On 2015 May 25, Tove Alm commented:

Response to letter by Islam

Shahidul Islam refers to our research article in Science “Tissue-based map of the human proteome” (M. Uhlén et al 23 January, 2015, p. 394) and claims that “they have skipped the control experiments because that would slow down the productivity”. This is of course not true; a large portion of the resources for the project are invested in validation of the antibodies.

A considerable amount of the resources in the program have been used to perform control experiments for the antibodies displayed in the portal and a lot of effort has been put into the visualization of the primary data to allow the scientific community to explore the data behind the control experiments. At present, more than 60,000 antibodies have been validated using Western blots and high-density micro arrays and these efforts have been published in hundreds of articles from our group (www.proteinatlas.org/about/publications). The result in each application is also compared to RNA expression levels and what is known in literature. In addition, to further examine the antibodies, a pipeline for validation of antibody specificity using gene silencing has been established. The results are currently available for a subset of the antibodies in the Protein Atlas.

The open-source policy applied to all antibodies published on the Atlas allows transparency and users may themselves review the experiments supporting the specificity of a particular antibody through the “Antibody/antigen” page (see example: www.proteinatlas.org/ENSG00000141510-TP53/antibody).

The statement by Islam that the antibodies are available through Atlas Antibodies and therefore “reduces the reliability of the immunohistochemistry images” is also difficult to understand. The Human Protein Atlas includes antibodies from more than 40 commercial providers and all primary data supporting the respective antibody is shown as an open resource on the antibody/antigen page for each antibody (see example link above). In the Human Protein Atlas program, a requirement for including an antibody has from the start been that the corresponding antibody must be available to the scientific community through a commercial provider. This is important, since it allows for the use of the same antibody by all researchers interested in human biology and medicine.

As far as we know, there are no past or current efforts with a more comprehensive pipe-line of systematic antibody validations.

On 2015 Apr 19, Md. Shahidul Islam commented:

Immunohistochemistry images published in the “Human Protein Atlas” are not reliable

IN THE ARTICLE “Tissue-based map of the human proteome” (M. Uhlén et al 23 January, 2015, p. 394), the authors point out that they have produced more than 13 million tissue-based immunohistochemistry images by using 24,028 antibodies. These images are freely available from the “Human Protein Atlas” portal. However, when it comes to the interpretation of the images, the key issue is whether the antibodies detect only the proteins that they are intended to do, under the conditions of the immunohistochemistry experiments. What is not widely known is that the scientists behind the “Human Protein Atlas” publish the immunohistochemistry images without performing the proper control experiments. For interpreting immunohistochemistry images, it is essential to perform the appropriate control experiments (1, 2). They have skipped the control experiments because that would slow down the productivity. They have tested the specificity of the antibodies mainly in the protein microarray experiments, and assumed that the antibodies that turn out to be specific in the protein micro array experiments will be specific also in the immunohistochemistry experiments. This assumption is not correct. Only about 45% of these antibodies detect the intended protein in the Western Blot. Scientists cannot publish immunohistochemistry images in peer-reviewed journals, without appropriate control experiments (1, 2). “Human Protein Atlas” is an exception. Here, millions of immunohistochemistry images have been published without requiring any independent peer-review process. The antibodies used for generating the images are being sold by the “Atlas Antibodies”, founded by the researchers from the “Human Protein Atlas” project. This conflict of interest further reduces the reliability of the immunohistochemistry images.

Md. Shahidul Islam Department of Clinical Sciences, and Education, Södersjukhuset, Karolinska Institutet, SE-118 83, Stockholm, Sweden and Department of Internal Medicine, Uppsala University, Uppsala, Sweden. E-mail: shahidul.islam@ki.se

Reference List

1.  R. W. Burry, J. Histochem. Cytochem. 59, 6 (2011).
2.  A. Lorincz, Z. Nusser, J. Neurosci. 28, 9083 (2008).

On 2015 Apr 24, Björn Hallström commented:

We agree that it is not immediately obvious what the colors in Fig 1F refer to, and that this could have been better explained in the figure legend. However, the yellow bands refer to genes that have RNA evidence either from this study or from Uniprot. Since the categories ("Not detected" etc) are determined solely from our RNA data it is possible that Uniprot has RNA evidence for a gene that we could not detect. This is what the yellow band above "Not detected" in Fig 1F refers to. These are generally genes that are expressed in more specialized tissues that were not part of our panel of 32 tissues, such as retina, olfactory bulb and related to hair growth.

On 2015 Mar 05, University of Kentucky Systems Biology and Omics Integration Journal Club commented:

Figure 1F appears to have a significant discrepancy. The definition for the "Not Detected" category in the first column section of Figure 1F is defined as "FPKM < 1 for all tissues" in Table 1. The definition would indicate that there is no RNA evidence for putative protein coding genes in this category; however, there is a yellow band indicating significant number of these genes have RNA evidence.

On 2015 Mar 05, University of Kentucky Systems Biology and Omics Integration Journal Club commented:

Figure 1F appears to have a significant discrepancy. The definition for the "Not Detected" category in the first column section of Figure 1F is defined as "FPKM < 1 for all tissues" in Table 1. The definition would indicate that there is no RNA evidence for putative protein coding genes in this category; however, there is a yellow band indicating significant number of these genes have RNA evidence.

On 2015 Apr 19, Md. Shahidul Islam commented:

Immunohistochemistry images published in the “Human Protein Atlas” are not reliable

IN THE ARTICLE “Tissue-based map of the human proteome” (M. Uhlén et al 23 January, 2015, p. 394), the authors point out that they have produced more than 13 million tissue-based immunohistochemistry images by using 24,028 antibodies. These images are freely available from the “Human Protein Atlas” portal. However, when it comes to the interpretation of the images, the key issue is whether the antibodies detect only the proteins that they are intended to do, under the conditions of the immunohistochemistry experiments. What is not widely known is that the scientists behind the “Human Protein Atlas” publish the immunohistochemistry images without performing the proper control experiments. For interpreting immunohistochemistry images, it is essential to perform the appropriate control experiments (1, 2). They have skipped the control experiments because that would slow down the productivity. They have tested the specificity of the antibodies mainly in the protein microarray experiments, and assumed that the antibodies that turn out to be specific in the protein micro array experiments will be specific also in the immunohistochemistry experiments. This assumption is not correct. Only about 45% of these antibodies detect the intended protein in the Western Blot. Scientists cannot publish immunohistochemistry images in peer-reviewed journals, without appropriate control experiments (1, 2). “Human Protein Atlas” is an exception. Here, millions of immunohistochemistry images have been published without requiring any independent peer-review process. The antibodies used for generating the images are being sold by the “Atlas Antibodies”, founded by the researchers from the “Human Protein Atlas” project. This conflict of interest further reduces the reliability of the immunohistochemistry images.

Md. Shahidul Islam Department of Clinical Sciences, and Education, Södersjukhuset, Karolinska Institutet, SE-118 83, Stockholm, Sweden and Department of Internal Medicine, Uppsala University, Uppsala, Sweden. E-mail: shahidul.islam@ki.se

Reference List

1.  R. W. Burry, J. Histochem. Cytochem. 59, 6 (2011).
2.  A. Lorincz, Z. Nusser, J. Neurosci. 28, 9083 (2008).