Two new automated compared with two enzyme linked immunosorbent antimullerian hormone assays

New automated AMH assays provide substantially lower values than existing AMH enzyme-linked immunosorbent assays. Assay-specific interpretation is necessary, along with international standardization.

Scott M. Nelson, Ph.D., Ewa Pastuszek, M.S., Grzegorz Kloss, M.S., Iwona Malinowska, M.S., Joanna Liss, Ph.D., Aron Lukaszuk, B.S., Lukasz Plociennik, Ph.D., Krzysztof Lukaszuk, M.D., Ph.D.

Volume 104, Issue 4, Pages 1016-1021


To compare new automated antimüllerian hormone (AMH) assay performance characteristics from the new automated Elecsys AMH (Roche; Elecsys) and Access AMH (Beckman Coulter; Access) assays with the existing AMH Gen II ELISA (enzyme-linked immunosorbent assay; Gen II; Beckman Coulter) and AMH/MIS ELISA (Ansh Labs; MIS) assays.

Prospective assay evaluation.

A university-affiliated clinical chemistry laboratory.

Patients referred for serum AMH measurement (n = 83) before start of in vitro fertilization cycle between September 2014 and October 2014.


Main Outcome Measure(s):
Serum AMH concentration.

Intra-assay coefficients of variation were low; MIS ≤ 9.0%; Gen II ≤ 5.8%; Access ≤ 10.7%; and Elecsys ≤ 2.8%. The Passing-Bablok regression equations (pmol/L) were y (Access) = 0.128 + (0.781 × Gen II); and y (Access) = 0.302 + (0.742 x MIS). For y (Elecys) = 0.087 + (0.729 x Gen II) and y (Elecys) = 0.253 + (0.688 x MIS). For y (Elecys) = 0.943 − (0.037 × Access). For all the assays, AMH exhibited a moderate positive correlation with AFC (r = 0.62–0.64); number of cumulus oocyte complexes (r = 0.60–0.64); and metaphase II oocytes (r = 0.48–0.50). Accuracy of pregnancy prediction, as determined by area under the receiver operating characteristic curve, was uniformly low for all assays (0.62–0.63).

The novel automated assays exhibit strong concordance in calibration, but derived values are substantially lower than those obtained from pre-existing assays, with assay-specific interpretation required for routine clinical use. These results highlight the need for an international standard of measurement of AMH.

  • Nigel Groome

    Dear Scott

    I am sure we all value your many contributions to the AMH field and supported your wish for reliable assays. Since the antibodies used in the DSL.Roche, Gen II, Roche and Access assays all came from my lab I naturally follow developments and am keen to make comments helpful to Roche and Beckman and users.

    I have to comment on your latest paper.

    I will first discuss some relevant theory relating to immunoassays and in particular that which explains i believe many of the historical problems and unfortunately still causes confusion.

    Then based on those considerations I will discuss issues raised by the data and comments in your paper.


    Even though automation of AMH assay was possible with my antibodies ten year ago it has been a long time coming and not a minute too late to meet the increasing clinical demand. I hope the contents of this email will make it clear that the problems were entirely predictable when manual ELISAs try to meet a routine clinical need in many countries.Exactly the same issues arose with the manual inhibin assays before Beckman automation.

    When different batches of ELISA plates are made the average affinity of the antibodies can vary according to antibody batch, its prior history which can influence antibody orientation and how quickly and thoroughly the plates are dried.Where the assay goes out into labs shaker speeds, temperature and wash times can vary and some labs may leave plates standing in wash solution for a while while they answer the phone or worst case over lunch. All these things can affect final signal and the poor calibrator has to behave exactly like the analyte in real samples to avoid drift of the means and medians for real samples. Any variable which affects antibody rate constants for the on and off reactions can affect results unless calibrator and analyte in samples behave the same. Reactions in ELISA have more of a kinetic element than some assays on automated systems which can be driven to completion by favorable kinetics of paramagnetic particles in suspension.

    When I developed the first inhibin A and B ELISAs i 1994-1996 and supplied materials to my collaborators they usually did all their key experiments on one batch of plates at one time. The assays were reliable enough in the hands to provide much of our current understanding of inhibin. Once made into ELISAs sold by DSL they immediately took off for Down”s screening but customers observed batch specific medians. This was not a problem as they simply adjusted the medians. Downs screening lab are familiar with shifting medians. Only when the assay was fully automated did these problems resolve. This is exactly the problem with manual AMH assays. Similar problems occur to some extent with all ELISAs but they become apparent when large scale clinical use and universally applicable patient cut offs prove elusive. Exactly with AMH linear correlations between plates and batches was perfect but the slope was variable from batch to batch and lab to lab.
    When comparing different manual assays with each other and even with automated assays there is no universally applicable conversion factor. The HOLY GRAIL of one conversion factor to define the relation of Gen II to DSL ot Gen II to Access or Gen II to Ansh or Gen II to Roche simply cannot exist. , it will be the manual ELISA which cannot hold its consistency in different labs with different batches and different operators. We should feel pleased we now have too automated assays which agree and where from now on will be kept consistent.


    This has to simulate the behaviour of real samples so that whatever affects the calibrator signal affect the patient samples equally.

    The problem with inhibin A and B assays is that the molecule in serum is a complex mixture of isoforms which could not be adequately simulated by either 32 K recombinant or even recombinant mixtures. How can you simulate a mixture perfectly?

    The reasons automation solved the problem are”

    1. The temperature, wash times and agitation are more controlled
    2. The immune reactions are more likely to go to completion because of the more favorable kinetics of paramagnetic particles.

    Thus even if small differences in antibody affinity arise the assay is still much more robust.


    immunotech assay used recombinant human AMH
    DSL and Gen II used bovine serum AMH
    Roche uses bovine serum AMH
    Access uses recombinant human AMH
    Ansh uses recombinant human AMH but from a different source

    The decision to use bovine serum calibrator was made in DSL in 2006 approx . I am not sure of the date. The assumption made was that because new born male bovine serum gave a decent signal it could be used as a calibrator. Howver, unless the capture and detection antibodies recognise bovine AMH and human AMH equally it could mean that they do not match and the assay wil be less robust than it could be.

    When the Gen II kit was launched in 201o it was I believe very similar to the DSL kit. There was no identification of any complement interference issue and the use of bovine calibrators continued. It was declared in the validation paper that it would be an ideal tool for studying the physiological role of AMH. Certainly a lot of studies have been done with it. Personally I would never describe any assay as gold standard just because it has been around a long time.

    The bovine serum in appears to have been a very successful calibrator in the Roche automated assay perhaps for the reasons given above- controlled conditions and antibody excess. It has to be under consideration as one factor making the assay less robust than desirable.

    Is it always better to use a recombinant calibrator.?. Well it can help but even recombinant proteins can be mixtures in predictable and unpredictable ways.
    Recombinant and natural AMH can differ because of tags in the recombinant , variable glycosylation, deamidation of glutamine, oxidation of methionine, local unfolding related to low pH in purification procedure, Any of these changes could make a given antibody react with lower affinity in the recombinant compared to the native molecule. It could even be the other way round if same AMH used to make the calibrator was the same as used to make the monoclonal antibodies. In a worst case scenario one could make a monoclonal better as seeing tagged AMH then normal.

    All these things are not hypothetical . They can and they do occur.

    The one thing that can be said for certain is that when assays are automated these problems all become less significant.

    Quantitation in any immunoassay is based on comparison of analyte in calibrators and real samples. For AMH there is no correct answer only consistent answers. Each antibody sees what it is capable of recognising and if it sees the analyte in serum and the same epitope in the calibrator differently then it will have different kinetics in the various stages of an ELISA. If for example someone leaves a micorplates at the last stage to wash for 30 min thinking it will give lower backgrounds more of the antibodies may let go of the calibrator than the analyte captured from samples.


    I am sorry to say it but the attempt to derive universal clinically constant conversion factors between all the different manual AMH was and is a waste of time for the reasons detailed above..

    Also when comparing the Gen II assay to the Access assay one has to acknowledge that it is more likely to be the manual assay which has shifted its values.

    At one time they called the immunotech assay the gold standard and DSL made every attempt to Gen II to match it.. Not surprisingly when run in Beckman labs there is still close agreement although it was lost for a while because of complement interference.

    However , take Gen II out into widespread use then medians and means will not be constant from lab to lab, operator to operator, batch to batch. This is not just a feature of this assay it is characteristic of ELISAs.

    Each lab may happily run the Gen II assay for years and get good linear correlations with previously tested samples however at some point they will discover changes in slope. They will blame the manufacturer but what they provided may have been very similar.


    1. In a recent paper Van Helden and Weiskirchen (2015) showed excellent correlation
    between the Elecsys and Access assays and a slope of near 1.0. They concluded that the two assays were interchangable for clinical use.

    2. Why then in the paper and in your recent oral presentation in Chicago do you first present slides showing that the Gen II assay has a large variability and another slide
    using this as a criticism of the Access version because it did not give the 1:1 agreement you were led to believe it would have. I can agree that Beckman were unwise to claim that the 1:1 relationship would exist under all circumstances in all labs with all batches. They will make sure that the Access version will be consistent but it is not possible to make improved more stable automated assays always agree with its inferior manual predecessor in each lab it is run..

    3. Much more significant is that the two fully automated assays agree with each other in correlation and in slope. This is a major step forwards. Most of the problems were a legacy from DSL days and I can assure you that Roche and Beckman have put their best efforts into producing these automated assays. It will be big advantage to researchers , clinicians clinical chemists and even patients to have different platforms giving essentially the same values for patients Note that the average difference between the Access and Elecsy assays was only 3% in the previous validation.

    4. All efforts should be made to retain the agreement between the two automated assays and encourage labs to switch to the automated assays. For those customers still running the manual ELISAs they have to appreciate their limitations.and they have to appreciate that when compared to either automated assay they will not necessarily always get the same relationship.

    5.You report intra assay CVs (same day) for each assay but running the controls for each assay in its own assay.. I simply cannot believe these figures are realistic.

    You have a figure of 8-9% for the Ansh assay. I doubt the manufacturers would expect more than 3-4% for control samples run on the same plate.

    You give figures of 5.8% and 1.9% for Gen II . Whilst these look impressive I would have expected the Ansh and Gen II assays to be similar and these are intra assay CVs and would be exceeded by inter assay CV and particularly inter lab CVs as discussed.

    You have figures of up to 2.9% for the Roche assay when they would expect around 1%.

    Most puzzling of all you have figure of 10.76%, 7.4%, and 10.25% .for the Beckman Access AMH assay. There are about 10x higher than expected and make it less reproducible than the manual assays. Do you think that likely when the previous Access validation paper gave intra assay values for its controls of around 1%. consistent with Beckman claims.

    I cannot see it likely that this high CV is being observed in routine practice when so far I have not read about or heard about any complaints ? The labs running AMH on the Access will know what CVs to expect and would be investigation this immediately.

    Did you not think it worth getting Beckman to comment and check your instrument ?

    Did you not see that it is inconsistent with your own data and previous data comparing the Roche and Beckman automated assays? The Van Heldenp paper had almost perfect linear correlation berween the two automated assays and your data in Fig 2C is nearly as good. Did it not occur to you that this fit and the limited scatter would not have been observed if one of the assays had CV of up to 11%?

    6. I would question how well the Ansh assay was run not only because of the high intra assay CVs you find but also because of the scatter observed. All the assays using my antibody pair now give good linear correlations – Gen II, Access, Elecys. Gen II gives good correlation with Gen II and in all these correlations there are no obvious and large outliers.

    However, if you look at your Figs 2A, 2D and 2E all of which compare the Ansh assays with other assays, a substantial number of outliers are observed outside the confidence lines and in several cases these almost half or double the value expected from the regression line.

    I am not sure why you do not comment on the apparently high CV with the Ansh assay and the many outliers which are not observed in comparisons of the other assays.

    7. Confidence in the Gen II pair is reinforced by the fact that the Immuntech assay uses different monoclonal antibodies to the Roche,Gen II and Access assays.

    8. It seems unlikely that the Ansh assay is subject to positive or negative heterophiic antibody or rheumatoid factor interferce but plainly the data do show some differences to all the other assays for a subset of patients. No explanation can be suggested and as I said I am not convinced all the assays in this paper were giviing optimal results.


    1. For the first time we have two fully automated AMH assays on major platforms.
    2. Two independent studies in different countries confirm that they give interchangeable results for clinical samples. If new manufacturers produce automated AMH assays every attempt should be made to get them to calibrate their assays to give the same values. This will make life easier for the clinical community, for reseachers and clinical chemistry labs.
    3. Manual ELISAs are not ideal for routine clinical use for AMH work because they can have systematic shifts in means, batch to batch, lab to lab, operator to operator.
    4. Manufacturers of the automated assays can, should and indeed must keep their results consistent from batch to batch from now on.
    5. This has to be done by serum pool QC testing of each batch and not just testing the kit controls.
    6.In the Roche assay the controls are not human serum but bovine AMH similar to the calibrators. Thus they are really just confirming the curve fit. The control is not
    human serum and may not behave the same. It cannot for certain be relied on to check batch to batch reliability of patient sample values.
    7. In the Beckman assay the control serum are human plasma spiked with recombinant AMH. For the reasons detailed above that is again really like testing like with like and just checks the curve fit.
    8. Only testing of each batch of kits with real serum panels will confirm values customers will expect with that batch.
    9. The three assays using the Gen II antibody pair, Gen II, Roche and Access all give good linear correlation with few outliers. The Gen II assay gives good correlation with
    the Immunotech assay so that we have likely been measuring the same things since 2000.

    10. When compared to the other assays in this study the Ansh assay seems to have more outliers and may in certain samples be recognising something slightly different in the forms of AMH.. Since the intra assay CV is 3 times higher than expected for that assay the possibility that the assay has been run incorrectly should be considered.

    11. The intra assay CVs of over 10 per cent for the Access assay when running its own controls are inconsistent with all expectations, claims and previously published work.. The writer assumes Beckman have been made aware of the issues as he can only assume there is a malfunction with the Beckman instrument used.

    12. All the assays using the Gen II antibody pair recognise ProAMH the inactive precursor and AMH N,C the processed bioactive material. This has been shown in a several papers and a patent by Pankhurst et al.(2014) and in the Roche assay paper,

    I have ten reasons why this is the correct thing to measure at this time but that is another topic


    I urge both Beckman and to Roche not to take these recent developments and improvements for granted. Make sure that each batch of kits is tested against real panels of sera to ensure the slope remains the same. Do not rely on agreement of the kit controls.

    I urge customers of the manual assays to switch to automation if long term consistency of results is important.

    If Beckman wishes to continue to use the Gen II antibody pair in an ELISA I would urge them to switch to recombinant standards and to consolidate their offering into one assay.

    I hope these comments are seen as objective and fair to all the parties and that I have not misunderstood any of the information in your paper.

    I am sorry that the AMH field had so much confusion but believe things will get easier now.

    I hope this is all helpful.


    Professor Nigel Groome
    Emeritus Professor Oxford Brookes University


  • Nigel Groome


  • Nigel Groome

    I am wanting to comment on this paper and just trying out the site to make sure it works and I can see my message.

    Nigel Groome Oxford Brookes University

Translate »