Earliest chemical evidence of life identified in South African rock
Researchers report that fragmentary, fossilized carbon from the Josefsdal Chert in Mpumalanga province, South Africa, contains the most confident chemical signature of ancient life yet identified on Earth. Using advanced pattern recognition and chemical analysis, the team dated the sample to about 3.33 billion years and found biotic signals that persist despite extreme geological alteration.
Combining Py‑GC‑MS with machine learning
The study analyzed 406 samples spanning modern organisms to ancient rocks using pyrolysis‑gas chromatography‑mass spectrometry (Py‑GC‑MS). Py‑GC‑MS thermally breaks organic material into fragments, separates them, and records mass spectra that reflect the molecular composition of the sample. Instead of relying on single diagnostic molecules, the researchers trained a machine learning model to recognize distributed chemical patterns characteristic of biological origin.
After cataloguing patterns from well‑preserved, younger samples, the team used those patterns to train an algorithm to detect subtler, degraded signals. The model achieved better than 90% accuracy in distinguishing biological from nonbiological chemical signatures across the dataset.
Oldest photosynthesis signals and broader implications
In addition to the 3.33‑billion‑year result, the team reports the oldest chemical evidence for photosynthesis in rocks dated to about 2.52 billion years (South Africa) and 2.3 billion years (Canada), extending the reliable molecular record of that metabolic process by roughly 800 million years.
Samples in the study ranged from nearly modern material back to roughly 3.8 billion years, including carbon‑bearing rocks from Greenland and 3.5‑billion‑year‑old stromatolites from Australia. Younger samples (generally under ~500 million years) produced strong and obvious biological signatures; signals gradually fade with age as geological processes strip away molecular detail.
"Ancient life leaves more than fossils; it leaves chemical 'echoes'," says mineralogist and astrobiologist Robert Hazen, a senior author on the paper. "By pairing powerful chemical analysis with machine learning, we can now read molecular 'ghosts' that still whisper their secrets after billions of years."
Limitations and cautious interpretation
The team emphasizes that a lack of a detectable signal in older samples does not prove absence of life: extreme degradation can erase the chemical patterns the algorithm relies on. Thus, while the Josefsdal Chert result provides the most confident chemical evidence so far that life had emerged and spread by 3.33 billion years ago, the true origin of life may be earlier but remain undetectable in heavily altered rocks.
The lead authors include Michael Wong and Anirudh Prabhu, with Robert Hazen among the senior contributors. The research has been published in the Proceedings of the National Academy of Sciences.
Why this matters
This approach—looking for distributed chemical patterns rather than single biomarkers—improves our ability to read degraded molecular records and could inform searches for past life on other planets and moons. By extending the timeline over which we can confidently identify biological chemistry, these methods add a powerful tool for decoding Earth's deep biological history.
