IGHMBP2 deletion impacts cellular proliferation and global translation

To investigate the molecular function of IGHMBP2, we stably perturbed its expression in K562 CRISPRi cells using CRISPR-Cas9 with two sgRNAs targeting exon 2 of IGHMBP2 (Fig 1A). Edited clones were screened for inactivated alleles via PCR, and selected partial and full-knockout IGHMBP2 monoclonal cell lines were validated by Sanger sequencing and Western blot (Figs 1B and C and S1A–E). To determine the physiological impact of IGHMBP2 deletion in cells, we stably expressed mEGFP in IGHMBP2 deletion clones and measured competitive proliferation relative to parental cells via flow cytometry across 2 wk of cell passages. We observed lower proliferation rates and heterogenous cell morphology among IGHMBP2 deletion cells (Figs 1D and E and S1F). We confirmed mEGFP expression does not impact cell proliferation and that cell lines expressing an alternate fluorophore used in subsequent experiments, TagBFP, exhibited comparable proliferation results (Fig S1G). Expression of fluorescent proteins remained robust across the experimental timeframes (Fig S1H). We conclude that IGHMBP2 loss induces a modest proliferation defect in K562 cells.

The effect of IGHMBP2 on proliferation could be related to its implicated role in protein synthesis. We therefore assessed the impact of IGHMBP2 deletion on global translation by performing polysome profiling (Fig 2A). Western blot of polysome fractions shows IGHMBP2 co-sediments with 40S, 60S, and 80S ribosomal species (Fig 2B). When we quantified the area under the curve for polysome peak groups and compared relative peak ratios per cell line, we observed mild yet reproducible enrichment of free ribosomal subunits relative to monosomes and polysomes in IGHMBP2 deletion clones, while subunit abundances held constant (Fig 2C). This suggests that translational suppression caused by IGHMBP2 deletion is not because of gross ribosomal biogenesis defects. To monitor global translation in cell lines differentially expressing IGHMBP2 at single-cell resolution, we measured relative rates of nascent polypeptide synthesis via an O-propargyl-puromycin (OPP) assay. We found translation was suppressed in IGHMBP2 deletion clones (Figs 2D and S2A–C), consistent with the polysome profiling results. We reproduced these results in cell lines expressing transgenic mEGFP, which were initially sorted to include all GFP+ cells and detected decreased mEGFP expression per cell among IGHMBP2-disrupted clones as an additional orthogonal measure of translational activity (Fig 2E and F).

Ribosome profiling reveals differentially expressed genes in IGHMBP2 deletion cells

Modest global changes in translation can result from changes to the translation of a subset of mRNAs (Calviello et al, 2021). To determine the gene expression changes in IGHMBP2 deletion cells, we performed ribosome profiling alongside RNA-seq. Ribo-seq and RNA-seq read counts showed high correlation between replicates (Fig S3A–J). We confirmed that Ribo-seq data exhibited 3-nt periodicity (Fig S4A–J), and most of the reads were forward-sense stranded (Fig S5A–J) and mapped to coding exons as expected (Fig S6A–J). We then performed differential expression analysis to identify and interpret the mode of gene regulation of IGHMBP2-dependent differentially expressed genes (DEGs). For each gene among RNA-seq and Ribo-seq results, IGHMBP2-dependent log₂ fold-changes (L2FC) were determined relative to parental expression levels and the Ribo:RNA-seq interaction term L2FC signified translational efficiency (TE). Genes with significant RNA-seq, Ribo-seq, or TE level changes were classified under ΔRNA-seq, ΔRibo-seq, or ΔTE, respectively. If applicable, ΔTE genes were further categorized as translation exclusive if only ribosome occupancy level was changed, or translation buffered if only mRNA abundance was altered (Chothani et al, 2019). To reduce possible artifacts for subsequent analysis, only genes found significant in at least two of the four ΔIGHMBP2 cell lines in any classification were retained as a significant hit or else designated nonsignificant (Table S1). When broadly comparing results per clone, we observed an increased magnitude and number of gene expression changes in full versus partial IGHMBP2 deletion cells (Fig 3A).

We functionally characterized IGHMBP2 KO differential expression results via gene set enrichment analysis (GSEA; Table S2). We observed gene expression attenuation between IGHMBP2-dependent RNA-seq (Fig 3B) and TE results (Fig 3C): translation machinery and mitochondrial gene sets were up-regulated transcriptionally, yet suppressed in TE. Developmental and homeostatic gene sets declined transcriptionally, yet increased in TE. The peptide biosynthetic process set—which contains ribosomal machinery, translation initiation, and elongation factors largely overlapping within other translation sets—was down-regulated among IGHMBP2 KO TE results via 167 out of 500 possible genes. Also among IGHMBP2 KO TE results, the regulation of developmental process set was highly enriched via 306 of 854 possible genes that broadly overlap with the anatomical structure morphogenesis gene set (Fig 3C; Table S2). Altered transcriptional programming due to IGHMBP2-dependent TE changes was also suggested by enrichment of the negative regulation of DNA-templated transcription factor gene set, with a normalized enrichment score (NES) of 1.6 (P.adj = 5.6 × 10⁻⁵), reflecting 236 of 582 possible genes (Table S2). Upon performing GSEA with ribosomal footprint fold-changes, we observed translational suppression of various homeostatic gene sets appropriately corresponding with the hemopoietic cell type of K562 cells (Fig 3D).

To visualize reproducible RNA-seq level changes, we filtered for ΔRNA-seq genes in both heterozygous or full IGHMBP2 deletion clones. For each group, most of the hits overlapped in the same directionality (10 out of 13 DEGs between IGHMBP2 HET clones; 43 out of 51 DEGs between IGHMBP2 KO clones) and showed significant correlation (Fig S7A and B). Together, these data identify the genetic changes underlying morphological and translation defects observed as a consequence of IGHMBP2 deletion, again implicating translation as a process impacted by loss of IGHMBP2 (Fig 3E).

ATF4 is up-regulated in IGHMBP2 deletion cells

Considering the recessive nature of IGHMBP2-linked pathologies, we sought to determine the genetic changes specific to full loss of IGHMBP2 to identify possible disease-relevant targets. Of several ΔTE genes identified from our Ribo:RNA-seq analysis (Fig S8A), ATF4 expression was notably up-regulated in genes filtered for dependence on full loss of IGHMBP2 (Fig 4A). Upon assessing ΔTE categorization, ATF4 was identified as a translation-exclusive ΔTE gene only in the full IGHMBP2 KO genotype condition (Fig S8B and C). Induction of ATF4 is caused by translational control through two upstream ORFs (Lu et al, 2004; Vattem & Wek, 2004). We observed an enrichment of normalized reads mapped to the uORFs and CDS of ATF4 in IGHMBP2 deletion clones relative to parental cells, especially at the enhancer uORF1 (Fig 4B and C), suggesting translational activation of ATF4 expression.

To test the hypothesis that IGHMBP2 deletion induces ATF4 expression, we generated parental and ΔIGHMBP2 reporter cell lines stably expressing a CMV-driven uORF1,2(ATF4)-mApple reporter gene (Fig 4D). As part of the reporter system, a separate CMV-EGFP transgene was introduced to enable normalization for promoter strength and translational activity (Guo et al, 2020). We used this reporter because of challenges with robust detection of ATF4 via immunoblot. We detected elevated ATF4 reporter levels in IGHMBP2 deletion clones relative to parental cells (Figs 4E and S9A). We confirmed inducibility of reporter expression upon acute ISR stimulation via thapsigargin (Fig S9A and B). We then exogenously expressed IGHMBP2 N- or C-terminally fused with TagBFP in our IGHMBP2 deletion reporter cell lines and observed ATF4 reporter expression reversed toward parental levels of ATF4 with both transgenes (Figs 4F and S9C and D), indicating genetic rescue and supporting the causative nature of IGHMBP2 genetic lesions. We therefore find that loss of IGHMBP2 induces ATF4 expression.

IGHMBP2 deletion results in ISR activation

ATF4 is a stress-responsive transcription factor that is tightly regulated at the translational level (Harding et al, 2000; Vattem & Wek, 2004). Given the role of ATF4 as part of the ISR and the relevance of the ISR in neurodegenerative diseases, we wondered if ISR activation is a consequence of IGHMBP2 deletion. IGHMBP2-dependent, mild suppression of ribosomal proteins and elongation factors shown in our sequencing results would also be consistent as a repercussion of ISR-induced eIF2α phosphorylation (Sidrauski et al, 2015). Surprisingly, differences in p-eIF2α were not discernible between parental and IGHMBP2 deletion cells via Western blot (Fig S10A and B). This may be due to homeostatic regulation, which produces reduced levels of p-eIF2α upon chronic ISR activation relative to acute stress (Novoa et al, 2003; Guan et al, 2017). However, upon treating IGHMBP2 deletion cell lines with p-eIF2α-dependent ISR inhibitor (ISRIB; Sidrauski et al, 2015), ATF4 reporter expression in IGHMBP2 KO cells robustly reversed toward parental reporter levels at steady state, supporting the occurrence of chronic, p-eIF2α–induced ISR activation (Fig 5A).

Four kinases can activate the ISR: PKR senses specific types of dsRNA, HRI senses heme concentration and mitochondrial dysfunction, PERK responds to the unfolded protein response in the endoplasmic reticulum, and GCN2 is activated via engagement with uncharged tRNAs or detection of stalled ribosomes (Costa-Mattioli & Walter, 2020). Given the genetic connection between IGHMBP2 and tRNA metabolism, we hypothesized ISR activation in IGHMBP2 deletion cells could be driven by empty A-site induced ribosomal stalling associated with tRNA-Tyr availability and consequential GCN2 activation. Considering IGHMBP2-ribosome association biochemically documented in literature and reproduced in our own data (Fig 2B; Guenther et al, 2009a), IGHMBP2 loss could also disrupt translation elongation by some mechanism unrelated to tRNA metabolism, yet again pointing to the GCN2 arm of the ISR. We thus treated ATF4 reporter cell lines with kinase inhibitor GCN2iB (Nakamura et al, 2018) and observed modest effects upon IGHMBP2 KO clones (Fig 5B). Although expected, two-way interaction between ISRIB and GCN2iB-treated parental versus IGHMBP2 deletion cells was statistically insignificant. This may suggest the effect of ISR inhibition on ATF4 reporter expression is not IGHMBP2-dependent, or is driven by the mild effect of ISRIB on parental cells and overall mild effect of GCN2iB on IGHMBP2 KO cells. One-way analysis, however, shows ATF4 up-regulation in IGHMBP2 KO cell lines is partially rescued by GCN2iB treatment, supporting the involvement of GCN2 in ISR activation (Fig 5A and B).

To further test the cellular and molecular consequences of ISR treatment on IGHMBP2 KO cells, we monitored relative growth rates in the presence of ISRIB or GCN2iB and found that proliferation of IGHMBP2 deletion clones was not impacted by either treatments across a 9-d time-course (Fig S11A). However, when measuring relative morphology and translation among cell lines, both inhibitors rescued relative median cell size in IGHMBP2 KO clones after 9 d (Figs 5C and S11B), and ISRIB partially restored translation indicated by transgenic mEGFP expression after 1 wk of treatment (Figs 5D and S11C). The particular strength of ISRIB in translational rescue suggests that targeting p-eIF2α as the ISR bottleneck is more effective than suppressing upstream ISR stimulation via GCN2 inhibition, which may be expected under chronic stress activation. Although GCN2iB treatment does not restore translation in IGHMBP2 deletion cells, which could also implicate involvement of additional ISR kinases, we note its dissimilar effects between parental and heterozygous versus IGHMBP2 KO cell lines. After 5 d of treatment, GCN2iB reproducibly negatively impacts translation only in the presence of IGHMBP2 (Figs 5D and S11C), which may evidence differential IGHMBP2-dependent GCN2 activity or translation buffering effects that cannot be directly measured via net changes with this experiment.

Upon determining that ISR activation is driven by GCN2 in IGHMBP2 deletion cells, we wondered if we could detect ribosome stalling among our Ribo-seq reads, as empty A-site stalled and collided ribosomes activate GCN2 (Ishimura et al, 2016; Inglis et al, 2019). We first performed a metagene analysis using Ribo-seq reads to assess whether IGHMBP2 deletion alters ribosome positioning along transcripts globally, as might occur under strong stalling conditions. We observed consistent ribosomal occupancy across sequenced transcripts in knockout compared with parental cell lines (Fig S12A), which may suggest IGHMBP2 impacts translation by a more ubiquitous mechanism nonspecific to stages of translation, or perhaps toward specific transcripts undetectable by global analysis. Determining whether IGHMBP2-RNA recognition or unwinding activity impacts global translation and how this may be governed by particular biophysical properties or motifs among mRNAs warrants further experimentation.

As tRNA dysregulation may be a favorable hypothesis driving global translation changes, specifically tRNA-Tyr as implicated from prior literature (de Planell-Saguer et al, 2009), we performed a Northern blot to visualize total tRNA-Tyr and tRNA-Val expression among our cell lines (Fig S12B). We did not discern differences in abundance of these tRNAs, suggesting IGHMBP2-dependent differences in tRNA-Tyr levels may not occur in K562 cells or are below detectability with biochemical assays. To further test the hypothesis that IGHMBP2 loss affects decoding of a subset of codons, we applied a recently developed machine learning package, choros, to infer in-frame ribosome A-site occupancy of codons globally, but did not detect specific codon enrichment (Mok et al, 2023 Preprint; Fig S12C–H). This lack of signal is perhaps explained by the overall subtle effect size of IGHMBP2 loss on translation, or could occur for other unknown reasons. We conclude IGHMBP2 deletion results in chronic, low-level ISR activation mediated by GCN2 signaling. GCN2 recognition of stalled ribosomes (Ishimura et al, 2016) or interaction with uncharged tRNAs (Wek et al, 1995) may occur below our detection threshold in response to disrupted tRNA metabolism or IGHMBP2-ribosome processes as possible ISR activation mechanisms due to IGHMBP2 loss (Fig 5E).

Cell culture

K562 CRISPRi cell lines used in this study were maintained in RPMI 1640 (Gibco) containing L-glutamine and 25 mM HEPES, supplemented with 10% FBS and 1% penicillin-streptomycin. Cells were stored in humidified incubators at 37°C with 5% CO₂ and routinely checked for mycoplasma (Lonza).

Cell line generation

Cas9-mediated knockout of IGHMBP2 was performed using two sgRNAs (Synthego) to target exon 2 of IGHMBP2 in combination: 5′-CAGAGAGAUGUUCUCCUGCC-3′ and 5′-CUGAAAGAGCUCCAGAGCCG-3′. Ribonucleoprotein (RNP) complexes were prepared by combining 50 pmol of recombinant Cas9 to 50 pmol of each sgRNA and incubating the mixture for 20 min at 37°C. K562 CRISPRi cells (Gilbert et al, 2014) were nucleofected (SF Cell Line 4D kit; Lonza) with the RNP. Nucleofected clones were isolated by limiting dilution and indel-containing clonal populations were screened by extracting gDNA using Quick Extract (Biosearch Technologies) for PCR amplification using primers targeting up and downstream the Cas9 cut-site: 5′-GGTTGTGGCATTAACTGCCC-3′ and 5′-CCCACATCAATTGTTGGAC-3′. PCR products were separated on a 3% agarose gel at 100 V, and IGHMBP2 gDNA disruption in select clones was further validated via Sanger sequencing with primer 5′-CTTTACGAGGGTACAAGTCACGG-3′ and Western blotting.

Transgenic cell line generation

For virus generation, Lenti-X HEK 293T cells (Takara Bio) were transfected with lentiviral packaging plasmids and transgenic constructs. Virus was harvested and clarified through a 0.45-μm filter after 3 d. Lentiviral titering was optimized to result in <50% transduction to generate cells with MOI < 1, and cells were infected via centrifugation at 150g for 1 h in 8 μg/ml polybrene (EMD Millipore). After 3 d, transduced cells were selected via FACS using an 85–100 μm-diameter nozzle. ATF4 reporter cell lines were generated and sorted following stress induction as previously described (Guo et al, 2020).

Western blot

Cell lines were harvested during active growth (0.4 × 10⁶–0.6 × 10⁶ cells/ml) and lysed in RIPA buffer supplemented with protease inhibitor cocktail (Halt). Lysates were centrifuged at 16,000g for 20 min at 4°C, and total protein concentration was quantified by Pierce BCA Assay (Thermo Fisher Scientific). SDS loading buffer was added to 50 μg total protein, and samples were boiled at 95°C for 5 min. Samples were loaded on 4–15% pre-cast PAGE gels (Bio-Rad), run at 85 V in running buffer, and transferred to a PVDF membrane (Bio-Rad). Membranes were incubated for 1 h in blocking buffer (1% [wt/vol] BSA in 0.1% TBS-T), followed by an overnight incubation at 4°C in blocking buffer containing primary antibody. Membranes were washed for 10 min with 0.1% TBS-T three times, followed by 1 h incubation with secondary antibody with rotation at RT and protected from light. Membranes were washed for 10 min 0.1% TBS-T three times and imaged using a LI-COR Odyssey DLx.

Antibodies used for Western blotting include rabbit anti-IGHMBP2 at 1:500 (23945-1-AP; Proteintech), rabbit anti-β-actin conjugated to Alexa Fluor 680 (Abcam); rabbit anti-eIF2α at 1:1,000; rabbit anti-eIF2α phospho-S51 at 1:1,000. Membranes were incubated with goat anti-rabbit 800 CW secondary antibody (926-32211; LI-COR) at 1:10,000.

Northern blot

Northern blotting was performed using NorthernMax Kit (Invitrogen) with adaptations as described. Total RNAs were extracted from cells with Trizol reagent (15596026; Invitrogen). 5 μg of total RNA per sample was mixed with 100 mM final concentration of Tris–HCl (pH 9.0) and incubated for 30 min at 37°C to remove amino acids attached to the 3′ end of tRNAs. Samples were then mixed with 2X Formaldehyde Load Dye (8550G; Invitrogen) and heat denatured for 3 min at 70°C. RNAs were loaded onto pre-cast 10% TBE-Urea gels (EC68755BOX; Thermo Fisher Scientific) in 1X TBE. RNAs were separated by PAGE, transferred to a positively charged nylon membrane (AM10104; Invitrogen), and cross-linked to the membrane using a UV trans-illuminator (Stratalinker 2400 UV Crosslinker) for 2 min at 1,200 × 100 μJ. Membranes were pre-hybridized with pre-warmed ULTRAhyb ultrasensitive hybridization buffer (AM8670; Invitrogen) at 52°C for 1 h, followed by addition of 10 pmol of each probe. The membranes were hybridized overnight and washed two times with Low Stringency Wash Buffer (AM8673; Invitrogen) with rotation at RT for 5 min. Membranes were probed with either tRNA-Tyr, tRNA-Val_mAC, or tRNA-Val_TAC, stripped, re-probed for tRNA-Gly_sCC as an internal control, and imaged using a LI-COR Odyssey DLx.

tRNA-Tyr and tRNA-Val probes were labeled with IRDye800 RD and the tRNA-Gly probe was labeled with IRDye680RD (LI-COR). Probe sequences used include Gly-sCC (5′-TCATTGGCCRGGAATYGAACCCGGGYCTCCCRCGTGGWAGGCGAGAATTCTACCACTGMACCACCMAYG/iAzideN/C-3′); Tyr-GTA (5′-TCCTTCGAGCCGGASTCGAACCAGCGACCTAAGGATCTACAGTCCTCCGCTCTACCARCTGAGCTATCGAAG/iAzideN/G-3′); Val-mAC (5′-TGTTTCYGCCYGGTTTCGAACCRGGGACCTTTCGCGTGTKAGGCGAACGTGATAACCACTACACTACRGAAA/iAzideN/C-3′); Val-TAC (5′-TGGTTCCACTGGGGCTCGAACCCAGGACCTTCTGCGTGTAAAGCAGACGTGATAACCACTACACTATGGAAC/iAzideN/C-3′) (Dittmar et al, 2006).

Flow cytometry

Flow cytometry was performed on BD LSRFortessa instruments. When the high-throughput screening 96-well plate attachment was used, plate parameters were set to a flow rate of 0.5–1.0 μl/sec, 20 μl sample volume, 50 μl mixing volume, 200 μl/sec mixing speed, 2 mixes, and 200 μl wash volume. Laser parameters were adjusted to ∼400 V for FSC, 250 V for SSC, 260 V for Violet laser detected with 450/50 bandpass filter (for TagBFP), 350 V for Blue detected with 530/30 bandpass filter (for EGFP and Alexa Flour 488), 580 V for YGD detected with 586/615 bandpass filter (for mApple), 520 V for YGC detected with 610/620 bandpass filter (for Alexa Flour 594), and 320 V for Red detected with 670/630 bandpass filter (for Alexa Flour 647). Analysis was performed and visualized using the flowCore and ggcyto libraries in R (Van et al, 2018). Data values were arcsinh transformed and gated to remove debris and doublets. Derivation of standardized and normalized values was performed among data collected on the same day per each experiment. After all gating was applied for each experiment, final populations comprise at least 2,300 to more than 10,000 cells depending on high-throughput screening collection limitations.

Statistical analyses were performed in R. One-way ANOVA (Kruskal–Wallis) was performed for flow cytometry data shown in Figs 2E and F, 4F, and 5A and B. Although P < 0.1 is considered weak significance for Fig 2E and F where the ANOVA P = 0.07 for both results, we accepted this threshold to reject the null hypothesis (H₀) and proceeded with Conover–Iman post hoc testing, as later experiments robustly reproduced these results with strong statistical significance (Fig S11B). Under P-adjusted = P(|T| ≥ |t|), the post hoc pairwise H₀ is rejected if P-adj ≤ α, where α = 0.05, and the Benjamini–Hochberg P-adjustment method was used. For time-course drug-treated samples, two-way ANOVA testing was performed on rank-transformed data followed by Tukey’s post hoc test.

Nascent protein synthesis assay

The Click-iT Plus OPP Protein Synthesis Assay Kit (C10457; Thermo Fisher Scientific) was used as described in a 96-well format with 0.5 × 10⁶ cells harvested during active growth per reaction. Cells were co-incubated with OPP in DMSO or 125 nM cycloheximide as a negative control for 30 min before fixation with 4% PFA and permeabilization with 0.25% Triton-X 100 in PBS. Cells were resuspended in PBS with 5% (wt/vol) FBS and differential fluorescent signal between cell lines was observed by flow cytometry.

Cell proliferation competition assay

Cell lines were seeded in 96-well plates at 1.5 × 10⁵ cells/ml. In wells where mEGFP+ or TagBFP+ transgenic cell lines were seeded with parental cells, 50% abundance of each cell line was targeted. Flow cytometry data were collected 1 d after seeding, showing 35–50% starting populations of fluorescent cells compared with parental cells. Wells were passaged for subsequent measurements every 3 d for 16 d when measuring untreated cells.

For experiments measuring relative growth amid ISR inhibitor treatments, wells were seeded in either 500 nM ISRIB or 1 μM GCN2iB, measured by flow cytometry 16 h after seeding as Day 1, and measured upon passaging into fresh medium (containing either DMSO, ISRIB, or GCN2iB) every other day for 9 d. Relative viability was determined relative to untreated Day 1 wells, and fluorescence or FSC measurements were normalized relative to untreated parental/parental +mEGFP per each day of measurement.

Polysome profiling

Cells were grown in separate flasks in triplicate and treated with 100 μg/ml cycloheximide for 5 min. 10 × 10⁶ cells were harvested per sample and lysed with a hypotonic lysis buffer (10 mM Hepes, pH 7.9, 1.5 mM MgCl₂, 10 mM KCl, 0.5 mM DTT, 1% Triton X-100, and 100 μg/ml cycloheximide) and trituration with a 26-G needle. Sucrose gradients were prepared with 10% and 50% sucrose in sucrose gradient buffer (100 mM KCl, 20 mM HEPES, pH 7.6, 5 mM MgCl₂, 1 mM DTT, and 100 μg/ml cycloheximide). For each cell line, 100 μl lysate was layered atop the sucrose gradient, and polysomal species were separated by ultracentrifugation with an SW 41 Ti swinging-bucket rotor (Beckman Coulter) at 36,000 RPM for 2 h at 4°C. Polysome profiles were obtained via injection through a spectrophotometer at a flow rate of 2 ml/min, and absorbance was recorded at 260 nm with sensitivity set to 0.1 on the UV/Vis detector. Gradients were eluted by fractionation, and peak fractions were methanol precipitated and resuspended in equal volumes for identification via subsequent immunoblotting.

Ribosome profiling

Cells were seeded in fresh medium 16 h before harvest. Cell count was measured near 0.3 × 10⁶ cells/ml per cell line at time of harvest, reflecting active growth conditions. Ribosomal footprint RNAs were obtained as previously described (McGlincy & Ingolia, 2017; Calviello et al, 2021). Cell lysates were treated with RNase I and subjected to size exclusion chromatography using MicroSpin Columns S-400 HR (Illustra). Ribosome-protected RNA was then extracted with Trizol, and RNA corresponding to the size of monosomal footprints (26–34 nt) were isolated by size selection via running RNA samples in a 15% polyacrylamide TBE-Urea gel. Footprint fragments were dephosphorylated and ligated to pre-adenylated oligonucleotide linker (NI-816: 5′-/5Phos/NNNNNTAGACAGATCGGAAGAGCACACGTCTGAA/3ddC/-3′) with Mth RNA ligase. Unligated linker was removed with RecJ exonuclease. Reverse transcription was performed on linker-ligated RNA with Protoscript II (primer NI-802: 5′-/5Phos/NNAGATCGGAAGAGCGTCGTGTAGGGAAAGAG/iSp18/GTGACTGGAGTTCAGACGTGTGCTC-3′), and samples were treated with 1 M NaOH to hydrolyze remaining RNAs. cDNA was size-selected with a 15% polyacrylamide TBE-Urea gel and circularized with CircLigase II. rRNA was depleted from the sample with a subtraction oligo pool, and cDNA RT was quantified via qPCR. cDNA libraries were amplified with Phusion polymerase (Forward primer NI-798: 5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTC-3′) using unique reverse index primers (reverse primer: 5′-CAAGCAGAAGACGGCATACGAGATJJJJJJGTGACTGGAGTTCAGACGTGTG-3′) and size-selected with a 15% polyacrylamide TBE-Urea gel. The quality and concentration of amplified libraries was assessed with a Bioanalyzer (Agilent) and Qubit (Invitrogen); 5 ng/sample were pooled and sequenced with the Illumina HiSeq 4000 (SE65 66x8x8x0) via the sequencing core of the UCSF Center for Advanced Technology.

RNA-seq

Total RNA was extracted from lysates using Direct-zol RNA Miniprep Kit (Zymo) and NEBNext Ultra II Directional RNA Illumina Library Prep Kit (E7760; NEB) with NEBNext rRNA Depletion Kit V2 (E7400; NEB). RNA-seq samples were prepared with single index primers (ME6609S; NEB). RNA integrity and concentration of amplified libraries were assessed with a Bioanalyzer and Qubit; 5 ng/sample was pooled and sequenced with the Illumina HiSeq 4000 (SE65 66x8x8x0) via the sequencing core of the UCSF Center for Advanced Technology.

Sequencing pre-processing

FASTQ files were converted to FASTA format. Using cutadapt, adapter sequences were removed from RNA-seq and Ribo-seq sample reads, and reads were filtered to a minimum length of 22 nt. Reads were collapsed by UMI, which were then removed. Ribosomal RNAs, repeat RNAs, and other noncoding RNAs were aligned against and removed with repeatmasker using bowtie2 2.4.1. Index files were generated using STAR 2.7.5a, GRCh38 primary assembly genome.fa file, and a GENCODE v25 .gtf annotation file with --sjdbOverhang set to 64 or 29 for RNA-seq and Ribo-seq indices, respectively. Filtered reads were then mapped to corresponding indices, generating annotated .bam files for analysis.

The number of uniquely mapped RNA-seq reads (million) for two replicates per cell line were 22.9 and 34.7 for parental; 29.3 and 30.6 for IGHMBP2 HET Clone #1; 26.8 and 29.8 for IGHMBP2 HET Clone #2; 28.3 and 29.8 for IGHMBP2 KO Clone #1; 29.8 and 33.6 for IGHMBP2 KO Clone #2. The number of uniquely mapped Ribo-seq reads (million) for two replicates per cell line were 2.0 and 3.3 for parental; 2.5 and 5.5 for IGHMBP2 HET Clone #1; 3.5 and 3.8 for IGHMBP2 HET Clone #2; 1.2 and 2.6 for IGHMBP2 KO Clone #1; 1.9 and 3.5 for IGHMBP2 KO Clone #2.

Sequencing analysis

RNA-seq and Ribo-seq results were analyzed using DESeq2 library in R (Love et al, 2014). Transcript biotypes from the corresponding GTF annotation (GENCODE v25) were matched to all genes, and “rRNA” and “Mt_rRNA”-classified genes remaining after pre-processing were removed. Mapped reads were pre-filtered by performing DESeq analysis on all samples to determine the minimal filterThreshold. The minimal filterThreshold was determined to be 3.7 for RNA-seq samples and 5.8 for Ribo-seq samples, and thus genes with counts below these cut-offs were removed from corresponding datasets. DESeq was run with pre-filtered genes with independentFiltering omitted, as this widened ability to statistically contextualize DE changes among genes otherwise abundantly filtered out with high independent filterThresholds in nuanced perturbation conditions, especially among heterozygous clones. Dispersion results were then shrunken using the apeglm method.

Because we compare more than two groups (parental, KO, and HET genotypes) a likelihood ratio test was used rather than default Wald testing, which uses standard error. The initial design used was ∼ SeqType + Condition + SeqType:Condition, and DESeq was run using likelihood ratio testing with a reduced design ∼SeqType + Condition to derive DE results reflecting SeqType:Condition (Ribo:RNA with respect to genotype). DESeq was run with filterThreshold pre-filtered genes with independentFiltering omitted. Because of the modeling design, resultant P-values are the same between IGHMBP2 KO or HET clones, and L2FC were independently determined among samples. Results were then filtered by baseMean cut-offs of > 60 for RNA-seq and baseMean > 40 for Ribo-seq. Genes were classified as significant for RNA-seq P-adjusted values < 0.01 and Ribo-seq P-adjusted values < 0.05. To further account for possible clonal artifacts and outliers, final significant DE genes among RNA-seq and Ribo-seq results must be significant in at least 2 out of 4 unique clones; genes found significant in only 1 out of the 4 clonal cell lines were filtered from the final datasets.

R packages used to assess quality control of sequencing data include MultiQC, Ribo-seQC, and RibosomeprofilingQC (Ewels et al, 2016; Calviello et al, 2019 Preprint). To determine localized read mapping coverage for specific genes, such as ATF4 uORFs and CDS, bamCoverage from deeptools (Ramírez et al, 2016) was used to convert RNA-seq and Ribo-seq pre-processed .bam files to .bigwig and .bed files, where reads were normalized to counts per million within bins separated by 50 nt. Counts per million-scaled reads were then mapped to hg38 using Interactive Genomics Viewer. Normalized counts were extracted from .bed files per genomic regions analyzed corresponding to chr22: 39,520,650-39,520,700 (uORF1); 39,520,750-39,520,800 and 39,521,350-39,521,400 (uORF2, separated by splice-site); 39,521,450-39,522,600 (CDS).

Gene set enrichment analysis

Using RNA-seq, Ribo-seq, and Ribo:RNA-seq results output from DESeq analysis, the L2FCs per gene were averaged between cell lines per analysis type. Mean L2FCs were then sorted as ranked lists, and GSEA was performed using fgsea and clusterprofiler R packages using the Biological Process sub-ontology (Yu et al, 2012; Korotkevich et al, 2016 Preprint; Wu et al, 2021). The minimum and maximum gene set sizes were set to 25 and 1,000, respectively. GSEA was performed with 100,000 permutations, and exported gene sets reflect a P-value cutoff of 0.01. The enrichment map plot was generated using the enrichplot library.

Metagene analysis

Metagene analysis was performed using ribosomeProfilingQC library in R with .bam files and the coverageDepth function. The GENCODE v25 .gtf annotation file was used to define UTR and CDS regions. Metagene profiles were then Min-Max normalized between 0 and 1 for comparability between each clone.

Codon enrichment analysis

Ribo-seq reads were mapped to the transcriptome using STAR 2.7.5a --quantMode TranscriptomeSAM. Transcriptomic .bam files were assigned posterior probability values using RSEM (rsem-calculate-expression) with --fragment-length-mean 29 for analysis with the choros R library (Mok et al, 2023 Preprint). For parental and IGHMBP2 KO Clone #2 samples, A-site offset rule values were determined to be 15, 14, and 16 for frame 0, 1, and 2 for reads between lengths 29–30 nt; 15, 14, and NA for frame 0, 1, and 2 for 28 nt reads; NA, 14, and NA for frame 0, 1, and 2 for 27 nt reads, respectively. A-site offset rules were less discernible in IGHMBP2 KO Clone #1, and thus was omitted from codon enrichment analysis.