correlations, some amino acid type identifications and estimates for ψ dihedral angles. The power of the experiment derives from the alternate ¹³C–¹²C labeling with [1,3-¹³C] glycerol or [2-¹³C] glycerol, which allows utilizing the small scalar ³J_CC couplings that are masked by strong ¹J_CC couplings in uniformly ¹³C labeled samples.