Beyond the Monolithic Gene

What ASHG has to say about gene transcripts, transcript isoforms, protein isoforms, and proteoforms, along with the research tools to come

Montreal Canada was this year’s venue for the annual American Society of Human Genetics (ASHG) conference. I can say with some assurance that this year was quite a bit more subdued than in years past: there were no splashy product launch parties this time around. I’m thinking of 2023 where ASHG was in Washington DC and the highlight was a rapper Macklemore… I can still remember his signature song “Thrift Shop”.

Yet there was plenty of genetics and genomics to report on, perhaps more subdued this year due to tight budgets and funding cuts. Even the plenary sessions did not have huge “big science” announcements or datasets to discuss. But the research continues.

r = 0.29

In 2025 Pacific Biosciences announced the addition of long read sequencing to the 1000 Genomes Project, in order to determine isoform transcript information from about 1,000 samples. 

One of the laboratories named in the announcement was from Johns Hopkins, and a post-doctoral fellow Margaret Starostik from Rajiv McCoy’s lab presented at ASHG some unpublished results.

For those unfamiliar with The Thousand Genomes Project (abbreviated 1KGP), its goal was to increase ancestral diversity by extensive genomic characterization across 26 populations. Previous to this long-read work, the McCoy group characterized a short-read RNA-seq dataset from 731 lymphoblastoid cell lines, and their dataset called MAGE was published in Nature in 2024. (Taylor et al. 2024)

Most RNA-seq data comes from short reads, which measure gene-level expression accurately but infer which isoform a particular read originated from.

Dr. Starostik presented long-read isoform data from a preliminary dataset of 630 samples (out of a total of 900 total samples the McCoy lab expects to complete in this dataset). They found that some 67% of isoforms were previously uncatalogued combinations of known splice sites, with some of these detected across all 630 samples analyzed so far. 

Comparing short-read and long-read estimates of expression in the same samples, agreement was r=0.83 at the gene level, but only r=0.29 at the transcript level.  She mentioned paralogous genes and pseudogenes and lncRNA adding noise to the measurements. 

A single gene with many products

Even though there were no “headline talks”, I found many talks looking at transcript isoforms or alternative splicing defects associated with disease. Instead of looking at a monolithic gene, we are now looking at many forms of the same gene in the transcript, that end up as protein variants or isoforms. 

Here is where it’s important to keep track of how genetic variation can show up in what is being transcribed, which then affects what is translated into protein, as well as how those end proteins can be further modified. 

Variants in the genome can be a coding variant, also known as a non-synonymous variant. Thus a coding region that reads AAA – ATC – AAG – GGG has a wild type four amino acids that codes for Lysine – Isoleucine – Lysine – Glycine. Let’s say the second triplet “ATC” has a mutation that changes it to “ACC”. The mutated non-synonymous variant now has the four amino acids that are now Lysine – Threonine – Lysine – Glycine. 

This is one clear change of a proteoform, where the wildtype and the mutation caused a different protein product to be produced. These can be on separate alleles, and each allele may be co-expressed, or one as a majority, or one completely shut off. 

The work at Hopkins looks at alternative splice variants from a single gene. Exons can be assembled selectively. Human protein-coding transcripts have a median of nine exons (Piovesan et al. 2019). The GENCODE catalog (Mudge et al. 2025) lists about nine transcripts per protein-coding gene, and the curators of GENCODE expect this number to climb into the tens or even  hundreds as more long-read RNA-seq data is generated. 

Keep in mind that for a given tissue one isoform typically dominates, and not every isoform codes for a unique protein. (Some differ only in untranslated RNA sequence, and have undefined function.)

Examples

One interesting talk was titled “The aging human transcriptome” by Max Tomlinson of King’s College London, looking at 95 twin pairs, average age of 62 years old, followed up after 9 years. The isoforms measured were RNA from PBMCs, read out on Oxford Nanopore, with an average readlength of 850 basepairs. 

The results reported that 3,200 genes and 1,600 transcripts changed over time, with isoform diversity increasing with time, and non-dominant isoforms also increasing over time. They also tracked differences between twin pairs as a function of BMI, interestingly with higher BMI there was higher isoform diversity, but the cause of these changes could not be determined. 

In a rare disease track Elena Cabello from the University of Zurich presented a talk about a family affected by early-onset scoliosis. She identified an autosomal dominant, fully penetrant variant through fine-mapping. The variant sits deep in an intron of LRRFIP1, 34 kb in, and long-read sequencing revealed an isoform shift. 

The Broad Institute’s Laura Domenech presented a talk titled “Pivoting from gene-level analyses to isoform-resolved biology”. This is work as part of the Developmental Genotype-Tissue Expression project (“dGTEx”) announced in January 2025 (Nature perspective article link). Samples are from 74 tissues from 120 post-mortem donors, that span the age groups 0 to 2 years, 2 to 8 years, 8 to 12.5 years, and 12.5 years to 18 years of age. There is also a parallel project called NHP dGTEx with corresponding non-human primates (rhesus macaques and common marmosets). 

She reported 5,000 differentially spliced genes (dDSG), and 81.4% of dDSGs are not differentially expressed. In addition, a full one-third show evidence of genetic constraint. 

She then went on to elaborate on a single gene MYBPC1 and how exons are conserved in infancy but increasingly show exon skipping with age. This pattern is conserved in non-human primates.

A Reasonable Doubt

Most genes produce only a single dominant protein isoform. Naturally if there were significant functional differences due to different isoforms across disease it would be a much better studied subject. Alternative splicing has been known since the 1970’s, have five decades of research shown its significance?

There are specific splice variants directly tied to disease, and that number is growing. But how much genome-wide isoform diversity has a specific function?

 For single nucleotide polymorphisms, variants of pathogenic significance are a rich field as interpretation of variant effect is practical translation of genetics to disease. For transcript isoforms the significance to disease is less clear, in addition to its role in development and function. However as the tools improve (the availability and affordability of long read sequencing in particular) there will be continued ongoing work. 

A Word about Nautilus

Nautilus Biotechnology is developing a single-molecule protein analysis platform, and through the use of several rounds of specific binding molecules and detection they can enable what they call Iterative Mapping. Billions of protein molecules are captured, and rounds of fluorescent probes are bound, detected through imaging, washed, and bound again. Recently they announced their initial focus on disease-specific proteins, starting with neurodegenerative disease and oncology, looking at proteoforms and PTMs while also continuing development of their full proteome application. 

Nautilus Biotechnology (link to website) presented a poster at ASHG detailing their work on an Alzheimer’s diagnostic protein called Microtubule-associated protein tau (gene MAPT). They can identify up to 768 proteoforms across 6 brain-derived full-length tau isoforms (0N3R, 0N4R, 1N3R, 1N4R, 2N3R, 2N4R), and for post-translational modifications (PTMs) they can detect pT181, pS202+pT205, pT205, pS214, pT217, pT231, pS396. Thus the “up to 768” can be determined by this math: 6 isoforms x 2⁷ patterns of phosphorylation = 6 x 128 = 768. 

These seven phosphorylation sites are significant, as they are associated with specific stages of Alzheimer’s disease. For example, pT217 is the leading blood biomarker for Alzheimer’s, and pT181 is a well-established marker in both CSF and blood. The unique value of the platform, in contrast to ELISA or multiplex protein assays from Olink or Alamar or Meso Scale Discovery, is in its ability to look at a single contiguous captured protein molecule, and interrogate the phosphorylation pattern across that protein. 

In September 2026 Nautilus announced (press release pdf) a Nature Methods paper (link) that demonstrates the detection of 130 distinct proteoforms of tau, with as many as six phosphorylation events. They demonstrate the pattern of phosphorylation suggests it is an ordered and site-specific process, rather than stochastic events. 

Nautilus is currently in early access, expecting to launch their single-molecule detection instrument called Nautilus Voyager™ in spring or summer of 2027, along with tau, AKT1, and an additional oncology proteoform assay. Additional assays are also under development. For AKT1, they are aiming to determine up to 32 proteoforms and 5 PTMs, and for EGFR, which is also in development, up to 64 proteoforms and 6 PTMs.

Not so quiet after all

This year there were no blowout parties, and the exhibit hall looked noticeably smaller than in prior years. But sitting in the sessions and looking at the posters there was a noticeable pattern: the gene is giving way to the transcript isoform, and the protein giving way to the proteoform. The Hopkins r=0.29 is less a shortcoming of short reads than an indication of how much biology remains to be discovered and characterized. 

The tools for isoform characterization are maturing. The tools for proteoform detection are less mature. Bottom-up mass spectrometry breaks proteins into peptides, losing the contiguous molecule; top-down mass spectrometry can measure intact proteoforms but is limited in quantification and ease of use. Newer approaches like Nautilus offer a tantalizing opportunity: reading PTMs across a single contiguous molecule, with the quantification and ease of use that could move the entire field forward. 

And what about reference standards? Genome in a Bottle gave the field of variant calling a standard still leveraged and in use for benchmarking. Spike-ins like SIRVs and Sequins, and the LRGASP benchmark, are a start, but for proteoforms it’s a wide-open opportunity. 

A quiet meeting, but not a quiet field.

References

Piovesan A, et al. Human protein-coding genes and gene feature statistics in 2019. BMC Res Notes. 2019;12:315. https://doi.org/10.1186/s13104-019-4343-8

Mudge JM, et al. GENCODE 2025: reference gene annotation for human and mouse. Nucleic Acids Res. 2025;53(D1):D966-D975. https://doi.org/10.1093/nar/gkae1078

Taylor DJ, et al. Sources of gene expression variation in a globally diverse human cohort. Nature 2024;532(8023):122-130. https://doi.org/1038/s41586-024-07708-2

Disclosure: The author is a former employee of both Olink Proteomics and Meso Scale Diagnostics, and consults for life science companies through Silent Valley Consulting. None of the companies mentioned are clients. 

Acknowledgement: Special thanks to Sheri Wilcox, Ph.D., VP of Scientific Engagement at Nautilus Biotechnology, for reviewing the Nautilus section for accuracy.