Ecoimmunology in the field: Measuring multiple dimensions of immune function with minimally invasive, field‐adapted techniques

2.1. Aspects of immune function

The immune system is extraordinarily complex and a full accounting of its components and functioning is outside the scope of this review. (A number of excellent textbooks are available on this subject, such as [Parham, 2021]). However, we briefly summarize some major components, which are also listed in Table 1.

The immune system consists primarily of specialized white blood cells (leukocytes) and their secreted products, as well as some components secreted by other tissues for the purposes of signaling or immune defense. These components are typically transported by and measured in the blood but exist throughout the tissues of the body. Note, however, components in the blood and tissues actually represent the last line of defense against parasites and pathogens, as there are many other barriers to infection that include behavioral avoidance and response (the behavioral immune system), physical barriers (such as skin), and secreted immune components at points of entry into the body (the mucosal immune system).

Leukocytes can be broken down into six main subsets: neutrophils, lymphocytes, basophils, eosinophils, monocytes, and macrophages. One subset, lymphocytes, can further be categorized into three main subtypes (T cells, B cells, and natural killer cells), with several distinct classes to each of these cell types. Humoral immunity consists of immunoglobulins (antibodies) secreted by B cells, the main classes of which are IgG (the predominate type), IgM (produced in response to new infections), IgE (a response to macroparasites), and IgA (most commonly investigated in its secreted form, sIgA, as the immunoglobulin in saliva, milk, and other secretions). A wide range of signaling molecules called cytokines (e.g., interleukins, chemokines) are secreted by leukocytes and other cells in the body. Finally, effector molecules (e.g., nitric oxide), members of the complement cascade, and inflammatory proteins such as CRP are secreted by various tissues and participate in immune defense and regulation of other responses.

2.2. Categorizing aspects of immune function

Given the complexity of the immune system, there are a number of ways to categorize its components for the purposes of simplifying reasoning and analysis. One common schema divide between innate and adaptive immunity (Blackwell et al., 2010; Garcia, Blackwell, et al., 2020; McDade, Georgiev, & Kuzawa, 2016). Innate immunity does not require experience with a pathogen, while adaptive immunity is acquired through exposure. Adaptive immunity primary involves T and B cells (as well as immunoglobulins), which undergo recombination of their receptors to generate novel variants capable of recognizing new pathogens. Some approaches further divide adaptive responses into cell‐mediated and humoral responses, reflecting the differing functions and costs of leukocytes versus antibodies (McDade, 2003, 2005a). However, the term cellular immunity is also often used to encompass all cellular responses, both innate and adaptive, and the term humoral immunity sometimes refers to all factors in the serum, not just antibody (Ruoss et al., 2019).

Other approaches group aspects differently, for example a pattern of gene expression dubbed the conserved transcriptional response to adversity (CTRA; Cole, 2019) has components representing antibody response, type 1 interferon responses (viral responses), and general proinflammatory responses. Approaches also sometimes differentiate between constitutive defenses and induced defenses (Heinrich et al., 2017; Schmid‐Hempel & Ebert, 2003). Still other approaches look at responses to specific classes of infection, for example, viral versus bacterial stimuli or indicators of responses specific to parasitic infections (Blackwell et al., 2010; Harrison et al., 2019; Trumble et al., 2016).

2.3. Tradeoffs and defense portfolios

Classifying aspects of immunity is important from both a theoretical and a practical perspective. Many studies in ecoimmunology and evolutionary anthropology use life history theory as an organizing framework. Life history theory examines the trade‐offs organisms make between competing demands on time, energy, and other resources. Most life history approaches decompose these demands into the categories of growth, reproduction, and maintenance, with the maintenance category encompassing tissue repair, immune response, and other factors.

Studies examining trade‐offs between immune function and other life history demands are often most interesting in quantifying broad measures of immunity, for example, proxies for immunocompetence or measures associated with the overall cost of immunity. However, most researchers recognize that immunocompetence is not unidimensional. Different individuals or populations may be more resistant to one kind of threat than to another. Thus, a more nuanced approaches examines trade‐offs between competence or investment in different categories of immune response, or in specific immune responses to specific threats. Other approaches think of the suite of investments and individual has as a defense portfolio (Schmid‐Hempel & Ebert, 2003), or as an immunophenotype, ¹ which can be thought of as a snapshot of an individual's immune system allocations (McDade, Georgiev, & Kuzawa, 2016).

3.1. Life history allocations, energetics, and individual variation

Central to the field of ecoimmunology is the study of how the energetic demands of immune defense and activation trade‐off against other demands such as growth and reproduction. Thus, the measures that are most appropriate for these studies are measures that (1) vary between individuals or populations and (2) are indicative of the energetic demands of immunity. In some contexts, indicators of acute infection may be most relevant, for example, CRP, while in others indicators of chronic infection or immune activation might be better, for example, concentration of immunoglobulin E (Blackwell et al., 2010; McDade et al., 2008). As we might expect from the lability of these measures, these indicate investments on different timescales, and thus trade‐offs on different timescales. Urlacher et al. (2018) found that CRP, which varies on short timescales, was associated with growth over a 1‐week period, while IgG was associated with growth over a 3‐month period, and IgE with growth over much longer periods.

Leukocyte counts are also likely to represent energy investments, since these cells are costly to manufacture and operate (Straub et al., 2010). However, some cell types may be costlier than others, particularly those involved in adaptive immunity, due to the need to maintain a wide repertoire of cells and antibodies with different receptors. For example, Garcia, Blackwell, et al. (2020) found that T cells and B cells were negatively associated with child height‐for‐age. Naïve subsets of adaptive cell types, or markers of thymic output like sjTREC may further be interpreted as investment into future immune defense, since these represent the production of new cell lines in preparation for new infections.

3.2. Immunosenescence and immunocompetence

Measures related to immunosenescence may also be useful for comparing individuals and making life history inferences. Naïve cell counts and sjTREC may be useful in this regard, as well as counts of senescent T cells (Franceschi et al., 2000). Methylation assays have been used to develop a so‐called “epigenetic clock” related to immune aging (Horvath et al., 2016).

Finally, many measures are intended to compare individuals in terms of immunocompetence—the hypothetical capacity of an individual to respond to a randomly encountered pathogen. The most direct assays of immunocompetence are functional assays such as antigen stimulation assays and bacterial killing assays. However, we caution that for the most part the results of these assays have not been directly validated against real‐life morbidity and mortality from illness. It is also probably impossible to create a single, global measure of immunocompetence since individuals will differ in response to different pathogens (Schmid‐Hempel & Ebert, 2003).

In animals, measures of immune response, such as the hemolysis‐hemagglutination assay (HHA), have been used as measures of immunocompetence and related to sex, reproductive status, and other factors (Gilot‐Fromont et al., 2012; Ruoss et al., 2019). These are usually used in concert with other measures such as cell counts and antibody levels in order to provide a multifaceted assessment.

3.3. The role of inflammation in disease

In the biomedical literature measures of inflammation have gained prominence due to their associations with what tend to be considered chronic diseases of modernity such as obesity, heart disease, and type‐2 diabetes (Gurven et al., 2009, 2016; Hersoug & Linneberg, 2007; West‐Eberhard, 2019). These are often straightforward to collect given their ease of measurement in DBSs. Measures usually include CRP and inflammatory cytokines such as IL‐6 and TNF‐α. However, these measures are quite labile and care must be taken to distinguish between baseline levels and acute elevations due to infection. Inflammatory markers may need to be interpreted differently in high‐ and low‐pathogen contexts: in low‐pathogen contexts baseline elevations of these markers are indicative of nonspecific inflammation and a variety of chronic diseases, whereas in high‐pathogen contexts these are more likely to indicate recent or ongoing acute infections (McDade et al., 2012).

3.4. Old friends, hygiene, and mismatches with ancestral environments

Complementary to the study of diseases of modernity is the study of how exposure to parasites and pathogens may regulate immune responses and limit disease. These studies are informed by the old friends and hygiene (i.e., biodiversity) hypotheses and often focus on the mismatch between modern and ancestral environments (Blackwell, 2022). These studies often involve measures of the disease state in question, often linked to either excess inflammation or to misdirected immunity, for example, autoimmunity or allergy. Measures of inflammation have been noted above, and for information on measures of autoimmunity, see Cepon‐Robins (2021). Additionally, measures related to the regulation of immunity are important for this area of study, including counts of regulatory T cells through flow cytometry and measurement of cytokines characteristic of Th1 or Th2 immune responses (e.g., IL‐10, IL‐4, IFN‐γ). Parasite exposure can be indirectly inferred through immune responses to parasites, including IgE levels, eosinophil counts, and the expression of genes related to both (Table 2).

3.5. Stress, social inequality, status

Studies examining the regulation of immunity in relation to social environment have used several different approaches. Studies of chronic stress or inequality often focus on inflammatory markers, as above, as these relate to chronic diseases which are often associated with economic and social inequality. A number of studies (e.g., Dieckmann et al., 2020; Levine et al., 2017; Powell et al., 2013) have examined patterns of RNA expression associated with the “conserved transcriptional response to adversity” (Cole, 2019), which is characterized by upregulation of pro‐inflammatory responses and downregulation of type I interferon antiviral response genes and antibody synthesis genes (Table 1). Measures related to the reactivation of latent retroviruses such as Epstein–Barr virus and cytomegalovirus have also been used as proxy measures for stress related immunosuppression (Glaser et al., 1991; McDade et al., 2000). However, anti‐EBV antibody is an indirect measure and findings have been mixed (see below).

3.6. Choosing the right marker

In choosing biomarkers, often researchers will face a trade‐off between scientific and practical concerns. For example, the best measure is often not the most field friendly or cost‐effective. Thus, choosing the right measure is a matter of picking something that is a close proxy for what you really want to investigate, but which fits within logistical and financial constraints. Table 1 gives a general overview of the aspects of immunity we discuss in this review, as well as some general ideas for the theoretical concepts these might proxy. In the following sections we discuss more of the practical details for each of these measures when used in the field.

5.1. Immunoglobulins

The major subclasses of immunoglobulins (IgG, IgA, IgE, and IgM) are robust against freeze–thaw cycles and are easily measured in DBSs with only slight modifications to commercially available kits (Tanner & McDade, 2007). In general, this means that DBS can be used to measure not only total immunoglobulin levels, but also specific antibodies against things such as Epstein–Barr virus capsid antigen (EBV), which is commonly measured as an index of stress‐related reactivation of latent virus and possibly T‐cell suppression (Glaser et al., 1991; McDade et al., 2000). However, anti‐EBV antibody may be a more complicated measure than is typically appreciated (e.g., Cacioppo et al., 2002) and many studies, particularly those including other biomarkers, do not find associations between EBV and other stress‐related measures (e.g., Panter‐Brick et al., 2020; Rudzik et al., 2014) raising the possibly of publication bias for positive findings. Specific antibodies against a number of pathogens have been measured in DBS, for example, Toxoplasma gondii (Lebech & Petersen, 1995), malaria (Corran et al., 2008), and measles (Colson et al., 2015). In principle antibodies should have similar properties in DBSs regardless of target, so adapting ELISAs devised for measuring immunoglobulins in serum for use with DBSs is often possible. DBSs can also be used to measures immunoglobin subtypes, that is, sub‐classes of IgG (Andersen et al., 2014).

5.2. C‐reactive protein

CRP is a favorite measure in human eco‐immunology given its stability and the availability of established ELISA protocols (i.e., Brindle et al., 2010; McDade et al., 2004). However, these protocols were published a number of years ago, and some of the antibodies used by them are no longer available commercially. The capture antibody used by Brindle et al., 2010 is currently available from Meridian (clone C5 anti‐CRP MAb, M86005M), however, the detection antibody is no longer listed. An alternate detection antibody (M01239M) is given as pairable with M86005M and will likely work similarly to the original antibody from the protocol. However, we would recommend validating this or any other replacement antibody pair before using them to analyze DBS since it is always possible for confirmational changes, elution buffer differences, or matrix effects to make a DBS assay behave differently from serum assays. DBS measures of CRP have been used extensively in a number of papers (e.g., Blackwell et al., 2010; Urlacher et al., 2018).

5.3. Cytokines

Cytokines are small molecules used for signaling between cells of the immune system. Because each cytokine has its own molecular structure, cytokines differ significantly in their stability at room temperature, loss due to freeze–thaw cycles, and recovery from DBS. One study found that interleukin (IL)‐6 and IL‐10 are stable in serum through freeze–thaws, but levels of IL‐4, IL‐13, IL‐15, IL‐17, tumor necrosis factor (TNF)α, interferon (IFN)γ and CXCL8 all changed with freeze–thaw cycles (De Jager et al., 2009). Skogstrand et al. (2005) used a Luminex xMAP assay to measure 25 cytokines and other biomarkers in DBS. The results show that in principle many cytokines can be measured in DBS and that cytokines in DBS are stable when kept frozen at −80°C, but the study did not compare values in plasma to DBS or assess the effect of freeze–thaw cycles. A second study (Skogstrand et al., 2008) assessed stability at room temperature, 35 and 4°C and found that a few cytokines, such as IL‐8, IL‐10, and TNF‐α appear to be relatively stable, even up to 30 days at room temperature, but the majority were not. Another concern is that the variability in cytokine levels due to differences in collection or storage is not consistently directional (i.e., values may go up or down). There may be a variety of reasons for this including differential degradation of cytokine molecules and release of cytokines from intracellular locations, binding molecules, or the DBS matrix. The study did find that one advantage of DBSs is that cytokine levels change rapidly in venous blood that is kept at room temperature or refrigerated, as the living cells in the blood become activated by the novel environment (Keustermans et al., 2013; Skogstrand et al., 2008). Since DBS are usually collected and dried immediately this is less of an issue.

More recently McDade et al. (2021) used a multiplex electrochemiluminescent immunoassay to simultaneously measure IL‐6, IL‐10, IL‐8, and TNF‐α in matched DBSs and plasma. They obtained good correspondence for IL‐6, IL‐10, and TNF‐α, but poor correspondence for IL‐8. They speculate this may be due to IL‐8 stored in red blood cells which is released on lysing, and which would not be present in plasma. An ELISA protocol for just IL‐6 has also been validated, with good correlation between DBS and serum (Miller & McDade, 2012).

In conclusion, DBSs seem to be reliable for the measurement of higher concentration, higher stability cytokines such as IL‐6, IL‐10, and TNF‐α. However, caution may be needed when measuring other cytokines in DBS, particularly those with lower concentrations or shorter half‐lives. An alternative may be to measure cytokine related gene (mRNA) expression (see below).

5.4. CD4 lymphocytes

For many assays where serum equivalents are desired the inclusion of red and white blood cells may be a nuisance. However, the fact that DBS contain whole blood presents an opportunity as well. Mwaba et al. (2003) tested an ELISA kit for measured CD4 protein in DBS and showed a reasonable correspondence between estimated CD4 counts from this method, and counts obtained by flow cytometry. Likely estimation of other cell types using ELISAs and DBS is possible, though we are not aware of publications validating these methods.

5.5. Single versus multiplex platforms

ELISAs can be used to measure many single analytes in DBS. ELISAs have the advantage of using relatively common equipment (an ELISA plate reader, at the minimum) and having a lower cost than multiplex assays. DBS kits typically cost around $300 to $700 for a 96‐well plate, good for running about 35 samples in duplicate, after accounting for standards and controls. Multiplex kits may be more than five times as expensive as ELISA kits. However, when measuring many analytes multiplex assays become more efficient, in terms of labor, sample used, and cost. However, multiplex assays also sometimes sacrifice sensitivity, particularly for lower concentration analytes, since they must compromise on assay dilutions that are effective for the many things being measured, which may differ in concentration by orders of magnitude. Common multiplex platforms include the MagPix xMAP instrument (Luminex Corp) which uses magnetic beads with bound antibodies, instruments from Meso Scale Diagnostics which use electrochemiluminescence, and flow cytometry‐based bead assays such as the Luminex 200 System and BD Cytometric Bead Array kits (Becton, Dickinson and Company) which can be used with any number of flow cytometers. However, only a handful of multiplex assays have been validated with DBS (McDade et al., 2021; Skogstrand et al., 2008), so other assays will require protocol development for use.

5.6. Complications for DBS samples

Although DBSs are convenient, recent studies have shown that there are a number of ways to introduce errors into blood spot analyses. In addition to the aforementioned effects of freeze–thaws and storage temperatures, humidity may also impact certain analytes measured in DBS. DBSs differ from serum measures due to the inclusion of white and red blood cells, which lyse, and the filter paper matrix itself which may create assay nonlinearities. At a minimum, the analysis of whole blood, rather than serum, will dilute concentrations by a factor of about 2 to 2.5, equivalent to the proportional volume of red blood cells in blood (hematocrit). Because of this and other factors, many protocols will include conversion equations for estimated serum equivalent values from DBS results. However, in the absence of data on serum equivalence it is problematic to compare DBS and serum values, so caution is needed. Most commercial kits are developed for analysis of serum and thus it is also inadvisable to perform DBS assays without validating the protocol against serum, given the issues noted above. Considerations and requirements for protocol validation has been described in a number of papers (e.g., McDade et al., 2007; Sultana et al., 2022). Additionally, the use of small blood spots, overlapping drops, smeared drops or the edges of blood spots can create issues, as these all affect the concentration per unit of filter paper and may also affect matrix effects and elution (Crimmins et al., 2020). Finally, because most commercial assay kits are not designed for DBS, maintaining consistency in the kit manufacturer used across a study sample is necessary. This means that if a researcher shifts to a different antibody or assay manufacturer (e.g., if one brand goes off the market) the burden is on them to do a new validation. To ensure continuity and comparability of biomarker data across a study sample, researchers should purchase enough antibody or kits for their entire study, while also needing to be aware of expiration dates. Planning out the materials needed and timeline to conduct laboratory analyses are as important as planning the data collection itself.

7.1. Isolating RNA from finger‐prick blood samples

Typical RNA‐seq blood collection protocols, such as the PAXgene Blood RNA extraction system (Qiagen, Inc) call for the collection of 2.5 ml venous blood into PAXgene vacutainers. PAXgene documents report that 2.5 ml blood yields ≥3 μg RNA for ≥95% of samples (mean ~ 8 μg) (PreAnalytiX., 2020). However, RNA sequencing requires only 50 to 100 ng RNA for evaluation of transcripts, and only 500 to 1000 ng for de novo transcriptome assembly, suggesting that 2.5 ml produces roughly 60× the minimum RNA yield needed for 3′ sequencing and that less blood could be collected.

Krawiec et al. (2009) developed a protocol for isolation of RNA from small samples of blood, to be used with mice who cannot spare 2.5 ml. Although not yet validated with human blood, for mice 50 μl of blood yielded 2300 ng of RNA, more than sufficient for sequencing. ADB also successfully collected and extracted RNA from great‐tailed grackles, with yields from 50 μl blood averaging 3152 ng of RNA. While birds have nucleated red blood cells, mice and humans have roughly similar white blood cell counts, suggesting results for humans may be similar to mice. While reported RNA yields from PAXgene documents suggest possible yields as low as 60 ng from 50 μl blood, other references suggest typical RNA levels for humans are actually higher than this, with a mean of 15 μg/ml (range 7–23 μg/ml) (Chomczynski et al., 2016), which would yield 750 ng from 50 μl. Lower yields in some PAXgene protocols may in fact be due to a ceiling effect from the RNA spin columns used, with higher recovery possible for lower volumes of blood (Krawiec et al., 2009).

Together, these results suggest that blood collected via finger‐prick should produce usable yields for RNAseq in humans. We plan to validate this procedure in the near future, but prior to that we recommend verifying RNA yields before using this protocol. An example finger‐prick collection protocol, modified from the protocols used for mice and birds, might read as follows: (1) Aliquot PAXgene reagent into smaller tubes at a ratio of 2.76 reagent/blood (e.g., 138 μl PAXgene/50 μl blood); (2) Pierce the fingertip with a sterile lancet and collect blood into a finger‐prick blood collection tube; (3) pipette 50 μl of blood from the collection tube into the PAXgene aliquot; (4) store at room temperature for at least 2 h (no longer than 72 hrs), then freeze. PAXgene aliquots may be frozen at −20C, −80C, or placed in liquid nitrogen. PAXgene‐preserved samples are also stable across at least three freeze–thaws, making them ideal for easy transport. Extraction and analysis should follow standard protocols.

7.2. Isolated RNA from DBS

Multiple authors have successfully sequenced RNA from DBS (McDade, Ross, et al., 2016; Reust et al., 2018). In these studies, relative transcript reads derived from DBS were highly correlated with PAXgene derived values and yielded largely similar associations with predictors, though with some discrepancies. Most of these discrepancies may be due to the small amount of RNA recovered from DBS. For example, McDade, Ross, et al. (2016) did not quantify RNA yields since these were expected to be too low for accurate quantification. The authors note that due to the increased noise from a small RNA sample, a sample size ~5% greater than in PAXgene studies might be required to maintain statistical power. These studies require a large number of blood spots, in the range of two to four complete blood spots (McDade, Ross, et al., 2016) or 8 × 3 mm² punches (Reust et al., 2018). Overall, DBS are likely to be simpler to store and transport, but small liquid aliquots with PAXgene are likely to yield better results.

7.3. Sample preparation and sequencing for RNA‐seq

Sequencing is typically done either by reading from the 3′ end of each cDNA transcript, which only partially reads the transcript, or by fragmenting the mRNA into smaller pieces for whole transcript sequencing before reverse transcribing into cDNA. Each approach has its own advantages and disadvantages, which we will briefly detail in the next sections (for a more thorough comparison see Ma et al., 2019). For both methods, after collecting the sample, RNA is isolated and purified, usually with a kit such as the RNeasy kit (Qiagen).

Due to the high volume of RNA in red blood cells, globin‐based reads can consequently comprise a large proportion (>80%) of sequencing reads. Thus, when preparing whole blood for 3′ RNA sequencing, it may be advisable to include a globin depletion step in the protocol. However, globin reads can also be removed from analyses after sequencing, and depletion may introduce biases to other reads (Harrington et al., 2020). Consequently, there are costs and benefits to globin depletion with the primary benefit being a reduction in sequencing depth and expense, and the cost being an increase in protocol complexity and read bias. Globin reads may also sometimes be of interest.

For whole transcript sequencing an enrichment step is commonly added after RNA isolation and purification to remove ribosomal RNA repeats which would otherwise swamp the sequencing and be uninformative for many purposes. Some protocols also enrich for RNAs with poly‐adenine tails, which helps to eliminate immature RNAs which may contain introns that have not yet been spliced out. These steps are not necessary for 3′ sequencing, as 3′ sequencing only targets the poly‐adenine tails at the 3′ ends of mature mRNAs. After enrichment the RNA is converted to complementary double‐stranded DNA (cDNA) via a reverse transcriptase enzyme and prepared for sequencing.

Whole transcript sequencing generates more cDNA fragments for longer genes, and consequentially more reads for these genes, relative to shorter genes. In contrast 3′ sequencing sequences from the 3′ end of the mRNA. Since mRNA is not fragmented, gene length does not influence sequencing reads and the number of reads is proportional to the original mRNA transcripts. Ma et al. (2019) compared the two methods, and as expected found that 3′ sequenced performed better for shorter transcripts (particularly <1000 bp) because they receive equal reads. Many genes of interest for quantifying immune function are small genes, for example, cytokines (e.g., the mRNA for IL‐10 is approximately 500 bases). 3′ can also be substantially cheaper than whole transcript sequencing (Sholder et al., 2020). There are, however, advantages of whole transcript sequencing. First among these is that whole transcript sequencing allows for the discovery and counting of alternative splice variants and noncoding RNAs. Depending on the study aims this may or may not be relevant.

Another important consideration—which depends on the sequencing technique and the question—is sequencing depth. For RNA‐seq, sequencing depth refers to the total number of reads obtained in a sequencing run. For example, if a gene of interest is lowly expressed, it may appear in only one out of a million transcripts. If a sequencing depth of 5 million is used, on average there will be 5 reads, but with a 95% confidence interval ranging from 1 to 10 reads, leading to an estimated relative expression of between 2 and 20 per 10⁷ reads. Increasing sequencing depth to 20 million narrows this range to 6 to 14.5 per 10⁷ reads. A main disadvantage of higher sequencing depth is increased cost.

The required read depth also depends on whether one is doing whole transcriptome sequencing or expression profiling sequencing (i.e., 3′ sequencing). Because only a single fragment is generated for a transcript with 3′ sequencing, this method attains a similar coverage as whole transcriptome sequencing at a fraction of the depth, and there are diminishing returns for increasing depth past ~10 million reads for 3′ sequencing (Moll et al., 2014). Thus, depending on the study design and budget a researcher might have to choose between sequencing many samples at low depth or few at high depth. One analysis of this question found that sequencing more replicates at lower depth provided more power than sequencing fewer replicates at higher depth (Liu et al., 2014). Thus, if the object of a study is to compare two groups of individuals, a lower sequencing depth and a greater sample might be preferred. If, however, it is essential to quantify the gene expression for a particular individual (rather than a group of individuals) a greater sequencing depth might be more important.

7.4. Post‐sequencing processing and analysis workflow

After samples are sequenced (i.e., “read”) raw read data go through multiple processing steps before they are converted to raw count data, and ready for downstream analysis. Post‐sequencing processing steps include quality assessment, transcript assembly, alignment, analysis of raw counts, and annotation. For the trainee interested in integrating RNA‐seq into their project, an important consideration is that each of these steps relies on bioinformatic tools and pipelines, which require a level of fluency and comfort with programming languages (e.g., R, Python) and statistical software and packages. As an example, here is an abbreviated description of an RNA‐seq pipeline for samples sequenced on an Illumina sequencing platform (full pipeline available at https://github.com/heinin):

1. Raw RNA‐seq data exist as FASTQ files, which contain the short sequence reads and quality scores. The quality of the FASTQ files is assessed with a tool called FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/)

2. The raw sequence reads often contain adapter sequences that were used to create the Illumina sequencing libraries. These adapters and low‐quality bases at the ends of each read are trimmed using Trimmomatic (https://anaconda.org/bioconda/trimmomatic). Trimming improves read mapping and affects all downstream analyses.

3. After trimming, the FASTQ quality is assessed again, until quality checks are passed, after which we continue with read mapping. The trimmed reads are mapped to the human reference genome using HISAT2 (https://ccb.jhu.edu/software/hisat2/index.shtml) (this is one of many tools for DNA and RNAseq alignment). The result is one SAM file (Sequence Alignment Map) for each sample. The SAM file is converted to a binary version using SAMtools (http://samtools.sourceforge.net/; https://samtools.github.io/hts-specs/SAMv1.pdf), and these BAM files are sorted based on genomic coordinates (also with SAMtools). Most tools that take BAMs as input files require them to be coordinate sorted.

4. Next, the expression levels of genes are obtained by quantifying the read counts mapping to each gene. One commonly used tool is called featureCounts (http://bioinf.wehi.edu.au/featureCounts/). For each sample, the output is a text file containing the read counts in each gene. These files can then be combined to create a count table with samples in columns and genes in rows, and that count table is used in downstream analyses (e.g., differential expression).

For researchers new to RNAseq, the learning curve for bioinformatic processing can be steep but may be worthwhile depending on your approach to a research question. There are a number of resources that describe workflows, pipelines, and workflow managers, including https://bioinformatics.uconn.edu/resources-and-events/tutorials-2/rna-seq-tutorial-with-reference-genome/ and https://rnaseq.uoregon.edu/. The Biostar Handbook is also an invaluable guide to bioinformatics that is updated multiple times a year and has topic‐specific (https://www.biostarhandbook.com/). The learning curve can also be buffered by teaming up with other researchers that are experts in the use of a particular method(s), as they can guide you through parts of the process and direct you to relevant resources. It is also possible to have a majority of these steps done “behind the scenes” (e.g., among the services offered by a Core Lab), though we suggest familiarizing yourself with the overall process in order to better understand the data you are working with, including its limitations.

7.5. Statistical analysis

The statistical approach for downstream analysis of count data will depend on the specific goals and research question being asked and is thus outside the scope of this review. Here we briefly discuss the process of count normalization and its importance for the interpretation of RNA‐seq data in field studies. In both whole transcript sequencing and 3′ RNA‐seq reads must be normalized for between sample variation in library size, that is, the total of all reads for a sample. For whole transcript sequencing counts are also normalized for gene length because gene length affects the number of cDNA fragments produced and thus the number of reads mapped to a particular gene; this is not usually necessary for 3′ sequencing because only one fragment is generated per transcript (Moll et al., 2014). Because library size is more likely to reflect variation in sequencing depth due to methodological stochasticity, rather than real‐world variation in total mRNA expression, counts can be normalized in a number of ways which reflect different assumptions (Evans et al., 2018).

One approach is to normalize to total library size, which converts values into counts per million reads (CPM). When using whole blood rather than isolated PBMCs normalized values will reflect the total proportion of all transcripts in all blood cells including red blood cells which contain RNA from a large number of genes (Kabanova et al., 2009). If red blood cell globin transcripts are depleted (either before sequencing or by removing these reads after sequencing) the interpretation of CPMs will change to represent the proportion of nonglobin reads. However, in some circumstances, normalization by read count may lead to faulty conclusions if the total mRNA differs between samples for biological reasons; for example, if some prevalent genes are upregulated in one sample this could instead be interpreted as a downregulation of the remaining genes, since these transcript proportions would decrease. To avoid this issue, a number of other normalization procedures are possible, including normalizing or controlling for sample RNA concentration while also normalizing for library size. Other approaches include normalization for housekeeping genes (genes thought to be expressed at relatively constant levels across cells), normalization based on distribution (quantile normalization), normalization based on cell‐type specific genes, and normalization based on cell counts measured independently in the sample (Dillies et al., 2013; Eisenberg & Levanon, 2013; Evans et al., 2018).

The choice of normalization procedure should depend on: (1) the initial sample (e.g., whole blood), (2) whether the RNA was enriched before sequencing, (3) whether the desired end product is mRNA/cell, mRNA/leukocyte, mRNA/total transcript, mRNA per volume of blood or some other value. These values will have different interpretations, for example, mRNA/leukocyte may be most appropriate for inferring leukocyte differentials, while mRNA/volume of blood may speak more to cell counts or systemic expression levels for the whole organism. Like most statistical analyses, there are numerous ways to approach normalization and we strongly advice investigating the advantages and constraints of each method before deciding which method is most appropriate for answering a specific question. There are a number of studies that describe the details of each method and provide useful insights on appropriate choice of RNA‐seq normalization and analysis methods (Abrams et al., 2019; Evans et al., 2018; Zyprych‐Walczak et al., 2015).

In sum, the decision to incorporate RNA‐seq into one's research should be driven by a triangulation between the question, the cost, and the learning curve. All specific methods and analyses will follow from the question.

8.1. Thymic output

Like RNA, it is also possible to extract DNA from DBS for analysis. In addition to genotyping, there are several assays of interest when measuring immune function. When T cells mature, they undergo a process wherein the DNA coding for the T‐cell receptor is randomly rearranged to create a unique receptor for each T cell lineage. Leftover DNA from this process is excised from the T cell receptor arrangement and circularized into DNA molecules called signal joint TCR excision circles (sjTRECs). New T cells leave the thymus with one or two sjTREC molecules, and as they mature and divide these become distributed among the daughter cells (Lorenzi et al., 2008). As such, sjTREC levels in the blood directly correspond to thymic output, that is, production of new naïve T cell lines by the thymus. sjTREC levels correlate with naïve cell counts measured in the blood, but imperfectly so, as naïve cells may also replicate in the periphery (though divergences are themselves informative) (den Braber et al., 2012). Thymic output declines with age in a consistent enough fashion that sjTREC has been used to estimate age from bloodstains (Ou et al., 2011; Ou et al., 2012). However, thymic output can also be temporarily upregulated during infections (Permar et al., 2003) and is influenced by various immunological (Cho et al., 2017) and physiological (Tarcic et al., 1998) conditions. These divergences from general age patterns are probably of the most interest in eco‐immunological studies, as they are useful measures of variation in immune investment. However, this means that analyses using sjTREC need to consider how to parse out infection or exposure‐related variation from effects due to initial thymic size and typical age‐related decline.

Quantitative PCR (qPCR) is used to quantify sjTREC DNA, as well as levels of a housekeeping gene for normalization. To quantify sjTREC in DBS, DNA is first extracted with a kit such as the QIAamp 96 DNA Blood Kit or Gentra Puregene Tissue Kit (Qiagen); Qiagen also provides an adapted protocol specifically for DBS. Extracted DNA samples are analyzed for quantity and quality with a spectrometer (e.g., NanoDrop, Thermo Scientific.). qPCR is then performed on an instrument such as the Bio‐Rad CFX96 (Bio‐Rad Laboratories) using appropriate dyes and primers for sjTREC and any normalization genes (see Lang et al., 2011 for primer sequences).

Lang et al. (2011) used ALB2 for initial normalization, to obtain a ratio of sjTREC per total nucleated cells. They also quantified the unrearranged configuration of the VD junction region (VD‐J) of the TCR‐β, which is the version present in non‐T cells. By subtracting the estimate of non‐T cells from total cells they were able to obtain an estimate of T cells, and ultimately calculate an sjTREC/T cell ratio. As with the RNA‐seq assays above, the choice of normalization alters interpretation. If peripheral cells replicate, for example in response to infection, then the sjTREC/T cell ratio will go down, even though thymic output might actually increase. Thus, some advocate for measuring sjTREC/ml (Lorenzi et al., 2008). An alternative approach when blood volume is not precise is to measure sjTREC/cell by PCR and use a point‐of‐care device to measure total leukocyte count. Multiplying sjTREC/cell times cell count will yield sjTREC/ml.

8.2. Lymphocyte cell subsets

As should be apparent from the sjTREC protocol above, qPCR can also be used to estimate the proportion of non‐T cells (and by extension T cells) in a sample. Though we know of no validations, a similar approach is likely possible for estimating B cells using primers for the non‐rearranged heavy and light immunoglobulin chains. Assessment of rearrangements is currently described for monitoring of B cell disorders (Cho et al., 2017). Measurement of non‐rearranged sequenced should be considerably simpler.

8.3. Methylation

DNA methylation can be measured in minimally invasive blood samples, like DBS, but can also be measured in buccal swabs or saliva (Non & Thayer, 2015). However, because buccal swab and saliva samples contain epithelial cells as well as white blood cells, blood may be the best sample for measuring immune function directly. Methylation is stable in DBS for considerable lengths of time, even at room temperature (Sasaki et al., 2020), and protocols for measuring methylation in DBS are similar to those for fresh blood (Walker et al., 2019). DNA is first extracted with a DNA purification kit before analysis. Methylation analyses may look at global methylation, may compare groups or individuals for differentially methylated genes, or may look at methylation of specific genes. The choice of assay method will differ depending on which of these approaches a study takes; methylation assays may be done through sequencing, mass spectrometry, the reading of microarrays, or through qPCR (Bock et al., 2016; Kurdyukov & Bullock, 2016; Non & Thayer, 2015). We refer the reader to the citations referenced here for more information.

For the purposes of eco‐immunology, methylation may be useful for examining particular immune related genes, particularly with regard to chronic effects due to lifestyle (Ligthart et al., 2016; McDade et al., 2019; Nakajima et al., 2010). Methylation is part of the regulatory pathway for gene expression, and may be important, for example, in TH1/TH2 balance (Brand et al., 2012). Patterns of methylation have also been found to associate with levels of CRP (Sun et al., 2013). However, methylation is not the only factor in gene expression as transcription factors and other regulators make many short‐term adjustments (Makar & Wilson, 2004). Thus, for measuring expression at a particular timepoint the measurement of RNA expression or protein product level is probably complementary. Similar to RNA‐seq data, methylation results can also be used to infer underlying cell subset proportions and thus may be another useful alternative to flow cytometry (Titus et al., 2017).

9.1. Antigen stimulation

Antigen stimulation assays involve mixing blood with an antigen or mitogen and incubating in medium, followed by measurement of the response to the treatment. Common antigens include lipopolysaccharide (LPS), a cell wall component of gram‐negative bacteria that initiates B‐cell division and activation of antigen presenting cells, and phytohemagglutinin (PHA) a mitogen that activates T cells; other studies have also used antigens ranging from fragments of helminths (van Riet et al., 2009) to H1N1 flu vaccine (Schneider‐Crease et al., 2021). Typical protocols call for incubation in a CO₂ enriched incubator at 37°C. However, May, van Bodegom, et al. (2009) developed a field‐friendly method for whole blood stimulation that does not require the CO₂ tank. They created a CO₂ enriched environment by sealing the samples in a plastic container containing water and a burning candle. The candle depletes the oxygen and enriches the CO₂ in the contained before going out (Figure 1B). This procedure has been successfully carried out for a number of studies (May, van Den Biggelaar, et al., 2009; Schneider‐Crease et al., 2021; Trumble et al., 2016), however, because it still requires the ability to maintain samples at 37°C—usually in an incubator—it is not the most field‐friendly protocol. The incubation time ranges from 24 to 72 h, depending on whether the assay is designed to capture initial, innate responses, or slower adaptive immune responses.

Antigen stimulations are usually done with 100 μl of whole blood per sample, so it is feasible to conduct them with finger‐prick collections. However, the blood must be fresh (ideally stimulation is done within an hour of collection) and cannot be frozen. After stimulation, the supernatant is collected and frozen (which necessitates a freezer or liquid nitrogen tank). The supernatant is typically transported frozen and assayed for cytokines using ELISA or multiplex assays. Cytokine levels in stimulated supernatants are usually much higher than they are in circulating blood samples. In principle, therefore, assays could be developed which avoid freezing, for example, by spotting stimulated samples on filter paper, or by preserving stimulated cells with PAXgene reagent and performing RNA‐seq instead of cytokine measurement (e.g., Meitern et al., 2014; Pulido‐Salgado et al., 2018).

One important caveat about antigen stimulation assays is to remember that the assays measure the response to stimulation, but not how effective that response is in fighting the infection. Cytokines are signaling molecules and measuring them is equivalent to measuring how loud the immune system yells when it feels threatened. Indeed, in some cases strong inflammatory responses may be detrimental or represent a failure of specific responses, so care in interpretation is needed.

9.2. Bacterial killing assays

Bacterial killing assays involve in vitro introduction of microbes to blood, plasma, or saliva in a dish or well. Microbes such as Candida albicans, Escherichia coli, and Staphylococcus aureus can be obtained as lyophilized pellets and reconstituted as needed. Liebl and Martin II (2009) report on a procedure designed for small birds that uses only 1.5 μl of plasma or blood. The sample is diluted in medium and mixed with the microbe solution, then incubated for 30 min to 1 h at a temperature optimal for the particular microbe used. The microbes are then diluted in nutrient medium and incubated for an additional 12 h. The bacterial concentration is then measured with a small Nanodrop Spectrophotometer (Thermo Scientific). As an alternate measurement, bacteria can also be spread on plates and the colonies counted (Millet et al., 2007). Since bacterial killing largely depends on the action of whole cells which are likely to be lysed or stressed by freezing it is unlikely to produce useful results from frozen blood, though we are not aware of protocols that have directly evaluated this.

9.3. Hemolysis‐hemagglutination assay

The HHA is an assay utilized by field biologists to measure complement and natural antibody activity via the ability of plasma to lyse or agglutinate foreign red blood cells. Natural antibodies are antibodies produced without previous exposure to an antigen and which are genetically encoded rather than requiring gene rearrangement to produce, while the complement system consists of a number of small molecules that bind to invaders and trigger further responses to clear infections. The HHA assay is similar to the hemolytic complement assay (HCA) developed in the 1960s to measure complement activity in humans and aid in diagnosing complement related disorders (Kirschfink & Mollnes, 2003; Mayer, 1961; Morgan, 2000; Rapp & Borsos, 1970). However, it is simpler to perform in the field with small volumes and results that are read by eye or with a scanner or camera, rather than a spectrophotometer. The HCA assay uses antibody‐sensitized sheep erythrocytes to evaluate the classical complement pathway and rabbit erythrocytes to assesses the alternative complement pathway; the HHA does not distinguish lysis by pathway, however it does assess agglutination which requires natural antibodies.

The HHA was first developed in birds, in which bird plasma was treated with a sample of rabbit blood (Matson et al., 2005), but has since been used for mammals, with chicken blood replacing rabbit blood for the assay (Gilot‐Fromont et al., 2012; Heinrich et al., 2017; Ruoss et al., 2019). To the best of our knowledge the HHA has not yet been used to assess human immune response, but it is use in other mammals suggests it should produce results. As little as 10 μl of plasma is serially diluted in PBS into 11 wells in a 96‐well plate. A 1% suspension of chicken red blood cells is added. Plates are incubated at 37°C for 110 min and then photographed or scanned to measure agglutination due to natural antibodies and then again after 70 more minutes to measure lysis, which is caused by complement. The scans are scored manually based on appearance. As with the other response assays, this assay requires an incubator or other method of maintaining a controlled temperature. Most assays use commercially available red blood cells, which may be difficult to obtain and maintain in the field. However, it is likely possible to use fresh erythrocytes which have been washed and resuspended in an appropriate buffer. Given the relatively complexity of performing the assay in the field, it is possible to use frozen plasma for the assay. Most protocols caution against freeze–thaw cycles, however evaluations of freeze‐thaw cycles on complement activity show that these have little effect (Hegemann et al., 2017; O'Shaughnessy et al., 2012; Yang et al., 2015).

10.1. Technical variation (batch effects)

For a number of assays, samples are analyzed in a series of “batches.” For example, ELISA and microarray assays are typically run on 96‐well plates that hold anywhere between ~36 and 96 samples, depending on protocols; qPCR assays are run on a 384‐well plate, and other genomic assays that use NGS technologies (e.g., RNA‐seq) are run on flowcells that have between one and eight lanes, and can analyze up to 400 samples. For ELISAs, metrics like the coefficient of variation (CV) are typically used describe variation within and between plates. The CV is the ratio of the standard deviation to the mean (typically multiplied by 100 to report as a percentage). Within each plate the CV is assessed for each sample, run in duplicate, to identify idiosyncrasies, many of which are due to pipetting errors or samples that are on the high or low end of the assay range. Samples with high CVs (>10%) are usually run again, sometimes with a different dilution. “High” and “low” controls, which have a known quantity of the analyte being measured, are included in most assay kits and should be run on each plate. CVs for the control duplicates are used to report variation across plates, due to minor differences in handling and processing. This noise is known as “technical variation.” It can be minimized through careful laboratory technique, but some variation in usually unavoidable. In some cases, it may be useful to statistically control for batch in the data analysis step, however this is only feasible if batch is not correlated with another variable of interest. For example, if using two plates to compare samples from two villages, it would be better to run each plate with a mix of samples from both villages, rather than one plate for each village, in order to minimize confounding of village and batch.

Unfortunately, the practice of including something like a “control” in next‐gen sequencing assays has yet to be adopted, and it remains unclear how best to accomplish this given the way NGS operates. There can be a significant amount of variation between batches due to technical factors in RNA‐seq and other NGS data. There are some statistical packages (e.g., Combat‐seq, variancePartition [Hoffman & Schadt, 2016; Zhang et al., 2020]) that quantify sources of variation that may be used to adjust the substantive NGS data for better comparability. Still, it is important to try to sequence all samples for a specific project together to limit such batch effects.

10.2. Age, sex, and reproductive status

Many immunological values vary with age, sex, and reproductive status. Age‐related patterns are apparent in many measures due to development and late‐life senescence (e.g., Blackwell et al., 2011, 2016; McDade, 2005b). Studies should evaluate whether it is appropriate to use age‐standardized z‐scores or to otherwise control for age. Note, however, that many age‐related changes are not strictly linear, so linear terms in regression models may not be sufficient. Differences by sex and reproductive status (e.g., menstrual cycle, pregnancy) are also apparent (Aghaeepour et al., 2017; Hové et al., 2020; Klein & Flanagan, 2016; Natri et al., 2019; Wander et al., 2008) and so studies should take these into account as part of study design and analysis.

10.3. Diurnal pattens

Many papers studying immune function do not report or control for time of blood collection. However, many components of the immune system vary diurnally, with many being lower in the morning and higher in the afternoon, opposite the diurnal rhythm of cortisol (Ackermann et al., 2012; Labrecque & Cermakian, 2015; Petrovsky & Harrison, 1998). Diurnal regulation of immunity may be relevant for many research questions (e.g., Garcia, Blackwell, et al., 2020), but even when it is not the focus, time of sample collection is an important variable that should be captured and accounted for in study designs.

10.4. Chronic versus acute elevations

Some immunological measures are likely to be much more variable within an individual. McDade et al. (2012) analyzed repeat samples of CRP in an Amazonian population. Even though 35% had elevated CRP at one timepoint, no participant had more than one elevated measure, indicating that these were acute elevations rather than chronic elevations. Blackwell et al. (2016) calculated intra‐class correlations for 19 immune measures based on repeat samples. Some measures, such as IgE concentration, CD4 counts, and CD8 counts were relatively stable across repeat measures, while others such as neutrophil count and natural killer cell count appeared more labile. The stability of a measure should also be considered in study design (repeat vs. single measures) and interpretation (see Table 1).

10.5. Capillary versus venous blood

For the most part, measures from capillary and venous blood correlate strongly. However, some differences have been noted (Sitoe et al., 2011; Srisala et al., 2019). Study designs should consider whether this is relevant to a particular study design and may wish to validate against serum measures.

10.6. Measures in urine or feces

Blood is the optimal tissue in which to measures systemic immunity, since blood distributes white blood cells, cytokines, and other factors throughout the body. However, if blood is unavailable, what, if any, immunological measures are available? The short answer is: it is difficult to measure systemic immunity in other fluids. The long answer is that while many things can be measured, care must be taken in interpretation. This is because markers in urine or feces are much more indicative of localized bladder or renal (in the case of urine) or gut (in the case of feces) infections than systemic immune allocations. Higham et al. (2015) compared serum measures with several fecal and urinary measures. They found that while serum and urinary neopterin were correlated (r = 0.664), fecal concentrations were not correlated with serum. Serum CRP was not correlated with urinary or fecal CRP levels, but there was a short‐term rise in fecal and urinary CRP following a surgery. Serum and urinary haptoglobin were also not correlated, and haptoglobin could not be measured in feces.

10.7. Measures in saliva and breastmilk

Many studies have measured secretory IgA in saliva, breastmilk, and other secreted substances (e.g., Hodges‐Simeon et al., 2017; Miller & McConnell, 2011; Van Anders, 2010). However, this is probably best interpreted as a marker of mucosal immunity, a complex system in its own right. Caution should be used when interpreting whether secretary IgA is indicative of overall immune status since mucosal immunity is not necessarily correlated with systemic immunity, and can be indicative of better health or worse health depending on the circumstances (Drummond & Hewson‐Bower, 1997; Gornowicz et al., 2014; Koss et al., 2016; Nakamura et al., 2006; Reid et al., 2001; Thaweboon et al., 2008). Cytokines can also be measured in saliva, but salivary cytokine levels are generally uncorrelated with serum levels (Riis et al., 2014) and may be indicative of periodontal disease (Riis et al., 2014). Saliva is also a source of white blood cells, which might be used for RNA‐seq, methylation, or other assays. The concentration of leukocytes in saliva is ~1.6 × 10⁵ cells/ml (Vidović et al., 2012), meaning ~2 ml of saliva would be equivalent to 50 ul blood. However, cells in saliva are only ~60% leukocytes with the remaining 40% being buccal cells (Theda et al., 2018), which may change the interpretation of these assays. Salivary leukocytes are also affected by gingivitis and other oral conditions, and in general proportions differ from blood, with monocytes overrepresented. Because of these differences, further validation is needed to understand how best to interpret measures in saliva, and how they correlate (or perhaps trade‐off) with measures in blood.