Assessing evidence of immune imprinting in serological patterns of influenza immune response

Reviewer #1 (Public review):

Summary:

This manuscript uses serological data to quantify the effects of imprinting on subsequent influenza antibody responses. While this is an admirable goal, the HI dataset sounds impressive, and the authors developed a number of models, the manuscript came off as very dense and technical. One of the biggest pitfalls is that it is not easy to understand the lessons learned. The two Results section headers make clear statements - there was an imprinting signal in the HI titers, but much of this signal could also be seen in an imprinting-free simulation - and then the Discussion states a number of limitations. This is fine, but it leaves the reader wondering exactly how large an error would be introduced by ignoring imprinting effects altogether; alternatively, if imprinting is purposefully added, what would the expected effect size be? The comments below will provide some concrete steps to help clarify these points.

Major comments:

(1) Lines 107-133: The first Results section is a dense slog of information, and the reader is never given a good overview of what the imprinting coefficients exactly are. As the paper currently stands, if you do not start by reading the Methods, you will take away very little. I suggest adding a schematic for any of your models, showing what HI titers would be expected with/without imprinting effects. or age effects, or both, to tie in your modeling coefficients with quantities that all readers are familiar with.

(1.1) Clarify what the imprinting coefficient (y-axis in Figure 1A) looks like in this schematic.

(1.2) Another aspect that I missed: In addition to stating which models were best by BIC, what is the absolute effect size in the HI titers? During my initial reading, I had hoped that Figure 3 would answer this question, but it turned out to be just an overview of the dataset. I strongly suggest having such a figure to show the imprinting effect inferred by different models. What would the expected effect be if you kept someone's birth year constant but tuned their age? What if you kept their age at collection constant but tuned their birth year?

(1.3) It would also help to explain in your schematic what the x-axis labels (H1, H2, H1/H3) would look like in these scenarios, and what imprinting relative to H3 means.

(2) As mentioned above, it was hard to understand the takeaway messages, such as:

(2.1) A similar question would be: If you model antibody titers without imprinting, how far off would you be from the actual measurements (2x off, 4x off...)? If you add the imprinting effect, how much closer do you get?

(2.2) Are there specific age groups that require imprinting to be taken into consideration, since otherwise HI analyses will be markedly off?

(2.3) Are there age groups where imprinting can be safely ignored?

(3) HI titers against multiple H1 and H3 variants were measured, but it is unclear how these are used, and why titers against a single variant each season would not have worked equally well.

Reviewer #2 (Public review):

Summary:

In this study, the authors were testing the hypothesis that hemagglutination inhibition antibody titers, measured later in life, might be higher against influenza viruses that belong to the same hemagglutinin classification group as the influenza virus that a person was likely first exposed to early in life. This is one conceptualization of a phenomenon termed immune imprinting, which may explain previously observed differences in susceptibility to severe influenza infection between cohorts that were likely first exposed to different hemagglutinin groups. The results of the analysis provide some support for this analysis. However, support for the hypothesis is not consistently observed across sensitivity analyses, and a simulation study finds that antibody patterns consistent with immune imprinting may arise due to other factors in the absence of true imprinting effects. Therefore, overall support for the hypothesis is weak. Nonetheless, this study is important in that it provides guidance and has developed an analytic methodology for additional studies in this area of research. These findings and methods may also be useful for other infectious diseases for which patterns consistent with immune imprinting have been observed.

Strengths:

The strengths of this study include the relatively large cohort data source with broad age representation, rigorous statistical methods, and the use of sensitivity and simulation analyses to assess the robustness of the results.

Weaknesses:

The model outcome includes antibody titers measured against many different viruses, and the imprinting parameter was defined at the subtype level. This may obscure specific imprinting effects related to finer structural similarities between first and subsequent virus exposures. This analysis focuses only on one component of the immune response to influenza; immune imprinting may also involve other immune mechanisms. The analysis was carried out in a Chinese cohort, and vaccination status of the cohort is not discussed; the results may not be generalizable to other populations, particularly if vaccination patterns differ.