Abstract

   Influenza represents a major and ongoing public health hazard. Current
   collaborative efforts are aimed toward creating a universal flu vaccine
   with the goals of both improving responses to vaccination and
   increasing the breadth of protection against multiple strains and
   clades from a single vaccine. As an intermediate step toward these
   goals, the current work is focused on evaluating the systemic host
   response to vaccination in both normal and high-risk populations, such
   as the obese and geriatric populations, which have been linked to poor
   responses to vaccination. We therefore employed a metabolomics approach
   using a time-course (n = 5 time points) of the response to human
   vaccination against influenza from the time before vaccination (pre) to
   90 days following vaccination. We analyzed the urinary profiles of a
   cohort of subjects (n = 179) designed to evenly sample across age, sex,
   BMI, and other demographic factors, stratifying their responses to
   vaccination as “High”, “Low”, or “None” based on the seroconversion
   measured by hemagglutination inhibition assay (HAI) from plasma samples
   at day 28 post-vaccination. Overall, we putatively identified 15,903
   distinct, named, small-molecule structures (4473 at 10% FDR) among the
   895 samples analyzed, with the aim of identifying metabolite correlates
   of the vaccine response, as well as prognostic and diagnostic markers
   from the periods before and after vaccination, respectively. Notably,
   we found that the metabolic profiles could unbiasedly separate the
   high-risk High-responders from the high-risk None-responders
   (obese/geriatric) within 3 days post-vaccination. The purine
   metabolites Guanine and Hypoxanthine were negatively associated with
   high seroconversion (p = 0.0032, p < 0.0001, respectively), while
   Acetyl-Leucine and 5-Aminovaleric acid were positively associated.
   Further changes in Cystine, Glutamic acid, Kynurenine and other
   metabolites implicated early oxidative stress (3 days) after
   vaccination as a hallmark of the High-responders. Ongoing efforts are
   aimed toward validating these putative markers using a ferret model of
   influenza infection, as well as an independent cohort of human seasonal
   vaccination and human challenge studies with live virus.

   Keywords: metabolomics, influenza, vaccine, LCMS

1. Introduction

   Influenza, flu, is a viral infection affecting the respiratory system
   with two major subtypes, influenza A virus (IAV) and influenza B virus
   (IBV), contributing to human disease. The virus is highly contagious
   and airborne, with symptoms ranging from mild to deadly [[32]1]. Both
   IAV and IBV contribute to seasonal infections, while the pandemic
   strains typically arise from the IAV clade. In recent years, the threat
   of pandemic has become more acknowledged, but seasonal (epidemic)
   influenza is still associated with significant morbidity and mortality
   worldwide, with estimates of an average of 389,000 annual deaths
   [[33]2] between 2002 and 2011. Therefore, IAV represents a significant
   public-health issue, and further work is needed to improve the
   prevention, surveillance, diagnosis, and treatment strategies to better
   understand the molecular underpinnings of the immune response to
   IAVfollowing infection and vaccination.

   Infection by IAV or IBV begins with targeting of the epithelial cells
   of the respiratory system by viral hemagglutinin (HA), which mediates
   cell entry, trafficking to the endosomes, and, ultimately, import to
   the cell nucleus, where the transcription of cRNA and vRNA takes place.
   The expression of the viral protein and RNA activates the innate and
   adaptive immune responses, leading to overt symptoms of infection and
   changes in cellular metabolism [[34]3,[35]4,[36]5]. IAV strains consist
   of different combinations of the two surface proteins found on the
   virus, hemagglutinin (HA) and neuraminidase (NA). There are 18 subtypes
   of hemagglutinin (H1-H18) and 11 subtypes of neuraminidase (N1–11)
   [[37]1]. NAs make up approximately 10–20% of influenza surface
   proteins, while HAs make up approximately 80–90% of surface proteins,
   partly explaining why most vaccine designs target the HA protein. IBV
   is similar, and strains are similarly grouped by HA, but divided into
   two different lineages (B/Victoria or B/Yamagata) instead of subtypes.
   Currently, H1N1, H3N2, B/Victoria, and B/Yamagata are all
   co-circulating seasonally in humans. One driver of the need for an
   annual vaccination is antigenic drift, in which mutations
   [[38]6,[39]7,[40]8,[41]9] in the genes that code for the antibody
   binding site, reducing binding recognition by the existing host
   antibodies. Antigenic shift during co-infection further generates the
   potential for novel immune-evading combinations of viral glycoproteins.
   With this reassortment during co-infection, IAV has the potential to
   generate hundreds (256) of unique genetic combinations of the two
   parental strains.

   Because of these challenges, influenza vaccination can lack specificity
   and efficacy [[42]10]. Currently, the seasonal vaccine strains are
   selected based on statistic modelling [[43]11] based on the observed
   configurations and other pre-season metrics. Most seasonal vaccines are
   trivalent or quadrivalent [[44]12], and the vaccination efficacy can
   vary dramatically from year to year, while the responses can be
   population-dependent. This changing landscape of strain selection and
   seasonal vaccination highlights the high-risk populations who exhibit
   poor responses to vaccination. The 2009 H1N1 (IAV) pandemic revealed
   the severity of obesity (BMI > 30) as a leading risk factor for more
   severe infections and higher mortality [[45]13]. The geriatric
   population (>65 years old) has also long been recognized [[46]14] as
   having less robust responses to influenza vaccination, leading to the
   use of high-dose or adjuvanted vaccine designs for these populations.
   For each of these high-risk populations, the mechanism of reduced
   vaccine efficacy is multi-factorial and incompletely understood.

   While research suggests that there is a link between high-risk
   populations and seroconversion following influenza vaccination, not
   much is known about the underlying metabolic mechanisms. Further
   characterization at the molecular level is also needed to understand
   what constitutes a robust immune response to vaccination and link these
   changes to the vaccine design and function of the host immune system.
   The two high-risk populations of interest in the current study both
   share systemic changes in their overall metabolism. Therefore, we
   hypothesized that the metabolic profiles of subjects undergoing
   influenza vaccination may reveal seroconversion-dependent changes in
   their systemic metabolism, which could aid in characterizing,
   identifying, and predicting the biochemical processes mediating a
   robust immune response to vaccination. As the end products of the cell
   regulatory process, metabolism is generally considered to be the most
   sensitive of the omics disciplines at detecting differences associated
   with the phenotype [[47]15,[48]16,[49]17] and is playing an
   increasingly impactful role in the investigation of novel mechanisms of
   pathophysiology. To begin addressing these questions, we examined a
   cohort of healthy adults undergoing annual influenza vaccination to
   identify potential metabolite markers that may be linked to effective
   response to vaccination among high-risk groups through metabolomic
   analysis. Our cohort comprised cross-sectional sampling with respect to
   age, sex, BMI, and other demographic factors so that we could associate
   metabolic changes with the vaccine response. We aimed to use
   seroconversion to the vaccine strains as a proxy of protective immune
   response to influenza vaccination with the goal of then correlating the
   seroconversion score to various time-dependent metabolic changes in the
   general cohort, as well as in the obese and geriatric subsets.

2. Methods

2.1. Vaccine Cohort

   The current metabolomics study utilizes a 2019–2020 cohort of urine
   samples from the University of Georgia (UGA4), which were acquired from
   subjects receiving split, inactivated Fluzone^TM, as previously
   described [[50]18,[51]19]. The study procedures, informed consent, and
   data collection documents were reviewed and approved by the Western
   Institutional Review Board and the Institutional Review Boards of the
   University of Pittsburgh and the University of Georgia. All subjects
   were recruited from the Athens, Georgia geographic region, including
   the University of Georgia. Background demographic data on the
   population were acquired from the Centers for Disease Control (CDC),
   Athens-Clarke County Unified Government, and United States Census
   Bureau. Subjects were excluded from the batch assignment and sample
   processing if one or more of the 5 urine sample time points were
   missing or unavailable. No other exclusion criteria were applied. The
   final study consisted of 179 unique subjects, each with five time
   points for a total n = 895 urine samples analyzed. HAI assays were
   carried out and seroconversion score were obtained as previously
   described [[52]18,[53]19].

2.2. Batch Design and Quality Control

   To minimize the batch effects, the subjects were randomized into 31
   technical batches, and all the sample time points for each subject were
   analyzed together. Each batch, therefore, contained 6 subjects, and
   each of their 5 time points for 30 total samples per batch, excluding
   the final batch. Urine samples were extracted as described below, and
   the order of acquisition was randomized to minimize the sequence
   (within-batch) effects. Each LC-MS sequence contained several control
   blocks of a standard cocktail (blank extraction buffer, with no sample
   but containing internal standards) that were extracted alongside each
   batch. The control block consisted of a blank control followed by a
   standard and another blank. The control blocks were injected at the
   start of the run, in-between every 6 samples, and at the end of the
   run, so that the instrument performance could be monitored throughout.
   These control injections were used to assess the data quality for each
   batch and measure the instrument variance, carry over, and column
   stability. Analytical blank injections were further used to define the
   blank threshold for peak detection in each batch. A large volume (2 L)
   of extraction buffer and 80% methanol for the internal blank controls
   was generated before the study to be used in the technical aliquots for
   all the subsequent technical batch processing.

2.3. Extraction of Metabolites from Urine

   For each batch, the appropriate samples were removed from −80 °C
   storage and transferred to wet ice and thawed. An aliquot of extraction
   buffer of 100% LCMS-grade methanol (Fisher Scientific, Waltham, MA)
   containing 625 nM metabolomics amino acid mix standard (Cambridge
   Isotope Laboratories, Inc, Tewksbury, MA.) was taken from 4 °C storage
   and equilibrated on dry ice for >15 min prior to sample processing.
   Urine samples were extracted by combining 200 µL of the sample with 800
   µL of extraction buffer in 2.0 mL screw cap vials containing ~100 µL
   disruption beads. The tubes were homogenized for 10 cycles in a
   Benchmark Scientific Bead Blaster^TM. Each cycle consisted of 20 s of
   homogenization at 6 m/s, followed by a 30 s pause. The homogenized
   samples were centrifuged at 21,000× g for 3 min at 4°C, and then a
   fixed volume of the supernatant (450 µL) was dried down using speed
   vacuum concentration (Thermo Fisher, Waltham, MA). Once dry, the
   samples were stored at −80 °C until processing. On the day of the LCMS
   data acquisition, the samples were reconstituted in 50 µL of LCMS-grade
   water, sonicated for 2 min, and centrifuged at 21,000× g for 3 min at 4
   °C to exclude any insoluble particulates. The extracted samples were
   then transferred to 2 mL glass vials (Agilent, Santa Clara, CA, USA)
   with glass LC inserts for analysis. All the samples were stored at −80
   °C after data acquisition and QC evaluation. This process was repeated
   for each of the 31 randomized batches.

2.4. LC-MS/MS with the Polar Global Metabolomics Method

   The samples were subjected to an LCMS analysis to detect and quantify
   the putatively identified metabolites. The LC column was a Millipore^TM
   ZIC-pHILIC (2.1 × 150 mm, 5 μm) coupled with a Dionex Ultimate 3000^TM
   system, and the column oven temperature was set to 25 °C for the
   gradient elution. A flow rate of 100 μL/min was used with the following
   buffers: (A) 10 mM ammonium carbonate in water, pH 9.0, and (B) neat
   acetonitrile. The gradient profile was as follows: 80–20% B (0–30 min),
   20–80% B (30–31 min), and 80–80% B (31–42 min). The injection volume
   was set to 2 μL for all the analyses (42 min total run time per
   injection). Each Millipore^TM ZIC-pHILIC column was tracked and used
   only for the urine samples associated with the current study.

   MS analyses were carried out by coupling the LC system with a Thermo Q
   Exactive HF^TM mass spectrometer operating in the heated electrospray
   ionization mode (HESI). The method duration was 30 min, using a
   polarity switching, data-dependent top 5 method for both the positive
   and negative modes. The spray voltage for both the positive and
   negative modes was 3.5 kV, and the capillary temperature was set to 320
   °C, with a sheath gas rate of 35, aux gas of 10, and max spray current
   of 100 μA. The full MS scan for both polarities utilized a 120,000
   resolution with an AGC target of 3e6 and a maximum IT of 100 ms, and
   the scan range was from 67 to 1000 m/z. The tandem MS spectra for both
   the positive and negative mode used a resolution of 15,000, AGC target
   of 1e5, maximum IT of 50 ms, isolation window of 0.4 m/z, isolation
   offset of 0.1 m/z, fixed first mass of 50 m/z, and 3-way multiplexed
   normalized collision energies (nCE) of 10, 35, and 80. The minimum AGC
   target was 1e4, with an intensity threshold of 2e5. All data were
   acquired in the profile mode.

   The quality control for each batch was assessed using the cocktail of
   isotopic amino acid standards present in both the samples and control
   block injections. The sample and standard chromatograms from each batch
   were visually compared against historic references for the expected