Fig 2: NUDT21 depletion induces aberrant alternative polyadenylation and altered protein levels.(A)(i) Poly(A)-ClickSeq (PAC-seq) track showing reduced NUDT21 and (ii-iv) PAC-seq tracks showing altered alternative polyadenylation in example genes MECP2, VMA21, and PAK1. n = 3/treatment. Peaks are 3' end sequencing reads. Multiple peaks in a gene indicate multiple mRNA isoforms with different cleavage and polyadenylation sites. Bracketed numbers show counts per million; kb stands for kilobases. (B) Schemata of the mRNA isoforms identified by PAC-seq for MECP2, VMA21, and PAK1 (top). Triangles show the binding sites of the qPCR primers used to validate the PAC-seq results. The gray primers detect all mRNA isoforms for the target gene, whereas the green primers only detect the isoform with the longest 3' UTR. Gapdh-normalized RT-qPCR quantification showing relatively less of the long isoforms in the shNUDT21-infected neurons (bottom) for MECP2, p=0.0007 (i), VMA21, p=0.01 (ii), and PAK1, p=0.005 (iii). Error bars indicate SEM. Data analyzed by unpaired, two-tailed t-test. (C) Volcano plot showing relative mRNA length change in shNUDT21-infected neurons compared to shScramble-infected controls. The horizontal, dashed line shows Padjusted = 0.05, n = 3/treatment. NUDT21 loss in neurons predominantly results in shorter mRNAs (p<0.0001, two-tailed chi-square test). >90% of reads are in 3' UTRs (Figure 6—figure supplement 1A). Principle component analysis (PCA) shows sample separation by treatment (Figure 6—figure supplement 1B). Distal cleavage sites of NUDT21-regulated mRNAs are enriched for the CFIm25 binding motif, UGUA, in mRNAs that shorten after NUDT21 knockdown, but not in non-target mRNAs (Figure 6—figure supplement 1C). (E) Mass spectrometry quantification of protein level fold change for genes with significantly altered APA (Padjusted <0.05). mRNA shortening predominantly results in increased protein levels (p<0.0001, two-tailed conditional chi-square test). Source files for the PAC-seq and mass spectrometry quantification data are available in Figure 6—source data 1. *p<0.05; **p<0.01; ***p<0.001.Figure 6—source data 1.RNA length and protein level changes in genes with misregulated APA following NUDT21 inhibition in human neurons.Alternative polyadenylation (APA), likelihood ratio (lr), degrees of freedom (df), first polyadenylation site (F), middle/center polyadenylation site (M), last polyadenylation site (L) mass spectrometry (MS), fold change (FC), intellectual disability (ID), probability of loss of function intolerance (pLI).

Fig 3: Partial loss of Nudt21-encoded CFIm25 protein causes learning deficits.(A) Nudt21+/- mice have conditioned fear learning deficits: fear learning, p=0.0009; contextual fear, p=0.045; cued fear, p=0.0009. (B) The open field assay shows that Nudt21+/- mice are no more active (i) but spend relatively more time in the center of the open field (ii), indicating reduced anxiety: p=0.01. (C) Nudt21+/- mice have spatial learning deficits in the Morris water maze. (i) They take longer to find the hidden platform during the training blocks (p=0.026, two-way, repeated measures ANOVA), (ii) despite swimming faster and farther (p=0.013). (iii and iv) When the hidden platform is removed in the probe trial, they spend less time in the quadrant that previously had the platform (p=0.039, Sidak’s multiple comparisons test). For all assays, mice were between 30–40 weeks of age. Error bars indicate SEM. Representative tracks are from the animal with the median result in each genotype. All data were analyzed by unpaired, two-tailed t-test unless otherwise stated. *p<0.05; ***p<0.001.

Fig 4: Distinct and opposing roles for CFIm59 and CFIm68 on global APA. (A) Up- and downregulated distal and proximal PAS counts upon CFIm KD or KO. Analyses of our NIH3T3 3' mRNA-seq and public datasets of HEK293T PAS-seq were performed to tabulate each PAS expression change upon CFIm disruption. Statistical significance was calculated with DEXSeq or Fisher's exact test and corrected for multiple testing. Dark gray indicates PAS differential expression with false discovery rate (FDR) =0.05 and light gray for FDR >0.05. (B) Relative expression difference (RED) score distribution of NIH3T3 and HEK293T upon CFIm disruption. RED of each gene was calculated as the log2 difference between KD (or KO) distal/proximal and control distal/proximal ratio. Paired two-tailed Student's t-test was used (****P= 0.0001). (C) Overlap of 3'UTRs shortened upon CFIm68 KD and lengthened upon CFIm59 KD in NIH3T3. (D, E) Genome tracks of example cancer-related mRNAs whose 3'UTRs lengthen upon CFIm59 depletion and shorten upon CFIm25 and -68 depletion in NIH3T3 (D) and HEK293T (E) cells.

Fig 5: CFIm controls Pten mRNA 3'UTR isoform expression. (A) Representative Pten tracks of 3'UTR-seq upon CFIm25, -59 and -68 KD in NIH3T3. Schematic of Pten and 3'UTR isoforms previously captured by 3' RACE (13) is shown at the bottom. Each track is separated into two parts with different scales indicated as count ranges on top to show all major isoforms. (B) Quantitative northern blotting of endogenous Pten upon CFIm KD in NIH3T3. Corresponding 3'UTR isoforms are indicated on the right with the size markers on the left. Bar graph on the bottom tabulates the average relative composition of the 300 nt, 3.3k and 5-6k isoforms across three biological replicates. (C) Pten 3'UTR isoform-specific qRT-PCR for the 300 nt, 3.3k and 5-6k isoforms upon CFIm KD in NIH3T3. (D) CFIm59 and CFIm68 expression rescue in respective KO cells. (E) Pten 3'UTR isoform ratio shift upon CFIm59 and CFIm68 rescue. APA 300 nt and APA 3.3k levels are measured by qRT-PCR and taken as ratios before normalization to the mock transfection control. (F) 3'UTR-seq tracks of Pten upon CFIm59 and CFIm68 rescue. All experiments were performed in at least three biological replicates with technical triplicates for qRT-PCR. Error bars indicated are mean ± standard deviation. P-values were calculated with one-way ANOVA followed by Dunnett's test in (C) and with Student's t-test in (E) (*P= 0.05, **P= 0.01, ***P= 0.001, ****P= 0.0001, and ns not significant).