Systems biology-based assessment of immune responses to whole cell and acellular pertussis vaccines

Abstract

Given the local and systemic adverse reactions associated with whole-cell pertussis vaccines combined with diphtheria and tetanus toxoids (DTP), acellular pertussis vaccines combined with the same toxoids (DTaP) were developed in the 1990s. In comparison to DTP, DTaP vaccines demonstrated reduced reactogenicity and equivalent or improved immunogenicity and efficacy. However, there has been a resurgence of pertussis disease, particularly in DTaP-vaccinated children, suggesting that immunity wanes more quickly with DTaP vaccination. To elucidate the differences in immune responses to DTP and DTaP vaccines, we employed a systems biology-based strategy to compare global changes in gene expression following primary vaccination with either DTP or DTaP. We used RNA-Seq and ribosome profiling (RP) to identify transcriptional and translational signatures, respectively, in peripheral blood mononuclear cells (PBMCs) collected from 50 infant recipients of DTP or DTaP at two time-points (baseline (pre-vaccination at Day 1) and either Day 2 or 8 post-vaccination). We also used standard serologic methods to assess immunogenicity, and correlated these results with transcriptional and translational signatures. Here, we provide a detailed description of the rationale, experimental design, methodology, and enrollment procedures used. Given the technical complexity of our approach, our objective is to fill knowledge gaps, describe key quality metrics, and support future publications. In brief, we recovered 4–12 million PBMCs (average 8.9 million) with 99% viability per 2.5 mL blood sample, enabling excellent nucleic acid recovery yields for the preparation of high-quality sequencing libraries. In turn, these generated RNA-Seq and RP datasets with sufficient genome coverage breadth and depth to enable differential gene expression analyses, demonstrating the validity of this approach to study pertussis vaccine immunology specifically, and its utility to characterize mechanisms of the human immune response to vaccination generally.

Introduction

In the 1990s, the routine immunization of infants with whole cell pertussis vaccines combined with diphtheria and tetanus toxoids (DTP) was replaced by acellular vaccines combined with the same toxoids (DTaP) in the United States, Australia, and many European countries¹. The primary reason for the change was the frequent occurrence of local and systemic adverse reactions following DTP vaccination, which were significantly reduced with DTaP^2,3. However, over the past decade, there has been a resurgence of pertussis disease, including in children previously vaccinated with DTaP. Data from countries using solely DTaP has suggested that immunity wanes more quickly with DTaP than with DTP^4,5. Data from Australia and the United States has indicated that the rate of reported pertussis cases is significantly lower when the first dose of vaccination is with DTP^6,7. Finally, data from non-human primates has shown that DTaP vaccines prevent disease but not transmission, while both natural infection and DTP vaccines can prevent disease and limit transmission⁸.

Recent vaccine studies have begun to include systems biology approaches like RNA-Seq into their repertoire of analytic tools⁹. Given that RNA-Seq provides both qualitative and quantitative measurements of mRNA levels, it is frequently used as a proxy for global gene expression. However, RNA-Seq only measures mRNA concentration and, therefore, can only provide an assessment of transcription. Not all mRNAs in a cell are actively translated into protein at any given time. Many transcripts remain inert until an appropriate signal induces their translation. As such, accurate measurements of global gene expression require consideration of regulatory processes occurring downstream of transcription at the level of translation. This limitation of RNA-Seq is directly addressed by Ribosome profiling (RP), as it probes mRNAs that are being actively translated into protein by ribosomes^10,11. RP is based on the knowledge that each ribosome mounted on an mRNA covers a ~30 nt “footprint.” Upon the addition of a nuclease, the ribosome protects that footprint from nucleic acid digestion, generating a record of its location in the process. In essence, the position of every translating ribosome on every mRNA in a cell can be determined with great accuracy by assessing the protected footprints generated during nuclease digestion. Furthermore, since only a portion of all mRNAs present in a cell are translated at any given time, and since translation rates vary between specific mRNA templates¹², the number of footprints measured is directly proportional to the amount of new protein synthesis. Thus, when used in combination, RNA-Seq and RP provide an unprecedented and more comprehensive view of global gene expression that considers regulatory processes occurring at both the transcriptional and translational levels.

RP has not been used to assess vaccine responses in children, and traditional methods have not provided an explanation for the observed differences in immune responses to DTP and DTaP vaccines^13,14. Given the availability of both DTP and DTaP in many low- and middle-income countries, we sought to define the immunological mechanisms underlying the improved durability of protection afforded by DTP relative to DTaP using a systems biology approach that included RNA-Seq and RP to simultaneously probe transcription and translation, respectively. Our primary goal was to uncover additional insights into the innate and adaptive immune responses to pertussis vaccines and to define specific differences through direct measurements of differentially expressed (DE) genes at early time points, even as early as 24 hours post-vaccination. Our hypothesis was that DTP vaccine induces a more robust and durable immune response, and further, that this response may manifest early, as measurable differential gene expression signals.

Given that RP is a novel approach to evaluating vaccine immunology, we established both primary and secondary objectives that would collectively provide an unprecedented global interrogation of the human immune response. As a primary objective, we sought to measure post-vaccination changes in gene expression at two different time points (24 hours and 7 days post-vaccination), and to compare changes in gene expression between vaccine groups (DTP and DTaP) following the first vaccine dose administered at ~2 months of age. As a secondary objective, we wanted to compare the gene-expression responses measured by RNAseq and RP with those obtained using traditional immunological tools, like antibody responses, inflammatory cytokine production, cell-mediated immunity, and B-cell antibody repertoires. However, before these comparisons could be made, we first needed to establish basic metrics of quality for RNA-Seq and RP workflows, as our approach had no precedent and working in the context of a vaccine study where the participants were infants presented several significant challenges. RP is a technically demanding methodology because actively translating ribosomes are machines in motion, so samples need to be processed quickly and with great technical expertise to prevent them from dissociating from mRNAs during handling. In addition, because our participants were infants, we were limited to working with very small blood volumes (≤3 mL), which increased the degree of difficulty and risk in our approach. The work presented here is intended as a roadmap that provides technical insights and defines metrics of quality. We provide details of our experimental design, assays used, and quality metrics defined, and we describe how various experimental and operational challenges were addressed. Detailed findings as they pertain to pertussis vaccines, including characterization of specific DE genes and immune pathway enrichment results, are described elsewhere (Creech et al.¹⁵).

Results

Study design

To address the technical challenges intrinsic to an approach where RP was conducted on limited amounts of material derived from infant participants, we first divided the study into two parts: a Pilot Study (PS) and a Main Study (MS) (Fig. 1). The PS (n = 10) was designed to evaluate and optimize sample processing workflows for differential gene expression analyses, while the MS (n = 40) would incorporate adjustments, if needed, identified during the PS, and incorporate an additional time point for testing. The sample size (n = 50) was not based on a formal hypothesis, but rather, on practical considerations needed to set a foundation on which to probe the immune responses to DTP- and DTaP-vaccines at the molecular level using RNA-Seq and RP. The assignment of infants into either vaccine group was determined by the choice of parents/legal guardians, who elected to immunize their children either in a public health clinic (and receive DTP) or in a private clinic (and receive DTaP). In the MS, however, randomization was used to assign infants to one of two study arms (Arm 1: sample collection at Day 2 (24 h post-vaccination), or Arm 2: sample collection at Day 8 (7 days post-vaccination)) using a block random design with random block sizes of 2 or 4 stratified by vaccine group. All laboratory personnel had previous experience handling RP workflows and were blinded to sample metadata including subject ID, vaccine group, and study visit. RNA-Seq and RP workflows were carried out in batches, according to sample pick lists designed to reduce the impact of batch effects on downstream analyses. In the PS, infants were randomly assigned to batches of four samples each that always included one infant receiving DTP and one infant receiving DTaP. In the MS, infants were further paired by study day such that batches included samples from either Day 2 or Day 8.

Fig. 1. Study design was separated into two parts, a Pilot Study (PS) and a Main Study (MS).

The PS was used to optimize omics workflows, while the MS incorporated technical adjustments identified during the PS and further subdivided participants into an assessment of differential gene expression at early (24 h post-vaccination) vs. late (7 days post-vaccination) time points. Details on the samples collected and various assays performed in each samples are provided in Tables 2 and 3.

Optimization of RP and RNA-Seq Experimental Protocols Based on Pilot Study Findings

In the PS we processed samples from 10 infants (n = 5 for DTP; n = 5 for DTaP) who supplied blood at baseline (Day 1, n = 10) and 24-hrs post-vaccination (Day 2, n = 10). These 20 samples were used to experimentally establish metrics of quality for RP and RNA-Seq workflows (Supplementary Data 1 for complete data set). Pre-processing of blood yielded an average PBMC count of 8.9 million (range 4–11 million), and an average total RNA concentration of 1426 ng (range 0–5035 ng) per 2.5 mL blood draw (Table 1). We expected that PBMC counts would closely correlate with RNA yields, and thus potentially serve as an early metric of quality, but this was not the case (Table 1, Supplementary Fig. 1). Instead, the best predictor of sequencing library quality was total monosome concentration for RP libraries (Spearman of rs = 0.44), and both total RNA and total monosomes for RNA-Seq libraries (Spearman of rs = 0.22) (Table 1, Supplementary Fig. 1).

Table 1.

Quality metrics used to assess wet-lab workflows and pre-processing for RP (orange columns) and RNA-Seq (green columns) in the PS (complete data set in Supplementary Data 1)

For comparison purposes, we also include a summary of the same metrics obtained during the MS (dark green row, complete data set in Supplementary Data 1). Cells are color-coded using a heat-map, where desirable counts are shown in blue, and less desirable counts are shown in red. Samples that failed quantification or sequencing are shaded grey. For quick comparison purposes, we also show average values for PS (light green row) and MS (dark green row) samples.

Given that RP workflows are more technically demanding than RNA-Seq workflows, we attempted to maximize signal-to-noise ratios by allocating ~80% of the PBMC pellet to RP and only ~20% to RNA-Seq. However, 8/20 samples with low nucleic acids did not yield RP libraries, and 4/20 did not yield RNA-Seq libraries (Table 1, shaded in gray). The most likely explanation for this was the elimination of cycloheximide (CHX) from the PS, a change initially introduced to test if we could accommodate cytokine measurements from plasma generated from the same blood sample used to generate PBMCs for RNA-Seq and RP. However, since our priority was to construct high-quality sequencing libraries for successful omics work, in the MS, we switched cytokine measurements to serum and brought back CHX, with significant overall improvements, as shown below.

We then assessed basic sequencing run metrics, like percent (%) chip loading and total number of raw and usable reads (usable reads do not include polyclonal and low-quality reads). Chip loading for RP workflows on the Ion Proton instrument ranged from 89–96% (average 94%), while chip loading for RNA-Seq workflows was lower and ranged from 70–93% (average 87%) (Table 1 and Supplementary Data 1). The difference in chip loading between RP and RNA-Seq libraries is not surprising, as individual reads in RP libraries are smaller (~160 bp vs. ~200 bp) and more homogeneous in size than in RNA-Seq libraries (Fig. 2), making them easier to evenly distribute over the entire surface of the chip during loading. Despite this difference, however, both RP and RNA-Seq sequencing metrics displayed average chip loading levels above the 85% ideal recommended by the manufacturer. As a result, our sequencing runs routinely generated large datasets with an average of 118 million usable reads/sample for RP, and 71 million usable reads/sample for RNA-Seq (Table 1). Still, the fact that we were able to generate large numbers of usable reads/run did not guarantee that the libraries were of sufficient breadth and depth to support quality downstream analyses of DE genes, especially in the case of RP, where there was no precedent for these types of experiments. Thus, to establish early metrics of RP library quality we also assessed basic bioinformatics metrics that included % read recovery following trimming of 5’ and 3’ adapters, % read recovery following quality (Phred scores <20 were eliminated) and length filtering (reads outside the typical 25–35 nt length range of ribosomal footprints were eliminated), % contaminant levels (rRNA is a known and abundant contaminant in all RP workflows), and numbers of reads mapped to the human genome (Supplementary Data 1). In the PS, the % of input reads remaining after trimming of 5’ and 3’ adaptors ranged from 55–78% (average 68%) of the initial usable reads, thus, trimming eliminated on average ~32% of the total available reads. The % of reads remaining after subsequent quality and length filtering steps ranged from 40–69% (average 58%), where an additional ~6% of the total available reads were also eliminated. We then assessed the proportion of rRNA contamination, which ranged from 33–63% (average 52%), despite carrying out two rounds of rRNA depletion during wet-lab processing. This left 6–11 million (average 8 M) “clean” reads to map to the human genome, which represented 5–11% (average 7%) of the initial usable reads generated during sequencing. Finally, during mapping to the human genome, 2–8 M reads (average 4 M, ~54% of “clean” reads) aligned to protein coding regions.

Fig. 2. Quality control of RP and RNA-Seq libraries.

Representative bioanalyzer profiles for RP (A and B) and RNA-Seq (C and D) libraries. RP libraries are smaller (~160 vs. ~200 bp) and more homogeneous in size than RNA-Seq libraries.

RP and RNA-Seq quality metrics based on main study findings

In the MS we processed samples from 40 infants (n = 20 for DTP; n = 20 for DTaP) who supplied blood at baseline (Day 1, n = 40) and either 24-hrs (Day 2 in DTP, n = 10; Day 2 in DTaP, n = 10) or 7 days post vaccination (Day 8 in DTP, n = 10; Day 8 in DTaP, n = 10). These 80 samples were processed as in the PS, except that CHX was used to increase the number of monosomes. We recovered the same number of PBMCs as we did in the PS (range 4–12 million, average 8.9 million, with 99% viability per 2.5 mL blood draw), but the addition of CHX had a notable positive impact on the amount of total RNA and monosomes isolated, as expected (Supplementary Data 1 for complete data set, Table 1 for summary MS averages relative to the PS and Supplementary Fig. 1 for correlation analysis). On average, in the MS, we obtained 3X more nucleic acid (3998 ng total RNA) than in the PS (1426 ng total RNA) (Table 1). In addition, fewer samples had questionable nanodrop quantifications (values below 2.5–3 ng/uL total RNA are not reliable) in the MS (7 of 80 = 8.75%) than in the PS (8 of 20 = 40%) (Supplementary Data 1). Overall, improvements in total RNA amounts translated into improvements in total monosomes. However, as in the PS, PBMC counts did not predict RNA concentrations. For example, sample MS28 had a high PBMC count (8 M) but very low total RNA (375 ng), while sample MS5 had a relatively low PBMC count (3.2 M) but high total RNA (4513 ng). Taken together, the numbers show that when the amount of starting monosome RNA is around 100 ng or less, or when the overall quality of material is questionable from the start for any reason, the probability of generating quality sequencing libraries drops, particularly for RP, which is not surprising given the higher technical difficulty associated with RP processing.

Our goal for the MS was to generate 80 RP libraries (n = 40 from pre-vaccination baseline samples at Day 1 and n = 40 from follow-up samples at Day 2 (n = 20) or Day 8 (n = 20)) and 80 RNA-Seq libraries (n = 40 from baseline and n = 40 from follow-up). We were successful for most of our target libraries, generating 88% (70/80) in the case of RP (47% improvement over the PS) and 99% (79/80) in the case of RNA-Seq (14% improvement). Of the RP libraries lost, 4 were lost due to insufficient starting material, 4 to likely degradation from RNAse contamination, and 2 to technical error. Only 1 RNA-Seq library was lost without a likely explanation (Supplementary Data 1). Library quality metrics also improved in the MS, where the average concentration of nucleic acid in each RP library increased from 526 pg in the PS to 724 pg in the MS (38% improvement). Similarly, measurements of undesirable “empty” RP libraries that contain nothing other than adapters decreased from 72.3 pg in the PS to 3.2 pg in the MS (96% improvement). A similar pattern was observed for RNA-Seq libraries, where the average concentration of nucleic acid in each library increased from 2.14 ng in the PS to 12.08 ng in the MS (465% improvement). Finally, library size distributions were much more homogeneous in the MS than in the PS. Collectively, the surrogates of quality generated during wet-lab processing indicate that the addition of CHX back into our workflows had a significant positive effect on both the amount and quality of libraries prepared.

We then assessed basic sequencing metrics, as we had done in the PS. Beyond slight improvements in the overall metrics of the MS, these did not significantly change relative to the PS (Supplementary Data 1), which is not surprising given that sequencing metrics are independent of sample and library quality metrics. In other words, good quality starting samples are required to make good quality libraries, but once a library is made, a good quality library is just as likely to be sequenced successfully as a bad quality library. Our sequencing metrics remained outstanding in the MS, with no differences in chip loading for RP samples (94% in the PS vs. 93% in the MS), and a slight improvement in chip loading for RNA-Seq samples (87% in the PS vs. 90% in the MS). In both cases, these metrics were considerably above the 85% optimum suggested by the manufacturer, and as a result, our sequencing runs continued to generate robust datasets, with an average of 129 million usable reads/sample for RP (vs. 117 M in the PS) (10% improvement), and 95 M usable reads/sample for RNA-Seq (vs. 72 M in the PS) (34% improvement). Other metrics of sequencing quality also improved. The total number of undesirable polyclonal reads, which do not yield usable data, decreased from 8% in the PS to 4% in the MS (50% improvement) during RP workflows, and from 23% to 22% (4% improvement) during RNA-Seq workflows. Similarly, the total number of undesirable low-quality reads decreased from 7% in the PS to 3.5% in the MS (50% improvement) during RP workflows, and from 18% to 8% (56% improvement) during RNA-Seq workflows. Taken together, the numbers show that improving library quality by adding CHX, while also maintaining sequencing run quality, had a significant positive impact on the number of reads generated.

As in the PS, basic bioinformatics metrics assessed in the MS for RP libraries included % read recovery following trimming of 5’ and 3’ adapters, % read recovery following quality and length filtering, % rRNA contamination, and number of reads mapped to the human genome (Supplementary Data 1). The number of input reads remaining after trimming of 5’ and 3’ adaptors ranged from 76–118 M (average 102 M reads (79%) vs. 81 M (68%) in the PS) (11% improvement). In essence, trimming eliminated on average ~21% (vs. ~32% in the PS) of the initial usable reads (52% improvement). The number of reads remaining after subsequent quality and length filtering steps ranged from 44–112 M (average 82 M (64%) vs. 69 M (58%) in the PS) (10% improvement). The amount of rRNA contamination remained unchanged with respect to the PS, as expected, and ranged from 31–96 M (average 70 M (55%) vs. 61 M (52%) in the PS), leaving 5–28 M (average 12 M (10%) vs. 8 M (7%) in the PS) “clean” reads to map to the human genome. Finally, during mapping to the human genome, 2–29 M reads (average 7 M (5%) vs. 4 M (3%) in the PS) aligned to protein coding regions (66% improvement).

The bioinformatics metrics assessed for RNA-Seq libraries also showed improvements (Supplementary Data 1). On average we worked with libraries containing 94 M reads (vs. 71 M in the PS) (32% improvement), and these had significant sequence breadth and depth because on average, ~81% of the reads remaining following filtering steps uniquely mapped to the human genome and covered ~20 thousand individual genes (Creech et al.¹⁵). Given that the human genome is between 20–25 K genes, that not all genes are expressed by all cell types at any one time, and that PBMCs are a heterogenous population of cells, we believe that targeting ~20 K individual genes is a very good indication that the libraries prepared resulted in sufficient genome coverage breadth and depth for meaningful downstream differential gene expression analyses (Creech et al.¹⁵). Taken together, the data highlight the importance of having informative quality metrics and show that even modest improvements throughout the technical process (such as inclusion of CHX) can have a significant positive impact on the number and quality of reads generated during the construction of both RP and RNA-Seq libraries.

Validation of omics pipeline based on measurements of differential gene expression

The final litmus test to validate our omics approach was to determine whether we could identify DE genes, at the levels of transcription and translation, as measured by RNA-Seq and RP, respectively. To do so, we compared baseline levels of gene expression (pre-vaccination at Day 1) with follow-up levels of gene expression (24 h post-vaccination at Day 2, or 7 days post-vaccination at Day 8) in the two vaccine groups. We detected DE genes in both vaccine groups (DTP and DTaP) and time points (Day 2 and Day 8) (Fig. 3). Day 2 showed more DE genes (2693 total DE genes: 1289 up-regulated and 1404 down-regulated) than Day 8 (395 total DE genes: 164 up-regulated and 231 down-regulated), indicating this approach is particularly suited to detect early changes in gene expression post-vaccination. Day 2 also showed more differentially transcribed (DTSC) (2172 total DTSC genes at Day 2: 947 up-regulated and 1225 down-regulated) than translated (DTSL) genes (1021 total DTSL genes at Day 2: 749 up-regulated and 272 down-regulated), indicating that transcription and translation are under independent regulatory control and that specific measurements of transcription via RP are essential for a complete picture of global gene expression. In fact, the data shown in Fig. 3 provides an unprecedented level of granularity that illustrates similarities and differences between vaccine groups, time points, and levels of transcriptional and translational control. For example, the highest number of uniquely identified DE genes was found at Day 2 in the DTP vaccine group using RNA-Seq (226 up- and 318 down-regulated uniquely DE genes), followed by the DTaP group by RNA-Seq (138 up- and 356 down-regulated uniquely DE genes), the DTP group by RP (191 up- and 133 down-regulated uniquely DE genes), and finally DTaP group by RP (61 up- and 34 down-regulated uniquely DE genes). For Day 8, the highest number of uniquely identified DE genes was found in the DTaP vaccine group using RNA-Seq (162 up-regulated and 228 down-regulated uniquely DE genes). What is most notable and relevant, however, is that only a portion of the genes identified as DE at the level of transcription were also DE at the level of translation, suggesting the importance and need to consider mechanisms of translational control for a proper illustration of global gene expression. For example, in the DTP vaccine group, 1620 genes were DTSC and 874 were DTSL, but of these, only 409 (331 up- and 78 down-regulated) were co-differentially regulated. Similarly for the DTaP vaccine group, 1527 genes were DTSC and 494 were DTSL, but of these, only 259 (243 up- and 16 down-regulated) were co-differentially regulated. We used this subset of co-DE genes to specifically calculate translational efficiency by subtracting the RNA-Seq log₂ fold change from the RP log₂ fold change of each gene (Fig. 4). As shown, translational efficiencies tended to be greater than 1 for most DE genes, indicating that translation exceeded transcription, and confirming the existence of an important degree of independent regulation of translational control of gene expression. Complete details on the gene expression patterns between vaccine groups, the timing and magnitude of each response, and the identity of individual DE genes identified are described elsewhere (Creech et al.¹⁵). Here, we have focused on the overall sensitivity and pertinence of our omics approach, particularly to ascertain the benefit of including measurements of DTSL genes in addition to DTSC genes and the validity of our approach to interrogate the human immune response at an unprecedented level of granularity.

Fig. 3. DE genes identified at the levels of transcription (measured by RNA-Seq) and translation (measured by RP) at two different time points (24 h post-vaccination at Day 2, and 7 days post-vaccination at Day 8) for two vaccines groups (DTP and DTaP).

Venn diagrams show up-regulated genes in red and down-regulated DE genes in blue.

Fig. 4. MA plots displaying translational efficiency for Day 2 and Day 8 in both vaccine groups (DTP and DTaP).

Genes that were differentially transcribed (measured by RNA-Seq) and translated (measured by RP) in the two vaccine groups were used to calculate translational efficiency by subtracting the RNA-Seq log2 fold change from the RP log2 fold change. Dots above zero indicate individual genes where translational efficiency exceeded transcriptional efficiency (and vice-versa for dots below zero). Collectively, the data indicate clear differential regulation of transcription and translation.

Discussion

The main goal of this study was to characterize infant immune responses to DTP and DTaP vaccines at a global level. To do so, we used RP, as this technique can provide direct measurements of gene expression at the level of newly synthesized protein by identifying the total number and exact position of all actively translating ribosomes¹⁶. We also performed a complementary RNA-Seq analysis on all samples, which allowed us to evaluate translational efficiency¹⁶. In mammals, translational efficiency is the single best predictor of protein levels, as mRNA abundance alone can only account for ~40% of the variability observed¹⁷, which is also subject to a broad dynamic range of up to 10-fold¹⁸. Depending on the tissue, the number of genes detected transcriptomically can range from ~11,000–15,000 of the ~20,000 total human protein-coding genes¹⁹. RP can routinely detect ~11,000–12,000 genes/sample, which is higher than the ~9000–11,000 proteins/sample detected using mass-spec proteomics^16,20. In addition, RP is predicted to be more sensitive to lowly expressed proteins than mass-spectroscopy²⁰. Taken together, the evidence points to RP as a better phenotypic predictor than RNA-Seq alone. However, including RNA-Seq measurements was also key because it provided a back-up strategy in case RP failed to generate measurable signals in the complex technical context of our study, which had no precedent. Several challenges were identified during protocol development and implementation, including: (1) limited volumes of infant blood (3 mL/draw); (2) ribosomes that are in motion as they are being isolated, and therefore extremely susceptible to lengthy manipulation protocols; and (3) a heterogeneous population of PBMCs from which to extract meaningful gene expression signals.

The main issue associated with limited blood volumes was whether we would obtain sufficient material (nucleic acid from PBMCs) to sustain differential gene expression analyses at both the transcriptional (measured by RNA-Seq) and translational levels (measured by RP). Each 3 mL blood sample needed to sustain not only omics assays, but also other immunologic assays (pertussis antibodies and serum/plasma cytokines) needed to compare approaches. To maximize signal-to-noise ratios for omics work, we opted to allocate as much blood as possible (2.5 mL out of 3 mL blood sample) to recover PBMCs for RP and RNA-Seq, and we collected these samples in blue-top tubes containing sodium citrate to prevent coagulation. The remaining 0.5 mL were used to isolate serum and were collected in red-top tubes with no additive. Given that plasma is a byproduct of PBMC isolation, we considered that perhaps we could accommodate cytokine measurements from plasma, provided we opted to forego the use of CHX, as it interferes with cytokine measurements. CHX is a translation elongation inhibitor routinely used in RP workflows because it significantly increases the number of footprints obtained by immobilizing ribosomes in place during handling. At the time that PBMCs were prepared, little was known about the effects of omitting CHX from RP workflows beyond the logical assumption that it would likely decrease monosome counts. Today, the issue of whether CHX affects monosome position on actively translated mRNAs is a matter of active debate^18,21,22 and includes reports that human ribosomes are resistant to CHX-mediated biases²³. While this issue may be important in the context of translation studies where the actual position of monosomes on specific mRNAs is crucial, in this study, we only needed to count total monosomes/mRNA to get a quantitative measurement of protein expression levels. Hence, the issue of CHX-mediated bias was much less of a concern than losing total monosome counts in the absence of CHX. Actively translating ribosomes are machines in motion, so they will eventually disengage mRNA. However, the rate at which they uncouple is both time- and temperature-dependent. Thus, we were extremely careful to adapt experimental conditions to work at near-freezing temperatures and as rapidly as possible after sample collection to minimize ribosome decoupling.

Given that we opted to forego CHX to accommodate cytokine measurements from plasma, we set up the PS (n = 10) to optimize RP workflows and assess whether the CHX omission was viable. Specifically, we evaluated rapid PBMC generation on-site immediately after blood draw. PBMCs were isolated, counted, assessed for viability, and then immediately frozen in liquid nitrogen and stored until they could be batch-processed in the laboratory. Once all blood samples had been collected and prepped into PBMCs, we moved our workflows to the lab, where we addressed working as cold and as fast as possible by always keeping frozen PBMCs on ice and using ice-cold solutions throughout the monosome isolation process. We also attempted to maximize signal-to-noise ratios by assigning 80% of the PBMCs collected to RP workflows that were more technically sensitive and only 20% to RNA-Seq workflows that were technically easier. The results of the PS demonstrated that although we were able to generate data to measure differential gene expression, we were only able to generate 12/20 (60%) target RP libraries and 16/20 (80%) target RNA-Seq libraries. The libraries that failed were linked to samples with low nucleic acid concentrations, and particularly to low monosome concentrations, which immediately suggested that the lack of CHX was the most likely explanation for the issue. In fact, we easily remedied the problem in the MS by adding back CHX and switching cytokine measurements to serum. This change resulted in very significant overall improvements in library quality and sequencing counts both for RP and RNA-Seq workflows. We were able to increase the total number of RP reads mapped to the human genome from 4 to 7 million between the PS and MS, which represented a 66% improvement in the number of reads mapped.

From a technical perspective, the main take-home lessons were five: (1) Separating the study into two phases (PS and MS) was key to tackling procedural issues that could only be addressed experimentally, such as whether to use CHX. (2) CHX significantly improved signal-to-noise ratios and should not be omitted unless there is an alternative way to increase starting nucleic acid concentrations. In our case, that was not possible given that we were working with limited volumes of blood also needed for other assays beyond RP and RNA-Seq. (3) An n = 10 in the PS was sufficient to identify potential pitfalls that could affect sample collection and/or processing, and to experimentally address them. (4) Keeping detailed quality metrics along the entire RP and RNA-Seq workflows (pre-processing, wet-lab library prep, sequencing, and bioinformatics) allowed us to identify specific metrics that predicted library quality prior to sequencing (i.e., nucleic acid concentrations rather than PBMC counts). 5) Despite two rounds of rRNA depletion during wet-lab procedures, rRNA contamination was a significant source of noise in our RP libraries, occupying on average 55% in the MS sequencing reads (vs. 52% in the PS). Although rRNA is a known contaminant in all RP workflows, sometimes occupying as much as 90% of the sequencing reads generated^16,24, this is likely an area where significant gains could be made. We suggest adding rRNA depletion oligos to the existing oligo pool based on the most common sequences identified as contaminants. Aside from these main take-home lessons, we show that we got good yields of PBMCs with high viability throughout the study, that our handling procedures were robust enough to generate quality sequencing libraries, and that sequencing of these libraries resulted in data that enabled differential gene expression analyses at different time points (baseline at Day 1 and follow-up at Day 2 or Day 8 post-vaccination) in two different vaccine groups (DTP and DTaP).

The specific results of differential gene expression analyses in response to DTP or DTaP vaccination, along with the comparisons and correlations with traditional immunological assays executed as part of this study, will be reported elsewhere (Creech et al.¹⁵). Here, we have presented a thorough technical summary of our systems biology-based approach to characterizing the human immune response following vaccination. In that respect, we are concerned with the utility of RP as a tool to measure the signatures of immunogenesis, rather than with the particulars of the pertussis-specific immune response. However, the fact that we attempted to extract meaningful gene expression signals from a heterogeneous population of cells warrants specific additional commentary. PBMCs isolated from adults contain 70–90% lymphocytes (70–85% CD3+ T cells, 5–10% B cells, and 5–20% NK cells), 10–20% monocytes, and 1–2% dendritic cells. The composition of PBMCs isolated from infants is expected to differ²⁵, and contaminant platelets and granulocytes (neutrophils and basophils) present in PBMC preparations^26,27 may also contribute unwanted signals. This complicates the interpretation of systems biology data obtained from PBMCs and highlights the importance of a thorough analysis of statistical significance, particularly in the context of individual genes. The heterogeneity contained within these libraries means that the reads representing the responses of any individual cell subtype are likely not present at a level of saturation that ensures representation of the entirety of those specific response pathways. In other words, any given gene set will be far from saturation due to the mixed nature of the cell population, and therefore, individual genes belonging to a specific gene set or genetic pathway may be underrepresented as DE genes, especially if their normal level of expression is low. Furthermore, because the information about gene status (DE vs. non-DE) carried by any individual gene is small, the best metric to ascertaining the nature of the DE signal is within-pathway gene counts for multiple genes, rather than individual gene counts. Finally, because several cell subtypes may contribute signal to within-pathway gene counts, and because a portion of the response to DTP and DTaP is likely similar and overlapping, the dissection of DE gene signals is nuanced and requires careful consideration. Our approach has been to first collect all the information pertaining to a particular gene pathway or cellular activity, and then to follow with a global interpretation of that phenotype. Previously, we showed that with a similar approach, we could detect meaningful DE gene signals in a context where only 3% of cells in the population were expressing the phenotype of interest²⁸. Here, we highlight the fact that we have been able to detect specific differences in DE genes in response to vaccination at the levels of transcription (as measured by RNA-Seq) and translation (as measured by RP), in different vaccine groups and at different time points. Furthermore, the number and identity of DE genes identified varies within and between vaccine groups and time points, which is expected given that transcription and translation are separate gene expression processes regulated by distinct mechanisms¹². Our translational efficiency analysis supports these observations, as we show high levels of translational control in both vaccine groups. This also justifies the unprecedented use of RP in this type of study, as our combined approach provides improvement over systems biology approaches that solely use RNA-Seq and yet extrapolate those measurements as a proxy of global gene expression. This study is the first to use RP in infants receiving DTP and DTaP, and the data obtained confirms that we can simultaneously evaluate transcriptomic and proteomic responses to vaccination at an unprecedented level of detail. By establishing baseline metrics of quality, we hope our study will serve as a road map for others in the field of vaccinology.

Methods

Participant enrollment and follow-up

The study was conducted in the San Juan de Lurigancho district of Lima, an area served by both the Peruvian Ministry of Health, where the pentavalent DTP vaccine is available free of charge, and by private health clinics, where the hexavalent DTaP vaccine is available for a fee. Supplementary Table 1 provides a summary of vaccines given routinely to infants in Peru. Parents/legal guardians of all enrolled infant participants signed IRB-approved informed consent documents according to all applicable regulations in both Peru and the US. Inclusion criteria were: (1) healthy, full-term infants (≥37 weeks gestation) aged 50–89 days at the time of enrollment; (2) eligible to receive first licensed DTP or DTaP-containing vaccine as per the standard of care in Peru; and (3) written informed consent from the parent or legal guardian. Exclusion criteria were: (1) low birth weight (<2.5 kg); (2) blood draw via venipuncture not possible following a maximum of 2 attempts per visit; 3) severe anemia (Hemoglobin <7 g/dl); (4) history of blood transfusions or receipt of blood products, including maternal transfusions while in utero; (5) steroid treatment at the time of enrollment, or history of prolonged steroids treatment (>14 days) between birth and enrollment; and (6) primary immunodeficiency or HIV infection/exposure. Transient exclusion criteria were: (1) fever within 24 hours (>37.9 ˚C); and (2) acute illness. Infants were enrolled one at a time over a 2-week period in the PS (n = 10), and over an 11-week period in the MS (n = 40). The duration of participation was ~17 months per child, during which blood samples were provided 5 times (Table 2).

Table 2.

Summary of study schema, including timing of sample collections relative to vaccination schedule and assays performed on the various samples collected

	Visit 1:	Visit 2:	Visit 3:	Visit 4:	Visit 5:
Timepoint	Day 1	Day 2 or 8	~Day 150	~Day 480	~Day 510
Timeline	Pre-vaccine baseline at 2 months of age	Post-1st vaccine dose at 2 months of age	~1 month post-completion of primary vaccination series given at 2, 4 and 6 months of age	Pre-booster dose at ~18 months of age	~1 month post-booster dose at ~19 months of age
Assays performed	• RNA-Seq and RP	• RNA-Seq and RP	• Pertussis protein-reactive B-cell measurements	• Pertussis protein-reactive B-cell measurements	• Pertussis protein-reactive B-cell measurements
	• Pertussis antibodies	• Pertussis antibodies	• Pertussis antibodies	• Pertussis antibodies	• Pertussis antibodies
	• Cytokines	• Cytokines	• Cytokines

Blood samples

Samples collected during each visit had specific needs for various blood products (serum, plasma, PBMCs), requiring allocation into either red-top (no additive) or blue-top (sodium citrate) collection tubes. A summary of PS and MS specimen volumes collected and their intended use is summarized in Table 3. Blood draws never exceeded 3 mL per visit. Red-top tubes were used to collect serum for assessment of pertussis antibodies (PS and MS) and cytokine measurements (MS only), while blue-top tubes were used to collect plasma for cytokine measurements (PS only), and PBMCs for differential gene expression analysis via RNA-Seq and RP (Day 2 and 8), and for analysis of pertussis protein-reactive B-cell measurements (Day 150 and 510).

Table 3.

Summary of PS and MS study specimens collected from a total of 3 mL/blood draw

		Pilot Study (PS)		Main Study (MS)
Visit	Total Blood Volume	Total Volume (Plasma/PBMCs)	Total Volume (serum)	Total Volume (PBMCs)	Total Volume (serum)
01 (Baseline Day 1)	3 mL	2.5 mL (plasma in ~0.5 mL aliquots, PBMCs as ~1 million cell dry pellets	0.5 mL	2.5 mL (PBMCs as ~1 million cell dry pellets	0.5 mL
02 (Day 2 or Day 8)	3 mL	3 mL (plasma in ~0.5 mL aliquots, PBMCs as ~1 million cell dry pellets	n/a	2.5 mL (PBMCs as ~1 million cell dry pellets	0.5 mL
03 (Day 150)	3 mL	2 mL (plasma in ~0.5 mL aliquots, PBMCs in 1 mL aliquots with cell concentration <5 million cells/mL in freeze media	1 mL (~0.5 mL aliquots)	2 mL (PBMCs in 1 mL aliquots with cell concentration <5 million cells/mL in freeze media	1 mL (~0.5 mL aliquots)
04 (Day 480)	3 mL	n/a	3 mL (~0.5 mL aliquots)	n/a	3 mL (~0.5 mL aliquots)
05 (Day 510)	3 mL	2 mL (plasma in ~0.5 mL aliquots, PBMCs in 1 mL aliquots with cell concentration <5 million cells/mL in freeze media	1 mL (~0.5 mL aliquots)	2 mL (PBMCs in 1 mL aliquots with cell concentration <5 million cells/mL in freeze media	1 mL (~0.5 mL aliquots)

In the Pilot Study, for visits 01 and 02, both RNA-Seq and RP were performed from the same PBMC aliquot, where approximately 20% of the pellet was used for RNA-Seq and the remaining 80% for RP. In the Main Study, volumes and aliquots remained as in the Pilot Study, with the only difference being that plasma collections were eliminated, and where necessary (visit 02) were replaced with serum collection.

Preparation of Serum

Whole blood collected in red-top tubes (Becton Dickinson 367812) was allowed to sit at RT until it clotted (~30–120 min, but up to 8 hours, if necessary). Clotted blood was centrifuged at 1300 g for 15 min at 4 °C to separate serum away from the blood cell pellet. Serum was aliquoted into cryovials and frozen for long-term storage at −80 °C.

Preparation of plasma and PBMCs

Whole blood collected in blue-top tubes (Becton Dickinson 363080) (supplemented with 100 ug/ml CHX (Sigma-Aldrich C7698) in the MS only) was immediately processed (<1 min) to isolate plasma (PS only) and PBMCs. Plasma was obtained by centrifugation at 1500 g for 5 min at 17 °C, aliquoted into cryovials, and frozen for long term storage at −20 °C. PBMCs were isolated using SeptMate^TM-15 columns (StemCell Technologies #86415) and Lymphoprep (StemCell Technologies #07801) according to the manufacturer’s instructions but with minor modifications in the case of PBMCs intended for RP and RNA-Seq omics workflows in the MS, where all solutions were supplemented with 100 ug/ml CHX. Briefly, blood samples were diluted with an equal volume of PBS supplemented with 2% FBS (+/− CHX), transferred to SeptMate^TM-15 columns calibrated with 4.5 mL of Lymphoprep (+/− CHX), and centrifuged at 1200 g for 10 min at 17 °C. PBMCs were harvested by pouring off the cell-containing top layer into clean 15 mL tubes. Cells were washed twice by adjusting volumes to 10 mL with ice-cold PBS + 2% FBS (+/− CHX) and spinning at 300 g for 8 min at 4 °C to gently pellet cells. PBMCs were counted and assessed for purity and viability using trypan blue staining prior to flash-freezing and long-term storage in liquid nitrogen. PBMCs were stored as dry pellets in the absence of a liquid medium.

Preparation of cell lysates for RP and RNA-Seq analysis

PBMCs were removed from liquid nitrogen, placed on ice to thaw, and resuspended in 500 µl of ice-cold lysis buffer (polysome buffer (20 mM Tris pH 7.4, 150 mM NaCl, 5 mM MgCl₂, 1 mM DTT, 100 µg/ml CHX) plus 25 U/ml DNase I (Ambion AM2222) and 1% Triton-X-100). Lysates were kept on ice for 10 min, triturated by passing them through a 26G needle 10 times, and then clarified by centrifugation at 20,000 g for 10 min at 4 °C. Cleared lysates were divided in two: 80% was immediately used in RP workflows; the remaining 20% was stored at −80 °C and later prepared into RNA-Seq libraries.

Preparation of RP libraries

RP libraries were prepared as originally described¹¹ with minor modifications. The main modification was a previously described re-design of the primers and adapters used for library prep so that sequencing could be done on the Ion Proton platform²⁸. We also included two rounds of rRNA-depletion and the addition of an extra rRNA-depletion oligo to the depletion pools as previously documented²⁸.

Preparation of RNA-Seq Libraries

Total RNA was isolated using RNeasy Mini Kits (Qiagen 74104). Poly-A RNA was isolated using Oligo(dT)25 Dynabeads (Thermo, 61002) from 5 µg of total RNA (assuming recoveries of 1%) previously spiked-in with ERCC Control Mixes (Thermo, 4456740). RNA-Seq libraries were prepared from recovered mRNA using Total RNA-Seq Kits v2 (Thermo, 4479789), following the manufacturer’s instructions.

Sequencing on the Ion Proton Platform

Prior to sequencing, all RP and RNA-Seq libraries were quality controlled by Bioanalyzer using Agilent HS DNA chips (Agilent, 5067–4626). RP and RNA-Seq libraries were diluted to 13 pM and 8 pM, respectively, prior to amplification by emulsion PCR using the Ion OneTouch™ enrichment system (Thermo-Fisher 4474779) and Ion PI Template OT2 200 Kit v3 (Thermo-Fisher 4488318) according to the manufacturer’s instructions. Sequencing was done on the Ion Proton (Thermo-Fisher 4474779) with Ion PIv3 200pb kits (Thermo-Fisher 4488315). Despite being bar-coded, libraries were sequenced individually using one Ion PIv2 chip (Thermo-Fisher 4482321) per sample.

RNA-Seq and RP data processing

For RP, sequences of known human rRNAs, rRNA pseudogenes, tRNA pseudogenes, mitochondrial rRNAs (mtrRNAs), mitochondrial tRNAs (mt-tRNAs), mt-rRNA pseudogenes, and human tRNAs were obtained from the Ensembl database (Version 100, April 2020). In addition, known RNA sequences from GenBank (NR003285.3, NR003286.4, NR003287.4, NR023363.1) and two contaminant sequences provided by the laboratory (5’−ATGTACACGGAGTCGAGCTCAACCCGCAACGCGA−3’ and 5’−ATGTACACGGAGTCGACCCAACGCGA−3’) were added to this sequence collection. These sequences were then used to build a Bowtie2 index of contaminant sequences. Next, 3’ and 5’ adapter sequences were trimmed from reads using Cutadapt (Version 2.10). Reads with Phred quality score of less than 20 for the majority of bases were removed using FASTQ quality filter from the FASTX Toolkit software package (Version 0.0.14). Reads that fell outside the typical length range of ribosomal footprints (25nt to 35nt) were removed. Reads were then aligned to the index of contaminant sequences using Bowtie2 (Version 2.4.1) with its local alignment option. Reads that mapped to contaminant sequences were removed, and those that did not were output to a FASTQ file for alignment to the reference genome. Reads were aligned to human reference GRCh38 using the gapped splice-aware aligner HISAT2 (Version 2.2.0). Additionally, the reads left unmapped by HISAT2 were attempted to be mapped to human reference GRCh38 in a second pass using Bowtie2 with its local alignment option. With the exception of adapter trimming and read length filtering, the RNA-Seq data was processed as described for the RP data.

For both RP and RNA-Seq data, gene expression quantification was carried out using the featureCounts function of the Subread package (Version 2.0.1). Reads that overlapped with multiple genes or mapped to multiple locations on the reference genome were excluded. Systematic differences in sequence coverage and in counts between samples were accounted for by calculating scaling factors for each sample using the trimmed mean of m values (TMM) method as implemented in the edgeR package (Version 3.18.1). Post normalization, genes located on the X or Y chromosomes were also excluded to avoid sex-specific effects. Moderated log₂ counts per million (LCPM) for the remaining genes were determined using edgeR. LCPM was used to identify technical bias (e.g., batch effects or GC-content bias) and outlying samples (e.g., sample labeling error). Specifically, principal component analysis, non-metric multidimensional scaling, and correlation heatmaps based on LCPM were used to identify potential global outliers and systematic effects. Reverse cumulative distribution function plots of average LCPM across samples were used to identify a suitable minimum average LCPM filter cutoff to exclude lowly expressed genes. A cutoff of 1 LCPM was used to exclude lowly expressed genes.

Statistical analysis

Negative binomial models as implemented in EdgeR (Version 3.18.1) were used to determine DE genes for each assay and vaccine group after the removal of lowly expressed genes and outliers under the assumption that counts are negative binomial distributed. Models included fixed effects for subject, study visit (pre- or post-vaccination day). The statistical significance of the study visit day effect (post - pre-vaccination) was evaluated for each gene using a likelihood ratio test. To compensate for testing multiple genes, p-values were adjusted by calculating false-discovery rates (FDR) based on the Benjamini-Hochberg procedure. Genes with an FDR-value < 0.05 and a fold change of ≥1.5-fold (up or down compared to pre-vaccination) were deemed to be significant DE genes. The following number of subjects were included in each DE gene evaluation: n = 10 for RNA-Seq, Day 2, DTP; n = 9 for RNA-Seq, Day 2, DTaP; n = 9 for RNA-Seq, Day 8, DTP; n = 7 for RNA-Seq, Day 8, DTaP; n = 9 for RP, Day 2, DTP; n = 9 for RP, Day 2, DTaP; n = 7 for RP, Day 8, DTP; n = 7 for RP, Day 8, DTaP. Results for each DE gene including FDR-adjusted p-values are detailed in Supplementary Data 2–8. Translational efficiency was calculated as the difference between RP and RNA-Seq log₂ fold change for each gene. Correlation analyses between early metrics of quality obtained during wet lab processing (total PBMCs, total RNA, and monosomes) and both RP and RNA-Seq sequencing library quality were assessed via Spearman correlation using the entire dataset (PS + MS) except for missing data points, which were omitted.

Abbreviations

CHX

cycloheximide

DE

differentially expressed

DTP

diphtheria and tetanus toxoids and whole cell pertussis vaccine

DTaP

diphtheria and tetanus toxoids and acellular pertussis vaccine

DTSC

differentially transcribed

DTSL

differentially translated

MS

main study

PBMCs

peripheral blood mononuclear cells

PS

pilot study

RP

ribosome profiling

Author contributions

M.L., A.V.S., C.F.L., N.J.T., K.M.E. and C.B.C. contributed to the conception and design of the study. C.F.L., A.I.G., M.A. and R.C. executed the clinical portions of the study. M.L., D.J. and A.G.G. executed laboratory work, including all RNA-Seq and RP experiments. M.L., A.V.S., J.B.G., S.C. and C.E.G. executed bioinformatics processing of NGS data. J.B.G., S.C. and C.E.G. executed statistical analyses. M.L., A.V.S., D.J., C.F.L., J.B.G., C.E.G., L.M.H., N.J.T., K.M.E. and C.B.C. analyzed the data. M.L., A.V.S., J.B.G., C.E.G. and C.B.C. prepared the manuscript. All authors reviewed and approved the final manuscript.

Data availability

RNA-Seq and RP datasets generated and analyzed in this study have been deposited in GEO under accession numbers GSE281593 and GSE281594, respectively.

Competing interests

All authors (except CFL and KME) declare no financial or non-financial competing interests. CFL has funding from and is a consultant to HilleVax. He is also a member of the Data Safety Monitoring Board of Merck. KME has funding from NIH and CDC and is a consultant for Bionet, Dynavax and IBM. She is also a member of the Data Safety Monitoring Boards of Sanofi, X-4 Pharma, Seqirus, Moderna, Pfizer, Merck, Roche, Novavax, and Brighton Collaboration.

Supplementary information

The online version contains supplementary material available at 10.1038/s41541-025-01121-0.