Articles

Spec Oct. 2018 Issue TOC

Click the title above to open the Spectroscopy October 2018 issue in an interactive PDF format.

Spectroscopy October 2018 Issue PDF

The development of analytical instrumentation over the past 50 years has allowed us not only to detect trace metals at the parts per quadrillion (ppq) levels, but also to know the oxidation state, biomolecular form, elemental species, and isotopic constituents. Here, we look at how the development of atomic spectroscopy techniques has enabled a much better understanding of the links between trace metal toxicity and human disease, and, in particular, the role of lead in the health of young children.

	Understanding the effects of trace metals on human health is as complex as it is fascinating. Too low or too high a concentration of essential trace elements in our diet can affect our quality of life. On the other hand, metallic contamination of the air, soil, and water can have a dramatic impact on our well-being. There are many examples that highlight both the negative and positive effects of trace metals on our lives. For instance, the effect of lead toxicity, particularly on young children, is well documented, but is it possible to pinpoint the source of the lead poisoning? The movie "Erin Brockovich" alarmed moviegoers about the dangers of hexavalent chromium (Cr VI) in drinking water, but how many in the audience realized that trivalent chromium (CRIII) metal is necessary for the metabolism of carbohydrates and fats? Dr. Oz recently alarmed his viewers about high levels of arsenic in apple juice, but what he failed to say was that it was not the highly toxic inorganic form of arsenic that had been found, but organic arsenic that had been metabolized by the apple tree to a less toxic form. Selenium, which is found in many vegetables including garlic and onions, has important antioxidant properties, but do we know why some selenium compounds are essential, while others are toxic? Clearly these are all complex questions that have to be answered to fully understand the role of trace elements in the mechanisms of human diseases. Atomic spectroscopy has an important role to play in answering these questions.

	The development of analytical instrumentation over the past 50 years has allowed us not only to detect trace metals at the parts per quadrillion (ppq) levels, but also to know their oxidation state, biomolecular form, elemental species, and isotopic constituents. We take for granted all the powerful and automated analytical tools we have at our disposal to carry out trace elemental studies on clinical, toxicological, and environmental samples. However, it wasn't always that way. As recently as the 1960s, the majority of trace elemental determinations were predominantly carried out by traditional wet chemical methods like volumetric-, gravimetric-, or colorimetric-based assays. In fact, the pharmaceutical industry has been using a sulfide precipitation colorimetric test for the measurement of lead and other heavy metals for more than 100 years; that method was only replaced in the 

) in January 2018 by a plasma spectrochemical test (1).

	It wasn't until the development of atomic spectroscopic techniques in the early to mid-1960s that the clinical analytical community realized they had a highly sensitive and diverse trace element technique that could be automated. Every time a major development was made in atomic spectroscopy, beginning with flame atomic emission (FAE) and flame atomic absorption (FAA) in the early 1960s, electrothermal atomization (ETA) or graphite furnace atomic absorption (GFAA) in the early 1970s, inductively coupled plasma–optical emission spectrometry (ICP-OES) in the late 1970s, and inductively coupled plasma–mass spectrometry (ICP-MS) in the early 1980s, trace element detection capability, sample throughput, and automation dramatically improved. There is no question that developments and breakthroughs in atomic spectroscopy have directly impacted our understanding of the way trace metals interact with the human body. Let us now take a look at a specific examples where atomic spectroscopy techniques have allowed us to delve deeper into understanding the impact of trace metal toxicity on our lives, focusing specifically on lead (Pb).

	Lead has no known biological or physiological purpose in the human body, but is readily absorbed into the system by ingestion, inhalation, and, to a lesser extent, by skin absorption (2). Inorganic lead in submicrometer-sized particles in particular can be almost completely absorbed through the respiratory tract, and larger particles may be swallowed. The extent and rate of absorption of lead through the gastrointestinal tract depend on characteristics of the individual and on the nature of the medium ingested. It has been shown that children can absorb 40–50% of an oral dose of water-soluble lead compared to only 3–10% for adults (3). Young children and toddlers are particularly susceptible because of their playing and eating habits, and because they typically have more hand-to-mouth activity than adults (4). Lead is absorbed more easily if there is a calcium or iron deficiency, or if the child has a high fat, inadequate mineral, or low protein diet. When absorbed, lead is distributed in the body in three main areas: bones, blood, and soft tissue. About 90% is distributed in the bones, while the majority of the rest gets absorbed into the bloodstream, where it gets taken up by porphyrin molecules (complex nitrogen-containing organic compounds providing the foundation structure for hemoglobin) in the red blood cells (5). It is, therefore, clear that the repercussions and health risks are potentially enormous if children are exposed to abnormally high levels of lead.

	The toxic effects of lead have recently been exemplified by the drinking water crisis in Flint, Michigan, where public health officials and water authority personnel failed to take remedial action when they replaced Lake Michigan with the Flint River as the source of the city's drinking water, a change that resulted in corrosion of lead pipes and high levels of lead in the drinking water supply. This particular problem is still being investigated, but the Centers for Disease Control and Prevention (CDC), recently reported that at least four million households have children living in them who are being exposed to high levels of lead from a combination of old lead paint and lead water pipes. As a result, there are approximately half a million U.S. children 1–5 years of age with blood lead levels (BLL) in excess of 5 micrograms per deciliter (µg/dL), the level at which CDC recommends remedial actions be taken (6).

	Lead poisoning affects virtually every system in the body, and often occurs with no distinctive symptoms. It can damage the central nervous system, kidneys, and reproductive system and, at higher levels, can cause coma, convulsions, and even death. Even low levels of lead are harmful and are associated with lower intelligence, reduced brain development, decreased growth and impaired hearing (7). The level of lead in a person's system is confirmed by a blood-lead test, and by today's standards a blood lead level is considered elevated if it is in excess of 5 µg/dL (50 ppb) for children (8). However, the long-term effects of lead poisoning have not always been well understood. In the early and mid-1960s, remedial action would be taken if a blood lead level (or 

, as it was known then) was in excess of 60 µg/dL. As investigators discovered more sensitive detection systems and designed better studies, the generally recognized level for lead toxicity has progressively shifted downward. In 1970, it was lowered to 40 µg/dL and, by 1978, the level had been reduced to 30 µg/dL. In 1985, the CDC published a threshold level of 25 µg/dL, which they eventually lowered to 10 µg/dL in 1991. It stayed at this level until it was reduced to 5 µg/dL in 2012. However, as our understanding of disease improves and measurement technology gets more refined, this level could be pushed even lower in the future (9). Figure 1 shows the trend in blood lead levels considered elevated by the Centers for Disease Control (CDC), since the mid-1960s.


	Figure 1: The trend in blood lead levels (µg/dL) in children considered elevated by the Centers for Disease Control and Prevention (CDC), since the mid-1960s.

 (BLRV) has been used more recently (since 2012), and refers specifically to the 97.5th percentile of blood lead levels for children 1–5 years old in the United States, calculated from blood lead tests performed in the National Health and Nutrition Examination Survey (NHANES). The BLRV is not a health-based toxicity threshold, nor does it define what level is considered normal. It is intended to help identify the highest risk childhood populations and geographic areas.

	It is also important to point out that these thresholds were not all determined the same way. Only in 2012 (when the recognized level for lead toxicity was lowered to 5 µg/dL) was the population-based threshold called the BLRV and calculated from population statistics. Although all these levels could be said to describe thresholds of elevated blood lead levels generally, even the term 

wasn't specifically defined in CDC policy until 1978.

	Currently, the major source of lead poisoning among children comes from lead-based household paints, which were used until they were banned in the United States in 1978 by the Consumer Product Safety Commission. Prior to that date, leaded gasoline was the largest pollutant, before it was completely removed from the pumps in 1995. Other potential sources include lead pipes used in drinking water systems, airborne lead from smelters, and clay pots, pottery glazes, lead batteries, and household dust. However, awareness of the problem, combined with preventative care and regular monitoring, have reduced the percentage of children aged 1–5 years with elevated blood levels (≥5 µg/dL) in the US from 26% in the early-mid 1990s to less than 2% in 2014. These data were taken from a recent National Health and Nutrition Examination Survey (NHANES) report (10).

Routine Monitoring of Lead Using Atomic Spectroscopic Techniques

	There is no question that the routine monitoring of children has had a huge impact in reducing the number of children with elevated blood levels. Lead assays were initially carried out using the dithizone colorimetric method, which was sensitive enough, but very slow and labor intensive. The method became a little more automated when anodic stripping voltammetry was developed (11), but blood-lead analysis was not considered a truly routine method until atomic spectroscopy techniques became available. Let's take a more detailed look at how improvements in atomic spectroscopy instrumentation detection capability have helped to lower the number of children with elevated blood lead levels, since atomic absorption was first commercialized in the early 1960s.

	When flame atomic absorption (FAA) was first developed, the elevated blood lead level was set at 60 µg/dL. Even though this level is equivalent to 600 parts per billion (ppb) of lead, which was well above the FAA detection limit of ~20 ppb at the time, FAA struggled to accurately detect lead at these levels when sample preparation was taken into consideration. The preparation of blood samples typically involved either dilution with a weak acid followed by centrifuging or filtering, or acid digestion followed by dilution and either centrifuging or filtering. More recently, dilution with a strong base like tetramethylammonium hydroxide (TMAH) and the addition of a surfactant to allow for easier aspiration has been used. When sample preparation was factored into the equation, a blood lead level of 600 ppb was reduced to 10–20 ppb, virtually the same as the FAA instrumental detection limit.

	To get around this limitation, an accessory called the Delves Cup was developed in the late 1960s to improve the detection limit of FAA (12). The Delves Cup approach uses a metal crucible or boat, usually made from nickel or tantalum, which was positioned over the flame. The sample, typically 0.1–1.0 mL, is pipetted into the cup, where the heated sample vapor is passed into a quartz tube, which is also heated by the flame. The ground state atoms generated from the heated vapor are concentrated in the tube, and therefore resident in the optical path for a longer period of time, resulting in much higher sensitivity and about 100x lower detection limits. The Delves Cup became the standard method for carrying out blood lead determinations for many years, because of its relative simplicity and low cost of operation.

	Unfortunately, the Delves Cup approach was found to be very operator dependent, not very reproducible (because of manual pipetting), and required calibration with blood matrix standards (13). The technique became less attractive when electrothermal atomization (ETA) was commercialized in the early 1970s. This new approach offered a detection capability for lead of ~ 0.1 ppb, approximately 200x better than FAA. However, its major benefit for the analysis of blood samples was the ability to dilute and inject the sample automatically into the graphite tube with very little off-line sample preparation. In addition, because the majority of the matrix components were "driven-off" prior to atomization at ~3000 °C, interferences were generally less than the Delves Cup, which only reached the temperature of the air or acetylene flame at ~2000 °C. This breakthrough meant that blood lead determinations, even at extremely low levels, could now be carried out in an automated fashion with relative ease.

	The next major milestone in AA was the development of Zeeman background correction (ZBGC) in 1981, which compensated for non-specific absorption and structured background produced by complex biological matrices, like blood and urine. ZBGC, in conjunction with the stabilized temperature platform furnace (STPF) concept, allowed for virtually interference-free graphite furnace analysis of blood samples using aqueous calibrations (14). Such was the success of the ZBGC–STPF approach, due primarily to the fact that it could be used to analyze many different kinds of samples using simple aqueous standards, that it became the recognized way of analyzing most types of complex matrices by ETA.

	Even though ETA had been the accepted way of doing blood lead determinations for more than 15 years, the commercialization of quadrupole-based ICP–MS in 1983 gave analysts a tool that was not only 100x more sensitive, but suffered from less severe matrix-induced interferences than ETA. In addition, ICP–MS offered multielement capability and much higher sample throughput. These features made ICP–MS very attractive to the clinical community, such that many labs converted to ICP–MS as their main technique for trace element analysis. Then, as the technique matured, using advanced mass separation devices, performance enhancing tools, powerful interference reduction techniques, and more-flexible sampling accessories, detection limits in real-word samples improved dramatically for some elements. Figure 2 shows the improvement in detection capability (in ppb) of ICP-MS compared to ETA and the other atomic spectroscopy techniques.


	Figure 2: Comparison of detection capability (ppb) of atomic spectroscopy techniques used to monitor blood lead and the approximate year they were developed or improved.

	It should also be emphasized that the detection limits shown in Figure 2 are instrument detection limits (IDLs), which are based on simplistic calculations of aqueous blanks carried out by manufacturers, and not realistic method detection limit (MDL) or procedural limits of detection (PLOD) that take into consideration the sample preparation procedure, dilution steps, and multiple analytical measurements. IDLs are also only intended to be used as a guideline for comparison purposes because there are so many different ways of assessing detection capability, based on variations in manufacturer, instrument design, and methodology.

	Figure 3 is a combination of Figures 1 and 2, and shows improvement in the blood lead method detection limit (now in µg/dL and not ppb) offered by atomic spectroscopy techniques compared to the trend in blood lead levels set by the CDC. To make the comparison more valid, a factor of 100x has been applied to the instrumental detection limits to give an approximation of the achievable "real world" method detection limit in a blood sample matrix. Both plots are shown in log scale, so they can be viewed on the same graph. The main purpose of these data is to show how the blood lead levels considered by the CDC as "elevated" over the past 50 years have dropped as method detection limits of the various atomic spectroscopy techniques have been lowered, thus giving researchers more confidence in the integrity of their data.


	Figure 3: The improvement in real-world method detection capability (in µg/dL) offered by atomic spectroscopy techniques for blood-lead determinations compared to the trend in blood-lead levels regulated by the Centers for Disease Control and Prevention (CDC).

	It should also be emphasized that a degradation factor of 50–100x is quite normal when converting an IDL to an MDL, when characterizing samples by atomic spectroscopy techniques. However, when analyzing a very complex biological matrix like blood by ICP-MS, there are many different ways of calculating LODs to encompass the entire analytical procedure. One common approach to determine the PLOD is to carry out 20 runs and plot standard deviation of the standards and spiked matrix versus concentration, extrapolating the regression line to the ordinate axis, to determine the standard deviation at zero concentration (15). In a high throughput laboratory, this approach might not be realistic, because of the additional time taken. The time involved can be somewhat shortened by taking fewer readings, but doing so will clearly negatively impact the statistical data and detection limit. Whichever approach is used, one should take into account variability in sample preparation, environmental contamination, solvents, and reagents, as well as minor sampling errors from dilution or pipetting over many runs, all of which can cause variability from day to day. Given such variability, a real-world procedural LOD for Pb in blood is often three orders of magnitude worse than the instrument detection limit, and is typically around 0.01–0.07 µg/dL, depending on the type of ICP–MS technology and interference reduction technique used (9,16).

Identifying Sources of Lead Using Isotopic Fingerprinting

	An added benefit of the ICP-MS technique is that it also offers isotopic measurement capability. This feature is very attractive to many clinical laboratories, because it gives them the ability to carry out isotope tracer (17), isotope dilution (18) and isotope ratio (19) measurements, which are beyond the realms of other atomic spectroscopy techniques. In fact, the isotopic measurement capability allows researchers to get a better understanding of the source of lead poisoning by measuring the isotope ratio of blood-lead samples and comparing them with possible sources of lead contamination. The principal behind this approach, known as 

, is based on the fact that lead is composed of four naturally occurring isotopes: 

Pb, all with the same atomic number, but with different atomic masses. Thus, when naturally occurring lead is ionized in the plasma, it generates four ions, all with different atomic masses. Figure 4 shows a mass spectrum of the four lead isotopes 

Pb, together with their relative natural abundances of 1.4%, 24.1%, 22.1% and 52.4%, respectively.


	Figure 4: Mass spectrum of the four lead isotopes at 204, 206, 207, and 208 atomic mass units (amu), with their respective natural abundances.

	All the lead isotopes, with the exception of 

Pb, are products of radioactive decay of either uranium or thorium, the abundance of which will vary slightly depending on the rock type and geological area. This means that in all lead-based materials and systems, 

Pb has essentially remained unchanged at 1.4%, since the earth was first formed (20). The ratios of the isotopic concentrations of 

Pb will therefore vary, depending on the source of lead. This fundamental principle can be then be used to match lead isotope ratios in someone's blood to a particular environmental source of lead contamination.

	However, there are known, well-understood limitations of this approach. For lead fingerprinting to be useful, potential sources of lead exposure must be limited in number and scope and the lead sources must be isotopically distinct. If more than two sources of environmental lead are likely, such as from water pipes, gasoline, smelter, paint, pottery, and glazes, then mixed or combined isotope ratios will occur and, as a result, no useful data will be obtained. In addition, if someone has chronic exposure to extremely high lead levels, the person might have brittle or broken bones, and accumulated lead in the bones would be released into the bloodstream, which could shift the lead equilibrium. This release of lead from the bones could elevate blood lead levels, independent of the source of the lead exposure or contamination.

	A good example of using isotope ratios to pinpoint the source of lead poisoning that worked extremely well involved a study carried out on a group of people living in a small village near Mexico City (21). A number of the residents had abnormally high levels of lead in their blood, which came from one of two likely sources: the use of leaded gasoline, which had contaminated the soil, or glazed ceramic pots, which were used for cooking and eating purposes, or both. For this experiment, the lead isotope ratios were measured using an electrothermal vaporization (ETV) sampling accessory coupled to an ICP-MS instrument. In this sample device, a heated graphite tube, similar to the type used in ETA, is used to thermally pretreat the sample. But instead of using the tube to produce ground state atoms, its main function is to drive off the bulk of the matrix before the analytes are vaporized into the plasma for ionization and measurement by the mass spectrometer. The major benefit of ETV-ICP-MS for this application is that complex matrices like blood, gasoline, and pottery or clay material can be analyzed with very little interference from matrix components (22). An additional benefit with regard to taking blood samples is that typically only a 20–50 µL aliquot is required for analysis. Figure 5 represents a schematic of how the ETV–ICP–MS system works, showing the two distinct steps: prevaporization to drive off the matrix components, and vaporization to sweep the analyte vapor into the ICP–MS for analysis.


	Figure 5: Schematic of ICP-MS coupled with an the electrothermal vaporization sampling accessory (ETV-ICP-MS), showing the two distinct stages: (a) prevaporization to drive off the matrix components and (b) vaporization to sweep the analyte vapor into the ICP-MS instrument for analysis. Adapted with permission from reference 22.

	In the Mexican study, ETV-ICP-MS was then used to determine the lead isotope ratios of 

Pb in blood samples from a group of residents. These ratios were then compared with the two likely sources of lead contamination from the cooking pots and the gasoline samples. Figure 6 shows a subset of data taken from the study. It shows a plot of the 

Pb ratio for the blood, cookware, and gasoline samples. It can be seen from this plot that the data for the blood and cookware are grouped very tightly together around the theoretical value of the ratios (known as the 

), while the gasoline data are grouped together on their own.

Pb for blood (â»), cookware (O), and gasoline (•) samples, showing the theoretical (primeval) lead line. Adapted with permission from reference 21.

	Based on principal component analysis of the data, this result confirms that the lead isotope ratios of the blood and cooking pots are almost identical, and are very close in composition to primeval lead, with very little addition of radiogenic lead (produced from radioactive decay). On the other hand, the alkyl lead compounds used in the production of leaded gasoline are from a different source of lead and as a result generate a very different isotopic signature. These data showed very convincing evidence that the residents of this small Mexican village were getting poisoned by the glazed clay pots they were using for cooking and eating, and not from contamination of the environment by leaded gasoline, as was first suspected.

	There is no question that developments in atomic spectroscopy have helped us better understand the toxicity effects of lead over the past 50 years. Atomic spectroscopy advances have allowed us to lower the clinical practice threshold level of 60 µL/dL in the mid-1960s to the current blood lead reference value (BLRV) of 5 µL/dL. More importantly, these techniques have helped to reduce elevated blood levels of children in the United States from 26% in the early to mid-1990s to less than 2% in 2014, as well as allowing us to get a much better understanding of the environmental sources of lead contamination. However, such is the power and versatility of modern atomic spectroscopy instrumentation and its accessories, that it has also dramatically improved our understanding of other trace metal-related human diseases. The toxic effects of trivalent/pentavalent arsenic and hexavalent chromium or the nutritional benefits of different selenium species would still be relatively unknown, if it weren't for the continual improvements in ICP-MS and, in particular, its use as a very sensitive detector for trace element speciation studies using chromatographic separation technology. Even though ICP-MS has been successfully applied to many application areas since it was first commercialized in 1983, its use as a biomedical, clinical, and toxicological research tool has had a direct impact on the quality of many people's lives.

	I would like to thank Dr. Steve Pappas of the CDC for his thoughtful comments and edits to this month's 

	(1) Elemental Impurities in Pharmaceuticals: Updates: United States Pharmacopeai (USP) Website: 

http://www.usp.org/chemical-medicines/elemental-impurities-updates

	(3) Agency for Toxic Substances and Disease Registry (ATSDR), Toxicological Profile for Lead, Section 3.3: Toxicokinetics, August, 2007, (

https://www.atsdr.cdc.gov/toxprofiles/TP.asp?id=96&tid=22

	(4) Preventing Lead Poisoning in Young Children, Chapter 2: Absorption of Lead, Centers for Disease Control and Prevention (CDC), 1991, 

https://www.cdc.gov/nceh/lead/publications/books/plpyc/contents.htm

	(5) H. L. Needham, Case Studies in Environmental Medicine-Lead Toxicity, U. S. Dept. of Health and Human Services (1990).

	(6) Preventing Lead Poisoning in Young Children, Lead Information Page, Centers for Disease Control and Prevention (CDC), 

https://www.cdc.gov/nceh/lead/default.htm

	(7) Childhood Blood Lead Levels in Children Aged <5 Years: United States, 2009–2014, Morbidity and Mortality Weekly Report (MMWR), Surveillance Summaries /January 20, 2017/66 (3); 1–10, 

https://www.cdc.gov/mmwr/volumes/66/ss/ss6603a1.htm

	(8) CDC Response to Advisory Committee on Childhood Lead Poisoning Prevention Recommendations in "Low Level Lead Exposure Harms Children: A Renewed Call of Primary Prevention" (2012), 

https://www.cdc.gov/nceh/lead/ACCLPP/blood_lead_levels.htm

	(9) Record of Proceedings from the Meeting of the Lead Poisoning Prevention Subcommittee of the NCEH/ATSDR Board of Scientific Counselors, Centers for Disease Control and Prevention (CDC), Atlanta, GA, September 19, 2016.

	(10) Centers for Disease Control and Prevention (CDC), Morbidity and Mortality Weekly Report (MMWR), October 7, 2016/65(39); 1089, Source: The National Health and Nutrition Examination Survey (NHANES); 

	(11) S. Constantini, R. Giordano, M. Rubbing. 

	(13) S. Cabet, J.M. Ottoway, and G.S. Fell, 

Quality Assurance of Chemical Measurements

 (CRC Press, Boca Raton, Florida, 1st ed., 1987).

	(16) D.R. Jones, J.M. Jarrett, D.S. Tevis, M. Franklin, N.J. Mullinix, K.L. Wallon, C.D. Quarles Jr., K.L. Caldwell, and R.L. Jones, 

https://www.sciencedirect.com/science/article/pii/S0039914016307305

	(18) J.W. McLaren, D. Beauchemin, and S.S. Berman, 

Inter-Science Publishers Inc, New York, New York, 1960).

	(21) M. Chaudhary-Webb, D.C. Paschal, W.C. Elliott, H.P. Hopkins, A.M. Ghazi, B.C. Ting, and I. Romieu, 

	(22) S. Beres, R. Thomas, E. Denoyer, and P. Bruckner, 

 is the principal of Scientific Solutions, a consulting company that serves the application and writing needs of the trace element user community. He has worked in the field of atomic and mass spectroscopy for more than 40 years, including 25 years for a manufacturer of atomic spectroscopic instrumentation. He has written almost 90 technical publications, including a 15-part tutorial series on ICP-MS. He recently completed his fourth textbook, entitled 

Measuring Elemental Impurities in Pharmaceuticals: A Practical Guide

. He has an advanced degree in analytical chemistry from the University of Wales, UK, and is also a Fellow of the Royal Society of Chemistry (FRSC) and a Chartered Chemist (CChem). He has served on the ACS Committee on Analytical Reagents for the past 18 years as leader of the heavy metals' task force.

Atomic Perspectives

With analytical advances, the clinical practice threshold for lead has dropped from 60 to 5 μL/dL.

The Critical Role of Atomic Spectroscopy in Understanding the Links Between Lead Toxicity and Human Disease

This year’s award recipient, Megan Thielges, is a pioneer in the use of vibrational probes, in conjunction with two-dimensional infrared spectroscopy, to investigate the structures and dynamics of proteins, with the aim of understanding biological systems.

 an associate professor of chemistry at Indiana University, has won the 2018 Emerging Leader in Molecular Spectroscopy Award, which is presented by 

 magazine. This annual award recognizes the achievements and aspirations of a talented young molecular spectroscopist, selected by an independent scientific committee. The award will be presented to Thielges at the SciX 2018 conference in October, where she will give a plenary lecture, and be honored in an award symposium.

	Thielges earned her B.S. in biochemistry at Arizona State University. She went on to earn her PhD in biophysics from The Scripps Research Institute in La Jolla, California, in 2009. After three years as a postdoctoral fellow at Stanford University, she obtained her current position as an associate professor of Chemistry at Indiana University. She has published 11 papers as a principle investigator, and 22 as a postdoctoral, graduate, or undergraduate researcher. She has given 45 presentations and received six previous awards, as well as nominations for more. In addition to her research, Thielges has served as a reviewer for numerous scientific journals and on panels for many major national departments, including the National Institutes of Health, the National Science Foundation, and the Department of Energy.

The Challenges of Studying Protein Dynamics 

	The focus of Thielges's work is the study of protein structure and dynamics. Specifically, she is a pioneer in the use of multisite vibrational probes in conjunction with two-dimensional (2D) infrared (IR) spectroscopy to investigate the structures and dynamics of proteins.

	The study of protein structure and function is a field of great interest. Proteins are the mediators of biological functions, and the structure of proteins is complex. The simplest level of protein structure is the primary structure, which consists of a sequence of amino acids joined to form a polypeptide chain. The order of the amino acids is what affects the structure and function of the protein; a single substitution can have life-threatening consequences. The secondary structure consists of a regular pattern of substructures that are formed because of the bonds between the various amino acids that compose the polypeptide chain. Proteins are not necessarily static in their conformations, however. In the field of protein dynamics, scientists look at the transitions a protein can make between different shapes that can occur on various scales of length or time, or both. These conformational changes have been shown to affect cellular functions, such as enzymatic activity and cellular signaling. These changes can happen on extremely fast timescales, making vibrational spectroscopy an excellent technique to observe them.

	Given the dense nature of protein structure and the large amounts of molecules present, however, measured signals are congested, with large amounts of spectral overlap, making it is difficult to resolve the signals. A strategy for alleviating this signal congestion is to use a vibrational probe, which involves replacing one or more molecules in the structure with another whose vibrational signature operates in the so-called 

, a region free from native protein signals (1). But developing effective vibrational probes is challenging. For a vibrational probe to be useful at a specific protein site, it must be strongly responsive to the physical property being investigated, be sensitive only to its immediate environment, cause little to no change to the original structure of interest, and be able to be incorporated into the system (2). This is where Thielges comes in.

Harnessing Vibrational Labels and 2D IR to Study Proteins

	A major aspect of Thielges's work is the development of new vibrational labeling strategies. This is the crucial element of her contributions, according to Andrei Tokmakoff, a professor at the University of Chicago. "Her work has resulted in a toolbox that is opening up new ways of using IR spectroscopy in protein chemistry and biophysics," he said. "By combining these new labels with ultrafast 2D IR, Megan has [developed] a new way of studying protein chemistry, where fast dynamical processes involved in catalysis or binding are directly observed rather than inferred."

	One example of this work was Thielges's use of vibrational labels to study cytochrome P450 enzymes, which play a role in the biological functions of molecular synthesis and metabolic processes. They are the major enzymes involved in the metabolism of medications, making their structure and function of particular interest to medical researchers. Thielges's work combining a new labeling techniques she developed (heme-bound CO, in this case) with IR and 2D IR spectroscopy has revealed different substrate binding configurations at the active site in these enzymes and enabled characterization of the fluctuations and exchange of these configurations. Her studies on P450 enzymes have illustrated the importance and power of 2D IR spectroscopy to measure protein dynamics occurring on very small timescales, as well as the technique's ability to gather information about the role of dynamics in biological activity (3). The National Institute of Health has awarded Thielges research grants for this work. By resolving different substrate binding configurations and the fast dynamics of conformational interchange, these experiments may one day enable understanding of the drug-degradation process undertaken by these enzymes.

	Another important application of Thielges's work is in the study of cellular signaling. The SH3 region in proteins is responsible for controlling certain signal transduction pathways, or the process through which a physical or a chemical signal is carried through a cell (cellular signaling). By introducing carbon-deuterium probes into the SH3 domain in yeast cells, Thielges was able to observe variations in the binding conformations of a proline-rich peptide (4). This study illustrated the power of IR spectroscopy to uncover rapidly interconverting states in a protein–peptide complex for the study of cellular signaling.


	Thielges with one of her students, Sashary Ramos, in the laser laboratory at Indiana University.

Enabling the Study of Proteins at the Nanosecond Timescale

	With 2D IR spectroscopy, it is possible to detect the signals from vibrational probes. However, the experimentally accessible timescale has typically been limited to the picosecond regime. This limitation has led to the need to develop 2D IR probes with longer vibrational lifetimes to extend the experimental timescale. Thielges's group has introduced and studied a novel IR probe, p-cyano-seleno-phenylalanine (CNSeF), for the study of protein dynamics by 2D IR spectroscopy, that is able to extend signal lifetime into the nanosecond range. The group demonstrates that this new probe could be detected on timescales 20–50 fold longer than could be achieved with the popular probe cyanophenylalanine (5).


	Thielges giving a talk at the International Conference on Coherent Multidimensional Spectroscopy (CDMS 2018) in Seoul, Korea, in June 2018.

	Tokmakoff wrote enthusiastically in support of Thielges's nomination for this award. He described her as an "ambitious and creative interdisciplinary scientist" who conducts "fearless research." What makes her stand out from her colleagues, he said, is not only that she is highly trained in two different scientific fields that rarely cross-protein and enzyme biochemistry and biophysics, and ultrafast laser spectroscopy-but that she is able to overcome the challenges of combining them successfully. As Tokmakoff said, leading spectroscopists often work with model systems (rather than meaningful biological problems), and chemical and molecular biologists are often unwilling to work with the long development process of new physical methods. "Megan is really the first of her kind-one who can do it all," he said. "She is truly a first class experimental scientist in both areas."

	Michael Fayer, a professor of chemistry at Stanford University, where Thielges did postdoctoral research, agrees with Tokmakoff's appreciation of Thielges's combination of skills. "Megan is an amazing blend of a real biologist and a hard-core chemical physicist," he said. "This is her great strength that allows her to make important contributions to biology and biophysics."

	Minhaeng Cho, the director of the IBS Center for Molecular Spectroscopy and Dynamics and a professor of chemistry at Korea University, first met Thielges at a conference three years ago, and the two have since collaborated on multiple occasions. According to Cho, Thielges's contributions to molecular spectroscopy started while she was still in graduate school. While at the Scripps Institute, he said, Thielges "introduced an extremely small and non-perturbative vibrational probe into proteins in a site-specific manner." This was important because of the spectral congestion that made using the plentiful C-H bonds to study the shapes and motions of proteins impossible. By replacing the hydrogen with deuterium at specifically targeted sites, Thielges demonstrated that the C-D stretch band, which falls into proteins' silent IR region, has "line shapes [that] are highly sensitive to protein conformations and secondary structures." Cho believes Thielges's research will lead her to address deep questions in biology, such as how protein–protein interactions and recognition are related to fast local fluctuation dynamics at the interface between proteins.

	Floyd Romesberg, who was Thielges's graduate advisor at Scripps, says Thielges is a remarkable experimentalist. "She totally revolutionized our experimental approach to data collection, and her methods are still used by my group today," he said. "She also organized a study group for the students to teach themselves the theory underlying time-resolved non-linear spectroscopy."

	Carlos Baiz, an assistant professor of chemistry at the University of Texas at Austin, describes Thielges as "creative, resourceful, [and] charismatic." He believes she will continue to "push the envelope" when developing more complex probes and methods to further evaluate protein structure and dynamics. He believes her new techniques will be more widely adopted by the ultrafast community in the coming years.


	Thielges receiving an award for her work as a session chair at the International Conference on Coherent Multidimensional Spectroscopy in Seoul, Korea, in June 2018.

	Peter Hamm at the University of Zurich conducts research that is very closely related to that of Thielges. But even though he characterizes their relationship as "a friendly competition," he nominated her for an award last year. Hamm describes Thielges's research as "an absolute prerequisite" for 2D IR spectroscopy to become a more versatile tool to study the structure and dynamics of proteins. "Her work is courageous," he says. "It does not seek the low-hanging fruit."

	Megan Thielges's colleagues all believe she has a very bright and promising future ahead of her. Romesberg summarized this outlook well. "Megan will continue to combine sophisticated biology and spectroscopy," he said, "not only to explore protein dynamics, but to begin to elucidate how [they have] been tailored by evolution for biological function."

	Megan Thielges is a pioneer at the forefront of combining biophysics with spectroscopy. She has been "fearless" and has overcome many practical obstacles to advance these two fields. She is a top-notch experimentalist who takes on big challenges. From organizing study groups to teach herself the theory behind time-resolved non-linear spectroscopy to developing methods as a graduate student that her advisor and his current group still use today, it seems there is nothing Megan Thielges can't do if she puts her mind to it.

	(1) R. Adhikary, J. Zimmerman, and F.E. Romesberg, 

	(2) J. Ma, I.M. Pazos, W. Zhang, R.M. Culik, and F. Gai, 

	(3) E.J. Basom, J.W. Spearman, and M.C. Thielges, 

	(4) R.E. Horness, E.J. Basom, J.P. Mayer, and M.C. Thielges, 

	(5) S. Ramos, K.J. Scott, R.E. Horness, A.L. Le Sueur, and M.C. Thielges, 

is a freelance writer. Direct correspondence about this article to 

Molecular Special Feature

This year’s honoree, Megan Thielges, is a pioneer in the development of vibrational probes for use with 2D IR spectroscopy, to reveal new information about the structures and dynamics of proteins.

The 2018 Emerging Leader in Molecular Spectroscopy Award

Tunable diode laser absorption spectroscopy (TDLAS) has been a popular technique for trace gas detection for more than 20 years in academic and research circles. Commercial applications began to appear more than 10 years ago and have hit the mainstream in many markets. This article provides an overview of the methods and the development history of TDLAS, and discusses some of the future challenges for the technology.

	Tunable diode laser absorption spectroscopy (TDLAS) has evolved along with the lasers and detectors used in the technique. Here, we will look a bit at the basics, and then focus on current capabilities and challenges.

	A typical tunable distributed feedback (DFB) diode laser module tunes over a range of 2–5 nm through a combination of current and temperature tuning. These lasers have a linewidth on the order of 1–4 MHz, which at 1500 nm translates to about 20–30 femtometers, far narrower than the typical Doppler-broadened linewidth of a molecular transition. Further, the lasers can be tuned at rates up to 20 or more GHz/s, allowing rapid modulation over the absorption line of interest.

	The Beer-Lambert law for small concentrations can be written with the absorbance a as a function of the absorption coefficient σ, number density 

	where the absorption coefficient is measured at the line center. 

 is the source intensity on the detector with no absorber, and 

 is the measured intensity on the detector in the presence of an absorber.

	The perennial problem with this physics is that it is always very difficult for a detector staring at a bright source like a laser to detect a small dip in power resulting from the absorption of a trace species. In many cases in atmospheric monitoring, even when the absorption coefficient is high and the pathlength is long, direct absorption measurements are limited by the noise in the environment and the noise in the detector, and typical detection limits may be on the order of 0.1% or more.

	The most practical solution to this problem is to superimpose a modulation on the drive current that tunes the laser in a predictable way. A common way to do this is referred to as 

, in which a relatively slow ramp (10-100 Hz) sawtooth wave scans over the entire absorption line of interest, while at the same time a higher frequency (10+ kHz), small-amplitude sine wave modulates the laser wavelength.

	These two drive signals are superimposed on the injection current of the laser, as illustrated in Figure 1. Each period of the slower frequency ramp contains a scan over the feature of interest. Frequency-sensitive lock-in detection tunes to the high-frequency modulation. The amplitude of the high-frequency modulation influences the signal strength, with the maximum signal occurring when the amplitude of the modulation is 2.2 times the half-width of the absorption line.


	Figure 1: Wavelength modulation includes a slow sweep (left) and a faster modulation (right) of the drive current, sweeping across an absorption feature of interest (bottom).

	Once the signal is detected by the lock-in amplifier, it has a characteristic shape that looks like the shape of a derivative of the absorption line shape, which gives wavelength modulation its nickname: 

. The idealized functions are shown in Figure 2, wherein Figure 2a on the left is modeled the absorbance of a line defined by a Voigt profile with an equal Doppler and Lorentzian half-width of 0.02 cm

, with peak absorbance of 1%. In Figure 2b, on the right, the idealized first through third harmonics of the signal are illustrated, with a modulation amplitude of 0.07 cm

. Both graphs are referenced such that the line center is 0.0 cm


	Figure 2: At left, an absorption line with 1% absorbance modeled with a Voigt profile; at right, the harmonic signals that would be measured with wavelength modulation (as described in the text) at 1, 2, and 3 times the modulation frequency.

	Each of the functions depicted in Figure 2b looks like the line shape that would be measured with a lock-in detector at one, two, and three times the modulation frequency as the laser is swept across the absorption line. The green dotted line corresponding to the first harmonic (marked "H1") has the shape of the first derivative, the red "H2" line looks like the second derivative, and so forth.

	In practice, the first derivative signal typically has a substantial DC offset as a result of the fact that the laser power is continuously increasing during the sweep of the laser, as current is increased. For this reason, the second harmonic, which is proportional to the partial pressure (concentration) of the absorbing species, is typically used for quantification, as it has a zero DC offset. In fact, at the line center, the intensity measurement of each of the even harmonics is proportional to the absorber concentration. The relations for the first few harmonics are:

 are the measured intensity at the first and second harmonic frequency, the former measured at the absorption line center frequency ν

, and the latter also a function of the modulation amplitude Δν. Thus, 

 is the DC offset of the first harmonic mentioned earlier, which is proportional to laser energy. 

 is the average laser intensity (which may be different as a result of laser or environmental conditions in the beam path). η is a proportionality constant, and is a function of the particular laser and detector pair which can be measured.

, which is typically known (and can be measured in a laboratory if it is not known). The last parameter to be defined is 

, the partial pressure of the species to be determined.

. With all of the parameters known, this ratio of harmonic signals allows calculation of the partial pressure of the absorbing species. This is one of the most straightforward ways to use a modulated tunable diode laser signal to determine trace gas concentration. The modulation and detection hardware are all readily available.

	There are other schemes for measurement of concentration as well, including higher-frequency (100 MHz–10 GHz) modulation (so-called 

) which requires a very fast (expensive) detector but pushes the intensity noise of the laser dramatically lower (1). Theoretically, this may have better performance than the wavelength modulation scheme above. A variation on this theme is the use of two FM frequencies separated by a small frequency (2). Detection at the difference (so-called 

) frequency allows use of a lower cost detector. Yet another method uses the ratio of two even harmonics (for example, 

) to simultaneously determine total pressure 

, through an understanding of the variation of the harmonic function with the "modulation depth," the ratio of the amplitude of modulation to the line strength (3).

	Measurement of temperature is also straightforward, including at elevated combustion temperatures. If the variation of line strength with temperature 

), shown in equation 3, for two lines of the same molecule are known, then the ratio of the measured intensities of the two lines can be solved for temperature. Ideally, the variation would not only be monotonic with temperature, but relatively steeply increasing over the range of maximum interest, for best sensitivity. We did this with two adjacent lines of water that could be measured with a single laser to yield temperature over a wide range (4).

	One of the first applications of TDLAS to atmospheric gases was published in 

 in 1971 (5). Through the 1980s, development was limited by the laser sources, which were primarily lead salt laser diodes that needed to be cooled to cryogenic temperatures. These lasers emitted in the mid-infrared (mid-IR), thus also requiring cooled detectors. Lead selenide and mercury-cadmium-telluride (MCT) detectors were the most popular, the latter particularly so because of the possibility for fast frequency response (and thus modulation). However, even with the development of compact and reliable cryocoolers, the cost and complexity of TDLAS systems kept them mostly employed in laboratories and in very special circumstances. The lengthy chapter in Sigrist's 1994 edited book is an excellent summary of work through the early 1990s in TDLAS systems (6).

	The development of near-IR (NIR) laser sources in the mid-to-late 1990s was driven primarily by the telecommunications boom, but it had major implications for TDLAS monitoring. As fast as single-mode fiber-coupled laser sources between 1.3 and 2 µm were developed for telecommunications, they were snapped up by researchers aiming to find just the right laser (remember that TDLs are tunable over only a few nm) to measure their favorite molecule of interest. These sources and detectors could be cooled by on-board TEC coolers, and were thus much more robust and easily fieldable. This more than made up for the fact that NIR bands are combination and overtone bands (rather than fundamental absorptions), which can be several orders of magnitude weaker than the fundamental bands. The modulation techniques developed to reduce laser and detector noise with the mid-IR lasers remained important to increase sensitivity with the new near-IR lasers and detectors, due to the relative weakness of the transitions.

	A stiff competition developed between very strong development groups at Unisearch (Ontario, Canada), Physical Sciences, Inc. (Andover, MA), and Southwest Sciences (Santa Fe, NM), as well as with practically oriented academic groups such as those of Prof. Ron Hanson (Stanford University), Prof. Jürgen Wolfrum (Universität Heidelberg, Germany), and (later) Prof. Volker Ebert (Technische Universität Darmstadt, Germany). The telecommunications lasers were available only in particular bands, and manufacturing techniques were under development. Therefore, there were always a few lasers in each batch that were outside of the intended wavelength range. These "rejects" were snapped up by the competing researchers, and were immediately tested for application to molecules of interest, each researcher hoping to be the first to measure, say, hydrogen sulfide or ammonia. Relationships with the TDL manufacturers were key; otherwise lead times for particular wavelength lasers could be 3–4 months or more. Each of these groups made substantial contributions to the development of TDLAS spectroscopy as it is practiced today.

	Approximately a decade ago, the basic methodology, hardware, and range of gases that could be measured hit a tipping point for TDLAS technologies. Small business innovation research (SBIR) gave way to commercial application, and companies such as Siemens, Mettler-Toledo, Teledyne, General Electric, Servomex, and SpectraSensors joined Unisearch and other early companies in a now crowded field of manufacturers. A typical list of gases that can be easily measured includes O

, and many others. Typical cross-stack detection limits are in the parts-per-million range (depending on the molecule) for a few meters of pathlength. Sub-parts- per-million detection can be obtained for many gases with longer paths, including with extractive sample cells.

	Perhaps the most hopeful of the new entrants to the crowd of manufacturers was Zolo Technologies, which was founded in 1999, and venture-funded to the tune of $51.5 million. They developed TDLAS sensors and tomographic temperature and species mapping strategies using multiple TDLs in boilers. They were purchased by John Zink Hamworthy Combustion in March of this year. The current size of the worldwide market for TDLAS is in the hundreds of millions of dollars annually, and growing quickly.

	There are two basic sets of challenges for TDLAS technology. The first is to continue to adapt to an increasing number of challenging measurement environments. Coal mines, refineries, cement plants, and other harsh environments bring continuing operating challenges. This requires developing fit-for-purpose sensing devices that are easy to install, which retain calibration or are easy to recalibrate under heavy wear and continuous use, and which can measure a suitable set of gases for the application. These are typical challenges for a developing technology, and will result in a spate of new implementations, as well as solutions to interesting technical challenges. Some of the most important challenges to consider for new systems are effective multiplexing, beam steering and control, and compactness.

	The second set of challenges is to adapt to new laser sources. The near-IR TDLAS systems have been quite successful, with detection limits in the parts-per-million over short (~1–10m) pathlengths and relatively low costs. To improve on the technology, new room-temperature interband cascade lasers (ICLs) operating in the 3–6 µm region will allow access to more of the fundamental bands. Quantum cascade lasers (QCLs) are further out in wavelength (typically 5–12 µm), and may also be farther out in wide application because of the price associated with power, packaging, and stability requirements, but QCLs are also promising. Figuring out how to use these lasers in industrial environments has the potential to push detection limits lower by orders of magnitude as a result of the strength of transitions accessible in the mid-IR region. However, the chicken-and-egg problem of finding enough demand for the lasers to make them economical enough to implement may stand in the way, at least until the equivalent of the telecom boom of the 1990s comes to the mid-IR to drive ICL and QCL laser development. Stay tuned!

	The author is grateful to have had the opportunity to learn much about TDLAS while working in partnership with Dr. Mohammadreza Gharavi, who was his graduate student from 1999–2004, and from staff for several years thereafter, at the University of California, San Diego. Later, commercial developments at Photon Machines, Inc. with Kregg Philpott continued the education. We found many of the potholes on the road to eventual success in implementation!

	(2) G.R. Janik, C.B. Carlisle, and T.F. Gallagher, 

	(3) U.S. Patent 7,969,576 B1, Steven G. Buckley, Mohammadreza Gharavi, Marco Borchers, June 28, 2011.

	(6) H.I. Schiff, G.I. Mackay, and J. Bechara, 

Air Monitoring by Spectroscopic Techniques,

 Markus W. Sigrist, Ed. (John Wiley & Sons, New York, NY, 1994), pp 239-333.

 is the CEO of Flash Photonics, an affiliate associate professor at the University of Washington, and has started and advised numerous companies in spectroscopy and in applications of machine learning. He has approximately 40 peer-reviewed publications and 6 patents. His work in practical optical spectroscopy, such as LIBS, Raman, and TDL spectroscopy, dovetails with the coverage in this column, which reviews methods (new and old) in laser-based spectroscopy and optical sensing. Direct correspondence to: 

Lasers and Optics

Tunable diode laser absorption spectroscopy is hitting the mainstream. Here, we look at the history of the technique, the current state of the technology, and future challenges.

Tunable Diode Lasers for Trace Gas Detection: Methods, Developments, and Future Outlook

This column is the continuation of our previous installments dealing with the question of outliers. Here we consider what to do about an outlier, once one is detected.

	You have developed a set of data and have a reading that is suspected to be an outlier. You have applied one or more of the tests described in our previous column (1) and confirmed that the reading is, indeed, an outlier. Now what do you do?

	There are three actions you might consider taking:

	These possible actions could be considered the 

 approaches to dealing with the outlier. Other approaches can also be considered, and, depending on the circumstances, might be preferred. For example, if the outlier arises when calibrating a spectrometer for quantitative analysis (using chemometrics, of course, which is the default activity for all our columns), then the origin of the outlier could be either in the reference laboratory values or among the instrumental values. In either case, an alternative (nonstatistical) approach would be to identify the source of the discordant value (instrument or laboratory), and then engage in investigation of the chemistry, physics, and background of the readings to find the fundamental cause. In a pharmaceutical context, for example, this is what would be called 

	A variant of this approach is to concentrate all of one's attention on just the outliers and ignore the rest of the data, with a view toward learning some new fundamental science, the rationale being that some new and unexpected effect has created the discordancy in the values. An outlier can be an indicator of a scientific accident, of the sort that has led to discoveries such as quinine, the smallpox vaccine, X-rays, insulin, penicillin, Teflon, and the cosmic microwave background (2-4) (Don't hold your breath for this, but it does happen!). After all, to a lesser degree, we saw previously (5) that the presence of a set of outliers that persisted through changes of data transformations and other manipulations of the data was indicative of a previously unsuspected systematic effect influencing the data. The discovery of this effect gave us much information about the meaning of calibration transferability, and how to achieve it. General scientific principles come into play here. Is the effect reproducible? Can it be created or avoided at will? Does it lead to predictions of new phenomena? Is there a causal link between the discordant data and fundamental physics, chemistry, biology, or math? Can you do a controlled experiment (or more generally, what is sometimes loosely called a 

)? Is there a theory (from another science, such as chemistry, physics, biology) that can explain the findings? Do other scientists get similar results, or at least have you had someone check your work? Here we will avoid departing from our mission of describing the chemometric and statistical effects on the data, and so we will not pursue those other topics. However, the reader should keep these alternate considerations in mind; after all, not everything has a statistical explanation! We will now consider the possibilities listed earlier.

	In the context of using chemometric calibration to perform chemical analysis, the most common action taken when an outlier is identified is to delete the discordant reading. As we saw in the discussion of the nature of outliers (6), however, it seems likely that samples summarily rejected this way are sometimes (perhaps even often) misidentified as outliers, when in actuality, they are simply the extreme samples of the distribution.

	Alternatively, as described above, the distribution could be other than expected. If an outlier is identified based on an expectation of a normal (or at least a symmetric) distribution but the actual distribution is exponential, or χ

, or some other asymmetric distribution, then large values are to be expected and summary deletion is obviously an incorrect action. Another incorrect action would be to deal with the data as though the assumption of normality applied.

	A consequence of deleting outliers is that the data no longer represent a random sample. Given that the discordant observation is, almost by definition, extreme in one characteristic or another, deleting the outliers is a form of censorship of the data. In extreme cases, where multiple outliers are deleted from a data set, deleting the outlier is tantamount to fitting the data to the calibration model, rather than a model to the data. If more than two outliers appear to need deletion, then it is more appropriate to reconsider whether they are actually outliers of the data set, or evidence of samples from a different population than the majority of the samples, having been mixed into the data.

	A careful statistician will never simply delete observations, even with good reason. Even when some observations must be deleted from a data set, a careful worker will keep a record of the deleted samples: which ones they were, their values, and the rationale for deleting them. This record will allow the purported outlier to be added back into the dataset, if it turns out that the act of deleting it was erroneous. A set of good recommendations is available in short articles by Tom Fearn (7,8), describing common sense methods of evaluating suspected outliers and making good decisions about removing them from a data set.

	For example, manually entered data is notorious for being rife with data entry and data transcription errors. Modern spectroscopic instruments, even low-end ones, generally transfer data to the analysis computer electronically, so this sort of error should not occur. Reference laboratory values, on the other hand, may well be subject to being manually copied, or entered into a database, or transferred to the analysis computer by manual entry, possibly through multiple steps involving manual handling. Typos and other types of errors can occur under these conditions.

	Data transformations are common in spectroscopic calibration practice. This is often done in an ad hoc manner, using the shotgun method of trying every available or conceivable data transform until one of them works in some manner. While we can't completely denigrate this approach (if for no other reason than that we've used it ourselves on occasion in the past), we do wish to discourage it, and thereby introduce more science into what is currently much more of an art of calibration. Therefore when we discuss a technique, or a method, or an algorithm, we do so with the expectation (or at least the hope!) that it will be used only under the specific conditions for which it is suited, to achieve a specific predefined result.

	We noted in a prior column (9) that one reason to transform data is to convert the data from having a log-normal distribution to having a normal distribution by taking the logarithm of that data. In chemometric practice, this would typically correspond the reference laboratory values. Another scenario warranting a transformation of the data is when the error of the reference laboratory measurement is not constant across the range of the analyte, but increases in proportion to the analyte value; that is, the "relative error" (instead of the absolute error) of the reference method is constant, as happens in some reference laboratory methods. In this case, also, the reference values should be replaced with their logarithms; then the errors of those reference logarithm values will be constant, and then when the calibration calculations are performed, those logarithms are used as the dependent variables, instead of the original reference values.

	There is nothing wrong with transforming the reference laboratory values to their logarithms, performing the calibrations for these new (reference laboratory) values, and creating a calibration model for those new reference laboratory values. In using the model for analyzing future unknown samples, the computation will result in the transformed values being predicted by the model. The user must then remember to back-transform the result into the original domain of the data. The user must also take care to ensure that the data transform did not introduce an unwanted nonlinearity into the data.

	Another transform of the data is an extension of standard good statistical practice, to collect data from multiple aliquots (at least two, although three is preferred, and may be required in some cases, as will be described). If one of the multiple readings is found to be an outlier, then it should be compared with the other readings. If the readings differ by an amount comparable to the magnitude of the outlier, then it seems clear that one of the readings is incorrect, although it may not be obvious which one is wrong. If data from three or more replicates were collected, however, then a "vote" can be taken; the two that agree are most likely correct, and the one that disagrees, especially if it is the discordant reading, is probably wrong. It may not be clear why that reading is different, but this may be a case where a type of root cause analysis, as described above, might be warranted. If none of the three (or more) readings differs noticeably from the others, then it might be appropriate to average all the like readings together, prior to performing the calibration calculations. The data may also be "windsorized" by averaging together the highest and lowest values with their respective nearest neighbors, as a way to reduce the influence of a possible outlier. This may be ineffective if the data is multivariate, since the outlier may not correspond to any individual extreme reading.

	Accommodation of an outlier is not necessarily a way to identify it. It is often a technique for reducing or eliminating the effect of a possible outlier on the results of the calculations, whether any outliers are present or not.

	There are many methods of accommodating outliers. One that is rarely used but probably should be used more commonly, is to use what are called 

 methods of analysis. These robust methods replace our usual methods of calculation with methods that produce essentially the same results, but are resistant to the effects of outliers. The simplest example of a robust method is the use of the median instead of the mean, as the measure of the central value of a set of data. Given that the value of the median depends almost entirely on the numbers at or near the center of the data, the values at the extremes are virtually immaterial. The largest and smallest values can be correct or off by a large amount, but that difference has no influence at all on the value of the median. Contrast that with the mean, where the influence of the extreme values increases as the square of the difference between the mean and the values themselves, so that the incorrect extreme value has undue influence on the results. There are robust equivalents to most of the common chemometric methods that are popular, including multiple linear regression (MLR) and principal component regression (PCR).

	A somewhat less extreme way of accommodating the (potential) outliers is to use weighted calculations. This approach permits the use of the standard algorithms and methods of analysis. The key here is the fact that outliers are, in general, the extreme observations of some characteristic of the data. This being the case, the weighted variations of the standard algorithms are used, with the weights corresponding to the degree of extremeness of the pertinent characteristic: the more extreme the value of the characteristic, the lower the weight assigned to the sample containing that extreme value. This approach has the advantages of retaining and using all the data, while still minimizing the effect of the outlier on the results. Equations for weighted as well as unweighted calibration algorithms for implementing MLR, PCR, and partial least squares (PLS) algorithms are described in ASTM standard E1655-05 (10).

	As noted at the beginning of this column, deletion of an outlier is the most common way that most spectroscopists know to deal with them. As we have seen, however, there are a number of alternate methods available to take outlier readings into account in spectroscopic analysis. Add that to the fact that in much calibration work, a great excess of enthusiasm is shown in the deletion of purported or even suspected outliers, so we hope that having alternate methods available to deal with them can reduce the overenthusiasm for deleting outliers and preserve data sets.

http://www.pbs.org/wgbh/nova/body/accidental-discoveries.html

https://www.chemours.com/Teflon/en_US/products/history.html

https://www.aps.org/publications/apsnews/200207/history.cfm

	(10) E1655-05, A.; American Society for Testing and Materials, West Conshohocken, PA, USA.

 serves on the Editorial Advisory Board of 

. He is also an adjunct professor at National University in La Jolla, California. He has previously held a number of senior R&D positions within instrumentation and biotechnology companies.

 and runs a consulting service, Mark Electronics, in Suffern, New York. Direct correspondence to: 

Chemometrics in Spectroscopy

What are the steps to take once an outlier is discovered? There are several options.

Spectroscopy-10-01-2018

Spectroscopy October 2018 Issue PDF

The Critical Role of Atomic Spectroscopy in Understanding the Links Between Lead Toxicity and Human Disease

The 2018 Emerging Leader in Molecular Spectroscopy Award

Tunable Diode Lasers for Trace Gas Detection: Methods, Developments, and Future Outlook

Outliers, Part III: Dealing with Outliers

Quantitative Detection of Phenobarbital in an Injectable Solution Based on Surface-Enhanced Raman Spectroscopy

Market Profile: Handheld Spectroscopy