In life science research, detergents are primarily used in sample preparation to liberate cellular components through membrane disruption and to solubilize lipid-associated proteins. However, detergents can interfere with applications such as enzyme-linked immunosorbent assays, isoelectric focusing, nuclear magnetic resonance spectroscopy, and mass spectrometry, and therefore, often need to be removed before analysis. Here we report a novel infrared-based method that permits fast and impartial analysis of detergent removal from biological samples. Because this method is spectrally based and label-free, determination of protein concentration can be performed simultaneously. An additional benefit of this technique is the small volume (2 µL) required for analysis, a significant fact given the often precious nature of biological samples.

Detergents, water-soluble surface-active agents distinguished by their amphipathic structure, are used extensively in protein biochemistry. Principally, this large family of molecules can be used to isolate, solubilize, and stabilize membrane proteins (1–5). Depending on their structural properties, certain detergents promote solubilization and disaggregation of recombinant proteins during the process of extraction and purification (1), and others facilitate protein stabilization and crystallization (6). Detergents are also used to reduce background by minimizing nonspecific binding or protein precipitation in a wide range of immunoassays (2,4).  

All detergent molecules contain a hydrophilic head and a hydrophobic chain (or tail). This unique composition permits their spontaneous, but ordered aggregation in aqueous media resulting in the formation of stable micelles. In their pure form, detergent micelles contain from three to several hundred molecules, and the number of molecules in the micellar form (aggregation number) is an important characteristic of the detergent. Consequently, detergent micelles can vary greatly in size and the molecular weight of the micelle represents another useful parameter describing detergent properties. Micelle formation is initiated after a detergent content reaches a certain threshold termed the 

 (CMC). Generally, detergents must be used at concentrations above their CMC to be effective (1–4).   

The extraction of proteins from cells or tissue requires detergents to facilitate dissociation and solubilization. Although detergents are critical to this process, as well as numerous other preparative methodologies used in protein research, they can interfere with many downstream applications, thus limiting analytical sensitivity, so they have to be removed before the analysis (7–9). The most commonly used detergent removal methods include size-based exclusion (dialysis and gel filtration), hydrophobic adsorption, and ion-exchange chromatography (10). The choice of an appropriate method is dictated by inherent properties of the detergents used including hydrophobicity, CMC, micelle size, and charge.  

UV absorbance, an array of colorimetric methods, and mass spectrometry (MS) are current options to monitor and optimize the efficiency of detergent removal (8,11,12). Because of the presence of aromatic rings, the concentration of compounds, such as Triton X (100 and 114) and NP-40, can be estimated by absorbance at 275 and 280 nm (8). However, application of this method is very limited given that detergent's absorbance overlaps with that of the protein. Specifically, for complex biological samples, such as cell culture or tissue lysates, it is impossible to establish pre-existing levels of protein and distinguish the absorbance signals produced by detergent and protein. Total organic carbon (TOC) analysis, yet another method applied to monitoring detergent removal, is also confounded by protein interference and is thus restricted to measuring detergent levels in the flow-through (removed) fraction. The use of liquid chromatography–tandem mass spectrometry (LC–MS) or matrix-assisted laser desorption–ionization (MALDI) has been reported in the case of monitoring the removal of Tween 20 and BRIJ-35 (8). SDS concentration is typically measured by a colorimetric approach using Stain-All dye (11). Similarly, concentrated sulfuric acid and phenol (12) are frequently used for concentration estimations of glycosidic and bile salt-based detergents like octyl glucoside and CHAPS. In summary, none of the existing platforms or methodologies appears capable of broad-based detergent content measurement in biological samples. 

Here, we report on the development of a novel Fourier transform infrared (FT-IR)-based method for fast and impartial analysis of detergent concentration in biological samples. The method uses a hydrophilic polytetrafluoroethylene (PTFE) membrane engineered for sample retention and optimal transparency in regions of the IR spectra used for analysis of biological samples. Aqueous samples are applied directly onto the membrane, dried, and analyzed by FT-IR, with minimal volume requirement (2 µL). Because of the specificity of the absorbance bands, the technique offers simultaneous detection of multiple species, such as protein and detergents reported here, without interfering with one another. This dual capability is well suited for the optimization of protein preparative processes where the reduction of detergent content is ultimately required. 

All measurements were performed using a Direct Detect spectrometer (EMD Millipore), an FT-IR system analyzing the spectral range 500–6000 cm

To verify the application of FT-IR to monitor detergent removal, 0.5% deoxycholic acid sodium salt (EMD Millipore), 1% NP-40 (Sigma), 1× RIPA lysis buffer (EMD Millipore) prepared with Dulbecco's phosphate buffered saline (PBS, EMD Millipore), and CytoBuster protein extraction reagent (EMD Millipore) were used. In each case, 2 µL of sample solution was applied onto the membrane position; this volume was used for all concentration estimates. The spectrometer was initially calibrated with each of the analyzed detergents and both lysis buffers. Sample concentrations were determined in reference to their respective calibration methods. 

Bovine gamma globulins (Bayer) at 2 mg/mL in PBS and bovine liver tissue lysates (prepared with CytoBuster and RIPA lysis buffers), total protein content at around 3 mg/mL each, were used to demonstrate the application of the FT-IR-based method for simultaneous monitoring of detergent removal and protein recovery. Protein concentration in the original sample was used as a reference point for 100% recovery. Protein concentrations were determined in reference to a calibration curve developed using bovine serum albumin (BSA) from the National Institute of Standards and Technology (NIST) diluted in PBS (EMD Millipore). 

Size-based separation via centrifugal diafiltration was established as the method for detergent removal. For this study, Amicon Ultra (0.5 and 2.0) centrifugal filters (EMD Millipore) were used according to the manufacturer's recommendations. Briefly, 500 µL of investigated detergent or buffer solution was placed in AU-0.5 or AU-2.0 devices. In the case of the AU-2.0 device, the volume was adjusted to 2 mL by addition of 1.5 mL of PBS. The devices were spun at 14,000

 for 5–15 min (AU-2.0). The retentate volume from each device was adjusted to 500 µL by the addition of PBS; the remaining detergent concentration was estimated by FT-IR. For each sample, the above procedure was repeated twice for a total of three spin-removal and analysis cycles. Detergent removal was analyzed using three devices per each nominal molecular weight limit (NMWL; 10 kDa, 30 kDa, and 100 kDa).

Figure 1: Sample presentation to FT-IR beam: (a) MIR signature of PTFE membrane; (b) the design of sample spot allowing for retention of analyzed sample in the mid-IR beam (20Ã magnification); (c) distribution of the sample dried on the spot. Sample: 2 ÂµL of 4 mg/mL cytochrome c in CytoBuster protein extraction reagent (20Ã magnification).			

FT-IR analysis of aqueous-based biological samples is achieved by application of the analyte onto a hydrophilic PTFE membrane incorporated into a measurement card. The membrane spectrum contains a strong signal between 1100 and 1300 cm

 , but it is transparent in the IR region above 1300 cm

  used for the analysis of biological samples (Figure 1a). The hydrophilic spot for sample application (see Figures 1b and 1c) is surrounded by a hydrophobic ring generated by mechanical removal of the hydrophilic surface that retains an analyte permitting precise presentation of the entire sample to the IR beam (13,14). 

Figure 2: Overlap of the IR bands originating from stretching vibrations of C-H bonds collected for various detergents.			

FT-IR-Based Monitoring  of Detergent Removal

Because of their complex chemical compositions, detergents absorb in many different regions of the IR spectrum. The symmetric stretching vibrations of C-H bonds (2840–2870 cm

) provided an ideal candidate for detergent analysis and are utilized in the reported method (Figure 2). Comprehensive analysis of various detergents demonstrated a high degree of variability in the slope of the calibration curves (data not shown), strongly suggesting the requirement for individualized calibrations. 

Figure 3: Efficiency of detergent removal from PBS performed using Amicon Ultra 0.5 centrifugal devices (10, 30, and 100 kDa NMWL). Starting detergent concentrations were as follows: 1% NP-40, 0.5% sodium deoxycholate, 1Ã RIPA and 1Ã CytoBuster. Retentates were analyzed following each of three successive centrifugal "wash" cycles. The graph plots % detergent removal as a function of spin step.			

To verify efficacy of an FT-IR-based approach to monitoring detergent removal, the following solutions were employed: NP-40 (nonionic detergent), sodium deoxycholate (anionic detergent), RIPA buffer, and CytoBuster protein extraction reagent. The latter two buffers, each containing multiple detergent types, are commonly used for the extraction and solubilization of cellular proteins. Linear calibration curves for each of the analyzed detergents and lysis buffers were prepared before the analysis. Size-based separation via centrifugal diafiltration using Amicon Ultra 0.5 and 2.0 (NMWL; 10 kDa, 30 kDa, and 100 kDa) filters has been used as a detergent removal method of choice. Overall, detergent properties could have been used in predicting efficiency of the removal by size exclusion (Figure 3). Specifically, sodium deoxycholate, characterized by a relatively high CMC (2 mM) and small micelle size (MW = 1200–1500 Da), was effectively cleared to below the reliable instrument detection level (0.031%) in less than three centrifugation cycles. In stark contrast, the NP-40 detergent, which is characterized by relatively low CMC (0.29 mM) and high micelle size (MW = 90 kDa), could only be efficiently removed through use of the 100 kDa NMWL format. Interestingly, CytoBuster was removed below reliable detection level (0.016×) in three spins while RIPA was resistant to clearance by the two devices harboring smaller NMWL membranes. 

Figure 4: Simultaneous quantitation of protein and detergent (4 mg/mL BSA in the presence of 1% SDS). The top part of the box shows IR signal registered between 3500 and 1400 cm-1. The bottom part of the box shows a magnification of the areas of FT-IR spectrum characteristic to the protein and the detergent signals. The spectrum of the detergent (no protein) is shown in green. The FT-IR spectrum of protein in the detergent containing buffer is shown in blue.			

Simultaneous FT-IR-Based  Monitoring of Detergent  Removal and Protein Recovery

The primary structure of all proteins and peptides is defined by a chain of amino acids linked covalently via peptide (amide) bonds. Amide bonds absorb electromagnetic radiation in multiple regions of the IR spectrum, including a strong band at 1600–1690 cm

  ("Amide I"). The reported FT-IR method utilizes strength of the Amide I band to determine protein concentration. Because of spectral separation of the analysis regions used for detergent and protein quantitation, collection of a single FT-IR spectrum permits simultaneous analysis of both molecular species (Figure 4). 

Figure 5: Efficiency of detergent removal from IgG (starting concentration 2 mg/mL in PBS) containing solutions performed using Amicon Ultra 0.5 centrifugal devices (10, 30, and 100 kDa NMWL). Starting detergent concentrations were as follows: 1% NP-40, 0.5% sodium deoxycholate, 1Ã RIPA, and 1Ã CytoBuster. Retentates were analyzed following each of three successive centrifugal "wash" cycles. The graph plots relative cleanup performance (% protein retained/% detergent removed) as a function of spin step.			

Critical to the process of detergent removal is the need to retain protein, both content and concentration. Interestingly, the results obtained during the analysis of detergent removal from samples containing protein could not have been predicted based solely on properties of the detergent. Figure 5 clearly shows that when mixed with protein, in this example IgG, sodium deoxycholate is as difficult to remove by size-based diafiltration as NP-40 or RIPA. Within the analyzed buffer set, only the CytoBuster reagent offered effective detergent removal (~90% in three spins) without compromising protein retention. Moreover, the lysates required fourfold dilution (achieved through use of the larger AU-2.0 device) to produce noticeable detergent removal. With three spin cycles using the AU-2.0 device, CytoBuster removal exceeded 90% (Figure 6). However, detergent clearance was accompanied by a substantial loss of protein content (~60%), suggesting that for heterogeneous mixtures containing a wide range of molecular weight species, such as cell lysates, size-exclusion methodologies may not be the optimal approach for detergent removal.  

Figure 6: Efficiency of detergent removal from bovine liver lysates (starting total protein concentration 3 mg/mL) performed using Amicon Ultra 2.0 centrifugal devices (10, 30, and 100 kDa NMWL). 1Ã RIPA and 1Ã CytoBuster were used to prepare lysates. Retentates were analyzed following each of three successive centrifugal "wash" cycles. The graph plots relative cleanup performance (% protein retained/% detergent removed) as a function of spin number.			

FT-IR spectroscopy is a rapid, universal, and highly effective method for detection and analysis of biomolecules suspended in solution that is based on the unique spectral signatures of analyzed species. More specifically, the work presented here demonstrated the application of FT-IR for simultaneous determination of protein and detergent content in biological samples. Most notably, reported quantitation requires single-time instrument calibration with the detergent of interest permitting standardization across both sample types and processing runs. Protein quantitation is performed using preloaded, standard NIST BSA calibration. In addition, the reported method can be used in parallel with the majority of known detergent removal techniques. Overall, the results suggest that optimization of a sample preparation workflow, including detergent removal, are confounded not only by the nature of the starting material (cell or tissue lysate), but also by the detection requirements and sensitivities of the downstream analytical tools. 

(1) R.M. Garavito and S. Ferguson-Miller, 

(8) B.S. Antharavally, K.A. Mallia, M.M. Rosenblatt, A.M. Salunkhe, J.C. Rogers, P. Haney, and N. Haghdoost, 

(10) A.M. Seddon, P. Curnow, and P.J. Booth,

(11) F. Rusconi, E. Valton, R. Nguyen, and E. Dufourc, 

(13) D.R. Gagnon, R.M. Pieper, and J.E. Aysta, Patent US005764355A (1998). 

(14) E. Chernokalskaya, V. Joshi, P. Clark, C. Utzat, R. Amara, and T. Rider, U.S. Provisional Patent Application No. 61/475434 (2011). 

Ivona Strug, Sara Gutierrez, Amedeo Cappione III, Mayra Jimenez, Mary Jane Mullen, and Timothy Nadler

 are with the Bioscience Division of EMD Millipore Corporation, in Danvers, Massachusetts. Direct correspondence about this article to: 

Articles

Peer-Reviewed Article

A novel infrared-based method that permits fast and impartial analysis of detergent removal from biological samples is presented.

A Rapid FT-IR-based Method for Monitoring Detergent Removal from Biological Samples

This is our 100th "Chemometrics in Spectroscopy" column, and when including "Statistics in Spectroscopy," there are now a total of 138 columns. We began in 1986 and have been working since that time to cover in-depth discussions of both basic and difficult subjects. In this newest series, we have been discussing the subject of multivariate calibration transfer (or calibration transfer) and the determination of acceptable error for spectroscopy. We have covered the basic spectroscopy theory of spectroscopic measurement in reflection, discussed the concepts of measuring and understanding instrument differences, and provided an overview of the mathematics used for transferring calibrations and testing transfer efficacy. In this installment, we discuss the statistical methods used for evaluating the agreement between two or more instruments (or methods) for reported analytical results. The emphasis is on acceptable analytical accuracy and confidence levels using two standard approaches: standard uncertainty or relative standard uncertainty, and Bland-Altman "limits of agreement."

As we have discussed in this series (1–3), calibration transfer involves several steps. The basic spectra are initially measured on at least one instrument (that is, the parent, primary, or master instrument) and combined with the corresponding reference chemical information (that is, actual values) for the development of calibration models. These models are maintained on the original instrument over time, are used to make the initial calibration, and are transferred to other instruments (that is, child, secondary, or transfer instruments) to enable analysis using the child instruments with minimal corrections and intervention. We note that the issue of calibration transfer disappears if the instruments are precisely alike. If instruments are the "same" then one sample placed on any of the instruments will predict precisely the "same" result. Because instruments are not alike and also change over time, the use of calibration transfer techniques is often applied to produce the best attempt at model or data transfer. As mentioned in the first installment of this series (1), there are important issues of attempting to match calibrations using optical spectroscopy to the reference values. The spectroscopy measures the volume fractions of the various components of a mixture.  

Historically, instrument calibrations have been performed using the existing analytical methods to provide the reference values. These existing methods have often overwhelmingly used weight percents for their results. Until this paradigm changes and analysts start using the correct units for reporting their results we have to live with this situation. We will also have to recognize that the reference values may be some prescribed analysis method, the weight fraction of materials, the volume percent of composition, or sometimes some phenomenological measurement having no known relation to underlying chemical entities, resulting from some arbitrary definition developed within a specific industry or application. Amongst ourselves we have sometimes termed these reference methods "equivalent to throwing the sample at the wall and seeing how long it sticks!" The current assumption is the nonlinearity caused by differences in the spectroscopy and the reported reference values must be compensated for by using calibration practices. This may not be as simple as presupposed, but requires further research.  

Multivariate calibration transfer, or simply calibration transfer, is a set of software algorithms, and physical materials (or standards) measured on multiple instruments, used to move calibrations from one instrument to another. All the techniques used to date involve measuring samples on the parent, primary (calibration) and child, secondary (transfer) instruments and then applying a variety of algorithmic approaches for the transfer procedure. The most common approaches involve partial least squares (PLS) calibration models with bias or slope corrections for predicted results, or the application of piecewise direct standardization (PDS) combined with small adjustments in bias or slope of predicted values. Many other approaches have been published and compared, but for many users these are not practicable or have not been adopted and made commercially available for various reasons.  

In any specific situation, if the prescribed method for calibration transfer does not produce satisfactory results, the user simply begins to measure more samples on the child (transfer) instrument until the model is basically updated based on the child or transfer instrument characteristics. We have previously described the scenario in which a user has multiple products and constituents and must check each constituent for the efficacy of calibration transfer. This is accomplished by measuring 10–20 product samples for each constituent and comparing the average laboratory reference value to the average predicted value for each constituent, and then adjusting each constituent model with a new bias value, resulting in an extremely tedious and unsatisfying procedure. Such transfer of calibration is also accomplished by recalibration on the child instrument or by blending samples measured on multiple instruments into a single calibration. Although the blending approach improves method robustness (or ruggedness) for predicted results across instruments using the same calibration, it is not applicable for all applications, for analytes having small net signals, or for achieving the optimum accuracy. 

How to Tell if Two Instrument Predictions, or Method Results, Are Statistically Alike

The main question when comparing parent to child instrument predictions, a reference laboratory method to an instrument prediction, or results from two completely different reference methods, is how to know if the differences are meaningful or significant and when they are not. There is always some difference expected, since an imperfect world allows for a certain amount of "natural" variation. However when are those differences considered statistically significant differences, or when are the differences too great to be acceptable? There are a number of reference papers and guides to tell us how to compute differences, diagnose their significance, and describe the types of errors involved between methods, instruments, and analytical techniques of many types. The analytical method can be based on spectroscopy and multivariate calibration methods, other instrumental methods, or even gravimetric methods. We have included several of the most noted references in the reference section of this column. One classic reference of importance for comparing methods is by Youden and Steiner (4). This reference describes some of the issues we will discuss in this column as well as details regarding collaborative laboratory tests, ranking of laboratories for accuracy, outlier determination, ruggedness tests for methods, and diagnosing the various types of errors in analytical results.  

Let us begin with a set of measurement data as shown in Table I. This is simulation data that is fairly representative of spectroscopy data following multivariate calibration. These data could refer to different methods, such as methods A, B, C, and D; or to different instruments. For our discussion we will designate that the data is from four different instruments A to D for 20 samples. The original data from a calibrated instrument is A1. The results of data transferred to other instruments is represented by B, C, and D. There are duplicate measurements for A as A1 and A2, and for B as B1 and B2, respectively. From these data we will perform an analysis and look for levels of uncertainty and acceptability for the analytical results. Note: C1 and D1 data are used in Figure 2 and will be referred to along with A2 and B2 in the next installment of this column. 

Table I: Data used for illustration, instruments (or methods A, B, C, and D)			

Standard Uncertainty and Relative Standard Uncertainty

First, we look at the definitions of uncertainty as described by the United States National Institute of Standards and Technology (NIST), a National Metrological Institute (NMI), which is a nonregulatory agency of the United States Department of Commerce. The lion's share of NIST's purpose is to advance measurement science, measurement standards, and measurement technology. Their charter is to define measurements from first principles that can be verified world-wide and used as standards for making measurements of any kind related to commerce or technology. The NIST definition for 

 is quite specific, as explained below (5,6).  

 is the estimated standard deviation of y.  

If the probability distribution characterized by the measurement result 

) is approximately normal (Gaussian), and 

) is a reliable estimate of the standard deviation of 

) is expected to encompass approximately 68% of the distribution of values that could reasonably be attributed to the value of the quantity 

 is an estimate. This implies that it is believed with an approximate level of confidence of 68% that 

). The use of a concise notation if, for example, 

 = (1 234.567 89 ± 0.000 11) U. A more concise form of this expression, and one that is commonly used, is 

 = 1 234.567 89(11) U, where it is understood that the number in parentheses is the numerical value of the standard uncertainty referred to in the corresponding last digits of the quoted result.  

For the data in Table I we simplified our comparison by selecting only A1 and B1 data noting that one might compare multiple analysis for multiple instruments as a more powerful test of the estimated standard deviation of a measurement result for any method or instrument combination. However, the approach shown here is adequate for estimates of uncertainty for typical analysis and calibration transfer applications. We note, using the NIST nomenclature, that 

 is the estimated analytical value for any sample as a function of a series of measurement quantities such as 

 is an independent observation (or measurement). We also note that when using this nomenclature the A1 and B1 measurements for each sample are denoted as 

) for each sample measurement is estimated as the sample mean from 

 independent measurements and is denoted as 

, giving us the relationship as follows. 

 (that is, the number of instruments, methods, or models being compared). So our estimated analytical value (

) is the mean for a number of measurements of that sample (

) using the analytical method prescribed. And it follows that the standard uncertainty 

) with reference to the measured values (

) is equal to the estimated standard deviation of the mean. 

So to apply this to our data illustration from Table I we use A1 and B1 as 

) is the estimated standard uncertainty for a series of measurements on multiple samples where the mean value of the measurements for each sample is used for comparison. The equations above are often used for multiple measurements of a single physical constant. For our application using only an A1 and B1 measurement for each sample, we compute the variance for each of the 20 samples and pool the results to give us our estimate of standard uncertainty 

Compiling these results yields the following for the standard uncertainty reported as 

| and so applying the previously computed results we note that 

| = 14.648 as the mean of all A1 and B1 values. Therefore, the 

) = 0.226 and so the relative standard uncertainty is reported as 

The confidence levels would be expressed as follows: 

So the expression of certainty for this example of data would be as 

Thus, for the analyst the final step is deciding if this is a satisfactory result across the two instruments. This concept requires much more discussion, one which we will present in a future column. 

For a second look at the data in Table I, we refer to one of the definitive papers used for method comparisons within the highly regulated clinical sciences. We introduce the Bland-Altman method rather than a possibly more familiar calculation because Bland-Altman is the standard for clinical data in an industry with stringent requirements for analytical results. There is an entire series of publications by these statisticians and this method is still taught and used for clinical analysis criteria. One of the possible reasons there are no approved clinical methods using near-infrared (NIR) spectroscopy is that it does not stand up to critical analysis requirements and the NIR groups look at data using unconventional terminology and methods that do not stand up to serious analytical scrutiny. With 27,000 citations in the literature, and having it included in standard university bioengineering curriculum, this would be considered a more standard approach in the wider analytical community. The NIR and chemometric communities would do well to learn from these older and thoroughly tested methodologies and referenced papers.  

The paper we cite is entitled "Statistical Methods for Assessing Agreement Between Two Methods of Clinical Measurement," and was written by J. Martin Bland and Douglas G. Altman (7). This paper describes the errors often made when comparing methods. The authors summarize the contents of this paper as follows, "In clinical measurement comparison of a new measurement technique with an established one is often needed to see whether they agree sufficiently for the new to replace the old. Such investigations are often analyzed inappropriately, notably by using correlation coefficients. The use of correlation is misleading. An alternative approach, based on graphical techniques and simple calculations, is described." So, what can be learned by using the techniques described in this paper to compare results from analytical methods? When methods are compared following calibration, one attempts to assess the degree of agreement between them. Bland and Altman discount completely the use of correlation as a useful parameter to assess analytical agreement. Their arguments are given in this discussion.  

 Figure 1: Instrument or method A1 (x-axis) as compared to instrument or method B1 (y-axis), and with data points compared to perfect line of equality.			

For this method comparison, two single measurements are taken for each sample on each instrument as A1 and A2. The first measurement is used for comparison and the second for a repeatability study (which we will cover in a later installment). For this method of comparison a line of equality plot is made to compare two methods or two instruments. The various 

 data points are plotted against a perfect straight line of equality. The authors make the point that correlation (

) measures the strength of the relationship between two variables, but it does not measure the agreement between them (Figure 1). Perfect agreement is indicated by the data points lying directly on the line of equality. A perfect correlation is indicated if all the points lie along any straight line. The authors emphasize that: correlation indicates the strength of a relationship between variables — not that the analytical results agree; a change in scale does not affect correlation, but drastically affects agreement; correlation depends on the range of the true quantity (analyte) in the sample; tests of significance are mostly irrelevant between two similar analytical methods; and data in poor agreement analytically can be highly correlated (Figure 2). Figure 2 shows three analytical sets with perfect correlation but poor agreement. 

Figure 2: Instrument or method A1 (x-axis) as compared to instruments (or methods) C or D (y-axis), showing line of equality for perfect agreement as well as other perfect correlations that are not in analytical agreement. This indicates correlation as an imperfect representation of agreement between methods. 			

A Bland-Altman plot shown in Figure 3, which will be extremely familiar to clinical analysts, demonstrates a good visual comparison technique to evaluate the agreement between two methods or instruments. The 

-axis (abscissa) for each sample is represented by the average value for each sample obtained from the comparison results (using two methods or two instruments). The 

-axis (ordinate) for each sample is represented by the difference between one method and the second method (or instruments A and B in our example) for each sample. Such a plot uses the mean and plus or minus two standard deviations as the upper and lower comparison thresholds. 

Figure 3: The Bland-Altman plot indicating the difference plotted against the mean for each sample for instrument (or methods) A1 and B1. (Note: For random, normally distributed data we would expect one sample out of 20 to be outside the Â±2 SD range.) 			

To assess if the data are in close enough agreement for our analytical purposes between A1 and B1 we compute the bias or mean difference (

(withbar)), the standard deviation of the differences (

 or SD), and the expected "limits of agreement." These are expressed as 

The mean difference is computed as the average of all the differences between the comparative instruments, for each sample as 

The standard deviation for this comparison for the set of samples is computed as 

And so for our data A1 and A2 in Table I we find the following results: Bias is 

(withbar) = -0.226; standard deviation for all samples is 

Chemometrics in Spectroscopy

The statistical methods used for evaluating the agreement between two or more instruments (or methods) for reported analytical results are discussed, with an emphasis on acceptable analytical accuracy and confidence levels using two standard approaches, standard uncertainty or relative standard uncertainty, and Bland-Altman "limits of agreement."

Calibration Transfer, Part IV: Measuring the Agreement Between Instruments Following Calibration Transfer

The performance of a Raman spectrograph for a particular application will depend, among other things, on its sensitivity and spectral resolution. The sensitivity will determine how long it will take to record a spectrum with a given signal-to-noise ratio. In turn, the grating reflectivity will determine the optical throughput of the instrument. The spectral resolution will determine how easy it will be to extract subtle information from a spectrum. The spectral resolution is determined by the focal length of the spectrograph and the groove density of the grating used to disperse the light, and will also affect the apparent sensitivity. Note that in many cases, spectral resolution may be improved at the expense of sensitivity. Because many new users of Raman equipment are not familiar with these grating–spectrograph properties, we thought it would be useful to summarize the physics, in simplistic terms, that determine how the instruments work.

Over the past 20 years Raman spectroscopy has gained popularity because instrumental innovations have made it easier to use them for problem-solving, providing spectra in 1% of the time that it took prior to the Raman revolution following the introduction of the holographic notch filters. In addition, new graduates of chemistry and materials science are taking jobs in industry without any graduate education in spectroscopy or the operation of spectroscopic instrumentation. Consequently, the new user is faced with myriad choices in configuring a new instrument or in optimizing an existing instrument for a given experiment. Understanding how the spectrograph core of a Raman instrument works will aid the novice in producing quality, defensible results for solving industrial problems or characterizing new materials.   

A spectrograph is designed to accept light with many wavelengths, separate the wavelengths in space, and then "detect" each wavelength on a multichannel detector, which today is synonymous with a charge-coupled device (CCD). Figure 1 is a schematic of a spectrograph.   

 Figure 1: Schematic of a dispersive Raman spectrograph.			

Figure 1 shows a typical Raman spectrograph. The collected Raman light is focused onto an entrance slit. After passing the slit, it diverges until it reaches a concave mirror whose focal length corresponds to the distance between the mirror and slit; after being reflected by the mirror, the light is "collimated." When the light hits the grating, which is an array of finely spaced lines on a reflective surface, there is constructive and destructive interference, which is wavelength and angle dependent. Consequently, each wavelength is reflected at a different angle (1). As each wavelength is then reflected from the camera mirror onto the array detector (CCD), it is focused at a different position on the array. The wavelength of the light on each array pixel can then be calculated from the known equations of grating physics, as shown in Figure 2 (2).   

Figure 2: Schematic showing how a pixel position is converted to a wavelength.			

It is not the purpose of this article to derive the equations that determine the conversion, but only to indicate to the new user the origin of the "magic" inside the software that enables the spectral "image" on the camera to be converted to a spectrum, that is, a plot of intensity (counts or counts/s) versus Raman shift (cm

). The physical quantities determining the separation on the camera of two wavelengths will be the incident angle of the light on the grating, the diffracted angles, as determined by these equations, and the focal length of the focusing element (2). When a Raman instrument is designed, the spectral dispersion at a given wavelength is selected, and then the angles and focal length are calculated to produce the desired result. In addition, there is optical software that enables asymmetrizing the geometry so that the images are kept as tight as possible on the surface of the detector, which, of course, is flat.  

Of course, the Raman spectrum is only meaningful when the wavelength values are converted to Raman shift units, also in the software, according to equation 1:  

where ν is derived from λ according to equation 2: 				

Typical widths of lines in a Raman spectrum are between 1 and 10 cm

  full width at half maximum (FWHM). If parameters are selected that produce 1 cm

  can be covered on a CCD that has 1024 pixels in the long direction, which is the spectral dispersion direction. In principle, the selection of the grating would be straightforward, but as we will see, there are important characteristics that have to be acknowledged in the choice.  

Just to get oriented to these effects, examine the behavior shown in Figure 3 of the 155 cm

  line of sulfur that was recorded with the 633-nm line of a HeNe laser on instruments whose focal length varied between 150 mm and 1920 mm. Depending on the goal of the measurement, it may or may not be important to resolve the components in the spectrum; how much will be necessary to resolve will determine the selection of gratings as will be discussed in the following sections. 

Figure 3: 155 cm-1 line of crystalline sulfur recorded with instruments whose focal length varied between 1920, 640, 460, and 250 mm (from top to bottom). (Courtesy of Sergey Mamedov of Horiba Scientific.) 			

The important grating characteristics that have to be matched with the desired instrument characteristics are 					

the dispersion, which is a result of the groove density (g/mm) and focal length, and 

the reflectivity, which is a complicated function of the groove density, the groove profile (including the blaze angle), the angle at which the grating is used, and the metallic coating (3,4). 

It is important to recognize that because the relevant 

, the dispersion is wavelength dependent, which will be illustrated in Figures 4a and 4b. Figures 4a and 4b, respectively, show dispersion curves for the same gratings in a 300-mm focal length spectrograph, in terms of nm and cm

Figure 4: Dispersion in (a) nm/pixel and (b) cm-1/pixel for an 1800- and 600-g/mm grating in a 300-mm focal length spectrograph.			

We first examine the dispersion in wavelength units (nm/pixel). The dispersion curve for the 600-g/mm grating over the entire range shown (200–1775 nm) is almost flat, varying between 0.144 and 0.138 nm/pixel. On the other hand the behavior of the other grating is quite different.  For all values of λ > 1080 nm, the dispersion is pinned to 0 nm/pixel. At these long wavelengths, a grating with this value for the groove density will not diffract. To put it another way, at about 1080 nm the light exits the grating at a grazing angle and the diffraction angle for any longer wavelengths will be imaginary. In addition, over the usable range of the grating the dispersion is more or less flat between 200 and 600 nm, after which the dispersion decreases at an accelerating rate. But all of these dispersion values are in nm/pixel, and what we really want is cm

/pixel, which is shown in Figure 4b. To understand why these curves are so different, we need to calculate cm

/Å which has a conversion factor of -1/λ

. Table I illustrates how rapidly this factor is changing with wavelength. 

Table I: Dispersion in cm-1/Ã as a function of wavelength			

Inspection of Figure 4b indicates that the dispersion in cm

/pixel is changing rapidly for both gratings. If one is configuring a system for ~3 cm

/pixel in the red (λ~ 600 nm), a 600-g/mm grating will work quite fine, but if there will be a laser emitting near 400 nm, the 1800-g/mm grating would provide better dispersion.   

Note that these curves are specific for a particular focal length mono. In Figures 5a and 5b we compare a 300-mm focal spectrograph to an 800-mm focal spectrograph. An important point to note is that the long wavelength termination is the same in both systems because it is determined by the angle for grazing exit diffraction, an inherent property of each grating, not the focal length of the spectrograph in which the grating is being used. 

Figure 5: Dispersion in cm-1/pixel of 2400- and 3600-g/mm gratings mounted in (a) a 300-mm focal length spectrograph and (b) an 800-mm focal length spectrograph.			

Aside from the fact that the long wavelength cut-offs are the same in the two systems, it is also important to note that the range of use for these gratings is somewhat limited in the visible part of the spectrum. The 2400-g/mm grating will diffract out to almost 800 nm, but the 3600-g/mm grating will not diffract much beyond 500 nm. These gratings are, in fact, usually used in the blue and UV part of the spectrum where their dispersion is so much better than that of lower groove density gratings. Table II enables an easy comparison of the dispersion per pixel at 300 nm for the two gratings in the two systems. 

Table II: Dispersion/pixel at 300 nm for a 300 versus 800 mm focal length spectrograph equipped with 2400 and 3600 g/mm gratings			

Keeping in mind that the sharpest feature that can be observed has a half width of 2 pixels, one can see how a short focal length instrument can begin to limit what can be differentiated when exciting in the UV. If we go to an even shorter wavelength, the limiting resolution on a short focal length system will probably not be adequate for many studies, as shown in Table III. 

Table III: Dispersion/pixel  at 200 nm for a 300 versus 800 mm focal length spectrograph equipped with 2400 and 3600 g/mm gratings			

For the reader interested in a more detailed, but practical explanation of how a spectrograph functions, it can be found in 

The grating reflectivity is a complicated function of the groove shape and spacing, and the polarization of the light hitting the grating. The article cited earlier (3) describes in detail the physics determining the grating behavior. Until the middle of the last century, at about the time when lasers became available, gratings were produced by cutting a metal surface with a diamond-tipped ruling engine. Such equipment produced flat facets on the grating. Within a few years of the introduction of the laser it was recognized that it is possible to produce gratings by exposing surfaces coated with photoresist to interfering laser beams. Major advantages of gratings produced with this technology included a virtual elimination of grating ghosts arising from defects in the ruling engine, and much higher groove densities. What was perhaps not recognized with the earliest gratings was that the quasi-sinusoidal groove profile produced by the holographic process exaggerated the ripples in the reflectivity curves called 

. However, the subsequent use of ion etching provided a means to engineer the profile to produce high reflectivity in the region of interest (4).   

Figure 6 shows the reflectivity curve for an 1800-g/mm holographic grating that has been optimized for use between 450 and 850 nm. There are three curves on this figure. The green curve is labeled TM and shows the reflectivity for light polarized perpendicular to the grooves of the grating. The red curve is labeled TE and has been measured with light polarized parallel to the grooves. The blue curve is the average of the two and represents unpolarized light. Note that equipment manufacturers often select a grating where the TM curve is higher in the region of interest. But, if the user is aware of the grating properties and what kind of polarization the sampling is presenting, a grating can be used on the short wavelength side of the cross-over between TE and TM. For this grating the ripples that were mentioned above appear between 300 and 425 nm. In most cases they will not present a problem to the spectroscopist, but an instrument response correction will eliminate all wavelength-dependent sensitivities. 

Figure 6: Efficiency curve for an 1800-g/mm grating.			

The holographic 1200-g/mm grating presented a surprise when it first appeared. The groove density was convenient for the desired dispersion, but it was implemented before severe problems in the Wood's anomalies were observed. The original non-optimized holographic grating had an extremely sharp anomaly where the reflectivity dropped to essentially 0% near 650 nm. Apparently, the groove profile and diffraction angle were such that all the diffracted light grazed off the grating at a particular wavelength. But this was subsequently corrected by optimizing the groove profile for wavelengths of interest; the optimization had the effect of softening the anomaly and moving it to other wavelength regions. Figure 7 shows the efficiency profiles for several 1200-g/mm gratings that have been optimized for different wavelengths. Visual inspection illustrates how effectively the ion-etching has optimized the reflectivity for different wavelength regions. But it should be noted that all three of these gratings still have the anomaly near 650 nm, which represents about a 10% drop in intensity. 

Figure 7: Reflectivity curves for three 1200-g/mm gratings optimized for different regions of the spectrum. The top grating has been optimized for 750 nm, the middle grating for 630 nm, and the bottom for 500 nm.			

Comparison of Spectra Recorded with Different Gratings

After all this somewhat theoretical discussion it would probably be appropriate to illustrate the effects of selecting different gratings for a measurement. Figure 8 shows the spectrum of a slightly fluorescent paper recorded with 300-, 600-, 1200-, and 1800-g/mm gratings on the Aramis (460-mm focal length) using the 532-nm laser. The top of Figure 8 shows the spectrum, as recorded with the different gratings. The middle of the figure shows the same spectrum after the removal of the baseline. The bottom of the figure shows a smaller region of the middle figure to better estimate the relative intensities and to visualize the noise. First, I want to point out that I recorded the spectra with acquisition times meant to compensate for the differences in photon flux per pixel that follows from the differences in dispersion with the various gratings. Because there are other factors effecting the intensities (such as grating reflectivity and aperture at which the grating is being used), this adjustment in acquisition time does not totally compensate for the differences in dispersion. But what is potentially more interesting is the change in relative intensity with dispersion. For instance, Table IV shows the changes in intensities of the lines at ~1080 and ~1600 cm

  as a function of groove density. The fact that the band intensities are increasing at different rates is a result of the different bandwidths, and the fact that one band actually is composed of overlapping components; as the dispersion decreases each pixel is integrating a greater number of wavenumbers.  

Table IV: Approximate change in intensities of a sharp (~1080 cm-1) and somewhat broad band (~1600 cm-1) as a function of grating groove density			

This can have some rather interesting implications for the fluorescence background. Inspection of the top of Figure 8 indicates that the background increases by more than a factor of 10-fold when comparing the 1800-g/mm spectrum to that of the 300-g/mm spectrum! In addition, when there is such a high background it becomes difficult to differentiate between weaker Raman features and artifacts in the background. The background-subtracted spectra in Figure 8 illustrate this point because a simple background subtraction cannot eliminate the random and pattern noise in the background. 

Figure 8: Raman spectra of slightly fluorescent paper recorded with 532-nm laser on the Aramis (460 mm focal length), using 300-, 600-, 1200-, and 1800-g/mm gratings, adjusting the integration times to scale with the groove density (1, 2, 4, and 6 s). Top, as recorded; middle, after baseline subtraction, with vertical displacement for clarity; bottom, after baseline subtraction, expanded region.			

I should point out that Figure 8 shows that complete spectra were recorded for all gratings. That means that even though a particular grating–laser combination will provide 

  in a single shot, the full spectra were recorded by scanning the entire range of interest — in this case, 100–4000 cm

. That means that there is no loss in capability when using a high groove density grating for higher spectral resolution. This figure also shows that the broad background does not scale with the same factor as a sharp band when changing grating (that is, dispersion). We were previously aware that the ratio of a sharp to broad band would change when the dispersion of the recording instrument is changed, but were surprised to see the same effect on the ratio of sharp band to (very) broad background. What this means is that when recording spectra of not-so-clean samples, the expected sensitivity advantage from recording low resolution spectra may be cancelled by the rapid increase in background! While it is always possible to subtract the background levels, the noise does not subtract, as can be seen in the middle and bottom traces in Figure 8. It would be wise to keep this in mind when dealing with samples with interfering backgrounds — it is always best to use conditions that will enable collecting spectra without a background rather than try to deal with it after-the-fact. For those interested in this point, we expect to deal with this more fully sometime in the near future.   

This column has been written to enable the analyst who is new to spectroscopy to understand what the considerations are in choosing a grating or focal length for a Raman spectrograph. Maybe you do not understand all of the grating physics that determines the grating functionality, but you should at least understand what parameters the choice of grating will optimize. Of course, the selection of laser wavelengths is independent of this, and actually has to happen before an intelligent choice of gratings can be made. I would love to hear from any of you about how useful this has been.   

Molecular Spectroscopy Workbench

The physics that determine how gratings and spectrographs work are summarized in simple terms for new users of Raman equipment.

Considerations of Grating Selection in Optimizing a Raman Spectrograph

At the fringe of the infrared spectrum is the terahertz region, which until recently was not significantly utilized for spectroscopy. The properties of terahertz spectroscopy make it useful for applications in a variety of industries, with many major companies already adopting the technique. The entrance of new competitors and very strong growth of the technique portend its growth into a major segment of the molecular spectroscopy market. 

Laboratory terahertz spectroscopy demand by industry for 2012.

The terahertz region of the electromagnetic spectrum lies between the far infrared region and the microwave region, and includes wavelengths of about 0.1–1.0 mm. Terahertz spectroscopy is somewhat complementary to other spectroscopy techniques, as a number of compounds have unique fingerprints in the terahertz region, including some narcotics, explosives, and various polymorphic forms of active pharmaceutical ingredients (APIs). A variety of common materials, as well as human tissue, are semitransparent to terahertz radiation, and it is non-ionizing, making it safe for human exposure. A wide variety of commercial terahertz spectrometers are already on the market, including conventional frequency-domain systems, time-domain systems, imaging systems, and portable instruments. 

For laboratory applications, terahertz spectroscopy is quickly being adopted by major companies for both quality analysis and product development. Intel's adoption of the technique validates its usefulness in the semiconductor and electronics industries. A number of major pharmaceutical manufacturers are now using the technique for both development and QA–QC of solid dosage forms. There is significant exploration of terahertz spectroscopy for clinical and medical applications, although much of this work is currently based in academic laboratories. 			

Global laboratory demand for terahertz spectroscopy was nearly $20 million in 2012, with at least six major competitors now offering commercial instruments. Although 2013 may be a challenging environment for this market, demand is still likely to see mid single-digit growth before returning to strong double-digit growth in 2014. 

The foregoing data were based on SDi's market analysis and perspectives report entitled 

Global Assessment Report, 12th Edition: The Laboratory Life Science and Analytical Instrument Industry

, October 2012. For more information, visit 

Departments

At the fringe of the infrared spectrum is the terahertz region, which until recently was not significantly utilized for spectroscopy.

Market Profile: Terahertz Spectroscopy

The contamination of apple juice and rice with potentially harmful arsenic compounds has been highlighted in the news recently, including in a November 2012 article from 

 (1). For most people, rice is far and away the major source of potentially harmful arsenic compounds in our diet. Relevant agencies have recently issued revised guidelines for human consumption, and there is discussion concerning possible regulation of the arsenic content of foods. 

Analytical methods capable of distinguishing between inorganic and methylated compounds are necessary to support any such regulation, because arsenic compounds vary widely in toxicity. Recent articles in the analytical chemistry literature illustrate the difficulties of making reliable measurements of the arsenic compounds in rice. Sample preparation is challenging, there is no agreement over the best chromatographic conditions, compound-dependent responses are evident, validation is difficult (if not impossible), and there is ambiguity over the severity of starch matrix effects. In a recent web seminar, Julian Tyson, professor of chemistry at the University of Massachusetts, Amherst, explained how to develop a reliable method for arsenic speciation. Below, he answers questions raised during the web seminar. A recording of the web seminar titled "Speciation Analysis: A Critical Look at Methods Involving HPLC with ICP-MS Detection, with a Focus on Rice" is available for free at 

http://www.spectroscopyonline.com/ArsenicRiceWebinar2013

What is the source of the arsenic that is present in rice?

 The arsenic compounds are either in the soil or in the irrigation water. The compounds in the soil already may include a significant residue from past agricultural practices, which, in the case of cotton growing in the United States, involved spraying with dimethylarsinate (cacodylic acid), monomethyl arsenate, and arsenic acid.   

Could genetically engineered rice reduce the amount of arsenic in rice?

I think that genetic engineering could help. In fact, Professor Om Parkash here at the University of Massachusetts is working on this. You can read more at 

http://www.bio.umass.edu/plantbio/faculty/parkash.html

Given that inorganic arsenic compounds are polar, would they be extracted from the rice (into the water) during cooking? 

Yes, all of the arsenic species are extracted by hot water. The problem is that in standard rice cooking methods, the cooking water is absorbed into the rice and thus eaten. Cooking in excess water and then discarding the water will decrease the amount of arsenic compound consumed. More details appear in a 2009 paper by Raab and colleagues (2). 

How much monomethylarsonate (MMA) do you expect to find in rice?

I expect to find just single-digit microgram-per-kilogram levels of arsenic as monomethylarsonate. Almost all the reports of arsenic speciation in rice from around the world show that MMA amounts to no more than a few percent of the total arsenic. Often, it cannot be detected at all. The one possible exception is Korean rice. See Paik and colleagues (3). 

What rice certified reference material is available for the arsenic species? 

According to the web site of the National Institute of Standards and Technology (NIST), both the rice and wheat flour reference materials are out of stock. Fortunately, as was pointed out by a member of the web seminar audience, the National Metrology Institute of Japan (NMIJ) has issued material CRM 7503-a, Arsenic Compounds and Trace Elements in White Rice. CRM 7503-a is certified for arsenite, arsenate, and dimethylarsinate — which account for 97 of the 98 μg/kg total. Inorganic arsenic accounts for 86% of the total.  This standard is a white rice flour (whereas the NIST material was a brown rice flour, I believe). The Japanese reference material is available in the United States from Wako Chemicals (

), with a lead time of about three weeks.   

Even if you do buy this material, however, you should also measure spike recoveries. The spikes should be added before the sample digestion–extraction. 

What would be the best overall method for measuring total inorganic arsenic, including the digestion and other sample preparation steps? 

If you are not interested in differentiating between arsenite and arsenate, try a procedure in which peroxide is a component of the extraction reagents. Grind to less than 200 μm and do not exclude any particles by sieving. Heat under pressure.  Degrade the starch with trifluoracetic acid and centrifuge (filtration not needed). Then select a high performance liquid chromatography (HPLC) procedure in which arsenate is eluted first, so that you are not waiting for the other components to be eluted first. Use isocratic elution, but investigate the effect of increasing the temperature (if you can). 

Could you describe the interface between the liquid chromatography (LC) and the inductively coupled plasma (ICP) instruments?

Typically, all that is needed to connect these two instruments is an appropriate length of PEEK tubing. The flow rate of most LC separations is compatible with that of conventional ICP nebulizers, so the column eluent can be delivered directly to the ICP nebulizer. We have used miniature spray chambers in the past in an effort to minimize extracolumn broadening, but the effects are relatively minor and currently we use the standard nebulizers and spray chambers that come with our ICP instrument. 

Can you please explain why gradient elution is a problem? I am using ion chromatography (IC).  

The problem that can arise with gradient elution is that the change in mobile-phase composition changes the conditions in the plasma, so that the signal for a given flux of arsenic going into the plasma changes, and so you see what appears to be compound-dependent responses, but those are probably mostly mobile-phase composition dependent responses. Once you have decided on your gradient, you will almost certainly need to calibrate the method by injecting a mixture of the species you want to determine and creating separate calibration curves for each species. How many standards to use is your decision. We usually check that the calibration is linear and then inject one (or maybe two) standards. How long before you calibrate again is a decision you need to make based on your experience with the stability of your instrument. If you add an arsenic compound as an internal standard, you could use any changes in the response to this compound as a diagnostic for drift in your instrument response. The other issue, as I am sure you are aware, is that with a gradient you need to re-equilibrate the column before injecting the next sample. This can be quite time consuming. We have approached this in the past by using two columns, one of which was being re-equilibrated while the other was in analysis mode. (This approach also requires another pump, of course.) 

What are your thoughts on the use of speciated isotope dilution ICP-MS techniques for speciation?  

Speciated isotope dilution ICP-MS is a powerful concept for the investigation of species interconversion in an analytical method. I believe that most of the work done so far has been with mercury, tin, and chromium. I do not think that the technique can work for arsenic, which is monoisotopic in nature, and although isotopes ranging from 

As are known, none has a half-life longer than three months. There was a poster (820-7) at the 2012 Pittcon conference from Skip Kingston's group at Duquesne describing work in which arsenite had been labeled with 

O. Following LC separation, the species were detected by organic mass spectrometry.  My take on the literature evidence is that under mild acidic or basic conditions (and heating to 95 °C under pressure), arsenic species can be extracted from rice without significant species transformation.  

You mentioned that a European proficiency test was conducted to see how accurately arsenic could be measured in a variety of laboratories. What was the outcome of that test in terms of the total arsenic determination?    

In 2009, the European Union Reference Laboratory for Heavy Metals in Feed and Food organized a proficiency test, IMEP-107, on the determination of total and inorganic arsenic (As) in rice. The goal of the test was to assess laboratories' ability to determine total and inorganic arsenic in rice. Approximately 98 laboratories reported results for total As and 32 for inorganic As.  The outcome for total arsenic determination was about the same as for the speciation analysis: 41% of the 98 participants do not get a satisfactory result. There is a slide in my presentation showing the results. Details of this study are available in the literature (4).  

You said that about 70 papers are published every year dealing with arsenic speciation, but the "Atomic Spectrometry Update" article only had 40 citations. Why?   

In a recent web seminar, Julian Tyson, professor of chemistry at the University of Massachusetts, Amherst, explained how to develop a reliable method for arsenic speciation. Here, he answers questions raised during the seminar.

Spectroscopy-10-01-2013

A Rapid FT-IR-based Method for Monitoring Detergent Removal from Biological Samples

Calibration Transfer, Part IV: Measuring the Agreement Between Instruments Following Calibration Transfer

Considerations of Grating Selection in Optimizing a Raman Spectrograph

Market Profile: Terahertz Spectroscopy

Speciation Analysis Using HPLC–ICP-MS: Arsenic in Rice

Vol 28 No 10 Spectroscopy October 2013 Regular Issue PDF