CLIN. CHEM. 36/2, 217-224 (1990)

CLINICAL CHEMISTRY, Vol. 36, No. 2, 1990 217

Establishment of an External Quality-Assessment Scheme for Amino Acid Analyses:Results from Assays of Samples Distributed during Two YearsJ. M. Rattenbury1 andJ. C. Tewnsend2

Ten different samples of lyophilized plasma and two of liquidurine were distributed during two years to 26 laboratoriesperforming quantitative amino acid analyses in a schemedesigned to provide external quality assessment. After eachdistribution, statistical summaries and performance scoresbased on delta standard deviations and percentage biasesfrom the alt-laboratory trimmed means were returned toparticipants, who also received annual performance summa-ries based on their accumulated results. Coefficients ofvariation calculated from returns across all the samplesranged from 13% for glycine to 65% for methionine. Auto-mated ion-exchange amino acid analyzers with ninhydrindetection appeared to perform better than other methods,although there was no clearly superior method and no modelof analyzer clearly outperformed the others. These exercisesdemonstrate that there is room for improvement in theperformance of quantitative amino acid analyses and thatindividual expertise may be more important in maintaininggood performance than the choice of method or analyzer.

AddItional Keyphrases: interlaboratory performance- intermethod comparison - analytical error

Quality-assessment schemes are an integral part of clin-ical analyses, and external schemes for the surveillance ofroutine assays have been in existence for many years. Morerecently, external quality-assessment (EQA) schemes havebeen extended and new ones devised to assess analyses ofan increasingly specialized nature.3

Amino acid analyses in clinical laboratories are useful insuch areas as detection and monitoring of inherited meta-bolic diseases, nutrition, renal and liver disease, and skel-etal disorders (1), but the ability to perform these analysesis restricted to relatively few specialized laboratories. Re-garding inherited metabolic diseases, clinical laboratorieswith amino acid analyzers usually act as reference centersfor general hospitals that screen for these disorders byqualitative methods.

A comprehensive scheme for the external quality assess-ment of quantitative amino acid analyses has not hithertobeen described in detail, although interlaboratory surveysof the performance of amino acid analyses have beenpublished. These surveys have mainly involved nutritionaland agricultural applications, with distribution of aqueousmixtures of amino acids, protein hydrolysates, or proteinsamples to be hydrolyzed by participants (2). Reports of

Department of Chemical Pathology, Children’s Hospital, Shef-field SlO 2TH, U.K.

2Department of Clinical Chemistry, Royal Hallamshire Hospi-tal, Sheffield Sb 2JF, U.K.

3Nonstandard abbreviations: EQA, external quality assess-ment; AlE, automated ion-exchange chromatography; OPA, 0-

phthalaldehyde; RMS, root mean square; and MB, mean percent-age bias.

Received May 22, 1989; accepted October 11, 1989.

amino acid surveys based on physiological fluids are few (3,4), but the results suggest that analyses of physiologicalfluids are less satisfactory than those of protein hydroly-sates or standard mixtures (3). From this background arosethe notion of a scheme for the external quality assessmentof amino acid analyses based on physiological fluids inclinical situations. Laboratories performing such analyseswere approached, and those expressing an interest wererecruited into the scheme. Here we report the establish-ment of this scheme and the results obtained from itsoperation during two years.

MaterIals and MethodsParticipants

In 1985, laboratories in the United Kingdom believed tobe performing amino acid analyses in medically relatedapplications were identified through personal contact orfrom manufacturers’ lists of users and invited to participatein an external quality-assurance scheme. Each was sent aquestionnaire requesting demographic information aboutlaboratories, applications of their analyses, workload,methods (including sample preparation, standardization,quality control, and data processing), and their quality-assessment requirements in terms of the type and fre-quency of sample, reporting, and scoring formats. Inter-ested laboratories were sent a trial sample for analysis(sample no. 1/85), and the results of the survey werereturned to them. They were then invited to register for thereceipt of samples on a regular basis. Twenty-six laborato-ries (all in the British Isles) registered for the scheme; 11were in teaching, eight in district general, and six inchildren’s hospitals; one participant was a microbiologicalresearch establishment. The applications of the analyzerswere described by participants as follows (some partici-pants reported more than one application): diagnosis andmonitoring of inborn errors of metabolism (n = 23), renaland metabolic research (n = 2), and analysis of growthmedia (n = 1). Secondary applications listed were unspec-ified research (n = 8), parenteral nutritionmonitoring andresearch (n = 6), renal investigations (n = 3), nutritionalinvestigations and research (n = 3), inherited metabolicdiseases (n = 2), and metabolic studies, liver investiga-tions, and quality control of bacteriological culture media(1 each). Participants reported annual workloads of be-tween 50 and 1250 analyses.

Methods

Table 1 lists analytical methods, manufacturers, andmodels of analyzer. Most participants used automatedion-exchange chromatography (AlE) with colorimetric nm-hydrin detection (5). Four participants used AlE withfluorimetric detection after reacting amino acids with o-phthalaldehyde (OPA) (6). All ME methods involved use oflithium citrate buffers. Four participants used “high-per-formance” liquid chromatography (HPLC) with pre-columnderivatization, three using OPA (7) and one 5-dimethyl-ainino-1-naphthalene sulfonyl (DAN8YL) (8) to form deriva-

Annual workload, RMS of No. oflnstltutlon no. of samples MethOde Analyzerc SDd outliers

‘G Not given N LC 5001 0.68 0T 500 HPLC Altex 332 0.72 1G 400 N J180 0.76 0

*C 950 N J180 0.78 0*( 840 N TSM 0.83 3G 200 N J180 0.86 0T 1250 N M 0.87 0

‘C 300 N J180 0.87 1T Not given N M 0,88 1R 400 N LKB4400 0.88 1T 600 N J180 0.91 1T 430 N LKB 4400 0.92 0

* 150 N Chromakon 500 0.95 0G 90 F J180 0.95 1

*C 380 F J180 1.07 1T Not given N Chromakon500 1.10 4

‘T 1250 N Locarte 1.16 3T 500 N J180 1.23 3G 100 N LC5001 1.27 7

*C 750 N LC5000 1.32 14G 550 HPLC Altex 420 1.32 10T 200 HPLC Altex 1.35 11

G 120 HPLC Perkin-Elmer 1.39 1*C 1240 N Locarte 1.41 6

T 400 N LKB4400 1.42 2G 50 F J180 1.55 9T 550 F J180 1.74 12

6, general hospital; T, teaching hospital,C, children’shospital;A, research institute; #{149},reference laboratory. #{176}N, automated ion-exchange (AlE) with ninhydrindetection; F, AlE with fluonmetricdetection. c Amino acid analyzers: Biotronik LC 5000/5001, Wissenschaftliche Gerate GmBH, Frankfurt am Main,F.R.G.;Model3321420 gradient HPLC, 005 column, Altex Scientific Inc., Berkeley, CA 94710; Chromaspek J 180/Chromaspek M, Hilger Analytical, Margate, U.K.; TSM,Technicon Instruments Corp., Basingstoke, U.K.; LKB 4400, LKB Instruments,Croydon,U.K.; Chromakon 500, Kontron Instruments,Zurich, Switzerland;Model4B, Locarte Instruments, London, U.K.; Series 3B, ODS column, Perkin-Elmer, Beaconsfield, U.K. d Calculated from returned results after removal of outliers, andsubject to the limitations described in Methods.

Table 1. Laboratories Participating In the EQA Scheme, Ranked in Order of Their Performance during Two Years

218 CLINICAL CHEMISTRY, Vol. 36, No. 2, 1990

tives. Ten participants reported automated data process-ing, with use of various types of computing integrator, sixdescribed manual peak identification and measurement,and 11 reported combined methods (usually visual peakidentification, followed by automated integration of theidentified peaks).

Reagents

Amino acids (chromatographically pure) were obtainedfrom BDH Ltd., Poole, U.K. All other reagents were of“ANALArt” grade, from the same supplier.

Lyophilized plasma samples. These were obtained insealed, rubber-capped vials (5- or 10-mL) from the U.K.External Quality Assessment Scheme, Queen ElizabethMedical Centre, Birmingham, U.K., and from WellcomeDiagnostics, Dartford, U.K., and were surplus to theirrequirements for general clinical chemistry EQA schemes.One of these samples (no. 1/85), pooled human plasma fromblood donated for transfusion, was distributed before warn-ings concerning IIIV infection were publicized. Thereafterall lyophilized plasma samples distributed were of bovineorigin. For some distributions, participants were instructedto reconstitute the samples with the appropriate volume ofdistilled or de-ionized water. However, to introduce suffi-cient variation into the amino acid proffle, some sampleswere accompanied by an aqueous reconstituting solution of

amino acids for addition to the lyophilized plasma.Reconstituting solutions. These were made by dissolving

weighed amounts of pure, commercially available aminoacids in de-ionized water, adjusting the solution to pH 2with hydrochloric acid (6 mol/L), and ifitering it through acellulose acetate filter of 4.5-pm pore size. Aliquots of thissolution, in volumes slightly in excess of those needed toreconstitute the plasma, were dispensed into screw-cappedvials and sealed with waxed film. The fact that reconsti-tuted volumes were 10 mL in 1986 and 5 mL in 1987 wasfortuitous and due to the types of lyophilized materialsavailable.

Urine samples. Urine collections from adult laboratorystaff were adjusted to pH 2 with hydrochloric acid (6 mol/L)and autoclaved in bulk at 121 #{176}Cfor 15 mm. After cooling,we added weighed amounts of pure amino acids and pre-servative (100 mg of thimerosal per liter of urine) andmixed well. The sample was then ifitered by the samemethod as the reconstituting solution (above), and 2.5-mLaliquots were dispensed into screw-capped vials.Table 2gives detailsof the samples and reconstituting solutions.

Dispatch of samples. Samples were sealed in plasticwallets and packed into boxes with absorbent wadding. Astandardized report form (Figure 1, left) accompanied eachsample, which was sent out on a Monday or Tuesday byfirst-class post at about two-month intervals. Participants

Vol, mL10

Amino acid profileReconstituting

Manufacturera Sourceb solutionc

1 1 Water

2 1 Water2 1 A

3 2 Water

4 3 Water

2 1 B

4 3 C

4 3 Water

4 3 D

4

4 3 E

4 -

4 3 Watera 1, U.K. External Quality Assessment Scheme for General Clinical Chemistry (UKEQAS), Woltson Research Laboratories, Queen Elizabeth Medical Centre,

Birmingham, U.K.; 2, Purce Associates, Macclesfield, U.K.; 3, Ortho Diagnostic Systems Ltd., High Wycombe, U.K.; 4, Wellcome Diagnostics, Dartford, U.K. 1)1,UKEQAS; 2, purchased; 3, Wellcome; 4, donated. C Reconstitutingsolutions:

5 Generally very low concns.

Vol, mL500

500

500

-250250

BCDE

Amountadded, mg Expected concn, pmoi/L59.6 1018

108.5 165439.9 60810.6 16289.0 105638.4 1023

33.7 92313.1 58981.5 1863

Table 2. Samples Distributed in the Amino Acid EQA Scheme, and Their Analytical Profiles

CLINICAL CHEMISTRY, Vol. 36, No. 2, 1990 219

Code Sample

1185d HIQC 8 (trialsample)

1/86 Armtrol 6502/86 Armtrol465

3/86 Ortho Iii 009Y01

4/86 WellcomtrolSVRSBC24

5/86 Armtrol 650

6/86 Wellcomtrol SVRSBC24

1/87 WellcomeBCG62/87 WellcomeBCA1

3187d,8 Autoclaved urine withadded glycine andhistidine

4/87 Wellcome BCA1

5187d,e Autoclaved urine withadded valine,leucine, andisoleucine

6/87 WellcomeBCA1

Amino acid(s) addedL-ValineDL-LeucineL-lsoleucineL-AlloisoleuclneDl.-OrnithineHCIGlycineMixed remnants of solutions A, B, & CL-GlutamineDL-AlanineDL-Citrulline

Aboutnormal,but increased glutamic acidand decreased glutarnine.

10 About normal but glutamine decreased.10 Above-normal branched-chain amino acids;

othersaboutnormal.10 Increasedconcnsof phenylalanine,leucine,

and glutamicacid;decreasedglutamine;othersaboutnormal.

10 Generallylowto normalconcns.

10 Increased ornithine,decreasedglutamine,others low to normal.

10 increased glycine, decreased glutamine,others low to normal.

5 Generallyvery low concns.5 Abnormal profile, generally low concns, but

valine, leucine, omithine, and isoleucineincreased.

2.5’ Increased glycine, others moderatelyincreased.

5 Increased citrulline; glutamine and alaninemoderately increased; others low.

2.5 Increased glycine, isoleucine, leucine, andvaline;othersmoderatelyincreased.

d Humanmaterial;all others bovine,a Uquid urine; all others lyophilized plasma. ‘Volume dispensed.

were given a date by which reports had to be returned, adate that allowed at least two full working weeks foranalyses to be completed.

Cakulation of results. Statistical summaries and perfor-mance scores were calculated from returned results, andreportswere generated with a BBC “B” microcomputer(Acorn Computers Ltd., Cambridge, U.K.) and specificallywritten software. Results from participants were treated asis conventional for EQA schemes (9). The mean, standarddeviation (SD), and coefficient of variation (CV) were cal-culated for each amino acid.4 Each individual result from a

4x, result returned (imo1JL); 1, mean; n, no. of results received foreach amino acid reported by different participants for each sample;N, no. of results received for different amino acids reported by eachparticipant for each sample; m, no. of returns for which results areincluded in the calculations; ASD, delta standard deviation of result,

(x - 1)/SD I; B, percentage bias, 100(x - 1)/I; RMS, root mean

participantwas then compared with the mean for thatamino acid and the absolute difference between them wasexpressed in delta standard deviation (zSD). If ASD ex-ceeded 3, the result was eliminated as an outlier. Themean, SD, and CV were then recalculated for the remain-ing results, which were compared with the new trimmedmean to calculate revised ASDs. This process was contin-ued until no more outliers were detected. In the final cycleof calculation, iSDs (including those of the outliers) andpercentage biases (B) were calculated and recorded. Theseprocedures were repeated for each amino acid for which aresult had been reported by a participant.

To estimate the performance of each participant for allthe amino acids in a particular sample, we calculated the

square of SDs, (SD2/N)’, or of CVs, (CV2/m)112 or (CV2/m-MB, mean percentage bias, (B1+B2+ . . . B)fN.

CDD. S L5IT05Y. . 08TPOO.. F’S SASr%.L. - a-.

1598 D#{216}ITE S SNSP.E SECt IVED. ‘ (‘3 . i ((PLEASE (HITS)

EXTEO08L QUALITY ASSUNAPIU OP QUA*4TITAYI4S ASINO ACID 0014.9005

SELT$

LAIOIATOIY CODE, SAMPLE CODES 9.’S4 METhOD CODES N DATE. *I,/S4

W1.A. AC

ThlIHE I181TOilAl* 3C

ASPASTIC AdO IO08I58IPE l32.IILY03HE 1s9

tHE “ ‘‘._

Sl.IITAPIIC ACID - I lOTi 0114180 ACIDS DETECTED. -

- I -I

IPN05IPE ‘t0 I

IOLVCIP#{128} 310 I

AMINO ACID ALLo SD

7.nIn. 13 *2.8 5.7A.5.rt)0 #{149}011 28 67.7 24.5

,, (t,*0d) I, 82.2 5.4PSlAP0UDt0IlI.

... ““#{176}#{176}‘“

G1,,t.t0 acId6604r9010#{149} I 98 -GI..t..In. IS 43.3 10.4...*flOad,P4C 2 *3.4 -Frail,. *9 50.2 .2.7SIsal,. 22 341.3 *99.7

., (trlaaad) 20 321.4 25.7

CV . .90515 Y5,I,T

lesol xi

I

- -

A).,,,. 22 3*7.0 51.0

CITRLI.LIHE 5L *

IVALIHE 203 I

I

CYThHE

HETHJDHIHE 13 I

flfl#{149} 21 20.5 5.3,. )t,-,..d) 25 15.2 3.7

,ool..aT,,. 22 50.1 12.7

.PSVAyII.n,.. 22 71.2 0.0Mb.t,d,n. 22 83.8 46.4

7_clI

LEUCIHE I -scl II

TVAOIIHE .sl I

II

I’M.4I*tflP.

3-M.-SIStldIn. 4 7.3 S-IS1’..yI,to,’..o *4 32.0 14.9flr,.ItS,,n. 21 005.1 306.7

,, *tFlAad* 28 053.5 231.6Lsi,p 21 265.2 294.9

.. )tr,.d* 19 180.5 45.5I

HISlIDIHE c10 ,

co...pas I

6.8.0,,. 19 1*3.3 21.0.r00,,’. I *7 -

Sf1 Yo*r ran,, *0. SPS I.n bias8.62 2/22 -4.3

25.4 0 42.134..

13.1 14 82.0 83 1.I “*4.5- 3 45.4*0.013.1 *4 05.4 00 ‘S.i -I.?03.7*7.7 *4 *02.1 *55 U 0.315.4 14 2i4.7 *55 -5.4 -5.9

28.1 13 *3.8 98 5.9 0.0- 2 5.2

23.1 *2 *05.3 50 -0.4 -0.643. *

8.0 *6 322.4 315 ‘5.4 -3.216.3 lb 306.1 253 -0.7 ‘15.014.1 II 60.2 56 -1.3 -17.0

- 0*3.0 *6 .11.3 203 -5.5 -7.5

*43.0 II *2.8 S “0.7 ‘lOS26.619.3 14 *8.9 *3 -1.7 -32.117.4 *8 88.0 79 ‘0.2 -12.3

9.0 *6 255.5 100 -0.0 -5.424.1 *6 44.2 4* -8.2 ‘-4.712.6 *6 75.3 65 -5.7 -I.?55.425.0 *4 60.9 98 1.1 26.4

- 3 *0.7- 4 7.3

51.5 II 38.8 35 0.1 6.733.8

24.2 19 588.5 082 -5.3 -7.9**5.0

24.1 *5 *72.4 166 U -1.215.2 *4 146.9 *28 0.1 ‘*0.7- I 17 #{149}‘

Your rAni *Or a..., blat

1/22

220 CLINICAL CHEMISTRY, Vol. 36, No. 2, 1990

F OIJL 2 r’r FI5 SI,1l QalJFlris-lTI’., r.i*r.ia C ID

#lT’ IC I

BllIL. D Xcr IDl”$ I

i#{149}.SPS. ((II-CTl1UTlPe

PCcOPSTITh1t P05 #{149}P1 AT 50014 TEPCRAT%E.ICI.L T)52P#’TA5 AS Ii PATIENT’S SPECIPENEHTII RCSIA.TI APO t(1’.N TillS COPY ITI 1”l-lLJB B PLJD & BS1O 05 14 RArTEpy. PAS1ENT OP CHEMICAl. PAThOlOGYCNII.OUEHE ISPiTM,, I.ESTESN SAPS. SHEFFIELD 515 2Th

Fig. 1. (Left) Example of a participant’sreport;(right) exampleof summarizedresultsand performancescore returnedto a participantMethod codes used: N, AlEwith ninhydnndetection; F, AlE with fluonmetric (OPA)detection; H, HPLC;n and , no.of resultsreturnedandmean,respectively,for all methods or participant’s method as appropriate; level, result reported by participant. “dSD’ Is deltastandarddeviation.%B Is percentage bias

root mean square (RMS) of the iSDs and the arithmeticmean of the biases (mean percentage bias, MB) from theresults of all amino acids reported. The calculation of RMSwas appropriate for the SDs and CVs because these werederived from the sums of squares of the results. It becameapparent from the results of the trial distribution that, toavoid gross distortion of the statistical summaries and theallocation of inappropriate scores, we had to set arbitrarylimitations on the results. Thus, SD and B were set atmaxima of 3 and 100%, respectively, and results below theminimum of 10 returned results or a mean of 20 .&mo1/L forany amino acid in a distributionwere excluded. Resultsfrom amino acids that fell below these latter criteria alsowere excluded from calculations for assessing the perfor-mance of laboratories and the assay of individual aminoacids. These exclusions were applied both to the perfor-mance scores returned to participants for each sample andto the summarized results in this report. As a furtherindicationof performance, each participant was given aranking for their RMS of ISD and MB after comparisonwith the results from all other participants for that partic-ular sample. Participants were asked to record the amountsdetectedfor a list of amino acids most commonly found inphysiologicalfluids,but they were also invitedto includeany additional unlistedamino acids detected. Figure 1(right) shows a typical report of a participant’s results.

At the end of each quality-assessment period (one year,

six samples),participantsreceiveda summary of theirperformance over that year. This included an annual RMSof SDs and annual mean bias for each amino acid re-ported. Overall RMS and MB for all amino acids from allsamples for which results were reported were calculated,and a ranking score was obtained so that each participantreceived a comprehensive summary of their performancefor the year.

ResultsOf’the 24 laboratories sent the trial sample (no. 1/85), 19

returned replies. During 1986 and 1987, 26 laboratoriesparticipated in the scheme (one laboratory reported resultsfrom two analyzers; these were treated as if from separateparticipants). Four laboratories entered the scheme after ithad started, and one left it after reporting results for twosamples;resultsfrom this latterparticipanthave not beenincluded in the overall analysis.Ifthe four late entrantsare excluded, 10 (43%) of all participants returnedresultsfor all 12 samples, four (17%) for 11 samples, three (13%)for eight samples, two (9%) for 10 samples, and one (4%)each for nine, seven, four, and three samples. The meantime allowed for the return of results after the dispatch ofsamples was 20.9 days; the actual mean response time was17.4 days. Of the returned results, 78% were received bythe due date.

Results obtainedduring the two-year operationof the

Means from the Lyop4IllzeddistrIbutions, imol/L All samples plasma Uquld urine

No. ofRMS0f RMSof RMSoI outliers

Amino acid Median Range m the CV8 m the CVs m the CVs detected

Glycine 333 30-1431 12 13 10 13 2 16 12Isoleucine 86 38-719 11 15 9 14 2 19 4Phenylalanine 67 25-989 11 15 9 14 2 18 7Leucine 207 36-1916 12 16 10 13 2 25 5Valine 216 60-1501 12 17 10 16 2 20 4Alanine 243 46-613 12 18 10 15 2 27 6Glutamic acid 122 52-355 12 18 10 17 2 26 1Senne 67 33-383 12 20 10 19 2 22 5Threonine 80 24-152 11 21 9 20 2 25 6Citruliine 51 21-1849 10 23 10 23 No results 3Aspartic acid 63 28-196 7 23 5 22 2 26 6Tyrosine 43 25-194 9 23 7 23 2 25 1Lysine 103 54-201 12 26 10 21 2 42 8Arginine 114 22-147 12 27 10 22 2 44 5Glutamine 43 23-730 9 27 9 27 No results 2Ornithine 65 34-908 10 28 10 28 No results 3Proline 76 22-406 12 37 10 37 2 37 2Taurlne 43 25-96 11 38 9 37 2 41 1Histidine 56 21-517 12 40 10 41 2 29 9

Tryptophan 34 26-51 8 41 6 40 2 44 0

Methionine 28 23-48 5 65 3 72 2 53 2Overall RMS of the CV5 27 26 32

Median and range are those of the mean amino acid concentrations from each distributioncalculated from results returned byparticipants. ‘rn” is the numberof distributions the results of which are included in the calculation of median, range, and AMSof theCV8, according to the limitationsdescribed in Methods.

CLINICALCHEMISTRY, Vol. 36, No. 2, 1990 221

scheme were analyzed in two ways. Estimates of thevarianceof individual amino acids were obtainedfrom theRMS of the CVs that were calculated after each of the 12distributions, and these are shown in Table 3. Interlabora-tory comparisons, on the other hand, were obtained fromthe RMS of the score(SD) of each result returned (Table1). The latter method was also used when we assessed

groups of participants-for example, to compare methods.Upon request, we will supply full statistical summaries foreach of the samples distributed.

As an additional means of stunmarizing performance,the total numbers of outlying results attributable to indi-vidual participantsand to particular amino acids werenoted. There was no obvious association between the num-ber of outlier values detected and the RMS of CVs ofindividual amino acids, but laboratories with better perfor-mance, as assessed by RMS of SDs, tended to producefewer outliers (Table 1).

Uncommon amino acids and other compounds not listedon the reporting sheet but reported by some participantswere as follows (number of occasions reported): a-amino-n-butyric (68), 3-methylhistidine (37), 1-methylhistidine(34), #{225}sparagine(29), hydroxyproline (24), ethanolamine(19), a-aminoadipic (12), phosphoserine (8), /3-aminoisobu-tyric (8), allo-isoleucine (8), homocystine (5), cystathionine(5), carnosine (5), homocitrulline (3), ‘y-aminobutyric (3),phosphoethanolamine (2), methionine sulfone (2), hydroxy-lysine (2), and /3-alanine (1). We added allo-isoleucine toone sample. /3-Aminoisobutyric acid, homocitrulline,me-thionine sulfone, and hydroxylysine were reported in urinesamples only.

Discussion

The pre-scheme questionnaire provided demographicand analytical information about participants and allowedthem to indicate their EQA preferences. The consensusthat emerged, on which the scheme was based, was thatmost samples should be lyophilized plasma and that theseshould be distributed six to 12 times a year,that partici-pants should be able to report the concentrations of anyamino acids detected in the samples, and that they shouldreceive reports giving statistical summaries of results andassessments of their own performance in terms of standarddeviations from the mean. Some participants also re-quested a report of the bias (%) of their results and use of adevised scoring system. The scheme was designed substan-tially to comply with these requirements.

Most participantsgave the diagnosis and monitoring ofinherited metabolic diseases as their primary reason forperforming quantitative analyses for amino acids. Partici-pants in teaching or children’s hospitals ordinarily pro-vided a reference service for general hospitals, some ofwhich would be performing qualitative screening tests forthese disorders. Eight of the participantswere themselvesdistrict general hospitals, and the operation of amino acidanalyzers in such sites resulted from particular local inter-est and expertise. The wide range in the number of aminoacid analyses performed per annum by participants reflectsthe provision of either local or reference services.

Most of the laboratories in the British Isles performingquantitative analyses for amino, acids in clinical situationswere included in the scheme. The small number of partic-ipants, as compared with the numbers ordinarily found in

Table 3. Summarized Results from All Samples Distributed

222 CLINICALCHEMISTRY, Vol.36, No. 2, 1990

national EQA schemes, reflects the specialized nature ofthese analyses. However, the relatively large number of“analytes” (individual amino acids reported) enabled com-parisons to be made of imprecision between laboratoriesand between individual amino acids when results werecombined across 12 distributions over two years.

IFCC recommendations speci1’ such desirable propertiesfor quality-assurance materials as similarity to patients’specimens, stability, and homogeneity (10). These consid-erations, in conjunction with the replies from the question-naire, mandated that plasma should be the commonestsample to be distributed but that urine or aqueous speci-mens should also be considered. Although in some EQAschemes liquid plasma samples are used, the known chem-ical or bacteriological instability of some amino acids (11)limits the usefulness of such material. Lyophilized plasmasamples were chosen for their stability but, when samplesfrom various sources were examined, it became apparentthat (a) processing had generally decreased the amino acidconcentrations below those usually found in humanplasma, (b) the concentrations of some amino acids (e.g.,glutamine) were greatly decreased, and (c) the variation inthe amino acid profile of lyophilized plasma samples fromvarious sources was insufficient to represent the range ofpatterns found in pathological samples. To overcome thesedeficiencies, we distributed some lyophilized samples withreconstituting solutions that contained added amino acids.However, the use of these solutions did not contribute tothe variability of results, because the RMS of CVs ofsamples reconstituted with them (25.9%) was no greaterthan that of samples reconstituted with water (26.0%).

For 11 of the 12 samples distributed, the proportion ofparticipants making returns varied between 77% and 96%.The response to sample no. 3/87, the first liquid urine to bedistributed, was 65%, the response for the second urinesample (no. 5/87) was 81%. The mean response for allsamples over the two years was 83%.

The number of returns made at each distribution (be-tween 17 and 24) was sufficient for statistics and perfor-mance scores to be calculated for most of the commonamino acids. The RMS of the CVs of all the common aminoacids measured over all 12 distributions was 27% (Table 3).This result was obtained after outlying values had beenremoved at the calculation of each distribution and theexclusion of CVs when fewer than 10 results were returnedfor a sample or the trimmed mean for an amino acid was<20 moI/L (see Cakulation of results). Williams (2) ob-mined an arithmetic mean CV of 23% for amino acidsmeasured by four laboratories in a single distribution ofbovine plasma. lithe results in that report are recalculatedas the RMS of the CVs, the CV becomes 28%. Thus twosurveys of widely differing design obtained similar esti-mates of overall variation in the measurement of aminoacids. In the present survey, the RMS of the CVs for the twourine samples exceeded that of the lyophilized plasmasamples (Table 3), though the difference was not significantby the Mann-Whitney U test. Liquid urine samples can beexpected to be less stable than lyophilized plasma duringdistribution despite the precautions taken, but greatervariability of results might also be due to the greaternumber of peaks obtained in a urine chromatogram andhence the increased potential for interference with peakmeasurement.

The wide variation in the RMS of the CVs for individualamino acids (Table 3), in part due to differences in their

concentrations in the samples, also demonstrates the ex-tent of their analytical individuality. In quantitative chro-matography, adequate peak shape and resolution are pre-requisites of accurate peak measurement; however, otherfactors can be examined that could have contributed to thepoor performance of analyses for specific amino acids.

Amino Acids

Methionine. None of the samples distributed was supple-mented with methionine, and its endogenous content wasgenerally low. Consequently, results from only five returnswere included in the summary. In addition, sample no. 2/86contained allo-isoleucine, which was mistaken for methio-nine by some participants and resulted in a particularlyhigh CV for methionine. If, however, the results for me-thionine in sample no. 2/86 are omitted, the RMS of theCVs decreases to 43%, which is still higher than for anyother amino acid.

Tryptophan. This was present at concentrations gener-ally within or below concentrations to be expected innormal human plasma. Some proportion of tryptophan,which can vary with sample conditions such as pH, isbound to albumin and may be lost during sample prepara-tion (11). In some chromatographic systems, this aminoacid gives a relatively broad, shallow peak, and its reten-tion times may alter if conditions are not held stable.

Histidine. The relatively poor performance of analysis ofthis amino acid is not readily explained. Inclusion of 1- or3-methylhistidine with the histidine peak would causesome variability. One participant reported the presence ofan unidentified substance interfering with the histidinepeak in some materials.

Taurine. This was reported less frequently by partici-pants than most of the other common amino acids, probablydue to its unimportance in the investigation of inheritedmetabolic diseases and its early elution in the chromato-gram.

Proline. This imino acid can be difficult to quantify,owing to its limited chromogenicity with ninhydrin and theneed to measure the product at 440 ma. Proline andhydroxyproline were not reported by participants usingHPLC.

Ornithine, lysine, arginine. The imprecision with whichthese basic amino acids was measured was similar (be-tween 26% and 28%, Table 3). Several amino acids forwhich detection and quantification were poor were locatedin the basic area of the chromatogram.

Glutamine. This amino acid is readily deaminated toglutamate and ammonia during sample handling, chroma-tography, or storage (11). Pre-analytical factors are there-fore likely to contribute to its observed variability.

Others. Determination of the remaining common aminoacids was better overall than for those discussed above.Various factors might have been responsible for this, in-cluding higher concentrations in the distributed material(e.g., glycine); less-troublesome chromatography, exempli-fled by a sharp, symmetrical peak shape with freedom frominterference; clinical relevance, which would lead to morecommon measurement (e.g., phenylalanine); and possiblyproximity to an internal standard within the chromato-gram. This could be particularly true for amino acidselutednear norleucine, the most commonly used internal stan-dard. Amino acids with better performance were generallyfrom the acidic to neutral regions of the chromatogram.

Problems with other individual amino acids became

No. labs RMS of SD19 0.984 1.184 1.29

CLINICALCHEMISTRY, Vol. 36, No. 2, 1990 223

apparent from time to time, often as a result of queriesraisedby participants. Glutathione, which co-elutes withaspartate in some systems, can be released from erythro-cytes during plasma processing (11), so that mean concen-trations of aspartate in some samples may be overesti-mated. Amino acids with disulfide bonds-e.g., cystine andhomocystine-become bound to protein unless plasma sam-ples are promptly deproteinized after collection (11). Cys-tine concentrations in the lyophilized plasma samples werelow and were frequently reported as zero by participants;therefore, this amino acid has been omitted from thesummarized results. Probably the blood collection andprocessing of the lyophilized samples resulted in the bind-ing of free cystine. The measured concentrations of cystinein the distributed samples was method-dependent, partic-ularly high results being obtained by HPLC (7). In HPLC,the sample is pre-treated with iodoacetate to form S-carboxymethylcystine, which is then converted to a fluo-rescentderivative with o-phthalaldehyde and mercapto-ethanol (12). Cooper et al. (13) suggested that, in conven-tional methods of protein precipitation for amino acidanalysis, variable amounts of free cystine are removedfrom plasma and that one must use HPLC to determine thetrue concentration. The need to precipitate plasma proteinswithout delay when one needs to know cystine or homocys-tine concentrations is well-known (11), but the gross differ-ences obtained by Cooper et al. (13) between HPLC andAlE methods may be due not only to deficiencies in thelatter technique but also to the nature of the samplepretreatment for HPLC.

In addition to the 22 common amino acids, participantswere invited to report unlisted amino acids and othercompounds detected. Returns for these substances wereusually sporadic, in part because of the absence from, orlow concentrations in, the distributed materials, but alsobecause participants differed in the importance they at-tached to minor peaks in the chromatograms.

Data Analysis

The data from returns are also examined to comparemethods of analysis, types of analyzer used, and the type ofparticipating laboratory. Although none of the differenceswas significant (P <0.05) when tested by the Mann-Whitney U test or the Kruskall-Wallis one-way analysis ofvariance, this could be due to the inhomogeneous nature ofthe data. Thus we present the results as observed trends.

Assessment based on RMS of ASDs showed that of thethree method groups recognized in the amino acid EQAscheme, AlE with ninhydrin detection (used by more thantwo-thirds of the participants) appeared to give the bestperformance, although as noted above this was not astatistically significant difference (Table 4). Because thetarget values in this analysis are the consensus all-methodtrimmed means, one must consider the possibility thatparticipants using minority methods were penalized. Nosuch effect was apparent in the data, however, except in theexample of cystine discussed above. The generally betterperformance of AIE/ninhydrin would accord with the long-established and robust nature of the method. Fluorimetricor HPLC techniques generally appeared to perform lesswell but could have other advantages such as greateranalytical sensitivity and faster analysis times. It is note-worthy, however, that within this general rule, some indi-vidual participants performed markedly better or worsethan their method group as a whole (Table 1), which

Table 4. Amino Acid EQA Performance, Compared byMethod and by Make of Analyzer

Method

AIE/ninhydrin (N)

HPLCAiE/OPA (F)

Analyzer Method

TSMJ180&.Ma

NN

Chromakon 500 NLKB4400 NLC5000/50018 N

Altex (OPA) HPLCJ 180 (OPA) FLocarte NPerkin-Elmer (DANSYL) HPLC

1 0.83

8 0.892 0.993 1.003 1.113 1.164 1.292 1.301 1.39

#{149}Analyzersfromthesame manufacturerand with basically similaranalyticalsystems are grouped together.

suggests that the expertise of participants was at least asimportant a factor of performance as the choice of method.Apparent differences between makes of analyzer are alsopresented, but these were not significant when tested byanalysis of variance. In addition, as noted above, resultsfrom groups containing few participants could be biased byaptitude as much as by choice of instrumentation.Differ-ences between the various types of hospital laboratory weresmall and nonsignificant: RMS of SDs were 1.01 (n = 7)for children’s hospitals, 1.07 (n = 11) for teaching hospitals,and 1.09 (n = 8) for general hospitals. Laboratories thatacted as reference centers for amino acid analyses (Table 1)appeared to perform better overall than those that did not[RMS of the SDs 0.98 (n =13) and 1.15 (n = 14), respec-tively], but the differences were not significant by Mann-Whitney U test. The single nonhospital laboratory, whichscored 0.88 on this basis, returned only three results duringthe period in question because it entered the scheme late.

An analysis of rejected outlying results indicated thatsuch outliers did not correspond with the performance ofindividual amino acids when assessed by RMS of the CVs(Table 3). This result might be expected, because trimmingto ±3 SD should, on average, produce a similar number ofoutlying values. On the other hand, the number of outlyingresults obtained by individual participants was related tothe RMS of their SD (Table 1). A relationship between thenumber of outliers removed and the RMS of SD was alsosustained when we compared methods, analyzers, andinstitutions.

We could not demonstrate an improvement in the per-formance of amino acid analyses by participants over thefirst two years of the EQA scheme, either by examination ofthe RMS of the CVs or from the number of outliers removedfrom sample to sample. The subjective impression, how-ever, is that in the second year of operation fewer outlierswere attributable to peak misidentification. There was noformal procedure for follow-up of participants with poorperformance, the onus for improvement being left to indi-vidual laboratories. On the other hand, informal discus-sions with participants on both general and specific prob-lems were frequent, and sources of advice were suggested.

The overall RMS of the CVs (27%) and the range of CVsfor individual amino acids revealed considerable scope for

224 CLINICALCHEMISTRY, Vol. 36, No. 2, 1990

improvement in the performance of the analyses. It hasbeen suspected for some time that the measurement ofamino acids in physiological fluids is not very reliable,owing to the relative instability and complexity of thesesamples as compared with protein hydrolysates (2, 14).Ambler (15) examined the several stages of amino acidanalysis that might be susceptible to error, includingsample deproteinization. In their survey of four laborato-ries, however, Williams et al. (3) found that sample depro-teithzation centrally by the organizing laboratory did notreduce the variability of results.

Examination of data from quality-assurance surveys isone method for determining whether goals for analyticalperformance are being met (16). If applied as a sampleexercise to the amino acid that performed best in thescheme (glycine), the one that performed worst (methio-nine), and one of obvious clinical importance (phenylala-nine), it becomes apparent that performance goals are notbeing satisfied. Setting these goals requires knowledge ofintra- and interindividual (“biological”) variation (16), butthe amount of such information for amino acids is limited(17) and reference intervals (11) have to be used as a firstapproximation. One definition for an analytical goal is thatthe analytical CV should be equal to or less than half thebiological CV (16); if reference intervals are used, thedesired analytical CV for the above three amino acidsshould be 9.4%, .9.5%, and 7.4%, respectively. The actualanalytical CVs obtained for them in the EQA scheme(Table 3) exceeded these values, and therefore the set goalswere not achieved. Any refinement of the calculation ofthese performance goals (for example, by using true intra-individual variation rather than reference ranges, whichthemselves include analytical variation) is likely to makethe goals even more stringent.

The conclusion that must be drawn from these results isthat at normal amino acid concentrations, analytical per-formance as revealed by the EQA scheme is inadequateand at subnormal concentrationswill be even less satisfac-tory. On the other hand, when amino acids are grosslyincreased, as in some untreated inherited metabolic disor-ders, the analytical performance of amino acid quantifica-tion is likely to be adequate as long as the amino acids inquestion are correctly identified.

We thank Joyce C. Allen for her help in preparing samples anddistributing reports, D. G. Bullock (UKEQAS) and Carole Hjelm(Weilcome Diagnostics) for providing lyophilized samples, partici-pants in the scheme for their constructive suggestions, and ElaineSingleton for secretarial help.

References1. Rattenbury JM, ed. Amino acid analysis. Chichester, U.K.:Ellis Horwood, 1981.2. Williams AP. Collaborative trials and amino acid analysis.Ibid., 138-52.

3 Williams AP, Hewitt D, Cockburn JE, Harris DA, Moore RA,Davies MG. A collaborative study on the determination of freeamino acids in blood plasma. J Sci Food Agric 1980;31:474-80.4. Hammond J. Quantitation of plasma amino acids-a demand-ing task. Clin Biochem Rev 1988;9:108.5. Spackman DH, Stein WH, Moore S. Automatic recording appa-ratus for use in the chromatography of amino acids. Anal Chem1958;30:1190-206.6. Roth M, Hempai A. Column chromatography of amino acidswith fluorescence detection. J Chromatogr 1973;83:353-6.7. Turnell DC, Cooper JDH. Rapid assay for amino acids in serumor urine by pre-column derivatization and reversed-phase liquidchromatography. Clin Chem 1982;28:527-31.8. Olson DC, Schmidt GJ, Salvin W. The determination of aminoacids in physiological fluids using liquid chromatography andfluorescence detection. Chroinatogr Newal 1979;7:22-5.9. Skendzel LP. A report on the College of American Pathologistssurvey programme. Ann Cliii Biochem 1969;6:89-93.10. BUttner J, Borth R, Boutwell JN, Broughton PMG, BowyerRC. Approvedrecommendation(1979) on quality control in clinicalchemistry. Part 3. Calibration and control materials. J Cliii ChemCliii Biochem 1980;18:855-60.11. Perry TL, Hansen S. Technical pitfalls leading to errors in thequantitation of plasma amino acids. Clin Chim Acts 1969;25:53-8.12. Cooper JDH, Turnell DC. Fluorescence detection of cystine byo-phthalaldehyde derivatisation and its separation using highperformance liquid chromatography. Clin Chim Acts 1982;227:158-61.13. Cooper JDH, Turnell DC, Green B, Wright DJ, Coombes EJ.Why the assay of serum cystine by protein precipitation andchromatography should be abandoned. Ann Clin Biochem1988;25:577-82.14. Gerritsen T, Niederwieser A. Amino acids. In: Curtius HC,Roth M, eds. Clinical biochemistry, principles and methods. Vol 2.Berlin, New York: De Gruyter, 1974;1062-121.15. Ambler RP. Standards and accuracy in amino acid analysis.Op. cit. (ref. 1):119-37.16. Fraser CG. Desirable performance standards for clinical chem-istry tests [Review]. Adv Clin Chem 1983;23:299-339.17. Scriver CR, Roeenburg LE. Amino acid metabolism and itsdisorders. Philadelphia: WB Saunders, 1973:39-60.