The stress-induced transcription factor ATF4 has multiple conserved retrocopies that can alter gene expression

Summary

The cell must defend against various stressors from internal and external sources that disrupt cell homeostasis. The integrated stress response (ISR) is a highly conserved pathway that helps restore this homeostasis through upregulating the transcription factor ATF4. Despite its importance to cell health and human disease, ATF4 has several duplications in humans that have not been characterized. Here, we characterize four retroduplications (retrocopies) of ATF4 in humans. Evolutionary analysis demonstrates that these retrocopies are present and intact in many primate species over the past 37 million years, including several independent copies. We also find evidence of non-neutral evolution among primates. Human ATF4 retrocopies show basal transcription in healthy, unstressed cells and can be upregulated by the ISR. When translated in human cells, ATF4 retrocopy proteins are regulated by the proteasome in the same way as the parent ATF4 protein. Remarkably, each retrocopy can also alter the expression of several canonical ATF4 target genes, demonstrating that they can impact ISR-ATF4 stress signaling. Overall, ATF4 retrocopies are conserved, biologically functional, and should be considered in future studies of the ISR and ATF4.

The transcription factor ATF4 is critical for managing cell stress. We characterize four duplications (retrocopies) of ATF4 in the human genome. These retrocopies are well-conserved among primates, are basally transcribed, and impact stress-signaling genes. This study also adds to the growing literature on biologically relevant retroduplications.

Introduction

Cellular pathways such as the integrated stress response (ISR) are activated in response to stress to restore homeostasis. The ISR is activated by viral infection, amino acid deprivation, endoplasmic reticulum (ER) stress, or heme deprivation through the protein kinases PKR, GCN2, PERK, and HRI, respectively.^1,2 In response to this stress, the ISR inactivates the translation initiation factor eIF2B, downregulating global translation, while simultaneously upregulating specific translation of the ATF4 transcription factor^1,2 (MIM: 604064)³ (Figure 1A).

Figure 1.

Characterizing four duplications of ATF4

(A) A simplified schematic of the integrated stress response (ISR) and translation of ATF4. Under normal conditions, the ISR is not activated, and global cell translation is normal. Due to the overlapping uORF3, ATF4 translation is limited due to abundant ternary complexes skipping the ATF4 CDS. Under stress conditions, the ISR is activated, thereby inhibiting eIF2B and global cell translation. The limited ternary complexes delay re-initiation until after the uORF3 start site, thereby preferentially translating the ATF4 CDS instead. Teal sections indicate the CDS of ATF4.

(B) Cartoons of genomic ATF4, ATF4 mRNA, and genomic ATF4P1–P4. We list the current status of uORFs here, including when it is lost in ATF4P1/P2. We also list the % coverage and identity of each duplicate nucleotide sequence (minus flanking TSDs) when compared to ATF4. Teal sections indicate the CDS of ATF4, and the blue band is a short 16-aa-encoding segment in ATF4P1/2 (Small X).

(C) A diagram of each translated retrocopy compared to the translated Ensembl Canonical transcript of ATF4. Amino acid changes that occur in more than one retrocopy are highlighted in blue. We also list the % coverage and identity of each translated protein sequence when compared to translation of ATF4 mRNA (cDNA).

ATF4 is basic leucine zipper transcription factor that is critical to restoring cell homeostasis or, under high or prolonged stress, activating apoptosis.^1,4,5 It modulates the expression of hundreds of genes, including those involved in amino acid synthesis, autophagy, and apoptosis.^5,6,7 ATF4 also has important roles in development^8,9,10,11 and tumor biology^{8,12,13,14,15,16,17} in humans. The basic leucine zipper domain allows ATF4 to localize to the nucleus, bind to DNA, and dimerize with proteins such as CHOP and ATF3.¹⁸

ATF4 is regulated at the protein and mRNA level. ATF4 protein contains a p300 interaction domain that enables interaction with the histone acetyltransferase p300.¹⁹ This interaction increases ATF4 stability and is independent of the p300 acetyltransferase activity.¹⁹ ATF4 also contains two known degradation domains: an oxygen-dependent degradation (ODD) domain and a βTrCP motif.¹ Proline residues in the ODD domain are hydroxylated by the oxygen sensor PHD3 and are then targeted by the ubiquitin ligases Siah1 and Siah2.^20,21 The βTrCP motif can be phosphorylated, leading to targeting by the SCF^βTrCP ubiquitin ligase.^22,23,24 Ultimately, each degradation domain leads to ubiquitination and degradation by the proteasome, and, thus, ATF4 has a half-life of less than 1 h.^1,20,24

Under normal translation initiation, the small ribosomal subunit associates with the ternary complex (eIF2, GTP, and Met-tRNA_i^Met), binds to mRNAs, and scans for open reading frames (ORFs). Once an ORF is found, the ternary complex dissociates for the ribosome to form and initiate translation. The small ribosomal subunit must bind to a new ternary complex each time to re-initiate translation and build ribosomes. When the ISR is activated, eIF2B (part of the ternary complex) becomes inhibited and reduces available ternary complexes, inhibiting global protein synthesis. ATF4 uses transcript-level regulation to evade this inhibition by using three upstream ORFs (uORFs).^25,26,27 uORFs are nucleotide sequences, upstream or overlapping with the coding sequence (CDS) of a gene, that can affect its translation efficacy.²⁸ In ATF4, uORF1 is 6 nt long, and uORF2 is 12 nt long. uORF1 and 2 are located within the 5′ UTR and do not overlap the CDS. In contrast, uORF3 is 180 nt long and overlaps with the ATF4 CDS. In normal, unstressed conditions, ribosomes first form on uORFs 1 or 2 and complete this short translation. Because ternary complexes are abundant under non-stressed conditions, the small ribosomal subunit continues scanning, and it usually finds and translates uORF3. Because uORF3 overlaps with the CDS, this process often “skips” over ATF4 and, consequently, ATF4 protein is rarely produced.^25,26,27 However, under the ISR, there are not enough ternary complexes to quickly re-initiate translation at uORF3, so scanning instead continues to the CDS ORF, at which point enough time has passed to allow a new ternary complex to bind. Thus, under the ISR, full-length ATF4 is translated more frequently.^25,26,27 These uORFs are highly conserved in vertebrates, indicating their importance in ATF4 regulation.²⁶

ATF4 has four gene duplications in humans (annotated as pseudogenes ATF4P1–P4) that have not been characterized. While rare, gene duplications can maintain or even gain functionality.^29,30,31 One class is retroduplications (retrocopies), which are reverse transcribed (via transposon activity^32,33) from the mRNA of “parent” genes and inserted back into the genome. When retrocopies are found to be functional, they are labeled as retrogenes,³⁴ such as HAPSTR2³⁵ (MIM: 301145),³ retroCHMP3,³⁶ PTENP1^37,38 (MIM: 613531),³ and PGK2^39,40,41 (MIM: 172270).³ To be considered a functional retrogene, these previous studies have shown the gene is transcribed and causes a biological outcome through its mRNA or protein. This could be with or without interaction with its original parent gene. If a gene duplication is functional and beneficial, evolution can maintain it in the genome.²⁹ One major source of evolutionary pressure on the ISR comes from viruses.^42,43 For example, hepatitis B hijacks ISR signaling proteins in order to prevent downstream apoptosis.⁴⁴ Adenovirus produces decoy double-stranded RNAs (dsRNAs) that sequester the ISR kinase PKR without activating it, thereby inhibiting its anti-viral activity.^42,45 These pressures have led to evolutionary selection in ISR machinery, such as in PKR, where it has evolved to evade viral mimicry.⁴⁶ Given that ATF4 is the critical downstream transcription factor of the ISR, such evolutionary pressure could select for these ATF4 duplications if they influence ISR function.

Here, we demonstrate that the four ATF4 duplications are retrogenes. These retrogenes span evolutionary events across at least 37 million years in vertebrates. There is evidence of divergent and convergent evolution, with one ATF4 retrogene truncation arising independently in at least three different primate species. Human retrogenes are expressed under normal conditions and ISR activation, are degraded by the same machinery as ATF4, and can alter the expression of ATF4 target genes. Given their evolutionary history and function in cells, these retrogenes are likely important in the highly conserved ISR pathway in vertebrates.

Materials and methods

Databases accessed for gene queries, expression, and protein folding

We used the University of California Santa Cruz (UCSC) genome browser to identify retrocopies of ATF4 by using BLAT⁴⁷ to search using the species-specific endogenous parental ATF4 gene. We identified retrocopies, with a BLAST-like Alignment Tool (BLAT) matching score of at least 100, by the presence of a poly(A) tail and/or by a lack of introns. We noted cases of uncertainty, such as with short contigs or poor (e.g., “N”-rich) sequencing around potential retrocopies. In species where ATF4 was not annotated, we used a closely related species (when possible) to predict the parental ATF4 location in the queried species then used that section to search that species for potential retrocopies. We also noted any duplications of ATF4 in Table S2, but only for annotation purposes, and we do not discuss them here.

We annotated exons and introns of human ATF4 by referencing the Ensembl genome database.⁴⁸ We specifically used the canonical ATF4 transcript ENST00000674920.3 (ATF4-206). We referenced the UniProt database to confirm post-translational modification sites.⁴⁹ We used MacVector (v.18.2.5.) to help annotate and visualize DNA sequences.

We predicted protein structures using AlphaFold 3 using default settings.⁵⁰ We downloaded .CIF files of each protein and oriented them using PyMOL (The PyMOL Molecular Graphics System, Version 2.2.6, Schrödinger). To help visualize their structures, we used PyMOL to color specific amino acids for Figure S1.

To create our species tree, we first generated our .tre file using the “rotl” package in RStudio (version 2023.12.1+402, RStudio Team (2020). RStudio: Integrated Development for R. RStudio, Boston, MA; http://www.rstudio.com/). We uploaded this file to the interactive tree of life⁵¹ to correct minor labeling errors and to sort species.

Cells

RPE-1 cells (hTERT RPE-1, CRL-4000, American Type Culture Collection [ATCC]) were a gift from Drs. Michael Boland and David Goldstein (Columbia University). HEK293T cells were gift from Dr. Charles Murtaugh (University of Utah).

We maintained RPE-1 cells in DMEM/F-12 media (Gibco, Thermo Fisher catalog [cat.] 11320033) supplemented with 10% fetal bovine serum (FBS) (Gibco, Thermo Fisher cat. 26140079) and 0.01 mg/mL hygromycin B (Sigma-Aldrich, cat. H3274, diluted in water). We maintained HEK293T and HeLa cells in DMEM with pyruvate (Gibco, Thermo Fisher cat. 11995065) supplemented with 10% FBS (Gibco, Thermo Fisher cat. 26140079). We grew both cells in cell incubators set to 37°C and 5% CO₂.

Determining ATF4 retrocopy expression via database and qPCR

We used the Genotype-Tissue Expression (GTEx) Project to determine expression levels of ATF4 retrocopies (ATF4P1–P4 are annotated on GTEx). We queried the Oxford Nanopore Technologies long-read data (GENCODEv26-aligned) for annotated transcripts of ATF4P1 (ENST00000443279), ATF4P2 (ENST00000601944), ATF4P3 (ENST00000433309), ATF4P4 (ENST00000438498), PTENP1 (ENST00000447117), PGK2 (ENST00000304801, ENST00001129597), and HAPSTR2 (ENST00000453380). GTEx was supported by the Common Fund of the Office of the Director of the National Institutes of Health and by NCI, NHGRI, NHLBI, NIDA, NIMH, and NINDS. The data used for our analyses here were obtained from the GTEx Portal on under the dbGaP accession phs000424.v10.p2 on 02/17/2025 and the long-read data on 06/20/2026. We used the UCSC genome browser⁴⁷ to access tracks containing ENCODE data for chromatin marker and candidate cis-regulatory element (cCRE) data,⁵² transcription start site markers using CAGE data,⁵³ and JASPAR motifs⁵⁴ of common ISR and unfolded protein response (UPR) transcription factors (ATF4, ATF3, CEBPA, CEBPA/DDIT3, CEBPB, XBP1, and ATF6).

To design primers for ATF4/retrocopy expression, we first aligned ATF4 and its retrocopies using ClustalOmega.⁵⁵ As we needed to use primers specific to each copy, without binding each other, we focused on DNA segments with low similarity between each retrocopy. When such areas were small, we used that low similarity area as the 3′ end of each primer, as this is the most important for primer binding/synthesis⁵⁶ (see Table 1). We used qPCR primers for ATF4 target genes from PrimerBank⁵⁷ or designed them using Primer3Plus.⁵⁸ Primers were ordered from IDT DNA (Coralville, IA).

Table 1.

List of primers used

Name	Nucleotide sequence used	Purpose
Primers for Gateway cloning
ATF4 forward	CTCTCTCCTTTCGTGAGGCC	for amplifying region around ATF4 for gel extraction
ATF4 reverse	TGTTCAGGTAACAAGTGAAACCT
ATF4P3 forward	CCGGGCCACTACCTCTTACT	for amplifying region around ATF4P3 for gel extraction
ATF4P3 reverse	GGTGCAATGGCTCACAACTG
ATF4P4 forward	CAAGATCCCCGATGAGGAGC	for amplifying region around ATF4P4 for gel extraction
ATF4P4 reverse	CTCAACTTCCTGGGCTCGAG
ATF4P1/2 forward	CCCCCAAGTCTTAACCCATT	for amplifying region around ATF4P1/2 for gel extraction
ATF4P1/2 reverse	AGCGTAGTGCAGAAGGGAAA
ATF4/ATF4P3/ATF4P4 forward	CACCGCCACCATGACCGAAATGAGC	for creating the Gateway cloning insert and fixing Kozak
ATF4/ATF4P3 reverse	GGGGACCCTTTTCTTCC
ATF4P4 reverse	CCGTCTGGGGGTCTCCT
ATF4P1/2, Long X forward	CACCGCCACCATGTCCCCCCTTGA
ATF4P1/2, Long X reverse	AAGGAATAATCTGGAGTG
ATF4P1/2, Small X	CACCGCCACCATGACCGAGATGAG CTTCCTGAGCAGCGAGGTGTTGGTG GGGGCT	synthesized directly (IDT) for use in Gateway cloning
Primers to differentiate ATF4 and retrocopy transcripts
ATF4 forward	TGACCACGTTGGATGACACT	qPCR
ATF4 reverse	GGGAAAGGGGAAGAGGTTGT	qPCR
ATF4P3 forward	CCCTAGTCCAGGAGACTAATAAGGT	qPCR
ATF4P3 reverse	AGGGATCATGGCAACGTAAG	qPCR
ATF4P4 forward	CTGGTGAGTGCAAAGAGCTG	qPCR
ATF4P4 reverse	ACACAGCTACAGCAGAGG	qPCR
ATF4P1/2 forward	GGCTCTCCCTCTACCTCCAG	qPCR
ATF4P1/2 reverse	GCTGTTTGGTTTTGCTCCAT	qPCR
ATF4 target genes
ATF3 forward	CCTCTGCGCTGGAATCAGTC	qPCR PrimerBank ID: 346223459c1
ATF3 reverse	TTCTTTCTCGTCGCCTCTTTTT
TRIB3 forward	AAGCGGTTGGAGTTGGATGAC	qPCR PrimerBank ID: 41327717c1
TRIB3 reverse	CACGATCTGGAGCAGTAGGTG
XPOT forward	CAAGTCTTCGCCTTGCTTTTTG	qPCR PrimerBank ID: 317108146c2
XPOT reverse	CCCTTGGATTTAGGTCCACTACT
SARS1 forward	CAGCCCTCATCCGAGAGAC	qPCR PrimerBank ID: 300795157c1
SARS1 reverse	TCTGCCCGAAATCTACATCGT
ERO1A forward	GGCTGGGGATTCTTGTTTGG	qPCR PrimerBank ID: 7657068c1
ERO1A reverse	AGTAACCACTAACCTGGCAGA
GAPDH forward	ACAACTTTGGTATCGTGGAAGG	qPCR PrimerBank ID: 378404907c2
GAPDH reverse	GCCATCACGCCACAGTTTC

To generate RNA, we first grew cells in six-well plates until ∼80% confluent. We then added TRIzol reagent (15596026, Invitrogen) to wells and scraped cell lysates into microcentrifuge tubes. We vortexed tubes for at least 15 s, and we stored them at −80°C for at least 24 h before further processing. We used the DirectZol kit (Zymo Research, cat. R2052) to extract RNA and the ProtoScript II First Strand cDNA Synthesis Kit (NEB, cat. E6560L) to convert RNA to cDNA. We analyzed gene expression using ATF4/retrocopy-specific primers (Table 1), PowerUp SYBR Green Master Mix (Fisher Scientific, cat. A25918), and QuantStudio 3 for analysis (Thermo Fisher).

dN/dS and ORF decay modeling

To determine the non-synonymous to synonymous (dN/dS) ratios of retrocopies, we used the KaKs_Calculator 2.0.⁵⁹ We used ORFs of human ATF4P1–P4 compared to either human ATF4 or our reconstructed ancestral versions of ATF4, and we set the parameters to use the Yang and Nielson (YN) methodology.⁶⁰

We aligned ATF4 sequences using Multiple Sequence Alignment by Log-Expectation (MUSCLE 3.8)⁶¹ of either apes or Catarrhini to represent ancestral copies of species 14.8 million years ago (mya) and 37.5 mya, respectively. We then created estimates of ancestral ATF4 by inputting these MUSCLE-aligned sequences into EMBOSS Cons⁶¹ to create consensus sequences using minimal manual annotation (see Table S3).

We used the Python script “mutator” to simulate the decay of ATF4 retrocopies under neutral evolution, and “orf_scanner” to count ORFs.^62,63 We modified the code of the mutator script to use a Poisson distribution (rather than Gaussian) to better model the relatively short evolutionary times of these retrocopies (see new code at https://github.com/hansmdalton/pedestal_fork). As we are most focused on the human retrocopies, we used human substitution, insertion, and deletion rates as determined from the literature^63,64: substitution = 1.16e−8, insertion = 2e−10, deletion = 5.5e−10. We used a generation time of 25 years for ATF4P3 and ATF4P4 to represent apes,⁶⁵ and we used a generation time of 10 years for ATF4P1/P2 to represent monkeys within Catarrhini that still maintain this retrocopy.^66,67,68 We derived the estimates of 14.8 mya divergence of orangutans and 37.5 mya divergence of Platyrrhini from the literature.^{69,70,71,72,73} We simulated ORF decay in 10,000 replicates every 50,000 years for 50 million years.

Starting from our estimated ancestral copy (Table S3) at 14.8 million years (my) or 37.5 my, we determined probabilities of each lineage-specific branch terminus retaining the intact retrocopy ORF being tested (e.g., full 351 amino acids [aa] for ATF4P3, or >130 aa for ATF4P1/2 Long X). To prevent double-counting branches, we sectioned the phylogenetic branches into non-overlapping segments for testing species probabilities. To determine the overall probability, we then multiplied these probabilities together for each ORF. For amino acid changes in the DNA-binding domain of ATF4P3, we used Biopython⁷⁴ to translate each permutation to look for any non-synonymous changes or indels. To look for any nucleotide changes in ATF4P2, we ran the same mutator script as above, but then we counted (using Python) the presence of any nucleotide change or indels. We then plotted all of these results using Rstudio⁷⁵ and annotated graphs in Adobe Illustrator.

Target-site duplication identification

The presence of poly(A) sequences and the lack of introns within the sequences of retrocopies indicates that they likely arose from retrotransposons.^34,76 This is typically driven by long interspersed element-1 LINE-1) elements, which leave a duplicated sequence artifact flanking each retrocopy.^34,76 We manually assessed the ∼100 bp downstream of the poly(A) sites following the 3′ UTR (as determined from parental ATF4). We used this sequence to identify identical sequences upstream of the 5′ UTR. When we identified a likely target site duplication (TSD) site, we checked this TSD against the rest of the species for this same sequence, refining the sequence as needed.

ISR stress treatments

We grew wild-type (WT) RPE-1 cells to ∼80% confluency for the experimental treatment day. We treated cells with either 0.1 μM thapsigargin (Tg) (Sigma cat. S7400) for 8 h (vehicle = DMSO) or 10 μM sodium arsenite (As) (Sigma cat. T9033) for 6 h (vehicle = water). We plated 3–5 biological replicates of WT RPE-1 cells in six-well plates (As) or 10-cm plates (Tg). After treatment, we collected cells in TRIzol and performed qPCR analysis as above.

Lentivirus production and transfection

As ATF4 retrocopies have high similarity to each other at the nucleotide level, we employed two sequential PCR reactions to specifically amplify each retrocopy for cloning (Table 1). Due to differences in introns, the cloned sequence for ATF4 was derived from RNA (cDNA), while the retrocopies were derived from genomic DNA. We derived genomic DNA from HeLa cells and RNA from human fibroblasts⁷⁷ (gifts of Dr. Nels Elde, University of Utah). We converted RNA to cDNA using the ProtoScript II First Strand cDNA Synthesis Kit (NEB, cat. E6560).

First, we used primers ∼100–500 bp up- and downstream of each ATF4 retrocopy to amplify that region. We then gel extracted (QIAquick Gel Extraction Kit, cat. 28704) this product to purify it from other genomic DNA. We used this as the product for further Gateway cloning using primers specific to each gene. Because the Small X insert was so short (55 bp), we instead ordered this entire sequence from IDT DNA (Coralville, IA) ready for further Gateway cloning. We sequenced all inserts and final products to ensure these were correct (Sanger sequencing via the University of Utah HSC DNA Sequencing Core).

For all copies of ATF4 (including the parental gene ATF4 itself), we observed no protein expression when we directly cloned either the CDS alone or the 5′ UTR plus CDS (performed in both RPE-1 and HeLa cells). To help resolve this, we designed primers to modify the native Kozak sequence⁷⁸ upstream of the first codon while maintaining the coding sequence integrity. Using these modified Kozak sequences (Table 2), we observed protein expression of all clones.

Table 2.

Kozak sequences in ATF4 and retrocopies

Gene	Endogenous Kozak	Modified Kozak
ATF4	CGCAACATGACC	GCCACCATGACC
ATF4P3	CGCAACATGACC	GCCACCATGACC
ATF4P4	CGCAACATGACC	GCCACCATGACC
ATF4P1/2, Small X	TGCACCATGACC	GCCACCATGACC
ATF4P1/2, Long X	GGCTTGATGTCC	GCCACCATGTCC

Native ATF4/retrocopy Kozak sequences required modification to induce correct protein expression in our cells.

We used the Gateway cloning system (Invitrogen, Thermo Fisher cat. K240020) to generate the entry vectors used for creating ATF4 overexpression (OE) cell lines. We then used LR clonase (Invitrogen, Thermo Fisher cat. 11791020) to transfer our clones into a lentiviral plasmid containing a cytomegalovirus (CMV) promoter to drive the OE and a C-terminal V5 tag (Addgene, pLenti6.2-ccdB-3xFLAG-V5, Plasmid #87071). We used a C-terminal tag to avoid future conflicts with 5′ UTR/Kozak/uORF analyses. We then generated lentivirus particles in HEK293T cells via FuGENE 6 Transfection Reagent (Promega cat. E2693) using psPAX2 and pCAG-VSVG packaging plasmids (Addgene plasmid #12259, and Addgene #35616, respectively; gifts of Dr. Charles Murtaugh, University of Utah).

We incubated RPE-1 cells with our generated lentivirus particles along with 8-μg/mL polybrene (gift from Dr. Charles Murtaugh, University of Utah). After 48 h, we selected for transfected polyclonal cell populations using 10-μg/mL blasticidin (Gibco, Thermo Fisher cat. R21001). We then selected for monoclonal populations by picking cell colonies to individual plates where we continued blasticidin selection and ensured single-colony expansion. We then performed western blot analyses (see below) to check for accurate and robust protein production of each cloned retrocopy.

Proteasome inhibition

To induce proteasome inhibition, we first grew RPE-1 cells to 80%–90% confluency. We then replaced their regular media with media containing either 0 or 100 nM bortezomib (Btz) (both with 0.1% DMSO). After 6 h of exposure to Btz, we collected cell protein for western blot or cells in TRIzol for RNA extraction and downstream qPCR analysis.

Western blots

To collect protein, we washed cells with PBS, then we treated them with RIPA buffer supplemented with 1× Protease Inhibitor Cocktail (Cell Signaling Technologies, cat. #5871). We lysed cells for 30 min at 4°C under agitation, then centrifuged them at 14,000 × g for 20 min. We extracted supernatants and determined protein concentration using the bicinchoninic acid (BCA) assay (Pierce BCA Protein Assay Kit, cat. 23227). We then froze samples at −80°C for later use.

To denature proteins, we added 6× loading buffer supplemented with β-mercaptoethanol (Sigma, cat. M6250) and heated them at 95°C for 5 min. We resolved proteins using a 12% Criterion XT Bis-Tris Protein Gel (Bio-Rad #3450118) and XT MES Running Buffer (Bio-Rad, cat. #1610789). We detected proteins using anti-V5 (Cell Signaling Technologies, cat. 13202, 1:1,000), anti-ATF4 (Proteintech, cat. 10835-1-AP, 1:800), anti-beta-actin (Santa Cruz Biotechnology, cat. sc-47778, 1:200) primary antibodies, and anti-rabbit and anti-mouse secondaries (“DyLight,” Cell Signaling Technology cat. 5151 and 5470, respectively, 1:10,000). We used the Precision Plus Protein All Blue Standards for our ladder (Bio-Rad, cat. #1610373). We visualized blots using an Odyssey DLx imager (LI-COR).

Statistics and graphs

We performed statistics and created figures using GraphPad Prism (version 10 for Windows, GraphPad Software, Boston, MA, USA; www.graphpad.com). We also utilized RStudio (version 2023.12.1+402, RStudio Team (2020). RStudio: Integrated Development for R. RStudio, Boston, MA; URL http://www.rstudio.com/).

Results

Humans have four retrocopies of ATF4

There are four annotated ATF4 duplications: ATF4P1, ATF4P2, ATF4P3, and ATF4P4. All of these duplications lack introns and end in truncated poly(A) tails—key characteristics of retrocopies^34,79 (Figure 1B; Table S1). Retrocopies often originate from LINE-1 retrotransposons in mammals,^32,33 and these leave short TSDs flanking each gene.^33,80 We examined the region around each retrocopy and found evidence for TSDs in all of them (Table S1). Thus, ATF4P1–4 are retrocopies that emerged through LINE-1 retrotransposons.

We compared the nucleotide sequence and putative protein translation of each retrocopy to its parent gene, ATF4 (Figures 1B and 1C, AlphaFold predicted structures in Figure S1).

ATF4P2 and ATF4P1

ATF4P2 and ATF4P1 are located on the X chromosome. Here, there is a larger ape-specific chromosomal reverse tandem duplication event.⁸¹ The proximal border of this duplication is in the middle of the gene, IKBKG (MIM: 300248),³ resulting in the distal end containing a truncated duplication (the pseudogene, IKBKGP1). As ATF4P1 is closer to the more recent IKBKGP1, it must be duplicated from ATF4P2, which is closer to IKBKG. This duplicated region is highly invariant. As such, human ATF4P2 and ATF4P1 are 100% identical at the nucleotide level from the first uORF to the poly(A) tail (Table S1), and we refer to them together as ATF4P1/2 throughout. While ATF4P1 was not directly derived from a retroduplication event, it nevertheless was duplicated from the retrocopy ATF4P2, so we refer to it as a retrocopy/retrogene throughout.

ATF4P1/2 maintain uORF1 and uORF2, but they have lost uORF3 due to a new stop codon (Figure 1B). This very early premature stop codon would result in a putative 15-aa product. However, ATF4 and ATF4P1/2 have a secondary in-frame start site after 48 bp. This is immediately following the premature stop in ATF4P1/2 and encodes a 170-aa product ending with multiple insertions that result in amino acid changes and a premature stop codon after the p300 interaction domain (Figures 1C and S1). Going forward, we will refer to these products as Small X (15 aa) and Long X (170 aa), respectively. At 4.5% query coverage (15 aa), ATF4P1/2 Small X has 87.5% identity with ATF4. At 45.5% query coverage (170 aa), ATF4P1/2 Long X has 69.1% protein identity with ATF4 (Figure 1C).

ATF4P3

ATF4P3 is located on chromosome (Chr.) 17 in an intron of the RNF157 gene (Figure S2) (MIM: 621044).³ Because ATF4P3 was inserted in the same strand orientation as RNF157, it might utilize the RNF157 promoter or its regulatory sites. At the nucleotide level, all of its upstream ORFs (uORFs) are intact (Figure 1B), suggesting similar translational regulation to ATF4. ATF4P3 encodes for a putative full-length ATF4 protein with zero nonsense changes. At 100% query coverage, ATF4P3 has 95.7% protein identity with ATF4 (Figure 1C). Notably, the majority of its missense changes (12/15) occur in the first half of the gene, with fewer in the second half (3/15), and none are present in the basic leucine zipper domains (Figure 1C). Because the ATF4 leucine zipper domain contains a nuclear localization signal (KKLKK⁸²), it is likely that ATF4P3 retains its nuclear localization and DNA-binding ability. Overall, ATF4P3 is nearly completely intact (∼95%) at both the nucleotide (Figure 1B) and protein level (Figures 1C and S1).

ATF4P4

ATF4P4 is located on Chr. 11. Its uORFs 1 and 2 are extended due to changes in their stop codons (Figure 1B). uORF1 now overlaps both uORF2 and uORF3. This uORF1 overlap likely results in inhibiting normal ribosomal binding to uORF3, which usually causes the CDS to be skipped,^25,26,27 and this likely would release some uORF inhibition of the ATF4P4 CDS. ATF4P4 contains a premature stop codon that occurs after the p300 interaction domain, and this results in a 138-aa product. Despite this truncation, ATF4P4 is nearly completely intact at the nucleotide level (∼95% identity). While ATF4P4 lacks both degradation domains and its DNA-binding domain at the protein level, its p300 interaction domain remains intact (Figures 1C and S1). This suggests that ATF4P4 would lack targeted degradation while still maintaining its ability to bind p300. Within the 138-aa product (39.2% query coverage), ATF4P4 has 90.6% amino acid identity with ATF4 (Figure 1C).

Several sites of post-translational modification in ATF4 are changed in these retrocopies. ATF4P1/2 Long X has differences in two phosphorylation sites (T114A, T119S). These are two of four phosphorylated threonine residues, along with T107 and T115, important for protein stability.⁸³ Mutation of all four of these threonine residues causes a strong increase in ATF4 stability. It is possible that the T119S site is still phosphorylated (as it is a serine). However, the T114A change should remove the phosphorylation site, thereby increasing the stability of ATF4P1/2 Long X protein. ATF4P3 has a change in a proline residue in the ODD domain (P162L), one of five prolines important for proper ODD domain-induced degradation.²⁰ This may lead to impaired degradation of ATF4P3 under normoxia, as this is observed for the same mutation in ATF4.²⁰ ATF4P4 has a change in a phosphorylation site (S58F), and mutations in this same site in ATF4 lead to reduced transcription factor activity.⁸⁴ Given that ATF4P4 is missing its DNA-binding domain, it is unclear what effect this amino acid difference has on ATF4P4. Overall, ATF4P1–4 are relatively intact and have high nucleotide and protein identity with ATF4.

Conserved and independent retrocopies of ATF4 are common in mammals

If ATF4 retrocopies are well-conserved across many species, it suggests they are under non-neutral selection. As such, we examined the evolutionary history of each ATF4 retrocopy. We analyzed ATF4 retrocopies through manual BLAT queries of 97 species, from yeast to humans, derived from UCSC-curated genome assemblies (Table S2; UCSC genome browser⁴⁷).

ATF4P2 and ATF4P1

Retrotransposition of ATF4P2 occurred in a common ancestor of apes and Cercopithecidae, the monkeys of Africa and Asia, approximately 37.5 mya (Figures 2A and S3; Table S2). In the six Cercopithecidae species we examined, ATF4P2 contains the same ancestral early stop for its Long X protein, which results in a putative 131-aa product (Figure 2A). This is different from what is observed in apes, where they range from 85 to 170 aa in length. The only intact protein domain in ATF4P2 Long X is the p300-binding domain.

Figure 2.

Catarrhini species tree comparisons of ATF4 retrocopy and duplicate retrocopy proteins

(A) ATF4P2 arose ∼37.5 mya in a common ancestor of Catarrhini (apes and Cercopithecidae, the monkeys of Africa and Asia). There is a short Small X peptide (ending at the blue bar) followed by a longer ORF (Long X) that resembles the truncated ATF4P4. Note that orangutan only lost the start site of ATF4P2 Long X, but they still maintain the highly truncated ATF4P2 Small X (Table S1). Cercopithecidae all contain the same truncating stop site resulting in a 131-aa product.

(B) ATF4P1 resulted from a duplication of ATF4P2 ∼14.8 mya. In humans, it is an exact copy even at the nucleotide level, while chimps, bonobos, and gorillas have various levels of truncation. Chimps and bonobos have a highly truncated Long X in ATF4P1, but they maintain it in ATF4P2 (see A). Bonobos also have an additional duplication of ATF4P2 with its own unique truncation.

(C) ATF4P3 arose in a common ancestor of apes ∼14.8 mya, and it is fully intact at the amino acid level in each species we examined.

(D) ATF4P4 arose in a common ancestor of apes ∼14.8 mya, and it carries an early stop that truncates the protein to 137 aa. There are two independent ATF4 retrocopies in orangutans and gibbons with similar truncations to ATF4P4 but are non-syntenic.

(E) Orangutans have three independent (non-syntenic) retrocopies of ATF4. Two are genomically truncated (Chrs. 15 and 17), and Chr. 2A carries two early stop codons. This Chr.17 retrocopy is likely a partial duplication of the ATF4P4-like orangutan retrocopy in (D) given its proximity.

(F) Timeline of approximate emergence of new retrocopies and their duplications. AA, amino acid; mya, million years ago.

ATF4P1 is an ape-specific duplication of the ATF4P2 retrocopy. It likely arose ∼14.8 mya because it is present in humans, chimps, bonobos, and gorillas but not in orangutans or gibbons^{69,70,71,72,73}(Figure 2B). Human and gorilla ATF4P1 genes have a putative Long X protein of 170 and 160 aa, respectively. In chimps and bonobos, their ATF4P1 Long X sequence is truncated with an early stop resulting in a putative 14-aa-long product (different from Small X protein). Bonobos also have an additional retrocopy duplication in tandem with its ATF4P2 and ATF4P1 (Figures 2B and S3; Table S1), resulting in three total copies. This additional bonobo duplication has 98% nucleotide identity with ATF4P2, and only 91% with ATF4P1, so it most likely also arose from ATF4P2 sometime in the last ∼2 my.

Humans, chimps, and bonobos have a 30- to 34-bp internal duplication in ATF4P2 and ATF4P1 between the p300 interaction and ODD domains (Table S1). In humans, this is exactly 30 bp (and thereby in-frame) in ATF4P2 and ATF4P1, causing its Long X protein to be 10 aa longer. However, this segment is 32 bp long in chimp ATF4P2 and results in an early stop. The duplication likely has no impact on bonobo ATF4P2 or ATF4P1 or its additional duplication, or chimp ATF4P1, as other indels cause earlier stop codons. Unexpectedly, this ∼30-bp duplication is not present in gorilla ATF4P2 or ATF4P1 (2019 and 2016 genome assemblies via UCSC genome browser). As it is not present in orangutans or Cercopithecidae either, it is unlikely due to a deletion event in gorillas, and it is instead likely a new duplication arising after the gorilla split. However, because it is present in ATF4P2 and ATF4P1 (and the additional duplication in bonobos), this suggests the unlikely occurrence of two independent duplication events of this segment in ATF4P2 and ATF4P1 in the common ancestor of humans, chimps, and bonobos.

Given the divergence time of 14.8 my, and a human mutation rate of ∼1.1 × 10⁻⁸/generation,⁸⁵ we expected to find ∼9 nucleotide changes between ATF4P2 and ATF4P1. However, human ATF4P2 is identical to ATF4P1 at the nucleotide level. As described above, this region of the genome is highly invariant after a larger genomic duplication event.⁸¹ One possibility is that selection on something in this region is maintaining identity between the duplicates. Alternatively, given the tandem duplication and 100% identity, this could be due to interlocus gene conversion, where one DNA sequence is replaced by a nearby paralog.⁸⁶ However, compared to other apes examined, human ATF4P2 and ATF4P1 are the only unchanged pair. Chimps, bonobos, and gorillas all have nucleotide differences between ATF4P2 and ATF4P1, and the third copy in bonobos is also different from its ATF4P2 and ATF4P1 (Figures 2A and 2B; Table S1). Thus, such interlocus gene conversion would likely be specific to humans.

Of note, chimps and bonobos are very closely related species,⁸⁷ yet their ATF4P2 and ATF4P1 retrocopies contain several nucleotide and protein differences relative to each other. For example, while both have a putative truncated 14-aa-long product in ATF4P1, these were independent, convergent events (22Gdel in bonobos and 5Cdel in chimps). In addition, ATF4P2 is truncated at a different amino acid length due to various nucleotide differences between chimps and bonobos (Figure 2A; Tables S1 and S2).

ATF4P3

ATF4P3 arose in a common ancestor of humans, chimps, bonobos, and gorillas (Figures 2C and S3; Tables S1 and S2). In each species, ATF4P3 has a full protein-length ORF, although each species has differences in amino acid changes (Figure S4). Similar to humans, there are no changes to the DNA-binding domain of ATF4P3 in chimps compared to ATF4 (Figure S4). Bonobo ATF4P3 has one change (K315Q) in its DNA-binding domain, while gorilla ATF4P3 has two changes (R296C, N317S). We queried how these amino acid changes might impact protein function by using PolyPhen⁸⁸ and SIFT.⁸⁹ K315Q in bonobo is predicted to be “benign” and “tolerated” by PolyPhen and SIFT, respectively. For the gorilla changes, both are predicted as “probably damaging” and “affecting protein function” by PolyPhen and SIFT, respectively. Given its sequence is identical in humans and chimps, and that the bonobo change is predicted as benign/tolerated, this domain is likely functional if it is translated.

ATF4P4

Similar to ATF4P3, ATF4P4 arose in a common ancestor of humans, chimps, bonobos, and gorillas (Figures 2D and S3; Tables S1 and S2). In all species, ATF4P4 retrocopies are truncated by the same early stop codon, resulting in a 137-aa product. We also found two similarly truncated (but independent/non-syntenic) “ATF4P4-like” retrocopies in both orangutans and gibbons, creating 152- and 179-aa products, respectively (Figure 2D). Thus, this truncated variant of an ATF4 retrocopy has arisen at least three independent times. Similar to ATF4P1/2, these truncated retrocopies only contain an intact p300-binding domain. Taken together, humans, bonobos, and gorillas all have three intact ATF4 retrocopies containing only the p300-binding domain (chimps have two). If expressed, this suggests some benefit to having extra copies of a p300-binding peptide.

Independent orangutan ATF4 retrocopies

Orangutans also contain three additional ATF4 retrocopies that are non-syntenic to any other retrocopy we observed (Figures 2E and S3; Tables S1 and S2). These retrocopies are unlike ATF4P1–P4. The Chr. 2A retrocopy has multiple internal start and stop sites, resulting in potentially smaller peptides. This results in one putative peptide that resembles ATF4P1/2 Small X, but none that contain as much of the p300 domain as ATF4P1/2 Long X. The retrocopy on Chr. 15 lacks the p300 domain due to a chromosomal truncation of DNA (unlike the previous protein-level truncations). The retrocopy on Chr. 17 may be a duplication of the ATF4P4-like retrocopy because it is only ∼1 Mb away. In addition, it is only a truncated copy of this gene with only one flanking tandem duplication site (TDS) rather than the two sites expected if it were another LINE-1 element-derived retrocopy. However, it is nevertheless possible that this arose independently and later lost one original flanking TDS.

Approximate evolutionary timeline

The first retrotransposition event occurred ∼37 mya (Figure 2F). Later, approximately 14.8 mya,^{69,70,71,72,73} the common ancestor of humans, chimps, bonobos, and gorillas had three independent duplications of ATF4—two retroduplications (ATF4P3, ATF4P4) and one duplication of a retrocopy (ATF4P1 duplicated from ATF4P2).⁸¹ This suggests a period of stronger selection for ATF4 retrocopies ∼14.8 mya (Figure 2F). In apes, additional ATF4 retrocopies have arisen since then in orangutans (∼14.8 mya or earlier) and bonobos (∼2 mya or earlier) (Figure 2F).

Independent ATF4 retrocopies arose in many other mammals

There are many genomically intact (e.g., not heavily degraded or fragmented) retrocopies of ATF4 throughout mammals (Tables 3 and S2). In addition to the extra retrocopies in orangutan and gibbon genomes (Tables S1 and S2; Figure S3), there are multiple non-syntenic ATF4 retrocopies in several other primates, including the Philippine tarsier (Carlito syrichta, four retrocopies), the mouse lemur (Microcebus murinus, one retrocopy), the bush baby (Otolemur garnetti, three retrocopies), and the flying lemur (Galeopterus variegatus, three retrocopies) (Table 3; Figure S3; Table S2). As none of these retrocopies are syntenic, there are at least 12 additional independent retroduplications of ATF4 in primates, bringing the total we observed to 18 (Figure S3).

Table 3.

Summary table of mammalian ATF4 retrocopies

Common name	Species name	Total retrocopies
Human	Homo sapiens	4
Chimp	Pan troglodytes	4
Bonobo	Pan paniscus	4
Gorilla	Gorilla gorilla gorilla	4
Orangutan	Pongo pygmaeus abelii	5
Gibbon	Nomascus leucogenys	2
Green monkey	Chlorocebus sabaeus	1
Crab-eating macaque	Macaca fascicularis	1
Rhesus macaque	Macaca mulatta	1
Baboon (hamadryas)	Papio hamadryas	1
Proboscis monkey	Nasalis larvatus	1
Golden snub-nosed monkey	Rhinopithecus roxellana	1
Tarsier	Tarsius syrichta (C. syrichta)	4
Mouse lemur	M. murinus	1
Bush baby	O. garnettii	3
Malayan flying lemur	G. variegatus	3
Chinese hamster	Cricetulus griseus	1
Guinea pig	Cavia porcellus	1
Squirrel	Spermophilus tridecemlineatus (Ictidomys tridecemlineatus)	1
Tree shrew	Tupaia belangeri	1
Bison	B. bison	1
Cow	B. taurus	1
Sheep	O. aries	1
Pig	Sus scrofa	1
Megabat	Pteropus vampyrus	1
Dog	Canis lupus familiaris	1
Hedgehog	E. europaeus	3
Tenrec	Echinops telfairi	1
Armadillo	D. novemcinctus	4

Mammals in which at least one full-length or truncated retrocopy were found. See Table S2 for details.

We also found evidence of independent retrocopies of ATF4 in 13 other mammals (Table 3; Figure S3; Table S2). Most of these species only had a single retrocopy of ATF4, with the exception of the European hedgehog (Erinaceus europaeus, three retrocopies) and the nine-banded armadillo (Dasypus novemcinctus, four retrocopies). Of note, the Bovidae family has a syntenic retroduplication of ATF4 present in at least bison, cows, and sheep (Bison bison, Bos taurus, and Ovis aries) (Figure S3; Table S2). This retrocopy is inserted within an intron of the BANK1 gene (MIM: 610292),³ so it may utilize its nearby regulatory sequences (albeit, the retrocopy is truncated). We see no evidence of ATF4 retrocopies in birds, lizards, amphibians, fish, or any invertebrate species (using invertebrate orthologs; Table S2). This is unsurprising, as there are much lower rates of retrocopies in non-mammalian species.^90,91

Likelihood of retaining ATF4 retrocopies over time

There are an estimated ∼8,000 retrocopies in humans.^90,92 Because many retrocopies arise with no nearby regulatory components, with only <20% having an intact ORF, it is not uncommon for neutral evolution to eventually degrade retrocopies.^90,92 Supporting this, we found several ATF4 retrocopies in vertebrates that appear to have degraded over time (Table S2). In contrast, persistence of an intact gene over time can suggest non-neutral evolution of that gene. ATF4P1–4 are relatively evolutionarily young (within ∼37.5 my, for example, instead of a 90+ my divergence between humans and mice⁹³). This is also amplified by the long generation times in humans (∼25 years) and monkeys (∼10 years). Thus, it is possible that not enough time has passed to determine the likelihood of neutral evolution. However, given the multiple intact or conserved ATF4 retrocopies in multiple ape and monkey branches, we wanted to determine the likelihood of ATF4P1–P4 ORFs persisting over time to date. First, we determined the ratio of non-synonymous to synonymous substitution rates (dN/dS) for each retrocopy. We compared each human retrocopy to the parent gene, ATF4 (Figure S5A). All retrocopy ratios fell between 0.4 and 0.6, indicating moderate negative selection. To better match the ancestral version of ATF4 at the time of retrocopy emergence in primates, we created approximate ancestral versions of ATF4 for each retrocopy based on likely time of emergence (Table S3). Using these versions, the dN/dS was similar in ATF4P1/2 and ATF4P4 (∼0.4), while rising to ∼0.7 in ATF4P3 (Figure S5B). Overall, these dN/dS generally suggest negative selection on ATF4 retrocopies.

Next, we wanted to computationally model the likelihood of ATF4P-1P4 persistence. We used the same ancestral reconstructions of ATF4P1–P4, as in the dN/dS analysis, to model mutations over time (Table S3). We then modeled the likelihood of maintaining the size of each retrocopy’s longest ORF over 10,000 simulations of decay within each species. Both ATF4P1/2 and ATF4P4 have truncated ORFs, and this short size makes them statistically more likely to avoid degradation. For ATF4P1/2, there was an approximately 45% chance of its Long X ORF remaining intact among apes and monkeys (Figure S6A). For ATF4P4, with its much shorter evolutionary time, there was approximately a 97.6% chance of remaining intact (Figure S6B). ATF4P3, on the other hand, is completely intact after the same evolutionary time period. As such, our model found approximately only a 20% chance of remaining intact within apes (Figure S6B). While evolutionarily young, the chances do not favor keeping an intact ORF the size of ATF4P3 in four different apes. As ATF4 is a transcription factor, we also wanted to model the likelihood of the DNA-binding domain remaining completely intact in ATF4P3 (Figures 1C and S4). We determined the likelihood that this region remained intact with no non-synonymous changes or indels in humans and chimps. The DNA-binding domain was unchanged at the protein level approximately 12.7% of the time, indicating this is even less likely to occur (Figure S6C). This is also an underestimation, as our modeling does not take into account times when an upstream frameshift or nonsense mutation would remove the DNA-binding domain. Overall, our modeling suggests at least ATF4P3 may be under non-neutral evolution.

Lastly, we modeled the likelihood of human ATF4P2 and ATF4P1 remaining identical over 14.8 my. As mentioned, this area of the genome is highly invariant.⁸¹ As such, it is possible they are genetically linked to a different gene under non-neutral evolution or have undergone interlocus gene conversion.⁸⁶ Nevertheless, it is interesting to model because, as mentioned, unlike humans, the three other apes we examined do have changes between their ATF4P2 and ATF4P1 retrocopies (Figures 2A and 2B [chimps, bonobos, gorillas]; Table S1). We used the same simulation as above, but we instead determined how many times there were zero changes at the nucleotide level. After modeling just ∼10 my, all 10,000 simulated replicates had at least 1 nt change (Figure S6D). As this duplication likely occurred ∼14.8 mya,^{69,70,71,72,73} this means there is a <0.01% chance of ATF4P2 having no changes.

ATF4 retrocopies are expressed in tissues and human cells

To determine whether the ATF4 retrocopies are expressed in humans, we first queried the GTEx online portal for mRNA expression data of each retrocopy based on thousands of healthy individuals. ATF4 has median transcript per million (TPM) values ranging from 98 to 806 depending on tissue (Tables 4 and S4). In contrast, ATF4P2 and ATF4P1 have a maximum TPM value of 0.05, and the highest ATF4P3 and ATF4P4 median TPM values are 0.88 and 2.3, respectively (Table 4). GTEx uses a TPM cutoff of 0.1 to be considered expressed for use in their eQTL analysis,⁹⁵ and the EMBL-EBI Expression Atlas sets a TPM cutoff of 0.5 for a gene to be considered expressed.^96,97 ATF4P3 has a TPM of ≥0.1 in all tissues, with 13 at ≥0.5 TPM. ATF4P4 has a TPM of ≥0.1 in 49/52 tissues, with five at ≥0.5 TPM. However, ATF4P2 and ATF4P1 never reach these thresholds. Long-read sequencing may better resolve these four highly similar retrocopy transcripts, so we also queried GTEx Oxford Nanopore Technology (ONT) data for these ATF4 retrocopies.⁹⁴ We found evidence for ATF4P3 and ATF4P4 in these long-read data, but we could not find ATF4P2 or ATF4P1. However, because ATF4P2 and ATF4P1 have 100% identical transcripts, their inability to be uniquely identified might cause analysis pipelines to exclude them. Interestingly, we also queried the ONT data for several known retrogenes, PTENP1,^37,38 PGK2,^41,98 and HAPSTR2,³⁵ and could only find transcripts of HAPSTR2. Thus, these long-read data do not definitively rule out ATF4P1/2 expression but they at least lend more credence to ATF4P3 and ATF4P4 expression. Overall, this suggests a low level of documented basal expression for ATF4P3 and ATF4P4 with none for ATF4P2 and ATF4P1.

Table 4.

GTEx expression metrics in each tissue type

	ATF4	ATF4P1	ATF4P2	ATF4P3	ATF4P4
# ≥0.1 TPM	52	0	0	52	49
# ≥0.5 TPM	52	0	0	13	5
Minimum TPM	98.38	0	0	0.12	0.084
Maximum TPM	806.1	0.047	0.049	0.88	2.3
Evidence in GTEx long-read data?94	N/A	no	no	yes	yes

See Table S4 for details. N/A, not applicable.

We next examined transcriptional markers around each retrocopy by using psiCube, which analyzes pseudogenes across humans, flies, and worms.⁹⁹ Based on psiCube annotation, both ATF4P3 and ATF4P4 have active chromatin (pseudogene activity resource, psiCube, April 2014⁹⁹). We examined this further by querying ENCODE data from seven human cell lines for chromatin markers.¹⁰⁰ ATF4P1, ATF4P2, and ATF4P3 have H3K4Me1 markers, indicative of enhancer elements,¹⁰¹ nearby their insertion sites (Figure S7A). ATF4P4 has a H3K27Ac marker, also indicative of enhancer elements¹⁰² (Figure S7A). ATF4P3 and ATF4P4 each harbor H3K4Me3 markers, which are often found near transcription start sites¹⁰³ (Figure S7A). We also note that ATF4P4 is inserted directly next to a different retrocopy, SH3GL1 (MIM: 601768),³ but the SH3GL1 retrocopy has no H3K4Me3 markers (Figure S7A), indicating specificity to ATF4P4. Adding to these chromatin data, we examined cCREs from ENCODE data.⁵² ATF4P3 and ATF4P4 have several cCRE markers indicating potential regulatory elements and open chromatin (Figure S7B). We next looked at transcription start site (TSS) mapping data (through Cap Analysis of Gene Expression [CAGE] sources),⁵³ and there are strong TSS signals in ATF4P3, moderate signals in ATF4P1 and ATF4P4, and only weak signals in ATF4P2 (Figure S7B). Finally, given their connection to ATF4 and stress, we examined retrocopies for JASPAR data⁵⁴ of any ISR or UPR-specific transcription factor motifs. All retrocopies have a nearby (<1 kb) ATF3-binding motif (Figure S7B), which is a target of ATF4.^104,105 ATF4P1/2 and ATF4P3 have motifs of either CEBPA (upregulated by the ISR¹⁰⁶) or CEBPA dimerized with DDIT3/CHOP (albeit further away) (Figure S7B). Finally, ATF4P1/2 also have an XBP1-binding motif that is associated with the UPR (Figure S7B). Taken together, these regulatory elements agree with GTEx expression data showing evidence for expression of ATF4P3 and ATF4P4 and weak or no expression of ATF4P1/2.

To verify the mRNA expression of these retrocopies, we tested endogenous retrocopy expression levels in a human cell line. Because ATF4 interacts with tumor cell biology,^{8,12,13,14,15,16,17} we avoided using common cancer-derived cell lines such as HeLa cells. Thus, we used hTERT RPE-1 cells, as they are a noncancerous human cell line that is karyotypically normal (CRL-4000, ATCC).^107,108 The ATF4 retrocopies are highly identical to ATF4 (Figure 1B), so we designed primers to minimize off-target binding to each other. As the 3′ end of each primer is most important for binding and subsequent synthesis by polymerase,⁵⁶ we designed primers with a mismatch at the 3′ end of either the forward or reverse primer (see materials and methods). For ATF4P1/P2, we instead designed a primer around a large deletion only present in ATF4P1/2. Each ∼100- to 200-bp product contained single-nucleotide variants (SNVs) with which we verified amplification of the correct cDNA. Importantly, we used DNAse treatment to eliminate genomic DNA contamination from our RNA samples, and we verified the efficacy of this treatment (Figure S8A). Using these primers and methodology, we sequenced RNA-derived cDNA from WT RPE-1 cells.

Using Sanger sequencing and chromatogram analysis, we successfully identified expression of each specific retrocopy of ATF4 (Figure 3A). ATF4P1/P2 and ATF4P4 produced clean SNVs distinguishing them from both ATF4 and the other ATF4 retrocopies (Figure 3A). However, ATF4P3 was partially contaminated by ATF4, likely due to its high similarity to ATF4 (>95%) preventing improvements to primer design. Nevertheless, we identified distinct ATF4P3 transcripts via ATF4P3-specific nucleotide changes and lack of contamination by ATF4P4 and ATF4P1/2 (Figures 3A and S8B). Of note, we find that ATF4P1/2 is indeed transcribed, despite its low expression and absence of long-read sequences, on GTEx (Table 4). Overall, we find that all ATF4 retrocopies are endogenously transcribed in RPE-1 human cells.

Figure 3.

Chromatogram analysis identifies endogenous ATF4 retrocopy transcripts in human cells

(A) We used primers specific to each gene to amplify transcripts under normal conditions. Using SNPs, we identified sections of each amplified transcript that differed between each gene. Black bars denote the SNPs used, and the transcript being examined is in blue. We checked the chromatograms of these regions to determine how clean each transcript signal was (e.g., whether ATF4, or other retrocopies, were being partially amplified). ATF4, ATF4P1/2, and ATF4P4 had clean peaks, while ATF4P3 had slight contamination from ATF4 transcripts. Overall, we were able to detect that every retrocopy transcript was present in cells. We created alignments using ClustalOmega and visualized them using MView.⁶¹

(B) Using the primers from (A), thapsigargin (Tg) treatment induced ATF4 and ATF4P1/2 expression (n = 5 biological replicates; ∗∗∗∗p < 0.0001, ∗p < 0.05, Student’s t test).

(C) Like Tg treatment, treatment with arsenite (As) induced ATF4 and ATF4P1/2 expression (n = 3 biological replicates; ∗∗p < 0.01, ∗p < 0.05, Student’s t test). For (B) and (C), we plot the average fold change and the standard error.

ATF4 expression is transcriptionally and translationally induced by multiple inputs into the ISR.¹ We tested whether inducing the ISR also induced mRNA expression of ATF4 retrocopies. We exposed WT RPE-1 cells to thapsigargin (Tg), a SERCA inhibitor that causes ER stress and activates the PERK branch of the ISR.^1,109 As previously shown,^110,111,112 Tg induces mRNA expression of ATF4 (Figure 3B). Remarkably, Tg also increased mRNA expression of the retrocopy ATF4P1/2 to approximately the same fold increase (∼2-fold). Tg did not significantly increase ATF4P3 or ATF4P4. To determine whether the ATF4P1/2 increase in expression was from ISR induction and not a Tg-specific response, we tested a second inducer of the ISR. We exposed cells to sodium As—an inorganic compound that activates the HRI branch of the ISR.^1,113 As with Tg, As induced mRNA expression of both ATF4 and ATF4P1/2 (Figure 3C), but it did not significantly increase ATF4P3 or ATF4P4. While neither ISR-inducing drug significantly increased ATF4P3 or ATF4P4 expression, there was high variably across both drugs tested (Table S5). It is possible that more stable induction might be achieved using more precise timing or dosage of ISR-inducing drugs.

ATF4 retrocopies can be translated and are degraded by the proteasome

To determine the effects of these retrocopies in human cells, we used lentiviral transfection to create stable OE lines of ATF4 and each of the ATF4 retrocopy ORFs in RPE-1 cells. We created each OE cell line and confirmed correct sequencing—including the presence of an in-frame V5 tag. Using the V5 tag, we verified correct band sizes for each of the overexpressed proteins except for the ATF4P1/2 Small X retrocopy, from which we were unable to verify protein on our western blots (likely due to its small size) (Figures S9A and S9B). These OE cell lines had no gross defects when compared to WT cells.

As described, ATF4 has two degradation domains: (1) an ODD domain and (2) a βTrCP motif. Each of these domains result in the sequestration to, and termination at, the proteasome.^1,20,24 As such, inhibiting the proteasome stabilizes ATF4 protein levels,^24,114 so we tested whether that is also true for ATF4 retrocopies. To inhibit the proteasome, we used Btz.^114,115 As expected, ATF4 protein increased under Btz treatment (Figures 4A, S9B, and S9C). In each retrocopy OE cell line, the protein was also increased under proteasome inhibition (Figures 4A, S9B, and S9C), indicating that these are also degraded by the proteasome.

Figure 4.

ATF4 retrocopy proteins are degraded by the proteasome and affect the expression of canonical ATF4 gene targets

(A) We grew each cell line overexpressing ATF4 and its retrocopies with or without proteasome inhibition (through Btz). Proteasome inhibition increased the ratio of ATF4/retrocopy protein in each instance (see red boxes and rightmost graph). Note that image contrast was globally adjusted for band clarity; see full gel in Figure S9.

(B) We analyzed gene expression of five known ATF4 target genes in each OE cell line, with or without proteasome inhibition. Blue bars represent empty vector (EV) control cell lines, and red bars represent cell lines overexpressing the gene indicated above each graph. Generally, the increased levels of each retrocopy protein from proteasome inhibition (in A) had a stronger impact on the expression of each ATF4 target gene. However, there were gene expression changes under both conditions, indicating that these retrocopies can impact ATF4 gene expression (n = 3 biological replicates; ∗∗∗∗p < 0.0001, ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05, one-way ANOVA with adjusted p values derived using Dunnett’s multiple comparison correction).

Surprisingly, while ATF4P4 and ATF4P1/2 lack both of the ATF4 degradation domains (Figure 1C), they still showed increased protein levels with proteasome inhibition. Given the lack of these domains, it is likely that a different mechanism underlies their eventual degradation by the proteasome. We also observed a second band at ∼10–12 kDa in the ATF4P4 OE line (Figures S9A–S9C) that also increased under Btz treatment. We note that there is a second methionine start site within ATF4P4, which, if translated, would be approximately the observed size (annotated in Figure S9D). Finally, we also note that ATF4P4 and ATF4P1/2 have a stronger relative response to Btz treatment (Figures 4A, S9B, and S9C). This could suggest that they have a faster turnover rate by the proteasome. However, the proteasome must unfold ubiquitinated proteins prior to processing.¹¹⁶ Given that ATF4P4 and ATF4P1/2 are both truncated, they may simply have a less stable folded product and are thereby more easily processed by the proteasome—regardless of any particular regulatory mechanism. Taken together, all ATF4 retrocopy proteins are degraded by the proteasome.

ATF4 retrocopies affect expression of canonical ATF4 target genes

ATF4 is a transcription factor that affects the expression of hundreds of genes.^5,6,7 We wanted to determine what effect, if any, ATF4P1–4 has on the expression of a diverse set of ATF4 target genes. We selected five ATF4 transcriptional target genes relating to unique downstream pathways: TRIB3 (MIM: 607898),³ involved in negative feedback of ATF4¹; ATF3 (MIM: 603148),³ a transcription factor important for stress and regeneration¹; SARS1 (MIM: 607529),³ a tRNA charging protein⁶; XPOT (MIM: 603180),³ involved in export of tRNAs⁶; and ERO1A, involved in protein folding.⁶ We determined the expression of these genes under ATF4 and retrocopy OE. To simulate cell stress resulting in increased expression of these proteins, we also determined gene expression under Btz treatment. Note that we included our nucleotide sequence-verified ATF4P1/2 Small X cell line in this assay, although we could not verify the expression of its small peptide (Figure S9B).

ATF4P3 increased XPOT expression, with or without Btz treatment (Figure 4B). It also increased SARS1 expression without Btz and ERO1A expression with Btz, but neither to a statistically significant level. As ATF4P3 most closely resembles ATF4, we expected that this protein would be most able to increase expression of ATF4 target genes. However, its effect on increasing gene expression was smaller compared to the other retrocopies—suggesting either that its high identity to ATF4 means it is also tightly controlled¹ or that one or more of its 15 single-amino acid variants (Figure 1C) affect its ability to alter gene expression.

ATF4P4 had the strongest overall effect on the expression of ATF4 target genes, but, surprisingly, it decreased expression of those genes rather than increased them (Figure 4B). ATF4P4 significantly reduced the expression of ATF3 with or without Btz. ATF4P4 also reduced the expression of TRIB3, XPOT, and SARS1, but only with Btz treatment. ATF4P4 lacks the basic leucine zipper domain, including its internal nuclear localization signal (Figure 1C), so this is unlikely due to any DNA-binding activity. The truncated form of ATF4P4 leaves only the p300-binding domain intact (Figure 1C). Thus, one potential mechanism is that ATF4P4 competes with ATF4 for p300 binding, thereby reducing the stability of ATF4 and its transcription of DNA.¹⁹ Alternatively, ATF4P4 may bind ATF4 co-binding proteins, such as CHOP,¹⁸ to prevent formation of heterodimers necessary for gene transcription. Overall, ATF4P4 expression negatively impacts the expression of several canonical ATF4 target genes.

While ATF4P1/2 Long X has a similarly truncated protein like ATF4P4 (Figure 1C), it had a smaller effect on gene expression (Figure 4B). Under Btz treatment, ATF4P1/2 Long X increased both XPOT and SARS1 expression. However, ATF4P1/2 Long X lacks the first 16 aa of the p300-binding domain, which may weaken any potential binding to p300 and any subsequent effects on ATF4. This is further supported by the fact that the ATF4P1/2 Small X peptide, which is simply the first 15 aa of the p300-binding domain (with one change, D15A) (Figure 1C), mimics ATF4P4 by also having a negative effect on ATF4 target gene expression. With or without Btz treatment, ATF4P1/2 Small X reduced TRIB3 expression (Figure 4B). Under Btz treatment, ATF4P1/2 Small X reduced the expression of ATF3 and ERO1A. ATF4P1/2 Small X also reduced expression of ERO1A without Btz treatment, but not to a statistically significant degree. Given these results and the amino acid overlap of ATF4P4 and ATF4P1/2 Small X, it is possible that the first 15 aa of the p300 interaction domain have a negative effect on ATF4 target gene expression.

Overall, it is clear that, with or without cell stress, ATF4 retrocopies can alter ATF4 target gene expression. Given this, their similar protein-degradation patterns (Figure 4), their ability to be upregulated by the ISR (Figure 3), and their expression in humans (Figure 3), we conclude that ATF4P1–4 are functional retrogenes.

Discussion

We characterized four retrogenes of the stress transcription factor ATF4 in humans. We found these retrogenes across multiple primate lineages, suggesting an increase in their selection ∼14.8 mya. These retrogenes are expressed at the mRNA level in cells—including ATF4P1/2, which responds to induction of the ISR. Overexpressing stable proteins of ATF4 retrocopies alters the expression of ATF4 target genes, indicating their potential for biological function. Given the important role for ATF4 in human diseases, such as cancer,^{8,12,13,14,15,16,17} our findings suggest that these ATF4 retrogenes should be considered when studying the ISR and ATF4 pathways.

The increase in LINE-1 element transposons in mammalian lineages resulted in an increase in their total retrocopied genes.^90,117 Humans have ∼8,000 retrocopies, but most remain uncharacterized or unstudied.^34,90,92,118 However, there are several instances of retrocopies having biological function. HAPSTR2 is a retrogene that aids in cell stress signaling.³⁵ The retrogene retroCHMP3 inhibits viral budding.^36,76 The testis-specific retrogene PGK2 positively regulates sperm function.^41,98 At the mRNA level, the retrogene PTENP1 regulates PTEN expression at the mRNA level.^37,38 Its transcript acts as a decoy that pulls inhibiting microRNAs away from PTEN, thereby increasing its expression.^37,38 Thus, ATF4P1–4 are yet another example of retrogenes capable of affecting cell biology. Of note, retrogenes can interact with their parent gene and/or evolve new function. For example, HAPSTR2 increases the stability of its parent gene, HAPSTR1, but it also impacts stress signaling when HAPSTR1 is knocked out.³⁵ On the other hand, the function of PTENP1 appears to primarily revolve around changing expression of its parent gene, PTENP1, without having independent function. While the ATF4 retrocopies can impact expression of known ATF4 target genes, it is still unclear whether they also have independent functionality. Neofunctionalization in stress transcription factor families is not uncommon. For example, ATFs have seven members (ATF1–7), which share similar DNA-binding domains and recognition sequences but which have different target genes and expression profiles.¹¹⁹ Similarly, HSF transcription factors have also diversified into having several specialized functions, likely through duplication events.¹²⁰ Thus, it will be worth testing the ATF4 retrocopies as to whether they have also gained any neofunctionality.

Approximately 30% of retrocopies have RNA expression data.^117,121 However, previous expression data must be validated. For example, there is evidence of ATF4P1/2 mRNA expression in our cells, yet ATF4P1/2 does not have significant expression in GTEx.¹²² In addition, while we find evidence for ATF4P3 and ATF4P4 in GTEx long-read data, we did not find ATF4P1 and ATF4P2. In this case, this might be due to the unique genetic region containing ATF4P1/2; however, we also did not find long-read transcripts for two other retrogenes (PTENP1 and PGK2) despite these having established expression.^37,38,41,98 Therefore, individual retrocopy expression should be more closely examined on a case-by-case basis, and it is possible that some retrocopy transcripts require transcript-specific primers to distinguish them.

When making OE constructs, we were surprised that adding endogenous ATF4, or retrocopy, CDSs driven by a CMV promoter did not produce protein (tested via western blot, in both transfected RPE-1 and HeLa cells). Adding the entire 5′ UTR with the CDS also did not produce protein. Only by editing the Kozak sequence (see materials and methods) upstream of the ATG were we able to resolve protein. ATF4 protein is translated basally, even in the absence of stress,⁶ and ATF4P3 and ATF4P4 share the same Kozak sequence. Thus, while we cannot rule out that endogenous translation of ATF4P3 or ATF4P4 might be affected by this Kozak sequence, at least they would likely be affected to the same degree as parental ATF4. However, the ATF4P1/2 Small X has an SNP in its Kozak sequence, so it is possible that its endogenous translation could be affected.

The ISR is a critical stress-response pathway conserved from yeast to humans.^1,123 It is crucial for restoring cell homeostasis under various stressors.¹ It is dysregulated in some individuals with Wolcott-Rallison syndrome¹²⁴ (MIM: 226980)³ and pulmonary disease,¹²⁵ as well as in several other disease models.^126,127,128 We found that the retrogenes ATF4P1/2 are upregulated under the ISR stressors As and Tg^113,129 (Figure 3). In addition, OE of ATF4P1/2 altered expression of several canonical ATF4 target genes (Figure 4)—positively or negatively depending on the ORF involved. he ATF4P1/2 retrogene is the oldest retrocopy in humans (Figure 2), and both copies remain unchanged at the nucleotide level for ∼14 my in humans. Thus, ATF4P1/2 may actively participate in the ISR in humans and should be considered when further studying these pathways and disease models. ATF4P3 and ATF4P4 had more variable expression under ISR stress (Figure 3), and this may indicate they need different conditions for their expression. For example, ATF4P3 is inserted in the intron of RNF157 (Figure S2), which has strong brain expression¹³⁰ (and also via GTEx¹²²). Thus, it may require a neuronal model or cell line to fully capture the expression or effects of ATF4P3.

Retrocopies can affect the expression of their parent genes at the transcriptional or translational level.^{34,36,37,38,121} Based on our proteasome inhibition experiments (Figure 4), all four ATF4 retrogene proteins are typically degraded by the proteasome (although not necessarily at the same rate). Inhibiting the proteasome, and thus increasing their protein levels, alters the expression of ATF4 target genes. This suggests that their actual translated peptides can influence gene expression. While this could be explained by an intact DNA-binding domain in ATF4P3, it does not explain ATF4P4 or ATF4P1/2, which lack this domain. ATF4 typically forms a heterodimer, such as with CHOP or ATF3,¹⁸ to bind to DNA. It is possible that ATF4P4 or ATF4P1/2 can bind—even partially—to these co-binding partners and sequester them away from ATF4. This could reduce gene expression overall or force ATF4 to bind with different co-binding partners and switch which genes are upregulated. Alternatively, it is possible that these retrogenes act at the RNA level (like retrogene PTENP1^37,38)—as our OE clones would presumably have increased mRNA as well—or a combination of the two. The exact mechanism of how these retrogenes influence ATF4, and its transcriptional ability, will be of interest to further characterize how they impact the ISR and related diseases.

As noted above, we find a second band at ∼10–12 kDa in the ATF4P4 OE line (Figures S9A–S9C) and a second methionine start site within ATF4P4 (annotated in Figure S9D). Because this smaller product of ATF4P4 also increased under Btz treatment, it is likely also degraded by the proteasome. As such, we looked within this product for potential lysine residues with evidence of ubiquitination in ATF4 (using Uniprot⁴⁹: P18848). ATF4 residue K88 has high confidence as a ubiquitinated site,^49,131 and it is maintained in ATF4P4 and ATF4P1/2 Long X (Table S1). Thus, we hypothesize that ubiquitination of residue K88 might affect its proteasomal degradation, as it would explain why ATF4P4 and ATF4P1/2 have higher abundance under Btz treatment despite lacking the ODD domain and βTrCP motif.

While they lack their DNA-binding domain, ATF4P4 and ATF4P1/2 maintain their p300 interaction domain (Figure 1C). As p300 usually stabilizes ATF4 and helps prevent its degradation,¹⁹ it is possible that ATF4P4 and ATF4P1/2 sequester p300 away from ATF4 and thereby reduce its expression. Additionally, the ability to bind p300 offers an explanation for the evolutionary origin of these retrogenes, including our finding that humans, chimps, bonobos, and gorillas have at least two “p300-only” retrocopies. Certain viruses can hijack p300 and use it to aid the transcription of viral machinery.^132,133,134 For example, the HIV-1 Tat protein is acetylated by p300 to aid its transcription,¹³⁵ and adenovirus E1A uses p300 to modulate gene expression.¹³⁶ Outside of any effect on ATF4 target gene expression, it is possible that ATF4P4 and ATF4P1/2 sequester p300 away from viruses to help limit viral gene transcription. Given that the ISR is known to have evolutionary pressure from viruses,^42,43 this provides a potential source of evolutionary pressure and selection to help explain the presence of maintained, but truncated, ATF4 retrogenes.

Data and code availability

Data/sequences generated or organized in this study are contained within the figures, tables, and supplementary files with links to original databases alongside them when applicable. The majority of code used in this study was derived from the literature or public sources (cited when used). Our modified code of the mutator script^62,63 is available on GitHub (https://github.com/hansmdalton/pedestal_fork).

Acknowledgments

We thank Erik Eide and Dr. Charles Murtaugh for their help in lentiviral cloning, western blots, and use of equipment. We thank Dr. Nels Elde and Diane Downhour for their samples and guidance in retrocopy experiments and strategies. We thank the HSC DNA Sequencing Core (director Dr. Derek Warner) at the University of Utah for their help in sequencing analysis. We thank Mr. Ben Anderson for his help in manually collecting GTEx data. This work was supported by National Institute of General Medical Sciences (NIGMS, www.nigms.nih.gov/) R35GM124780 to C.Y.C. It was also supported by NIGMS F32GM136057 and Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD, https://www.nichd.nih.gov/) K99HD111662/R00HD111662 to H.M.D. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Declaration of interests

The authors declare no competing interests.

Web resources

• Clustal Omega⁵⁵, https://www.ebi.ac.uk/jdispatcher/msa/clustalo

• Ensembl⁴⁸, https://www.ensembl.org/

• Expasy Translation Tool¹³⁷, https://web.expasy.org/translate/

• GTEx Portal¹²², https://www.gtexportal.org/

• Interactive Tree of Life⁵¹, https://itol.embl.de/

• Online Mendelian Inheritance in Man³, https://www.omim.org/

• PrimerBank⁵⁷, https://pga.mgh.harvard.edu/primerbank/

• UCSC Genome Browser¹³⁸, https://genome.ucsc.edu/cgi-bin/hgGateway

Supplemental information

Table S1. Table listing all retrocopies found in Catarrhini

These include labeled start/stop sites and TSDs. Each tab is a different gene, with ATF4 also listed for reference

Table S2. Table listing all observed retrocopies of ATF4 among 97 different species

Each annotated retrocopy had to reach a BLAT score of at least 100 and contain no introns and/or a poly(A) tail to be considered a “yes.” Alternatively, we assigned “possible” when one of these features was present, but the sequence was either highly degraded or missing (such as in an incomplete scaffold assembly). Sequences containing many hits from BLAT of a repeat sequence (likely from a transposon) are marked as “multiple complex duplications.” Non-retrocopy gene duplications are also noted in under “comments”

Table S3. Document containing estimations of ancestral versions of ATF4

These are derived from Ape ATF4 transcripts to represent the ATF4P3/4 ancestral gene (14.8 mya) or Catarrhini ATF4 transcripts to represent the ATF4P1/2 ancestral gene (37.5 mya)

Table S4. Table of median TPM of each gene taken from GTEx (

www.gtexportal.org)

Tissues reaching expression of at least 0.5 TPM are highlighted in green, and these data are quantified in a table below. The right table sorts each tissue by the highest expression of ATF4 to compare retrocopy expression rates

Table S5. Raw Ct values for qPCR analyses of RPE-1 cells under Tg or As treatment

Highlighted cells indicate a single cell population with a stronger overall response (in every gene tested) relative to the other populations. “No template” control results are on the right, indicating the maximum possible Cts achieved with no template input (“undetermined” means no signal after 40 cycles)