Food Science & Technology · Postgraduate & Research · also Advanced Food Analysis, Foodomics, Instrumental Analysis (PG)
The analytical science of a research laboratory: high-resolution mass spectrometry, NMR, advanced chromatography, imaging and hyperspectral methods, proteomics, metabolomics and lipidomics of foods, biosensors, and the chemometrics that interprets the data.
In the exams: ICAR AICE 35.a Unit 4 (non-destructive methods, biosensors) and 35.b Unit 6 (research techniques); PhD research
Modern food research and control depend on instruments that can identify thousands of compounds in a sample, measure contaminants at micrograms per kilogram, see the structure of a fat crystal or a protein network, and turn complex signals into decisions about authenticity, safety and quality. Routine food analysis, taught in food-analysis, measures defined constituents by established methods; the advanced techniques of a research or reference laboratory ask broader questions: what is in this sample that should not be, where did it come from, how did processing change its molecules and structure, and which of thousands of features distinguish one product from another.
This subject covers high-resolution and tandem mass spectrometry; LC-MS/MS and GC-MS/MS for residues and contaminants; nuclear magnetic resonance; advanced chromatography, including UHPLC, two-dimensional and ion chromatography; thermal analysis by DSC and TGA; microscopy of foods by SEM, TEM, confocal and atomic force microscopy; X-ray diffraction and scattering; hyperspectral imaging and machine vision; proteomics, peptidomics and the detection of allergens; metabolomics and lipidomics, together often called foodomics; biosensors and electronic noses and tongues; and the chemometrics and machine learning used to interpret the data. Each chapter explains the principle, what the technique measures, its strengths and limits, and typical food applications.
These techniques are examined in research-fellowship examinations in food safety and quality, which include non-destructive methods, biosensors and research techniques, and they are the working tools of doctoral research and of reference and enforcement laboratories. The practice questions combine principles with the short calculations that analysts use: mass accuracy and resolving power, chromatographic resolution and plate number, matrix effects, Bragg spacings, chemical shifts, and the figures of merit of calibration models.
A mass spectrometer ionises molecules, separates the ions by their mass-to-charge ratio (m/z) and detects them. The ionisation method decides which compounds can be analysed. Electron ionisation (EI), used with gas chromatography, bombards gas-phase molecules with electrons, conventionally at 70 eV, giving reproducible fragmentation patterns that can be matched against large spectral libraries; it suits volatile, thermally stable compounds such as flavour volatiles, many pesticides and fatty acid methyl esters. Electrospray ionisation (ESI), the main interface for liquid chromatography, forms ions from charged droplets at atmospheric pressure and is a soft technique, giving mainly protonated [M+H]⁺ or deprotonated [M−H]⁻ molecules and adducts such as [M+Na]⁺ and [M+NH₄]⁺; it suits polar and ionisable compounds, including veterinary drugs, mycotoxins, peptides and many metabolites. Atmospheric-pressure chemical ionisation and photoionisation suit less polar compounds, and matrix-assisted laser desorption/ionisation (MALDI) is used for proteins, peptides and imaging.
Analysers differ in resolution, mass accuracy, speed and sensitivity. Quadrupoles filter ions by m/z using oscillating electric fields and are robust and sensitive at unit resolution; ion traps store and fragment ions; time-of-flight (TOF) analysers measure the flight time of ions in a field-free tube; Orbitrap analysers measure the frequency of ions orbiting a central electrode; and Fourier-transform ion cyclotron resonance instruments give the highest resolution. Resolving power is commonly defined as R = m/Δm, where Δm is the peak width at half height; an instrument with R = 70 000 at m/z 200 separates peaks about 0.003 apart. Mass accuracy is expressed in parts per million: (measured − theoretical)/theoretical × 10⁶. A measured m/z of 303.0505 for a theoretical 303.0499 is an error of about 2 ppm. High-resolution mass spectrometry (HRMS), with resolving powers in the tens of thousands or more and errors of a few ppm, gives elemental formulas, separates isobaric interferences and allows retrospective screening of data for compounds not targeted at the time of acquisition.
Tandem mass spectrometry (MS/MS) selects a precursor ion, fragments it, usually by collision-induced dissociation with an inert gas, and analyses the product ions. In a triple quadrupole, the first quadrupole selects the precursor, the second acts as a collision cell and the third selects a product ion; monitoring several precursor-product pairs (transitions) in multiple reaction monitoring (MRM) mode gives very high selectivity and sensitivity for target analytes, the basis of most quantitative residue methods. Other scan modes include product-ion scans (all fragments of one precursor, for identification), precursor-ion scans (all precursors that give a chosen fragment) and neutral-loss scans (all precursors that lose a chosen neutral fragment), useful for screening classes of compounds. Hybrid instruments, such as quadrupole-TOF and quadrupole-Orbitrap, combine selection with high-resolution detection of products. In data-dependent acquisition the instrument fragments the most intense ions it detects; in data-independent acquisition it fragments all ions in successive windows, giving more complete but more complex data. Ion mobility spectrometry adds a separation by size and shape (collision cross-section) between chromatography and mass analysis.
Interpreting spectra uses several clues. The monoisotopic mass and its accuracy constrain the elemental formula; the isotope pattern adds information, as chlorine gives M and M+2 peaks in a ratio of about 3:1 and bromine about 1:1, and the M+1 peak indicates the number of carbon atoms (about 1.1 per cent of the M peak per carbon); fragmentation pathways and comparison with reference standards or library spectra confirm structures. For example, an ion at m/z 23 units above the protonated molecule suggests a sodium adduct, since [M+Na]⁺ is 21.98 higher than [M+H]⁺. Confirmation of identity for official purposes requires agreement of retention time and ion ratios with a reference standard (chapter 2), because exact mass alone can match several isomers.
Residues and contaminants in food, such as pesticides, veterinary drugs, mycotoxins, plant toxins, process contaminants and persistent organic pollutants, must be measured at very low concentrations, often in the microgram per kilogram range, in complex matrices. Multi-residue methods measure hundreds of analytes in one run. The QuEChERS approach (quick, easy, cheap, effective, rugged and safe), widely used for pesticides, extracts a homogenised sample with acetonitrile, induces phase separation by adding salts (magnesium sulphate and sodium chloride, often with buffering citrate or acetate), and cleans the extract by dispersive solid-phase extraction with sorbents chosen for the matrix: primary-secondary amine to remove organic acids and sugars, C18 to remove fats, and graphitised carbon black to remove pigments such as chlorophyll (with caution, as it also retains planar pesticides). Other approaches include solid-phase extraction cartridges, immunoaffinity columns for mycotoxins, and dilute-and-shoot methods for clean matrices.
In electrospray ionisation, co-eluting matrix components compete with analytes for charge and alter droplet evaporation, causing ion suppression (or sometimes enhancement), the matrix effect. It is quantified by comparing the slope of a calibration in matrix extract with that in solvent: matrix effect (%) = (slope_matrix/slope_solvent − 1) × 100; a value of −30 per cent means 30 per cent suppression. Matrix effects are compensated by matrix-matched calibration (standards prepared in blank extract of the same commodity), by isotope-labelled internal standards (which co-elute and are suppressed equally), by standard addition, or by dilution of extracts, which modern sensitive instruments allow. Gas chromatography has its own matrix effects, often enhancement, because matrix components block active sites in the injector that would otherwise adsorb or degrade analytes; analyte protectants or matrix-matched standards compensate.
Methods for official control must be validated and must meet criteria for identification and quantification. Guidance used in the European Union for pesticide residues, for example, requires for identification by low-resolution tandem MS at least two product ions (transitions), retention times matching those of standards within a set tolerance (such as ±0.1 minute), and ion ratios within a tolerance of the reference (such as ±30 per cent relative); for HRMS, accurate mass of the molecular ion and at least one fragment within a few ppm. Quantification uses calibration over the working range, recoveries usually expected within about 70 to 120 per cent with relative standard deviations up to about 20 per cent, and limits of quantification below the relevant maximum residue limits. Many jurisdictions apply a default maximum residue limit, commonly 0.01 mg/kg, for pesticide-commodity combinations without a specific limit, which drives the sensitivity required.
LC-MS/MS is the main technique for polar and thermally labile analytes: most modern pesticides, veterinary drugs (antibiotics, anthelmintics, coccidiostats), mycotoxins (aflatoxins, ochratoxin A, deoxynivalenol, fumonisins, zearalenone, often in one multi-mycotoxin method), plant toxins such as pyrrolizidine and tropane alkaloids, marine biotoxins, perfluoroalkyl substances and process contaminants such as acrylamide. GC-MS/MS remains essential for volatile and non-polar analytes: organochlorine and many other pesticides, polycyclic aromatic hydrocarbons, polychlorinated biphenyls, chloropropanols and their esters (after derivatisation), and furan. Dioxins and dioxin-like compounds are measured by high-resolution GC with high-resolution MS or by GC-MS/MS under strict criteria, using isotope dilution with labelled congeners. Results near legal limits are reported with measurement uncertainty and judged by stated decision rules (see quality-assurance-lab-accreditation). Non-targeted screening with HRMS complements targeted methods by searching data for unexpected contaminants, and suspect screening matches data against lists of possible compounds.
Nuclear magnetic resonance spectroscopy exploits the magnetic properties of nuclei with non-zero spin, chiefly ¹H and ¹³C, but also ³¹P, ¹⁹F and ²H. In a strong magnetic field, such nuclei have energy levels separated by an amount proportional to the field; radiofrequency pulses excite transitions, and the nuclei emit signals at their resonance (Larmor) frequencies as they relax. Instruments are described by the ¹H frequency, such as 400 or 600 MHz. The exact frequency depends on the chemical environment of each nucleus, so signals are spread over a chemical-shift scale, δ (ppm) = (ν − ν_ref)/ν₀ × 10⁶, relative to a reference such as tetramethylsilane (TMS) at 0 ppm. On a 400 MHz instrument, a signal 1600 Hz from TMS appears at 4.0 ppm. Typical ¹H shifts are about 0.9 ppm for methyl groups of fatty acids, 5.3 ppm for olefinic protons, 9 to 10 ppm for aldehydes and about 4.7 ppm for water; spin-spin coupling splits signals into multiplets that reveal neighbouring nuclei. Two-dimensional experiments, such as COSY (proton-proton couplings), HSQC (protons attached to carbons) and HMBC (long-range proton-carbon correlations), assign structures.
NMR has three properties that make it valuable in food analysis. It is quantitative, because the area of a signal is directly proportional to the number of nuclei that produce it, so quantitative NMR (qNMR) with an internal standard of known purity measures concentrations without calibration curves for each analyte. It is non-destructive and highly reproducible between instruments, so spectral databases can be shared. And it sees all soluble compounds above its detection limit at once, which makes it useful for fingerprinting. Its main limitation is sensitivity, far lower than mass spectrometry, so it detects major and minor constituents (typically micromolar and above) rather than trace contaminants.
NMR fingerprinting is used to authenticate foods: spectra of honey, fruit juices, wine, olive oil, coffee and milk are compared with databases of authentic samples by multivariate statistics to detect adulteration, false origin or false variety. Site-specific natural isotope fractionation NMR (SNIF-NMR) measures the deuterium content at specific positions in a molecule; the deuterium distribution in ethanol from wine reveals whether sugar from beet or cane was added before fermentation (chaptalisation beyond legal limits) and the botanical origin of the sugar, and the method is an official reference method for wine and is used for vanillin and other flavours to distinguish natural from synthetic or biosynthetic sources, alongside isotope-ratio mass spectrometry.
Low-field or time-domain NMR uses small benchtop instruments with permanent magnets and measures relaxation rather than chemical shifts. The relaxation times T₁ (spin-lattice) and T₂ (spin-spin) differ between protons in solids, liquids and water in different environments: solid fat gives a rapidly decaying signal and liquid oil a slowly decaying one, so time-domain NMR is the standard method for measuring solid fat content in fats and chocolate as a function of temperature; it also measures oil and moisture in seeds and powders rapidly, and T₂ distributions show how water is bound or mobile in meat, bread, cheese and gels. Magnetic resonance imaging (MRI) builds images from NMR signals and shows, non-destructively, internal structures and the distribution of water and fat, for example in fruit with internal defects, during drying or freezing, and in cheese ripening. NMR is also used to study the structure of food proteins, starch and fibres, and in metabolomics (chapter 10).
Chromatographic separation is described by a few quantities. Retention is measured by the retention time t_R or the retention factor k = (t_R − t₀)/t₀, where t₀ is the hold-up time. Efficiency is the plate number, N = 16(t_R/w)² using the baseline peak width w, or N = 5.54(t_R/w₁/₂)² using the width at half height; for a peak at 10.0 minutes with a baseline width of 0.20 minutes, N = 16 × 2500 = 40 000. The plate height H = L/N relates efficiency to column length L. Resolution of two peaks is R_s = 2(t₂ − t₁)/(w₁ + w₂); R_s = 1.5 corresponds to baseline separation of equal Gaussian peaks. The van Deemter equation, H = A + B/u + C·u, describes how plate height depends on mobile-phase velocity u through eddy diffusion (A), longitudinal diffusion (B) and resistance to mass transfer (C); smaller particles reduce A and C, giving lower plate heights and a flatter curve, so separations can run faster without losing efficiency.
Ultra-high-performance liquid chromatography (UHPLC) uses columns packed with particles below 2 µm, or superficially porous (core-shell) particles, and instruments able to work at pressures up to about 1000 to 1300 bar, giving sharper peaks, higher resolution and shorter run times than conventional HPLC, with lower solvent use; it is the standard front end for LC-MS. Hydrophilic interaction chromatography (HILIC) retains polar compounds, such as sugars, amino acids, nucleotides and polar pesticides, that pass unretained through reversed-phase columns. Chiral stationary phases separate enantiomers (see chemistry-for-food-science), and supercritical fluid chromatography with carbon dioxide mobile phases separates lipids and chiral compounds quickly with little organic solvent.
Complex food samples, such as essential oils, roasted coffee aroma, petroleum-derived mineral oil contamination or protein hydrolysates, can contain more compounds than one column can separate. Two-dimensional chromatography couples two columns with different separation mechanisms. In heart-cutting two-dimensional chromatography, selected fractions from the first column are transferred to the second. In comprehensive two-dimensional chromatography (GC×GC or LC×LC), the whole effluent of the first column is sampled continuously in short fractions, by a modulator, onto a fast second column; the peak capacity is approximately the product of the peak capacities of the two dimensions, so GC×GC with a non-polar first and polar second column can resolve thousands of volatile compounds and displays structured patterns in which chemical classes fall in bands. GC×GC with time-of-flight MS is used for aroma profiling, for the analysis of mineral oil saturated and aromatic hydrocarbons in foods (often preceded by online LC-GC), and for contaminant screening. Multidimensional LC is used in proteomics and polyphenol analysis.
Ion chromatography separates ions on ion-exchange columns and usually detects them by conductivity, with a suppressor that removes the conductivity of the eluent so that analyte ions are detected sensitively. It measures inorganic anions (chloride, nitrate, nitrite, sulphate, phosphate, bromate and perchlorate in water and foods), cations (sodium, potassium, calcium, magnesium, ammonium) and organic acids. High-performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD) separates carbohydrates, which are weak acids at high pH, without derivatisation, and is the reference technique for sugar profiles, oligosaccharides such as fructans and galacto-oligosaccharides, and authenticity of honeys and juices. Coupling ion chromatography to mass spectrometry or inductively coupled plasma MS adds identification and element-specific detection, for example for speciation of arsenic.
Differential scanning calorimetry (DSC) measures the heat flow into or out of a sample, relative to a reference, as it is heated or cooled at a controlled rate, typically 1 to 10 °C per minute. Transitions appear as peaks or steps. Endothermic peaks (heat absorbed) show melting of fats, ice and sugars, starch gelatinisation, protein denaturation and evaporation; exothermic peaks (heat released) show crystallisation, for example of fats on cooling or of amorphous sugars on heating above their glass transition; and a step change in heat capacity, without a peak, shows the glass transition (T_g), at which an amorphous material changes from a glassy to a rubbery state. The area of a peak gives the enthalpy of the transition: if a 10 mg sample of a fat shows a melting peak with an area corresponding to 1.0 J, its melting enthalpy is 100 J/g. Onset, peak and end temperatures characterise the range.
DSC has many food applications. For fats and chocolate, melting profiles and polymorphic forms (different crystal forms melt at different temperatures; cocoa butter's form V melts at about 33 to 34 °C) guide tempering and blending. For starch, gelatinisation temperatures and enthalpies show the effects of variety, water content, sugar and salt, and a melting peak of retrogradated amylopectin, typically around 45 to 60 °C, measures staling in bread and starch gels. For proteins, denaturation temperatures and enthalpies indicate thermal stability and the extent of denaturation by processing; for example, the remaining denaturation enthalpy of whey proteins shows how much has already been denatured. For frozen and dried foods, the glass transition and the freezing behaviour of the freeze-concentrated phase (T_g′) underpin the state diagrams used in freeze-drying and frozen storage (see shelf-life-stability). Modulated DSC superimposes a sinusoidal temperature oscillation on the ramp, separating reversing events such as the glass transition from non-reversing events such as relaxation and evaporation, which often overlap in foods.
Thermogravimetric analysis (TGA) records the mass of a sample as it is heated in a controlled atmosphere. Steps in the curve show loss of free and bound water, release of volatiles, and thermal decomposition of carbohydrates, proteins and fats at higher temperatures, with the residue at high temperature in air corresponding to ash. The derivative curve (DTG) shows the rate of mass loss and separates overlapping steps. TGA is used to determine moisture and volatile content, to study the thermal stability of ingredients, packaging polymers, encapsulated flavours and pigments, and to characterise by-products and biomass for energy use; coupled with infrared spectroscopy or mass spectrometry of the evolved gases, it identifies what is lost at each step. For example, a powder that loses 4.5 per cent of its mass up to 120 °C under nitrogen has a moisture and volatile content of about 4.5 per cent.
Dynamic mechanical analysis (DMA) applies a small oscillating stress or strain to a solid or semi-solid sample over a temperature range and measures the storage modulus (elastic response), the loss modulus (viscous response) and their ratio, tan δ. The glass transition appears as a large fall in storage modulus and a peak in tan δ, often detected more sensitively than by DSC, which makes DMA useful for low-moisture foods such as biscuits, cereals and confectionery, and for edible films and packaging. Combined with sorption isotherms and DSC, these methods give the state diagrams that relate water content, temperature and the physical state of foods: the Gordon-Taylor equation, for example, predicts how water, a plasticiser, lowers the glass transition of a carbohydrate mixture, explaining why crisp foods become soggy when they absorb moisture. Instruments must be calibrated for temperature and enthalpy with standards such as indium, samples sealed appropriately (hermetic pans for moist samples), and scanning rates stated, because transition temperatures shift with heating rate.
Food texture, stability and digestion depend on structure at scales from millimetres to nanometres, and microscopy makes that structure visible. The resolution of a light microscope is limited by diffraction to about half the wavelength of light, given by the Abbe limit d = λ/(2NA), where NA is the numerical aperture of the objective: with green light of 550 nm and an oil-immersion objective of NA 1.4, d ≈ 200 nm. Bright-field microscopy with stains shows cells, starch granules, fat globules and protein networks: iodine stains starch blue-black, and specific dyes stain fat and protein. Polarised-light microscopy shows birefringent, ordered structures: native starch granules show a dark 'Maltese cross' under crossed polarisers, which disappears when they gelatinise, giving a simple test of gelatinisation; fat crystals and ice crystals are also birefringent.
Confocal laser scanning microscopy (CLSM) uses a focused laser and a pinhole to reject out-of-focus light, giving sharp optical sections through intact, often fully hydrated samples, which can be stacked into three-dimensional images. With fluorescent probes, several components are seen in one image: fat with Nile red or Bodipy, protein with fast green or fluorescein isothiocyanate, polysaccharides with labelled lectins or calcofluor, and live or dead bacteria with nucleic-acid stains. CLSM shows the microstructure of emulsions, yoghurt and cheese networks, bread crumb, meat products and biofilms with little sample preparation, and time-lapse imaging follows changes during processing or digestion.
Electron microscopes use electron beams, whose short wavelengths give resolutions of nanometres or better. Scanning electron microscopy (SEM) scans a focused beam over the sample surface and detects emitted secondary electrons, giving images of surface topography with great depth of field: the surfaces of starch granules, powder particles, fractured cheese or chocolate, crystal habits and microbial cells. Conventional SEM needs dry, conductive samples, so foods are fixed, dehydrated and coated with a thin layer of gold or carbon, which can introduce artefacts; cryo-SEM examines rapidly frozen, fractured samples, preserving water and fat, and environmental SEM works with hydrated samples at low vacuum. Energy-dispersive X-ray spectroscopy in an SEM maps elements, for example to identify a metal or mineral foreign body. Transmission electron microscopy (TEM) passes electrons through ultrathin sections (often tens of nanometres) or negatively stained or frozen-hydrated particles and shows internal ultrastructure at near-molecular resolution: casein micelles and their changes on heating or acidification, fat globule membranes, protein aggregates, and the structure of nanoparticles used in encapsulation.
Atomic force microscopy (AFM) scans a sharp tip on a flexible cantilever across a surface and records its deflection, building a three-dimensional topographic image with nanometre resolution, in air or liquid and without staining or coating. It shows the conformation of individual polysaccharide molecules such as pectin and xanthan, protein films at interfaces, starch granule surfaces and the nanostructure of fat crystals. In force spectroscopy mode, AFM measures forces between the tip and the sample, giving elasticity and adhesion at the nanoscale. Each microscopy method has artefacts from sample preparation and imaging conditions, so structures are confirmed by more than one method, and image analysis software quantifies features such as droplet size, porosity and network connectivity rather than relying on visual impression.
When X-rays strike a crystalline material, waves scattered by regularly spaced planes of atoms interfere constructively at angles given by Bragg's law, nλ = 2d sin θ, where λ is the X-ray wavelength, d the spacing between planes, θ the angle between the incident beam and the planes, and n an integer. Laboratory diffractometers commonly use copper Kα radiation with λ = 0.154 nm. A diffraction peak at 2θ = 20° (θ = 10°) corresponds to d = 0.154/(2 × sin 10°) = 0.443 nm. Powder X-ray diffraction of food materials gives patterns of peaks from crystalline regions superimposed on a broad halo from amorphous regions; the ratio of crystalline peak area to total area estimates the degree of crystallinity.
Starch granules give characteristic patterns. The A-type pattern, from tightly packed double helices of amylopectin, is typical of cereal starches such as wheat, maize and rice; the B-type pattern, with a more open, hydrated packing, is typical of tuber starches such as potato, of high-amylose starches and of retrograded starch; the C-type pattern, a mixture of A and B, is found in many legume starches; and the V-type pattern arises from single helices of amylose complexed with lipids or other ligands, formed during processing such as extrusion and cooking in the presence of fats. XRD therefore follows gelatinisation (loss of crystallinity), retrogradation and staling (development of B-type crystallinity) and the formation of amylose-lipid complexes, which affect digestibility and texture. It also identifies crystalline forms of sugars, such as α-lactose monohydrate and anhydrous forms, and of salts, and distinguishes crystalline from amorphous powders, which differ in hygroscopicity and caking.
Fats crystallise in several polymorphic forms, distinguished by the packing of the fatty acid chains, which XRD identifies from short-spacing peaks at wide angles. The α form, with hexagonal packing, shows a single strong short spacing near 0.415 nm; the β′ form, with orthorhombic packing, shows two strong spacings near 0.42 and 0.38 nm; and the β form, with triclinic packing, the most stable and highest melting, shows a strong spacing near 0.46 nm. The long spacings, at small angles, reflect the lamellar stacking of triacylglycerol molecules in double or triple chain-length arrangements. The β′ form gives the small crystals and smooth texture wanted in margarines and shortenings, while transformation to β gives large crystals and graininess; in chocolate, cocoa butter must be tempered to form V, a β form, and its transformation to form VI during storage contributes to fat bloom. XRD, often combined with DSC, is the standard way to identify these forms and to follow transformations during processing and storage; synchrotron sources give intense beams that follow crystallisation in real time.
Small-angle X-ray scattering (SAXS) measures scattering at very small angles, which arises from structures much larger than interatomic distances, roughly 1 to 100 nm: the size and internal structure of casein micelles, the lamellar organisation of fat crystals and starch granules (the 9 nm repeat of crystalline and amorphous lamellae in starch), protein aggregates, emulsion droplets and the mesophases formed by emulsifiers. Wide-angle X-ray scattering (WAXS) covers the atomic-scale range of conventional diffraction, and combined SAXS/WAXS describes structure across scales. Small-angle neutron scattering, at large facilities, uses contrast variation with heavy water to highlight particular components. X-ray computed tomography, a related imaging technique, reconstructs three-dimensional images of internal structure, such as the pore network of bread, cereals and foams. X-ray fluorescence, by contrast, measures the characteristic X-rays emitted by elements and is used for elemental analysis rather than structure.
Machine vision systems capture images of food with cameras and analyse them automatically, replacing or supporting human inspection on processing lines. A system includes illumination designed to show the features of interest (diffuse lighting to avoid glare, backlighting for silhouettes, structured light for three-dimensional shape), cameras (colour RGB, monochrome, line-scan for products on conveyors, or 3-D cameras), and software that processes images. Colour measured in RGB is often converted to device-independent colour spaces such as CIE L*a*b*, in which L* is lightness, a* the red-green axis and b* the yellow-blue axis, so that colour can be related to quality attributes such as ripeness, baking degree or meat colour. Processing steps include segmentation (separating objects from background, for example by thresholding), feature extraction (size, shape, colour, texture), and classification against grading rules or trained models.
Machine vision grades fruit and vegetables by size, shape, colour and surface defects at rates of many items per second; sorts grains, nuts and pulses; checks fill levels, label position, cap closure and date codes; measures the colour and size of biscuits and the browning of fried products; detects foreign bodies with contrasting colour; and estimates fat cover on meat cuts. Deep learning, especially convolutional neural networks trained on large sets of labelled images, now outperforms rule-based methods for complex tasks such as detecting subtle defects or classifying varieties, but it needs representative training images, careful validation and monitoring for changes in products or lighting.
Hyperspectral imaging combines imaging with spectroscopy: for every pixel of an image it records a full spectrum, typically of a hundred or more contiguous wavelength bands, producing a three-dimensional data cube (two spatial dimensions and one spectral dimension, the hypercube). Visible and near-infrared systems cover about 400 to 1000 nm, and short-wave infrared systems about 1000 to 2500 nm, where overtone and combination absorptions of O–H, C–H and N–H bonds relate to water, fat, protein and carbohydrates. Line-scan (push-broom) cameras record one line of pixels with all wavelengths at a time as the sample moves on a conveyor, which suits online use; other systems capture one wavelength at a time with tunable filters, or snapshot imagers capture all at once at lower resolution. Multispectral imaging uses a few selected wavelengths, chosen from hyperspectral studies, to build faster and cheaper online systems.
Analysis follows the methods of spectroscopy and chemometrics (chapter 12). Spectra are pre-processed to remove scatter and noise, for example by standard normal variate, multiplicative scatter correction and Savitzky-Golay smoothing and derivatives; models are built to predict composition at each pixel (giving chemical images that map moisture, fat or protein) or to classify pixels (sound or bruised tissue, meat or bone, contaminated or clean). Applications include detecting bruises and internal defects in fruit before they are visible, mapping fat and moisture in meat and fish fillets, detecting faecal contamination on poultry carcasses, sorting grains for mycotoxin-associated kernels, identifying foreign materials such as plastics on lines, and checking adulteration of powders such as spices and milk powder. Limits include the large volume of data, sensitivity to illumination, sample geometry and temperature, the need for reference measurements to calibrate models, and the limited depth of penetration of light, so internal properties are inferred from near-surface spectra.
Proteomics studies the full set of proteins in a sample, their identities, quantities, modifications and changes. In food science it is used to characterise raw materials (the proteins of milk, meat, cereals and legumes), to follow changes during processing and storage (proteolysis in cheese ripening and meat ageing, glycation by the Maillard reaction, oxidation), to authenticate species and varieties, and to detect allergens. Bottom-up proteomics, the most common approach, extracts proteins, reduces and alkylates their disulphide bonds, digests them with trypsin (which cleaves after lysine and arginine), separates the resulting peptides by nano- or UHPLC, and analyses them by tandem mass spectrometry; peptide fragment spectra are matched to sequences in protein databases by search algorithms, with statistical control of the false discovery rate, and proteins are inferred from their peptides. Top-down proteomics analyses intact proteins, preserving information on modifications and variants, but is technically harder. Two-dimensional gel electrophoresis, separating proteins by isoelectric point and mass, was the original proteomic tool and is still used to visualise patterns.
Quantification in proteomics may be label-free (comparing peptide signal intensities or spectral counts between runs), use stable-isotope labelling of samples (metabolic labelling or chemical tags such as isobaric tags that allow several samples to be compared in one run), or use targeted methods with isotope-labelled peptide standards for absolute quantification. Peptidomics focuses on naturally occurring or processing-generated peptides, for example bioactive peptides with antihypertensive or antioxidant activity released in fermented milks and protein hydrolysates, bitter peptides in cheese, and peptides that indicate species or heat treatment.
Food allergen management needs methods that detect traces of allergenic foods, such as milk, egg, peanut, tree nuts, soy, wheat and other cereals containing gluten, fish, crustaceans, sesame and others, in products and on surfaces. Three families of methods are used. Immunoassays, chiefly enzyme-linked immunosorbent assays (ELISA) and lateral-flow devices, use antibodies against allergenic proteins; they are sensitive, rapid and widely used, but results depend on the antibody and on how processing (heat, hydrolysis, fermentation) has changed the target protein, so different kits can give different results for the same sample, and results are often expressed in different units (mg of protein, mg of whole food) that must be interpreted carefully. Gluten is measured by ELISA methods based on antibodies such as R5, with competitive formats for hydrolysed products, against limits such as 20 mg/kg for 'gluten-free' claims in many jurisdictions. DNA-based methods (PCR, especially real-time PCR) detect species-specific DNA sequences; they are sensitive and robust to processing that leaves DNA fragments, but they detect DNA, not the allergenic protein, so they are indirect and cannot detect allergens in highly purified fractions without DNA, and they cannot distinguish milk from beef or egg from chicken meat.
Mass spectrometry detects allergenic proteins directly through signature (marker) peptides: peptides unique to the allergenic protein, stable to processing, and reliably released by digestion. In targeted LC-MS/MS with MRM, several marker peptides for several allergens are measured in one run, with isotope-labelled peptides as internal standards, giving multiplexed, confirmatory and potentially quantitative results less dependent on processing than ELISA; challenges include protein extraction from processed matrices, digestion efficiency, conversion of peptide amounts to protein or food amounts, and the lack of matrix reference materials. Results of allergen testing are interpreted against reference doses derived from clinical threshold data, used in precautionary allergen labelling schemes and in the risk-based action levels recommended by an FAO/WHO expert consultation on allergen thresholds, expressed as milligrams of protein from the allergenic food.
Metabolomics measures the small molecules (metabolites, roughly below 1500 Da) in a biological sample: sugars, organic and amino acids, lipids, phenolics, volatiles and many others. Together with proteomics and genomics applied to food and nutrition, it forms foodomics. Two strategies are used. Targeted metabolomics quantifies a defined set of known metabolites, with standards, for testing hypotheses. Untargeted metabolomics measures as many features as possible without preselection, to discover differences between groups of samples, such as authentic and adulterated, organic and conventional, fresh and stored, or before and after a dietary intervention. The main platforms are LC-HRMS (broad coverage of semi-polar and polar compounds), GC-MS (volatile and derivatised compounds, with robust libraries) and NMR (quantitative, reproducible, lower sensitivity); they are complementary, and many studies use more than one.
An untargeted workflow has many steps, each of which can introduce bias. Sampling and quenching must stop metabolism quickly and be consistent; extraction must be reproducible; and the sequence of analysis is randomised, with pooled quality-control (QC) samples, a mixture of all samples, injected regularly to monitor and correct instrument drift, and blanks to identify contaminants. Data processing detects peaks (features defined by m/z and retention time), aligns them across samples, removes features that are unreliable in QC samples (for example with a relative standard deviation above about 20 to 30 per cent), fills missing values and normalises. Statistical analysis then combines univariate tests with false-discovery-rate correction and multivariate methods (PCA to explore, PLS-DA or other classifiers to find discriminating features), which must be validated by cross-validation, permutation tests and independent samples because, with thousands of features and few samples, models can easily fit noise.
Discriminating features must be identified to be useful, and identification is the bottleneck of untargeted metabolomics. The Metabolomics Standards Initiative describes levels of confidence: level 1, identified by comparison with an authentic standard analysed under the same conditions (retention time and spectra); level 2, putatively annotated from spectral similarity to libraries or literature; level 3, putatively characterised as a compound class; and level 4, unknown. Reports should state the level for each compound. Pathway and enrichment analysis relates changes to metabolic pathways, with caution about database coverage of plant and food metabolites.
Lipidomics studies lipids comprehensively: triacylglycerols, phospholipids, sphingolipids, sterols, fatty acids and oxidised lipids such as oxylipins. Shotgun lipidomics infuses extracts directly into the mass spectrometer, while LC-MS separates lipid classes or species first; annotation must state the level of structural detail (for example the sum of carbons and double bonds versus individual fatty acyl chains and their positions). Food lipidomics characterises the lipids of milk, meat, fish, oils and seeds, detects adulteration of oils and fats, follows lipid oxidation and its products, and studies the digestion of lipids in emulsions.
Applications of food metabolomics include authentication of origin, variety and production system (for example geographical origin of honey, rice, olive oil and tea; organic versus conventional produce; wild versus farmed fish), detection of adulteration and fraud, finding markers of processing and storage (thermal treatment, ripening, spoilage), characterising flavour and its precursors, studying the effect of breeding and agronomy on composition, and nutritional metabolomics, in which biomarkers of food intake in blood or urine provide objective measures of what people eat and how diet affects metabolism. For authentication, models must be built on large, representative sets of authentic samples covering seasonal, regional and varietal variation, validated with independent samples and maintained over time, before they can be used in enforcement.
A biosensor is an analytical device that combines a biological recognition element with a transducer that converts the recognition event into a measurable signal. Recognition elements include enzymes (glucose oxidase, cholinesterases for organophosphate pesticides), antibodies (immunosensors for pathogens, toxins and allergens), nucleic acids (DNA probes for species and pathogens), aptamers (synthetic nucleic-acid sequences selected to bind a target), whole cells, and biomimetic receptors such as molecularly imprinted polymers. Transducers are electrochemical (amperometric, measuring current from redox reactions; potentiometric, measuring potential; impedimetric, measuring changes in impedance), optical (colorimetric, fluorescence, chemiluminescence, surface plasmon resonance, which detects binding through changes in refractive index at a metal surface), piezoelectric (quartz crystal microbalances, whose frequency changes with bound mass) or thermal. The best-known biosensor is the blood glucose meter, an amperometric enzyme electrode using glucose oxidase or glucose dehydrogenase; similar enzyme electrodes measure glucose, lactate, glutamate, ethanol and other food components.
Biosensors promise rapid, specific, portable and inexpensive measurements for use on site: detecting pathogens on processing lines, pesticides in produce, antibiotics in milk at reception, histamine in fish, allergens on surfaces and toxins in grains. Lateral-flow immunoassay strips, a simple optical biosensor format read by eye or with a reader, are already widely used for antibiotics, mycotoxins and allergens. Nanomaterials such as gold nanoparticles, carbon nanotubes and graphene improve sensitivity. Their performance is described by the same figures as other methods, limit of detection, selectivity, working range, repeatability and stability, plus regeneration and shelf life of the biological element. Practical limitations include interference from complex food matrices, the need for sample preparation, stability of biological components, and validation against reference methods, which is why many biosensors described in research have not reached routine use.
An electronic nose is an array of gas sensors with partial, overlapping selectivity, combined with pattern-recognition software, that responds to the overall volatile profile of a sample rather than to individual compounds. Sensor types include metal-oxide semiconductors, whose resistance changes when volatiles react at their heated surfaces; conducting polymers; quartz crystal microbalances or surface acoustic wave devices coated with sorbent films; and electrochemical cells. The pattern of responses across the array forms a fingerprint that is classified by multivariate methods such as PCA, discriminant analysis or neural networks after training with samples of known class. Electronic tongues apply the same principle to liquids with arrays of potentiometric (ion-selective or lipid-membrane) or voltammetric electrodes, responding to taste-related compounds such as acids, salts, bitter substances and umami compounds.
These instruments are used for rapid classification and quality control: detecting spoilage of meat and fish, monitoring fruit ripeness, grading coffee, tea, wine and olive oil, checking raw materials for off-odours, and assessing bitterness of pharmaceuticals or protein hydrolysates. Their advantages are speed, low cost per analysis and the ability to screen many samples. Their limitations are significant: they do not identify compounds; sensors drift over time and respond to humidity and temperature; models trained on one set of samples may not transfer to new batches, seasons or instruments; and they cannot replace sensory panels for describing aroma or taste, only correlate with them. Good practice therefore includes controlled sampling (headspace equilibration time and temperature), regular calibration with reference samples, drift correction, validation with independent sample sets, and confirmation of key findings by GC-MS or sensory evaluation.
Modern instruments produce data with many more variables than samples: spectra with thousands of wavelengths, chromatograms with thousands of features, images with millions of pixels. Chemometrics is the application of statistical and mathematical methods to extract information from such data. Pre-processing removes unwanted variation before modelling: mean-centring subtracts the mean of each variable; autoscaling also divides by its standard deviation so that all variables have equal weight; standard normal variate and multiplicative scatter correction remove light-scattering effects in near-infrared spectra; Savitzky-Golay smoothing and first or second derivatives remove baseline offsets and sharpen overlapping bands; and normalisation corrects for differences in total signal.
Principal component analysis (PCA) is an unsupervised method that finds new, uncorrelated variables (principal components) as linear combinations of the original ones, ordered by the variance they explain. Score plots show how samples relate (clusters, trends, outliers), and loading plots show which variables drive each component. Partial least squares (PLS) regression is a supervised method that relates a matrix of predictors (such as spectra) to responses (such as fat content) through latent variables that maximise covariance between them; it is the standard method for calibrating near-infrared and other spectroscopic analysers. The number of latent variables is chosen by cross-validation to avoid overfitting. Model quality is reported as the root mean square error of calibration, of cross-validation (RMSECV) and, most importantly, of prediction for an independent test set (RMSEP), together with R². The ratio of performance to deviation, RPD = SD of the reference values/standard error of prediction, compares the error with the natural spread of the samples; values above about 2 to 3 are commonly considered adequate for screening and higher values for quantitative analysis, though thresholds vary by application. For example, a fat calibration with SD of reference values 2.4 per cent and SEP 0.6 per cent has RPD = 4.0.
Classification assigns samples to classes, such as authentic or adulterated, origin A or B, or acceptable or defective. PLS discriminant analysis (PLS-DA) adapts PLS to classes; soft independent modelling of class analogy (SIMCA) builds a separate PCA model for each class and checks whether new samples fit it, useful for authentication when only the authentic class is well defined (one-class classification); linear discriminant analysis, k-nearest neighbours, support vector machines and random forests are other widely used classifiers. Artificial neural networks, and deep learning with many layers, learn complex non-linear relationships from large data sets and are increasingly applied to spectra, images and process data; they can be very accurate but need large, representative training sets and are harder to interpret, which matters when results must be explained to regulators or courts. Performance of classifiers is reported as sensitivity (true positive rate), specificity (true negative rate), accuracy and misclassification rates, ideally as a confusion matrix, with attention to class imbalance, since a model that calls every sample authentic will look accurate if fraud is rare.
The most common failure in chemometrics and machine learning is overfitting: building a model that fits the training data, including its noise, but predicts new samples poorly. With many variables and few samples, it is easy to find apparently perfect separations by chance. Protection comes from validation: cross-validation within the training set to choose model complexity; permutation tests, in which class labels are randomly shuffled to show what performance arises by chance; and, above all, testing on an independent set of samples collected separately (different batches, seasons, instruments), never used in model building. Variable selection must be done inside the cross-validation loop, not before it, or performance will be overestimated. Models must be maintained: changes in raw materials, instruments or processes require monitoring of prediction residuals, updating, and transfer between instruments with standardisation methods. Data fusion combines several data sources, such as NIR and MS or spectra and images, at the level of raw data, features or decisions, often improving classification. Well-validated chemometric models make advanced analytical data useful for rapid quality control, process monitoring and authenticity testing; poorly validated ones produce confident but wrong answers.
A doctoral or master's student uses this subject to choose techniques for a thesis: for example DSC and XRD to follow starch retrogradation, CLSM and SAXS to study emulsion gels, LC-MS/MS to measure a contaminant, or untargeted metabolomics with chemometrics to authenticate a product. Calculations of mass accuracy, resolution, plate number, matrix effect, Bragg spacing, chemical shift and RPD, and the correct validation of models, are core skills.
Five years into a career in a reference, enforcement or industrial research laboratory, an analyst develops and validates multi-residue LC-MS/MS and GC-MS/MS methods, handles matrix effects and identification criteria, runs allergen and authenticity testing, maintains NIR and hyperspectral calibrations, and interprets structural data for product development teams.
A laboratory head with twenty-five years' experience decides on investment in HRMS, NMR or imaging platforms, builds databases of authentic samples for authenticity programmes, sets policies for non-targeted screening and decision rules, and represents the laboratory in standardisation and proficiency-testing networks.
A researcher at this level develops new analytical approaches: non-targeted screening workflows, MS-based allergen quantification, foodomics for authenticity and nutrition biomarkers, in-line spectroscopic and imaging sensors with machine learning, and multi-scale structural characterisation linking molecular organisation to texture and digestion.
This manual explains principles and applications of advanced techniques at a level that supports doctoral research and reference-laboratory work. It does not provide instrument operating procedures, validated methods for specific analytes, or regulatory identification criteria of every jurisdiction, which must be taken from the applicable guidance. Numerical values (resolving powers, shifts, spacings, thresholds such as RPD) are typical and vary with instruments and applications. Routine methods are in food-analysis and statistical theory in statistics-experimental-design-food; isotope-ratio mass spectrometry and DNA methods for authenticity are treated more fully in food-fraud-authenticity.
Original practice questions written to the syllabus and pattern of the exams named in each question's examTags. They are not past papers and do not reproduce any board's questions.
These notes are free to read here, with no account.
What the Laisha app adds:
Free for ten days, then a subscription. We would rather write that here than have you find it later.