The Wild West of Spike-in Normalization

Abstract

Spike-in normalization is a powerful approach to assess global changes in data obtained from genomic mapping of DNA-associated proteins by methods such as ChIP-sequencing (ChIP-seq)^1,2 or CUT&RUN³. While multiple spike-in methods provide detailed documentation, the implementation of these approaches often omit critical quality control steps and veer from the established procedures. Spike-in normalization typically makes use of a single scalar to normalize genome-wide data, making the approach particularly vulnerable to errors in implementation. Here, we show that proper application of spike-in normalization can increase quantification accuracy across a spectrum of conditions and outline how misuse of spike-in approaches can create erroneous biological interpretations. We conclude by providing guidelines to minimize pitfalls when applying this approach to normalize data from protein-DNA interaction results.

Spike-in normalization emerged to accurately assess global changes in signal, yet misuse of the approach can skew results

Spike-in normalization was developed as a means to correctly quantify protein-DNA interactions in cases where the overall concentration of target DNA-associated proteins changes significantly between samples^1,2,4. The spike-in of exogenous chromatin from another species to each sample prior to immunoprecipitation (IP) serves as an internal control, with the assumption that the epitope of interest does not vary in the added exogenous material. In a somewhat analogous manner to the use of the ERCC RNA standards⁵, this normalization strategy can reduce variability between replicates² and capture changes in genome-wide signal intensity that would otherwise be obscured by standard read-depth normalization¹. Multiple spike-in normalization approaches have been tailored for specific applications; as such they differ by the source of the exogenous chromatin (biological or synthetic), whether a separate antibody is used for the sample and the exogenous epitope, the ratio between the exogenous chromatin and the sample chromatin, and the downstream computational analysis pipeline (Table 1).

Table 1. Overview of spike-in methods and examples of misuse that create uncertainty in interpretation.

Left column summarizes commonly used spike-in normalization methods. Variations on background choice, normalization method, or spike-in species are highlighted. The most common deviations from published protocols involve separate alignment to the target and spike-in genomes and lack of sufficient/consistent input samples. Hs: Homo sapiens, Dm: Drosophila melanogaster, Mm: Mus musculus.

Normalization tool	Normalization Model	Key limitations	Examples of Misuse
ChIP-Rx: Common antibody for the sample and spike-in chromatin1	$\alpha =\frac{1}{N_d}$, $\alpha$ = normalization constant ${\mathsf{N}}_{\mathsf{d}}$ = Spike-in $Dm$ reads	The model assumes linear behavior of signal to epitope abundance. Normalization metric applied linearly across signal and background and does not employ the input to account for spike-in variation.	Inappropriate alignment to spike-in and target genome separately10
Bonhoure et al.: A method using a common antibody for the sample and spike-in chromatin2	$Z_{i,k}={\alpha }_k{\gamma }_{i,k}+{\beta }_k{\mathrm{x}}_{i,k}+{\epsilon }_{i,k}$ $\mathsf{i}$ = genomic segment ${\mathsf{z}}_{\mathsf{i},\mathsf{k}}$ = sample counts ${\beta }_{\mathsf{k}}{\mathsf{x}}_{\mathsf{i},\mathsf{k}}$ = estimated from input, probability distribution of nonspecific reads (noise) ${\alpha }_{\mathsf{k}}{\gamma }_{\mathsf{i},\mathsf{k}}$ = specific tag counts corresponding to protein occupancy scores (signal) ${\epsilon }_{\mathsf{i},\mathsf{k}}$ = experimental errors	Significant overlap between genomes. The model assumes linear behavior of signal to epitope abundance. Spike-in normalization: assumes the background-adjusted counts should be invariant between samples. Normalization only to “reliable signal” regions of the spike-in data	Spike-in reads vary by ~10 fold and are too low for accurate quantification. Incomplete draft of the spike-in reference genome (D. iulia) assembly. Background regions are assumed, but not confirmed to be invariant11.
Epicypher ICeChIP: Synthetic nucleosome spike-in12	Calculate the % Input for each gene locus of interest (or genome-wide for ChIP-seq). Calculate the % Input for the on-target SNAP-ChIP spike-in $Normalizedsignal=\frac{(\%InputofGeneLocus)}{(\%InputofSNAPChIP)}$	Limited to study of histone marks and common epitope tags. Synthetic histones must be purchased for each modification of interest.	Ratios of spike-in and sample input chromatin varies by ~9x between replicates and low spike-in read depth13.
Parallel ChIP: Spike-in specific antibody as well as antibodies are common for sample and spike-in chromatin14	Spike-in specific antibody: adjusted spike-in peaks to have distribution centered at zero (unchanged), applied same normalization factor to sample. Parallel ChIP: antibody for DNA-binding protein whose genomic distribution and intensities are invariant. Plot control peaks separately, adjust center distribution of peaks to be unchanged.	Applied spike-in normalization in two ways, first with a spike-in specific antibody for H2Av, histone variant only in $Dm$. Second approach normalized to a separate control antibody (Parallel ChIP). Use of a spike-in specific antibody assumes experimental procedures will not affect spike-in IP differently than the target IP.	Inappropriate alignment to spike-in and target genome separately. The close evolutionary distance between spike-in $(Mm)$ and target $(Hs)$ results in overlap of reads that align to both genomes15.
Egan et al.: A method using a spike-in specific antibody $(Dm)$16	The spike-in normalization is performed by counting $Dm$ reads in each sample and using those read counts to generate correction factors; calculated as control/treated reads.	Spike-in specific antibody assumes experimental procedures will not affect spike-in IP differently than the target IP. No requirement for including inputs.	Inappropriate alignment to spike-in and target genome separately. Some input samples missing17.
SNP-ChIP: Use SNPs to enable spike-in of different S. cerevisiae strains18	${\mathsf{C}}_{\mathsf{endo}}∕{\mathsf{C}}_{\mathsf{spike}}$ = ratio of reads in target/spike-in species $Q=C_{spike}∕C_{endo}$ $Nf=Q_{Input}∕Q_{ChIP}$	Normalization factor can only be derived from SNP regions18,19. Limited in application depending on target genome.	None identified in our survey.
Active Motif Spike-in Normalization Kit: (Cat #61686 and #53083) adapted from Egan et al16	Select IP with lowest # of spike-in $(Dm)$ reads, divide by spike-in reads for that sample to get normalization factor. Down sample target reads by normalization factor.	No use of inputs in normalization to account for variable spike-in/target chromatin ratio. Separate spike-in specific antibody.	Inappropriate alignment to separate spike-in and target genomes. No input samples available20.

Importantly, the implementation of spike-in normalization has several potential pitfalls. The basic assumption is that the ratio between the spike-in and the sample chromatin is identical between conditions, providing a constant signal to normalize against. Yet trusting this assumption without incorporating proper quality control (QC) steps may lead to erroneous normalization factors in the analysis (e.g., cases where initial spike-in to sample chromatin ratios vary). In addition to experimental error, key assumptions such as the expectation that ChIP enrichment is linearly dependent on the relative epitope abundance over a range of different spike-in concentrations and antibodies, as well as other experimental parameters, may not hold. Further, in contrast to other widespread approaches (e.g., quantile normalization) that incorporate the genome-wide distribution of ChIP enrichment to draw conclusions, most spike-in normalization techniques rely only on a single scaling factor to carry out the genome-wide transformation (Table 1). Thus, this single scaling factor may highly influence the quantitative results and biological interpretation obtained from such genome-wide assays.

Many original spike-in normalization methods address these vulnerabilities in their work; however, a major concern emanates from the improper use of spike-in normalization by studies that deviate from the original protocols cited. Such misuses include: (1) a lack of critical QC steps leading to large variability between ratios of spike-in to sample chromatin or unsuccessful ChIP of the spike-in; (2) deviations from the recommendations in the original method (e.g., alternative alignment strategies); and (3) absence of true biological replicates that would otherwise reveal unexpected variation. These misusage cases call into question the validity of some of the post-normalization conclusions.

While the discussion outlined above refers to usage of spike-in chromatin to normalize ChIP-seq experiments, attempts to employ similar normalization procedures have also been applied to other genome-wide mapping methods such as CUT&RUN and CUT&Tag^3,6. However, in some of these cases the term “spike-in” indicates the addition of exogenous naked DNA instead of chromatin. For instance, either spike-in of Drosophila DNA or carryover E.coli DNA from MNase purification are added to CUT&RUN as a means for normalization of library preparation^3,6,7. We note that the use of the terms “spike-in calibration” or “normalization” in these cases does not account for variation from antibody efficiency and thus caution the reader to take that into consideration when conducting analysis and data interpretation. Importantly, CUT&RUN was shown to benefit from the addition of exogenous cells⁸ or synthetic nucleosomes to the samples⁹. In general, we recommend choosing the spike-in methods that account for as many potential sources of experimental variation as possible. Ideally, this means a spike-in containing the epitope of interest from biological material resembling the sample (e.g., cells or chromatin) as this strategy provides the most quality controls and assessments of antibody efficiency, sample handling and processing, etc. In this correspondence, we focus on methods and datasets that employ exogenous spike-in chromatin to normalize ChIP-seq data.

Here, we first provide examples to demonstrate how proper application of spike-in normalization can increase quantification accuracy. We then summarize our survey of misuse cases from the literature and discuss how the inappropriate employment of spike-in normalization could affect the resulting biological observations. Lastly, we present an example to convey the importance of using the appropriate computational approach for correcting variations in spike-in normalization and conclude by providing guidelines to minimize pitfalls when using this approach.

Proper application of spike-in normalization can increase quantification accuracy across signal ranges

To evaluate the ability of spike-in normalization to correctly quantify variations in the abundance of DNA-associated proteins we focused on experimental conditions with pre-defined ground truth. We initially reanalyzed the data from one of the original spike-in methods¹, which includes titration of H3K79me2 levels over a 10-fold range (Supplemental Fig. 1a) by mixing known ratios of cells treated and untreated by the DOT1L inhibitor. This data, as was previously shown¹, provides a clear demonstration of the ability of the spike-in normalization approach to improve upon read-depth normalization and correctly quantify enrichment over the range of signal intensity (Fig. 1a).

Fig. 1. Demonstration of the ability of spike-in normalization to accurately capture signal variation over wide and narrow dynamic ranges.

(a) Reanalysis of data from Orlando et al.¹, where untreated cells (high H3K79me2 signal) were mixed with DOT1 inhibited Jurkat cells (low H3K79me2 signal) in 5 ratios of treated/untreated (0/100, 25/75, 50/50, 75/25, 100/0). Dynamic range of H3K79me2 signal was measured by using Western blot results from Orlando et al.¹ with ImageJ; the change in H3K79me2 between 0% treated and 100% treated samples was approximately 10-fold (Supplemental Fig. 1a). Maximum signal (10X) and minimum signal (1X) are labeled on x axes. The H3K79me2 ChIP-seq signal was quantified and plotted against the line of expected signal either by using standard read depth normalization (left) or spike-in normalization (right). Accuracy of the fit of the expected line was determined by R squared (the value is reported in top right of each plot). (b) The results of a similar titration experiment generated to focus on a narrower dynamic range. Mitotic-arrested cells were generated by treating with thymidine then S-trityl-cysteine (STC), hereafter termed “mitotic”. We estimate approximately 85-90% of cells arrested in prometaphase by this method. From previous mass-spectrometry data²¹, the approximate fold change between mitotic H3K9ac (low H3K9ac signal) and interphase H3K9ac (high H3K9ac signal) was 3x, labeled on x axes (Supplemental Fig. 1b). We used interphase cells (high H3K9ac) mixed with mitotic-arrested cells (low H3K9ac) in six ratios (100/0, 95/5, 75/25, 50/50, 25/75, 0/100). To each of the samples, we spiked-in both D. melanogaster and S. cerevisiae chromatin. Quantification of the H3K9ac signal after read-depth normalization (left) or spike-in normalization using Drosophila (right) was plotted as in (a). The normalization using S. cerevisiae provided similar results (Supplemental Fig. 3). Within each plot, H3K79me3 or H3K9ac signal was min-max normalized according to the following equation: ${\mathsf{z}}_{\mathsf{i}}=({\mathsf{x}}_{\mathsf{i}}\text{-}\mathsf{min}(\mathsf{x}))∕(\max (\mathsf{x})\text{-}\mathsf{min}(\mathsf{x}))$. Here, $\mathsf{min}(\mathsf{x})$ is the average minimum signal (100% treated cells in (a) or 100% mitotic cells in (b)), $\max (\mathsf{x})$ is the average maximum signal (0% treated cells in (a) or 0% mitotic cells in (b)). ${\mathsf{X}}_{\mathsf{i}}$ is the signal for each sample and ${\mathsf{z}}_{\mathsf{i}}$ is the minmax normalized signal for each sample. QCs for the dataset are in Supplemental Fig. 2-6 and Supplemental Table 3.

Next, building on the observations by Orlando et al.¹, we conducted additional titration experiments to evaluate the spectrum of conditions where spike-in normalization can operate. Our previous mass-spectrometry results showed that mitotic cells have about 3-fold reduction in H3K9ac compared to interphase (Supplemental Fig. 1b)²¹. Using mitotic and interphase cells, we conducted a titration experiment focused on this narrow range of acetylation levels (from 1x to 3x; Fig. 1b; Supplemental Fig. 2). Importantly, the narrow signal range provides a powerful demonstration of the inability of standard read-depth normalization to capture the expected trend (Fig. 1b, left). On the other hand, spike-in normalization effectively separates the samples based on their expected signal at this dynamic range (Fig. 1b right; Supplemental Fig. 3).

The impact of misuse of spike-in normalization on downstream results

We identified in the literature several common scenarios of inappropriate employment of spike-in normalization (Fig. 2). We first consider each of these scenarios and the potential effect of such spike-in normalization misuse on the resulting biological conclusions. Next, we provide an example for the widespread inappropriate implementation by reanalysis of public datasets using this normalization approach.

Figure 2. A schematic depicting the impact of misuse of spike-in normalization on downstream results.

Left, “True Enrichment” shows a hypothetical scenario of two conditions with variable histone acetylation profiles. Panels a-e show various scenarios observed in literature and their effect on both local signal (depicted as schematics of individual ChIP-seq peaks) and global signal (shown as a scatterplot of log-normalized signal for all peaks for each condition). The data represents 3 replicates, while for simplicity, the peaks and scatter plots show only the average of the replicates. Of note – these scenarios are not mutually exclusive, and combinations of these misuse of spike-in normalization can occur.

(a) Traditional ChIP-seq with no spike-in added – the signal appears identical between the two conditions, not capturing the true patterns.

(b) Properly performed normalization with spike-in ChIP (blue) yields results that accurately follow the original ground truth.

(c) Variable proportion of spike-in added to the samples, this skews the analysis and condition #2 appears to wrongly have higher signal.

(d) Low yield of spike-in data, either from input chromatin levels or technical issues. This precludes any QC of the spike-in. Also, the low number of reads that align to the spike-in genome could highly skew the results one way or the other (depicted as overlaying normalized peaks and two plots in both directions).

(e) Exogenous chromatin from a species which is phylogenetically close to the sample species. This could skew the results in both directions, depending on the percent of misassigned reads in each sample (depicted as in (d)).

No spike-in.

For ChIP-seq without spike-ins, global changes are lost after read-depth normalization, and the samples appear identical^1,2 (Fig. 2a).

Proper, constant spike-in/sample ratio.

In contrast, proper implementation of the spike-in normalization approach that includes a constant ratio of exogenous/target chromatin accurately captures the ground truth of relative signal between samples (Fig. 2b, green checkmark).

Variable spike-in/sample ratio.

Variation between exogenous/target chromatin ratios between ChIP-seq conditions (Fig. 2c). In this example, the normalization distorts the underlying biology in a manner that is dependent on the ratio, resulting in the normalized enrichment of the target epitope deviating from the ground truth. Deceptively, a QC of the spike-in by read-depth normalization will obscure the differences in total epitope amount, in a similar manner to the ChIP-seq of “no spike-in” samples in Fig. 2a.

One of the most important QC steps for the processing of spike-in experiments is to analyze the non-IP input (control) for all samples. Input samples contain the sheared chromatin that was not used for ChIP and serve as a control to identify non-specific signal. The input can also be used to directly measure the sample/spike-in chromatin ratio that was initially present before the ChIP step is performed, which should be consistent across all samples.

To evaluate the breadth of studies that fail to conduct this QC, we surveyed both GEO datasets and Google Scholar results. We first included 4 studies that initially developed spike-in normalization methods^1,2,14,16 and experimental data we generated (Fig. 2b). The remainder of datasets were chosen more agnostically, by surveying GEO and Google Scholar results for the phrase “spike-in ChIP”, then sorting the results according to best match/relevance, and lastly filtering out irrelevant datasets (no ChIP-seq data, or lack of spike-ins as previously defined in this correspondence). We aimed to include a diversity of species used (spike-in and target) and epitope types (histone mark vs transcription factor), arriving at a final set of 53 datasets in our survey, expecting that these data make up a sizable proportion of all ChIP-seq spike-in studies. Out of these 53, only 27 included the appropriate input samples for each condition (Fig. 3 insert)^{1,10,11,13-15,22-42}. In 12 of the datasets^2,17,42-51, a subset lacked inputs for some conditions or treatments, making it impossible to QC the spike-in/target chromatin ratio. The last 14 datasets^16,20,52-63 lacked inputs altogether (Fig. 3 insert).

Figure 3. Variations in the ratios of spike-in to sample chromatin in public datasets.

We examined 53 datasets. 51 of these were obtained from our survey. We additionally included the data we generated for Fig. 1b and Supplemental Fig. 2 (n=2). Of these only 27 had sufficient input samples for each condition (52.9%, pie chart; insert). One additional dataset plotted had > 1 input but was still missing inputs for some conditions. We aligned the inputs to a concatenated spike-in and target genomes (Methods). For each dataset, the input with lowest ratio of spike-in/sample reads was scaled down to 1, to capture the variation between input samples in the same dataset. Y-axis is shown on log2 scale to capture the diversity in variation within datasets. Alignment information for each individual sample is available in Supplemental Table 1, GEO accessions for each dataset are in Supplemental Table 5.

Of the datasets with > 1 input sample, the median variation of the spike-in/target chromatin ratio is 1.4-fold, with the extreme variation being 12.4-fold (Fig. 3; Supplemental Table 1)¹¹. Thus, the assumption that equal amounts of exogenous chromatin were incorporated into all samples should be questioned. Further, this highlights the importance of using the input samples for every sample and condition for QC, as the inputs may provide an opportunity to correct variations in the spike-in/target chromatin ratios as is recommended previously^2,53. Otherwise, an unaccounted-for variation in the spike-in/sample ranges may result in erroneous scaling that can lead to invalid results in a similar manner to the scenario in Fig. 2c.

Low spike-in/sample ratio.

In cases where the ratio of exogenous chromatin added is extremely low, it can be impossible to conduct QC for the ChIP-seq of the spike-in chromatin. This may also increase the noise due to the higher variability associated with the potentially low read counts originating from the exogenous chromatin (Fig. 2d). We conducted a literature survey to evaluate the abundance of such cases, finding that the genome coverage of the spike-in samples ranged between 0.00012X to 6.5X. In 15 of the studies the genome coverage of the spike-in samples was low (< 0.1X)^{2,10,11,14,23,26-29,31,33,34,38,41,43}. In an extreme example¹¹, where the spike-in was designed to be at a ratio of ~1:10,000 of the sample, the coverage of the spike-in genome was 0.00012X to 0.000375X¹¹.

Alternative alignment strategies.

Original spike-in protocols indicate that the alignment is ideally conducted by using a concatenated genome from both the exogenous and target species, where multimapping reads are discarded^1,2. While aligning the sequencing data separately to the two genomes may provide a benefit, and for applications where the expected enrichment occurs in repeat regions it could potentially improve accuracy, such an alignment approach requires careful post processing. For instance, reads that align to both genomes may be easily double counted, thus potentially rendering the analysis inaccurate (Fig. 2e, Supplemental Fig. 7 and Supplemental Table 2). This risk increases when using closely related spike-in and target genomes such as H. sapiens and M. musculus (as observed in 15 datasets out of the 53 we surveyed)^{2,14,15,25,28,29,41,43,44,48,52,53,56,59,61}.

Accounting for variation in the spike-in/sample ratio may increase quantification accuracy across a wide range

Many of the public datasets we surveyed had variability in the spike-in/sample ratio across conditions (Fig. 3, above). To evaluate the ability of correcting for such deviations between samples, we generated a dataset where we only varied the ratio of spike-in to target chromatin and quantified differences in enrichment. We used mitotic human cells (synchronized HeLa-S3 with low H3K9ac) that were treated with the pan-HDAC inhibitor TSA which increases the global levels of histone acetylation⁶⁴ or DMSO as a control. Using the same chromatin preparation from either TSA- or DMSO-treated cells, we varied the ratio of the spike-in to target chromatin over 5 orders of magnitude (from 0.00025x to 2.5x; Fig. 4; Supplemental Fig. 8; Supplemental Table 4). Using this data, we compared three commonly used normalization approaches: (1) standard read-depth normalization (Fig. 4a); (2) normalization of relative IP ratios of spike-in and target under the assumption that the spike-in is constant (in a similar manner to previous studies^1,16, Fig. 4b); and (3) using the non-IP input controls to detect the ratios of spike-in to target chromatin present in each experiment to calibrate IP normalization (Fig. 4c).

Figure 4. An example of the ability of proper normalization strategy to correct for variations in the spike-in/sample ratio.

Mitotically arrested HeLa-S3 cells were treated with either the pan-HDAC inhibitor TSA (to increase global acetylation) or DMSO as a control. For each treatment, using the same chromatin preparation the amount of added spike-in was varied over five orders of magnitude (0.00025 to 2.5x spike-in/target ratio). The data were normalized by three approaches: (a) read-depth normalization – the change in the signal between the conditions cannot be observed; (b) normalization of relative IP ratios of spike-in and target – most of the change in the signal is not observable and the amount of spike-in added impacts the general signal; and (c) account for the ratio of spike-in/target by using the non-IP input controls and then normalize as in (b) using the relative spike-in/target IP – the known differences between the conditions can be detected in all cases but the one where the amount of spike-in is higher than the target. QCs for the dataset are in Supplemental Fig. 8-14 and Supplemental Table 4.

These data show that both read depth normalization (Fig. 4a) or ignoring the true ratio of the spike-in to target chromatin (Fig. 4b) is prone to fail in capturing the ground truth. When we account for the ratio of spike-in/target by using the input and then normalize using the spike-in IP, the change in signal between conditions can be detected across most of the tested spectrum of spike-in/target ratios (Fig. 4c). Importantly, following this normalization approach, the target signal, which should remain invariant, fluctuated by a ratio of 0.74-2.33x when we increased the spike-in/target ratio by 10x in a stepwise manner (Fig. 4c). Thus, while some level of variation in the spike-in/target ratio can potentially be corrected for by using the inputs as recommended previously², it is critical to minimize the variability in this ratio between conditions. Lastly, this observation also demonstrates that the ability of spike-in normalization to quantify true signal can be impacted by the amount of spike-in chromatin added.

Recommendations and Future Directions

Given the risk of misusing spike-in normalization, and the high number of occurrences of such instances, we provide several guidelines for avoiding pitfalls when implementing spike-in normalization for ChIP-seq experiments.

1. The QC steps of the spike-in are required. This includes both measuring the spike-in/target ratio for each sample by isolating and sequencing the unenriched input sample. It is also critical to analyze the ChIP-seq signal for the spike-in itself – i.e., by visual inspection on a genome browser such as UCSC⁶⁵ or IGV⁶⁶ as well as peak calling and metagene analysis.

2. The spike-in species should have a well annotated and complete genome assembly, ideally a model species such as D. melanogaster.

3. The proper experimental implementation of spike-in chromatin should be guided by multiple considerations: (1) the raw spike-in chromatin material must be sufficient for verifying a successful IP when analyzing the data (measured as detailed in Methods); (2) the amount of target chromatin relative to spike-in should allow sufficient sequencing of the mixed species while remaining within practical sequencing depths; and (3) the depth of sequencing for the mixed species should account for the added genome of the spike-in samples and follow the ENCODE guidelines⁶⁹ to provide sufficient coverage.

   The above parameters will be affected by the size of the spike-in genome, the epitope of interest (and its prevalence in spike-in and target samples), as well as the overall experimental design. As an example, based on our experience and studies by others¹, when using spike-in chromatin from species such as D. melanogaster or S. cerevisiae to detect changes in histone modifications in H. sapiens, we recommend aiming for a genome coverage of about 0.1x for D. melanogaster (~100-150k reads) and a genome coverage of about 0.5x for S. cerevisiae (~50-100k reads), given the high gene density of the S. cerevisiae genome⁶⁷.

4. To decrease variability in spike-in/target chromatin ratios, we quantify DNA before combining chromatin from each species. Additionally, we caution that for some applications, the exogenous chromatin may take up more reads in the IP than expected, and titrations of the amounts may be useful to identify optimal conditions (detailed in Methods).

5. Each experiment should include 3–4 biological replicates to ensure reproducibility. Additionally, confidence intervals of spike-in/sample signal should be plotted to enable estimation of the variability.

6. To determine the acceptable level of variation between conditions for the ChIP signal of the exogenous chromatin, we recommend applying the Irreproducible Discovery Rate (IDR) calculation, as suggested by the ENCODE’s guidelines for assessing variability between replicates⁶⁹.

7. The variation of spike-in/target unenriched inputs should not exceed the variation of spike-in/target ChIP reads (this variation is measured as ChIP reads aligning to spike-in genome divided by ChIP reads aligning to target genome, see Methods). Otherwise, rather than the change in true ChIP signal between spike-in and target, the technical variation in spike-in quantities will have an outsized impact on the normalization factors.

8. Alignment to a merged genome that combines both the target and spike-in species and stringent filtering (keeping only primary alignments and MAPQ>10) decreases the chances of ambiguous alignment.

9. Lastly, for robust and reproducible results, the conclusions of the experiment should be validated by an orthogonal assay such as quantitative Western blots of the chromatin fraction, immunofluorescence assays, mass spectrometry, or other genetic or biochemical evidence depending on the context of the study.

Further, spike-in normalization methods are under active development and improvements that may address current key limitations of the strategy. Here, we discuss some of the main future directions:

1. Most spike-in normalization methods employ a single scaling factor to transform the genome-wide data resulting from ChIP-seq. This global transformation of the entire dataset may not accurately reflect the true values in all regions, particularly when considering ChIP-enriched peaks versus nonenriched background regions. Recent ChIP-seq normalization methods attempt to address this limitation by scaling signal and background separately^2,70. An additional way to improve the robustness of the applied scalar is to curate the regions where spike-in reads are quantified to derive the normalization factor (e.g., limiting the analysis to regions surrounding the TSS when analyzing active histone marks).

2. One strategy to increase confidence in the spike-in normalization factor is to employ two exogenous spike-ins as we demonstrated here (Fig. 1b and Supplemental Fig. 3). Each spike-in species is used to generate its own normalization factor. Applying both normalization factors reduces the effect of noise inherent from using each alone and the two species enhance the QC.

3. Rather than deriving a normalization factor based on the simple sum of all spike-in reads, it may be more prudent to control for the effect of outliers in a similar manner to DESeq2⁷¹ by using the middle quartiles of the spike-in peaks to improve the robustness of the normalization constant calculation.

Altogether, the proper use of spike-in normalization improves the sensitivity for detecting genome-wide changes between conditions. Given the increase in the use of ChIP-seq and similar genomic approaches to identify the impact of perturbations with widespread effects, we envision that spike-in normalization along with the advancement of this method, will be critical to identify new biological observations.

Methods:

Public Data Analysis

Detailed code is available at the Github repository link in the Additional Data section, however the basic procedure for calculating alignment statistics is shown below:

1. All FASTQ files downloaded using sra-toolkit fasterq-dump:

   $ fasterq-dump --skip-technical --split-3 SRR#####

2. FASTQ files aligned with Bowtie2 default parameters to a concatenated reference genome containing spike-in and target genomes. Spike-in chromosomes labeled with suffix of genome version to distinguish from target chromosomes.

   $ bowtie2 -p 24 -x target_spike-Bowtie2Index/target_spike SRR#####.fastq -S SRR######.dualalign.sam

3. Aligned file sorted, then split into species-specific .bam files with bamutil splitChromosome

   $ samtools sort SRR######.sam > SRR######.sorted.bam

   $ bam splitChromosome --in SRR######.sorted.bam –out SRR#####.

4. Unmapped, non-primary alignments removed, and reads with MAPQ<10 removed with HOMER makeTagDirectories. Fragment length explicitly set to 150.

   $ makeTagDirectories SRR######-tagdir/ -genome genome -fragLength 150 SRR######.genome.bam

*NOTE*: The number of aligned reads for paired end Read 1 may not equal Read 2, therefore “total tags” may be reported in intervals of 0.5.

• 5.Total read counts tallied in Supplementary Table 1.

Genome versions:

H. sapiens: hg38

D. melanogaster: dm6

D. pseudoobscura: dp3

M. musculus: mm10

D. iulia: GCA_019049465.1

Synthetic nucleosome sequences from Epicypher: Kmet_stat panel

Experimental Methods

Cell Culture

Hela-S3 Cell Media and Culture:

HeLa-S3 cells were grown in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 1x penicillin-streptomycin, L-glutamine, and 110mg/L sodium pyruvate. Cells were incubated at 37°C and 5% CO₂.

S2 Cell Culture:

Drosophila S2 cells were grown in Schneiders Drosophila Medium at 28°C in the dark.

Yeast Cell Culture:

Yeast Cell Culture: From frozen stocks, starter cultures of strain LPY6494 were inoculated in 3mL of Gibco liquid YPD media (Cat #A1374501), grown to log phase at 30°C at 220rpm, then expanded to 50mL cultures at an OD of 0.08 and grown for 16 hours at 30°C and 220rpm. OD was estimated with Nanodrop. Yeast culture and fixation performed as previously described⁷².

Hela-S3 Mitotic Synchronization

Cells were synchronized in G1/S by adding 2 mM thymidine (Sigma T1895-1G) for 16 hr, washed with PBS, released for 3 hr in fresh medium, and pro-metaphase arrested with 10uM S-trityl-L-cysteine (STC) (Sigma 164739) for 12 hr then harvested. For simplicity, we term these cells as “mitotic”, even though according to our estimates approximately 85-90% of cells are arrested in prometaphase. When specified, cells are treated with DMSO or 150nM TSA (Selleck Chemicals S1045) 9 hours before harvesting.

ChIP-seq

ChIP-seq was performed as previously described⁷³. Briefly, HeLa-S3 cells were double crosslinked on plates with 2 mM DSG for 30 minutes (Fisher Scientific NC1736108) and 1% formaldehyde (Pierce ThermoFisher PI28906) for 10 minutes. Crosslinking was quenched with 1/20th volume of 2.625 M glycine (EMD Millipore 357002) and 0.5% BSA (NEB B9000S), then cells were washed 3 times with 0.5% BSA/PBS + Protease inhibitor (PIC, Roche 11697498001), spinning down at 1200g for 5 minutes at 4C. Washed pellets were snap frozen then stored in −80C. Cells were lysed with LB3 lysis buffer + PIC, then chromatin was sheared on the PIXUL sonicator for 108 minutes (Active Motif 53130). S2 cells were crosslinked in 1% formaldehyde for 10 minutes at room temperature, then quenched with glycine and 0.5% BSA, lysed in LB3, and sonicated on the PIXUL for 60 minutes. Yeast cultures in log phase were crosslinked with 0.86% formaldehyde for 30 minutes. The yeast cell wall was digested with Zymolase at 37°C for 60 minutes, cells were washed 3x in PBS + PI, then resuspended in NP-S buffer + PI. Yeast were sonicated on the PIXUL for 120 minutes. All PIXUL sonication was done with settings Pulse (N) = 50, PRF (kHz) = 1.00, Burst Rate (Hz) = 20, only Process Time was varied.

Sheared chromatin from each species were combined and then inputs were taken before immunoprecipitation for each condition to assess the relative ratios of chromatin. For the experiment detailed in Fig. 1b, where H3K9ac was titrated in mitotic and interphase HelaS3 cells, chromatin from approximately 600k HeLaS3 cells was combined with chromatin from 500k S. cerevisiae and 100k D. melanogaster S2 cells in each IP. For the experiment detailed in Fig. 4, where we titrated S. cerevisiae spike-in amounts, chromatin from S. cerevisiae was added in ratios ranging from 0.00025-2.5x relative to HeLaS3 cells.

Notes on minimizing variability between spike-in/target ratios:

Since there are multiple opportunities to introduce variability between cell counting and combining of material from each species, it is difficult to maintain constant spike-in/target chromatin ratios. To overcome this challenge, rather than relying on cell count estimates, we measure sheared DNA quantities as means to estimate chromatin amounts prior to combining material from each species.

Thus, following shearing by PIXUL, a subset of chromatin was de-crosslinked, the RNA was removed with RNAse A (Roche-SIGMA 11119915001) and protein was removed with Proteinase K (Invitrogen 25530049). The resulting DNA is quantified with Qubit HS DNA kit (Invitrogen Q32854). Then the samples can be calibrated such that equal amounts of spike-in and target chromatin are added for every condition. Except for the experiment where yeast chromatin was deliberately varied, the estimated number of yeast cells added per IP is no more than 1x the number of human cells. As the S. cerevisiae genome is gene dense and the epitope of interest is a histone acetylation mark, S. cerevisiae makes up a significant proportion of reads in the IP (Supplemental Table 3). In a sample with minimal target IP signal, e.g. 100% STC-treated (mitotic) cells, S. cerevisiae accounts for ~0.8% of the input reads and ~30% of IP reads from the same chromatin pool (Supplemental Table 3, Supplemental Fig. 2).

To each IP, 10uL of Protein A beads (Thermo Fisher Scientific 10002D) and 4uL of H3K9ac (CST 9649) were added, 1% Triton X-100 and 1x PIC were added. Samples were incubated overnight on rotator at 4C. IP Washes were performed as follows: 3x with 180uL of WBI, 3x with 180uL of WBIII, and 2x with 185uL TET, then eluted in 25uL of cold TT Buffer (buffer components below). Inputs were incubated with Speedbeads (Cytiva 65152105050250), in 12% PEG (Millipore Sigma P5413) and 1M NaCl for 10 minutes at room temperature, then washed twice with 80% EtOH, and eluted in 25uL cold TT Buffer. From inputs and IPs, libraries were prepared with the NEB Ultra II Library Prep Kit (NEB E7645L). Barcoded adapters were from BioO NEXTflex ChIP-seq Adapters (NOVA-514104 & NOVA-514151). Libraries were de-crosslinked at 55C 1hr, then 65C overnight, RNA removed with RNAse A (Roche-SIGMA 11119915001) and protein removed with Proteinase K (Invitrogen 25530049), and cleaned with Speedbeads, 8.6% PEG and 0.8M NaCl. Next libraries were PCR amplified with 1uM Solexa 1GA and 1GB primers and NEB Ultra II Q5 Mastermix (PCR protocol: 98C for 30’’, then 12 cycles of 98C for 10’’, 60C for 15’’, 72C for 30’’, then hold at 72C for 2’). A final cleanup was done with Speedbeads, 8.5% PEG and 1M NaCl, washed twice with 80% EtOH, then the final libraries were eluted in 20uL TT. Library size was estimated on a 2% agarose gel and concentration quantified with Qubit HS dsDNA kit (Invitrogen Q32854). Libraries were pooled and sequenced on Illumina NextSeq 550.

Buffer components:

NP-S Buffer: 10mM Tris-HCl pH 7.4, 0.5mM EDTA, 50mM NaCl, 5mM MgCl2, 1mM CaCl2, 0.075% NP-40, 1mM 2-mercaptoethanol

LB3: 10mM Tris-HCl pH 7.5, 100mM NaCl, 1mM EDTA, 0.1% deoxycholate, 0.5% Sarkosyl

WBI: 20mM Tris-HCl pH 7.5, 150mM NaCl, 1% Triton X-100, 2mM EDTA, 1% SDS

WBIII: 10mM Tris-HCL pH 7.5, 250mM LiCl, 1% Triton X-100, 1mM EDTA, 0.7% deoxycholate

TET: 10mM Tris-HCl pH 8.0, 1mM EDTA, 0.2% Tween 20

TT Buffer: 10 mM Tris-HCl pH 8.0, 0.05% Tween 20

Bioinformatic Analysis

An overview of data processing and analysis steps is shown in Supplemental Fig. 15, and details that vary between datasets are described below.

Data Processing and Initial Analysis

Samples were demultiplexed with Illumina’s bcl2fastq program, adapters were removed with Trimmomatic, and FASTQC was run for sequencing quality control. Reads were aligned with BWA MEM to a concatenated genome containing hg38 + spike-in species (dm6, sacCer3, or both). The aligned sam file was sorted and split by chromosome using bamUtil splitChromosome. Spike-in chromosomes were pre-labeled with a suffix (_sac3 or _dm6), samtools merge was used to recombine alignments for each species. Duplicate reads were removed with samtools markdup. For easier downstream processing, HOMER tag directories were made; by default read counts for each species were filtered for primary alignments with MAPQ > 10. Signal was quantified at peaks with HOMER annotatePeaks. Peak finding for each individual sample was done with HOMER findPeaks, with - style histone, -size 1000 -minDist 2500 with the corresponding inputs as is recommended for variable length peaks: http://homer.ucsd.edu/homer/ngs/peaks.html; FRIP score for each sample was estimated in output file of peak finding.

For the yeast spike-in titration dataset, scatterplots of peak counts were generated to compare spike-in titration conditions in Supplemental Fig. 14. Sample IPs were combined into one tag directory, and all inputs combined similarly, then HOMER findPeaks was done with the recommended variable length peak parameters on this “mega-sample” using the “mega-input” as the input control. The increased read-depth of both sample and input files will increase the tag threshold for peak finding, resulting in more significant peaks. Read-depth normalized signal at each peak was quantified with HOMER annotatePeaks using mega-sample peaks to determine quantified regions, with additional parameters -size 1000. For each treatment condition (TSA or DMSO), signal in each increasing yeast ratio (0.0025-2.5x) were plotted against the sample with the minimum yeast ratio (0.00025x).

Visualization

Bigwig files were created with HOMER makeBigWig and viewed on the UCSC Genome Browser. Heatmaps were generated using Deeptools computeMatrix and plotHeatmap with the union set of all peaks called for each experiment, with additional parameters: reference-point --referencePoint center -b 2000 -a 2000. Scatterplots were generated from HOMER annotatePeaks quantification of signal at peak or tss regions as specified in figure captions. Plots generated with R, ggPlot2, and cowplot.

Normalization

ChIP-Rx titration of H3K79me2:

ChIP-Rx dataset from Orlando et al. was normalized according to the previously published method, with additional steps outlined below. The experiment was performed in human Jurkat cells, with Drosophila S2 cells as a spike-in. Therefore, all previously described alignment steps were followed with D. melanogaster and H. sapiens alignments, with Drosophila chromosomes labeled with the “_dm6” suffix. The initial ChIP-Rx method determines the normalization factor as $\alpha =\frac{1}{N_d}$, where ${\mathsf{N}}_{\mathsf{d}}$ = (number of reads aligning to the Drosophila genome/1000000)¹. We additionally applied an adjustment factor accounting for the variability in spike-in/target chromatin present in the input samples. The ratio of spike-in/target input reads is defined by the equation ${\mathsf{R}}_{\mathsf{input}}=\frac{spik{e}_{input}}{targe{t}_{input}}$. ${\mathsf{R}}_{\mathsf{input}}$ was scaled as in Fig. 3 by $R_{input,scaled}=\frac{R_{input,i}}{min(R_{input})}$ so that the sample with the lowest ${\mathsf{R}}_{\mathsf{input}}$ is set to 1. The final normalized signal is therefore calculated as $z_i=x_i\ast \left(\frac{1}{N_d}\right)\ast R_{input,scaled}$ As H3K79me2 is predominantly located on promoters, we quantified signal at TSS using similar parameters as Orlando et al., measuring 5kb upstream and downstream of the TSS with HOMER annotatePeaks.pl tss hg38 - size 10000 -hist 25. TSS histograms were made first with traditional read-normalization (the default behavior of HOMER annotatePeaks). Spike-in normalization was done by multiplying read-normalized histogram signal by the spike-in normalization factor “IP_input_factor” for each sample.

Titration of H3K9ac:

The dataset containing a titration of mitotic and interphase HelaS3 cells includes spike-in chromatin from two separate species, D. melanogaster and S. cerevisiae. Alignment was done against the concatenated genome of hg38 + dm6 + sacCer3. After splitting alignment files into each species, removing PCR duplicates, and filtering reads by making HOMER tag directories, reads aligning to each species were quantified as shown in Supplemental Table 3. Normalization factors for each spike-in species were determined as follows: norm_factor = (spike-in IP reads/target IP reads) * (target input reads/spike-in input reads). Normalization factors for each spike-in species, named “fly norm factor”, and “yeast norm factor” are listed in Supplemental Table 3. H3K9ac is present on both promoters and enhancers, so we sought to quantify signal at a set of H3K9ac peaks across all samples in the experiment. Sample IPs were combined into one tag directory, and all inputs combined similarly, then HOMER findPeaks was done with the recommended variable length peak parameters on this “mega-sample” using the “mega-input” as the input control. The increased read-depth of both sample and input files will increase the tag threshold for peak finding, resulting in more significant peaks. H3K9ac signal was quantified at these peaks with HOMER annotatePeaks, using the mega-sample peak file as first argument to define quantified regions, then additional parameters hg38 -size 4000 -hist 25. Spike-in normalization was done by dividing read-normalized histogram signal by the spike-in normalization factor “fly norm factor” or “yeast norm factor” for each sample. $z_i=x_i\ast \left(\frac{targetIP}{spike-inIP}\right)\ast (\frac{spike-ininput}{targetinput})$

Titration of S. cerevisiae spike-in:

The spike-in titration dataset, comparing TSA and DMSO treated mitotic Hela-S3 cells contained S. cerevisiae chromatin, added in ratios ranging from 0.00025-2.5x that of the target chromatin. Alignment was done against the concatenated genome of hg38 + sacCer3. After splitting alignment files into each species, removing PCR duplicates, and filtering reads by making HOMER tag directories, reads aligning to each species were quantified as shown in Supplemental Table 4. Spike-in normalization was performed according to two commonly used strategies: 1) normalizing to spike-in IP alone, and 2) normalizing to both spike-in IP and input chromatin.

1. The normalization factor was calculated as the ratio of yeast/human reads multiplied by a scaling factor of 200 to account for relative genome size and is reported in Supplemental Table 4 as “yeast/human reads*200”. When normalizing to spike-in IP alone, data were normalized by dividing by the normalization factor, “yeast/human reads*200” column for each IP sample. Signal normalized to spike-in IP is calculated as $z_i=x_i\ast \left(\frac{targetIP}{spike-inIP}\right)$.

2. When normalizing to spike-in IP and input, the normalization factor “yeast/human reads*200” for each IP was divided by the corresponding input value in the same column. The resulting normalization factor accounting for IP and input is reported in Supplemental Table 4 as the “spike_IP_input_norm”. Spike-in normalization was done by multiplying read-normalized histogram signal by the spike-in normalization factor “IP_input_factor” for each sample. Spike-in normalization was done by dividing read-normalized histogram signal by the spike-in normalization factor “IP_input_factor” for each sample. The final normalized signal ${\mathsf{z}}_{\mathsf{i}}$ is calculated as $z_i=x_i\left(\frac{targetIP}{spike-inIP}\right)\ast (\frac{spike-ininput}{targetinput})$

Quantification of Normalized Signal

ChIP-Rx titration of H3K79me2 & Titration of H3K9ac:

Total signal was estimated by calculating area-under-the-curve for both read-depth normalized and spike-in normalized histograms (see Github). Total signal for each sample was then min-max normalized by the equation $z_i=\left(\frac{x_i-min(x)}{max(x)-min(x)}\right)$. Here, $\mathsf{min}(\mathsf{x})$ is the average minimum signal (H3K79me2 titration: 100% treated cells, H3K9ac titration: 100% mitotic cells), max(x) is the average maximum signal (H3K79me2 titration: 0% treated cells, H3K9ac titration: 0% mitotic cells). ${\mathsf{x}}_{\mathsf{i}}$ is the signal for each sample and ${\mathsf{z}}_{\mathsf{i}}$ is the minmax normalized signal for each sample. The resulting data is scaled such that the average minimum signal is set at 0, and the average maximum signal is set at 1. This allows more intuitive comparison across datasets and normalization methods. The accuracy of each normalization method is quantified by calculating the R² value of data against the line of expected signal (diagonal line shown in plots of Fig. 1). For the minmax normalized data, the line of expected signal is simply the diagonal line connecting maximum (y=1) and minimum (y=0) signal. R² is calculated by the equation $R^2=1-\left(\frac{S{S}_{res}}{S{S}_{total}}\right)$ where $S{S}_{res}=sum({(z_i-expecte{d}_i)}^2)$ and $S{S}_{total}=sum({(z_i-mean(z_i)}^2)$. R² values are reported in Fig. 1 as inserts in plots.

Titration of S. cerevisiae spike-in:

While there is existing evidence of the effect of TSA on increasing histone acetylation in mitotic HelaS3 cells, this dataset contains no known maximum/minimum values with which to construct a standard curve. Therefore, H3K9ac signal was quantified at every peak with HOMER annotatePeaks -size 1000, and the distribution of signal across peaks were shown as violin plots, comparing read-normalized (Fig. 4a), spike-in IP normalized (Fig. 4b), and spike-in IP and input normalized (Fig. 4c) peak counts. Additionally, the ratio of TSA/DMSO counts in each spike-in titration condition were plotted for each normalization method (Supplemental Fig. 14).

Supplementary Material

Supplemental Tables 2-5

Supplemental Table 2: Examples of how multi-mappers may affect alignment statistics.

Supplemental Table 3: Sequencing statistics for H3K9ac ChIP in Interphase/Mitotic HeLa-S3 titration experiment.

Supplemental Table 4: Sequencing statistics for H3K9ac ChIP-seq in TSA/DMSO treated Mitotic HeLa-S3 cells.

Supplemental Table 5: GEO Accession numbers for Public Datasets in Supplemental Table 1.

Supplemental Table 1

Supplemental Table 1: Metadata and alignment statistics for public datasets analyzed.